id
stringlengths 40
40
| metadata
dict | text
stringlengths 9
2.66M
| created
timestamp[s]date 2025-03-04 00:00:00
2025-03-07 00:00:00
| source
stringclasses 1
value | attributes
dict |
|---|---|---|---|---|---|
bf3127272576d9143beaf28c894d26b29d131433 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 12,
"total-fallback-pages": 0,
"total-input-tokens": 44354,
"total-output-tokens": 8546,
"fieldofstudy": [
"Physics"
]
} | Modified Function Projective Synchronization between Different Dimension Fractional-Order Chaotic Systems
Ping Zhou¹,² and Rui Ding¹
¹ Research Center for System Theory and Applications, Chongqing University of Posts and Telecommunications, Chongqing 400065, China
² Key Laboratory of Industrial Internet of Things and Networked Control of the Ministry of Education, Chongqing University of Posts and Telecommunications, Chongqing 400065, China
Correspondence should be addressed to Ping Zhou, [email protected]
Received 2 August 2012; Accepted 13 August 2012
Academic Editor: Jinhu Lü
A modified function projective synchronization (MFPS) scheme for different dimension fractional-order chaotic systems is presented via fractional order derivative. The synchronization scheme, based on stability theory of nonlinear fractional-order systems, is theoretically rigorous. The numerical simulations demonstrate the validity and feasibility of the proposed method.
1. Introduction
Fractional-order calculus, which can be dated back to the 17th century [1, 2]. However, only in the last few decades, its application to physics and engineering has been addressed. So, the fractional-order calculus has attracted increasing attention only recently. On the other hand, complex bifurcation and chaotic phenomena have been found in many fractional-order dynamical systems. For example, the fractional-order Lorenz chaotic system [3], the fractional-order unified chaotic system [4], the fractional-order Chua chaotic circuit [5], the fractional-order modified Duffing chaotic system [6], and the fractional-order Rössler chaotic system [7, 8], and so on.
Nowadays, synchronization of chaotic systems and fractional-order chaotic systems has attracted much attention because of its applications in secure communication and control processing [9–21]. Many approaches have been reported for the synchronization of chaotic systems and fractional-order chaotic systems [12–19]. In 1999, Mainieri and Rehacek proposed projective synchronization (PS) [12] for chaotic systems, which has
been extensively investigated in recent years because of its proportional feature in secure communications. Recently, a modified projective synchronization, which is called function projective synchronization (FPS) [13–15] has been reported. In FPS, the master and slave systems could be synchronized up to a scaling function, but not a constant. So, the unpredictability of the scaling function in FPS can additionally enhance the security of communication.
To the best of our knowledge, most of the existing FPS scheme for the fractional-order chaotic systems only discuss the same dimension. However, in many real physics systems, the synchronization is carried out through the oscillators with different dimension, especially the systems in biological science and social science [16–21]. Moreover, in some previous works [16, 17], all the nonlinear terms of response system or error system was absorbed. Referring to chaotic synchronization via fractional-order controller, there are a few results reported until now. Inspired by the above discussion, in this paper, we present a modified function projective synchronization (MFPS) scheme between different dimension fractional-order chaotic systems via fractional-order controller. The fractional-order controller is easily designed. The synchronization technique, based on tracking control and stability theory of nonlinear fractional-order systems, is theoretically rigorous. Our modified function projective synchronization (MFPS) scheme need not absorb all the nonlinear terms of response system. This is different from some previous works. Two examples are presented to demonstrate the effectiveness of the proposed MFPS scheme.
This paper is organized as follows. In Section 2, a modified function projective synchronization (MFPS) scheme is presented. In Section 3, two groups of examples are used to verify the effectiveness of the proposed scheme. The conclusion is finally drawn in Section 4.
2. The MFPS Scheme for Different Dimension Fractional-Order Chaotic Systems
The fractional-order chaotic drive and response systems with different dimension are defined as follows, respectively:
\[
\frac{d^{q_d}x}{dt^{q_d}} = F_d(x),
\]
\[
\frac{d^{q_r}y}{dt^{q_r}} = F_r(y) + C(x, y),
\]
where \(q_d (0 < q_d < 1)\) and \(q_r (0 < q_r < 1)\) are fractional order, and \(q_d\) may be different with \(q_r\). \(x \in \mathbb{R}^n, y \in \mathbb{R}^m \ (n \neq m)\) are state vectors of the drive system (2.1) and response system (2.2), respectively. \(F_d : \mathbb{R}^n \rightarrow \mathbb{R}^n, F_r : \mathbb{R}^m \rightarrow \mathbb{R}^m\) are two continuous nonlinear vector functions, and \(C(x, y) \in \mathbb{R}^m\) is a controller which will be designed later.
Definition 2.1. For the drive system (2.1) and response system (2.2), it is said to be modified function projective synchronization (MFPS) if there exist a controller \(C(x, y)\) such that:
\[
\lim_{t \to +\infty} \|e\| = \lim_{t \to +\infty} \|y - M(x)x\| = 0,
\]
\(\)
where \( \| \cdot \| \) is the Euclidean norm, \( M(x) \) is a \( m \times n \) real matrix, and matrix element \( M_{ij}(x) \) \((i = 1, 2, \ldots, m, j = 1, 2, \ldots, n) \) are continuous bounded functions. \( e_i = y_i - \sum_{j=1}^{n} M_{ij}x_j \) \((i = 1, 2, \ldots, m) \) are called MFPS error.
**Remark 2.2.** According to the view of tracking control, \( M(x)x \) can be chosen as a reference signal. The MFPS in our paper is transformed into the problem of tracking control, that is the output signal \( y \) in system (2.2) follows the reference signal \( M(x)x \).
In order to achieve the output signal \( y \) follows the reference signal \( M(x)x \). Now, we define a compensation controller \( C_1(x) \in R^m \) for response system (2.2) via fractional-order derivative \( d^{\nu_i}(M(x)x)/dt^{\nu_i} \). The compensation controller is shown as follows:
\[
C_1(x) = \frac{d^{\nu_i}(M(x)x)}{dt^{\nu_i}} - F_r(M(x)x),
\]
(2.4)
and let controller \( C(x, y) \) as follows:
\[
C(x, y) = C_1(x) + C_2(x, y),
\]
(2.5)
where \( C_2(x, y) \in R^m \) is a vector function which will be designed later.
By controller (2.5) and compensation controller (2.4), the response system (2.2) can be changed as follows:
\[
\frac{d^{\nu_i}e}{dt^{\nu_i}} = D_1(x, y)e + C_2(x, y),
\]
(2.6)
where \( D_1(x, y)e = F_r(y) - F_r(M(x)x) \), and \( D_1(x, y) \in R^{m \times m} \). So, the MFPS between drive system (2.1) and response system (2.2) is transformed into the following problem: choose a suitable vector function \( C_2(x, y) \) such that system (2.6) is asymptotically converged to zero.
In what follows we present the stability theorem for nonlinear fractional-order systems of commensurate order [22–25]. Consider the following nonlinear commensurate fractional-order autonomous system
\[
D^q x = f(x),
\]
(2.7)
the fixed points of system (2.7) is asymptotically stable if all eigenvalues (\( \lambda \)) of the Jacobian matrix \( A = \partial f / \partial x \) evaluated at the fixed points satisfy \( |\arg \lambda| > 0.5 \pi q \). Where \( 0 < q < 1 \), \( x \in R^n \), \( f : R^n \rightarrow R^n \) are continuous nonlinear functions, and the fixed points of this nonlinear commensurate fractional-order system are calculated by solving equation \( f(x) = 0 \).
Now, the following theorem is given based on the above discussion in order to achieve the MFPS between the drive system (2.1) and the response system (2.2).
**Theorem 2.3.** Choose the control vector \( C_2(x, y) = D_2(x, y)e \), and if \( D_1(x, y) + D_2(x, y) \) satisfy the following conditions:
(1) \( d_{ij} = -d_{ji} \) \((i \neq j)\),
(2) \( d_{ii} \leq 0 \) \( \) (all \( d_{ii} \) are not equal to zero),
then the modified function projective synchronization (MFPS) between (2.1) and (2.2) can be achieved. Where $D_1(x, y) \in \mathbb{R}^{m \times m}$, and $d_{ij}$ ($i, j = 1, 2, \ldots, m$, for all $d_{ij} \in \mathbb{R}$) are the matrix element of matrix $D_1(x, y) + D_2(x, y)$.
**Proof.** Using $C_2(x, y) = D_2(x, y)e$, so fractional-order system (2.6) can be rewritten as follows:
$$
\frac{d^\nu e}{dt^\nu} = [D_1(x, y) + D_2(x, y)]e. \tag{2.8}
$$
Suppose $\lambda$ is one of the eigenvalues of matrix $D_1(x, y) + D_2(x, y)$ and the corresponding non-zero eigenvector is $q_r$, that is,
$$
[D_1(x, y) + D_2(x, y)]q_r = \lambda q_r. \tag{2.9}
$$
Take conjugate transpose ($H$) on both sides of (2.9), we yield
$$
\{[D_1(x, y) + D_2(x, y)]q_r\}^T = \overline{\lambda} q_r^H. \tag{2.10}
$$
Equation (2.9) multiplied left by $q_r^H$ plus (2.10) multiplied right by $q_r$, we derive that
$$
q_r^H \left\{[D_1(x, y) + D_2(x, y)] + [D_1(x, y) + D_2(x, y)]^H \right\}q_r = q_r^H q_r \left(\lambda + \overline{\lambda}\right). \tag{2.11}
$$
So,
$$
\lambda + \overline{\lambda} = \frac{q_r^H \left\{[D_1(x, y) + D_2(x, y)] + [D_1(x, y) + D_2(x, y)]^H \right\}q_r}{q_r^H q_r}. \tag{2.12}
$$
Because $d_{ij} = -d_{ji}$ ($i \neq j$, for all $d_{ij} \in \mathbb{R}$) in matrix $D_1(x, y) + D_2(x, y)$, so
$$
\lambda + \overline{\lambda} = \frac{q_r^H \begin{pmatrix} 2d_{11} & 0 & \cdots & 0 \\ 0 & 2d_{22} & \cdots & 0 \\ \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & \cdots & 2d_{mm} \end{pmatrix} q_r}{q_r^H q_r}. \tag{2.13}
$$
Because $d_{ii} \leq 0$ (for all $d_{ii} \in \mathbb{R}$), and all $d_{ii}$ are not equal to zero. So,
$$
\lambda + \overline{\lambda} \leq 0. \tag{2.14}
$$
From (2.14), we have
$$
|\arg \lambda [D_1(x, y) + D_2(x, y)]| \geq 0.5\pi > 0.5q_r\pi. \tag{2.15}
$$
Abstract and Applied Analysis
According to the stability theorem for nonlinear fractional-order systems of commensurate order [22–25], system (2.8) is asymptotically stable. That is
$$
\lim_{t \to +\infty} \|e\| = 0.
$$
(2.16)
Therefore,
$$
\lim_{t \to +\infty} \|e\| = \lim_{t \to +\infty} \|y - M(x)x\| = 0.
$$
(2.17)
This indicates that the modified function projective synchronization between drive system (2.1) and response system (2.2) will be obtained. The proof is completed. □
**Remark 2.4.** Theorem 2.3 indicates that the condition of the MFPS between drive system (2.1) and response system (2.2) are \(|\arg \lambda[D_1(x, y) + D_2(x, y)]| > 0.5q_r\pi\). So, in practical applications, we can easily choose the matrix \(D_2(x, y)\) according to the matrix \(D_1(x, y)\). Moreover, in order to reserve all the nonlinear terms in response system or error system, the controller in our work may be complex than the controller reported by [16, 17]. But, all the nonlinear terms in response system or error system are absorbed in [16, 17].
**Remark 2.5.** Perhaps our result can be extended to the modified function projective synchronization of complex networks of fractional order chaotic systems [26–28] and the complex fractional-order multi-scroll chaotic systems [29–31]. But, the modified function projective synchronization for complex networks and complex fractional-order multi-scroll chaotic systems would be much more complex. Further work on this issue is an ongoing research topic in our group.
**3. Applications**
In this section, to illustrate the effectiveness of the proposed MFPS scheme for different dimension fractional-order chaotic systems. Two groups of examples are considered and their numerical simulations are performed.
**3.1. The MFPS between 3-Dimensional Fractional-Order Lorenz System and 4-Dimensional Fractional-Order Hyperchaotic System**
The fractional-order Lorenz [3] system is described as follows:
$$
\begin{align*}
D^{\alpha} y_1 &= 10(y_2 - y_1) \\
D^{\alpha} y_2 &= 28y_1 - y_2 - y_1 y_3 \\
D^{\alpha} y_3 &= y_1 y_2 - \frac{8y_3}{3}.
\end{align*}
$$
(3.1)
The fractional-order Lorenz system exhibits chaotic behavior [3] for \(q_r \geq 0.993\). The chaotic attractor for \(q_r = 0.995\) is shown in Figure 1.
Recently, Pan et al. constructed a hyperchaotic system [17]. Its corresponded fractional-order system is described as follows:
\[
\begin{align*}
D^x_1x_1 &= 10(x_2 - x_1) + x_4 \\
D^x_2x_2 &= 28x_1 - x_1x_3 \\
D^x_3x_3 &= x_1x_2 - \frac{8x_3}{3} \\
D^x_4x_4 &= -x_1x_3 + 1.3x_4.
\end{align*}
\]
The hyperchaotic attractor of system (3.2) for \( q_d = 0.95 \) is shown in Figure 2.
Consider the fractional-order hyperchaotic system (3.2) with fractional-order \( q_d = 0.95 \) as drive system, and the fractional-order Loren system with fractional-order \( q_r = 0.995 \) as response system. According to the above mentioned, we can obtain
\[
F_r(y) - F_r(M(x)x) = D_1(x, y)e = \begin{pmatrix}
-10 & 10 & 0 \\
28 - y_3 & -1 - \sum_{j=1}^{4} M_1jx_j \\
y_2 & \sum_{j=1}^{4} M_1jx_j & -\frac{8}{3}
\end{pmatrix} e. \quad (3.3)
\]
Now, we can choose
\[
D_2(x, y) = \begin{pmatrix}
0 & 0 & -y_2 \\
-38 + y_3 & 0 & 0 \\
0 & 0 & 0
\end{pmatrix}.
\] (3.4)
So,
\[
D_1(x, y) + D_2(x, y) = \begin{pmatrix}
-10 & 10 & -y_2 \\
-10 & -1 & -\sum_{j=1}^{4} M_{1j} x_j \\
y_2 & \sum_{j=1}^{4} M_{1j} x_j & \frac{8}{3}
\end{pmatrix}.
\] (3.5)
According to the above theorem, the MFPS between the 3-dimensional fractional-order Lorenz system (3.1) and the 4-dimensional fractional-order hyperchaotic system (3.2) can be achieved. For example, choose \( M(x) = \begin{pmatrix}
1 & x_2 & 2 & 1 \\
0 & x_1 & 2 & 1 \\
0 & 0 & 1 & 1 \\
0 & 0 & 0 & 3
\end{pmatrix} \). The corresponding numerical result is shown in Figure 3, in which the initial conditions are \( x(0) = (2, 1, 2, 1)^T \) and \( y(0) = (18, 13, 13.5)^T \), respectively.
### 3.2. The MFPS between 4-Dimensional Fractional-Order Hyperchaotic Lü System and 3-Dimensional Fractional-Order Arneodo Chaotic System
In 2002, Lü and Chen reported a new chaotic system [32], which be called Lü chaotic system. The Lü chaotic system is different from the Lorenz and Chen system. Based on Lü chaotic system, the hyperchaotic Lü chaotic system and the fractional-order hyperchaotic Lü system have been constructed recently. The fractional-order hyperchaotic Lü system [16] is described by the following
\[
\begin{align*}
D^q y_1 &= 36(y_2 - y_1) + y_4 \\
D^q y_2 &= 20y_2 - y_1y_5 \\
D^q y_3 &= y_1y_2 - 3y_5 \\
D^q y_4 &= y_1y_3 - y_4.
\end{align*}
\] (3.6)
The hyperchaotic attractor of system (3.6) for \( q_r = 0.96 \) is shown in Figure 4.
The fractional order Arneodo chaotic system [16] is defined as follows:
\[
\begin{align*}
D^q x_1 &= x_2 \\
D^q x_2 &= x_3 \\
D^q x_3 &= 5.5x_1 - 3.5x_2 - x_3 - x_1^2.
\end{align*}
\] (3.7)
The chaotic attractor of system (3.7) for \( q_d = 0.998 \) is shown in Figure 5.
Consider the fractional-order Arneodo chaotic system (3.7) with fractional-order \( q_d = 0.998 \) as drive system, and the fractional-order hyperchaotic \( \tilde{\text{L}} \) system (3.6) with fractional-order \( q_r = 0.96 \) as response system. According to the above mentioned, we can yield
\[
F_r(y) - F_r(M(x)x) = D_1(x,y)e = \begin{pmatrix}
-36 & 36 & 0 & 1 \\
-y_3 & 20 & -\sum_{j=1}^{3} M_1 x_j & 0 \\
y_2 & \sum_{j=1}^{3} M_1 x_j & -3 & 0 \\
y_3 & 0 & \sum_{j=1}^{3} M_1 x_j & -1
\end{pmatrix} e. \tag{3.8}
\]
Now, we can choose
\[
D_2(x, y) = \begin{pmatrix}
0 & 0 & -y_2 & 0 \\
-36 + y_3 & -21 & 0 & 0 \\
0 & 0 & 0 & -\sum_{j=1}^{3} M_{1j} x_j \\
-1 - y_3 & 0 & 0 & 0
\end{pmatrix}.
\] (3.9)
So,
\[
D_1(x, y) + D_2(x, y) = \begin{pmatrix}
-36 & 36 & -y_2 & 1 \\
-36 & -1 & -\sum_{j=1}^{3} M_{1j} x_j & 0 \\
y_2 \sum_{j=1}^{3} M_{1j} x_j & -3 & -\sum_{j=1}^{3} M_{1j} x_j & 0 \\
-1 & 0 & \sum_{j=1}^{3} M_{1j} x_j & -1
\end{pmatrix}.
\] (3.10)
According to above theorem, the MFPS between the 4-dimensional fractional-order hyperchaotic Lü system (3.6) and the 3-dimensional fractional-order Arneodo chaotic system (3.7) can be achieved. For example, choose \( M(x) = \begin{pmatrix}
1+x_2 & 0 & 0 \\
0 & 1+x_3 & 0 \\
0 & 0 & 0.5+x_1
\end{pmatrix} \). The corresponding numerical result is shown in Figure 6, in which the initial conditions are \( x(0) = (2, 2, 2)^T \), and \( y(0) = (11, 10, 11, 2)^T \), respectively.
4. Conclusions
In this paper, based on the stability theory of the fractional-order system and the tracking control, a modified function projective synchronization scheme for different dimension fractional-order chaotic systems is addressed. The derived method in the present paper shows that the modified function projective synchronization between drive system and response system with different dimensions can be achieved. The modified function projective synchronization between 3-dimensional fractional-order Lorenz system and 4-dimensional
Fractional-order hyperchaotic system, and the modified function projective synchronization between the 4-dimensional fractional-order hyperchaotic \(\dot{L}\) system, and the 3-dimensional fractional-order Arneodo chaotic system, are chosen to illustrate the proposed method. Numerical experiments shows that the present method works very well, which can be used for other chaotic systems.
Acknowledgments
The authors are very grateful to the anonymous reviewers for their valuable comments and suggestions, which have greatly improved the presentation of this paper. This work is supported by the Foundation of Science and Technology project of Chongqing Education Commission under Grant KJ110525, and by the National Natural Science Foundation of China (61104150).
References
[1] I. Podlubny, *Fractional Differential Equations*, vol. 198, Academic Press, San Diego, Calif, USA, 1999.
[2] R. Hilfer, *Applications of Fractional Calculus in Physics*, World Scientific, River Edge, NJ, USA, 2001.
[3] I. Grigorenko and E. Grigorenko, “Chaos in the fractional Lorenz system,” *Physical Review Letters*, vol. 91, no. 3, pp. 34101–34104, 2003.
[4] X. J. Wu, J. Li, and G. R. Chen, “Chaos in the fractional order unified system and its synchronization,” *Journal of the Franklin Institute*, vol. 345, no. 4, pp. 392–401, 2008.
[5] T. T. Hartley, C. F. Lorenzo, and H. K. Qammer, “Chaos in a fractional order Chua’s system,” *IEEE Transactions on Circuits and Systems I*, vol. 42, no. 8, pp. 485–490, 1995.
[6] Z. M. Ge and C. Y. Ou, “Chaos in a fractional order modified Duffing system,” *Chaos, Solitons & Fractals*, vol. 34, no. 2, pp. 262–291, 2007.
Abstract and Applied Analysis
[7] Y. G. Yu and H. X. Li, “The synchronization of fractional-order Rössler hyperchaotic systems,” *Physica A*, vol. 387, no. 5-6, pp. 1393–1403, 2008.
[8] A. Freihat and S. Momani, “Adaptation of differential transform method for the numeric-analytic solution of fractional-order Rössler chaotic and hyperchaotic systems,” *Abstract and Applied Analysis*, Article ID 934219, 13 pages, 2012.
[9] N. X. Quyen, V. V. Yem, and T. M. Hoang, “A chaotic pulse-time modulation method for digital communication,” *Abstract and Applied Analysis*, vol. 2012, Article ID 835304, 15 pages, 2012.
[10] D. Y. Chen, W. L. Zhao, X. Y. Ma, and R. F. Zhang, “Control and synchronization of chaos in RCL-Shunted Josephson Junction with noise disturbance using only one controller term,” *Abstract and Applied Analysis*, vol. 2012, Article ID 378457, 14 pages, 2012.
[11] Y. Y. Hou, “Controlling chaos in permanent magnet synchronous motor control system via fuzzy guaranteed cost controller,” *Abstract and Applied Analysis*, vol. 2012, Article ID 650863, 10 pages, 2012.
[12] R. Mainieri and J. Rehacek, “Projective synchronization in the three-dimensional chaotic systems,” *Physical Review Letters*, vol. 82, no. 15, pp. 3042–3045, 1999.
[13] Y. Yu and H.-X. Li, “Adaptive generalized function projective synchronization of uncertain chaotic systems,” *Nonlinear Analysis*, vol. 11, no. 4, pp. 2456–2464, 2010.
[14] Y. Chen, H. An, and Z. Li, “The function cascade synchronization approach with uncertain parameters or not for hyperchaotic systems,” *Applied Mathematics and Computation*, vol. 197, no. 1, pp. 96–110, 2008.
[15] H. An and Y. Chen, “The function cascade synchronization method and applications,” *Communications in Nonlinear Science and Numerical Simulation*, vol. 13, no. 10, pp. 2246–2255, 2008.
[16] S. Wang, Y. G. Yu, and M. Diao, “Hybrid projective synchronization of chaotic fractional order systems with different dimension,” *Physica A*, vol. 389, no. 21, pp. 4981–4988, 2010.
[17] L. Pan, W. Zhou, L. Zhou, and K. Sun, “Chaos synchronization between two different fractional-order hyperchaotic systems,” vol. 16, no. 6, pp. 2628–2640, 2011.
[18] X. Y. Wang and J. M. Song, “Synchronization of the fractional order hyperchaos Lorenz systems with activation feedback control,” *Communications in Nonlinear Science and Numerical Simulation*, vol. 14, no. 8, pp. 3351–3357, 2009.
[19] R. Zhang and S. Yang, “Adaptive synchronization of fractional-order chaotic systems via a single driving variable,” *Nonlinear Dynamics*, vol. 66, no. 4, pp. 831–837, 2011.
[20] R. X. Zhang and S. P. Yang, “Stabilization of fractional-order chaotic system via a single state adaptive-feedback controller,” *Nonlinear Dynamic*, vol. 68, no. 1-2, pp. 45–51, 2012.
[21] D. Cafagna and G. Grassi, “Observer-based projective synchronization of fractional systems via a scalar signal: application to hyperchaotic Rössler systems,” *Nonlinear Dynamic*, vol. 68, no. 1-2, pp. 117–128, 2012.
[22] M. S. Tavazoei and M. Haeri, “Chaos control via a simple fractional-order controller,” *Physics Letters A*, vol. 372, no. 6, pp. 798–807, 2008.
[23] E. Ahmed, A. M. A. El-Sayed, and H. A. A. El-Saka, “Equilibrium points, stability and numerical solutions of fractional-order predator-prey and rabies models,” *Journal of Mathematical Analysis and Applications*, vol. 325, no. 1, pp. 542–553, 2007.
[24] Z. M. Odibat, “Adaptive feedback control and synchronization of non-identical chaotic fractional order systems,” *Nonlinear Dynamics*, vol. 60, no. 4, pp. 479–487, 2010.
[25] Z. Odibat, “A note on phase synchronization in coupled chaotic fractional order systems,” *Nonlinear Analysis*, vol. 13, no. 2, pp. 779–789, 2012.
[26] J. Lü, X. Yu, G. Chen, and D. Cheng, “Characterizing the synchronizability of small-world dynamical networks,” *IEEE Transactions on Circuits and Systems I*, vol. 51, no. 4, pp. 787–796, 2004.
[27] J. Zhou, J.-a. Lu, and J. Lü, “Adaptive synchronization of an uncertain complex dynamical network,” *IEEE Transactions on Automatic Control*, vol. 51, no. 4, pp. 652–656, 2006.
[28] J. Lu and G. Chen, “A time-varying complex dynamical network model and its controlled synchronization criteria,” *IEEE Transactions on Automatic Control*, vol. 50, no. 6, pp. 841–846, 2005.
[29] J. Lü and G. Chen, “Generating multiscroll chaotic attractors: theories, methods and applications,” *International Journal of Bifurcation and Chaos in Applied Sciences and Engineering*, vol. 16, no. 4, pp. 775–858, 2006.
[30] J. H. Lu, S. M. Yu, H. Leung, and G. R. Chen, “Experimental verification of multidirectional multiscroll chaotic attractors,” *IEEE Transactions on Circuits and Systems I*, vol. 53, no. 1, pp. 149–165, 2006.
[31] J. Lu, F. Han, X. Yu, and G. Chen, “Generating 3-D multi-scroll chaotic attractors: a hysteresis series switching method,” Automatica, vol. 40, no. 10, pp. 1677–1687, 2004.
[32] J. Lü and G. Chen, “A new chaotic attractor coined,” International Journal of Bifurcation and Chaos in Applied Sciences and Engineering, vol. 12, no. 3, pp. 659–661, 2002. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2083,
1
],
[
2083,
5089,
2
],
[
5089,
7794,
3
],
[
7794,
9590,
4
],
[
9590,
11867,
5
],
[
11867,
12698,
6
],
[
12698,
14530,
7
],
[
14530,
15052,
8
],
[
15052,
16516,
9
],
[
16516,
18169,
10
],
[
18169,
22940,
11
],
[
22940,
23295,
12
]
]
} |
8c57bdbad6641a96bc8dd0338d5e8680d8add723 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 23,
"total-fallback-pages": 0,
"total-input-tokens": 112102,
"total-output-tokens": 27435,
"fieldofstudy": [
"Physics"
]
} | Nucleon Structure Functions from the NJL-Model Chiral Soliton
I. Takyi, H. Weigel
Institute for Theoretical Physics, Physics Department, Stellenbosch University, Matieland 7602, South Africa
We present numerical simulations for unpolarized and polarized structure functions in a chiral soliton model. The soliton is constructed self-consistently from quark fields from which the structure functions are extracted. Central to the project is the implementation of regularizing the Dirac sea (or vacuum) contribution to structure functions from first principles. We discuss in detail how sum rules are realized at the level of the quark wave-functions in momentum space. The comparison with experimental data is convincing for the polarized structure functions but exhibits some discrepancies in the unpolarized case. The vacuum contribution to the polarized structure functions is particularly small.
I. INTRODUCTION
Perhaps the most convincing evidence for the quark substructure of baryons emerges from Deep Inelastic Scattering (DIS). The conjunction of perturbative Quantum Chromo Dynamics (QCD) and the parton model successfully explains the wealth of DIS data collected over the past decades [1–3]. However, these are not fully first principle calculations as the hadron wave-functions cannot (yet) be directly computed in QCD. Rather, in the spirit of the parton model, quark distributions are parameterized and subjected to perturbative QCD analysis [4]. On the other hand there are many phenomenological models based on various aspects of QCD that (attempt to) describe (static) properties of hadrons with particular focus on the nucleon. Examples are the non-relativistic quark model [5], the MIT bag model [6] relativistic quark-diquark models [7], chiral soliton models [8]; just to name a few. In principle any of these approaches should also be capable to predict nucleon structure functions. This is particular challenging for chiral soliton models that are formulated as bosonized action functionals hiding the quark substructure of hadrons. In this respect the Nambu-Jona-Lasino (NJL) or chiral quark soliton model [9, 10] is special: based on a quark self-interaction the bosonization process can be traced step by step.
In the past nucleon structure functions have indeed been computed from the chiral quark soliton model and mainly two approaches were followed. Within the valence quark only approximation [11, 12] the observation that though the Dirac sea (or vacuum) is essential to form the soliton, the vacuum contribution to nucleon properties is only moderate thereby justifying neglecting its contribution to the structure functions. Though this approximation has empirical support, it is formally incomplete and does not follow from a systematic expansion scheme. In parallel, studies on the quark distributions were performed [13–19] that identified the model quark degrees of freedom with those using the operator product expansion in the analysis of DIS. These studies included vacuum contributions. These are plagued by ultra-violet divergences requiring an a posteriori implementation of regularization by a single Pauli-Villars subtraction imposed onto the distributions. Unfortunately, this is not stringent since there are terms in the action that do not undergo regularization, e.g. to maintain the axial anomaly as measured by the decay of the neutral pion into two photons. Generally, the model has quadratic divergences (most notably the gap equation) and a single subtraction may or may not be sufficient to remove all divergences. In the present project we therefore include the vacuum contributions to the nucleon structure functions (i) without identifying the NJL model quarks field with those of QCD, and (ii) by implementing the mandatory regularization already at the level of the defining action functional. The first issue is addressed by noting that the model emulates the chiral symmetry of QCD and thus produces the same symmetry currents, in particular the electromagnetic one. To address the second issue we recall that DIS is described by the hadronic tensor $W_{\mu\nu}$ which is the Fourier transform of the nucleon matrix element of the commutator of two electromagnetic current operators. Though there is no direct implementation of this commutator when bosonizing the NJL model, we take advantage of its relation to the Compton tensor $T_{\mu\nu}$ which itself is computed from the time-ordered product of these currents. Time-ordered products are straightforwardly included in a path integral formalism within which bosonization is conducted. Regularizing this path integral by multiple Pauli-Villars subtractions proves most appropriate because it allows to trace the quark that carries the large momentum in the Bjorken limit. This formalism was developed already some time ago [20] but its numerical simulation has been long outstanding. It will be central to the study presented here.
This paper is organized as follows: In Section [11] we introduce the NJL model with emphasis on describing the regularization procedure in Minkowski space. The formulation in Minkowski space is advantageous to identify the absorptive part of $T_{\mu\nu}$ which leads to $W_{\mu\nu}$. In Section [11] we review the formalism from Ref. [20] of how to obtain the hadronic tensor in this model and in particular the role of the Bjorken limit. The NJL soliton description is explained...
We discuss the formalism to obtain the structure functions via the hadronic tensor from quark spinors that self-consistently interact with the soliton in Section IV. Subsequently (Section VI) we describe the way in which this formalism builds up the sum rules. Numerical results are presented in Section VII. This analysis also includes the perturbative evolution to the scale at which experimental data are taken. As stressed above, the model structure functions are identified from symmetry currents, not by equating QCD degrees of freedom. However, the evolution makes this unavoidable for the lack of any sensible alternative. We briefly conclude and summarize in Section VIII.
Finally we leave technical details to four Appendixes.
Some preliminary results extracted from this paper have been put forward in Ref. [21]
II. THE MODEL
We formulate the regularized action of the bosonized Nambu-Jona-Lasinio (NJL) model in Minkowski space as the sum of three pieces
\[ A_{\text{NJL}} = A_R + A_I + \frac{1}{4G} \int d^4x \, \text{tr} \left[ S^2 + P^2 + 2m_0 S \right] \]
\[ A_R = -\frac{N_C}{2} \sum_{i=0}^{2} c_i \text{Tr} \log \left[ -DD_5 + \Lambda_i^2 - i\epsilon \right] , \]
\[ A_I = -\frac{N_C}{2} \text{Tr} \log \left[ -D(D_5)^{-1} - i\epsilon \right] . \]
The subscripts \( R \) and \( I \) for real and imaginary refer to the respective properties after Wick-rotation to Euclidean space. In Minkowski space this corresponds to the use of two distinct Dirac operators
\[ D = i\partial \sigma - (S + i\gamma_5 P) \gamma_0 + \partial_0 \gamma_5 =: D^{(\sigma)} + \partial_0 + \partial_5 \gamma_5 \]
\[ D_5 = -i\partial \sigma - (S + i\gamma_5 P) \gamma_0 - \partial_0 - \partial_5 \gamma_5 =: D^{(\sigma)}_5 + \partial_0 - \partial_5 \gamma_5 . \]
Here, as in the local part of Eq. (1), \( S \) and \( P \) are scalar and pseudoscalar fields that represent physical particles. Furthermore \( v_\mu \) and \( a_\mu \) are source fields. When expanding the action with respect to these sources the linear terms couple to the vector and axial vector currents. Since \( A_I \) is (conditionally) finite in the ultra-violet, only \( A_R \) undergoes regularization. Its dynamical content is quadratically divergent\(^1\) and we thus require two subtractions. In the Pauli-Villars scheme they are implemented as
\[ c_0 = 1, \quad \Lambda_0 = 0, \quad \sum_{i=0}^{2} c_i = 0 \quad \text{and} \quad \sum_{i=0}^{2} c_i \Lambda_i^2 = 0 . \]
In practice we will reduce the number of regularization parameters by assuming \( \Lambda_1 \to \Lambda_2 = \Lambda \) which translates into the general prescription \( \sum_i c_i f(\Lambda_i^2) = f(0) - f(\Lambda^2) + \Lambda^2 f'(\Lambda^2) \). Nevertheless we always write the regularization as in Eq. (1).
The variation of the action with respect to the scalar field \( S \) yields the gap equation
\[ \frac{1}{2G} (m - m_0) = -4iN_Cm \sum_{i=0}^{2} c_i \int \frac{d^4k}{(2\pi)^4} \left[ -k^2 + m^2 + \Lambda_i^2 - i\epsilon \right]^{-1} \]
that determines the vacuum expectation value of the scalar field \( \langle S \rangle = m \). Replacing \( S \) by \( \langle S \rangle \) in \( D \) shows that \( m \) is the fermion mass and is thus called the \textit{constituent quark mass}. The chiral field \( U \) defines the non-linear representation for the isovector pion field \( \pi \)
\[ S + iP = mU = m \exp \left[ \frac{g}{m} \pi \cdot \tau \right] . \]
\(^1\) The cosmological constant, computed from setting all fields to their vacuum expectation values, is quartically divergent and must be subtracted when contributing.
We obtain the pion propagator by expanding $A$ to quadratic order and extract the pion mass from the pole
$$m_\pi^2 = \frac{1}{2G} \frac{m_0}{m} \frac{1}{2N_C \Pi(m_\pi^2)}$$ \hspace{1cm} (6)
and requiring a unit residuum at the pole fixes the quark-pion coupling
$$\frac{1}{g^2} = 4N_C \frac{d}{dq^2} \left[ q^2 \Pi(q^2) \right] \bigg|_{q^2=m_\pi^2}.$$ \hspace{1cm} (7)
The above polarization function is
$$\Pi(q^2) = \int_0^1 dx \Pi(q^2, x) \quad \text{with} \quad \Pi(q^2, x) = -i \sum_{i=0}^{2} c_i \frac{d^4 k}{(2\pi)^4} \left[ -k^2 - x(1-x)q^2 + m^2 + \Lambda_i^2 - i\epsilon \right]^2.$$ \hspace{1cm} (8)
Finally we get the pion decay constant $f_\pi$ from the coupling to the axial current. To this end we expand $A$ to linear order in both $\pi$ and $a_\mu$. The result is $f_\pi = 4N_C mg \Pi(m_\pi^2)$. Using the empirical data $m_\pi = 138\text{MeV}$ and $f_\pi = 93\text{MeV}$ fixes two of the three ($m_0$, $G$ and $\Lambda$) model constants. It is customary to use the constituent quark mass $m$ as the single tunable parameter.
III. STRUCTURE FUNCTIONS IN THE NJL MODEL
The hadronic tensor of DIS is the matrix element of the commutator of electromagnetic current operators
$$W_{\mu\nu}(q; H) = \frac{1}{4\pi} \int d^4x e^{iq\cdot x} \left< H \left[ J_\mu(\xi), J_\nu^\dagger(0) \right] \right| H \right>,$$ \hspace{1cm} (9)
where $q$ is the momentum of the virtual photon and $H$ refers to either the pion or the nucleon target. This tensor is decomposed into Lorentz structures whose coefficients are form factors that turn into the structure functions in the so-called Bjorken limit. Labeling the target momentum by $p$ this limit is defined as
$$Q^2 = -q^2 \to \infty \quad \text{with} \quad x = \frac{Q^2}{2p \cdot q} \text{ fixed}.$$ \hspace{1cm} (10)
Often $x$ is referred to as the Bjorken variable.
By the optical theorem, $W_{\mu\nu}$ is proportional to the absorptive part of the Compton amplitude
$$W_{\mu\nu}(q; H) = \frac{1}{2\pi} \text{Im} \, T_{\mu\nu}(q; H).$$ \hspace{1cm} (11)
The latter is the matrix element of a time-ordered product of the currents
$$T_{\mu\nu}(q; H) = i \int d^4\xi e^{iq\cdot \xi} \left< H \left[ T \left\{ J_\mu(\xi), J_\nu^\dagger(0) \right\} \right] H \right>.$$ \hspace{1cm} (12)
With this relation we implement regularization from first principles because the time-ordered product is immediately obtained from the action by functional differentiation
$$T \left\{ J_\mu(\xi), J_\nu^\dagger(0) \right\} = \frac{\delta^2}{\delta \bar{v}_\mu(\xi) \delta \bar{v}_\nu^\dagger(0)} A_{NJL}(v) \bigg|_{v_\mu=0},$$ \hspace{1cm} (13)
where $v_\mu$ is the photon field introduced by minimal coupling. Within the NJL model the photon couples to the quarks inside the hadron. As discussed comprehensively in Ref. [20] the evaluation of $T_{\mu\nu}$ becomes feasible in the Bjorken
\footnote{The hadronic tensor can be written as the matrix element of the commutator for the lowest energy hadron in a given baryon number sector.}
limit. Then the quark propagator that carries the large photon momentum can be identified and thus be taken to be that of a free massless fermion. Thus the functional derivative from Eq. (13) when applied to the real part simplifies to differentiating
\[ A^{(2,\nu)}_{\Lambda_R} = -i \frac{N_C}{4} \sum_{i=0}^{2} c_i \text{Tr} \left\{ \left( -D^{(\nu)} D^{(\nu)}_5 + \Lambda^2_5 \right)^{-1} \left[ Q^2 \phi (\bar{q})^{-1} \gamma_5 \right] \right\}. \]
At this point it is important to explain the crucial role of the subscript `5' attached to the second term in square brackets of Eq. (14). For this second term we have to recall that the (inverse) derivative operator in \( \phi (\bar{q})^{-1} \) is actually associated with the expansion of \( D_5 \). When comparing this \( \gamma_5 \)-odd operator to the ordinary Dirac operator in Eq. (2) one observes immediately that \( D_5 \) has a relative sign between the derivative operator \( i\partial \) and the axial source \( \gamma_5 \). Therefore the axial–vector component of \( (\phi (\bar{q})^{-1} \gamma_5) \) requires a relative sign. With \( S_{\mu\nu\sigma} = g_{\mu\rho}g_{\nu\sigma} + g_{\rho\nu}g_{\mu\sigma} - g_{\mu\sigma}g_{\rho\nu} \), that is
\[ \gamma_\mu \gamma_\rho \gamma_\nu \psi = S_{\mu\nu\sigma} \gamma^\sigma - i\epsilon_{\mu\nu\sigma} \gamma^\sigma \gamma^5 \quad \text{while} \quad (\gamma_\mu \gamma_\rho \gamma_\nu)_5 = S_{\mu\nu\sigma} \gamma^\sigma + i\epsilon_{\mu\nu\sigma} \gamma^\sigma \gamma^5. \]
In Ref. [20] this modification was formally shown to be consistent with the affected sum rules. In Section VI we see on the level of the momentum space quark wave-functions that the structure functions computed on the basis of Eq. (15) indeed fulfill the sum rules.
Similarly, in the Bjorken limit, the imaginary part becomes
\[ A^{(2,\nu)}_{\Lambda_1} = -i \frac{N_C}{4} \text{Tr} \left\{ \left( -D^{(\pi)} D^{(\pi)}_5 + \Lambda^2_5 \right)^{-1} \left[ Q^2 \phi (\bar{q})^{-1} \gamma_5 \right] \right\}. \]
These expressions are still quite formal and we will use them to obtain nucleon structure functions in Section V. We emphasize that these expressions are directly deduced from the regularized action in Eq. (1) and that no further assumption about the regularization has been made. In Ref. [20] it has been shown that applying this formalism to the pion relates its structure function to the spectral function from Eq. (5) as \( F(x) = \frac{5}{2} (4N_C g^2) \frac{d^2}{dp^2} [p^2 \Pi(p^2, x)] \bigg|_{p^2 = m^2_\pi} \), a result that was previously obtained from the analysis of the handbag diagram in Refs. [22, 23].
IV. NJL MODEL SOLITON
We construct the soliton from static meson configurations by introducing a Dirac Hamiltonian \( h \) via
\[ iD^{(\pi)} = \beta (i\partial_t - h) \quad \text{and} \quad iD^{(\pi)}_5 = (i\partial_t - h) \beta. \]
Its diagonalization
\[ h \Psi_\alpha = \epsilon_\alpha \Psi_\alpha, \]
yields eigen-spinors \( \Psi_\alpha = \sum_{\beta} V_{\alpha\beta} \Psi^{(0)}_\beta \) (\( \Psi^{(0)}_\beta \) are free Dirac spinors in a spherical basis, see Appendix A) and energy eigenvalues \( \epsilon_\alpha \). The hedgehog configuration minimizes the action in the unit baryon number sector and introduces the chiral angle \( \Theta(r) \) via
\[ h = \alpha \cdot p + \beta m U_5(r) \quad \text{where} \quad U_5(r) = \exp [i\hat{r} \cdot \tau_5 \gamma_5 \Theta(r)]. \]
With the boundary conditions \( \Theta(0) = -\pi \) and \( \lim_{r \to \infty} \Theta(r) = 0 \) the diagonalization, Eq. (18) yields a distinct, strongly bound level, \( \Psi_\nu \), referred to as the valence quark level in. Its (explicit) occupation ensures unit baryon number. The functional trace in \( A_R \) is computed as an integral over the time interval \( T \) and a discrete sum over the basis levels defined by Eq. (18). In the limit \( T \to \infty \) the vacuum contribution to the static energy is then extracted from \( A_R \to \) \( -TE_{\text{vac}} \). Collecting pieces, we obtain the total energy functional as [3, 10]
\[ E_{\text{tot}}[\Theta] = \frac{N_C}{2} \left[ 1 + \text{sign}(\epsilon_\nu) \right] \epsilon_\nu - \frac{N_C}{2} \sum_{i=0}^{2} c_i \sum_{\alpha} \left\{ \sqrt{\epsilon_\alpha^2 + \Lambda^2_5} - \sqrt{\epsilon^{(0)}_\alpha^2 + \Lambda^2_5} \right\} + m^2 \int d^3 r \left[ 1 - \cos(\Theta) \right]. \]
Here we have also subtracted the vacuum energy associated with the non-dynamical meson field configuration \( \Theta = 0 \) (denoted by the superscript) that is often called the cosmological constant contribution. This subtraction will also
play an important role for the unpolarized isosinglet structure function as it enters via the momentum sum rule. The soliton profile is then obtained as the profile function $\Theta(r)$ that minimizes the total energy $E_{\text{tot}}$ self-consistently subject to the above mentioned boundary conditions.
This soliton represents an object which has unit baryon number but neither good quantum numbers for spin and flavor (isospin). Such quantum numbers are generated by canonically quantizing the time-dependent collective coordinates $A(t)$ which parameterize the spin-flavor orientation of the soliton. For a rigidly rotating soliton the Dirac operator becomes, after transforming to the flavor rotating frame \[ W \]
\[ W(q) = \frac{1}{2} \Omega \cdot \tau, \]
which, according to the quantization rules, are replaced by the spin operator \[ \Omega \rightarrow \frac{1}{\alpha^2} J. \]
The constant of proportionality is the moment of inertia
\[ \alpha^2 = \frac{N_C}{4} \left[ 1 + \text{sign}(\epsilon_v) \right] \sum_{\beta \neq \alpha} |\langle \tau_3 | \beta \rangle|^2 \right] + \frac{N_C}{8} \sum_{\alpha \neq \beta} \sum_{i=1}^2 c_i \left( \frac{|\langle \alpha | \tau_3 | \beta \rangle|^2}{\epsilon^2_{\alpha} - \epsilon^2_{\beta}} \right) \left( \frac{\epsilon^2_{\alpha} + \epsilon_\alpha \epsilon_\beta + 2\Lambda^2}{\sqrt{\epsilon^2_{\alpha} + \Lambda^2}} - \frac{\epsilon^2_{\beta} + \epsilon_\alpha \epsilon_\beta + 2\Lambda^2}{\sqrt{\epsilon^2_{\beta} + \Lambda^2}} \right) \right), \]
which is of the order $N_C$. With Eq. (23) the expansion in $\Omega$ is thus equivalent to the one in $1/N_C$. The nucleon wave-function becomes a (Wigner D) function of the collective coordinates. A useful relation in computing matrix elements of nucleon states is \[ \langle N | \frac{1}{2} \text{tr} (A^\dagger \tau_i A \tau_j) | N \rangle = -\frac{4}{3} \langle N | I_i J_j | N \rangle. \]
V. STRUCTURE FUNCTIONS FROM SOLITON
We first repeat the relation between the structure functions and the hadronic tensor of the nucleon. Its symmetric combination $W_{\mu\nu}^S(q) = \frac{1}{2} (W_{\mu\nu}(q) + W_{\nu\mu}(q))$, is parameterized by two form factors\[ W_{\mu\nu}^S(q) = MW_1(\nu, Q^2) \left( -g_{\mu\nu} + \frac{q_\mu q_\nu}{q^2} \right) + \frac{W_2(\nu, Q^2)}{M} \left( p_\mu - \frac{p \cdot q}{q^2} q_\mu \right) \left( p_\nu - \frac{p \cdot q}{q^2} q_\nu \right). \]
In the Bjorken limit, Eq. (10), these form factors turn into the unpolarized structure functions that we extract by appropriate projections:
\[ F_1(x) = -\frac{1}{2} g^{\mu\nu} W_{\mu\nu}^S(q) \quad \text{and} \quad F_2(x) = -x g^{\mu\nu} W_{\mu\nu}^S(q). \]
It must be noted that the Callan Gross relation, i.e. \[ F_1(x) = 2xF_2(x), \]
is satisfied in this case by construction. Similarly, the anti-symmetric part is also parameterized by two form factors
\[ W_{\mu\nu}^A(q) = i\epsilon_{\mu\nu\lambda\sigma} q^\lambda \left\{ MG_1(\nu, Q^2)s^\nu + \frac{G_2(\nu, Q^2)}{M} ((p \cdot q)s^\nu - (q \cdot s)p^\nu) \right\}. \]
In the Bjorken limit these form factors yield the structure functions
\[ g_1(x) = M^2 \nu G_1(\nu, Q^2) \quad \text{and} \quad g_2(x) = \nu G_2(\nu, Q^2). \]
---
3 Factors of the nucleon mass, $M$, occur for dimensional reasons.
The longitudinal, $g_1(x)$, and transverse, $g_T(x) = g_1(x) + g_2(x)$, structure functions are extracted from the hadronic tensor using the projection operators
$$g_1(x) = \frac{1}{2M} \int \frac{d^4 \rho \rho_{\mu} \rho_{\nu}}{q \cdot s} W_{\mu\nu}^A(q) \quad \text{and} \quad q \parallel s$$
(30)
$$g_T(x) = -\frac{1}{2M} \int \frac{d^4 \rho \rho_{\mu} \rho_{\nu}}{q \cdot s} W_{\mu\nu}^A(q) \quad \text{and} \quad q \perp s$$
(31)
To obtain the hadronic tensor for the nucleon in the soliton model, the functional traces in Eqs. (14) and (16) are computed using the basis defined by the self-consistent soliton, Eq. (13). This calculation has been detailed in Ref. [20] that we adopt directly. We start with the leading order in $M/\Lambda$ to the vacuum (or sea) contribution to $W_{\mu\nu}$
$$W_{\mu\nu}^{(s)}(q) = -\frac{1}{8} \frac{MNc}{2\pi} \int \frac{dw}{2\pi} \sum_\alpha \int d^3 \xi \int \frac{d\lambda}{2\pi} e^{i M x \lambda}$$
(32)
$$\times \langle N, s | \left\{ \left[ \bar{\Psi}_\alpha(\xi) Q_A^2 \gamma_\mu \gamma_\nu \Psi_\alpha(\xi + \lambda \hat{e}_3) e^{-i \lambda \omega} - \bar{\Psi}_\alpha(\xi) Q_A^2 \gamma_\mu \gamma_\nu \Psi_\alpha(\xi - \lambda \hat{e}_3) e^{i \lambda \omega} \right] f^+_\alpha(\omega) \right\} p$$
$$+ \left[ \bar{\Psi}_\alpha(\xi) Q_A^2 (\gamma_\mu \gamma_\nu)_5 \Psi_\alpha(\xi - \lambda \hat{e}_3) e^{-i \lambda \omega} - \bar{\Psi}_\alpha(\xi) Q_A^2 (\gamma_\mu \gamma_\nu)_5 \Psi_\alpha(\xi + \lambda \hat{e}_3) e^{i \lambda \omega} \right] f^-_\alpha(\omega) \right\} \langle N, s \rangle.$$
Here $n^\mu = (1, 0, 0, 1)^\mu$ is the light-cone vector of the photon momentum while $Q_A = A^\dagger Q A$ denotes the flavor rotated quark charge matrix from which we compute nucleon matrix elements as in Eq. (23). Furthermore
$$f^\pm_\alpha = \sum_{i=0}^2 c_i \omega \pm \epsilon_\alpha - \epsilon_\alpha^2 - \Lambda_i^2 + i\epsilon \pm \frac{\omega}{\epsilon_\alpha^2 + \Lambda_i^2 + i\epsilon},$$
are Pauli-Villars regularized spectral functions. The subscript ‘$p$’ indicates their pole contributions that we will explain below.
For the vacuum contribution to the isosinglet unpolarized structure function we then obtain
$$[F_1^{l=0}(x)]_s = \frac{1}{72} \frac{M N c}{2\pi} \int \frac{dw}{2\pi} \sum_\alpha \int d^3 \xi \int \frac{d\lambda}{2\pi} e^{i M x \lambda} \left( \sum_{i=0}^2 c_i \frac{\omega + \epsilon_\alpha}{\omega^2 - \epsilon_\alpha^2 - \Lambda_i^2 + i\epsilon} \right)_p$$
$$\times \int d^3 \xi \left\{ \bar{\Psi}_\alpha(\xi)(1 - \alpha_3) \Psi_\alpha(\xi + \lambda \hat{e}_3) e^{-i \lambda \omega} - \bar{\Psi}_\alpha(\xi)(1 - \alpha_3) \Psi_\alpha(\xi - \lambda \hat{e}_3) e^{i \lambda \omega} \right\}.$$
(34)
Here the pole contributions is
$$\left( \sum_{i=0}^2 c_i \frac{1}{\omega^2 - \epsilon_\alpha^2 - \Lambda_i^2 + i\epsilon} \right)_p = \sum_{i=0}^2 c_i \frac{-i\pi}{\omega_\alpha} [\delta(\omega + \omega_\alpha) + \delta(\omega - \omega_\alpha)],$$
(35)
where (eventually we take the single cut-off limit as described after Eq. (3) and thus omit the label $i$ on $\omega_\alpha$)
$$\omega_\alpha = \sqrt{\epsilon_\alpha^2 + \Lambda_i^2}. \quad (36)$$
We recall that the single cut-off approach requires a derivative with respect to that cut-off. Of course, this also affects the implicit dependence of $\omega_\alpha$ on that cut-off.
We introduce the Fourier transform of the quark wave-function as
$$\tilde{\Psi}_\alpha(p) = \int \frac{d^3 r}{4\pi} \Psi_\alpha(r) e^{ip \cdot r}.$$
(37)
Implementing a full Fourier transform differs from the approaches of Refs. [15, 16] who used the expansion from diagonalizing the Dirac Hamiltonian, Eq. (13). This resulted in discontinuities of the numerically computed quark distributions and required a smoothing procedure.
Performing the frequency $(\omega)$ and lambda$^4$ ‘$\lambda$’ integrals gives the vacuum contribution of the flavor-singlet unpolarized structure function in the nucleon rest frame (RF)
$$[F_1^{l=0}(x)]_s = \frac{5 MN}{144} \sum_\alpha \sum_{i=0}^2 c_i \int_{|M x^\pm|}^\infty p dp \int d\Omega_p \left\{ \pm \bar{\Psi}_\alpha(p) \tilde{\Psi}_\alpha(p) - \frac{\epsilon_\alpha M x^\pm}{\omega_\alpha p} \bar{\Psi}_\alpha(p) \tilde{\Psi}_\alpha(p) \right\}.$$
(38)
$^4$ Technically it is advisable to average the photon direction rather than fixing it along the $z$-axis [15].
where
\[ Mx_\pm^\alpha = Mx \pm \omega_\alpha. \]
(39)
In the above \([F_1^{I=0}(x)]_s^\pm \) refers to the positive (negative) frequency components that are typically referred to as quark and antiquark distributions. In our calculation they arise from the two poles of the \(\delta\)-function in Eq. (35). Then the vacuum part of the isoscalar, unpolarized structure function becomes
\[ [F_1^{I=0}(x)]_s = [F_1^{I=0}(x)]_s^- + [F_1^{I=0}(x)]_s^+. \]
(40)
As a matter of fact, this is still not the full result. Substituting free spinors (not interaction with the soliton) produces a non-zero result. This non-zero result must also be subtracted. In the discussion of the sum rules below we will see that this is nothing but the \(c_\alpha^{(0)}\) type subtraction performed in Eq. (20) and may be considered a cosmological constant type contribution.
The valence quark contribution is obtained by replacing the quark levels in (38) by the cranked valence level \(\psi_v(\tau, t) = e^{-i\epsilon_v t} A(t) \left\{ \Psi_v(r) + \frac{1}{2} \sum_{\alpha \neq v} \Psi_\alpha(r) \langle \frac{\alpha}{\epsilon_v - \epsilon_\alpha} \right\} = e^{-i\epsilon_v t} A(t) \psi_v(r), \)
(41)
In the above \(\psi_v(r)\) is the spatial part of the valence quark wave-function with the rotational correction included and \(\epsilon_v\) is the energy eigenvalue of the valence quark level. Noting that the valence quark wave-function has positive parity and the pole contribution \(f_\pm|_{\text{pole}} = -4i\pi\delta(\omega \mp \epsilon_v)\) gives the valence quark contribution
\[ [F_1^{I=0}(x)]_v^\pm = \frac{-5MN_e}{144} \left[ 1 + \text{sign}(\epsilon_v) \right] \int_0^\infty p dp \int d\Omega_p \left\{ \pm \bar{\Psi}_v(p)\tilde{\Psi}_v(p) - \frac{Mx_\pm^\alpha}{p} \tilde{\Phi}_v(p)\tilde{\Phi}_v(p) \right\}, \]
(42)
where \(Mx_\pm^\alpha = Mx \pm \epsilon_v.\) Again, we have separated positive and negative frequency components.
The quark spinors \(\Psi_v(r)\) separate into radial and angular pieces \([20].\) At the end, the structure functions, as in Eq. (38) are computed as integrals over the (Bessel-)Fourier transforms of the radial functions in the quark spinors. In Appendix [A] we list examples explicitly.
In quite an analogous manner, the isovector components of the polarized structure functions are extracted from the anti-symmetric combination \(W_{\mu\nu}(q)\). Explicitly we find the vacuum contribution to the longitudinal polarized structure function to be
\[ \left[ g_1^{I=1}(x) \right]_s^\pm = \frac{2MN_e}{72} I_3 \sum_{\alpha} \sum_{i=0}^{2} C_i \left\{ \mp \int_{|Mx_\pm^\alpha|}^\infty dp Mx_\alpha \int d\Omega_p \bar{\tilde{\Phi}}_\alpha(p)\hat{\Phi}_\alpha(p) \right\}, \]
(43)
and the isovector transverse structure function as
\[ \left[ g_1^{I=1}(x) \right]_s^\pm = \frac{2MN_e}{72} I_3 \sum_{\alpha} \sum_{i=0}^{2} C_i \left\{ \frac{\epsilon_\alpha}{\omega_\alpha} \int_{|Mx_\pm^\alpha|}^\infty dp p^2 \left[ A_\pm \int d\Omega_p \bar{\tilde{\Phi}}_\alpha(p)\hat{\Phi}_\alpha(p) + B_\pm \int d\Omega_p \bar{\tilde{\Phi}}_\alpha(p)\hat{\Phi}_\alpha(p) \right] \right\}. \]
(44)
In these formulas we have introduced the abbreviations, see also Eq. (39)
\[ A_\pm = \frac{1}{2p} \left( 1 - \frac{(Mx_\pm^\alpha)^2}{p^2} \right), \quad B_\pm = \frac{1}{2p} \left( 3\frac{(Mx_\pm^\alpha)^2}{p^2} - 1 \right). \]
(45)
The total (vacuum contribution to the) polarized structure functions is the sum of the positive (+) and negative (−) frequency components. Again, some details in terms of the Fourier transformed radial functions are given in Appendix [C]. For completeness we also list the formulas for the valence quark contribution [12]. The contribution to
the longitudinal polarized structure function is obtained as
\[
[g_{l1}^T(x)]^\pm = \frac{M N_c}{144} [1 + \text{sign}(\epsilon_v)] I_3 \left\{ \mp \int_0^\infty dp \int_{|Mx^\pm|} \frac{d\Omega_p}{4\pi} \tilde{\Psi}_v(p) \hat{p} \cdot \tau \tilde{\Psi}_v(p) + B_{\pm} \int_{|Mx^\pm|} \frac{d\Omega_p}{4\pi} \tilde{\Psi}_v(p) \hat{p} \cdot \tau \tilde{\Psi}_v(p) \right\},
\]
and that for the transverse polarized structure function as
\[
[g_{l1}^T(x)]^\pm = \frac{M N_c}{144} [1 + \text{sign}(\epsilon_v)] I_3 \left\{ \mp \int_0^\infty dp \int_{|Mx^\pm|} \frac{d\Omega_p}{4\pi} \tilde{\Psi}_v(p) \tau \cdot \sigma \tilde{\Psi}_v(p) \right\},
\]
The isovector unpolarized and isoscalar polarized structure functions are subleading in the $1/N_C$ counting. They are also more complicated to compute as they are quartic in the spinors and involve double sums over the basis states defined Eq. [18]. We refrain from presenting those lengthy expressions here and rather refer the interested reader to the Appendixes of Ref. [27].
VI. FORMAL DISCUSSION OF SUM RULES
In this section we discuss how the sum rules for the unpolarized and polarized structure functions work out when written explicitly in terms of the momentum space eigenspinors $\tilde{\Psi}_\alpha$. In this context it is important to note that we compute the structure functions for a localized configuration in its rest frame. Then the Bjorken variable has support on the half axis from zero to infinity. Lorentz covariance is regained by transforming to the infinite momentum frame, cf. Section VII.B
Sum rules relate integrated structure functions to static observables. In soliton models the latter are directly expressed in terms of the eigenspinors, Eq. [18] in coordinates space. Typically the sum rules can then be expressed as level-by-level identities. The only exception is the momentum (or energy) sum rule. For it to be obeyed it is compulsory that the soliton is an extremum of the energy functional, Eq. [20].
A. Momentum sum rule
For the momentum sum rule we require that $\frac{36}{M} \int dxx F_1(x)$ produces the quark contribution to the classical energy, i.e. all but the last integral in Eq. [20]. First we consider the scalar terms, $\pm \tilde{\Psi}_\alpha(p) \tilde{\Psi}_\alpha(p)$, from the vacuum contribution, Eq. [35].
\[
[M_{G0}] = \frac{M N_c}{4} \sum_{i=0}^2 c_i \int_0^\infty dx x \left\{ \left\langle \alpha | \alpha \right\rangle_{|Mx^+|} - \left\langle \alpha | \alpha \right\rangle_{|Mx^-|} \right\}
\]
\[
= \frac{M N_c}{4} \sum_{i=0}^2 c_i \int_0^\infty dy \left( y - \frac{\omega_\alpha}{M} \right) \left\langle \alpha | \alpha \right\rangle_{My} - \int_0^\infty dy \left( y + \frac{\omega_\alpha}{M} \right) \left\langle \alpha | \alpha \right\rangle_{MMy},
\]
\[
= -\frac{N_c}{2} \sum_{i=0}^2 c_i \int_0^\infty dy \left\langle \alpha | \alpha \right\rangle_{My} = \frac{N_c}{2} \sum_{i=0}^2 c_i \omega_\alpha \int_0^\infty dy \left\langle \alpha | \alpha \right\rangle_{MMy} \int_0^\infty dp \frac{d\Omega_p}{4\pi} \tilde{\Psi}_\alpha(p) \tilde{\Psi}_\alpha(p),
\]
\[
= -\frac{M N_c}{2} \sum_{i=0}^2 c_i \omega_\alpha \int_0^\infty dy \left[ \frac{d\Omega_p}{4\pi} \tilde{\Psi}_\alpha(p) \tilde{\Psi}_\alpha(p) \right] = -\frac{N_c}{2M} \sum_{i=0}^2 c_i \sqrt{c_\alpha^2 + \lambda_i^2},
\]
\[\tag{48}\]
We adopt the notation $\left\langle \alpha | \alpha \right\rangle_a = \int_0^\infty dp \frac{d\Omega_p}{4\pi} \tilde{\Psi}_\alpha(p) \tilde{\Psi}_\alpha(p)$.
\[\tag{49}\]
which is $1/M$ times the vacuum contribution to the classical energy. This contribution also includes subtraction of the trivial vacuum energy, when there is no soliton. Hence the isoscalar unpolarized structure function necessitates the analog subtraction, as indicated earlier. For the valence contribution the momentum sum rule gives
$$\left[\mathcal{M}_{G}^{0}\right]_{v} = \frac{36}{5} \int_{0}^{\infty} dx \varepsilon \left[F_{1}^{0}(x)\right]_{v} = \frac{N_{C}}{2M} \left[1 + \text{sign}(\varepsilon)\right] \varepsilon_{v}.$$ \hspace{1cm} (49)
Similarly, integrating the term with the operator $\hat{p} \cdot \alpha$ gives
$$\left[\mathcal{M}_{G}^{1}\right]_{s} = -\frac{M^{2}N_{C}}{2} \sum_{i=0}^{2} c_{i} \int_{0}^{\infty} dy \varepsilon \langle \alpha | \frac{\hat{p}}{\sqrt{\alpha}} | \alpha \rangle_{\text{My}} = -\frac{N_{C}}{6M} \sum_{i=0}^{2} c_{i} \frac{\varepsilon_{\alpha}}{\omega_{\alpha}} \langle \alpha | \alpha \rangle \varepsilon_{\alpha} \cdot p \alpha \rangle.$$ \hspace{1cm} (50)
Next we use the Dirac Hamiltonian, Eq. (13) to write
$$\varepsilon_{\alpha} \cdot p = \{ r \cdot p, h \} - m \beta \{ r \cdot p, U_{5}(r) \}$$
so that $\langle \alpha | \varepsilon_{\alpha} \cdot p | \alpha \rangle = \varepsilon_{\alpha} \langle \alpha | r \cdot p, U_{5}(r) | \alpha \rangle$. Since $r \cdot p$ is the dilatation operator this matrix element measures the change of the single particle energy when scaling the soliton extension by an infinitesimal amount. Furthermore
$$\frac{\omega_{\alpha}}{\omega_{\alpha}} = \frac{\partial}{\partial \omega_{\alpha}} \sqrt{\varepsilon_{\alpha}^{2} + \lambda^{2}}$$
so that $\left[\mathcal{M}_{G}^{1}\right]_{s}$ is the change of the vacuum energy when the soliton extension deviates slightly from its stationary point. Similarly, the valence quark adds $\left[\mathcal{M}_{G}^{2}\right]_{v} = i \frac{N_{C}}{12M} \left[1 + \text{sign}(\varepsilon)\right] \langle \psi | r \cdot p, U_{5}(r) | \psi \rangle$ to the sum rule. Then $\left[\mathcal{M}_{G}^{1}\right]_{s} + \left[\mathcal{M}_{G}^{2}\right]_{v}$ is the coefficient of $(\lambda - 1)$ term in the expansion
$$E[U(\lambda \psi)] = E_{0} + (\lambda - 1)E_{1} + \cdots (\lambda - 1)^{l}E_{l} + \cdots$$ \hspace{1cm} (51)
of the classical energy. Since $U(x)$ is a stationary point, $E_{1} = 0$ thus verifying the momentum sum rule \[13\]. Obviously the momentum sum rule is not saturated level by level; rather it requires summing all contributions to this isoscalar unpolarized structure function. Hence this sum rule will be a very sensitive test of the numerical simulation.
### B. Bjorken sum rule
Here we verify the Bjorken sum rule in our model, which relates the isovector polarized structure function $g_{1}^{T}$ to the axial charge \[23\]. First, we show that the term in Eq. (43) with the operator $\hat{p} \cdot \tau \gamma_{5}$ integrates to zero
$$\sum_{i=0}^{2} c_{i} \int_{0}^{\infty} dx \left\{ Mx^{-1} \langle \alpha | \frac{\hat{p}}{\sqrt{\alpha}} \tau \gamma_{5} | \alpha \rangle_{[Mx^{-1}]} - Mx^{+} \langle \alpha | \frac{\hat{p}}{\sqrt{\alpha}} \tau \gamma_{5} | \alpha \rangle_{[Mx^{+}]} \right\},$$
$$= \sum_{i=0}^{2} c_{i} \left\{ \int_{-\infty}^{0} \frac{dy}{\alpha} M y \langle \alpha | \frac{\hat{p}}{\sqrt{\alpha}} \tau \gamma_{5} | \alpha \rangle_{[M y]} - \int_{0}^{\infty} \frac{dy}{\alpha} M y \langle \alpha | \frac{\hat{p}}{\sqrt{\alpha}} \tau \gamma_{5} | \alpha \rangle_{[M y]} \right\},$$
$$= \sum_{i=0}^{2} c_{i} \int_{-\infty}^{\infty} \frac{dy}{\alpha} M y \langle \alpha | \frac{\hat{p}}{\sqrt{\alpha}} \tau \gamma_{5} | \alpha \rangle_{[M y]} = 0.$$ \hspace{1cm} (52)
There are two contributions without $\gamma_{5}$. The first one contributes
$$\frac{MN_{C}I_{3}}{72} \sum_{i=0}^{2} c_{i} \frac{\varepsilon_{\alpha}}{\omega_{\alpha}} \int_{0}^{\infty} dx \left[ \langle \alpha | \frac{pA_{+} \tau \cdot \sigma | \alpha \rangle_{[M x^{+}]} + \langle \alpha | \frac{pA_{-} \tau \cdot \sigma | \alpha \rangle_{[M x^{-}]} \right]$$
$$= \frac{MN_{C}I_{3}}{144} \sum_{i=0}^{2} c_{i} \frac{\varepsilon_{\alpha}}{\omega_{\alpha}} \left[ \int_{-\infty}^{\infty} \frac{dy}{\alpha} M y \langle \alpha | \tau \cdot \sigma \left( 1 - \frac{(M y)^{2}}{p^{2}} \right) | \alpha \rangle_{[M y]} + \int_{-\infty}^{\infty} \frac{dy}{\alpha} M y \langle \alpha | \tau \cdot \sigma \left( 1 - \frac{(M y)^{2}}{p^{2}} \right) | \alpha \rangle_{[M y]} \right]$$
$$= \frac{N_{C}I_{3}}{108} \sum_{i=0}^{2} c_{i} \frac{\varepsilon_{\alpha}}{\omega_{\alpha}} \langle \alpha | \tau \cdot \sigma | \alpha \rangle.$$ \hspace{1cm} (53)
The term with $\hat{p} \cdot \tau \cdot \sigma$ disappears because
$$
\int_0^\infty dy \int_{M_y}^\infty p^2 dp \int d\Omega_p \left( \frac{1}{p} - \frac{3}{y^2} \right) \bar{\Psi}_\alpha(p) \hat{p} \cdot \tau \cdot \sigma \Psi_\alpha(p) = \int_0^\infty dy \int_{M_y}^\infty dp \int d\Omega_p \frac{\partial}{\partial y} \left( py - \frac{M^2 y^3}{p} \right) \bar{\Psi}_\alpha(p) \hat{p} \cdot \tau \cdot \sigma \Psi_\alpha(p) = \int_0^\infty dy \left( \frac{M^2 y^3}{p} - py \right) \left[ \int d\Omega_p \bar{\Psi}_\alpha(p) \hat{p} \cdot \tau \cdot \sigma \Psi_\alpha(p) \right]_{p=M_y} = 0. \tag{54}
$$
Hence the Bjorken sum rule for the vacuum contribution of the longitudinal polarized structure function becomes
$$
\int dx [g_1^p(x) - g_1^n(x)]_v = \frac{N_C}{108} \sum_{i=0}^{2} c_i \frac{\epsilon_\alpha}{\omega_0} \langle \alpha | \tau \cdot \sigma | \alpha \rangle = \frac{1}{6} \left[ \frac{N_C}{18} \sum_{\alpha} \sum_{i=0}^{2} c_i \frac{\epsilon_\alpha}{\sqrt{\epsilon_\alpha^2 + \Lambda^2_5}} \langle \alpha | \gamma_3 \gamma_5 \tau_3 | \alpha \rangle \right]. \tag{55}
$$
The object in square brackets is the vacuum contribution to the axial charge \[29\]. Similar calculations from the valence contribution give
$$
\int dx [g_1^p(x) - g_1^n(x)]_v = -\frac{N_C}{54} \left[ 1 + \text{sign}(\epsilon_v) \right] \langle v | \tau \cdot \sigma | v \rangle = -\frac{1}{6} \left[ -\frac{N_C}{9} \left[ 1 + \text{sign}(\epsilon_v) \right] \langle v | \gamma_3 \gamma_5 \tau_3 | v \rangle \right] \tag{56}
$$
with the object in square brackets being the valence quark contribution to the axial charge. This indeed verifies the Bjorken sum rule for the total axial charge.
In an analog, yet much more tedious, calculation the sum rules for the subleading contributions in the $1/N_C$ expansion are also verified via level by level identities. Details may be found in Ref. \[27\]. We would like to mention however, that the Adler sum rule \[29\], which concerns a structure function from neutrino interactions and thus the exchange of a gauge boson (not considered here), measures the isospin of the nucleon. In that case the sum rule is not level by level; rather summing this integrated structure function over all levels reproduces the moment of inertia, Eq. \[24\] \[30\].
\[\text{VII. NUMERICAL RESULTS}\]
In this Section we present our numerical results for the structure functions. These results are obtained in a number of subsequent steps. First we construct the coordinate space eigenspinors of the self-consistent chiral soliton as described in Section \[IV\] for the parameters listed at the end of Section \[II\]. In the second step we evaluate the Fourier transform according to Eq. \[37\]. Details of this transformation are provided in Appendix \[A\]. Essentially the spinors in momentum space are combinations of spherical harmonic functions of the unit momentum vector and momentum space radial functions that are Bessel transforms of the radial functions in the coordinate space spinors from Section \[IV\]. In momentum space the spherical harmonic functions combine to the conserved grand spin just as do those in coordinate space. Hence we formally obtain matrix elements of operators as, for example $\alpha \cdot \hat{p}$, in the very same way as the matrix elements of $\alpha \cdot \hat{r}$ in coordinate space. In the third step the momentum space radial functions are numerically integrated to produce the structure functions in the nucleon rest frame. In the next step they are transformed to the infinite momentum frame \[31\] and subsequently the standard (perturbative QCD) evolution to the scale of the experimental data is performed to allow for a sensible comparison. We note that this evolution brings into the game a new model parameter, the scale at which the evolution commences. We take a single scale for all structure functions.
We test the outcome of our numerical simulations via the sum rules, that is, we compare the integrated functions with associated local quantity obtained from the coordinate space spinors. To gain acceptable agreement a very fine (equi-distant) grid for the radial variable in momentum space is required. Typically we take several thousand points on an interval between zero and ten times the physical cut-off, $\Lambda$. Needless to say that this consumes a large amount of CPU time and obtaining (in particular the subleading $1/N_C$ contributions to) the structure functions takes days or weeks on an ordinary desktop PC. Still, there are minor numerical inaccuracies as reflected by small oscillations of the structure functions around a central value at larger $x$, cf. figures below. Working in momentum space, rather than using the expansion coefficients $V_{\alpha\beta}$ introduced after Eq. \[18\] has, however, the advantage that no smearing \[14\] procedure is required.
FIG. 1: Isoscalar unpolarized structure function in the nucleon rest frame for a constituent quark mass of 400MeV. Dashed and dotted lines refer to the positive and negative frequency contributions, respectively.
FIG. 2: Isovector unpolarized structure function in the nucleon rest frame for a constituent quark mass of 400MeV. Note the logarithmic scale for the Bjorken variable. Dashed and dotted lines refer to the positive and negative frequency contributions, respectively.
A. Rest frame results
In Figures 1 and 2 we show typical results for the isoscalar and isovector components of the unpolarized structure functions respectively. In this case they have been obtained using the constituent quark mass of \( m = 400\text{MeV} \). We separately show the contributions of the discrete valence level, those of the vacuum contributions and their sums (labeled as total). For the vacuum contribution we find the unexpected result that it dominates the valence counterpart. Mainly this originates from the (additional) subtraction of the non-soliton piece mentioned after Eq. (40). Without that subtraction we would not get a finite result, of course. Neither would the momentum sum rule be fulfilled. However, this piece does not connect to the soliton rest frame and it is not clear at all whether or not transformation of the Bjorken variable should be performed before taking the difference between the soliton and non-soliton isoscalar unpolarized structure functions. Therefore we do not attach much relevance to this large vacuum contribution. As expected, the vacuum contribution is sub-dominant for the isovector unpolarized structure function.
In Figure 3 we present the unpolarized structure function that enters the Gottfried sum rule, i.e. \( F_{p}^{2}(x) - F_{n}^{2}(x) = 2x[F_{1}^{p}(x) - F_{1}^{n}(x)] \) as the Callan-Gross relation holds in the soliton rest frame. The vacuum contribution turns slightly negative at large \( x \) which persists when adding the dominating valence piece to form the total contribution of this structure function. In Table II we compare our results for the Gottfried sum rule, \( S_{G} = \int_{0}^{\infty} \frac{dx}{x} (F_{p}^{2} - F_{n}^{2}) \), for various constituent quark masses to the experimental data from the NMC Collaboration [32]. Under this integral, the vacuum part is even less significant as its positive and negative parts compensate. In total, the agreement for the Gottfried sum rule is surprisingly good since usually chiral soliton models reproduce empirical data with 30% accuracy [8].
In Figures 4 and 5 we show the isoscalar and isovector contributions to the longitudinal polarized structure functions \( g_{I=0,1} \), respectively. In both pictures we display the valence and vacuum contributions as well as their sums. Also the positive and negative frequency components of the valence and vacuum parts are shown. Obviously these structure functions are indeed dominated by their valence contributions and we thus a posteriori verify the valence
FIG. 3: Unpolarized structure function \( F_2^p(x) - F_2^n(x) \) in the nucleon rest frame for a constituent quark mass of 400MeV.
TABLE I: The Gottfried sum rule for various values of \( m \). The subscripts 'v' and 's' denote the valence and vacuum contributions, respectively. The third column contains their sums.
| \( m \) [MeV] | \( |S_G|_v \) | \( |S_G|_s \) | \( S_G \) | empirical value |
|-------------|-------------|-------------|-------|----------------|
| 400 | 0.214 | 0.000156 | 0.214 | |
| 450 | 0.225 | 0.000248 | 0.225 | 0.235 ± 0.026 [32] |
| 500 | 0.236 | 0.000356 | 0.237 | |
*only approximation* adopted in Ref. [12]. The only exception is the isoscalar structure function \( g_2 \) for which the valence contribution is small by its own due to large cancellations in \( g_2 = g_T - g_1 \). We also recognize some minor oscillations in the vacuum contributions at larger \( x \). These occur as remnants of numerical inaccuracies. We also remark that the present valence quark results do not exactly match those from Ref. [12] in which a soliton profile from the proper-time regularization scheme was employed.
We have computed the axial vector and singlet charges on one hand side via the respective sum rules, i.e. by integrating the structure functions \( g_{I=1,0}^I(x) \) and on the other hand via the coordinate space matrix elements of \( \gamma_3 \gamma_1 \gamma_5 \) and \( \gamma_3 \gamma_5 \) as e.g. in Eqs. (55) and (60). The comparison in table II serves as test for the numerical accuracy which works perfectly in the vector case while some minor discrepancies are observed for the axial singlet charge. This is understood as the latter is actually quartic in the quark wave-functions (two of which are Fourier transformed) and it is also a double sum over those wave-functions. So even tiny numerical errors are amplified.
As is typical for chiral soliton models, the axial vector charge falls short off the measured datum by about 30-40% [8]. It has been argued that this could be remedied by \( 1/N_C \) corrections arising from a particular handling of the collective coordinate quantization [33, 34]. However, these corrections do not emerge in the current approach and also lead to inconsistencies with PCAC [35]. On the other hand, the predicted axial singlet charge, which is linked to the
FIG. 4: Isoscalar longitudinal polarized structure functions in the nucleon rest frame for a constituent quark mass of 400MeV. For the valence and vacuum contributions we separately display the positive (dashed) and negative (dotted) frequency contributions. The full lines are their sums. Note the small vertical scale for the vacuum contribution.
proton spin problem\cite{36}, is well within the errors of the empirical value.
\textbf{B. Projection and evolution}
The soliton picture for baryons employs a localized field configuration which generally breaks translational invariance. This causes the structure functions not to vanish when $x > 1$ as would be demanded kinematically. This effect is obvious in the above figures. We note that it is not limited to soliton models but is observed, \textit{i.e.} in the bag model as well\cite{39}. Using light cone coordinates in the bag model in one space dimension a mapping of the structure functions from the localized field configuration was constructed that annihilated the structure functions for
TABLE II: The axial-vector and -singlet charges for various values of the constituent quark mass $m$. Subscripts are as in table 4. The data in parenthesis give the numerical results as obtained from the coordinate space representation.
| $m$ [MeV] | $g_A$ | $g_{A\perp}$ | empirical value | $m$ [MeV] | $g_A$ | $g_{A\perp}$ | empirical value |
|-----------|-------|--------------|-----------------|-----------|-------|--------------|-----------------|
| 400 | 0.734 | 0.065 | 0.799 (0.800) | 400 | 0.344 | 0.0016 | 0.345 (0.350) |
| 450 | 0.715 | 0.051 | 0.766 (0.765) | 450 | 0.327 | 0.0021 | 0.329 (0.332) |
| 500 | 0.704 | 0.029 | 0.733 (0.733) | 500 | 0.316 | 0.0028 | 0.318 (0.323) |
$x > 1$. Guided by that construction a Lorentz boost was applied transforming the rest frame structure functions to the infinite momentum frame.
$$f_{IMF}(x) = \frac{\Theta(1-x)}{1-x} f_{RF}(-\ln(1-x)),$$
(57)
where $f_{RF}$ refers to any of the structure functions computed in Section VII A. In what follows we will omit the label IMF for the boosted structure functions.
Even though we have adopted the high energy Bjorken limit in our kinematical analysis of the Compton tensor, it must be emphasized that the NJL model is (at best) an approximation to QCD at the low mass scale, $\mu^2$ which is thus a hidden parameter in the approach. To compare with experimental data that are taken at higher energy scales, $Q_{exp}^2$ we adopt Altarelli-Parisi (or DGLAP) equations (41–43), to evolve the model structure functions accordingly. To be precise, we integrate
$$f(x, t + \delta t) = f(x, t) + \delta t \frac{df(x, t)}{dt}.$$
(58)
with $t = \ln \left(\frac{Q^2}{\Lambda_{QCD}^2}\right)$ from $Q^2 = \mu^2$ to $Q^2 = Q_{exp}^2$. The structure functions from Eq. (57) are the initial values and we tune $\mu^2$ for best fit at $Q_{exp}^2$.
Since the isoscalar structure functions are associated with gluon type quantum numbers they mix under the evolution. We take this into account under the assumption that the gluon distributions vanish at $\mu^2$. We consider the leading order of the perturbative expansion sufficient to estimate the quality of our results. Then the evolution equations have the following structure
$$\frac{df^{(I=1)}(x, t)}{dt} = \frac{g_{QCD}(t)}{2\pi} C_R(F) \int_x^1 \frac{dy}{y} P_{qq}(y) f^{(I=1)} \left(\frac{x}{y}, t\right),$$
(59)
$$\frac{df^{(I=0)}(x, t)}{dt} = \frac{g_{QCD}(t)}{2\pi} C_R(F) \int_x^1 \frac{dy}{y} \left\{ P_{qq}(y) f^{(I=0)} \left(\frac{x}{y}, t\right) + P_{qq}(y) g \left(\frac{x}{y}, t\right) \right\},$$
(60)
$$\frac{dg(x, t)}{dt} = \frac{g_{QCD}(t)}{2\pi} C_R(F) \int_x^1 \frac{dy}{y} \left\{ P_{gg}(y) g \left(\frac{x}{y}, t\right) + P_{gg}(y) f^{(I=0)} \left(\frac{x}{y}, t\right) \right\},$$
(61)
where $C_R(F) = \frac{(N_f^2 - 1)}{2N_f}$ is the color factor for $N_f$ flavors. Furthermore $g_{QCD}(t) = \frac{4\pi}{\beta_0 t}$ with $\beta_0 = \frac{11}{3} N_C - \frac{2}{3} N_f$ is the leading order perturbative running coupling constant. Explicit expressions for the splitting functions $P_{qq}, \ldots, P_{gg}$, taken from Ref. (44) are listed in Appendix D for completeness. From the evolved isoscalar and isovector components we finally obtain the proton and neutron structure functions as sum and difference
$$f^{(p,n)}(x, Q^2) = \frac{1}{2} \left[ f^{I=1}(x, Q^2) \pm f^{I=0}(x, Q^2) \right].$$
(62)
We note that applying the perturbative QCD scheme to the model structure functions requires the identification of model and QCD degrees of freedom even though there is no definite reason for doing so other than the lack of any sensible alternative.
The second polarized structure function $g_2$ contains subleading, twist three, elements that undergo a different evolution procedure that is also described in Appendix D.
FIG. 8: Model prediction for the longitudinal polarized proton structure functions. Left panel: \( g_1^p(x) \); right panel: \( g_1^{3\text{He}}(x) \). These functions are “DGLAP” evolved from \( \mu^2 = 0.4\text{GeV}^2 \) to \( Q^2 = 3\text{GeV}^2 \) after projected to the infinite momentum frame “IMF”. Data are from Refs. [46, 47] for the proton and from Ref. [48] for helium. In the latter case \( E \) refers to the electron energy.
FIG. 9: Model prediction for the polarized proton structure functions \( g_{2W}^{WW(p)}(x) \) (left panel) and \( g_2^p(x) \) (right panel) the are the twist-2 and -3 pieces of \( g_2 \). These functions are “DGLAP” evolved from \( \mu^2 = 0.4\text{GeV}^2 \) to \( Q^2 = 5\text{GeV}^2 \) after projected to the infinite momentum frame “IMF”.
C. Comparison with experiment
As in previous calculations within the valence only approximation [11, 12, 45] we take \( \mu^2 = 0.4\text{GeV}^2 \) as initial value in the evolution differential equations. Smaller values contradict the perturbative nature of the evolution procedure as \( \Lambda^2_{\text{QCD}} \) becomes sizable in view of \( \Lambda^2_{\text{QCD}} = 0.2\text{GeV}^2 \). On the other hand, significantly larger \( \mu^2 \) values worsen the agreement with experimental data.
In the left panel of Figure 8 we show the numerical result for the polarized proton structure function \( g_1 \) obtained from the evolution equation at \( Q^2 = 3\text{GeV}^2 \). We compare our results to experimental results from the E143 Collaboration [46, 47]. At small \( x \) the model results are somewhat larger than the data, but definitely the gross features are predominantly reproduced.
For the neutron data are available in terms of the helium structure function [48]
\[
g_1^{3\text{He}}(x) \approx P_n g_1^n(x) + P_p g_1^p(x) - 0.014 \left[ g_1^p(x) - 4g_1^n(x) \right],
\]
with \( P_n \approx 0.86 \) and \( P_p \approx -0.028 \). From the right panel in Figure 8 we see that our model results reproduce the main features of the data: small positive values at large \( x \) turning negative at moderate \( x \), though the minimum is more pronounced by the model.
Next we discuss the results for the structure function \( g_2(x) \). As discussed in Appendix D the twist-2 and -3 pieces must be disentangled within the evolution whose result is shown in Figure 9. The effect of evolution is small for the twist-2 component but essential for the twist-3 counterpart. When the end point of evolution is reached, the two components are combined to \( g_2^p(x, Q^2) \). We display the model prediction in Figure 10 and see that the data are well produced. This shows that the higher twist contributions cannot be neglected. This suggests that the higher twist contributions cannot be neglected.
\[\text{In Ref. [48] direct neutron data are only given as the ratio } g_1^n(x)/F_1(x).\]
FIG. 10: Model prediction for the polarized proton structure functions $g_2^p(x)$. This function is “DGLAP” evolved from $\mu^2 = 0.4\, GeV^2$ to $Q^2 = 5\, GeV^2$ after projected to the infinite momentum frame “IMF”. Data is from Ref [40].
Recently data were reported for the neutron twist-3 moment
$$d_n^3(Q^2) = 3 \int_0^1 dx \, x^2 \, \bar{g}_n^2(x, Q^2)$$
et two different transferred momenta: $d_n^3(3.21 GeV^2) = -0.00421 \pm 0.00114$ and $d_n^3(4.32 GeV^2) = -0.00035 \pm 0.00104$, where we added the listed errors in quadrature. For $m = 400\, MeV$ the model calculation yields $-0.00426$ and $-0.00035$, respectively. While the lower $Q^2$ result nicely matches the observed value, the higher one differs by about three standard deviations. This discrepancy as a function of $Q^2$ indicates that the large $N_C$ approximation to evolve $g_2$ (cf. Appendix D) requires improvement.
Finally, in Figure 11 we display the unpolarized structure function that enters the Gottfried sum rule, i.e. $F_2^p(x) - F_2^n(x)$ using the evolution equation. Though the negative contribution from the Dirac vacuum (cf. Figure 2) around $x = 1$ is tiny in the rest frame, it gets amplified when transforming to the infinite momentum frame by the factor $1/(1-x)$ in Eq. (57) thereby worsening the agreement with the experimental data collected by the NMC [32].
VIII. CONCLUSION
We have presented the numerical simulation of nucleon structure functions within the NJL chiral soliton model. Central to this analysis has been the consistent implementation of the regularized vacuum contributions that arise from all quark spinors being distorted by the chiral soliton. Generally speaking, vacuum contributions should not be omitted in any quark model as no expansion scheme suppresses them. This is even more the case for the NJL model because the vacuum part significantly contributes to forming the soliton.
In our analysis we have only identified the symmetry currents of the model with those from QCD, not the quark distributions that are bilinear operators which are bilocal in the quark fields. Also, it is important to enforce the regularization on the action functional so that the regularization prescription for a given structure function is an unambiguous result. This increases the predictive power compared to previous similar studies that ”advocated” an ad hoc regularization of quark distributions [14, 16]. A first principle regularization is particularly important when
---
7 Schwinger’s proper time regularization scheme is popular in the context of the NJL chiral soliton [24]. Ref. [16] explicitly states that its
the sum rule for the structure function does not relate to coordinate space matrix elements of the quark fields. As an example we have seen that the isovector unpolarized structure functions are not subject to regularization (the explicit, lengthy formulas can be obtained from Ref. [27]). The prediction for the corresponding, so-called Gottfried, sum rule decently matches the empirical value. This is a very favorable case for the *valence only approximation* as the vacuum contribution essentially integrates to zero due to an unexpected negative contribution at large \( x \). For the isoscalar part we recognized that the subtraction of the zero-soliton vacuum contribution has a sizable effect at small \( x \). The emergence of this contribution is somewhat surprising as it suggests that the zero-soliton vacuum has structure. Yet it is required for convergence as well as fulfilling the momentum sum rule. We emphasize that we observe acceptable agreement for polarized proton structure functions between our model results and the experimental data. For the polarized structure functions our numerically expensive computation indeed showed that the vacuum contribution is sub-dominant, except maybe for the isoscalar part of \( g_2 \) where the valence part is tiny by itself. Nevertheless these results support the *valence only approximation* to a large extent.
In both, the unpolarized and polarized cases, the comparison with experiment required two additional operations on the model structure functions. As the soliton is a localized field configuration, translational invariance is lost and the rest frame structure functions must be Lorentz transformed to the infinite momentum frame. In turn the results from that transformation are subject to the perturbative QCD evolution scheme. This brings into the game the hidden parameter at which to commence the evolution. We took that to be the same as in the *valence only approximation*.
Even though we have separated positive and negative frequency contributions to the structure functions we stop short of identifying them as (anti-)quark distributions that parameterize semi-hard processes [50], like e.g. Drell-Yan [51]. The reason being that we avoid to identify model and QCD quark degrees of freedom at the model scale. There are many other nucleon matrix elements of bilocal, bilinear quark operators for which experimental results or lattice data are available. Examples are chiral odd distributions [52, 53] or quasi-distributions [54–56]. It is challenging to see whether quark distributions, or at least some of them, can also be formulated and computed with a first principle regularization scheme in the NJL chiral soliton model.
Acknowledgments
This project is is supported in part by the National Research Foundation of South Africa (NRF) by grant 109497. I. T. gratefully acknowledges a bursary from the *Stellenbosch Institute for Advanced Studies* (STIAS).
Appendix A: Soliton matrix elements
The Dirac Hamiltonian \( h \) of the hedgehog field configuration [19] commutes with the grand spin operator
\[
G = J + \frac{\tau}{2} = L + \frac{\sigma}{2} + \frac{\tau}{2},
\]
which is the operator sum of the total spin \( J \) and the isospin \( \tau/2 \). The total spin is the operator sum of the orbital angular momentum \( L \) and the intrinsic spin \( \sigma/2 \). Since \( h \) preserves \( G \), the eigenfunctions of the Dirac Hamiltonian are also eigenfunctions of \( G \). The quantum numbers of \( G \) are \( G^2 = G(G+1) \) and \( G_3 = M \). The respective eigenfunctions are tensor spherical harmonics associated with the grand spin
\[
[Y_{LMJG}(\hat{r})]_{is} = \sum_{m,s,s_3,i_3,j_3} C_{LMi,Lmi}^{GM} C^{J,j_3}_{Lm} Y_{LM}(\hat{r}) \chi_s(s_3) \chi_i(i_3)
\]
where \( C_{LMi,Lmi}^{GM} \) and \( C^{J,j_3}_{Lm} \) are \( SU(2) \) Clebsch-Gordon coefficients that describe the coupling of \( \chi_s \) and \( \chi_i \), which are two components spinors and isospinors, respectively, and the spherical harmonic functions \( Y_{LM} \).
For a prescribed profile function \( \Theta(r) \) the numerical diagonalization of the Dirac Hamiltonian [19] produces the application to the quark distributions is not yet known.
radial functions \( g_{α}^{(G;±i)} \) and \( f_{α}^{(G;±i)} \) that feature in eight component spinors \( 26 \):
\[
\Psi_{α}^{(+)}(r) = \left( \frac{ig_{α}^{(G;+;1)}(r)}{f_{α}^{(G;+;1)}(r)} \right) Y_{GG+\frac{1}{2}GM}(\hat{r}) + \left( \frac{-ig_{α}^{(G;+;2)}(r)}{f_{α}^{(G;+;2)}(r)} \right) Y_{GG-\frac{1}{2}GM}(\hat{r}) \tag{A3}
\]
\[
\Psi_{α}^{(-)}(r) = \left( \frac{-ig_{α}^{(G;-;1)}(r)}{f_{α}^{(G;-;1)}(r)} \right) Y_{GG+\frac{1}{2}GM}(\hat{r}) + \left( \frac{ig_{α}^{(G;-;2)}(r)}{f_{α}^{(G;-;2)}(r)} \right) Y_{GG-\frac{1}{2}GM}(\hat{r}) \tag{A4}
\]
Here, the second superscript \((±)\) denotes the intrinsic parity defined by the parity eigenvalue as \((-1)^{G} \times (±1)\). The radial functions are written as linear combinations of spherical Bessel functions that build the free spinors \( \Psi_{α}^{(0)} \). The order of these Bessel functions matches the angular momentum label \( G \) (first subscript) of the multiplying \( \mathcal{Y} \). The linear combination goes over momenta discretized by pertinent boundary conditions at a radius significantly larger than the extension of the profile function \( Θ(r) \). In Ref. \( 26 \) the condition that the radial function multiplying the \( G \) with equal orbital angular momentum and grand spin indexes vanished at that large distance. In contrast, we impose that condition on the radial function of the upper component. This avoids spurious contributions to the moment of inertia \( 18 \).
Writing
\[
e^{ip\tau} = 4π \sum_{LM} (i)^{L}j_{L}(pr)Y_{LM}(\hat{r})Y_{LM}(\hat{p}) \tag{A5}
\]
we find the Fourier transform, Eq. \( 37 \) of the spinors
\[
\bar{\Psi}_{α}^{(G,+)}(p) = (i)^{G+1} \left( \frac{g_{α}^{(G;+;1)}(p)}{f_{α}^{(G;+;1)}(p)} \right) Y_{GG+\frac{1}{2}GM}(\hat{p}) + (i)^{G-1} \left( \frac{-g_{α}^{(G;+;2)}(p)}{f_{α}^{(G;+;2)}(p)} \right) Y_{GG-\frac{1}{2}GM}(\hat{p}) \tag{A6}
\]
\[
\bar{\Psi}_{α}^{(G,-)}(p) = -(i)^{G} \left( \frac{g_{α}^{(G;-;1)}(p)}{f_{α}^{(G;-;1)}(p)} \right) Y_{GG+\frac{1}{2}GM}(\hat{p}) + (i)^{G+1} \left( \frac{-g_{α}^{(G;-;2)}(p)}{f_{α}^{(G;-;2)}(p)} \right) Y_{GG-\frac{1}{2}GM}(\hat{p}) \tag{A7}
\]
The radial functions in momentum space are the Fourier-Bessel transforms
\[
\bar{φ}_{α}(p) = \int_{0}^{∞} dr r^{2} j_{Lα}(pr)φ_{α}(r) \tag{A7}
\]
where \( Lα \) is the angular momentum associated with the coordinate space radial wave-function \( φ_{α}(r) \). Note that the grand spin spherical harmonic functions in momentum space are constructed precisely as in coordinate space, just that the argument is the momentum space solid angle. Note that the intrinsic parity is also conserved quantum number.
The valence quark carries \( G = 0 \), then only the components with \( J = +1/2 \) are allowed for the eigenspinor
\[
\Psi^{0,+}(r) = \Psi_{v}(r) = \left( \frac{ig_{v}(r)Y_{0,\frac{1}{2},0,0}(\hat{r})}{f_{v}(r)Y_{1,\frac{1}{2},0,0}(\hat{r})} \right) \tag{A8}
\]
here \( g_{v}(r) = g_{α}^{(0;+;1)}(r) \) etc, are the particular eigenvalue-functions. The cranking correction associated with the first order rotation \( \hat{H}_{α} \) dwells in the channel with \( G = 1 \) and negative intrinsic parity
\[
\Psi^{(-)}(r) = \left( \frac{ig_{α}^{(1)}(r)Y_{2,\frac{1}{2},1,0}(\hat{r})}{-f_{α}^{(1)}(r)Y_{1,\frac{1}{2},1,0}(\hat{r})} \right) + \left( \frac{ig_{α}^{(2)}(r)Y_{2,\frac{1}{2},1,0}(\hat{r})}{f_{α}^{(2)}(r)Y_{1,\frac{1}{2},1,0}(\hat{r})} \right) \tag{A9}
\]
for convenience we have written \( g_{α}^{(1;+;1)}(r) \) as \( g_{v}(r) \) etc. Taking the Fourier transform of equation \( 11 \) gives
\[
\bar{Ψ}_{v}(p) = \bar{Ψ}_{v}(p) + \sum_{α} \langle H_{α} | \bar{Ψ}_{α}(p) \rangle \tag{A10}
\]
where
\[
\bar{Ψ}_{v}(p) = \left( \frac{g_{v}(p)Y_{0,\frac{1}{2},0,0}(\hat{p})}{f_{v}(p)Y_{1,\frac{1}{2},0,0}(\hat{p})} \right) \tag{A11}
\]
TABLE III: Matrix elements $\int d\Omega p \bar{\psi}_{\alpha}(p) \gamma L' J' G M (p) \hat{p} \cdot \sigma \gamma L J G M (p)$.
\begin{tabular}{cccc}
\hline
$J' = G - \frac{i}{2}$ & $J' = G + \frac{i}{2}$ \\
$\frac{L'}{G} = G - 1$ & $\frac{L'}{G} = G$ & $\frac{L'}{G} = G$ & $\frac{L'}{G} = G + 1$
\hline
0 & 0 & 0 & $L = G - 1$
-1 & 0 & 0 & 0 & $L = G$
0 & 0 & 0 & -1 & $L = G$
0 & 0 & -1 & 0 & $L = G + 1$
\hline
\end{tabular}
and
$$\bar{\Psi}_\alpha(p) = -i \left( \bar{g}_{\alpha}^{(1)}(p) \gamma_{2,1,1,M}(p) - \bar{g}_{\alpha}^{(2)}(p) \gamma_{0,1,1,M}(p) \right) \left( \bar{f}_{\alpha}^{(1)}(p) \gamma_{1,1,1,M}(p) - \bar{f}_{\alpha}^{(2)}(p) \gamma_{1,1,1,M}(p) \right).$$ \hspace{1cm} \text{(A12)}$$
The “matrix element” $\langle H_\alpha \rangle$ arises from perturbatively treating the collective rotation
$$\langle H_\alpha \rangle = \frac{1}{2} \frac{\langle \sigma \cdot \Omega | v \rangle}{\epsilon_v - \epsilon_\alpha}.$$ \hspace{1cm} \text{(A13)}$$
Appendix B: Unpolarized Structure Functions at Leading Order
The level sums (over $\alpha$) as \textit{e.g.} in Eq. (63) concern the label of the radial function, grand spin ($G$) and its projection ($M$). As we average of the direction of the virtual photon \textit{[14]}, the matrix elements are degenerate in $M$. This produces the extra factor $2G + 1$ that we make explicit.
It is then straightforward to compute the matrix elements that appear in (51):
$$\int d\Omega p \bar{\Psi}_\alpha(p) \bar{\Psi}_\alpha(p)$$ \hspace{1cm} \text{(B1)}$$
and
$$\int d\Omega p \bar{\Psi}_\alpha(p) \hat{p} \cdot \alpha \bar{\Psi}_\alpha(p) = \int d\Omega p \bar{\Psi}_\alpha(p) \hat{p} \cdot \sigma \gamma_5 \bar{\Psi}_\alpha(p).$$ \hspace{1cm} \text{(B2)}$$
The positive intrinsic parity of the matrix element of (B2) is obtained as
$$\int d\Omega p \bar{\Psi}_\alpha^{(G,+)}(p) \bar{\Psi}_\alpha^{(G,+)}(p) = (2G + 1) \left( \bar{g}_{\alpha}^{(G,+,1)}(p)^2 + \bar{f}_{\alpha}^{(G,+,1)}(p)^2 + \bar{g}_{\alpha}^{(G,+,2)}(p)^2 + \bar{f}_{\alpha}^{(G,+,2)}(p)^2 \right),$$ \hspace{1cm} \text{(B3)}$$
and for the negative intrinsic parity as
$$\int d\Omega p \bar{\Psi}_\alpha^{(G,-)}(p) \bar{\Psi}_\alpha^{(G,-)}(p) = (2G + 1) \left( \bar{g}_{\alpha}^{(G,-,1)}(p)^2 + \bar{f}_{\alpha}^{(G,-,1)}(p)^2 + \bar{g}_{\alpha}^{(G,-,2)}(p)^2 + \bar{f}_{\alpha}^{(G,-,2)}(p)^2 \right).$$ \hspace{1cm} \text{(B4)}$$
Here the overall factor $(2G + 1)$ arises from summing the grand spin projection contained in $\sum_\alpha$. From Table (III) the positive intrinsic parity of the matrix element of (B2) is obtained as
$$\int d\Omega p \bar{\Psi}_\alpha^{(G,+)}(p) \hat{p} \cdot \sigma \gamma_5 \bar{\Psi}_\alpha^{(G,+)}(p) = -2(2G + 1) \left( \bar{g}_{\alpha}^{(G,+,1)}(p) \bar{f}_{\alpha}^{(G,+,1)}(p) + \bar{g}_{\alpha}^{(G,+,2)}(p) \bar{f}_{\alpha}^{(G,+,2)}(p) \right),$$ \hspace{1cm} \text{(B5)}$$
and that for the negative intrinsic parity as
$$\int d\Omega p \bar{\Psi}_\alpha^{(G,-)}(p) \hat{p} \cdot \sigma \gamma_5 \bar{\Psi}_\alpha^{(G,-)}(p) = -2(2G + 1) \left( \bar{g}_{\alpha}^{(G,-,1)}(p) \bar{f}_{\alpha}^{(G,-,1)}(p) + \bar{g}_{\alpha}^{(G,-,2)}(p) \bar{f}_{\alpha}^{(G,-,2)}(p) \right).$$
The matrix element from the valence contribution \textit{[12]} is easily obtained, using the definition of the decomposition of the valence wave function \textit{[11]}. They are given as
$$\int d\Omega p \bar{\Psi}_\alpha(p) \bar{\Psi}_\alpha(p) = \bar{g}_v(p)^2 + \bar{f}_v(p)^2$$ \hspace{1cm} \text{(B6)}$$
and
$$\int d\Omega p \bar{\Psi}_\alpha(p) \hat{p} \cdot \sigma \gamma_5 \bar{\Psi}_\alpha(p) = -2\bar{g}_v(p)\bar{f}_v(p)$$ \hspace{1cm} \text{(B7)}$$
at leading order $1/N_C$.
As contribution is given as
Also the matrix element (C2) is computed from the matrix elements from Table V: the positive intrinsic parity
Here we list the matrix elements that appear in the vacuum contribution of the polarized structure functions,
needs to be multiplied.
TABLE V: Matrix elements $\int d\Omega_{p} \Psi_{\alpha}^{(G,+)}(p) \hat{p} \cdot \tau \Psi_{\alpha}(p)$. The overall factor $1/(2G + 1)$
| $J = G - \frac{1}{2}$ | $J = G + \frac{1}{2}$ |
|------------------------|------------------------|
| $L = G - 1$ | $L = G - 1$ |
| $L' = G$ | $L' = G$ |
| $L' = G$ | $L' = G$ |
| $J = G - \frac{1}{2}$ | $J = G + \frac{1}{2}$ |
| -1 | 0 |
| 0 | -2/2G(G + 1) |
| -2/2G(G + 1) | 0 |
| 0 | -2/2G(G + 1) |
| 0 | 1 |
| 2/2G(G + 1) | 0 |
TABLE IV: Matrix elements $\int d\Omega_{p} \Psi_{\alpha}^{(G,+)}(p) \hat{p} \cdot \tau \Psi_{\alpha}(p)$. The overall factor $1/(2G + 1)$
$\int d\Omega_{p} \Psi_{\alpha}^{(G,+)}(p) \hat{p} \cdot \tau \Psi_{\alpha}(p)$ and $\int d\Omega_{p} \Psi_{\alpha}^{(G,+)}(p) \hat{p} \cdot \sigma \Psi_{\alpha}(p)$.
The matrix element (C1) is computed from the matrix elements from Table IV the positive intrinsic parity is obtained as
\[
\int d\Omega_{p} \Psi_{\alpha}^{(G,+)}(p) \hat{p} \cdot \tau \Psi_{\alpha}(p) = 2 \left( g_{\alpha}^{(G,+;+1)}(p) \tilde{f}_{\alpha}^{(G,+;+1)}(p) - g_{\alpha}^{(G,+;+2)}(p) \tilde{f}_{\alpha}^{(G,+;+2)}(p) \right) - 4 \sqrt{G(G + 1)} \left( g_{\alpha}^{(G,+;+1)}(p) \tilde{f}_{\alpha}^{(G,+;+2)}(p) + g_{\alpha}^{(G,+;+2)}(p) \tilde{f}_{\alpha}^{(G,+;+1)}(p) \right),
\]
and that for negative intrinsic parity as
\[
\int d\Omega_{p} \Psi_{\alpha}^{(G,-)}(p) \hat{p} \cdot \tau \Psi_{\alpha}(p) = 2 \left( g_{\alpha}^{(G,-;+1)}(p) \tilde{f}_{\alpha}^{(G,-;+1)}(p) - g_{\alpha}^{(G,-;+2)}(p) \tilde{f}_{\alpha}^{(G,-;+2)}(p) \right) + 4 \sqrt{G(G + 1)} \left( g_{\alpha}^{(G,-;+1)}(p) \tilde{f}_{\alpha}^{(G,-;+2)}(p) + g_{\alpha}^{(G,-;+2)}(p) \tilde{f}_{\alpha}^{(G,-;+1)}(p) \right).
\]
Also the matrix element (C2) is computed from the matrix elements from Table V the positive intrinsic parity contribution is given as
\[
\int d\Omega_{p} \Psi_{\alpha}^{(G,+)}(p) \hat{p} \cdot \tau \Psi_{\alpha}(p) = \left( -g_{\alpha}^{(G,+;+1)}(p)^2 - \tilde{f}_{\alpha}^{(G,+;+1)}(p)^2 + g_{\alpha}^{(G,+;+2)}(p)^2 + \tilde{f}_{\alpha}^{(G,+;+2)}(p)^2 \right) + 4 \sqrt{G(G + 1)} \left( g_{\alpha}^{(G,+;+1)}(p) \tilde{f}_{\alpha}^{(G,+;+2)}(p) + \tilde{f}_{\alpha}^{(G,+;+1)}(p) \tilde{f}_{\alpha}^{(G,+;+2)}(p) \right),
\]
Appendix C: Polarized Structure Functions at Leading Order
Here we list the matrix elements that appear in the vacuum contribution of the polarized structure functions, Eqs. (33) and (34). The matrix element to be considered are
\[
\int d\Omega_{p} \Psi_{\alpha}^{(G,+)}(p) \hat{p} \cdot \tau \Psi_{\alpha}(p),
\]
\[
\int d\Omega_{p} \Psi_{\alpha}^{(G,+)}(p) \hat{p} \cdot \sigma \Psi_{\alpha}(p)
\] and
\[
\int d\Omega_{p} \Psi_{\alpha}^{(G,+)}(p) \tau \cdot \sigma \Psi_{\alpha}(p).
\]
TABLE VI: Matrix elements \( \int d\Omega_p \mathcal{Y}_{L',J'GM}(p) \tau \cdot \sigma \mathcal{Y}_{LJGM}(p) \). The overall factor \( 1/(2G+1) \)
| \( J' = G - \frac{1}{2} \) | \( J' = G + \frac{1}{2} \) |
|-----------------|-----------------|
| \( L' = G - 1 \) | \( 0 \) | \( 0 \) | \( 0 \) | \( 2G+1 \) |
| \( L' = G \) | \( 0 \) | \( 0 \) | \( 0 \) | \( L = G \) |
| \( L' = G \) | \( 0 \) | \( 0 \) | \( 2G+1 \) | \( L = G + 1 \) |
| \( L' = G + 1 \) | \( J = G - \frac{1}{2} \) |
| \( J = G + \frac{1}{2} \) |
and that for the negative intrinsic parity as
\[
\int d\Omega_p \bar{\Psi}_a^{(G,-)}(p) \bar{\tau} \cdot \sigma \Psi_a^{(G,-)}(p) = \left( -\tilde{g}_a^{(G,-;1)}(p)^2 - \tilde{f}_a^{(G,-;2)}(p)^2 + \tilde{g}_a^{(G,-;2)}(p)^2 + \tilde{f}_a^{(G,-;2)}(p)^2 \right)
- 4\sqrt{G(G+1)} \left( \tilde{g}_a^{(G,-;1)}(p)\tilde{g}_a^{(G,-;2)}(p) + \tilde{f}_a^{(G,-;1)}(p)\tilde{f}_a^{(G,-;2)}(p) \right). \tag{C7}
\]
Furthermore the matrix element \( \text{(C3)} \) is computed from the matrix elements from Table VI, the positive intrinsic parity contribution becomes
\[
\int d\Omega_p \bar{\Psi}_a^{(G,+)}(p) \bar{\tau} \cdot \sigma \Psi_a^{(G,+)}(p) = (2G+1) \left( \tilde{f}_a^{(G,+;1)}(p)^2 + \tilde{f}_a^{(G,+;2)}(p)^2 \right) - (2G+3)\tilde{g}_a^{(G,+;1)}(p)^2
+ (2G-1)\tilde{g}_a^{(G,+;2)}(p)^2 + 8\sqrt{G(G+1)}\tilde{g}_a^{(G,+;1)}(p)\tilde{g}_a^{(G,+;2)}(p), \tag{C8}
\]
and for negative intrinsic parity becomes
\[
\int d\Omega_p \bar{\Psi}_a^{(G,-)}(p) \bar{\tau} \cdot \sigma \Psi_a^{(G,-)}(p) = (2G+1) \left( \tilde{g}_a^{(G,-;1)}(p)^2 + \tilde{g}_a^{(G,-;2)}(p)^2 \right) - (2G+3)\tilde{f}_a^{(G,-;1)}(p)^2
- (2G-1)\tilde{f}_a^{(G,-;2)}(p)^2 - 8\sqrt{G(G+1)}\tilde{f}_a^{(G,-;1)}(p)\tilde{f}_a^{(G,-;2)}(p). \tag{C9}
\]
Appendix D: Splitting functions
Here we list the splitting functions used in Eqs. (59), (60) and (61). They are different for the isovector, isosinglet and gluon contributions and are given as \([42, 57]\). They determine the probability for the parton \( m \) to emit a parton \( n \) such that the momentum of the parton \( m \) is reduced by the fraction \( z \). The regularized function \( (1-z)^{-1}_+ \) is defined under the integral by \([42]\)
\[
\int_0^1 dz \frac{f(z)}{(1-z)_+} = \int_0^1 dz \frac{f(z) - f(1)}{1-z}. \tag{D2}
\]
In the above \( C_H(F) = \frac{(N_f^2 - 1)}{2N_f} \) is the color factor for \( N_f \) flavors. Also the running coupling constant in the leading order is given by \( g_{QCD}(t) = \frac{4\pi}{\beta_0 t} \) with \( \beta_0 = \frac{11}{3} N_c - \frac{2}{3} N_f \) being the coefficient of the leading term of the QCD...
β-function. Using the “+” prescription, the evolution equations for the isovector, isosinglet and gluon contributions become \[57\]
\[
\frac{df^{(I=1)}(x,t)}{dt} = \frac{2C_R(F)}{9t} \left\{ \int_x^1 \frac{dy}{y} \left( \frac{1+y^2}{1-y} \right) \left[ \frac{1}{y} f^{(I=1)} \left( \frac{x}{y}, t \right) - f^{(I=1)}(x,t) \right] + \left[ x + \frac{x^2}{2} + 2 \ln(1-x) \right] f^{(I=1)}(x,t) \right\}.
\]
\[
\frac{df^{(I=0)}(x,t)}{dt} = \frac{2C_R(F)}{9t} \left\{ \int_x^1 \frac{dy}{y} \left( \frac{1+y^2}{1-y} \right) \left[ \frac{1}{y} f^{(I=0)} \left( \frac{x}{y}, t \right) - f^{(I=0)}(x,t) \right] + \frac{3}{4} \left( y^2 + (1-y^2) \right) g(x,t) + \left[ x + \frac{x^2}{2} + 2 \ln(1-x) \right] f^{(I=0)}(x,t) \right\}.
\]
\[
\frac{dg(x,t)}{dt} = \frac{2C_R(F)}{9t} \left\{ \int_x^1 \frac{dy}{y} \left( \frac{1+y^2}{1-y} \right) f^{(I=0)} \left( \frac{x}{y}, t \right) + \frac{9}{2} \left( \frac{1}{y} - y^2 (1-y) \right) g \left( \frac{x}{y}, t \right) + \frac{9}{2} \frac{g \left( \frac{x}{y}, t \right) - g(x,t)}{1-y} \right\} + \left[ \frac{3}{2} + \frac{9}{2} \ln(1-x) \right] g(x,t) \right\}.
\]
(D3)
Now, since our NJL model calculations do not account for any gluon content in the nucleon, we assume in our numerical calculations that, at the initial boundary scale \(\mu^2\), the gluon content, \(f(x,t_0) = 0\) for both the polarized and unpolarized structure functions.
Unlike the polarized spin structure function \(g_1(x)\) of the nucleon and the unpolarized structure functions, the nucleon’s second polarized spin structure function \(g_2\) involves contributions from quark-gluon iterations and quark masses \[58-60\]. According to the standard operator product expansion analysis, these contributions come from the twist-3 local operators. It, however, also receives contribution from twist-2 local operators under the impulse approximation. Thus, the structure function \(g_2\) can be written as the sum of the twist pieces
\[
g_2(x,Q^2) = g_{2W}(x,Q^2) + \bar{g}_2(x,Q^2), \quad (D4)
\]
where the twist-2 piece is given as \[61\]
\[
g_{2W}(x,Q^2) = -g_1(x,Q^2) + \int_0^1 \frac{1}{y} g_1(y,Q^2) dy, \quad (D5)
\]
while that of the twist-3 piece is
\[
\bar{g}_2(x,Q^2) = g_1(x,Q^2) + g_2(x,Q^2) - \int_0^1 \frac{1}{y} g_1(y,Q^2) dy. \quad (D6)
\]
The twist-2 part undergoes the ordinary evolution as in Eq. (D3) and the twist-3 piece is first parameterized by its moments
\[
M_j(\mu^2) = \int dx x^{j-1} \bar{g}_2(x,\mu^2) \quad (D7)
\]
that scale as \[58\]
\[
M_j(Q^2) = \left[ \ln(\mu^2) \right]^{\gamma_{j-1}/\gamma_0} \left\{ \ln(Q^2) \right\}^{\gamma_{j-1}/\gamma_0} \text{with} \quad \gamma_{j-1} = 2N_c \left[ \psi(j) + \frac{1}{2j} + \gamma_E - \frac{1}{4} \right]. \quad (D8)
\]
Here \(\psi(j)\) is the logarithmic derivative of the Γ-function. Then \(\bar{g}_2(x,Q^2)\) is obtained by expressing it in terms of the evolved moments, i.e. by inverting Eq. (D7).
[1] T. Muta, *Foundations of Quantum Chromodynamics* (World Scientific, Singapore, 1987).
[2] R. G. Roberts, *The Structure of the Proton* (Cambridge Monographs on Mathematical Physics, 1990).
| 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
5439,
1
],
[
5439,
9029,
2
],
[
9029,
12034,
3
],
[
12034,
16643,
4
],
[
16643,
19898,
5
],
[
19898,
24264,
6
],
[
24264,
27971,
7
],
[
27971,
31465,
8
],
[
31465,
36054,
9
],
[
36054,
40921,
10
],
[
40921,
43949,
11
],
[
43949,
46722,
12
],
[
46722,
47427,
13
],
[
47427,
51312,
14
],
[
51312,
54190,
15
],
[
54190,
56820,
16
],
[
56820,
61042,
17
],
[
61042,
64810,
18
],
[
64810,
68481,
19
],
[
68481,
71775,
20
],
[
71775,
74386,
21
],
[
74386,
77492,
22
],
[
77492,
77492,
23
]
]
} |
4190b62b0e080b159585caead9b3219ec2696aab | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 25,
"total-fallback-pages": 0,
"total-input-tokens": 61112,
"total-output-tokens": 13889,
"fieldofstudy": [
"Physics"
]
} | Angular Dependence of Neutrino Flux in KM$^3$ Detectors in Low Scale Gravity Models
Pankaj Jain$^1$, Supriya Kar$^1$, Douglas W. McKay$^2$
Sukanta Panda$^1$ and John P. Ralston$^2$
$^1$Physics Department
I.I.T. Kanpur, India 208016
$^2$Department of Physics & Astronomy
University of Kansas
Lawrence, KS 66045
Abstract: Cubic kilometer neutrino telescopes are capable of probing fundamental questions of ultra-high energy neutrino interactions. There is currently great interest in neutrino interactions caused by low-scale, extra dimension models. Above 1 PeV the cross section in low scale gravity models rises well above the total Standard Model cross section. We assess the observability of this effect in the 1 PeV - 1000 PeV energy range of kilometer-scale detectors, emphasizing several new points that hinge on the enhancement of neutral current cross sections with respect to charged current cross sections. A major point is the importance of “feed-down” regeneration of upward neutrino flux, driven by new-physics neutral current interactions in the flux evolution equations. Feed-down is far from negligible, and it is essential to include its effect. We then find that the angular distribution of events has high discriminating value in separating models. In particular the “up-to-down” ratio between upward and downward-moving neutrino fluxes is a practical diagnostic tool which can discriminate between models in the near future. The slope of the angular distribution, in the region of maximum detected flux, is also substantially different in low-scale gravity and the Standard Model. These observables are only weakly dependent on astrophysical flux uncertainties. We conclude that angular distributions can reveal a breakdown of the Standard Model and probe the new physics beyond, as soon as data become available.
1 Introduction
The Ultra-high energy neutrino nucleon cross section $\sigma_{\nu N}$ is a topic of fundamental physical importance. Low scale gravity models \cite{1,2} predict enhancement of the neutrino-nucleon neutral current type cross section at center of mass energies above the fundamental gravity scale of about 1 TeV \cite{3,4,5}. Consequences for cosmic ray physics have been studied in application to the highest energy cosmic rays \cite{3,4,5,6,7,8,9,10,11}, where there is great interest in possible violation of the Greisen, Zatsepin, Kusmin (GZK) bound \cite{12}, at roughly 10 EeV ($10^{19}$eV). There is also great interest in application to an intermediate range \cite{5,13,14}, roughly 0.1 - 100 PeV ($10^{14} - 10^{17}$ eV). The history and future of the highest energy experiments \cite{15} and intermediate energy experiments \cite{16} provide a tremendous impetus for these studies pointing toward new physics.
Whatever the model, there are rich opportunities to study fundamental high energy interactions by focusing on (1) the ratio of neutral current type events to charged current events and (2) the angular distribution of events in upcoming experiments. The neutral-to-charged ratio and the angular distribution shape do not depend on the uncertainties of overall flux normalizations. The primary uncertainty in all cosmic ray comparisons with theory – the overall scale of the flux – drops right out. For a range of models with standard and reasonable flux spectral indices, the angular distributions are also remarkably insensitive to the details of the model. The strongest determining factor in the shape of the angular distribution is the fundamental physics of the interaction cross section itself. We emphasize and explore this fact, showing that arrays now planned or under construction could stringently test the Standard Model and proposals for new physics simply on the basis of the neutral-to-charge ratio and the slope of the angular distribution of neutrino-nucleon events. In the context of extra space-time dimensions, data could determine or bound such details as the scale and number of extra dimensions in the present models.
Here we continue earlier work \cite{5} that examined possible signatures of enhanced $\sigma_{\nu N}$ in kilometer scale detectors. Extensions of the currently operating AMANDA \cite{17} and RICE \cite{18,19} experiments to ICECUBE \cite{20}.
\footnote{In the present context, by neutral current type cross section we mean there is no leading, charged lepton produced. The shower is essentially hadronic, which includes the case of black hole final states.}
dimensions would certainly explore the 1 PeV - 100 PeV region. In the case of RICE, modest improvements even allow a reach above the EeV range. In [5] we used a linear extrapolation of the low scale gravity mediated neutral current $\sigma_{\nu N}$ from the low energy, $\sqrt{s} \ll 1$ TeV, region, to the $\sqrt{s} \geq 1$ TeV region and found a very sharp suppression of the up-to-down ratio compared to the standard model that set in at about 5 PeV for $M = 1$ TeV and at about 50 PeV for $M = 2$ TeV.
In that earlier study [5], we did not apply other extrapolations of the cross section to the up-to-down calculation, nor did we include the “feed down” effect [21, 22], which results from neutral current interactions degrading higher energy neutrinos as they travel through the earth and “feeding” the flux at lower energies. In the standard model, this effect is small above 1 PeV where the flux decrease steepens and the neutral current cross section is too small to compensate. In contrast, this effect turns out to be extremely important after including the gravity induced neutral current $\sigma_{\nu N}$, which rises rapidly with energy. This point has not been explicitly recognized previously in connection with gravity enhancement.
A number of new gravity effects relevant to energies above the fundamental scale, applicable to physics at the Large Hadron Collider (LHC), kilometer cubed detectors ($KM^3$) and GZK energies, have been proposed recently. In both Arkani-Hamed, Dimopoulos and Dvali (ADD) [1] and Randall and Sundrum (RS) [2] models, eikonal treatment of the effective low energy amplitude (used as the “Born term” input) has been studied [23] and applied to LHC [24] and GZK [8] energies. In string realizations of the ADD framework [25], “stringy” cross sections, relevant just below $M$, have been estimated [26, 27, 28], as has the black hole formation cross section [29, 7], which may be relevant above $M$, including the GZK energy region. We have investigated all of these options, and find that the eikonalized ADD low energy “Born amplitude” and “geometrical” black hole cross sections lead to the largest and least model dependent effects in our 1 PeV to 100 PeV $KM^3$ application. Our study goes beyond that reported recently in [13] in two respects: we emphasize higher energies and we include and analyze the consequences of the new, eikonalized graviton exchange component of the neutral current interaction. This latter point also distinguishes our work from a recent black hole detection study [14].
\footnote{We do not treat the possibility of brane production and decay here [30].}
We should note here that, though the neutral current does not produce a prompt electron shower or leading muon characteristic of charged current signatures, one expects that the hadronic shower, which is completely electromagnetic after several radiation lengths, will be an observable signature of neutral current interactions. For this reason we regard all of the current detection mechanisms to be relevant to our study, certainly including RICE, which can detect a radio pulse from any kind of shower, AMANDA and ICECUBE.
1.1 The Cross Section
At C.M. energies well above the Planck mass, the classical gravity Schwarzschild radius $R_s(\sqrt{s})$ is the dominant physical scale. The classical impact parameter, $b$, may make sense in this domain, and the eikonal approximation be valid, for values of $b$ larger than $R_s$. We will sketch the eikonal set-up shortly. At smaller impact parameters, the parton-level geometrical cross section
$$\hat{\sigma}_{BH} \approx \pi r_S^2$$
provides a classical, static estimate of the cross section to form black holes. In Eq. 1 $r_S$ is the Schwarzschild radius of a 4 + $n$ dimensional black hole of mass $M_{BH} = \sqrt{\hat{s}}$,
$$r_S = \frac{1}{M} \left( \frac{M_{BH}}{M} \right)^{1+n} \left[ \frac{2^n \pi^{(n-3)/2} \Gamma \left( \frac{3+n}{2} \right)}{2 + n} \right]^{1/n}$$
where $\sqrt{\hat{s}}$ is the parton-parton or, in our case, neutrino- parton C.M. energy, and $M$ is the 4+$n$-dimensional scale of quantum gravity. The black hole production process is expected to give a dominant contribution when $\sqrt{\hat{s}} >> M$. Black holes will form only if the impact parameter $b < r_S$. To convert Eq.1 into an estimate for the neutrino- nucleon cross-section, we fold it with the sum over parton distribution functions and integrate over $x$-values, where $\hat{s} = xs$, at a momentum transfer typical of the black hole production process:
$$\sigma_{\nu N \rightarrow BH}(s) = \Sigma_i \int_{x_{min}}^1 dx \hat{\sigma}_{BH}(xs) f_i(x, q).$$
3We use the mass scale convention discussed in [10], referred to as $M_D$ there.
Black hole formation requires $x > M^2/s$, so we take $x_{\text{min}} = M^2/s$. In addition, $q^{-1} = b < r_S$ is required. We adopt $q = \sqrt{s}$ up to $\sqrt{s} = 10\,\text{TeV}$, the maximum range in $q$ of the CTEQ parton distribution functions [31], the set we use, and $q = 10\,\text{TeV}$ when $\sqrt{s}$ is above this value. As remarked in [10], the dependence of $\sigma_{\nu N \rightarrow BH}(s)$ on the choice of $x_{\text{min}}$ and the treatment of $q$ is rather mild.
In the case of ADD model the black hole production cross sections can be large for $n > 2$, in which case the fundamental scale can be of the order of 1 TeV. A number of authors have adopted this estimate and applied it to ultra relativistic, parton level scattering. The approximation has been challenged on the basis that quantum corrections should lead to exponential suppression of individual channels, such as the black hole formation final state [32, 33], with several, independent arguments advanced in each case. In defense of the “black disk” approximation, several authors also point to success of internal consistency checks of the classical picture [28, 34, 24]. Recent phenomenological studies seem to be agnostic on this issue [9, 10, 11, 14, 35], treating the phenomenological consequences of both versions.
In a string picture with scale $M_s < M$, there is a range of energy $M_s \simeq \sqrt{s}$ where string resonances dominate [26], and a range $M_s < \sqrt{s} \leq M$, where stringball formation [28] could dominate. The cross section can be roughly expressed as [26]
$$\sigma_{\text{SR}}(\sqrt{s}) \sim g_s^2 \delta(s - M_{\text{SR}}^2), \sqrt{s} \simeq M_{\text{SR}},$$
(4)
for the string resonance case. Here $g_s$ is the (weak) string coupling constant and $M_{\text{SR}}$ is the mass of a string resonance state. Similarly, for the stringball case, estimates of the cross section give [28]
$$\sigma_{\text{SB}}(\sqrt{s}) \sim 1/M_s^2, M_s/g_s^2 < \sqrt{s} < M_s/g_s^2,$$
(5)
where $M$ is a few times less than $M_s/g_s^2$ for weak coupling. The impact of these various processes on the physics to be expected at the LHC, at a next linear collider (NLC) and very large hadron collider (VLHC) has been surveyed in a number of papers, summarized and referenced in the Snowmass 2001 report of the extra dimensions subgroup [36].
\[4\] In this discussion we suppress the $\hat{s}$ notation for convenience, though parton level processes are intended.
In our application to $KM^3$ physics in this paper, we mentioned above that the “classical” eikonal cross section [37, 23] and the geometric black hole formation cross section are the only cases where we find potentially observable effects. We outline our treatments of the eikonal model in the ADD [1] and RS1 [2] pictures next. The black hole cross section needs no further elaboration. The fundamental mass scale $M$ in the case of the ADD model for $n > 2$, can be of the order of 1 TeV, though new astrophysics analyses may constrain $n = 3$ more severely, as we comment below [38]. Similarly, in the RS picture the effective scale of gravity on the physical brane, the lowest K-K mode mass, can be arranged to be of the order of 1 TeV. In all of our quantitative work, we set the scale $M$ the same for every value of $n$ we use in our comparisons. As noted earlier, the scale $M$ is the same as $M_D$ defined in [24] and discussed in [10].
In the RS1 model with one extra dimension or in the ADD model with several, a possible choice for the input amplitude to the eikonal approximation, referred to as the Born amplitude, is given by,
$$iM_{\text{Born}} = \sum_i \frac{ics^2}{M^2 q^2 + m_i^2}$$
(6)
where $c$ is the gravitational coupling strength, which is effectively Newtonian for ADD and electroweak for RS. Here $q = \sqrt{-t}$ is the momentum transfer. In the Randall-Sundrum case, the sum runs over the massive K-K modes, constrained to start at or above the TeV scale when $c$ is of order electroweak strength. Their spacing is then also of TeV order. In the ADD case, the index $i$ must include the mass degeneracy for the $i$th K-K mode mass value. The spectrum is so nearly continuous that an integral evaluation of the sum is valid, but must be cut off at a scale generally taken to be of the order of $M$. Taking the transverse Fourier transform of the Born amplitude, we get the eikonal phase as a function of impact parameter, $b$,
$$\chi(s, b) = \frac{i}{2s} \int \frac{d^2q}{4\pi^2} \exp(iq \cdot b) iM_{\text{Born}}$$
(7)
For the ADD model, where $c = (M/\overline{M}_P)^2$ and $\overline{M}_P = 2.4 \times 10^{18}$ GeV is the reduced, four dimensional Planck mass,
$$\chi(s, b) = \frac{s(2^{2n-3} \pi^{\frac{4n}{3} - 1})}{M^{n+2} \Gamma(n/2)} 2 \int_0^\infty dmm^{n-1}K_0(mb)$$
6
\[ b^n_c = \frac{1}{2} (4\pi)^{\frac{n}{2}} \Gamma \left( \frac{n}{2} \right) \frac{s}{M^{2+n}}. \]
(9)
Because of the exponential decrease of \( K_0 \), the phase integral is actually ultraviolet finite. For the RS model, the corresponding expression is
\[ \chi(s, b) = \sum_i \frac{cs}{2M^2} K_0(m_i b). \]
(10)
For widely spaced KK modes in the Randall-Sundrum case, the lowest few modes dominate and contribute insignificantly in the 1 PeV < \( E_\nu < 100 \) PeV region. We do not discuss RS further.
The eikonal amplitude is then given in terms of the eikonal phase by
\[ M = -2is \int d^2 b \exp(iq \cdot b) \left[ \exp(i\chi) - 1 \right] \]
\[ = -i4\pi s \int db b J_0(qb) \left[ \exp(i\chi) - 1 \right], \]
(11)
The eikonal amplitude for the case of ADD model can be obtained analytically [23, 8, 24] in the strong coupling \( qb_c \gg 1 \) and weak coupling \( qb_c \ll 1 \) regimes.
In the strong coupling regime the eikonal amplitude can be computed using the stationary phase approximation and is given by,
\[ M = A_n e^{i\phi_n} \left[ \frac{s}{qM} \right]^{\frac{n+2}{2}} , \]
(12)
where
\[ A_n = \frac{(4\pi)^{\frac{3n}{2n+1}}} {\sqrt{n+1}} \left[ \Gamma \left( \frac{n}{2} + 1 \right) \right]^{\frac{1}{n}}, \]
(13)
\[ \phi_n = \frac{\pi}{2} + (n+1) \left[ \frac{b_c}{b_s} \right]^n \]
(14)
and \( b_s = b_c (qb_c/n)^{-1/(n+1)} \). In the weak coupling regime, \( q \to 0 \) the amplitude is given by
\[ M(q = 0) = 2\pi isb_c^2 \Gamma \left( 1 - \frac{2}{n} \right) e^{-\pi/n} . \]
(15)
As it turns out, the small q region contributes little to the cross section, and we use the simple rule that the amplitude is set to its value at $q = 1/b_c$ for values $q \leq 1/b_c$.
The parton-level cross section is calculated by assuming that it is given by the Born term as long as $\hat{s} < M^2$. For $\hat{s} > M^2$ the cross section is estimated by the eikonal amplitude. For $M = 1$ TeV, for example, the actual matching between the Born and eikonal amplitudes occurs in the range $\sqrt{\hat{s}} = 1 - 3$ TeV, depending on $n$ and the value of $y = q^2/\hat{s}$. In any case the region $\sqrt{\hat{s}} \sim M$ contributes negligibly to the cross-section, so the precise matching choice makes no difference in the final result. The eikonal calculation is not expected to be reliable if the momentum transfer $q > M$. At large momentum transfer we assume that the black hole production dominates the cross section. The eikonal cross section is, therefore, cut off once the momentum transfer $q > 1/r_S$. The neutrino-parton differential cross-section is folded with the CTEQ parton distributions and integrated over $x$ and $y$ variables, consistent with our momentum transfer restriction on the eikonal amplitude. The CTEQ limit at $x = 10^{-5}$ is exceeded only at the high end of the energy range we study. A standard power law extrapolation is used when $x$ does range below this value, though the $\hat{s}$ values are so low that the contribution of this range to the cross section is negligible. In Fig. 1 we plot the total neutrino-nucleon cross sections for several different values of the fundamental scale $M$ and the number of extra dimensions $n$, including both eikonal and black hole production contributions. The cross section is clearly more sensitive to the choice of the scale parameter, $M$, than to the number of dimensions, $n$. In fact the sensitivity to choice of $n$ comes primarily through the dependence of the differences in bounds on $M$ corresponding to different choices of $n$. The strongest bounds on $M$ come from astrophysical and cosmological considerations, which unavoidably require some degree of modeling.
A recent review of experimental and observational constraints is given in [38], where lower bounds on $M$ are quoted for $n = 3$ from various analyses such as supernova cooling, regarded as the least model dependent bound ($M \geq 2.5$ TeV), post-inflation re-heating ($M \geq 20$ TeV), and neutron star heat excess ($M \geq 60$ TeV). For $n = 4$, the most severe constraint is $M \geq 5$
\footnote{A discussion of the reliability of the eikonal amplitude in the $5$ TeV $\leq \sqrt{s} \leq 15$ TeV range is given in [24].}
Figure 1: The neutrino-proton cross section $\sigma_{\nu p}$ in the ADD model using the number of extra dimensions $n = 3, 4, 6$ as a function of the neutrino energy $E_{\nu}$ for $E_{\nu} < 10^8$ GeV. The geometric black hole production cross section is included. The solid, dotted and short dashed curves correspond to $n = 3, 4$ and 6 respectively. The long dashed curve represents the Standard Model prediction.
5TeV in the case of the post-inflation re-heating limit. There are essentially no constraints on the cases $n = 5$ and 6. Laboratory lower bounds are typically of order 1 TeV or less for all $n \geq 2$, with LEP II providing the strongest bound at 1.45 TeV for $n=2$.
2 Event Rates of Downward Neutrinos
In Fig. 2, we show the downward event rate, defined as the number of interactions from down- coming neutrinos, within a one kilometer cubed volume. We use two, quite different input flux assumptions to show the dependence of rate on the flux. A simple parameterization of an optically thick source model of flux above 1 PeV given in Ref. [39] (SDSS) is used, along with the flux bound for optically thin source environments of Ref. [40] (WB). The flux in [39] is about two orders of magnitude larger than the bound in [40], and it has roughly an $E^{-2}$ power law behavior from 0.1 PeV
to 10 PeV, and then it steepens to approximately $E^{-3}$. We parametrized the SDSS flux such that it falls as $E^{-2}$ for $10 \text{ TeV} < E < 10 \text{ PeV}$ and as $E^{-3}$ for $E > 10 \text{ PeV}$. The WB bound, on the other hand, falls as $E^{-2}$ over the whole energy range. The two flux curves cross at about $10^3 \text{ PeV}$.
Our estimate is made by taking the vertical flux result, multiplying it by $2\pi$ steradians and by the probability that a neutrino would interact within 1 km in ice, with density 0.93 gm/cc, given the cross section model in question. To get an actual event rate for a given detector, we would have to multiply by the acceptance of the detector. In the SDSS flux model, Fig. 2a, the number of interactions peaks at around 10 PeV for the $M = 1 \text{ TeV}$ case, with encouragingly large numbers of interactions, in the 350-650 range, induced by low-scale gravity. This is 10-20 times the SM interaction rate at the same energy. The number and the location of the peak rate depend upon the flux model of course [22]. This dependence is illustrated in Fig. 2b, where the event rate for the WB bound is shown. The event rates are lower by a factor of about 50 and the gravity induced events have a much broader peak, centered at about 15 PeV, compared to the SDSS case. The peak is broader for the WB flux than for SDSS because the WB limit falls as $E^{-2}$ throughout this energy region, while the SDSS flux falls as $E^{-3}$ above 10 PeV, cutting off the higher energy events more rapidly. The shape of the SM event rate is the same in both cases, since the flux shapes are the same below 10 PeV.
The excess above the SM is roughly half neutral current type events arising from eikonalized graviton exchange and half black hole events, which decay predominantly into hadrons. Therefore an optimal detection scheme requires sensitivity to the hadronic shower from the deposited energy in the ice. The ICECUBE and RICE detectors, for example would respond readily to hadron showers at these energies. The special role of enhanced neutral current type events, those with hadronic signatures like the graviton exchange and black hole events, prompts us to propose that the ratio of neutral current to charged current events provides a powerful tool to uncover new interactions. In particular, if the neutral current type component, distinguished by hadronic dominated showers and no leading charged lepton, dominates, as it does in the black hole and the eikonal regimes of the low scale gravity models, there would be no standard model explanation. The ratio of neutral current type events to charged current events, distinguished by the presence of a leading charged lepton, is shown in Fig. 3. The rapid rise that sets in above the threshold for new physics is remarkable, and would be so even without
the black hole contribution. Even scales $M > 2T eV$ are discernible with this observable. Putting together the information from several techniques, RICE and ICECUBE for example, one might well separate the “neutral” versus “charged” character of events and find a clear window on new physics.
3 The Regeneration and Angular Dependence of Neutrino Flux
We next calculate the up over down ratio of neutrino flux. The downward $\phi_0(E_\nu)$ is the flux of neutrinos incident on the surface of the earth from the sky. We will consider only the diffuse neutrino flux and assume that it is isotropic. Our calculations are easily generalized if the flux is found to be non-isotropic or if we are interested in individual sources. The upward flux $\phi_{up}(E_\nu)$ is defined as the flux of neutrinos coming upwards from the surface of the earth. This is the angular average
$$\phi_{up}(E_\nu) = \frac{1}{2\pi} \int_0^{2\pi} d\Phi \int_0^{\pi/2} \sin(\theta) d\theta \phi(E_\nu, \theta)$$
(16)
of the flux $\phi(E_\nu, \theta)$ emerging from the earth, where $\theta$ is the polar angle with respect to the nadir and $\Phi$ is the azimuthal angle. The ratio $R$ is essentially the ratio of up-to-down event rates. The event rates are given by the product of flux, cross section, number density, volume and acceptance, and to the extent that the latter four factors cancel in the ratio, only the up-to-down ratio of fluxes survives. This ratio is affected by the energy dependence of the flux, but not to its overall normalization. The upward flux depends on the cross sections of course, as we elaborate next.
In order to determine $\phi(E_\nu, \theta)$ we first need to solve the evolution equation for the neutrino propagating through the interior of the earth. In the case of the standard model, the neutrino cross section is dominated by the charged current, the neutral current does little to evolve, or “feed down”, the rapidly falling flux above 1 PeV, and neutrinos basically get lost after collision inside earth. In the present case, on the other hand, the eikonalized graviton exchange gives a large contribution to the feed down above 1 PeV, and we have to include this important enhancement of regeneration of lower
The large cross sections mean that neutrinos experience many interactions as they proceed through the earth, even at angles near the horizon. For instance, just 5 degrees below the horizon a 100 PeV neutrino traverses more than 20 interaction lengths of earth before reaching the detector region. This behavior supports our use of the continuous evolution model, summarized below in Eq.17, since fluctuations are small for UHE application, where the cross section is large. The neutrino loses little energy during any particular collision and hence it is reasonable to assume that it practically moves in a straight line path. The evolution equation for the neutrino is given by [21, 22]
\[
\frac{d\ln \phi}{dt}(E_\nu, \theta) = -\sigma_{W+BH}^{\nu N}(E_\nu) - \sigma_{Z+E}^{\nu N}(E_\nu) + \int_{E_\nu}^\infty dE'_\nu \frac{\phi(E'_\nu, \theta)}{\phi(E_\nu, \theta)} \frac{d\sigma_{Z+E}^{\nu N}(E'_\nu, E_\nu)}{dE'_\nu(E'_\nu, E_\nu)} \equiv -\sigma_{\text{eff}}^{\nu N}(E_\nu, \theta),
\]
where \(\sigma_{W+BH}^{\nu N}\) is the “neutrino absorbing” W-exchange + black hole cross section and \(\sigma_{Z+E}^{\nu N}\) is the “neutrino regenerating” Z-exchange + eikonalized graviton exchange cross section. In Eq.17, \(dt = n(r)dz\) and \(n(r)\) is the number density of nucleons at any distance \(r\) from the center of earth, radius \(R_e\). Expressing the flux \(\phi(E_\nu, \theta)\) as
\[
\phi(E_\nu, \theta) = \phi_0(E_\nu) \exp[-\sigma_{\text{eff}}(E_\nu, \theta)t(\theta)],
\]
where the column density at upcoming angle of entry \(\theta\), chord length \(2R_e\cos\theta\), is given by
\[
t(\theta) = \int_0^{2R_e\cos\theta} n(z, \theta)dz,
\]
we solve equations 17 and 18 numerically by iteratively improving the flux \(\phi(E_\nu, \theta)\). Using this solution we can determine the ratio \(R\) of up to down flux,
\[
R = \frac{\phi_{\text{up}}(E_\nu)}{\phi_0(E_\nu)}
\]
---
6 The black hole production component, in contrast, essentially leads to loss of neutrinos upon collision. In the present context, the black hole interaction acts like the charged current in the standard model. The eikonal component has the character of the usual neutral current, transferring a small fraction of the neutrino energy to the hadron and leaving a leading neutrino in the final state.
7 We do not treat the regeneration of \(\tau\) neutrinos through \(\tau\) production by \(\nu_\tau\) and subsequent \(\tau\) decay back to \(\nu_\tau\) [41].
By using $R$, one sharply reduces the effects of experimental systematics and flux normalization and isolates the dependence on cross sections and flux shape. Refinements of angular binning can add information according to the size of the data sample, as we discuss below.
In Fig. 4 we plot the ratio $R$ as a function of the neutrino energy. This figure illustrates two points: the ratio $R$ is insensitive to the normalization of the flux assumed and, given a flux, the feed-down from higher to lower energies as the neutrinos pass through the earth is a powerful effect. In this application, our two, quite different input flux assumptions, show the insensitivity of $R$ to the flux used. The long dashed curve gives the ratio for the larger flux with a “knee” from $[39]$, while the shorter dashed line gives the result of the smaller flux bound with uniform $E^{-2}$ fall-off of $[40]$. The solid curve refers to the SM cross section input, the two flux models give indistinguishable results in this case and are shown as one line. The lowest curve shows the ratio as a function of energy with pure absorption, in the sense that only the first two terms in $\sigma_{\text{eff}}$, Eq.17., are included. The two flux models give the same result, as they should. The sum of the eikonal and black hole cross section is used, the eikonal providing the neutral-current cross section and the black hole the purely absorptive cross section, as described in footnote 6. The cross section are computed in the ADD model, and the value $n = 3$ and the scale of quantum gravity $M = 1$ TeV. We see that the regeneration term gives a big effect for $M = 1$ TeV, producing factors of more than 10 above 20 PeV, and it is essential to include it in assessing the consequences of the low scale gravity models. Though radically different from each other, the two flux assumptions lead to nearly identical values of $R$ out to 50 PeV and then differ only weakly above that, where the effect of the steeper decrease of SDSS flux shows up.
In Fig. 5 we plot the results for the ratio $R$ in the ADD model as a function of the neutrino energy for the choice of the fundamental scale $M = 1, 2$ TeV and the number of extra dimensions equal to 3, 4 and 6. The result depends on both the number of extra dimensions and, especially strongly, on the fundamental scale $M$. If the fundamental scale is larger than about 2 TeV, the KM$^3$ neutrino detectors will not be able to distinguish the Standard model result from the predictions of quantum gravity by using $R$ as a diagnostic. However for $M \approx 1$ TeV, we find that the effect is very large and, given enough flux, could be seen in these detectors. For energies above 1 PeV, a noticeable difference in $R$ between the two cases is seen. With
sufficient data, a distinction between no deviation from the standard model and a deviation corresponding to $M = 1$ with $n = 3, 4$ or 6 may be drawn.
The angular distribution of upward flux is an important diagnostic tool, as we stressed in the Introduction. In Fig. 3 we show the integrated flux per square kilometer per year above 1.8 PeV as a function of nadir angle for $M = 1, 2$ and $n=3,4,6$. The (higher) 2 TeV curve is essentially the same as for the SM, while the (lower) curves for $M = 1$ show clear suppression at all nadir angles up to $\pi/2$. It is somewhat disguised by the log scale, but most of the contribution to the ratio $R$ comes from the highest event rates near the horizontal. With several bins of good statistics data above 1 radian in nadir angle, one can distinguish the slopes of the flux vs. nadir angle near the horizontal. The difference between these slopes for the SM and low scale gravity models when $M \approx 1$ TeV provides another new physics discriminator.
As in the case of the $R$ observable, the slope is insensitive to the flux value, enhancing its power at identifying the nature of the neutrino interactions.
Realistically, one needs enough upcoming events at a given energy to calculate a meaningful ratio. The situation for the two flux models we are discussing is shown in the table.
| n | M | 1-10 PeV | 5 - 10 PeV | > 10 PeV | > 15 PeV |
|---|---|---------|-----------|---------|---------|
| 3 | 1 | 83 | 7.2 | 1.3 | 0.22 |
| 4 | 1 | 79 | 5.4 | 0.62 | 0.07 |
| 6 | 1 | 74 | 2.6 | 0.09 | 0.004 |
| 3 | 2 | 80 | 8.9 | 4.3 | 1.8 |
| 4 | 2 | 80 | 8.9 | 4.2 | 1.7 |
| 6 | 2 | 80 | 9 | 4.1 | 1.6 |
| SM| | 80 | 8.8 | 4.2 | 1.7 |
Table 1: The numbers of upward events per year expected over the energy ranges shown, for models with different $n$ and $M$ choices and for the standard model. The flux used is our rough approximation to SDSS above 1 PeV.
For our two power law approximation to the flux of [39] (SDSS), we see from the table that there are enough events with $E_\nu > 10$ PeV in a 10 year run to determine the ratio $R$. Can one discriminate among the different physics models for the cross section with the events? Fig. 5 shows that
if there are enough upcoming events to calculate a meaningful value for $R$, there are clear distinctions among the models. For example, taking a naive, purely statistics estimate, the $M = 1$ TeV, $E_{\nu} > 10$ PeV case would produce $13 \pm 3, 6 \pm 2$ and $1 \pm 1$ events for $n = 3, 4$ and 6. Combining these values with the downward event rates (see Fig. 2), we see that the distinction between the $n = 3$ case and the $n = 6$ cases is significant. When $M = 2$ TeV, the numbers of upcoming events is essentially the same for the low scale gravity models and the SM. In 5 - 10 years of data, there would be enough events to determine an up-to-down ratio reasonably well. Figure 2 shows that the number of downward events is larger in the low scale gravity models than in the SM, because of their larger cross sections, so $R$ values are different for the different cases. Thus, even for $M = 2$, values of $R$ above 15 PeV are distinctly different in the two classes of models, as one sees in Fig. 5. In fact, within the low scale gravity models themselves there is more than a factor two difference between $n = 3$ and $n = 6$. Moreover, the low scale gravity models can all be easily distinguished from the SM with 5 years of data, with the number of events per year shown in the table. Even the cut $> 15$ PeV allows the meaningful distinction between the low scale models and the SM in 5 - 10 years of data. Using the WB bound on the optically thin source flux, we find that the upward events are too sparse to discriminate among models by use of the $R$ ratio. Flux values in between the two presented here offer various levels of discrimination, with useful information obtainable for the larger fluxes.
Putting together the rise in downward event rate above 1 PeV, the sensitivity of $R$ to the new physics cross sections, and the slope of the upward flux as a function of nadir angle, we see that a signal of low scale gravity models would stand out clearly.
4 Conclusions:
$KM^3$ detectors have been planned primarily as “neutrino telescopes”. However our study indicates that $KM^3$ will also provide vigorous new exploration of fundamental questions in ultra-high energy physics. Extra dimension, low scale gravity models show an observable impact on the signature of neutrino events in the $KM^3$ detectors such as the currently running RICE detector and the ICECUBE detector, which is in the R & D stage. Of the “strong
gravity” effects we looked at, the black hole formation cross section and the extrapolation of the small $q^2$, part of $\sigma^{\nu N}$ from the $s < M^2$ to $s \gg M^2$ within the ADD model yield the largest detectable effects.
While some observables are somewhat sensitive to the number of dimensions, $n$, most are quite sensitive to the value of the scale of gravity $M$. If $M$ is 1-2 TeV, the enhanced $\sigma^{\nu N}$ creates a clearly recognizable signature, with the “neutral-to-charged” event ratio still showing new physics effects at $M = 5$ TeV. The ratio $R$ of upcoming to down-going events is a powerful diagnostic, capable of discriminating between models if fluxes are large enough to produce a significant number of upcoming events. In the entire analysis, we emphasized that the regeneration of neutrino upcoming flux due to neutrinos scattering down from higher to lower energies is a crucial feature of the gravity-induced neutral current interactions. This is a general and important feature of our work presented here.
Considering only down-coming events, we propose that the fraction of neutral current type events, those with hadronic showers and no leading charged lepton, provides a probe of new interactions. In particular, event signatures arising from a dominant component of eikonalized graviton exchange or of black hole production as occur in low scale gravity models, would have no standard physics explanation, and would point the way toward physics beyond the Standard Model if observed. There is every reason to believe that the neutral-to-charged current ratio can be extracted from upcoming facilities well enough to make a practical signal. In particular, determining the ratio of events with a muon, to those making an isolated shower, is certainly feasible with a combination of AMANDA/ICECUBE and RICE technology. The striking behavior of the neutral current-to-charged current ratio is shown in Fig. 3.
Finally, the shape and slope of the angular distribution, Fig. 6, is found to have good discriminating power. The shape does not depend at all on the overall normalization of the flux. Moreover the slopes differ substantially right in the regime of maximum detectable flux, near $\pi/2$ nadir angle, and is ideal for comparing low scale gravity models to the Standard Model. Whether or not extra-dimension models as currently envisaged survive, the angular distribution can severely test the neutrino physics of the Standard Model, possibly strongly bounding or even discovering new physics, as soon as data becomes available.
**Acknowledgements:** P. Jain thanks the University of Kansas College of
Arts and Sciences and Department of Physics and Astronomy for hospitality and support in the course of this work. This research was supported in part by the U.S. Department of Energy under grant number DE-FG03-98ER41079. We used computational facilities of the Kansas Center for Advanced Scientific Computing for part of this work.
References
[1] N. Arkani-Hamed, S. Dimopoulos and G. Dvali, Phys. Lett. B 429, 263 (1998).
[2] L. Randall and R. Sundrum, Phys. Rev. Lett. 83, 3370 (1999).
[3] S. Nussinov and R. Shrock, Phys. Rev. D 59, 105002 (1999).
[4] G. Domokos and S. Kovesi-Domokos, Phys. Rev. Lett. 82, 1366 (1999).
[5] P. Jain, D. McKay, S. Panda and J. Ralston, Phys. Lett. B 484, 267 (2000).
[6] S. Nussinov and R. Shrock, Phys. Rev. D 64, 047702 (2001); A. Jain, P. Jain, D. McKay and J. Ralston, hep-ph/0011310, Int. Jour. of Mod. Phys. A 17, 533 (2002); C. Tyler, A. Olinto and G. Sigl, Phys. Rev. D 63, 55001 (2001); L. Anchordoqui, H. Goldberg, T. McCauley, T. Paul, S. Reucroft and J. Swain, Phys. Rev. D 63, 124009 (2001); M. Kachelriess and M. Plumacher, Phys. Rev. E 62 (2000).
[7] J. Feng and A. Shapere, Phys. Rev. Lett. 88, 021303 (2002); A. Anchordoqui and H. Goldberg, Phys. Rev. D 65, 047502 (2002).
[8] R. Emparan, M. Masip and R. Rattazzi, Phys. Rev. D 65, 064023 (2002).
[9] A. Ringwald and H. Tu, Phys. Lett. B 525, 135 (2002).
[10] L. Anchordoqui, J. Feng, H. Goldberg and A. Shapere, Phys. Rev. D 65, 124027 (2002).
[11] M. Kowalski, A. Ringwald and H. Tu, Phys. Lett. B 529 1 (2002).
[12] K. Greisen, Phys. Rev. Lett. 16, 748 (1966); G. Zatsepin and V. Kuzmin, Pis’ma Zh. Eksp. Teor. Giz. 4, 114 (1966) [JETP. Lett. 4, 78 (1966)].
[13] J. Alvarez-Muñiz, F. Halzen, T. Han and D. Hooper, Phys. Rev. Lett. 88, 021301 (2002).
[14] J. Alvarez-Muñiz, J. Feng, F. Halzen, T. Han and D. Hooper, Phys Rev. D 65, 124015 (2002).
[15] S. Yoshida et al. [AGASA Collaboration], in Proc. 27th International Cosmic Ray Conference, Hamburg, Germany, 2001, Vol. 3, p 1142; D. Zavrtanik [AUGER Collaboration], Nucl. Phys. Proc. Suppl. 85, 324 (2000); D. Bird et al. [Fly’s Eye Collaboration], Ap. J. 441, 151 (1995); J. Matthews et al. [HiRes Collaboration], in Proc. 27th International Cosmic Ray Conference, Hamburg, Germany, 2001, Vol. 1, p. 350; M. Lawrence, R. Reid and A. Watson, J. Phys. G 17, 733 (1991).
[16] F. Halzen et al. [AMANDA Collaboration] in Proceedings of the 26th International Cosmic Ray Conference (ICRC 99), Salt Lake City, 17-25 Aug. 1999, B. L. Dingus, D. B. Kieda, M. H. Salamon (Eds.), (AIP, 2000) pp. 428-431. T. Montaruli [ANTARES Collaboration], hep-ex/0201009; V. Balkanov et al., Baikal Collaboration, astro-ph/0112446; P. Greider [NESTOR Collaboration], Nucl. Phys. Proc. Suppl. 97, 105 (2000); I. Kravchenko et al. [RICE Collaboration], astro-ph/0112372. The ICECUBE Project, http://www.ssec.wisc.edu/a3ri/icecube.
[17] F. Halzen et al. [AMANDA Collaboration] in Proceedings of the 26th International Cosmic Ray Conference (ICRC 99), Salt Lake City, 17-25 Aug. 1999, B. L. Dingus, D. B. Kieda, M. H. Salamon (Eds.), (AIP, 2000) pp. 428-431.
[18] I. Kravchenko et al. [RICE Collaboration], astro-ph/0112372.
[19] I. Kravchenko et al. [RICE Collaboration], astro-ph/0206371.
[20] The ICECUBE Project, http://www.ssec.wisc.edu/a3ri/icecube.
[21] V. Berezinskii, A. Gazizov, G. Zatsepin and I. Rozental, Yad. Fiz. 43, 637 (1986); Sov. J. Nucl. Phys. 43, 406 (1986).
[22] For an application in the context of the present discussion, see G. M. Frichter, J. P. Ralston and D. W. McKay, Phys. Rev. D 53, 1684 (1996).
[23] R. Emparan, Phys. Rev. D 64, 0204025 (2001).
[24] G. Giudice, R. Rattazzi and J. Wells, Nucl. Phys. B 630, 293 (2002).
[25] I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos and G. Dvali, Phys. Lett. B 463, 257 (1998).
[26] S. Cullen, M. Perelstein and M. Peskin, Phys. Rev. D 62, 055012.
[27] E. Dudas and J. Mourad, Nucl. Phys. B 575, 3 (2000); R. Cornet, J. Illana and M. Masip, Phys. Rev. Lett. 86, 4235 (2001).
[28] S. Dimopoulos and R. Emparan, Phys. Lett. B 526, 393 (2002).
[29] S. Dimopoulos and G. Landsberg, Phys. Rev. Lett. 87, 161602 (2001); S. Giddings and S. Thomas, Phys. Rev. D 65, 056010 (2002).
[30] E-J. Ahn, M. Cavaglia and A. Olinto, hep-th/0201042; P. Jain, S. Kar, S. Panda and J. Ralston, hep-ph/0201232.
[31] H.L. Lai et al., Phys. Rev. D. 55, 1280 (1997).
[32] M. Voloshin, Phys. Lett. B 518, 137 (2001); ibid 524, 376 (2002).
[33] J. Ralston, “Quantum Suppression of Black-Hole Formation in High Energy Scattering”, (unpublished).
[34] S. Giddings, hep-ph/0110127
[35] T. Rizzo, JHEP 0202, 011 (2002).
[36] G. Pásztor and T. Rizzo, hep-ph/0112054.
[37] G. ’t Hooft, Phys. Lett. B, 198, 61 (1987).
[38] J.Hewett and M Spiropulu, hep-ph/0205106.
[39] F. W. Stecker, C. Done, M. Salamon and P. Sommers, Phys. Rev. Lett. 66, 2697 (1991); erratum Phys. Rev. Lett. 69, 2738 (1992).
[40] J. Bahcall and E. Waxman, Phys. Rev. D 64, 023002 (2001).
[41] J. Beacom, P. Crotty and E. Kolb, astro-ph/0111482.
Figure 2: The downwards event rate per cubic kilometer per year within the Standard Model (SM) and in the ADD model for the fundamental scale $M = 1, 2$ TeV and for the number of extra dimensions, $n = 3, 4, 6$ using (a) the SDSS flux model and (b) the WB flux model. The solid line shows the Standard Model prediction alone. The long dashed, dotted and short dashed curves show the predictions of the neutral current graviton exchange within the ADD model plus geometric, black hole cross sections, and including the SM contribution. The upper and lower histograms correspond to the $M = 1, 2$ TeV choices respectively. The new physics contributions are insignificant below 1 PeV, where their contribution above the SM is not visible on this scale.
Figure 3: The ratio of neutral current type events to SM charged current events as a function of neutrino energy for $n = 3, 4$ and $6$ and for $M = 1, 2,$ and $5$ TeV. Only the downward events are included in this plot. The neutral current type interactions, in the sense used here, are dominated by the black hole and eikonal components of the low scale gravity amplitude above the scale of gravity.
Figure 4: The contribution of the regeneration term to the ratio $R$ of the upwards and downwards flux within the ADD plus geometric black hole production model with the fundamental scale $M = 1$ TeV and for the number of extra dimensions equal to 3. The long dashed line (flux from [39]) and short dashed line (flux from [40]) predict nearly the same $R$ as a function of energy for the full calculation, including the regeneration term. Ignoring the regeneration of flux, one gets the lowest curve, identical for both flux models. The regeneration term has a profound effect on the up-to-down flux ratio $R$ due to the large, gravity driven neutral current cross section. Standard model prediction, the solid line, is also shown and it is the same for both flux assumptions.
Figure 5: The ratio $R$ as a function of neutrino energy for the fundamental scale $M = 1, 2$ TeV and for the number of extra dimensions 3 (solid curves), 4 (dotted curves) and 6 (dashed curves). The Standard model (SM) prediction is also shown.
Figure 6: The angular dependence of the upwards flux for the fundamental scale $M = 1, 2$ TeV and for the number of extra dimensions, $n = 3, 4, 6$. The Standard model (SM) prediction is also shown. Only neutrinos with energy $E_\nu > 1.8$ PeV are considered. The $M = 2$ TeV results are indistinguishable from the Standard Model, but the R value is different because downward event rate is larger in the M=2 low scale gravity model (see Fig.2). | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
1837,
1
],
[
1837,
4472,
2
],
[
4472,
7106,
3
],
[
7106,
9197,
4
],
[
9197,
11662,
5
],
[
11662,
13978,
6
],
[
13978,
15507,
7
],
[
15507,
18188,
8
],
[
18188,
19498,
9
],
[
19498,
22341,
10
],
[
22341,
24572,
11
],
[
24572,
27024,
12
],
[
27024,
29808,
13
],
[
29808,
32141,
14
],
[
32141,
34585,
15
],
[
34585,
37238,
16
],
[
37238,
38766,
17
],
[
38766,
40545,
18
],
[
40545,
42006,
19
],
[
42006,
42260,
20
],
[
42260,
43010,
21
],
[
43010,
43412,
22
],
[
43412,
44189,
23
],
[
44189,
44435,
24
],
[
44435,
44880,
25
]
]
} |
c25fa814879ec396e4ed94087c8e5d0da7fd919f | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 29,
"total-fallback-pages": 0,
"total-input-tokens": 150036,
"total-output-tokens": 36873,
"fieldofstudy": [
"Mathematics"
]
} | IN Variant Measures and Measurable Projective Factors for Actions of Higher-Rank Lattices on Manifolds
Aaron Brown, Federico Rodriguez Hertz, and Zhiyuan Wang
Abstract. We consider smooth actions of lattices in higher-rank semisimple Lie groups on manifolds. We define two numbers $r(G)$ and $m(G)$ associated with the roots system of the Lie algebra of a Lie group $G$. If the dimension of the manifold is smaller than $r(G)$, then we show the action preserves a Borel probability measure. If the dimension of the manifold is at most $m(G)$, we show there is a quasi-invariant measure on the manifold such that the action is measurable isomorphic to a relatively measure preserving action over a standard boundary action.
1. Introduction and Statement of Results
In this paper we consider lattices $\Gamma$ in higher-rank Lie groups $G$ acting by $C^{1+\beta}$-diffeomorphisms on compact manifolds. The Zimmer program refers to a number of questions and conjectures related to such actions. It is expected that all such actions are constructed from algebraic examples. In particular, if the dimension of $M$ is smaller than the dimension of all possible algebraic actions, Zimmer’s conjecture asserts that all actions factor through the action of a finite group. See [BFH] for recent solution to (non-volume-preserving case of) Zimmer’s conjecture for cocompact lattices in split, simple Lie groups.
The main results of this paper concern actions of lattices in low dimensions. Most rigidity results in the literature concerning actions of lattices in low dimensions require additional hypotheses such as the preservation of a Borel probability measure (see [FH, Pol]), strong regularity assumptions of the action (see [FS]), or extremely low dimensions (see [Wit, BM, Ghy] for actions on the circle and [FH, Pol] for actions on surfaces.) Our focus in this paper is to establish the existence of an invariant measure for actions in moderately low dimensions and with low differentiability. In particular, in Theorem 1.6 we show that if the dimension of $M$ is sufficiently small relative to algebraic data associated to a simple Lie group $G$, then for any lattice $\Gamma \subset G$, any $C^{1+\beta}$-action $\alpha : \Gamma \to \text{Diff}^{1+\beta}(M)$ preserves a Borel probability measure. The critical dimension below which we are guaranteed an invariant probability is precisely the critical dimension in the non-volume-preserving case of Zimmer’s conjecture for split, simple Lie groups. In the case that $\Gamma$ is cocompact, Theorem 1.6 follows immediately from the main result of [BFH]; on the other hand, the proof of the main result of [BFH] uses many of the ideas used to prove Theorem 1.6, particularly our Proposition 5.1 below. Theorem 1.6 moreover holds for actions of nonuniform lattices whereas Zimmer’s conjecture has yet to be verified for nonuniform lattices.
The second main result, Theorem 1.10, concerns actions $\alpha : \Gamma \to \text{Diff}^{1+\beta}(M)$ on manifolds $M$ of certain intermediate dimensions. This range of dimensions includes examples where there exist non-trivial (volume-preserving) actions as well as examples of actions that do not preserve any Borel probability measure. In this case, we show that there exists a quasi-invariant measure $\mu$ on $M$ such that the action on $(M, \mu)$ is measurably isomorphic to a relatively measure-preserving extension over a standard projective action.
Given an action \( \alpha : \Gamma \to \text{Diff}^{1+\beta}(M) \), the key idea in both theorems is to consider the \( G \)-action induced by \( \alpha \) on an auxiliary space which we denote by \( M^\alpha \). We take \( P \subset G \) to be a minimal parabolic subgroup and consider \( P \)-invariant measures \( M^\alpha \). This approach should be compared with a number of papers by Nevo and Zimmer, particularly \([NZ1, NZ2]\). Nevo and Zimmer consider a manifold with a \( G \)-action and \( G \)-stationary measure \( \nu \). \( \nu \) decomposes as \( \nu_0 \ast \lambda \) where \( \lambda \) is a \( P \)-invariant measure (See \([NZ1, \text{Theorem 1.4]} \) for discussion of this decomposition). Assuming that \( \lambda \) satisfies certain technical conditions—namely that the measure \( \lambda \) is either \( P \)-mixing in \([NZ1]\) or that every non-trivial element of the maximal split Cartan subgroup \( S \subset P \) acts ergodically in \([NZ2]\)—it is shown that the \( G \)-action on \( (M, \nu) \) is a relatively measure-preserving extension over a standard projective action. These technical conditions are typically difficult to verify. In our argument, we exploit the constraints on the dimension of \( M \) and verify certain conditions similar to those introduced by Nevo and Zimmer. For instance, the technical condition in \([NZ2, \text{Theorem 3}] \) that all elements of the maximal split Cartan subgroup \( S \subset P \) act ergodically implies our Claim 6.2 below and hence all arguments in Section 6.2 apply. In practice, it is difficult to verify such ergodicity hypotheses.
1.1. Introduction. Throughout we assume that \( G \) is a real, connected, semisimple Lie group with finite center and \( \mathbb{R} \)-rank at least 2. By a standard construction, there is a finite cover \( \tilde{G} \to G \) such that \( \tilde{G} \) is the direct product of connected, almost-simple Lie groups:
\[
\tilde{G} = \prod G_i.
\]
We take \( \Gamma \subset G \) to be lattice and, writing \( \tilde{\Gamma} \) for the lift of \( \Gamma \) to \( \tilde{G} \), we assume that for every almost-simple factor \( G_i \subset \tilde{G} \) with \( \mathbb{R} \)-rank 1, the projection of \( \tilde{\Gamma} \) to \( G_i \) is dense in \( G_i \). Such a lattice will be called a higher-rank lattice. This in particular includes the cases that
1. \( G \) has no compact factors and \( \Gamma \subset G \) is irreducible, or
2. every non-compact, almost-simple factor of \( G \) has \( \mathbb{R} \)-rank at least 2.
Below, we will study smooth actions of such groups \( \Gamma \). As we may lift an action of \( \Gamma \) to an action of \( \tilde{\Gamma} \), without loss of generality we will assume for the remainder that \( G \) is a direct product \( G = \prod G_i \) of almost-simple Lie groups.
Note that \( G = C \times G' \) where \( C \) is the maximal connected compact normal subgroup of \( G \) and \( G' \) is the maximal connected normal subgroup without compact factors. We remark that our main results—Theorems 1.6 and 1.10—are sharpest when \( G' \) is assumed to be simple.
Let \( M \) be a compact, connected, boundaryless \( C^\infty \) manifold and let \( \alpha : \Gamma \to \text{Diff}^{1+\beta}(M) \) be an action of \( \Gamma \) on \( M \) by \( C^{1+\beta} \) diffeomorphisms. For notational convenience later, we assume \( \alpha \) is a right action; that is \( \alpha(gh)(x) = \alpha(h)(\alpha(g)(x)) \). Conjecturally, all such actions are obtained from families of model algebraic actions via standard constructions. In particular, if \( \dim(M) \) is sufficiently small so that no model algebraic actions exists, Zimmer’s conjecture states that all such actions should factor through actions of finite quotients of \( \Gamma \); that is, the image \( \alpha(\Gamma) \) of \( \Gamma \) in \( \text{Diff}^{1+\beta}(M) \) should be finite. Such an action is said to be trivial. See \([FS, \text{Conjectures I, II}], [\text{Fis, Conjectures 4.12, 4.14}], \text{or [BFH, Conjecture 2.4]} \) for more precise formulations. See also \([\text{BFH}] \) for recent solution to (the non-volume-preserving case of) Zimmer’s conjecture for cocompact lattices in split, simple Lie groups.
We recall that in dimension 1, any lattice in a higher-rank, simple Lie group with finite center acts trivially on the circle \([Ghy, BM] \). For certain lattices acting on surfaces, we obtain in conjunction with the main results of \([FH] \) the following complete results.
Theorem A ([FH, Corollary 1.8] + Theorem 1.6). Let $S$ be a closed oriented surface and for $n \geq 4$, let $\Gamma \subset \text{SL}(n, \mathbb{Z})$ be a finite index subgroup. Then every $C^{1+\beta}$ action of $\Gamma$ on $S$ is trivial.
Theorem B ([FH, Corollary 1.7] + Theorem 1.6). Let $S$ be a closed oriented surface of genus at least 1 and for $n \geq 4$, let $\Gamma \subset \text{SL}(n, \mathbb{R})$ be a nonuniform lattice. Then every $C^{1+\beta}$ action of $\Gamma$ on $S$ is trivial.
More generally, Theorem B holds when $\Gamma \subset G$ is a nonuniform lattice and $G$ is a connected, semisimple Lie group with finite center, no compact factors, and $r(G) \geq 3$ for the integer $r(G)$ defined below ([FH, Corollary 1.7]). In particular, the conclusion of Theorem B hold for any nonuniform lattice in a higher-rank, simple Lie group $G$ with finite center such that the restricted root system of the Lie algebra of $G$ is not of type $A_2$. By the main results of [BFH], triviality of all actions on surfaces also holds for cocompact lattices in all such groups.
Note that if $\Gamma \subset \text{SL}(3, \mathbb{R})$ is any lattice then there is model real-analytic action of $\Gamma$ on a surface $S$ that admits no invariant probability measure—namely, the right projective action of $\Gamma \subset \text{SL}(3, \mathbb{R})$ on $\mathbb{R}P^2$ (or $S^2$.) Note that any volume form on $\mathbb{R}P^2$ is quasi-invariant but non invariant under this action. More generally, consider $G$ a semi-simple Lie group with finite center. Let $Q \subset G$ be a parabolic subgroup and let $\Gamma \subset G$ be a lattice. Then there is a natural right-action of $\Gamma$ on the quotient $Q \backslash G$ preserving no Borel probability measure but preserving the Lebesgue measure class.
Given the model action discussed above, we have the following conjecture, motivated by Theorems A and B, attributed to Polterovich in [Fis, Question 4.8].
Conjecture 1.1. Let $\Gamma \subset \text{SL}(3, \mathbb{R})$ be a lattice. Let $S$ be closed, connected a surface and let $\Gamma$ act on $S$ by $C^{1+\beta}$ diffeomorphisms. Suppose there is no $\Gamma$-invariant Borel probability measure on $S$. Then $S$ is either $\mathbb{R}P^2$ or $S^2$; furthermore any such action is smoothly conjugate to the standard projective action.
1.2. Facts from the structure of Lie groups. To state our main results we recall some facts and definitions from the structure theory of real Lie groups. A standard reference is [Kna]. Let $G$ be a connected, semisimple Lie group with finite center. As usual, write $\mathfrak{g}$ for the Lie algebra of $G$. Fix a Cartan involution $\theta$ of $\mathfrak{g}$ and write $\mathfrak{t}$ and $\mathfrak{p}$, respectively, for the $+1$ and $-1$ eigenspaces of $\theta$. Denote by $a$ the maximal abelian subalgebra of $\mathfrak{p}$ and by $m$ the centralizer of $a$ in $\mathfrak{t}$. We let $\Sigma$ denote the set of restricted roots of $\mathfrak{g}$ with respect to $a$. Note that the elements of $\Sigma$ are real linear functionals on $a$. Recall that $\dim_{\mathbb{R}}(a)$ is the $\mathbb{R}$-rank of $G$.
We choose a family of positive roots $\Sigma_+ \subset \Sigma$ and write $\Sigma_-$ for the corresponding set of negative roots. For $\beta \in \Sigma$ write $\mathfrak{g}^\beta$ for the associated root space. Then $n = \bigoplus_{\beta \in \Sigma_-} \mathfrak{g}^\beta$ is a nilpotent subalgebra. A standard parabolic subalgebra (relative to the choice of positive roots $\Sigma_+$) is any subalgebra of $\mathfrak{g}$ containing $m \oplus a \oplus n$. Recall $\beta \in \Sigma_+$ is a simple (positive) root if it is not a linear combination of other elements in $\Sigma_+$. We denote by $\Pi \subset \Sigma_+$ the set of simple roots in $\Sigma_+$. We have that the standard parabolic subalgebras of $\mathfrak{g}$ are parametrized by exclusion of simple (negative) roots: for any sub-collection $\Pi' \subset \Pi$ let
$$q_{\Pi'} = m \oplus a \oplus \bigoplus_{\beta \in \Sigma_+ \setminus \text{Span}(\Pi')} \mathfrak{g}^\beta.$$
Then $q_{\Pi'}$ is a Lie subalgebra of $\mathfrak{g}$ and all standard parabolic subalgebras of $\mathfrak{g}$ are of the form $q_{\Pi'}$ for some $\Pi' \subset \Pi$. (See [Kna, Proposition 7.76] and, in particular, the analysis of corresponding $\mathfrak{sl}(2, \mathbb{R})$-triples, [Kna, Lemma 7.73]).
Let $A$, $N$, and $K$ be the analytic subgroups of $G$ corresponding to $a$, $n$ and $\mathfrak{t}$. These are closed subgroups of $G$ and $G = KAN$ is the corresponding Iwasawa decomposition of $G$. As $G$ has finite center, $K$ is compact. Note that the Lie exponential $\exp : \mathfrak{g} \to G$ restricts to diffeomorphisms between $a$ and $A$ and $n$ and $N$. Fixing a basis for $a$, we identify $A = \exp(a) = \mathbb{R}^d$. Via this identification we often extend linear functionals on $a$ to $A$. We write $M = C_K(a)$ for the centralizer of $a$ in $K$. Then $P = MAN$ is the standard minimal parabolic subgroup. Since $M$ is compact, it follows that $P$ is amenable. A standard parabolic subgroup (relative to the choice of $\theta$ and $\Sigma_+$ above) is any closed subgroup $Q \subset G$ containing $P$. The Lie algebra of any standard parabolic subgroup $Q$ is a standard parabolic subalgebra and the correspondence between standard parabolic subgroups and subalgebras is 1-1.
We say two restricted roots $\beta, \tilde{\beta} \in \Sigma$ are coarsely equivalent if there is some $c > 0$ with
$$\tilde{\beta} = c\beta.$$
Note that $c$ takes values only in $\{1, 2, 4, 8\}$ and this occurs only if the root system $\Sigma$ has a factor of type $BC$. Let $\hat{\Sigma}$ denote the set of coarse restricted roots; that is, the set of coarse equivalence classes of $\Sigma$. Note that for $\xi \in \hat{\Sigma}$, $\mathfrak{g}^\xi := \bigoplus_{\beta \in \xi} \mathfrak{g}^{\beta}$ is a nilpotent subalgebra and the Lie exponential restricts to a diffeomorphism between $\mathfrak{g}^\xi$ and the corresponding analytic subgroup which we denote by $G^\xi$.
Let $\mathfrak{q}$ denote a standard parabolic subalgebra of $\mathfrak{g}$. Observe that if $\mathfrak{g}^\beta \subset \mathfrak{q}$ for some $\beta \in \Sigma$ then, from the structure of parabolic subalgebras, $\mathfrak{g}^\xi \subset \mathfrak{q}$ where $\xi \in \hat{\Sigma}$ is the coarse restricted root containing $\beta$. A standard parabolic (proper) subalgebra $\mathfrak{q}$ is maximal if there is no subalgebra $\mathfrak{q}'$ with $\mathfrak{q} \subsetneq \mathfrak{q}' \subsetneq \mathfrak{g}$. Note that maximal standard parabolic subalgebras are of the form $\mathfrak{q}_{\Pi, \beta}$ for some $\beta \in \Pi$.
1.3. Resonant codimension and related combinatorial numbers. Given a standard parabolic subalgebra $\mathfrak{q}$ define the resonant codimension of $\mathfrak{q}$ to be the cardinality of the set
$$\{\xi \in \hat{\Sigma} \mid \mathfrak{g}^\xi \not\subset \mathfrak{q}\}.$$
Given $G$ as above we define a combinatorial number $r(G)$ as follows.
**Definition 1.2.** The minimal resonant codimension of $\mathfrak{g}$, denoted $r(\mathfrak{g})$, is defined to be the minimal value of the resonant codimension of $\mathfrak{q}$ as $\mathfrak{q}$ varies over all (maximal) proper parabolic subalgebras of $\mathfrak{g}$.
**Example 1.3.** We compute $r(\mathfrak{g})$ for a number of classical real simple Lie algebras as well as simple real Lie algebras with restricted root systems of exceptional type. Given a simple real Lie algebra $\mathfrak{g}$ the number $r(\mathfrak{g})$ is determined purely by the restricted root system. In particular, we have the following.
**Type $A_n$:** $r(\mathfrak{g}) = n$. This includes $\mathfrak{sl}(n + 1, \mathbb{R})$, $\mathfrak{sl}(n + 1, \mathbb{C})$, $\mathfrak{sl}(n + 1, \mathbb{H})$.
**Type $B_n$, $C_n$, and $(BC)_n$:** $r(\mathfrak{g}) = 2n - 1$. This includes $\mathfrak{sp}(n, \mathbb{R})$, $\mathfrak{so}(n, m)$ for $n < m$, and $\mathfrak{su}(n, m)$ and $\mathfrak{sp}(n, m)$ for $n \leq m$.
**Type $D_n$:** $r(\mathfrak{g}) = 2n - 2$ for $n \geq 4$. This includes $\mathfrak{so}(n, n)$ for $n \geq 4$.
**Type $E_6$:** $r(\mathfrak{g}) = 16$.
**Type $E_7$:** $r(\mathfrak{g}) = 27$.
**Type $E_8$:** $r(\mathfrak{g}) = 57$.
**Type $F_4$:** $r(\mathfrak{g}) = 15$.
**Type $G_2$:** $r(\mathfrak{g}) = 5$.
In all classical root systems $A_n, B_n, C_n, (BC)_n$, and $D_n$ the number $r(g)$ corresponds to the parabolic obtained by omitting the left-most root in the standard Dynkin diagrams. Exceptional root systems are checked by hand.
Note that if $g$ is non-simple then $r(g)$ is $\min\{r(g_i) : 1 \leq i \leq n\}$ where $g_i$ are the simple non-compact factors of $g$. We write $r(G) = r(g)$. Note that our number $r(G)$ grows with the rank of $G$ but not with the dimension of the minimal algebraic actions. In particular, we only obtain the optimal expected results in the case that $G$ is split.
We define a second number $m(g)$ associated to the Lie algebra $g$ of $G$.
**Definition 1.4.** Given a simple Lie algebra $g$ of R-rank at least 2, define $m(g)$ to be the minimal value of the resonant codimension of $g$ as $q$ varies over all proper parabolic subalgebras $q$ of the form $q_{\Pi_\sim \{\alpha_i, \alpha_j\}}$ where $\alpha_i \neq \alpha_j$ are simple roots in $\Pi$. If $g$ has rank 1, let $m(g) = 1$. If $g = \oplus g_i$ is semisimple, take $m(g)$ to be the minimum of $m(g_i)$ over all non-compact, simple factors $g_i$ of $g$.
As before, write $m(G) = m(g)$.
**Example 1.5.** Again, we compute the number $m(g)$ for a number of classical, simple real Lie algebras as well as simple real Lie algebras with restricted root systems of exceptional type. As before, given a simple real Lie algebra $g$ the number $m(g)$ is determined only by the restricted root system.
- **Type $A_n$:** $m(g) = 2n - 1$.
- **Type $B_n, C_n,$ and $(BC)_n$:** $m(g) = 4n - 4$.
- **Type $D_n$:** $m(g) = 9$ for $n = 4$; $m(g) = 4n - 6$ for $n \geq 5$.
- **Type $E_6$:** $m(g) = 24$.
- **Type $E_7$:** $m(g) = 43$.
- **Type $E_8$:** $m(g) = 84$.
- **Type $F_4$:** $m(g) = 20$.
- **Type $G_2$:** $m(g) = 6$.
In all classical root systems except $D_4$, the number $m(g)$ corresponds to the parabolic subalgebra obtained by omitting the two left-most roots in the standard Dynkin diagrams. In $D_4$, the number $m(g)$ corresponds to omitting two commuting roots. Exceptional root systems are checked by hand.
As before, write $m(G) = m(g)$.
### 1.4. Statement of results
Let $G$ be as introduced above and let $\Gamma \subset G$ be a higher-rank lattice. Recall that $\alpha$ denotes a right action of $\Gamma$ on a compact, boundaryless manifold $M$ by $C^{1+\beta}$ diffeomorphisms.
#### 1.4.1. Existence of invariant measures in low dimensions
Our first main result establishes the existence of an $\alpha$-invariant measure if the dimension $M$ is sufficiently small relative to $r(G)$.
**Theorem 1.6.** Let $M$ be a manifold with $\dim(M) < r(G)$. Then for any $C^{1+\beta}$ action $\alpha$ of $\Gamma$ on $M$ there exists an $\alpha$-invariant Borel probability measure.
We remark in the case that $\Gamma$ is cocompact, Theorem 1.6 in an immediate corollary of the main result of [BFH] where Zimmer’s conjecture is verified for actions of compact lattices on manifolds of dimension less than $r(G)$. The proof of the main result of [BFH] uses the proof of Theorem 1.6, namely the key observation in Proposition 5.1 below. We note also that Theorem 1.6 applies to nonuniform lattices whereas Zimmer’s conjecture has
yet to be verified for nonuniform lattices. We do not assert any regularity of the measure in Theorem 1.6. In particular, the ergodic components of the measure are expected to be supported on finite sets as such actions are expected to be trivial. Theorems A and B follow directly from the main results in [FH] and Theorem 1.6.
1.4.2. Finite extensions of projective factors in critical dimension. In the case where \( \dim M = r(G) \), we recall as a model the standard right action of \( \Gamma \subset \text{SL}(n + 1, \mathbb{R}) \) on \( \mathbb{R}^n \). Note that \( \mathbb{R}^n \) has the structure of \( Q/\text{SL}(n + 1, \mathbb{R}) \) for a (maximal) parabolic subgroup \( Q \subset \text{SL}(n + 1, \mathbb{R}) \).
**Theorem 1.7.** Let \( M \) be a manifold with \( \dim(M) = r(G) \). Then given any \( C^{1+\beta} \) action \( \alpha \) of \( \Gamma \) on \( M \) either
(a) there exists an \( \alpha \)-invariant Borel probability measure on \( M \); or
(b) there exists an \( \alpha \)-quasi-invariant Borel probability measure \( \mu \) on \( M \) and a maximal parabolic subgroup \( Q \subset G \) such that the action \( \alpha \) of \( \Gamma \) on \( (M, \mu) \) is measurable conjugate to a finite extension of the standard right action of \( \Gamma \) on \( (Q \setminus G, m) \) where \( m \) is of Lebesgue class.
Motivated by the above theorem, we extend Conjecture 1.1.
**Conjecture 1.8.** Let \( M \) be a manifold with \( \dim(M) = r(G) \). Given any sufficiently smooth action \( \alpha \) of \( \Gamma \) on \( M \) either
(a) there exists an \( \alpha \)-invariant Borel probability measure on \( M \); or
(b) there is a maximal parabolic subgroup \( Q \subset G \) such that \( M \) is diffeomorphic to a finite cover of \( Q \setminus G \); moreover, the action \( \alpha \) is smoothly conjugate to a lift of the standard right-action of \( \Gamma \) on \( Q \setminus G \).
1.4.3. Projective factors in intermediate dimensions. Let \( (X, \nu) \) and \( (Z, \mu) \) be standard measure spaces and suppose \( \Gamma \) acts measureably on both \( X \) and \( Z \) (on the right) and preserves the measure classes of \( \nu \) and \( \mu \) respectively. Let \( (Y, \eta) \) be a standard measure space and write \( \text{Aut}(Y, \eta) \) for the group of invertible, measure-preserving transformations of \( (Y, \eta) \). Let \( \alpha \) and \( \rho \) denote, respectively, the actions of \( \Gamma \) on \( (Z, \mu) \) and \( (X, \nu) \).
**Definition 1.9.** We say \( \alpha \) is a relatively measure-preserving extension (modeled on \( (Y, \eta) \)) of \( \rho \) if there are
1. a measurable cocycle \( \psi \colon \Gamma \times (X, \nu) \to \text{Aut}(Y, \eta) \) over \( \rho \), and
2. an isomorphism of measure spaces \( \Phi \colon (Z, \mu) \to (X \times Y, \nu \times \eta) \)
such that \( \Phi \) intertwines \( \alpha \) and the skew action defined by \( \psi \): if \( \Phi(z) = (x, y) \) then
\[
\Phi(\alpha(\gamma)(z)) = (\rho(\gamma)(x), \psi(\gamma, x)(y))
\]
**Theorem 1.10.** Let \( M \) be a manifold with \( \dim(M) \leq m(G) \). Then given any \( C^{1+\beta} \) action \( \alpha \) of \( \Gamma \) on \( M \) there is an \( \alpha \)-quasi-invariant Borel probability measure \( \mu \) on \( M \), a standard parabolic subgroup \( Q \), and a Lebesgue space \( (Y, \eta) \) such that the action \( \alpha \) on \( (M, \mu) \) is a relatively measure-preserving extension (modeled on \( (Y, \eta) \)) of the standard right action of \( \Gamma \) on \( (Q \setminus G, m) \).
Note in the above theorem, if \( Q = G \) it follows that \( \mu \) is \( \alpha \)-invariant. As discussed above, the result in Theorem 1.10 should be compared to results of Nevo and Zimmer, particularly [NZ1, NZ2].
2. Suspension construction and its properties
We construct an auxiliary space on which the action \( \alpha \) of \( \Gamma \) on \( M \) embeds as a Poincaré section for an associated \( G \)-action. On the product \( G \times M \) consider the right \( G \)-action
\[
(g, x) \cdot \gamma = (g\alpha(\gamma)(x))
\]
and the left \( G \)-action
\[
a \cdot (g, x) = (ag, x).
\]
Define the quotient manifold \( M^\alpha := G \times M / \Gamma \). As the \( G \)-action on \( G \times M \) commutes with the \( \Gamma \)-action, we have an induced left \( G \)-action on \( M^\alpha \). We denote this action by \( \alpha \).
We write \( \pi : M^\alpha \to G / \Gamma \) for the natural projection map. Note that \( M^\alpha \) has the structure of a fiber bundle over \( G / \Gamma \) induced by the map \( \pi \) with fibers diffeomorphic to \( M \). As the action of \( \alpha \) is by \( C^{1+\beta} \) diffeomorphism, \( M^\alpha \) is naturally a \( C^{1+\beta} \) manifold. Equip \( M^\alpha \) with a \( C^\infty \) structure compatible with the \( C^{1+\beta} \) structure.
Note that the action \( \tilde{\alpha} \) of \( G \) on \( M^\alpha \) preserves two transverse distributions \( E^F \) and \( E^G \) where \( E^F = \ker(D\pi) \) and \( E^G \) is tangent to the local \( G \)-orbits on \( M^\alpha \). Furthermore, these distributions integrate to \( C^{1+\beta} \) foliations of \( M^\alpha \).
We first observe
**Claim 2.1.** There exists an \( \alpha \)-invariant Borel probability measure on \( M \) if and only if there exists an \( \tilde{\alpha} \)-invariant Borel probability measure on \( M^\alpha \).
That an \( \alpha \)-invariant measure on \( M \) induces an \( \tilde{\alpha} \)-invariant measure on \( M^\alpha \) is standard. For the reverse implication, see, for instance, [NZ1, Lemma 6.1]. Note that any \( \tilde{\alpha} \)-invariant measure on \( M^\alpha \) projects under \( \pi \) to the Haar measure on \( G / \Gamma \).
As the suspension space \( M^\alpha \) is non-compact in the case that \( \Gamma \) is non-uniform, some care is needed when applying tools from smooth ergodic theory to the \( G \)-action on \( M^\alpha \). Indeed, although the non-compactness comes from the homogeneous factor, care is needed in order to control the fiber-wise dynamics as the corresponding \( C^1 \)- and \( C^{1+\beta} \)-norms of the fiberwise dynamics need not be bounded.
Below, we use the quasi-isometry between the Riemannian and word metrics on \( \Gamma \) established in [LMR] to control the degeneration of the fiber-wise dynamics. We follow the approach of [BRH] and construct dynamical charts relative to which the tools of classical smooth ergodic theory may be applied. The remainder of this section is devoted to constructing a Riemannian metric on \( TM^\alpha \), corresponding distance function \( d \), and a family of dynamical charts.
The reader interested only in actions of cocompact lattices may skip the remainder of this section.
2.1. Construction of a fundamental domain and family of fiber metrics. Recall our standing assumptions on the Lie group \( G \) and the lattice \( \Gamma \). A set \( D \subseteq G \) is a fundamental domain for \( \Gamma \) if \( \bigcup_{\gamma \in \Gamma} D\gamma = G \) and if the natural map \( G \to G / \Gamma \) is one-to-one on \( D \). A Borel set \( D \subseteq G \) is almost-open if the interior of \( D \) has full measure in the closure of \( D \). A set \( S \subseteq G \) is a fundamental set if \( \bigcup_{\gamma \in \Gamma} S\gamma = G \) and the set \( \{ \gamma : S\gamma \cap S \neq \emptyset \} \) is finite. The injectivity radius \( r^\Gamma(g) \) of \( \Gamma \) at a point \( g \in G \) is the largest \( 0 < r \leq 1 \) such that the map \( g : \Gamma \to G / \Gamma \) given by \( X \mapsto \exp_g(X)g\Gamma \) is injective on
\[
\{ X \in \mathfrak{g} : \|X\| < r \}.
\]
We write
\[
V_r(g) := \{ \exp_g(X)g\Gamma : \|X\| \leq r \}
\]
for the remainder.
Our goal below is to build on $TM$ a family of continuous Riemannian metrics $\langle \cdot, \cdot \rangle_g$, parameterized by $g \in G$, and an almost-open, Borel fundamental domain $D \subset G$ for $\Gamma$ such that
1. the family of metrics $\langle \cdot, \cdot \rangle_g$ depends continuously on $g \in G$;
2. the family $\langle \cdot, \cdot \rangle_g$ is $\Gamma$-equivariant: given $\gamma \in \Gamma$ and $v, w \in T_x M$
$$\langle v, w \rangle_g = \langle D_x \alpha(\gamma)v, D_x \alpha(\gamma)w \rangle_{g \gamma};$$
3. writing
$$V = \bigcup_{g \in D} V_{r(g)}(g),$$
the family $\langle \cdot, \cdot \rangle_g$ is uniformly comparable on $V$: there is a $C > 0$ so that for all
g, \overline{g} \in V, x \in M, and v, w \in T_x M
$$\langle v, w \rangle_g \leq C \langle v, w \rangle_{\overline{g}};$$
4. for every $p \geq 1$ the function $g \mapsto d_G(e, g)$ is $L^p$ on $D$ with respect to the Haar measure where $d_G(\cdot, \cdot)$ is the right-invariant metric on $G$.
Note that given a finite-index subgroup $\Gamma' \subset \Gamma$, a fundamental domain $D'$ for $\Gamma'$ and a $\Gamma'$-equivariant family of metrics which satisfy (1)-(4) above for $\Gamma'$, then we can choose a fundamental domain $D \subset D'$ for $\Gamma$ and construct a $\Gamma$-equivariant family of metrics satisfying (1)-(4) for $\Gamma$. Below, we will pass to a finite-index subgroup $\Gamma' \subset \Gamma$ and construct such a domain and family of metrics for $\Gamma'$.
To build such a family, first note that by quotienting by any compact factors and the center of $G$, we obtain a surjective homomorphism with compact kernel $\Psi: G \to \overline{G}$ where $\overline{G}$ is semisimple, without any compact factors, and has trivial center. Moreover, the image $\overline{\Gamma} := \Psi(\Gamma)$ is a lattice in $\overline{G}$. From the Margulis Arithmeticity Theorem [Mar], it follows that there is a semisimple, linear algebraic group $H$ such that setting $H = H(\mathbb{R})^0$ there is a surjective homomorphism $\Phi: H^0 \to \overline{G}$ with compact kernel such that
$$\overline{\Gamma} \cap \Phi(H(\mathbb{Z}))$$
has finite index in $\overline{\Gamma}$. Let $\hat{\Gamma} = \Phi^{-1}(\Gamma) \cap H(\mathbb{Z}) \cap H$. Then $\hat{\Gamma}$ has finite index in $H(\mathbb{Z})$ and is hence arithmetic. Replacing $\Gamma$ and $\hat{\Gamma}$ with finite index subgroups, we may assume that $\hat{\Gamma}$ is torsion-free and neat (see [BJ] for definition), and that $\Gamma$ and $\hat{\Gamma}$ map surjectively onto $\overline{\Gamma}$. Let $N$ denote the kernel of $\Phi: \Gamma \to \overline{\Gamma}$. Note that $N$ is finite.
We will use $\overline{\Gamma}$ to build a family of metrics and domain $D$. Note that while $\overline{\Gamma}$ may not act on $M$, $\hat{\Gamma}$ acts on the space of $N$-invariant Riemannian metrics on $TM$.
2.1.1. Compactification of $H/\hat{\Gamma}$. Let $\hat{K} \subset H$ be a maximal compact subgroup. We may assume that $\hat{K}$ contains the kernel of the map $\Phi: H \to \overline{G}$. Let $X$ denote the symmetric space $\hat{K} \backslash H$. Following [BJ, III, Chapter 9], write $\overline{X}^{BS}$ for the Borel-Serre partial compactification of $X$. $\overline{X}^{BS}$ has the structure of a real-analytic manifold with corners. The action of $\hat{\Gamma}$ on $X$ extends to a continuous, proper action on $\overline{X}^{BS}$. Moreover the quotient $\overline{X}^{BS}/\hat{\Gamma}$ is a compact, Hausdorff space. Furthermore (having taken $\hat{\Gamma}$ to be neat) the quotient $\overline{X}^{BS}/\hat{\Gamma}$ has the structure of a real-analytic manifold with corners.
2.1.2. Parameterized families of metrics. As $\hat{\Gamma}$ maps surjectively onto $\overline{\Gamma}$, it follows that $\hat{\Gamma}$ acts on the space of $N$-invariant Riemannian metrics on $M$. Consider $\overline{X}^{BS} \times TM$. As the manifold with corners $\overline{X}^{BS}/\hat{\Gamma}$ admits partitions of unity, by selecting a $\hat{\Gamma}$-equivariant partition of unity on $\overline{X}^{BS}$ subordinate to a cover by sets of the form $V_{r(g)}(g)$, given any
fixed $N$-invariant metric on $TM$ we may construct a $\hat{\Gamma}$-equivariant, continuous family of $N$-invariant, Hölder continuous Riemannian metrics $\langle \cdot, \cdot \rangle_x$ on $TM$ parametrized by points of $x \in \overline{X}^{BS}$. As $\hat{K}$ contains the kernel of $\Phi$, given $\overline{g} \in \overline{G}$ we associate a Riemannian metric on $TM$ by
$$\langle \cdot, \cdot \rangle_{\overline{g}} = \langle \cdot, \cdot \rangle_{\Phi^{-1}(\overline{g})}.$$
The family $\{\langle \cdot, \cdot \rangle_{\overline{g}} : \overline{g} \in \overline{G}\}$ is continuous in $\overline{g}$ and is $\hat{\Gamma}$-equivariant. Finally, to $g \in G$ we associate a Riemannian metric on $TM$ by
$$\langle \cdot, \cdot \rangle_g = \langle \cdot, \cdot \rangle_{\Psi(g)}.$$
The family $\{\langle \cdot, \cdot \rangle_g : g \in G\}$ is continuous in $g$ and $\Gamma$-equivariant.
2.1.3. Construction of fundamental domain and verification of its properties. Let $S \subset H$ be a Siegel fundamental set for $\hat{\Gamma}$ in $H$ containing $e$; that is (see for instance [Mar, VIII.1])
1. $\bigcup_{\gamma \in \hat{\Gamma}} S\gamma = H$;
2. the set $\{\gamma : S\gamma \cap S \neq \emptyset\}$ is finite;
3. the function $g \mapsto d_H(e, g)$ is $L^p$ on $S$ with respect to the Haar measure for every $1 \leq p < \infty$;
4. $S$ is left $K$-invariant.
We say an element of $h \in S$ is well-positioned in $S$ if, denoting injectivity radius of $H/\hat{\Gamma}$ at $h \in H$ by $r^i(h)$, we have that $V_{r^i(h)}(h) \subset S$. Enlarging the cusp parameters of $S$, the set $S$ can be chosen so that
5. the well-positioned elements of $S$ form a fundamental set for $\hat{\Gamma}$.
We will moreover assume $e \in S$ is well-positioned.
An additional property of $S$ that follows from the construction of the Borel-Serre compactifications is that $\hat{K}\backslash S$ has compact closure in $\overline{X}^{BS}$. It follows that the set $\{\langle \cdot, \cdot \rangle_h : h \in S\}$ is a uniformly comparable family of metrics.
Let $S' \subset S$ denote the set of well-positioned element of $S$. Write $S \subset \overline{G}$ for $S = \Phi(S')$. Then $S'$ is fundamental set for $\overline{\Gamma}$ in $\overline{G}$. Moreover, as $S$ is $\hat{K}$-saturated we have that $\Phi^{-1}(S') \subset S$. It follows that $\{\langle \cdot, \cdot \rangle_{\overline{g}} : \overline{g} \in S\}$ is a uniformly comparable family of metrics.
Finally let $\tilde{S} = \Psi^{-1}S$. Then
1. $\tilde{S}$ is a fundamental set for $\Gamma$ in $G$;
2. the family $\{\langle \cdot, \cdot \rangle_g : g \in \tilde{S}\}$ is a uniformly bounded family of metrics;
3. the family $\{\langle \cdot, \cdot \rangle_g : g \in G\}$ is $\Gamma$-equivariant and continuous;
4. as all quotients and extensions are by compact kernels, it follows that $g \mapsto d_G(e, g)$ is $L^p$ on $\tilde{S}$ with respect to Haar measure;
5. for $g \in \tilde{S}$, the image $\Psi(V_{r^i(g)}(g))$ is contained in $\Phi(S)$.
Let $D \subset \tilde{S}$ be an almost-open, Borel, fundamental domain for $\Gamma$ in $G$ containing $e$. Then the desired properties of the family $\langle \cdot, \cdot \rangle_g$ and the fundamental domain $D$ hold.
2.1.4. Induced distance on $M^\alpha$. Using the $\Gamma$-equivariant family of metrics $\{\langle \cdot, \cdot \rangle_g : g \in G\}$ constructed above and using the right invariant metric on $G$, we endow $G \times M$ with a continuous Riemannian metric such that $\Gamma$ acts by isometries. This induces a Riemannian metric on $TM^\alpha$ and corresponding distance function $d(\cdot, \cdot)$ on $M^\alpha$.
2.2. Some estimates. Equip $M$ with any $C^\infty$ Riemannian metric; by compactness, all estimates are independent of the choice of metric. Let $\exp_x : T_x M \to M$ be the Riemannian exponential map at $x$ and fix $r_0 \leq 1$ to be smaller than the injectivity radius of $M$.
Write $B_x(r) \subset T_xM$ for the norm ball $B_x(r) = \{ v \in T_xM : \|v\| < r \}$. Given a diffeomorphism $f : M \to M$ let
\[ \tilde{f}_x : U_{x,f} \subset B_x(r_0) \subset T_xM \to B_{f(x)}(r_0) \subset T_{f(x)}M \]
be the diffeomorphism defined by
\[ \tilde{f}_x := \exp^{-1}_{f} \circ f \circ \exp_x \]
on the maximal domain $U_{x,f}$ on which it is defined. Given $U \subset U_{x,f}$ define the local $C^1$ and Hölder norms of $\tilde{f}_x|_U : U \to B_x(r_0) \subset T_xM$ to be
\[
\|D\tilde{f}_x\|_U = \sup_{v \in U} \|D_v\tilde{f}_x\|, \quad \text{Hö}^\beta_U(D\tilde{f}_x) := \sup_{v \neq w \in U} \frac{\|D_v\tilde{f}_x - D_w\tilde{f}_x\|}{\|v - w\|^\beta}.
\]
If $f : M \to M$ is $C^{1+\beta}$, define
\begin{enumerate}
\item $\|Df\| := \sup_{x \in M} \|D\tilde{f}_x\|_{U_{x,f}}$ and
\item $\text{Hö}^\beta(f) := \sup_{x \in M} \text{Hö}^\beta_U(D\tilde{f}_x)$.
\end{enumerate}
Compactness of $M$ ensures $\|Df\|$ and Hö$^\beta(f)$ are finite.
We have the following elementary estimate.
**Claim 2.2.** Let $f, g \in \text{Diff}^{1+\beta}(M)$. Given $x \in M$ and $U \subset U_{x,g} \subset T_xM$ such that
\[ \tilde{g}_x(U) \subset U_{g(x),f} \]
we have
\[ \text{Hö}^\beta(U(D(f \circ g)_x)) \leq \|Df\| \text{Hö}^\beta(Dg) + \|Dg\|^{1+\beta} \text{Hö}^\beta(Df). \]
**Proof.** For $v, u \in U$ and $\xi$ with $\|\xi\| = 1$
\[
\|D_v(f \circ g)_x\| - D_u(f \circ g)_x\| \xi = \|D_{\tilde{g}_x(v)}(f \circ g)_{\tilde{f}_x(D_v\tilde{g}_x\xi)} - D_{\tilde{g}_x(u)}(f \circ g)_{\tilde{f}_x(D_u\tilde{g}_x\xi)}\|
\leq \|D_{\tilde{g}_x(v)}(f \circ g)_{\tilde{f}_x(D_v\tilde{g}_x\xi)} - D_{\tilde{g}_x(u)}(f \circ g)_{\tilde{f}_x(D_u\tilde{g}_x\xi)}\|
+ \|D_{\tilde{g}_x(v)}(f \circ g)_{\tilde{f}_x(D_u\tilde{g}_x\xi)} - D_{\tilde{g}_x(u)}(f \circ g)_{\tilde{f}_x(D_u\tilde{g}_x\xi)}\|
\leq \|Df\| \text{Hö}^\beta(U(D\tilde{g}_x)\tilde{f}_x)\|D_u\tilde{g}_x\| + \|Dg\| \text{Hö}^\beta(U(D\tilde{g}_x)\tilde{f}_x)\|D_u\tilde{g}_x\|
\leq \|Df\| \text{Hö}^\beta(U(D\tilde{g}_x)\tilde{f}_x)\|D_u\tilde{g}_x\| + \|Dg\| \text{Hö}^\beta(U(D\tilde{g}_x)\tilde{f}_x)\|D_u\tilde{g}_x\|
\leq \|Df\| \text{Hö}^\beta(U(D\tilde{g}_x)\tilde{f}_x)\|D_u\tilde{g}_x\| + \|Dg\|^{1+\beta} \text{Hö}^\beta(U(D\tilde{g}_x)\tilde{f}_x)\|D_u\tilde{g}_x\|. \quad \square
\]
In particular, we have the following.
**Claim 2.3.** Let $g_i \in \text{Diff}^{1+\beta}(M)$, $i = \{1, 2, \ldots, \ell\}$ and fix $C$ with $\|Dg_i\| \leq C$ and $\text{Hö}^\beta(Dg_i) \leq C$. Given $n \geq 0$ and
\[ U \subset B_x(C^{-n}T_0) \subset T_xM \]
with $h = g_{i_1} \circ \cdots \circ g_{i_n}$ we have
\begin{enumerate}
\item $\|Dh_x\| \leq C^n$ and
\item $\text{Hö}^\beta(U(Dh_x)) \leq nC^n(1+\beta)$ for every $x$.
\end{enumerate}
### 2.3. Construction of dynamical charts.
Let $D \subset G$ be the almost open, fundamental domain for $\Gamma$ constructed in Section 2.1. In the sequel, we often use the measurable parameterization $D \times M$ of $M^\alpha = (G \times M)/\Gamma$.
Fix a globally defined, Borel family of isometric identifications $\tau_x : T_xM \to \mathbb{R}^n$. With respect to any fixed background $C^\infty$ Riemannian metric on $M$, let $\exp_x : T_xM \to M$
denotes the Riemannian exponential map at \( x \) and \( r_0 \) denote the injectivity radius of \( M \). Let \( \mathbb{R}^k = g \oplus \mathbb{R}^n \) be equipped with the product Euclidean metric where \( k = \dim G + \dim M \).
Given \( p = (g, x) \in D \times M \) let \( \rho(g) = \frac{1}{2} \min \{ r^G(g), r_0 \} \) and let
\[
\phi_p : \mathbb{R}^k(\rho(g)) \to M^\alpha
\]
be the natural embedding
\[
\phi_p : (X, v) \mapsto \left( \exp(X)g, \exp_x(r_x^{-1}v) \right) / \Gamma
\]
where we write \( \mathbb{R}^k(r) := \{ v \in \mathbb{R}^k : \|v\| < r \} \). We immediately verify that, relative to the induced metric in 2.1.4, the charts \( \phi_{(g,x)} \) are \( C^1 \) with \( \| D\phi_{(g,x)} \| \) uniformly bounded; in particular relative to the distance function \( d \) in 2.1.4 the charts are uniformly bi-Lipschitz.
As the injectivity radius \( r^G(g\Gamma) \) is comparable to the distance from \( g\Gamma \) to a fixed base point \( x_0 \in G/\Gamma \) we have that \( g \mapsto -\log(\rho(g)) \) is \( L^q \) with respect to the Haar measure for all \( 1 \leq q < \infty \).
Recall we let \( A \) be the analytic subgroup of \( G \) corresponding to \( a \). Fixing a basis for \( a \), via the parameterization \( \exp : a \to A \) we identify \( A \) with \( \mathbb{R}^d \) where \( d \geq 2 \) is the rank of \( G \).
Below, we consider an arbitrary lattice \( \mathbb{Z}^d \subset A \) and fix a finite, symmetric, generating set \( F = \{ s_j : 1 \leq j \leq m \} \) for \( \mathbb{Z}^d \).
Following the notation of [BRH], we let \( U = U_0 = \Lambda = D \times M = M^\alpha \) for any such \( \mathbb{Z}^d \) and \( F \).
In the sequel, we will be concerned with \( A \)-invariant measures \( \mu \) on \( M^\alpha \) that project to the Haar measure on \( G/\Gamma \). Note in the case that \( G \) has compact factors, the Haar measure on \( G/\Gamma \) need not be \( A \)-ergodic. However, from the pointwise ergodic theorem we have the following.
**Claim 2.4. Almost every \( A \)-ergodic component of the Haar measure on \( G/\Gamma \) is \( G' \)-invariant.**
Indeed, almost every \( A \)-ergodic component contains \( G^\xi \) for every nonzero coarse root \( \xi \in \Sigma \) and the \( G^\xi \) generate all of \( G' \). Similarly, every \( G' \)-invariant measure on \( G/\Gamma \) is an \( A \)-ergodic component of the Haar measure on \( G/\Gamma \).
**Proposition 2.5.** Let \( \mu \) be an \( A \)-invariant probability measure on \( M^\alpha \) projecting to a \( G' \)-invariant measure on \( G/\Gamma \). Then for any lattice \( \mathbb{Z}^d \subset A \cong \mathbb{R}^d \) and any finite, symmetric, generating set \( F = \{ s_j : 1 \leq j \leq \ell \} \) for \( \mathbb{Z}^d \) the standing hypothesis of [BRH, Section 3.1] hold relative to the charts \( \{ \phi_p : p \in M^\alpha \} \) above.
That is, there are measurable functions \( r : D \to (0, 1] \) and \( C : D \to [1, \infty) \) and a constant \( L \) with
\[
r(g) \leq \rho(g), \quad -\log r(g) \in L^q(D), \quad \text{and} \quad \log C(g) \in L^q(D) \quad \text{for all} \quad 1 \leq q < \infty
\]
such that, writing
\[
r(p) = r(g), \quad \rho(p) = \rho(g), \quad C(p) = C(g)
\]
for \( p = (g, x) \in M^\alpha = D \times M \), we have
(H1) \( \phi_p : \mathbb{R}^k(\rho(p)) \to M^\alpha \) is a \( C^1 \) diffeomorphism onto its image with \( \phi_p(0) = p \);
(H2) \( \| D\phi_p \| \leq L \) and \( \| D\phi_p^{-1} \| \leq L \); in particular \( \phi_p : \mathbb{R}^k(\rho(p)) \to (U, d) \) is a bi-Lipschitz embedding with \( L^{-1} \leq \text{Lip}(\phi_p) \leq L \).
Moreover, for each \( m \in F \), setting \( f(\cdot) = \tilde{\alpha}(m, \cdot) \) we have for \( p \in M^\alpha \) that
(H3) the map
\[
\tilde{f}_p := \phi_p^{-1} \circ f \circ \phi_p
\]
is well defined on \( \mathbb{R}^k(r(p)) \) with range contained in \( \mathbb{R}^k(\rho(f(p))) \):
(H4) \( \tilde{f}^q_p : \mathbb{R}^k(r(p)) \to \mathbb{R}^k(\rho(f(p))) \) is uniformly \( C^{1+\beta} \) with
\[
\| \tilde{f}^q_p \|_{1+\beta} \leq C(p)
\]
(H5) for every \( n \in \mathbb{Z}^d \), \( p \mapsto \log^+ \| D_p \hat{\alpha}(n) \| \) \( \in L^q(\mu) \) for any \( 1 \leq q < \infty \); in particular \( p \mapsto \log^+ \| D_p \hat{\alpha}(n) \| \) \( \in L^{d,1}(\mu) \).
Here \( L^{d,1}(\mu) \) is the Lorentz integrability space (see [Lor]). We have \( L^p(\mu) \subset L^{d,1}(\mu) \) for any \( p > d \). The assertion that \( p \mapsto \log^+ \| D_p \hat{\alpha}(n) \| \) \( \in L^{d,1}(\mu) \) guarantees the cocycle satisfies the hypotheses of the higher-rank multiplicative ergodic theorem. As \( -\log \rho, -\log r \) and \( \log C \) are \( L^d \) on the domain \( D \), it follows that, in the terminology of [BRH], they are slowly growing functions over the action of \( \mathbb{Z}^d \).
**Proof.** Fix a finite, symmetric generating set \( S = \{ \gamma_i : 1 \leq i \leq \ell \} \) for \( \Gamma \). For each \( 1 \leq i \leq \ell \) take \( g_i = \alpha(\gamma_i) : M \to M \) and set \( \tilde{C} > 1 \) with
1. \( \| Dg_i \| \leq \tilde{C} \), and
2. \( \text{Hö}^3(Dg_i) \leq \tilde{C} \).
Let \( S = \{ \gamma_i \} \) be a fixed finite generating set for the lattice \( \Gamma \subset G \). Let \( d_{\text{word}} \) denote the corresponding word metric on \( \Gamma \). Let \( d_G \) denote the distance on \( G \) induced by the right-invariant metric on \( G \). Note that \( d_G \) restricts to a metric on \( \Gamma \subset G \). It follows from [LMR] that if \( \Gamma \) is a higher-rank lattice as introduced in Section 1, the metrics \( d_{\text{word}} \) and \( d_G \) are quasi-isometrically equivalent: there are \( A > 1 \) and \( B > 0 \) such that for all \( \gamma, \hat{\gamma} \in \Gamma \) we have
\[
A^{-1}d_G(\gamma, \hat{\gamma}) - B \leq d_{\text{word}}(\gamma, \hat{\gamma}) \leq Ad_G(\gamma, \hat{\gamma}) + B.
\]
Now consider any lattice \( \mathbb{Z}^d \) in \( A \simeq \mathbb{R}^d \) and finite symmetric generating set \( F \). Given \( g \in D \) and \( s_j \in F \) let \( \gamma_j(g) \) be such that \( s_j g \in D\gamma_j(g) \). Define
\[
N(g) = \max_{s_j \in F} \{ d_{\text{word}}(e, \gamma_j(g)) \}.
\]
We have
\[
d_{\text{word}}(e, \gamma_j(g)) \leq A \left[ d(e, g) + d(e, s_j) + d(e, s_j g(\gamma_j(g))^{-1}) \right] + B.
\]
Let \( \nu \) denote the image of \( \mu \) in \( G/\Gamma \) and naturally consider \( \nu \) as a measure on \( D \). Recall that \( C \) denotes the maximal compact normal subgroup of \( G \). We have that \( C \) and \( G' \) commute whence \( C \) acts transitively on the set of \( G' \)-ergodic components of the Haar measure. As \( g \mapsto d(e, g) \) is in \( L^q(D, \text{Haar}) \) for all \( 1 \leq q < \infty \) and as \( C \) has bounded diameter and acts transitively on \( G' \)-ergodic components, we have that \( g \mapsto d(e, g) \) is in \( L^q(D, \nu) \) for all \( 1 \leq q < \infty \). Also, as the map \( D \to D \) given by \( g \mapsto s_j g(\gamma_j(g))^{-1} \) preserves the Haar measure, it follows that \( g \mapsto N(g) \) is in \( L^q(D, \nu) \) for all \( 1 \leq q < \infty \).
We set \( r(g, x) = r(g) := \tilde{C}^{-N(g)} \rho(g) \). We have that \( 0 < r(g, x) \leq \rho(g) \) for every \( (g, x) \in D \times M \). Moreover, we have that
\[
\int_D (-\log(r(g, x)))^q \, d\mu(g, x) = \int_D (-\log(r(g)))^q \, d\nu(g) < \infty.
\]
Given \( s_j \in F \), let \( f = \hat{\alpha}(s_j) \). Write \( \tilde{f}_{(g, x)} : \mathbb{R}^k(r(g)) \to \mathbb{R}^k(\rho(f(g))) \) for
\[
\tilde{f}_{(g, x)} := \phi_{f(g, x)}^{-1} \circ f \circ \phi_{(g, x)}.
\]
(H3) then follows. From Claim 2.3 we have
\[
\| D\tilde{f}_{(g, x)} \| \leq \tilde{C}^{N(g)}, \quad \text{Hö}^3(D\tilde{f}_{(g, x)}) \leq N(g)\tilde{C}^{N(g)(1+\beta)}
\]
whence (H4) follows. Moreover, we have that the function
\[
(g, x) \mapsto \log \| D_0 \tilde{f}(g, x) \|
\]
is \(L^q(\mu)\) for every \(1 \leq q < \infty\). From the cocycle property, (H5) follows for all elements of the action. \(\square\)
3. LYAPUNOV EXONENTS, COARSE FOLIATIONS, AND CONDITIONAL ENTROPY
For this section we consider the restriction of the action \(\tilde{\alpha}\) on \(M^\alpha\) to \(A\). Take \(\mu\) to be an \(A\)-invariant probability measure on \(M^\alpha\). Let \(\nu = \pi_\alpha(\mu)\) be the projection of \(\mu\) to \(G/\Gamma\). In the case that \(\Gamma\) is not cocompact, assume the projection \(\nu\) is \(G'\)-invariant so that the charts in Section 2.3 satisfy properties (H1)–(H5) of Proposition 2.5.
3.1. Lyapunov exponent functionals. From the \(L^{d,1}\) integrability of (H5) of Proposition 2.5 it follows that the restriction to \(A\) of the derivative cocycle \(D\tilde{\alpha}\) on \((M^\alpha, \mu)\) satisfies the hypotheses of the Oseledec’s multiplicative ergodic theorem in every direction \(s \in \mathbb{R}^d\) (see (5) below). Moreover, we have uniform convergence along spheres guaranteed by the stronger conclusions of the higher-rank Oseledec’s multiplicative ergodic theorem.
Equip \(A \simeq \mathbb{R}^d\) with any norm \(| \cdot |\).
Theorem 3.1 (Higher-rank multiplicative ergodic theorem; [BRH, Theorem 2.4]). Let \(\mu\) be any \(A\)-invariant measure on \(M^\alpha\) satisfying (H5) of Proposition 2.5. Then there exist
1. a full measure, \(A\)-invariant subset \(\Lambda_0 \subset M^\alpha\);
2. an \(A\)-invariant measurable function \(r : \Lambda_0 \to \mathbb{N}\);
3. an \(A\)-invariant measurable family of linear functionals \(\lambda_i(p) : A \to \mathbb{R}\) for \(1 \leq i \leq r(p)\);
4. and a family of mutually transverse, \(D\tilde{\alpha}|_{\Lambda\text{-invariant}},\) measurable subbundles \(E_{\lambda_i} \subset TM^\alpha\) with \(T_pM^\alpha = \bigoplus_{i=1}^{r(p)} E_{\lambda_i}(p)\) for \(p \in \Lambda_0\)
such that
\[
\lim_{s \to \infty} \frac{\log \| D_p\tilde{\alpha}(s)(v) \| - \lambda_i(p)(s) }{|s|} = 0
\]
for all \(v \in E_{\lambda_i}(p) \setminus \{0\}\). The linear functionals \(\lambda_i(p) : A \to \mathbb{R}\) are the Lyapunov exponent functionals. The dimension of the corresponding \(E_{\lambda_i}(p)\) is the multiplicity of \(\lambda_i(p)\).
Recall the two \(D\tilde{\alpha}\)-invariant subbundles \(E^F\) and \(E^G\) of \(TM^\alpha\). We may restrict the derivative cocycle \(\{D\tilde{\alpha}(s) : s \in A\}\) to either of the two \(A\)-invariant distributions \(E^F\) or \(E^G\). These restrictions satisfy the hypotheses of the higher-rank multiplicative ergodic theorem. For the restricted cocycles, we obtain Lyapunov exponent functionals \(\{\lambda_i^F(p)\}\) and \(\{\lambda_i^G(p)\}\) and splittings \(E^F(p) = \oplus \lambda_i^F(p)\), \(1 \leq i \leq r^F(p)\) and \(E^G(p) = \oplus \lambda_i^G(p)\), \(1 \leq j \leq r^G(p)\) defined on a full measure \(A\)-invariant subsets. By a direct computation, we have that the linear functionals \(\lambda_i^G(p)\) coincide with \(\Sigma\), the restricted roots of \(g\) with respect to \(\alpha\). In particular, the number \(r^G(p)\), the functions \(\lambda_j^G(p)\), and the subspaces \(E_{\lambda_i^G(p)}(p)\) are defined at every point \(p \in M^\alpha\) and are independent of \(p\).
Below, we write \(\mathcal{L}(p)\), \(\mathcal{L}^F(p)\) and \(\mathcal{L}^G(p) = \Sigma\), respectively, for the corresponding collections of Lyapunov exponent functionals at the point \(p\) for the derivative cocycle and
its restrictions to $E^F$ and $E^g$. If $\mu$ is $A$-ergodic we write $\mathcal{L}(\mu)$, $\mathcal{L}^F(\mu)$ and $\mathcal{L}^G(\mu)$ or simply $\mathcal{L}$, $\mathcal{L}^F$ and $\mathcal{L}^G$ if the measure is understood.
3.2. Coarse Lyapunov exponents and coarse Lyapunov manifolds. For this section assume that $\mu$ is $A$-ergodic and that the charts in Section 2.3 satisfy properties (H1)–(H5) of Proposition 2.5. Note that Lyapunov exponents and dimension of the corresponding subspaces are independent of the point $a$.
As with restricted roots, we group Lyapunov exponent functionals into coarse equivalence classes by declaring that two exponents are equivalent if they are positively proportional. We write $\hat{\mathcal{L}}$ for the equivalence classes of coarse exponents. For $\chi \in \hat{\mathcal{L}}$ we write $E_\chi(p) = \bigoplus_{\lambda \in \chi} E_\lambda(p)$.
Recall that we equipped $TM^\alpha$ with a Riemannian metric which, in turn, induces a distance $d$ on $M^\alpha$. Given $s \in A$ and $p \in M^\alpha$ we write
$$W_s^u(p) := \left\{ y \in M^\alpha : \limsup_{n \to -\infty} \frac{1}{n} \log d(\hat{\alpha}(ns)(p), \hat{\alpha}(ns)(y)) < 0 \right\}$$
for the unstable manifold through $p$ for the action of $s \in A$ on $M^\alpha$. For $\mu$-almost every $p \in M^\alpha$, we have that $W_s^u(p)$ is a connected, injectively immersed, $C^{1+\beta}$ manifold with $T_pW_s^u(p) = \bigoplus_{\chi \in \hat{\mathcal{L}}, \lambda \in \chi} E_\lambda(p)$. Observe that given $s \in A$, the global unstable manifolds $\{W_s^u(p) : p \in M^\alpha\}$ form a (generally non-measurable) partition of $(M^\alpha, \mu)$.
Let $\mathbb{Z}^d$ be any lattice in $A \simeq \mathbb{R}^d$. Given a coarse Lyapunov exponent $\chi \in \hat{\mathcal{L}}$ we write $W^\chi(p)$ for the path connected (relative to the immersed topologies) component of
$$\bigcap_{\{s \in \mathbb{Z}^d : \chi(s) > 0\}} W_s^u(p)$$
containing $p$. $W^\chi(p)$ is the coarse Lyapunov manifold corresponding to $\chi$ through $p$. For a.e. $p$, $W^\chi(p)$ is a $C^{1+\beta}$ injectively immersed manifold with $T_pW^\chi(p) = E_\chi(p)$ (see [BRH]). We let $W^\chi$ denote the partition of $(M^\alpha, \mu)$ into coarse Lyapunov manifolds $W^\chi(p)$. In the terminology of [BRH], $W^\chi$ is a $C^{1+\beta}$-tame, $\hat{\alpha}\lfloor_A$-invariant, measurable foliation. Note that the partition $W^\chi$ is defined independently of the choice of lattice $\mathbb{Z}^d \subset A$ in that for any two choices of lattice, the corresponding partitions coincide modulo $\mu$.
Similarly, in the terminology of [BRH], the partition $G$ of $M^\alpha$ into $G$-orbits and the partition $\mathcal{F}$ of $M^\alpha$ into fibers form $C^{1+\beta}$-tame, $\hat{\alpha}$-invariant, measurable foliations. We similarly define $W^\mathcal{F}(p)$ and $W^\xi(p)$ for the coarse Lyapunov manifolds associated to coarse fiberwise Lyapunov exponents $\chi^F \in \hat{\mathcal{L}}^F$ and coarse roots $\xi \in \Sigma$. Note that if $\xi \in \Sigma$ then $W^\xi(p)$ is simply the orbit $\hat{\alpha}(G^\xi)(p)$ of $p$ by the unipotent subgroup $G^\xi = \exp G^\xi$ of $G$. We similarly define measurable foliations $W^\mathcal{F}$ and $W^\xi$ given by the partitions into fiberwise coarse manifolds and orbits of coarse root groups.
3.3. Conditional entropy, entropy product structure, and coarse-Lyapunov Abramov–Rohlin formula. Recall for $s \in A$ the $\mu$-metric entropy of $\hat{\alpha}(s)$ is
$$h_\mu(\hat{\alpha}(s)) := \sup\{h_\mu(\hat{\alpha}(s), \eta)\}$$
where the supremum is over all measurable partitions $\eta$ of $(M, \mu)$ and $h_\mu(\hat{\alpha}(s), \eta)$ is given by the mean conditional entropy
$$h_\mu(\hat{\alpha}(s), \eta) = H_\mu(\eta^+ | \hat{\alpha}(s)\eta^+)$$
where $\eta^+ = \sqrt[\infty]{\hat{\alpha}(s')\eta}$.
Given the partition $\mathcal{W}^x$ into coarse Lyapunov manifolds for $x \in \mathcal{L}$, for $s \in A$ with $\chi(s) > 0$ we define the conditional metric entropy of $\hat{\alpha}(s)$ relative to $\mathcal{W}^x$ as follows: A measurable partition $\xi$ of $(M^\alpha, \mu)$ is said to be subordinate to $\mathcal{W}^x$ if, for a.e. $p$,
1. the atom $\xi(p)$ is contained in $\mathcal{W}^x(p)$,
2. the atom $\xi(p)$ contains a neighborhood of $p$ in $\mathcal{W}^x(p)$, and
3. the atom $\xi(p)$ is precompact in $\mathcal{W}^x(p)$.
The conditional metric entropy of $\hat{\alpha}(s)$ relative to $\mathcal{W}^x$ is
$$h_\mu(\hat{\alpha}(s) \mid \mathcal{W}^x) := \sup \{ h_\mu(\hat{\alpha}(s), \eta \vee \xi) \}$$
where the supremum is over all partitions $\xi$ subordinate to $\mathcal{W}^x$ and all measurable partitions $\eta$.
From [BRHW] we have the following result which states that entropy behaves like a product along coarse Lyapunov manifolds.
**Proposition 3.2** (BRHW, Corollary 13.2). For $s \in A$
$$h_\mu(\hat{\alpha}(s)) = \sum_{\chi(s) > 0} h_\mu(\hat{\alpha}(s) \mid \mathcal{W}^x).$$
Given a coarse exponent $\chi \in \hat{\mathcal{L}}$ we write $\chi(F) \in \hat{\mathcal{L}}^F$ for the unique fiberwise coarse exponent with
$$\chi(F) = c\chi$$
for some $c > 0$ if such $c$ exists and 0 otherwise. Similarly, define $\chi(G)$ to be the unique coarse restricted root $\hat{\xi} \in \hat{\Sigma}$ that is positively proportional to $\chi$ and 0 otherwise. Note that given a non-zero coarse Lyapunov exponent $\chi \in \hat{\mathcal{L}}$, at least one of $\chi(F)$ or $\chi(G)$ is non-zero.
Let $\nu$ denote the image of $\mu$ under $\pi: M^\alpha \to G/\Gamma$. From the Abramov-Rohlin formula (c.f. [LW, BC]), we may decompose entropy of $\mu$ into the sum of the entropy along fibers and the entropy of the factor: for any $s \in A$
$$h_\mu(\hat{\alpha}(s)) = h_\nu(s) + h_\mu(\hat{\alpha}(s) \mid \mathcal{F}). \quad (6)$$
Here $\mathcal{F}$ is the partition into preimages of the projection $\pi$ and $h_\nu(s)$ is the metric entropy of the translation by $s$ on $(G/\Gamma, \nu)$. From [BRHW], we have a similar decomposition into fiber and factor entropy along coarse manifolds.
**Proposition 3.3** (BRHW, Theorem 13.4). Let $s \in A$ be such that $\chi(s) > 0$. Then
$$h_\mu(\hat{\alpha}(s) \mid \mathcal{W}^x) = h_\nu(s \mid \chi(G)) + h_\mu(\hat{\alpha}(s) \mid \mathcal{W}^x(F)). \quad (7)$$
Above, $h_\nu(s \mid \chi(G))$ denotes the metric entropy of translation by $s$ on $(G/\Gamma, \nu)$ conditioned on the partition of $(G/\Gamma, \nu)$ into orbits of $G^{x(G)}$. Note that for our applications below, if $\chi(F) = 0$ then $h_\mu(\hat{\alpha}(s) \mid \mathcal{W}^x(F)) = 0$.
Proposition 3.3 is a special case of [BRHW, Theorem 13.4] which establishes an Abramov-Rohlin formula for entropy subordinated to coarse Lyapunov manifolds for two smooth $\mathbb{Z}^d$-actions, one of which is a measurable factor of the other. In the current setting, our factor map $\pi: M^\alpha \to G/\Gamma$ is smooth and we obtain Proposition 3.3 directly from Proposition 3.2. We include a proof of Proposition 3.3 in our current setting.
**Proof of Proposition 3.3.** Note that, as the map $\pi: M^\alpha \to G/\Gamma$ is smooth, every coarse restricted root $\hat{\xi} \in \hat{\Sigma}$ for the action of $A$ on $G/\Gamma$ coincides with some coarse Lyapunov exponent $\chi \in \hat{\mathcal{L}}$ for the action of $A$ on $(M^\alpha, \mu)$; in particular, every $\hat{\xi} \in \hat{\Sigma}$ is of the form $\hat{\xi} = \chi(G)$ for some $\chi \in \hat{\mathcal{L}}$.
Given $\chi \in \hat{\mathcal{L}}$, set $\overline{\chi} = \chi(G)$ and take $s \in A$ with $\chi(s) > 0$. If $\overline{\chi} = 0$ take $\overline{\eta}$ to be the point partition on $G/\Gamma$. Otherwise, take $\overline{\chi}$ to be a measurable partition of $G/\Gamma$ such that
1. $s^{-1} \cdot \overline{\eta} \geq \overline{\eta}$;
2. the atom $\overline{\eta}(x)$ of $\overline{\eta}$ containing $x$ is contained in the $G\overline{\chi}$-orbit of $x$ and contains an open neighborhood of $x$ in the $G\overline{\chi}$-orbit;
3. $\bigvee_{n \in \mathbb{N}} s^{-n} \cdot \overline{\eta}$ is the point partition.
Let $\eta = \pi^{-1}(\overline{\eta})$. Take $\zeta$ to be a measurable partition of $M^\alpha$ such that
1. $\hat{\alpha}(s^{-1})(\zeta) \geq \zeta$;
2. the atom $\zeta(x)$ of $\zeta$ containing $x$ is contained in $W^\chi(x)$ and contains an open neighborhood of $x$ in $W^\chi(x)$ for almost every $x$;
3. $\bigvee_{n \in \mathbb{N}} \tilde{\alpha}(s^{-n})(\zeta)$ is the point partition.
The partitions $\overline{\eta}$ and $\zeta$ satisfy
$$h_\nu(s, \overline{\eta}) = h_\nu(s | \overline{\chi}), \quad \text{and} \quad h_\mu(\tilde{\alpha}(s), \zeta \vee F) = h_\mu(\tilde{\alpha}(s) | W^\chi(F)).$$
We have the following standard computation (c.f. [KRH, Lemma 6.1]):
$$h_\mu(\tilde{\alpha}(s) | W^\chi) := h_\mu(\tilde{\alpha}(s), \eta \vee \zeta) \leq h_\mu(\tilde{\alpha}(s), \eta) + h_\mu\left(\tilde{\alpha}(s), \zeta \vee \bigvee_{n \in \mathbb{Z}} \alpha(s^n)(\eta)\right)$$
$$= h_\nu(s, \overline{\eta}) + h_\mu(\tilde{\alpha}(s), \zeta \vee F) = h_\mu(s | \overline{\chi}) + h_\mu(\tilde{\alpha}(s) | W^\chi(F)).$$
Now, fix $\chi_0 \in \hat{\mathcal{L}}$. Given any $s \in A$ with $\chi_0(s) > 0$ we have from (6) and the analogue of Proposition 3.2 applied to the total, fiber, and base entropies (see full formulation in [BRHW, Theorem 13.4]) that
$$h_\mu(\tilde{\alpha}(s)) = \sum_{\chi(s) > 0} h_\mu(\tilde{\alpha}(s) | W^\chi) \leq \sum_{\chi(s) > 0} h_\nu(s | \chi(G)) + \sum_{\chi(s) > 0} h_\mu(\tilde{\alpha}(s) | W^\chi(F))$$
$$= h_\nu(s) + h_\mu(\tilde{\alpha}(s) | F) = h_\mu(\tilde{\alpha}(s)).$$
Since entropies are non-negative quantities, it follows that
$$h_\mu(\tilde{\alpha}(s) | W^\chi) = h_\nu(s | \chi(G)) + h_\mu(\tilde{\alpha}(s) | W^\chi(F))$$
for all $\chi \in \hat{\mathcal{L}}$ with $\chi(s) > 0$. \qed
4. Conditional measures and criteria for invariance
Let $G$ be as in the introduction. That is $G = G_1 \times \cdots \times G_\ell$ is the direct product of almost-simple Lie groups. Let $G^2 \subset G$ be the product of all non-compact almost-simple factors and $C \subset G$ the product of all compact almost-simple factors. Consider $X$ any locally compact, second countable metric space and suppose that $X$ admits a continuous left $G$-action $x \mapsto g \cdot x$. We moreover assume the action is locally free; that is, for every $x \in X$ there is a neighborhood $e \in U_x \subset G$ such that the map $U_x \rightarrow X, g \mapsto g \cdot x$ is injective. It follows that for every $x$ we have a canonical identification of $G$ with a covering space of the orbit $G \cdot x$ given by $g \mapsto g \cdot x$.
4.1. **Conditional measures along orbits of subgroups.** Consider $\mu$ any Borel probability measure on $X$. Let $V \subseteq G$ be a connected Lie subgroup and let $\eta$ be a measurable partition of $(X, \mu)$ such that for almost every $x \in X$, the atom $\eta(x)$ is contained in the $V$-orbit $V \cdot x$ and contains an open neighborhood of $x$ in the $V$-orbit $V \cdot x$. Such a partition is said to be **subordinate to $V$-orbits**. As above, we identify $V \subseteq G$ with a cover of the $V$-orbit through $x$. Taking a family of conditional probability measures for the partition $\eta$ of $(X, \mu)$, for $\mu$-almost every $x \in X$ we obtain a probability measure $\mu^V_x$ on $G$ whose support contains the identity and is contained in $V$. Fix a sequence of measurable partitions $\eta_j$ subordinate to $V$-orbits such that for any compact set $K \subseteq V$, for almost every $x$ there is a $j$ with $K \cdot x \subseteq \eta_j(x)$. By fixing a choice of normalization on $V$, a standard construction gives for almost every $x \in X$ a locally finite measure $\mu^V_x$, supported on $V$, which is canonical up to the choice of normalization. To emphasize the lack of uniqueness, we write $[\mu^V_x]$ for the equivalence class of the measure $\mu^V_x$ up to normalization of the measure.
The family of measures $\mu^V_x$ have the following characterization: for any partition $\eta$ subordinated to $V$-orbits, there is a function $c^\eta: X \to (0, \infty)$ so that if $\{\mu^\eta_x: x \in X\}$ is a family of conditional measures on $(X, \mu)$ associated with the measurable partition $\eta$, then
$$\mu^\eta_x = c^\eta(x)(v \mapsto v \cdot x)_*(\mu^V_x)|_{\eta(x)}.$$
Note that the subgroups $V$ above need not be unimodular. We have the following claim which follows from local disintegration and the definition of the left Haar measure.
**Claim 4.1.** Let $V \subseteq G$ be a connected Lie subgroup. Then the measure $[\mu^V_x]$ coincides with the left Haar measure on $V$ for $\mu$-almost every $x \in M$ if and only if the measure $\mu$ is invariant under the action of $V$.
The remainder of this section is devoted to a number of criteria which will guarantee that $[\mu^V_x]$ is the left Haar measure.
4.2. **Invariance from the structure of parabolic subgroups.** Recall we write $P = MAN$ for the minimal parabolic subgroup of $G$. Let $P' = P \cap G'$. Then $P' = MAN$ is the minimal parabolic subgroup of $G'$. Suppose $\mu$ is a $P'$-invariant, Borel probability measure on $X$. Given a coarse negative root $\xi \in \Sigma_-$ and a nontrivial subgroup $V \subseteq G^\xi$ such that $\mu$ is $V$-invariant then, as the stabilizer of a measure is a closed subgroup of $G$, it follows from the structure theory of parabolic subgroups (of $G'$) that $\mu$ is invariant by the full coarse root group $G^\xi$. In the case that the subgroup $V$ above varies with the point $x \in X$, we have the following lemma. Note that $G^\xi$ is nilpotent so subgroups of $G^\xi$ are unimodular.
**Lemma 4.2.** Let $\mu$ be a $P'$-invariant measure on $X$ and suppose for some $\xi \in \Sigma_-$, that for $\mu$-a.e. $x \in X$ there is a nontrivial, connected Lie subgroup $V(x) \subseteq G^\xi$ such that $[\mu^G_x]$ coincides with the Haar measure on $V(x)$. Moreover, assume the assignment $x \mapsto V(x)$ is measurable and $A$-invariant. Then the measure $\mu$ is $G^\xi$-invariant.
**Proof.** Let $\{\mu^\xi_x\}$ denote the $A$-ergodic decomposition of $\mu$. It is enough to verify that the measure $\mu^\xi_x$ is $G^\xi$-invariant for almost every $x$. First note that there is a $s \in A$ so that, for every $x \in X$ and $y \in N \cdot x, d(s^k \cdot x, s^k \cdot y) \to 0$. It follows that the partition into ergodic components is refined by the measurable hull of the partition into $N$-orbits. In particular, for $\mu$-a.e. $x$ the measure $\mu^\xi_x$ is $N$-invariant.
Fix a generic $x \in X$. Let $V$ be the $\mu^\xi_x$-a.s. constant value of $x \mapsto V(x)$. Let $H(x)$ be the closed subgroup of $G$ under which $\mu^\xi_x$ is invariant and let $h_x = \text{Lie}(H(x))$.
As $-\xi$ is a positive coarse restricted root, we have $g^{-\xi} \subseteq h_x$. Moreover, given a nonzero $Y \in \text{Lie}(V)$, from the analysis of $s\mathfrak{l}(2, \mathbb{R})$ triples in $g$ (see [Kna, Lemma 7.73]), we
have that \((\text{ad}(Y))^2\) maps \(g^{-\xi}\) onto \(g^\xi\). In particular \(g^\xi \subset h_x\) whence \(\mu_x^G\) is \(G^\xi\)-invariant.
4.3. High-entropy method. We have the following theorem of Einsiedler and Katok from which we deduce invariance along unipotent subgroups of an \(A\)-invariant measures based on its support along coarse root spaces. We say a Lie subalgebra \(h \subset g\) is contracting if it is invariant under the adjoint action of \(A\) and if there is some \(s \in A\) with
\[ h = \bigoplus_{\xi \in \hat{\Sigma}: \xi(s) < 0} (g^\xi \cap h). \]
Note that any such \(h\) is nilpotent, hence unimodular. We state a simplified version of the High Entropy Theorem from \([EK2]\).
**Theorem 4.3** (High Entropy Theorem, \([EK2, Theorem 8.5]\)). Let \(\mu\) be an \(A\)-invariant measure on \(X\) and let \(h \subset g\) be a contracting Lie algebra with corresponding analytic subgroup \(H\). Then for \(\mu\)-a.e. \(x\) there are Lie subgroups
\[ H_x \subset S_x \subset H \]
with
(1) \(\mu_x^H\) is supported on \(S_x\);
(2) \(\mu_x^H\) is invariant under left and right multiplication by \(H_x\);
(3) \(H_x\) and \(S_x\) are connected and their Lie algebras are direct sums of subspaces of root spaces;
(4) \(H_x\) is normal in \(S_x\) and if \(\xi, \xi' \in \hat{\Sigma}\) with \(\xi \neq \xi'\) are distinct coarse roots then for \(g \in S_x \cap G^\xi\) and \(h \in S_x \cap G^{\xi'}\) the cosets \(gH_x\) and \(hH_x\) commute in \(S_x/H_x\);
(5) \(\mu_x^G\) is left- and right- invariant under multiplication by elements of \(H_x \cap G^\xi\).
It follows that the groups \(S_x\) and \(H_x\) are equivariant under conjugation by \(A\); that is \(S_{s,x} = sS_x s^{-1}\). Unlike in \([EK2]\), we only consider here the adjoint action of \(A\) on \(g\). As this action is semisimple with real roots, it follows that the groups \(S_x\) and \(H_x\) are normalized by \(A\). In particular, the maps \(x \mapsto S_x\) and \(s \mapsto H_x\) are constant along \(A\)-orbits.
4.4. Invariance from entropy considerations. Consider again \(\mu\) an \(A\)-invariant, \(A\)-ergodic measure. Given a coarse root \(\xi \in \hat{\Sigma}\) let \(W^\xi\) be the partition of \(X\) into orbits of \(G^\xi\). We have a standard fact (see for example \([LS]\)) that if \(\mu\) is \(G^\xi\)-invariant then for \(s \in A\) with \(\xi(s) > 0\), the entropy of the action of \(s\) on \((X, \mu)\) conditioned along orbits of \(G^\xi\) is given by
\[ h_\mu(s \mid W^\xi) = \sum_{\beta \in \xi} \beta(s) \dim(g^\beta). \]
The converse also holds.
**Lemma 4.4.** Let \(\xi \in \hat{\Sigma}\) be such that
\[ h_\mu(s \mid W^\xi) = \sum_{\beta \in \xi} \beta(s) \dim(g^\beta) \]
for some \(s \in A\) with \(\xi(s) > 0\). Then \(\mu\) is \(G^\xi\)-invariant.
Indeed, Ledrappier shows in \([Led1, Theorem 3.4]\) that \(\mu\) has absolutely continuous conditional measures along \(G^\xi\)-orbits. Moreover, from the explicit computation of the density function in the proof of \([Led1, Theorem 3.4]\) it follows that the conditional measures of \(\mu\)
along $G^\xi$-orbits coincide with the image of the Haar measure on $G^\xi$. See also [LY, (6.1)] for the argument in English. From Claim 4.1 it follows that $\mu$ is $G^\xi$-invariant.
We remark that deriving extra invariance of a measure by verifying that conditional entropy is maximized also underlies the proof of the so-called “invariance principle” for fiber-wise conditional measures invariant under a skew product, developed by Ledrappier in [Led2] and extended in [AV].
5. Main propositions and proofs of Theorems 1.6, 1.7, and 1.10
5.1. Non-resonance implies invariance. We return to the setting introduced in Section 2. Consider an $A$-invariant, $A$-ergodic measure $\mu$ on $M^\alpha$ satisfying (H5) of Proposition 2.5. We say a restricted root $\beta \in \Sigma$ of $g$ is resonant (with the fiber exponents $\mathcal{L}^F(\mu)$ of $\mu$) if there exists a $c > 0$ and a $\lambda \in \mathcal{L}^F(\mu)$ with
$$\beta = c\lambda.$$
If no such $c$ and $\lambda$ exist, we say $\beta$ is non-resonant. We similarly say that a fiberwise Lyapunov exponent $\lambda \in \mathcal{L}^F(\mu)$ is resonant (with $g$) if there is a $c > 0$ and $\beta \in \Sigma$ with
$$\lambda = c\beta.$$
Note that resonance and non-resonance are well-defined on the set of coarse restricted roots $\tilde{\Sigma}$ and coarse fiberwise exponents $\mathcal{L}^F(\mu)$.
The proof of Theorem 1.6 follows directly from the following key proposition.
Proposition 5.1. Let $\mu$ be an $A$-invariant, $A$-ergodic Borel probability measure on $M^\alpha$ such that the image of $\mu$ in $G/\Gamma$ is $G'$-invariant. Then, given a coarse restricted root $\xi \in \tilde{\Sigma}$ that is non-resonant with the fiberwise Lyapunov exponents of $\mu$, the measure $\mu$ is $G^\xi$-invariant for the action $\tilde{\alpha}$.
Proof. Indeed if $\xi$ is a non-resonant coarse restricted root then $\xi = \chi(G)$ for some coarse exponent $\chi \in \mathcal{L}$ with $\chi(F) = 0$. Let $\nu$ denote the image of $\mu$ in $G/\Gamma$. Since $\nu$ is $G'$-invariant, it follows for $s \in A$ with $\xi(s) > 0$ that
$$h_\nu(s \mid \xi) = \sum_{\beta \in \xi} \beta(s) \dim(g^\beta).$$
From Proposition 3.3 and the fact that the partitions $\mathcal{W}^x = \mathcal{W}^\xi$ coincide in $M^\alpha$, it follows that $h_\mu(\tilde{\alpha}(s) \mid \mathcal{W}^{\chi(F)}) = 0$ whence
$$h_\mu(\tilde{\alpha}(s) \mid \mathcal{W}^\xi) = \sum_{\beta \in \xi} \beta(s) \dim(g^\beta).$$
The $G^\xi$-invariance of $\mu$ then follows from Lemma 4.4.
We remark that the proof of Proposition 5.1 is similar to key steps in [MT] and [EM] where one deduces extra invariance of a measure by computing conditional entropy, verify the entropy is the maximal value permitted by the Margulis–Ruelle inequality, and applying Ledrappier’s result Lemma 4.4 to obtain invariance.
5.2. $P$-invariant measures and the proof of Theorem 1.6. Recall that $P$ is the minimal standard parabolic subgroup and is hence amenable. It follows that there exists an invariant probability measure $\mu$ for the restriction to $P$ of the action $\tilde{\alpha}$ on $M^\alpha$. Moreover, it follows that any such measure factors to the Haar measure on $G/\Gamma$. Fix a $P$-invariant, $P$-ergodic measure $\mu$ on $M^\alpha$. Recall that $A \subset P$ and the data $r(\cdot), \lambda_\lambda(\cdot), E_{\lambda}(\cdot)$ defined in Theorem 3.1 for the action of $A$ on $(M, \mu)$ as well as the corresponding data $r^F(\cdot), \lambda^F(\cdot)$, and
$E_{\lambda^F}(\cdot)$ and $r^G(\cdot)$, $\lambda^G_\beta(\cdot)$, and $E_{\lambda^G}(\cdot)$ for the fiberwise and orbit cocycles. As observed earlier, the data $r^G(\cdot)$, $\lambda^G_\beta(\cdot)$, and $E_{\lambda^G}(\cdot)$ are independent of the measure $\mu$ and the point. We show for $\mu$ as above, the remaining data is independent of the point.
**Claim 5.2.** Suppose that $\mu$ is a $P$-invariant, $P$-ergodic measure. Then the functions $r(\cdot)$, $r^F(\cdot)$, $\lambda_\beta(\cdot)$, and $\lambda^F(\cdot)$ and the dimensions of the corresponding subspaces $E_{\lambda_\beta}(\cdot)$, $E_{\lambda^F}(\cdot)$ are constant almost surely.
**Proof.** Note that $\mu$ is $P$-ergodic but need not be $A$-ergodic. Let $\{\mu^\beta_p\}_{p \in M^\alpha}$ denote the $A$-ergodic decomposition of $\mu$. We may select $s \in A$ so that $\beta(s) < 0$ for every $\beta \in \Sigma_+$. By the pointwise ergodic theorem, it follows that ergodic components are refined by the measurable hull of the partition into $N$-orbits. Then $\mu^\beta_p$ is $N$-invariant for almost every $p \in M^\alpha$. It follows that the data in the claim is constant along $AN$-orbits.
Finally, recall that $P = MAN$ with $M$ contained in the centralizer of $A$. It follows that the data is constant along $M$ orbits. By the $P$-ergodicity of $\mu$, the result follows. \qed
From Claim 5.2 it follows that for any $P$-invariant, $P$-ergodic measure $\mu$ on $M^\alpha$, the set of resonant roots depends only on the measure $\mu$ and not the decomposition of $\mu$ into $A$-ergodic components.
Theorem 1.6 now follows immediately from Proposition 5.1.
**Proof of Theorem 1.6.** Let $\mu$ be any $P$-invariant, $P$-ergodic measure on $M^\alpha$. Then $\mu$ is invariant under a closed subgroup $Q \supset G$ with $P \subset Q$. If $\dim M < r(G)$ then there at most $r(G) - 1$ fiberwise Lyapunov exponent functionals in $\cal L^F$, hence at most $r(G) - 1$ coarse fiberwise Lyapunov exponent functionals in $\hat{\cal L^F}$. It follows that there are at most $r(G) - 1$ resonant coarse restricted roots $\xi \in \hat{\Sigma}$. From Proposition 5.1, it follows that $Q$ is a standard parabolic subgroup with resonant codimension strictly smaller then $r(G)$. But then $Q = G$ by definition of $r(G)$.
It follows that $\mu$ is a $G$-invariant, Borel probability measure on $M^\alpha$. From Claim 2.1, it follows that there exists a $T$-invariant Borel probability measure on $M$. \qed
### 5.3. Parabolic subgroups associated to conditional measures.
We continue to assume $\mu$ is a $P$-invariant, $P$-ergodic measure on $M^\alpha$. The proof of Theorems 1.7 and 1.10 follow from an analysis of the geometry of the measures $[\mu^G_p]$ constructed in the previous section.
We define subgroups $Q_{\text{In}}(\mu) \subset Q_{\text{Out}}(\mu)$ of $G$ as follows: Given $p \in M^\alpha$ let
1. $Q_{\text{In}}(\mu)$ denote the largest subgroup of $G$ for which $\mu$ is invariant for the action $\tilde{\alpha}$;
2. $Q_{\text{Out}}(\mu; p)$ denote the smallest, closed, $[\mu^G_p]$-null subgroup of $G$.
Note that both $Q_{\text{In}}(\mu)$ and $Q_{\text{Out}}(\mu; p)$ are standard parabolic subgroups. As $P \subset Q_{\text{Out}}(\mu; p)$, it follows that $Q_{\text{Out}}(\mu; p)$ is constant along $P$-orbits. By $P$-ergodicity of $\mu$, we write $Q_{\text{Out}}(\mu)$ for the almost-surely constant value of $Q_{\text{Out}}(\mu; p)$.
Theorems 1.7 and 1.10 will follow from verifying that $Q_{\text{In}}(\mu) = Q_{\text{Out}}(\mu)$. We use the criteria in the previous section to verify this condition. First, consider the case that every fiberwise Lyapunov exponent $\lambda^F_\beta$ of $\mu$ is resonant with a negative root. In this setting we immediately obtain that $Q_{\text{In}}(\mu)$ and $Q_{\text{Out}}(\mu)$ coincide.
**Proposition 5.3.** Suppose that for every $\lambda^F_\beta \in \cal L^F$ there is a $\beta \in \Sigma_-$ and $c > 0$ with $\lambda^F_\beta = \beta \cdot c$. Then $Q_{\text{In}}(\mu) = Q_{\text{Out}}(\mu)$.
We also verify that $Q_{\text{In}}(\mu) = Q_{\text{Out}}(\mu)$ given the combinatorics of the number $m(G)$.
**Proposition 5.4.** Suppose $\alpha$ has no rank-1 simple ideals and that $Q_{\text{In}}(\mu)$ is a maximal parabolic subgroup. Then $Q_{\text{In}}(\mu) = Q_{\text{Out}}(\mu)$.
5.4. Proofs of Theorems 1.7 and 1.10. Given a $P$-invariant, $P$-ergodic measure $\mu$ as above, let $\tilde{\mu}$ denote the locally finite measure on $G \times M$ obtained from lifting $\mu$ on fundamental domains of $\Gamma$. Given $g \in G$, let $\mu_g$ denote the conditional probability measure on $M$ defined by disintegrating $\tilde{\mu}$ along fibers and identifying the fiber $\{g\} \times M$ with $M$.
As $\tilde{\mu}$ lifts $\mu$, we have that $\{\mu_g : g \in G\}$ is $\Gamma$-equivariant:
$$\mu_{g\gamma} = \alpha(\gamma)_* \mu_g.$$
Moreover, as $\mu$ is $Q_{\mathrm{In}}(\mu)$-invariant, for almost every $g \in G$, we have that $\mu_g = \mu_{gg}$ for every $q \in Q_{\mathrm{In}}(\mu)$. Let $Q = Q_{\mathrm{In}}(\mu)$. We equip $Q \setminus G$ with any measure $m$ in the Lebesgue class. Let $\overline{\pi}$ be the measure on $Q \setminus G \times M$ given by
$$\overline{\pi}(B) = \int \mu_g(\{x : (Qg, x) \in B\}) \, dm(Qg)$$
and let $\hat{\mu}$ be the measure on $M$ given by
$$\hat{\mu}(B) = \int \mu_g(B) \, dm(Qg).$$
Note that $\hat{\mu}$ is image of $\overline{\mu}$ under the natural projection $\pi : Q \setminus G \times M \to M$.
Consider the $\overline{\pi}$-measurable partition $\zeta^\pi$ on $Q \setminus G \times M$ into level sets of the map $\pi$. We have that $\zeta^\pi$ is measurably equivalent to the partition $\{Q_{\mathrm{In}}(\mu)\backslash Q_{\mathrm{Out}}(\mu) \times \{x\} : x \in M\}$. In particular, in the case $Q_{\mathrm{In}}(\mu) = Q_{\mathrm{Out}}(\mu)$ the following claim follows immediately.
Claim 5.5. If $Q_{\mathrm{In}}(\mu) = Q_{\mathrm{Out}}(\mu)$ then the projection $(Q \setminus M, \overline{\pi}) \to (M, \hat{\mu})$ is a measurable isomorphism.
Theorems 1.7 and 1.10 follow from $\Gamma$-equivariance of the family $\{\mu_g\}$ and Claim 5.5.
Proof of Theorems 1.7 and 1.10. Let $\mu$ be a $P$-invariant, $P$-ergodic measure on $M^\alpha$.
First consider the setting of Theorem 1.7 where $\dim(M) = r(G)$. If there exists a non-resonant, fiberwise Lyapunov exponent $\lambda^F_i$ for $\mu$ then, by dimension counting, there are at most $r(G) - 1$ coarse resonant roots $\xi \in \Sigma$. However, as $\mu$ is $P$-invariant and as there are no proper parabolic subalgebras of resonant codimension smaller that $r(G)$, it follows that $\mu$ is necessarily $G$-invariant. It then follows that if $\alpha$ has no invariant probability measure on $M^\alpha$, then every fiberwise Lyapunov exponent of $\mu$ is resonant with a root of $\mathfrak{g}$. We claim in this case that every fiberwise exponent for $\mu$ is in fact resonant with a negative root $\beta \in \Sigma_{-}$. Indeed, if there existed a fiberwise exponent that was resonant with a positive root, then would be at most $r(G) - 1$ resonant negative roots. As we assume $\mu$ is $P$-invariant we again generate a parabolic subgroup which preserves $\mu$ and with resonant codimension smaller than $r(G)$. This again implies the existence of an $\alpha$ invariant probability on $M^\alpha$.
Thus, in the case that $\dim(M) = r(G)$ it follows that if there is no $\alpha$-invariant measure on $M$ then there exists $s \in A$ such that $\lambda^F_i(s) < 0$ for every fiberwise Lyapunov exponent $\lambda^F_i$ of $\mu$. Proposition 5.3 then holds and a standard argument shows in this case that the fiberwise conditional measures $\mu_g$ are supported on a finite set for almost every $g$. By ergodicity, the number of atoms is constant a.s.
In the case that $\dim(M) \leq m(G)$ and every fiberwise Lyapunov exponent is resonant with a negative root, the same analysis as above holds. In particular the hypotheses of Proposition 5.3 hold. Note that this holds even if $\mathfrak{g}$ has rank-1 simple ideals (so $m(G) = 1$ and $M$ is a circle.) If $\dim(M) \leq m(G)$ and not every fiberwise Lyapunov exponent is resonant with a negative root, then there are at most $m(G) - 1$ resonant, negative coarse restricted roots. Note if $\mathfrak{g}$ has rank-1 simple ideals then, as $m(G) - 1 = 0$, this implies
\[ Q_{\text{in}}(\mu) = Q_{\text{Out}}(\mu) = G. \] Thus we may assume \( \mathfrak{g} \) has no rank-1 simple ideals. From the definition of \( \mathfrak{m}(G) \), it follows that either \( Q_{\text{in}}(\mu) = G \) or that \( Q_{\text{in}}(\mu) \) is a maximal parabolic subgroup and, from Proposition 5.4, we have that \( Q_{\text{in}}(\mu) = Q_{\text{Out}}(\mu) \).
In particular, under the hypotheses of either Theorem 1.7 or 1.10, we have \( Q := Q_{\text{in}}(\mu) = Q_{\text{Out}}(\mu) \).
In the setting of either theorem, the spaces \((M, \mu_g)\) are Lebesgue probability spaces. As there are at most countably many isomorphism types of Lebesgue probability spaces, by \( P \)-ergodicity it follows that the spaces \((M, \mu_g)\) are all measurably isomorphic to a fixed abstract Lebesgue probability space \((Y, \eta)\). In particular, there exists a measurable family of measurable isomorphisms \( \phi_g : (M, \mu_g) \to (Y, \eta) \). Moreover the family of isomorphism \( \phi_g \) is \( Q \)-invariant. The family of isomorphisms \( \phi_g \) translates the \( \Gamma \)-equivariance of the family \( \mu_g \) to a family of automorphisms of the measure space \((Y, \eta)\) parameterized by \( Q \setminus G \):
\[
\psi(\gamma, Qg) := \phi_{g\gamma} \circ \alpha(\gamma) \circ \phi_{g}^{-1} \in \text{Aut}(Y, \eta).
\]
One verifies that \( \psi \) is a cocycle over the right \( \Gamma \)-action on \( Q \setminus G \).
It now follow from Claim 5.5 that \((M, \hat{\mu})\) is measurably isomorphic to \((Q \setminus G, \nu) \times (Y, \eta)\). Moreover, the action \( \alpha \) of \( \Gamma \) on \((M, \hat{\mu})\) is measurably conjugate via this isomorphism to the skew action defined by \( \psi \) over the standard right action of \( \Gamma \) on \( Q \setminus G \). Theorems 1.7 and 1.10 now follow.
6. Proof of Propositions 5.3 and 5.4
We recall the notation of Section 5.3. In particular, we take \( \mu \) to be a \( P \)-invariant, \( P \)-ergodic measure on \( M^\alpha \). Recall also the definitions of \( Q_{\text{in}}(\mu) \) and \( Q_{\text{Out}}(\mu) \) in Section 5.3. We verify under the hypotheses of Propositions 5.3 and 5.4 that \( Q_{\text{in}}(\mu) = Q_{\text{Out}}(\mu) \).
6.1. Conditional measures along coarse root spaces. Under the assumption that \( Q_{\text{in}}(\mu) \neq Q_{\text{Out}}(\mu) \) the following claim guarantees the existence of a coarse restricted root \( \xi \in \Sigma \) with \( G^\xi \not\subset Q_{\text{in}}(\mu) \) and such that the measure \( [\mu^G_p] \) is non-trivial. Write \( Q = Q_{\text{in}}(\mu) \) and \( q = \text{Lie}(Q) \) for the remainder.
Claim 6.1. Suppose \( Q_{\text{in}}(\mu) \neq Q_{\text{Out}}(\mu) \). Then there is a coarse restricted root \( \xi \in \hat{\Sigma} \) with \( G^\xi \not\subset q \) and such that \( \mu^G_p \) is non-atomic for \( \mu \)-a.e. \( p \in M^\alpha \).
The claim follows from the local product structure of \( A \)-invariant measures on \( G \)-spaces demonstrated in [EK1, Proposition 8.3] and further developed in [EK2, Theorems 7.5, 8.4]. We sketch a short proof here for completeness.
Given standard parabolic subgroup \( Q \) with Lie algebra \( q \), let
\[
\Sigma_Q = \{ \beta \in \Sigma : g^\beta \subset q \}, \quad \Sigma^\perp_Q = \{ \beta \in \Sigma : g^\beta \not\subset q \}.
\]
We have \( \Sigma = \Sigma_Q \cup \Sigma^\perp_Q \) and \( \Sigma_Q \) and \( \Sigma^\perp_Q \) are saturated by coarse equivalence classes of restricted roots.
Proof. Recall we write \( Q = Q_{\text{in}}(\mu) \) and the measure \( \mu^G_p \) is a \( Q \)- and hence \( A \)-invariant measure on \( G \). Let \( g^\perp := \bigoplus_{\beta \in \Sigma^\perp_Q} \). Note that \( \Sigma^\perp_Q \) consists of negative roots. Let \( V \) be the analytic subgroup corresponding to \( g^\perp \). Let \( C_s \) denote conjugation by \( s \in A \). We have \( C_s(V) = V \) for \( s \in A \). As \( \mu \) is \( A \)-invariant, we have for \( s \in A \) that \( [(C_s)_{\ast} \mu^G_p] = [\mu^G_{(s)p}] \) for almost every \( p \).
As \( \Sigma^\perp_Q \subset \Sigma^\perp \), we may find an \( s_0 \in A \) and a coarse restricted root \( \xi \subset \Sigma^\perp \) with
- \( \beta(s_0) = 0 \) for \( \beta \in \xi \);
\[ \beta(s_0) < 0 \text{ for all } \beta \in \Sigma^A_q \setminus \xi. \]
Let \( V' \) be the analytic subgroups of \( V \) corresponding to \( \bigoplus_{\beta \in \Sigma^A_q \setminus \xi} \beta \).
Suppose first that \( \mu_p^V \) is not supported on a single \( V' \)-orbit for a positive measure of set of \( p \in M^\alpha \). As \( \hat{\alpha}(s_0) \) acts as the identity on \( G^\xi \)-orbits, we have \( \mu_p^{G^\xi}_{\hat{\alpha}(s_0)(p)} = \mu_p^{G^\xi} \) for almost every \( p \). Moreover, as \( \hat{\alpha}(s_0) \) contracts \( V' \) orbits, it follows from Poincaré recurrence and Lusin’s theorem that \( \mu_p^{G^c} = \mu_p^{G^\xi}_{\hat{\alpha}(v)(p)} \) for \( \mu_p^{V'} \)-a.e. \( v \in V' \) and \( \mu^G \)-a.e. \( p \). Thus, \( \mu_p^{G^c} \) is atomic on a positive measure set of \( p \) only if \( \mu_p^V \) is supported on a single \( V' \)-orbit for a positive measure of set of \( p \in M^\alpha \). Thus, \( \mu_p^{G^c} \) is non-atomic on a positive measure set of \( p \). Note that the actions by \( A \) and \( M \) preserve the coarse root subgroups \( G^\xi \) and also preserve the measure \( \mu \). Also, as the \( A \)-ergodic components of \( \mu \) are \( N \)-invariant, it follows from \( P \)-ergodicity of \( \mu \) that \( \mu_p^{G^c} \) is non-atomic for almost every \( p \).
If \( \mu_p^V \) is supported on a single \( V' \)-orbit for almost every \( p \in M^\alpha \), we may recursively repeat the above argument with \( V \) replaced by \( V' \).
\[ \square \]
### 6.2. Recurrence and the proof of Proposition 5.3.
We show under the assumption that every fiberwise Lyapunov exponent is resonant with a negative root, that \( Q_{\text{In}}(\mu) = Q_{\text{Out}}(\mu) \).
Suppose that \( Q_{\text{In}}(\mu) \neq Q_{\text{Out}}(\mu) \) and let \( \xi \) be a coarse restricted root as in Claim 6.1. We will show below that \( \mu \) is \( G^\xi \)-invariant. The contradiction completes the proof of Proposition 5.3.
Recall that \( G = C \times G' \) where \( C \) is the maximal compact factor. Let \( (M^\alpha)' \) denote the quotient of \( M^\alpha \) under the action of \( \hat{\alpha}(C) \). Note that \( (M^\alpha)' \) might have an orbifold structure but still has the structure of a fiber-bundle (with orbifold fibers) over \( G'/\Gamma = C \setminus G/\Gamma \). We have \( C \subset M \subset C \). As \( C \) is contained in the centralizer of both \( A \) and \( N \), the actions of \( A \) and \( N \) on \( G/\Gamma \) and \( M^\alpha \) descend to actions of \( A \) on \( C \setminus G/\Gamma \) and \( (M^\alpha)' \).
Let \( A' \subset A \) denote the kernel of \( \xi \); that is, \( s \in A' \) if \( \beta(s) = 0 \) for all \( \beta \in \xi \). As we assume \( \Gamma \) has dense image in every rank-1 almost-simple subgroup of \( G \), it follows from Moore’s ergodicity theorem (applied to each irreducible factor) that \( A' \) acts ergodically on \( G'/\Gamma \) (see for example [Zim, Theorem 2.2.6]).
Let \( \mu' \) denote the projection of the measure \( \mu \) onto \( (M^\alpha)' \) and let \( \{ \mu'_g : g \Gamma \in G'/\Gamma \} \) denote a family of conditional measures induced by the partition of \( (M^\alpha)' \) into its fibers over \( G'/\Gamma \). As discussed in the proof of Theorem 1.7, the assumption that every fiberwise Lyapunov exponent is resonant with a negative root combined with the \( C \)-invariance of \( \mu \) implies that \( \mu'_g \) has finite support for almost every \( g \Gamma \in G'/\Gamma \). Moreover, \( P \)-ergodicity of \( \mu \) ensures that the number of atoms is constant in \( g \Gamma \).
Note that (as we assume \( \mu^{G^c}_p \) is non-atomic) the partition of \( ((M^\alpha)',\mu') \) into full \( G^\xi \)-orbits is non-measurable. Let \( \eta^\xi \) denote the measurable hull of this partition; that is the finest measurable partition of \( ((M^\alpha)',\mu') \) containing full \( G^\xi \)-orbits.
Consider the action of \( A' \) on \( ((M^\alpha)',\mu') \). Note that the action need not be ergodic. Let \( \mathcal{E}_{A'} \) denote the partition into ergodic components of \( \mu' \) with respect to the action of \( A' \).
We have the following claim which will provide the necessary recurrence to complete the proof of Proposition 5.3.
**Claim 6.2.** The partition \( \eta^\xi \) refines \( \mathcal{E}_{A'} \).
**Proof.** Let \( \mathcal{E}_A \) denote the partition into ergodic components of \( \mu' \) with respect to the action of \( A \) on \( (M^\alpha)' \). Taking \( s \in A \) such that \( \xi(s) < 0 \), it follows that the partition of \( (M^\alpha)' \) into \( G^\xi \)-orbits defines a (uniformly) contracting foliation of \( (M^\alpha)' \) under the action \( \hat{\alpha}(s) \). By
the pointwise ergodic theorem, it follows that the partition of \((M^\alpha)'\), \(\mu'\) into \(\hat{\alpha}(s)\)-ergodic components is refined by \(\hat{\eta}^\xi\). To complete the proof of the claim we show \(E_A = \hat{E}_{A'}\).
Fix \(p' \in (M^\alpha)'\) and let \((\mu'_p)^{E_A}\) be the \(A\)-ergodic component of \(\mu'\) containing \(p'\). Let \(\hat{\tilde{E}}(p')\) denote the partition into \(A'\)-ergodic components of \((M^\alpha)'(\mu'_p)^{E_A}\). We claim that \(\hat{\tilde{E}}(p')\) is finite for almost every \(p'\). Indeed first note that, as both \(A\) and \(A'\) act ergodically on \(G'/T\), the \(A\) and \(A'\)-ergodic components of \((M^\alpha)'(\mu')\) project to the Haar measure on \(G'/T\). Furthermore, as the fiber conditional measures \((\mu'_p)^{G/T}\) are purely atomic and as the ergodic components of the \(A'\)-action on \((M^\alpha)'(\mu'_p)^{E_A}\) are mutually singular, it follows that the partition \(\hat{\tilde{E}}(p')\) is finite.
As \(A' \subset A\) with \(A\) abelian, it follows that \(A\) permutes elements of the partition \(\hat{\tilde{E}}(p')\) of \((M^\alpha)'(\mu'_p)^{E_A}\). Note that the partition \(\hat{\tilde{E}}(p')\) is finite, \(A\) acts ergodically on \((M^\alpha)'(\mu'_p)^{E_A}\), and \(A\) is a connected group. In particular, \(A\) acts ergodically on the (finite) factor measure space \((M^\alpha)'(\mu'_p)^{E_A})/\hat{\tilde{E}}(p')\). This yields a contradiction unless the partition \(\hat{\tilde{E}}(p')\) contains only one element. It follows that \(E_A = \hat{E}_{A'}\).
Now consider \(\hat{\eta}^\xi\), the measurable hull of the partition into \(G^\xi\)-orbits on \((M^\alpha),\mu\) and \(\hat{E}_{A'}\), the partition of \((M^\alpha),\mu\) into \(A'\)-ergodic components. We claim \(\hat{\eta}^\xi\) refines \(\hat{E}_{A'}\). Indeed, since \(C\) centralizes \(A'\), \(C\) permutes elements of \(\hat{E}_{A'}\). Similarly, \(C\) centralizes \(G^\xi\) and hence permutes elements of \(\hat{\eta}^\xi\). Moreover, \([\mu_p^{G^\xi}] = [\mu_p^{\hat{\eta}^\xi(g)}]\) for \(g \in C\). If elements of \(\hat{E}_{A'}\) did not contain full \(G^\xi\)-orbits modulo \(\mu\), then \(C\)-orbits of elements of \(\hat{E}_{A'}\) would not contain \(C\)-orbits of full \(G^\xi\)-orbits modulo \(\mu\) contradicting the above claim.
We now consider the \(A'\)-action on \(M^\alpha\). As \(A'\) acts isometrically on \(G^\xi\)-orbits and as \(\hat{\eta}^\xi\) refines \(\hat{E}_{A'}\), from standard measure rigidity arguments for actions of Abelian groups we obtain the following.
**Claim 6.3.** \(\mu\) is invariant under the action \(G^\xi\).
We only outline the main steps in the proof of Claim 6.3.
**Proof of Claim 6.3.** Fix \(U \subset G^\xi\) a pre-compact, open neighborhood of the identity in \(G^\xi\). Given almost every \(p \in M^\alpha\), the measure \(\mu_p^{G^\xi}\) gives positive mass to \(U\). For such \(p\), normalize \(\mu_p^{G^\xi}\) on \(U\).
Let \(A'\) be as above. Then any \(s \in A'\) commutes with \(G^\xi\) whence \(C_s(U) := sUs^{-1} = U\) and \(\mu_s^{G^\xi} = \mu_p^{G^\xi}\). Let \(K \subset M^\alpha\) be a compact set on which the assignment \(p \mapsto \mu_p^{\xi}\) is continuous (where locally finite measure on \(G^\xi\) are endowed with the topology dual to compactly supported continuous functions.)
Consider a generic \(p \in K\). Recall that the \(\alpha(G^\xi)\)-orbit of \(p\) is contained in the \(A'\)-ergodic component of \(\mu\) containing \(p\). Consider any \(p' \in \alpha(G^\xi)(p) \cap K\) that is a density point of \(K\) with respect to the \(A'\)-ergodic component of \(\mu\) containing \(p\). It follows that there is a sequence \(s_k \in A'\) with
1. \(\alpha(s_k)(p) \in K\) for every \(k \in \mathbb{N}\);
2. \(\alpha(s_k)(p) \to p'\) as \(k \to \infty\);
3. \(\mu_p^{\xi} = \mu_{\alpha(s_k)}(p)\) for every \(k \in \mathbb{N}\).
It follows that \(\mu_p^{\xi} = \mu_p^{\xi}\). Taking sets \(K_{\xi}\) above of measure arbitrarily close to 1, for typical points \(p\), it follows that \(\mu_p^{\xi} = \mu_p^{\xi}\) for all \(p' = \alpha(v)(p)\) for a \(\mu_p^{G^\xi}\)-conull set of \(v\). It follows that for almost every \(p\), the group of isometries of \(G^\xi\) preserving \(\mu_p^{G^\xi}\) up to
normalization acts transitively on the support of $\mu^G_p$ in $G^\xi$. In fact, the group of right-translations of $G^\xi$ preserving $\mu^G_p$ up to normalization acts transitively on the support of $\mu^G_p$ in $G^\xi$.
It now follows from arguments developed in [KS2, Section 5] that $[\mu^G_p]$ coincides with the Haar measure on a non-trivial subgroup $V(p) \subset G^\xi$. See also [KS1, Section 6.1] for an argument in the framework described here. Moreover, the assignment $p \mapsto V(p)$ is measurable and constant on $A$-orbits. From Proposition 4.2 it follows that $\mu$ is $G^\xi$-invariant. □
Recall our initial choice of $\xi$ was such that $G^\xi \not\subset Q = Q_{\text{In}}(\mu)$. From this contradiction we conclude that $Q_{\text{In}}(\mu) = Q_{\text{Out}}(\mu)$. This completes the proof of Proposition 5.3.
6.3. Proof of Proposition 5.4. The proof of Proposition 5.4 is a direct application of the Theorem 4.3. Recall the definitions of $\Sigma_Q$ and $\Sigma^G_Q$ above. Suppose that $Q_{\text{In}}(\mu) \neq Q_{\text{Out}}(\mu)$. Let $\xi \in \hat{\Sigma}$ be as in Claim 6.1. Then $\xi \subset \Sigma^\perp_{Q_{\text{In}}(\mu)}$. Write $Q = Q_{\text{In}}(\mu)$.
If $\xi$ contains two elements, we have $\xi = \{\beta', 2\beta'\}$ for some root $\beta' \in \Sigma_-$. In the latter case, take $\beta = 2\beta'$ if $[\mu^G_p]$ is supported on $G^{2\beta'}$ for almost every $p$ and $\beta = \beta'$ otherwise. If $\xi$ is a single root take $\beta$ with $\xi = \{\beta\}$.
We claim
Claim 6.4. If $Q_{\text{In}}(\mu)$ is maximal then, with $\beta$ as above, there is a nonzero root $\gamma \in \Sigma_{Q_{\text{In}}(\mu)}$ with
1. $\gamma \neq -c\beta$, for any $c > 0$;
2. $\gamma + \beta \in \Sigma$;
3. $\gamma + \beta \in \Sigma^\perp_{Q_{\text{In}}(\mu)}$.
Proof. Indeed let $q = \text{Lie}(Q_{\text{In}}(\mu))$. Then $q = q^\perp_{\Pi, \{\alpha\}}$ for some simple root $\alpha$. If $\beta = -\alpha$ then, as we assume the are no rank-1 simple ideals, there is simple positive root $\hat{\alpha} \neq -\alpha$ adjacent to $\alpha$ in the Dynkin diagram corresponding to the simple factor containing $\alpha$. Then $\hat{\alpha} - \beta = \hat{\alpha} + \alpha$ is a root. Take $\gamma = -\hat{\alpha}$. Then (since $q$ is of the form $q^\perp_{\Pi, \{\alpha\}}$) $\gamma = -\hat{\alpha} \in \Sigma_{Q_{\text{In}}(\mu)}$ and $\gamma + \beta \in \Sigma^\perp_{Q_{\text{In}}(\mu)}$. Similarly, if $\beta = -2\alpha$ (so that $\beta$ is a root in factor of type $BC_n$) then $\alpha$ is the right-most root in the Dynkin diagram; with $\hat{\alpha}$ the root adjacent to (that is, to the left of) $\alpha$, since $\hat{\alpha} + 2\alpha$ is a root, $\gamma = -\hat{\alpha}$ satisfies the conclusions of the claim.
If $\beta \neq -\alpha$ and $\beta \neq -2\alpha$ then $\beta$ is of the form
$$\beta = c_\alpha \alpha + \sum_{\alpha \neq \alpha \in \Pi} c_\alpha \hat{\alpha}$$
where $c_\alpha < 0$, $c_\hat{\alpha} \leq 0$, and $\sum_{\alpha \neq \alpha \in \Pi} c_\alpha \geq 1$. Since $\beta$ is not a simple negative root, there is a simple (positive) root $\alpha' \in \Pi$ such that $\beta + \alpha'$ is a negative root. If $\alpha' \neq \alpha$ then, since $\beta = (\beta + \alpha') - \alpha'$ and $-\alpha' \in \Sigma_{Q_{\text{In}}(\mu)}$, it follows that $$(\beta + \alpha') \notin \Sigma_{Q_{\text{In}}(\mu)}$$
since $Q_{\text{In}}(\mu)$ is a subgroup. Then $\gamma = \alpha'$ satisfies the conclusion of the claim. On the other hand, if $\alpha' = \alpha$ then, since $\beta + \alpha$ is a negative root,
$$-(\beta + \alpha) \in \Sigma_+ \subset \Sigma_{Q_{\text{In}}(\mu)}$$
and
$$\beta + - (\beta + \alpha) = -\alpha \notin \Sigma_{Q_{\text{In}}(\mu)}.$$
Since $\beta$ is linearly independent of $\alpha$, the root $\gamma = -(\beta + \alpha)$ satisfies the conclusion of the claim. \hfill $\Box$
As we assume that $\gamma \neq -c\beta$ for $c > 0$, it follows that we may find $s \in A$ with $\beta(s) < 0$ and $\gamma(s) < 0$. Let $h$ be the Lie subalgebra generated by $g^\xi \oplus [g^\gamma]$ where $[\gamma]$ is the coarse equivalence class of $\gamma$. Then $H = \exp(h)$ is the minimal subgroup containing $G^\xi$ and $G^\gamma$ that is contracted by all $s$ with $\beta(s) < 0$ and $\gamma(s) < 0$. For $p \in M^\alpha$, let $H_p \subset S_p \subset H$ be the subgroups guaranteed by Theorem 4.3.
Let $\beta = \beta + \gamma$. As $\beta \in \Sigma$, from our choice of $\beta$ and that $\gamma \in \Sigma_Q$ there are $g \in G^\beta \cap S_p$ and $h \in G^\gamma \cap S_p$ that do not commute. Theorem 4.3 implies that $\text{Lie}(H_p)$ contains a nontrivial intersection with $g^\beta = [g^\beta, g^\gamma]$. In particular, one can find a measurable $A$-invariant family of subgroups $V(x) \subset G^\beta$ satisfying the hypotheses of Lemma 4.2. From Lemma 4.2, it follows that the measure $\mu$ is $G^\beta$-invariant contradicting the choice of $\gamma$. This completes the proof of Proposition 5.4.
APPENDIX A. TABLES OF ROOT DATA FOR CLASSICAL ROOT SYSTEMS
A table of simple roots and all positive roots is given in Table 1. We express the roots in terms of a standard presentation (c.f. [Kna][Appendix C].) In all cases, the parabolic subalgebra q of minimal resonant codimension is q = q_{\Pi \setminus \{\alpha_1\}} from which we immediately verify r(g) in Example 1.3 from Table 1. We also verify that m(g) is the resonant codimension of q = q_{\Pi \setminus \{\alpha_1, \alpha_2\}} except for D_4 from which we verify m(g) in Examples 1.5.
**Table 1.** Roots systems and positive roots for classical root systems
| Simple roots and Dynkin diagram | Positive roots |
|---------------------------------|----------------|
| $A_\ell$ | $\alpha_i + \cdots + \alpha_k = e_i - e_{k+1}$ $1 \leq i < k \leq \ell$ |
| $B_\ell$ | $\alpha_i + \cdots + \alpha_k = e_i - e_{k+1}$ $1 \leq i < k \leq \ell - 1$ |
| $C_\ell$ | $\alpha_i + \cdots + \alpha_k = e_i - e_{k+1}$ $1 \leq i < k \leq \ell$ |
| $D_\ell$ | $\alpha_i + \cdots + \alpha_k = e_i - e_{k+1}$ $1 \leq i < k \leq \ell - 2$ |
**References**
[AV] A. Avila and M. Viana. Extremal Lyapunov exponents: an invariance principle and applications. *Invent. Math.*, 181(1):115–189, 2010.
[BC] T. Bogenschütz and H. Crauel. The Abramov-Rokhlin formula. In *Ergodic theory and related topics*, III (Güstrow, 1990), volume 1514 of *Lecture Notes in Math.*, pages 32–35. Springer, Berlin, 1992.
[BJ] A. Borel and L. Ji. *Compactifications of symmetric and locally symmetric spaces*. Mathematics: Theory & Applications. Birkhäuser Boston, Inc., Boston, MA, 2006.
[BFH] A. Brown, D. Fisher, and S. Hurtado. Zimmer’s conjecture: Subexponential growth, measure rigidity, and strong property (T). *Preprint*, 2016. arXiv:1608.04995.
UNIVERSITY OF CHICAGO, CHICAGO, IL 60637, USA
E-mail address: [email protected]
Pennsylvania State University, State College, PA 16802, USA
E-mail address: [email protected]
Pennsylvania State University, State College, PA 16802, USA
E-mail address: [email protected] | 2025-03-06T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3451,
1
],
[
3451,
7964,
2
],
[
7964,
12362,
3
],
[
12362,
16332,
4
],
[
16332,
19558,
5
],
[
19558,
23325,
6
],
[
23325,
27310,
7
],
[
27310,
31439,
8
],
[
31439,
35361,
9
],
[
35361,
38519,
10
],
[
38519,
42406,
11
],
[
42406,
46271,
12
],
[
46271,
49853,
13
],
[
49853,
53684,
14
],
[
53684,
57290,
15
],
[
57290,
60490,
16
],
[
60490,
64836,
17
],
[
64836,
67909,
18
],
[
67909,
71402,
19
],
[
71402,
75721,
20
],
[
75721,
79764,
21
],
[
79764,
84006,
22
],
[
84006,
88769,
23
],
[
88769,
93006,
24
],
[
93006,
96718,
25
],
[
96718,
97978,
26
],
[
97978,
99849,
27
],
[
99849,
99849,
28
],
[
99849,
100116,
29
]
]
} |
eba3794dea0ed464c587fa9bc35ddc596188aa5d | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 4,
"total-fallback-pages": 0,
"total-input-tokens": 14871,
"total-output-tokens": 4515,
"fieldofstudy": [
"Physics"
]
} | Self-breaking in planar few-atom Au constrictions for nm-spaced electrodes
K. O’Neill, E. A. Osorio, and H. S. J. van der Zant
Kavli Institute of Nanoscience Delft, Delft University of Technology, Lorentzweg 1, 2628 CJ Delft, The Netherlands
We present results on electromigrated Au nanojunctions broken near the conductance quantum 77.5 $\mu$S. At room temperature we find that wires, initially narrowed by an actively-controlled electromigration technique down to a few conductance quanta, continue to narrow after removing the applied voltage. Separate electrodes form as mobile gold atoms continuously reconfigure the constriction. We find, from results obtained on over 300 samples, no evidence for gold cluster formation in junctions broken without an applied voltage, implying that gold clusters may be avoided by using this self-breaking technique.
Electronic devices based on single nanometer-sized molecules show promising routes to exploiting the functionality available through organic synthesis. In addition such devices provide an experimental platform to understand the electronic, ionic and mechanical degrees of freedom of a single molecule and its coupling to the environment. While many methods exist to create nanometer-spaced electrodes, the presence of a gate electrode is crucial to correctly identify the signatures of single-molecule conduction at low temperature, implying that a planar geometry is required. In this geometry, a common method for creating the nanometer sized electrode spacing is by electromigration, in which a large current density of $\sim 10^8$ $\text{Acm}^{-2}$ is used to deform a small gold wire until physically separated electrodes are formed. It has remained a persistent challenge to unambiguously determine the presence of a trapped single molecule in transport measurements using the electromigration technique. Transport measurements are hindered by the uncontrollable nature of the breaking process, which produces nanogaps of a wide range of sizes, and the formation of gold clusters that give signatures of Coulomb Blockade and Kondo physics, unrelated to conduction processes through single molecules.
This paper demonstrates a ‘self-breaking’ effect in gold wires that are narrowed by electromigration to a few atoms. Gold nanoconstrictions fabricated at room temperature tend to be unstable; on a time-scale of tens of minutes or hours they break further until the conductance reaches values $< 100 \mu$S without an applied voltage. Subsequent measurements at low temperature show that self-broken wires are less likely to produce gold clusters than samples that are actively broken into separate electrodes. This ‘self-breaking’ effect has also been observed in transmission electron microscope studies, indicating that the decreasing conductance is due to a physical separation of the electrodes in time.
Samples are fabricated as follows: thin wire bridges, with a cross-section of 100 nm $\times$ 12 nm and a typical length of 500 nm are evaporated over a gate dielectric. The gates are formed using aluminium wires and its native aluminium oxide of a few nm is used as the gate dielectric. Contact to the bridge is made by large wires which contribute a small series resistance. The resistance between bonding pads before the wires are narrowed is typically $100 - 200 \Omega$. To initially narrow the gold wires by electromigration, we use an active breaking scheme, similar to the ones previously reported. In the active breaking process the voltage is increased from below the electromigration threshold ($> 200 \text{mV}$) while sampling the current; the maximum current is $\sim 8 \text{mA}$. If the absolute resistance of the wire increases by a value determined during the sweep, typically around 10%, the applied voltage is reduced back to $100 \text{mV}$, and the sweep is repeated with a new value for the wire resistance. With this method, nanoconstrictions may be narrowed to a target conductance with high reproducibility, often within 10%, provided the target conductance is greater than the conductance quantum $G_0 = \frac{e^2}{h} = 77 \mu$S. Electromigration events change the junction resistance on a time-scale $< 100 \mu$s, and the rate typically chosen for voltage output and current sampling is $\sim 25 - 50 \mu$s. Figure (a) shows Atomic Force Microscope images of a sample before (I) and after (II) narrowing using this technique. We note the formation of a hillock downstream of the gap, demonstrating that electron wind force is responsible for the transfer of momentum between electrons and gold atoms.
While active breaking allows the resistance of a constriction to be precisely controlled, the mobility of gold at room temperature is high enough to break the wire completely, resulting in two separate electrodes, even without applying a bias. Figure (b)-(d) shows the conductance versus time of three wires at room temperature when initially narrowed by active breaking to $4 \text{k}\Omega$, $2 \text{k}\Omega$ and $900 \Omega$. Here, we plot the numerical $\frac{dI}{dV}$ around zero bias obtained by sweeping the applied voltage from $-100 \text{mV}$ to $100 \text{mV}$ while measuring the current through the constriction, at a rate of 1 sweep/s. The sample in Figure (c) was narrowed first to $700 \Omega$ and then measured with a lock-in amplifier using an oscillation of 1 mVRMS around zero bias. To emphasize that self-breaking occurs even in the absence of an applied electric field, we have carried out several hundreds of experiments in which no bias is applied after narrowing, and have observed self-breaking in all cases.
In each sample, the resulting reduction in conductance is not continuous but evolves in discrete steps. Configurations may persist for long times before changing, and the time to reach less than one conductance quantum
FIG. 1: (a) Atomic Force Microscope images of a sample before and after narrowing. (b)-(e) show self-breaking data in which conductance (in S on the left axis and conductance quanta $2e^2/h$ on the right axis) is plotted against time (in hours and minutes), measured at room temperature and in a pressure less than $10^{-5}$ Torr. The straight lines in (d) and (e) are fits up to the points near the conductance $2G_0$.
may vary between tens of minutes, to tens of hours. The rate of conductance drop until $2G_0$ is $\sim 2000$ s/G$_0$ for (b) and (c), $8700$ s/G$_0$ for (d), and $1000$ s/G$_0$ for (e). We note that, as the junction approaches a conductance of $\sim 1G_0$, the junctions are typically only stable for a few seconds. This behavior corresponds to the removal of the final atom between the two electrodes until charge can only be transferred by tunneling processes.
Despite the unstable nature of the few-atom constrictions at room temperature, fast acquisition allows $I-V$s to be recorded as the system evolves. Figure 2(a) shows a selection of the $dI/dV$ vs $V$ curves collected during self-breaking of the sample presented in Figure 1(c) at 300 K, computed from the $I-V$ curves as previously described. As we have demonstrated above, wires narrowed at room temperature are unstable, further characterization may be performed at low temperature. In contrast to room temperature, gold wires narrowed by active breaking at 1.6 K are stable, and their $I-V$ characteristics do not vary in time. In this conductance range, constrictions of a specific target resistance can be created by actively-controlled electromigration to within 5%, even in the conductance range near $1G_0$. The low-temperature data in Figure 2(b) was collected with a lock-in amplifier in the order indicated by the arrow. We see that, in both the unstable room temperature constrictions, and in the stable low temperature constrictions, the $dI/dV$ - $V$ curves are flat for $dI/dV \gtrsim 5G_0$, and become highly non-linear near $1G_0$. In addition, measurements of the samples narrowed at low temperature to $dI/dV \sim 1G_0$ at 1.6 K reveal negligible magnetic field dependence up to 10 T. For electron interference to be the origin of the observed oscillations, loops of 20 nm or less are implied.
Finally, we present statistics on samples broken at room temperature with single organic molecules, either by breaking the wires with a self-assembled monolayer, or by breaking directly in a solution of the molecules of interest, collected over approximately a year. After breaking, samples were cooled to 1.6 K and the current was measured as a function of gate and bias for each device. Of 162 samples broken by narrowing the constriction to a few atoms wide and then allowed to self-break with no bias applied, 24 showed gate dependence, and none showed indications of transport through gold clusters (see below). In contrast, of 171 samples broken by active breaking into the tunneling regime (applying above-threshold biases until $> 100$ kΩ), 38 show gate dependence, of which 6 samples show indications of transport through gold clusters.
In discriminating between transport through gold clusters and transport through single molecules we identify two parameters of a double-tunneling system with a single charging island: the electron addition energy and the coupling to the gate electrode. We attribute charging energies less than 100 meV in combination with a gate coupling greater than 0.2 to gold grains. We reach these parameters as follows: by identifying single-molecule samples through their vibrational spectra, it has been established that, for the molecules studied here, electron addition energies typically lie in a range greater than 100 meV, comparable with electron addition energies measured using other techniques. In addition, transport that is due to single molecules has consistently shown a low gate coupling, not larger than 0.15. These clusters, on the other hand, would be in direct con-
tact with the gate dielectric, and so would be expected to have a higher gate coupling, which we typically find to be 0.2513.
We have also observed self-breaking in narrowed gold wires on silicon oxide substrates, instead of aluminium gates, revealing comparable results to those presented here. In addition, experiments with platinum wires have shown no self-breaking effect, indicating that the effect is connected with some intrinsic property of the wire material. The self-breaking of narrowed wires can be compared with diffusion of gold, as observed in scanning tunneling microscope (STM) experiments14. Here the surface diffusion velocities were measured in the range from 0.03 to 0.2 Å/s. Recalculating this as the time required to jump the nearest neighbour distance 2.88 Å, surface mobility implies that it takes 10 – 100 s for gold atoms to move in and out of a contact, contributing to or reducing the conductance by $G_0$. This is within an order of magnitude faster than the observed rate shown in Figure 1 for conductances > 2 $G_0$, and an order of magnitude slower than for junctions with conductance < 2 $G_0$. Given these variations, there is rough agreement between this crude calculation and the measurements which seems to suggest that a diffusion mechanism drives the self-breaking process.
The traces in Figure 1(b) and (c) seem to imply a preference for conductances at integer multiples of $G_0$, consistent with previous observations. Figure 3(a) shows a histogram in which the occurrence of conductances is plotted in a histogram, using the data presented in Figures 1(b)-(d). Figure 3(b) shows a similar histogram of conductance values built from 8 samples with a driving field close to the electromigration threshold, using a much higher sampling rate. It is clear that, while some of the conductance peaks fall on integer multiples of $G_0$, many others do not, and it is therefore difficult to conclude that preferred conductances exist. For comparison, STM15 and MCBJ experiments show that at least ~ 230 traces have to be considered, implying that the statistics in our experiments may be too low. In addition, in such experiments only the first three conductance peaks are visible, for which we obtain fewer statistics.
The cause of the non-linear $\frac{d^2I}{dV^2}$ vs $V$ curves shown in Figure 2 is unclear. An explanation for the effect may come from electron interference in ballistic atomic-scale junctions16, which may be of the same order of magnitude as the measurements presented here. Further measurements, at room and low temperatures, have not revealed a clear suppression of the non-linearities near integer multiples of $G_0$, which would strongly corroborate the electron interference theory. An alternative cause could stem from contamination in the junction. STM experiments establish that contamination by air (notably water vapour) both modifies the surface of gold samples17 and changes the second derivative $G''$ of one and two atom junctions17. We believe that this is unlikely to explain the variation in $G''$ observed in our junctions, which is ~ 1000 times higher than reported values (reference 17).
Financial support was obtained from the Dutch organization for Fundamental Research on Matter (FOM), and the ‘Nederlandse Organisatie voor Wetenschappelijk Onderzoek’ (NWO). We thank M. Poot for the data on silicon oxide substrates. K.O‘N. was supported by the Marie Curie Fellowship organization. The authors are grateful to A. Ito, T. Bjørnholm and M. Ruben for providing the molecules.
* Electronic address: [[email protected]](mailto:[email protected])
1 H. Park, A. K. L. Kim, A. P. Alivisatos, J. Park, and P. L. McEuen, App. Phys. Lett. 75, 301 (1999).
2 J. I. Gonzalez, T.-H. Lee, M. D. Barnes, Y. Antoku, and R. D. Dickson, Phys. Rev. Lett. 93, 147402 (2004); R. Sordan, K. Balasubramanian, and K. Kern, Appl. Phys. Lett. 87, 013106 (2005).
3 H. Heersche, Z. de Groot, J. A. Folk, L. P. Kouwenhoven, H. S. J. van der Zant, A. A. Houck, J. Labaziewicz, and I. L. Chuang, Phys. Rev. Lett. 96, 017205 (2006).
4 D. R. Strachan, D. E. Smith, M. D. Fischbein, D. E. Johnston, B. S. Guiton, M. Drndić, D. A. Bonnell, and A. T. Johnson, Jr., Nanolett. 6, 441 (2006); H. Heersche, G. Lentschung, K. O’Neill, H. S. J. van der Zant, and H. W. Zandbergen, unpublished (2006).
5 M. Trouwborst, S. J. van der Molen, and B. J. van Wees, J. Appl. Phys. 99, 114316 (2006).
6 H. S. J. van der Zant, Y.-V. Kervennic, M. Poot, K. O’Neill, Z. de Groot, H. B. Heersche, N. Stuhr-Hansen, T. Bjørnholm, D. Vanmaekelbergh, C. A. van Walree, and L. W. Jenneskens, Faraday Discussions 131, 347 (2006).
7 D. R. Strachan, D. E. Smith, D. E. Johnston, T.-H. Park,
M. J. Therien, D. A. Bonnell, and A. T. Johnson, Applied Physics Letters 86, 043109 (2005); A. A. Houck, J. Labaziewicz, E. K. Chan, J. A. Folk, and I. L. Chuang, Nano Letters 5, 1685 (2005).
Assuming magnetic flux $\frac{\Phi}{\pi}$ enclosed by circular interfering electron paths in a magnetic flux density 10 T.
8 Oligophenylenevinylene 3, oligophenylenevinylene 5, spiro-fused triarylamine, $[\text{Co}^{II}_4\text{L}_4]^{8+}$, $[2 \times 2]$-grid type complexes, Co-terpyridine.
9 K. Likharev, Proc. IEEE 87, 606 (1999).
10 E. A. Osorio, K. O’Neill, N. Stuhr-Hansen, O. F. Nielsen, T. Bjørnholm, and H. S. J. van der Zant, Adv. Mat. 19, 281 (2007).
11 S. Kubatkin, A. Danilov, M. Hjort, J. Cornil, J.-L. Brédas, N. Stuhr-Hansen, P. Hedegård, and T. Bjørnholm, Nature 425, 698 (2003).
12 H. S. J. van der Zant, E. A. Osorio, M. Poot, and K. O’Neill, phys. stat. sol. (b) 248, 3408 (2006); K. I. Bolotin, F. Kuemmeth, A. N. Pasupathy, and D. C. Ralph, Appl. Phys. Lett. 84, 3154 (2004).
13 C. Roberts, B. Hoffmann-Millack, and W. S. Steer, J. Vac. Sci. Technol. B 9, 841 (1991).
14 M. Brandbyge, J. Schiøtz, M. R. Sorenson, P. Stoltze, K. W. Jacobsen, J. K. Nørskov, L. Olesen, E. Laegsgaard, I. Stensgaard, and F. Besenbacher, Phys. Rev. B 52, 8499 (1995).
15 B. Ludoph, M. H. Devoret, D. Esteve, C. Urbina, and J. M. van Ruitenbeek, Phys. Rev. Lett. 82, 1530 (1999); C. Untiedt, G. Rubio Bollinger, S. Vieira, and N. Agrait, Phys. Rev. B 62, 9962 (2000).
16 K. Hansen, S. K. Nielsen, M. Brandbyge, E. Laegsgaard, I. Stensgaard, and F. Besenbacher, Appl. Phys. Lett. 77, 708 (2000). | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
5862,
1
],
[
5862,
9883,
2
],
[
9883,
14613,
3
],
[
14613,
16209,
4
]
]
} |
e2d446c3444cd7307b62a9c7904a12be38585ffc | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 7,
"total-fallback-pages": 0,
"total-input-tokens": 33636,
"total-output-tokens": 9545,
"fieldofstudy": [
"Physics"
]
} | The possibility of engineering experimentally viable systems that realize gauge fluxes within plaquettes of hopping have been subject of search for decades due to vast amounts of theoretical study. This is of particular interest for topological band insulators, where it is known that such fluxes bind protected mid-gap states. These modes can hybridize in extended flux lattices, giving rise to semi-metallic bands that are highly tunable. We demonstrate that within artificial materials, local \( \pi \)-fluxes can be naturally realized. Consequently, we provide concrete set-ups to access this physics and analyze similar self-organized band structures and physical properties. Our work therefore does not only pinpoint simple systems exhibiting flat bands, but also opens up a route to study flux lattice models and associated effective theories in a novel scene.
**Introduction**—Gauge fluxes have been associated with many exotic physical properties studied over several decades, touching upon topics ranging as wide as from e.g visons in lattice gauge theories \([1]\) to quasi-particle descriptions in quantum Hall systems \([2]\). Similarly, topological band insulators \([3, 4]\), arising by virtue of an interplay between symmetry and topology \([5–13]\), or symmetry protected states in general \([14–16]\), can be unambiguously distinguished from trivial states by monodromy defects \([17–21]\), i.e effective gauge fluxes. Such defects bind modes that relate to the relevant topological invariants.
Despite the richness of such theoretical insights, realizing gauge fluxes in accessible set-ups is still a standing question. The difficulty here mainly lies in the condition that the mentioned model settings require fluxes that are localized to a (or a few) plaquette of hopping sites. Although intriguing proposals utilizing the controllability of cold atom systems \([22]\) or vortex lattices in type-II superconductors \([23]\) maybe become more within reach over time, such experiments will unavoidably be rather involved. This in some sense also applies to routes that use lattice defects to act as effective fluxes \([19, 24–27]\). While recent experimental results in fact seem promising \([28]\), controlling and manipulating defects will require additional progress.
In this work we analyze physical consequences of localized fluxes within the context of readily controllable artificial materials. We retrieve \( \pi \)-flux bound midgap ‘zero’ modes in topological regimes, as in the electronic counterparts. Apart form providing concrete setups, we show that in extended flux lattices these modes hybridize into midgap band structures that have gapless touching points at specific points in the Brillouin zone. This leads to new transport features of the highly tunable bands that can readily be flattened by increasing the spacing of the flux lattice. Given the implementability of the outlined structures this opens the door to study gauge flux models in a controllable set-ups and harvest the associated rich physics.
**Model**—To make our arguments concrete, we analyze band dispersions of a modulated honeycomb lattice tight binding model with \( \pi \)-flux lattice on top of it. We consider the simple Hamiltonian
\[
H = \sum_{\langle ij \rangle} t_{ij} c_i^\dagger c_j,
\]
where \( \langle ij \rangle \) refers to the nearest neighbours and \( c_i^\dagger \) is the creation operator at site \( i \). Furthermore, the hopping parameters \( t_{ij} \) are set in terms of modulations of the type depicted in Fig. 1, in which we assign intra-hexagon hopping \( t_0 = t + \delta \) for black bonds and inter-hexagon hopping \( t_1 = t - \delta \) for red bonds. This modulation gaps out Dirac cones of the original honeycomb lattice model through the coupling between the Dirac cones at the K and K’ points in the Brillouin zone \([29–31]\). Importantly, the states near zero energy are described by a massive Dirac equation where the sign of the mass term can be flipped by changing the ratio between \( |t_0| \) and \( |t_1| \), or the sign of \( \delta \). The feature that the sign flip of the mass term can be induced by a little modulation in the hopping amplitude makes this model an ideal playground to simulate a quantum spin Hall state in artificial systems. Indeed, several of the experimental demonstrations of helical edge states in the literature in fact implement this model \([32–39]\).
In the following, we set \( t = -1 \), and \( |\delta| = 0.1 \). Note
that $\delta > 0$ corresponds to $|t_0| < |t_1|$ while $\delta < 0$ corresponds to $|t_0| > |t_1|$. Throughout the paper, we use $a_0$ as the nearest neighbor distance between the sites in the honeycomb lattice.
To implement a $\pi$-flux lattice, it is convenient to think of the conjugate expression by strings. Namely, we pick up two hexagons and connect them by a string, and every time the string hits the bond, we flip the sign of the hopping on that bond. Then we are left with two $\pi$-flux threaded hexagons at the ends of the string. There is arbitrariness in the path of the strings, i.e., if two strings form a loop when they are combined, the two ways are related by a gauge transformation and physically equivalent. For instance, Figs. 1(b) and 1(c) are physically related by a gauge transformation and physically equivalent. For instance, Figs. 1(b) and 1(c) are physically the same. As in the case of the uniform magnetic field, in order to make the Hamiltonian strictly periodic (for calculating band structures), the supercell of the $\pi$-flux lattice has to contain an integer multiple of full flux ($2\pi$), indicating that the supercell should contain two $\pi$-flux threaded hexagons. In the following, when we see the band structure as a function of $k_x$, we implicitly chose the string alignment as in Fig. 1(c) to minimize the period in $x$-direction, while when we see the band structure as a function of $k_y$, the alignment in Fig. 1 is chosen to minimize the period in $y$-direction.
Because of the hopping modulation, hexagons in the current model can be classified into two kinds, one consists of the black bonds exclusively, while the other consists of the mixture of the black and the red bonds. We first consider the case that the $\pi$ fluxes are placed on the hexagon of the first kind, as in Fig. 1, and check the existence of the midgap states by looking at the isolated $\pi$-flux. In practice, we numerically obtain the energy spectrum and the wave functions with sufficiently large $L_x$ and $L_y$. A clear sign of the midgap state is found for $|t_0| > |t_1|$. There appear two bound states per a $\pi$-flux, whose wave functions are shown in Fig. 2.
For the case with $L_x \ll L_y$, the system is essentially regarded as a one-dimensional $\pi$-flux lattice ($\pi$-flux chain) along the zigzag direction, while for the case with $L_x \gg L_y$, the system is regarded as a $\pi$-flux chain along the armchair direction. In such cases, we can think of band structures generated by hybridizations between the bound states of the neighboring $\pi$-fluxes. The results for $L_x \ll L_y$, i.e., the band structures of the $\pi$-flux chain along the zigzag direction are shown in Fig. 3. We clearly see an in-gap state with 1D Dirac structure at the zone center and the corner for $|t_0| > |t_1|$, while no in-gap state for $|t_0| < |t_1|$ is retrieved. Similar band structures are obtained also for $L_x \gg L_y$, i.e., for the $\pi$-flux chain along the armchair direction. However, in the armchair case, we numerically obtain a mini gap within the in-gap band.
at the zone center and boundary ($k=\pi$). Figure 3(c) shows the gap $\Delta$ across zero energy at the zone center. For the chain along the zigzag direction, $\Delta$ exponentially decays as function of the interchain distance $L_y$, which signals gapless nature of the single chain. On the other hand, for the chain along the armchair direction, $\Delta$ saturates as a function of the interchain distance $L_x$, conveying that the mini gap is not coming from size effects.
The $\pi$-flux may also be put on the hexagons of the second kind, for instance by shifting the strings in Figs. 1(b) or 1(c) by the size of a single hexagon. The band structures for the $\pi$-flux chain threading the hexagons of the second kind are shown in Fig. 4. We see that some states pop up (and down) from the bulk continuum to the gap for $|t_0| < |t_1|$, but there is no clear sign of zero gap or near zero gap states. A consideration on the extreme limit of $t_0=0$ or $t_1=0$ gives some insights for the difference between threading the hexagons of the first and the second kind, and also the difference between $|t_0| > |t_1|$ and $|t_0| < |t_1|$. If we have $t_1=0$, the system is a set of isolated hexagons of the first kind. In such a case, it is easy to convince ourselves that the $\pi$-flux threading of those hexagons gives rise to zero energy states, and nothing happens if the $\pi$-fluxes are misplaced. On the other hand, if we have $t_0$, the system is a set of dimers with no loops, making $\pi$-fluxes ineffective. Whether or not the zero energy state in the extreme limit survives in the interested parameter region is a nontrivial issue, and topology and symmetry come into play to assist the zero energy state to survive, as we will see in the following.
**Stability**—The stability of the $\pi$-flux modes finds its origin in the existence of edge states and thus hangs directly together with the topological phase [17, 18, 24, 25, 27, 40, 41]. In fact they provide for a direct bulk probe of the non-trivial symmetry protected topology. As outlined in [24], this can readily seen using a Volterra argument. Specifically, one can consider the effect of a $\pi$-flux by hypothetically cutting the system and then gluing back the two sides after the hopping elements over the segment that connects the boundary and defect core are multiplied with a $\pi$-phase, $e^{i\pi}$. Considering the case when the cutting the system results in edge states, gluing the sides together conventionally results in a mass term gapping the spectrum back to the original bulk situation. However, as the phases induce a minus sign for the hoppings on one side of the defect core this mass term now switches sign. As result, we obtain an effective Dirac system with changing mass term,
$$H_0 = v k \sigma 3k + m(x)\mu_1,$$
where $v$ specifies the velocity, $\mu$ refers to the two edges of the Volterra cut and $\sigma$ parametrizes the pseudo spin. Such Dirac theories with changing mass terms are well known to host localized soliton states at the mass domain wall, being the $\pi$-flux mode.
Our remaining task is to show the stability of the gapless edge mode supporting the description by Eq. (2). Here we give two complementary explanations, one using a pseudo time reversal symmetry (TRS) and another using a mirror winding number. As for the former one, the system at hand is characterized by a $\mathbb{Z}_2$ classification, which is associated with a pseudo time reversal symmetry (TRS), see also Supplemental Material. In particular, time reversal, represented by complex conjugation $K$, can be lifted to an anti-unitary operation under $C_2$ point group symmetries for $n=4,6$ [7]. The bands with complex values under the rotation operator associated with the $E$ representation, as for e.g. as $|p_x\rangle$ and $|p_y\rangle$ states, come in pairs and induce a two dimensional real representation due to the combination of both symmetries leading to degeneracies at the $\Gamma$ and $K$ points in the Brillouin zone. These doublets can then give rise to a $\mathbb{Z}_2$ non-trivial phase by ‘switching’, rather similar to spinful TRS systems. This mechanism is also at work in the present context, where doublet at the $\Gamma$ point, arising by virtue of the six-fold symmetry give rise to a non-trivial $\mathbb{Z}_2$ invariant. This argument only relies on doublets ($E$ representation) and has way broader applicability than our specific model, however we should note that this TRS comes from the crystal symmetry and only emerges in the effective model. In fact the edge states can be gapped out by continuing away from the $\Gamma$-point in the Brillouin zone, showing that the robustness of this symmetry is indeed not generally to be taken for granted.
Alternative, more rigorous, support for the gapless edge modes at zigzag edges comes from the combination of the chiral (sublattice) symmetry and the mirror symmetry, whose reflection plane is perpendicular to the zigzag edge [31, 42]. The gaplessness comes from the degeneracy of the two zero modes having opposite chirality (sublattice polarization) and the opposite parity with respect to the reflection, guaranteed by nontrivial mirror winding numbers. In contrast to the pseudo TRS based argument, the mirror winding number is completely robust, although the chiral symmetry is generically a strong constraint.
Finally, we turn to the gaplessness (mini gap) of the midgap bands for the zigzag (armchair) π-flux chain, by focusing on the subspace spanned by the π-flux states. The above arguments suggests that the π-flux state inherits the properties of the edge states, and the two bound states at a single π-flux have opposite chiralities (see Fig. 2). Then, within the effective model, we can take a basis such that the chiral operator γ becomes γ = σ_3, which gives us a chiral symmetric effective Hamiltonian
\[ H(\hat{k}) = h_1(\hat{k})\sigma_1 + h_2(\hat{k})\sigma_2, \]
where \( \hat{k} = k_x \) (\( \hat{k} = k_y \)) for the zigzag (armchair) π-flux chain (see Fig. 1), referring to \( x,y \) direction whereas the value is set by the spacing of fluxes. For the zigzag π-flux chain, we also have the reflection symmetry \( \sigma_3\hat{I} \), where \( \hat{I} \) brings \( \hat{k} \) to \(-\hat{k} \), which commutes with γ at \( k = 0 \), see Supplemental Material. (Note that under the influence of the π-fluxes, the reflection symmetry actually means the spacial reflection associated with a proper gauge transformation.) This gives \( h_{1,2}(\hat{k}) = -h_{1,2}(\hat{k}) \), and retaining the nearest neighbor hoppings, the band structure will result in \( E(\hat{k}) \sim \pm |\sin \hat{k}| \). On the other hand, for the armchair π-flux chain, the reflection symmetry becomes \( \sigma_1\hat{I} \), which anticommutes with γ (see Supplemental Material), giving us \( h_1(\hat{k}) = h_1(\hat{k}) \) and \( h_2(\hat{k}) = -h_2(\hat{k}) \). Taking only the leading contribution to \( h_{1,2}(\hat{k}) \), the dispersion relation becomes \( E(\hat{k}) = \pm \sqrt{m^2 + \hat{k}^2 |\sin \hat{k}|^2} \). These arguments perfectly agree with the numerical results. As a side note, we mention that these considerations can also be phrased using the domain wall argument. That is, one can readily formulate all, symmetry consistent, couplings between the Volterra cuts and then project down to the subspace of the soliton solutions. These argument then lead to the same effective Hamiltonian. Generalizing this description to account for the different symmetries involved for the zigzag and armchair edge then similar undetsmading of the difference between these cases, while this set-up also allows for the addition of other (TRS) symmetry breaking terms and study those effects [27].
**Perspective-** Our demonstration suggests π-flux as a nice building block for designing various kinds of states. For instance, since the π-flux state is localized, simply making the distance between the threaded hexagons away results in flattened bands. Figure 5 shows such examples, i.e., we observe smaller total band width of the in-gap states in the case of larger distance between the threaded hexagons. For the chains along the armchair direction, the mini gap is also reduced, which has also been detected in Fig. 3(c). The advantage of using π-fluxes as building blocks is that the existence of zero energy states are guaranteed by topological protection. In addition, in the same modulated honeycomb lattice model, edge states or interface states at the boundary between two regions with \( |t_0| > |t_1| \) and \( |t_0| < |t_1| \) have been investigated. Hence, it is an interesting future problem to look for new states or phenomena caused by interplay between edge/interface states and π-flux states.
**Experimental implementation-** Obviously, the key ingredient to realize π-flux is feasibility of the sign flip of the hopping. In some artificial systems, schemes to change the sign of the coupling have already been proposed. For instance, Ref. [43] proposes a mechanical system made of disks and springs, and there “normal” and “reversed” springs leading to opposite signs of couplings are introduced. Sometimes, spring-based discrete systems are effectively or essentially realized in continuous elastic media with springs being substituted by beams. Mapping a discrete model to a continuous model is a non-trivial task, but once the mapping is established, current 3D printing techniques will facilitate us to fabricate a designed elastic media [44]. Another possible and promising scheme is to use LC circuits. Ref. [45] proposes a clever wiring between elements to integrate 0-phase and π-phase hoppings into a single system, and furthermore, the authors of Ref. [45] have already implemented a real circuit with both 0-phase and π-phase hoppings to show Berry curvatures and Fermi arcs.
Even in a system where tuning of hoppings itself is difficult, if we have control over site potentials instead, it is in principle possible to build a system with effective π-flux. A rough sketch of the scheme is the following. Let us focus on two sites across which we want to induce hopping. Then, instead of directly connecting them, we put an extra intermediate site connected to both of the focused sites by hopping \( t \). If we set the difference of the site potentials for the intermediate and the focused sites \( \Delta E \) large enough, the second order perturbation leads to an effective hopping between the focused site \( t^2/\Delta E \), which means that the sign of the effective hopping can be chosen by changing the sign of \( \Delta E \).
**Conclusion and Discussion-** We have shown that π-flux modes, characterizing topological band insulators, can be naturally realized in artificial materials. Given the direct implementability of artificial materials, this creates
a promising scene to study π-flux modes as well as the highly tunable bands that arise from their hybridization. More generically, these modes can be used as building blocks, opening up the door to an experimentally viable setting in which numerous models incorporating fluxes (vortex lattices, lattice gauge theories etc.) and the associated effective (field) theories can be brought to life.
We note that in contrast to electron systems, such bosonic set-ups do not have a natural filling, constraining the possibilities of anomalies that also arise due to an interplay with the edge states of the parent model \cite{27}. However, the midgap modes still can be viewed as (higher) Chern numbers depending on the configuration of this flux lattice. Motivated by these results and novel possible routes, we hope that our work will inspire a range of future pursuits.
**Acknowledgements** R.-J.S appreciatively acknowledges funding via Ashvin Vishwanath from the Center for the Advancement of Topological Materials initiative, an Energy Frontier Research Center funded by the U.S. Department of Energy, Office of Science. This work was partially supported by JSPS KAKENHI Grant Numbers JP17K14358 and JP18H01162 (T.K.).
## Appendix
### Symmetry
In this Supplemental Material, the symmetries for the model are detailed. Using the unit vectors $a_1 = e^{i(3\sqrt{3}a_0/2, 3a_0/2)}$ and $a_2 = e^{i(-3\sqrt{3}a_0/2, 3a_0/2)}$, the Hamiltonian of our model is written as
$$H(k_x, k_y) = t_0 \begin{pmatrix} 0 & D \end{pmatrix} \begin{pmatrix} D^\dagger & 0 \end{pmatrix}, \quad D = \begin{pmatrix} \alpha e_1^* e_2 & 1 & 1 \\ 1 & \alpha e_1 & 1 \\ 1 & \alpha e_2 & 1 \end{pmatrix},$$
where the sites in a unit cell is labeled as Fig. 6, $e_i = e^{i k \cdot a_i}$ and $\alpha = t_1/t_0$. This Hamiltonian has (i) the chiral symmetry
$$H(k_x, k_y) = -\gamma H(k_x, k_y) \gamma, \quad \gamma^2 = 1.$$ \quad (5)
with
$$\gamma = \begin{pmatrix} 1 & 0 & 0 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 & 0 & 0 \\ 0 & 0 & 0 & -1 & 0 & 0 \\ 0 & 0 & 0 & 0 & -1 & 0 \\ 0 & 0 & 0 & 0 & 0 & -1 \end{pmatrix},$$
(ii) the mirror symmetry with the reflection plane perpendicular to the zigzag edge
$$H(-k_x, k_y) = R_1 H(k_x, k_y) R_1,$$ \quad (7)
and (iii) the mirror symmetry with the reflection plane perpendicular to the armchair edge
$$H(k_x, -k_y) = R_2 H(k_x, k_y) R_2$$ \quad (8)
where
$$R_1 = \begin{pmatrix} 1 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 1 & 0 & 0 \\ 0 & 0 & 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 0 & 0 & 1 \end{pmatrix}, \quad R_2 = \begin{pmatrix} 0 & 0 & 0 & 1 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 1 \\ 0 & 0 & 0 & 1 & 0 & 0 \\ 1 & 0 & 0 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 & 0 & 0 \end{pmatrix}.$$ \quad (9)
Note that $\gamma$ and $R_1$ are compatible since $[\gamma, R_1] = 0$, while $\gamma$ and $R_2$ are incompatible since $[\gamma, R_2] = 0$. To derive an effective model near the $\Gamma$-point and the zero energy, we focus on the subspace spanned by a new basis set $\{|u_-, |u_+, |l_-, |l_+\rangle\}$ defined as
$$|u_\pm\rangle = \begin{pmatrix} \pm \rangle \\ 0 \end{pmatrix}, \quad |l_\pm\rangle = \begin{pmatrix} 0 \\ \pm |\rangle \end{pmatrix}$$ \quad (10)
and
$$|\pm\rangle = \frac{1}{\sqrt{3}} \begin{pmatrix} 1 \\ \omega_\pm \end{pmatrix}, \quad \omega_\pm = -\frac{1}{2} \pm \frac{\sqrt{3}}{2} i.$$ \quad (11)
which gives us
\[ H(k_x, k_y) = t_0 \begin{pmatrix} 0 & \tilde{D} \\ \tilde{D}^\dagger & 0 \end{pmatrix} \]
(12)
with
\[ \tilde{D} = \begin{pmatrix} -(-|D|-) & -(|D|+) \\ -(+|D|-) & +(|D|+) \end{pmatrix} \]
(13)
We have, up to the second order in \(|k|\),
\[ \langle +|D|+ \rangle = -\langle -|D|- \rangle = -(1 - \alpha) - \alpha \frac{9a_0}{2}(k_x^2 + k_y^2), \]
(14)
and up to the first order in \(|k|\),
\[ \langle -|D|+ \rangle = \alpha \frac{3a_0}{2} k_x, \quad \langle +|D|- \rangle = -\alpha \frac{3a_0}{2} k_x, \]
(15)
leading to
\[ H(k_x, k_y) = \left( (t_0 - t_1) + t_1 \frac{9a_0}{2}(k_x^2 + k_y^2) \right) \sigma_3 \otimes \mu_1 \\
+ t_1 \frac{3a_0}{2} \langle k \cdot \sigma \rangle \otimes \mu_1. \]
(16)
If we further go to the basis set
\[ \begin{pmatrix} |u_-| + |l_-| \\ \sqrt{2} \\ |u_+| + |l_+| \\ \sqrt{2} \\ |u_+| - |l_+| \\ \sqrt{2} \\ |u_-| - |l_-| \\ \sqrt{2} \end{pmatrix}, \]
the Hamiltonian becomes
\[ H(k_x, k_y) = \begin{pmatrix} H_+(k_x, k_y) & 0 \\ 0 & H_-(k_x, k_y) \end{pmatrix}, \]
(17)
with
\[ H_+(k_x, k_y) = \left( (t_0 - t_1) + t_1 \frac{9a_0}{4}(k_x^2 + k_y^2) \right) \sigma_3 \\
+ t_1 \frac{3a_0}{2} (k_x \sigma_1 \pm k_y \sigma_2). \]
(18)
Each of \(H_+(k_x, k_y)\) and \(H_-(k_x, k_y)\) has the (pseudo) time reversal symmetry \(i \sigma_2 \mathcal{K}\) with \(\mathcal{K}\) being complex conjugation.
Using,
\[ |p_x\rangle = \frac{1}{2} \begin{pmatrix} 0 \\ -1 \\ 0 \\ 1 \end{pmatrix}, \quad |p_y\rangle = \frac{1}{2\sqrt{3}} \begin{pmatrix} 2 \\ -1 \\ -1 \\ 1 \end{pmatrix}, \]
(20)
\[ |d_{x^2-y^2}\rangle = \frac{1}{2\sqrt{3}} \begin{pmatrix} -2 \\ 1 \\ 1 \\ -2 \end{pmatrix}, \quad |d_{xy}\rangle = \frac{1}{2} \begin{pmatrix} 0 \\ -1 \\ 1 \\ 0 \end{pmatrix}, \]
(21)
the basis set Eq. (17) is rewritten as
\[ i|p_x\rangle - i|p_y\rangle, \quad \frac{-|d_{x^2-y^2}\rangle - i|d_{xy}\rangle}{\sqrt{2}}, \]
(22)
\[ -i|p_x\rangle + i|p_y\rangle, \quad \frac{-|d_{x^2-y^2}\rangle + i|d_{xy}\rangle}{\sqrt{2}}, \]
(23)
namely,
\[ i|p_-\rangle, \quad -|d_-\rangle, \quad -i|p_+\rangle, \quad -|d_+\rangle. \]
By definition, we have the following relations,
\[ \gamma|p_x\rangle = -|d_{xy}\rangle, \quad \gamma|p_y\rangle = -|d_{x^2-y^2}\rangle, \]
(24)
\[ \gamma|d_{x^2-y^2}\rangle = -|p_y\rangle, \quad \gamma|d_{xy}\rangle = -|p_x\rangle, \]
(25)
\[ R_1|p_x\rangle = -|p_+\rangle, \quad R_1|p_y\rangle = |p_\rangle, \quad R_2|p_x\rangle = |p_x\rangle, \quad R_2|p_y\rangle = -|p_y\rangle, \]
(26)
\[ R_1|d_{x^2-y^2}\rangle = |d_{x^2-y^2}\rangle, \quad R_1|d_{xy}\rangle = -|d_{xy}\rangle, \quad R_2|d_{x^2-y^2}\rangle = |d_{x^2-y^2}\rangle, \quad R_2|d_{xy}\rangle = -|d_{xy}\rangle, \]
(27)
which lead to
\[ \gamma = \sigma_1 \otimes \mu_1, \quad R_1 = \hat{1} \otimes \mu_1, \quad R_2 = -\sigma_2 \otimes \mu_1. \]
(30)
**Symmetry with \(\pi\)-flux**
With \(\pi\)-fluxes or \(\pi\)-phase strings, special care is required regarding the reflection symmetry. In short, we have to augment the spatial reflection by a gauge transformation. Let us consider to apply the reflection along the vertical line on the configuration in Fig. 7. The spatial reflection alone brings Fig. 7(a) to Fig. 7(b), which is not identical to the original configuration. If we further apply the gauge transformation such that the sign of the wave function within the shaded region in Fig. 7(c) is flipped, Fig. 7(b) goes back to Fig. 7(a). Note that this in fact amounts to a large gauge transformation of \(2\pi\), as conveyed by combining the paths in the right column, and that this indeed shows the reflection symmetry of the states in Fig. 2 in the main text. Equipped with this combined operation of the spatial reflection and the gauge transformation, it is straightforward to convince ourselves that the wave function in Fig. 2(a) in the main body is odd against this symmetry operation, while the one in Fig. 2(b) is even against this symmetry operation. This legitimate us to use \(\sigma_1 \hat{I}\) as a reflection symmetry within the subspace spanned by the \(\pi\)-flux states.
---
[1] J. B. Kogut, Rev. Mod. Phys. 51, 659 (1979).
[2] R. B. Laughlin, Phys. Rev. Lett. 50, 1395 (1983).
[3] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010).
FIG. 7. (a) Original configuration. (b) Only the spatial reflection is applied. (c) Schematic description of the gauge transformation. The sign of the basis function is flipped in the shaded region.
[4] X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. **83**, 1057 (2011).
[5] A. Kitaev (AIP, 2009) pp. 22–30.
[6] S. Ryu, A. P. Schnyder, A. Furusaki, and A. W. Ludwig, New Journal of Physics **12**, 065010 (2010).
[7] L. Fu, Phys. Rev. Lett. **106**, 106802 (2011).
[8] R.-J. Slager, A. Mesaros, V. Juričić, and J. Zaanen, NaturePhysics **9**, 98 (2012).
[9] J. Kruthoff, J. de Boer, J. van Wezel, C. L. Kane, and R.-J. Slager, Phys. Rev. X **7**, 041069 (2017).
[10] A. Bouhon, A. M. Black-Schaffer, and R.-J. Slager, arXiv preprint arXiv:1804.09719 (2018).
[11] L. Fu, Phys. Rev. Lett. **106**, 106802 (2011).
[12] R.-J. Slager, A. Mesaros, V. Juričić, and J. Zaanen, Phys. Rev. B **90**, 245104 (2014).
[13] R.-J. Slager, V. Juričić, V. Lahtinen, and J. Zaanen, Phys. Rev. B **93**, 245104 (2016).
[14] H. Hamasaki, Y. Tokumoto, and K. Edagawa, Applied Physics Letters, Applied Physics Letters **110**, 092105 (2017).
[15] L.-H. Wu and X. Hu, Phys. Rev. Lett. **114**, 223901 (2015).
[16] L.-H. Wu and X. Hu, Sci. Rep. **6**, 24347 (2016).
[17] T. Kariyado and X. Hu, Sci. Rep. **7**, 16515 (2017).
[18] S. Yves, R. Fleury, T. Berthelot, M. Fink, L. Levontin, and G. Lerosey, Nature Communications **8**, 16023 EP (2017).
[19] J. Noh, W. A. Benalcazar, S. Huang, M. J. Collins, K. P. Chen, T. L. Hughes, and M. C. Rechtsman, Nature Photonics **12**, 408 (2018).
[20] Y. Yang, Y. F. Xu, T. Xu, H.-X. Wang, J.-H. Jiang, X. Hu, and Z. H. Hang, Phys. Rev. Lett. **120**, 217401 (2018).
[21] D. V. Zhirihiin, M. A. Gorlach, X. Ni, D. A. Smirnova, D. Korobkin, A. P. Slobozhanyuk, P. A. Belov, A. Alì, and A. B. Khanikaev, Journal of Physics: Conference Series **1092**, 012176 (2018).
[22] R. Chaunsali, C.-W. Chen, and J. Yang, New Journal of Physics **20**, 113036 (2018).
[23] R.-J. Slager, L. Rademaker, J. Zaanen, and L. Balents, Phys. Rev. B **92**, 085126 (2015).
[24] J.-W. Rhim, J. H. Bardarson, and R.-J. Slager, Phys. Rev. B **97**, 115143 (2018).
[25] K.-I. Imura, Y. Takane, and A. Tanaka, Phys. Rev. B **84**, 035443 (2011).
[26] R.-J. Slager, A. Mesaros, V. Juričić, and J. Zaanen, Phys. Rev. B **90**, 245104 (2014).
[27] R.-J. Slager, V. Juričić, V. Lahtinen, and J. Zaanen, Phys. Rev. B **93**, 245104 (2016).
[28] S. Barik, A. Krasahin, C. Flower, T. Cai, H. Miyake, W. DeGottardi, M. Hafezi, and E. Waks, Science **359**, 666 (2018).
[29] B. Bradlyn, L. Elcoro, J. Cano, M. G. Vergniory, Z. Wang, C. Felser, M. I. Aroyo, and B. A. Bernevig, Nature Physics **5**, 298 (2009).
[30] Y. Ran, A. Vishwanath, and H. Watanabe, Nature Physics **8**, 16023 (2012).
[31] A. Mesaros, R.-J. Slager, and V. Juričić, Nature Communications **9**, 3072 (2018).
[32] X.-G. Wen, Phys. Rev. B **80**, 155131 (2009).
[33] T.-Y. Wang and Y.-W. Tsai, New Journal of Physics **20**, 083031 (2018).
[34] Y. Lu, N. Jia, L. Su, C. Owens, G. Juzeliūnas, D. I. Schuster, and J. Simon, Phys. Rev. B **99**, 020302 (2019). | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4509,
1
],
[
4509,
7614,
2
],
[
7614,
12997,
3
],
[
12997,
18486,
4
],
[
18486,
21895,
5
],
[
21895,
26132,
6
],
[
26132,
29237,
7
]
]
} |
bee4bd6343a8be9c19bb7fc2ab2d8b66656ddedd | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 10,
"total-fallback-pages": 0,
"total-input-tokens": 34327,
"total-output-tokens": 16479,
"fieldofstudy": [
"Psychology"
]
} | Causal Attributions in an Australian Aboriginal Family With Marfan Syndrome: A Qualitative Study
Aideen M. McInerney-Leo1*, Jennifer West2, Bettina Meiser3, Malcolm West2, Matthew A. Brown4,5 and Emma Duncan4,6
1 Dermatology Research Centre, The University of Queensland Diamantina Institute, The University of Queensland, Brisbane, QLD, Australia, 2 Prince Charles Hospital Clinical Unit, School of Clinical Medicine, The University of Queensland, Brisbane, QLD, Australia, 3 Prince of Wales Clinical School, University of New South Wales, Sydney, NSW, Australia, 4 Translational Genomics Group, Institute of Health and Biomedical Innovation, Queensland University of Technology, Brisbane, QLD, Australia, 5 Guy’s and St Thomas’ NHS Foundation Trust and King’s College London NIHR Biomedical Research Centre, London, United Kingdom, 6 Department of Endocrinology, James Mayne Building, Royal Brisbane and Women’s Hospital, Herston, QLD, Australia
Causal attributions are important determinants of how health threats are processed and affect health-related behaviors. To date, there has been no research on causal attributions in genetic conditions in Aboriginal Australians. Forty members of a large Aboriginal Australian family with Marfan syndrome (MFS) were invited to participate in an ethically approved study exploring causal attributions, including perceived causes of phenotypic variability within the family. Eighteen individuals consented to conduct semi-structured qualitative interviews, which were recorded, transcribed verbatim and analyzed thematically. Most participants knew that MFS was genetic, but there were diverse theories about inheritance, including beliefs that it skipped generations, was affected by birth order and/or gender, and that it co-occurred with inheritance of blue eyes within this family. The mutation was thought to have been inherited from British settlers and initially triggered by disease or diet. Factors believed to modify disease severity included other genes and lifestyle factors, particularly alcohol and substance abuse and stress. Generally, this family did not endorse “blaming” chance or a higher power for phenotypic variability, though some felt that the spirits or a deity may have played a role. In conclusion, although participants knew MFS was a genetic condition, many speculated about the role of non-genetic causes in initiating the original mutation; and the gene-environment interaction was thought to affect severity. This study demonstrates a successful approach for exploring causal attributions in other genetic conditions in First Australians.
Keywords: causal attribution, Marfan syndrome, genetics, Aboriginal Australian, qualitative, psychosocial
INTRODUCTION
Marfan syndrome (MFS) is a connective tissue disorder, affecting 2–3/10,000 individuals (Judge and Dietz, 2005). MFS is an autosomal dominant condition caused by mutations in FBN1 (Lee et al., 1991). The main clinical features are aortic root dilatation and ectopia lentis (lens dislocation); other features include increased arm span, pectus excavatum or carinatum (sunken or bulging...
breastbone, respectively), scoliosis, joint laxity, myopia, spontaneous pneumothorax, and mitral valve prolapse (Marfan, 1896; Loey et al., 2010). Aortic root dilatation in MFS patients can lead to aortic dissection and sudden death, though surgical and/or medical intervention can reduce the otherwise high mortality of this condition (Brooke et al., 2008). Lens dislocation occurs in the majority of MFS patients, and can occur after minimal or absent trauma, compromising visual acuity (Hindle and Crawford, 1969).
There is considerable intra- and inter-familial clinical variability in MFS, suggesting genetic and/or environmental modifiers (Dietz et al., 1992; Dietz and Pyeritz, 1995). Potential genetic modifiers include the expression of the non-mutated copy of FBN1 (i.e., inherited from the unaffected parent) (Hutchinson et al., 2003; Aubart et al., 2015) and/or variants in related genes (Fernandes et al., 2016). Environmental factors likely to affect risk of aortic dilatation and/or dissection include hypertension, chest trauma, illicit drugs, and pregnancy (Hirst et al., 1958; Larson and Edwards, 1984; Nistri et al., 1999; Lange and Hillis, 2001).
Illness representations (i.e., people's perceptions of, and beliefs about, an illness) affect how health threats are processed as well as subsequent health-related behaviors (Leventhal, 1970; Leventhal et al., 1997, 1998, 2003). Illness representations are not static but change in reaction to life experiences, including experiences of illness and medical care personally and/or in family and friends, media exposure, and cultural beliefs (Leventhal et al., 2003). The component of illness perceptions most strongly pertaining to genetics is causal attribution, the process of analyzing a situation and identifying the reason that a major event occurred (Leventhal et al., 1997). After receiving a diagnosis of a serious medical condition, 70–95% of people strive to identify and attribute an explanation (Turnquist et al., 1988), in an effort to restore the belief that the world is coherent, cohesive and predictable (Abramson et al., 1978; Taylor et al., 1984; Affleck et al., 1987). As with illness perceptions generally, causal attributions may change over time (Hunt et al., 1989).
Shiloh et al. (2002) asked participants (predominately university students) to group possible causes of disease by perceived similarity. Three overarching categories emerged: environmental, behavioral, and “hidden.” Genetics was assigned to the hidden category, along with mystical and psychosocial causes. The hidden category was perceived to be the least controllable. This is troubling, as control, or at least perceived personal control (Berkenstadt et al., 1999), is important in coping (Scharlooo et al., 1998) and affects health behaviors (e.g., medication adherence, clinic attendance, etc.). However, elsewhere Shiloh points out that the perceived lack of control is also associated with decreased blame, which in turn aids coping with genetic conditions (Shiloh, 2006). Shiloh postulates that individuals attending genetic counseling are constantly balancing the desire to decrease blame with that of increasing control (Shiloh, 2006).
Few studies are available on causal attributions in genetic conditions generally and only one addressing causal attributions in MFS specifically (Peters et al., 2001). Peters et al.’s (2001) study, which was carried out in the United States and assessed 174 people with MFS, found that aortic dissection, joint pain, and depressive symptoms were all negatively associated with a belief that MFS was a curable and/or controllable condition. The majority felt they had low personal control over their illness and endorsed heredity as the cause, while half believed that chance played a role. These attributions seemed consistent with the presence or absence of a family history of MFS, in that individuals with a positive family history were more likely to select heredity, whereas those with no family history (i.e., in whom MFS resulted from a new mutation), were more likely to select chance (Peters et al., 2001). This study also demonstrated that it was possible to hold multiple, mutually exclusive causal attributions. Notably, this study predominantly focused on factors initially causing disease, rather than those modifying disease severity.
To the best of our knowledge, no psychosocial research has been reported in Aboriginal and Torres Strait Islander Australians pertaining to causal attributions in genetic conditions. Most available psychosocial research in First Australians has centered on cancer care and support provision, showing that cancer was sometimes perceived to be caused by “payback” and, in those instances, Aboriginal people were less likely to seek assistance for pain management (McGrath, 2006b). The idea of relocating for cancer care was frightening and encompassed a fear of leaving home, cultural alienation, disempowerment, losing support, and being in a hospital environment (McGrath, 2006a, 2007; McGrath et al., 2007).
In summary, there is little published regarding causal attributions in MFS and nothing exploring individual interpretation of phenotypic variability within a family (despite all family members sharing a common mutation). Further, there have been no previous studies of causal attributions in any genetic condition in Aboriginal Australians. Here, we report causal attributions and individual interpretation of familial phenotypic variability by affected and unaffected family members of a large Indigenous Australian family with MFS.
**MATERIALS AND METHODS**
**Study Design**
This study was approved by the University of Queensland Human Research Ethics Committee (UQ #2014001158). Affected and unaffected family members (including spouses/parents), who had previously participated in a genetics study of MFS (Summers et al., 2004) were invited to participate. Individuals were invited to participate either by mail or in person (by JW, a cardiac nurse known to the family). All individuals invited to participate were given an invitation packet, which included a letter of invitation, an information sheet, consent form, an opt-in/opt-out response sheet and a stamped addressed return envelope. Those who had not responded after 2 weeks were telephoned to address any questions or concerns. Additional invitation packets were sent if initial sets had been misplaced or had not arrived. Finally, after some participants had been interviewed, they were encouraged to contact other family members to “vouch” for the study; this included both previously invited individuals and other members of the family not previously contacted. Consented participants were interviewed in person...
or by telephone, depending on their preference, and at a time of their choosing.
A qualitative methodology was chosen as the optimal initial exploration of this topic. In-depth, semi-structured interviews were conducted by AM-L exploring participants' perceptions of MFS, including its challenges, benefits and controllability, and participants' perceptions of causal attribution both for MFS per se and its clinical variability. Sampling was discontinued when no more individuals were willing to participate; additionally, at this point data saturation had been reached (Denzin and Lincoln, 1994).
Data Analysis
Interview recordings were transcribed verbatim. The conceptual framework of Miles and Huberman (1994) was used to guide analysis. Analysis was both deductive (utilizing theoretical frameworks generated from past research) and inductive (identifying emergent themes and patterns pertaining to each category). Transcripts were coded by an accredited genetic counselor and experienced qualitative researcher (AM-L) to assign themes and concepts based on the topics covered in the semi-structured interview guide. In this manner a preliminary codebook was developed. Subsequently all transcripts were re-coded, this time subdividing thematic categories (nodes) into more fine-grained sub-categories (e.g., lifestyle was subdivided into exercise, in turn subdivided into high impact exercise, inadequate exercise etc.). These sub-nodes were further refined or merged as needed and transcripts reanalyzed until no new codes were required; and an advanced codebook was generated. All transcripts were then imported into QSR International's NVIVO 11 qualitative data analysis software (Richards, 2005), and transcripts were coded using accepted coding techniques (Coffey and Atkinson, 1996). As a strategy to reduce bias and enhance validity (also referred to as credibility in qualitative research) (Miles and Huberman, 1994), four transcripts were coded independently by BM, a research psychologist with expertise in qualitative research methodology and psychosocial aspects of genetics. Codes were reviewed and reconciled. All themes were evaluated against participants' disease status (i.e., whether they had a personal diagnosis (affected) versus no personal diagnosis (unaffected) with MFS). Each illustrative quotation in the results section is followed by an ID number and a code denoting the disease status of the respondents, i.e., unaffected (UA) or affected (AF) individuals. In the interest of confidentiality, unaffected siblings/offspring are not distinguished from unaffected spouses/parents.
RESULTS
Letters of invitation were sent initially to 27 members of a large Indigenous Australian family with MFS located in two rural towns. Six returned the response sheet agreeing to participate. One letter was returned due to an incorrect address. Follow-up phone calls identified eight additional individuals willing to participate, and five individuals were invited in person at an outreach clinic. Thus, a total of 40 individuals were invited, of whom 18 participated in an interview, giving a participation rate overall of 45%.
From February to October 2015, 13 women and 5 men ranging in age from ~20 to 70 years were interviewed (specific ages not given to protect confidentiality). Mean interview length was 23 min (range: 10–46 min). Of the 18 participants, 8 had MFS, 6 were unaffected family members, and 4 were spouses and/or parents of affected individuals. Fifteen of the 18 participants were parents, and of these 10 had affected children. All had experienced the death of at least one first- or second-degree relative from complications of MFS. Specific clinical details of affected individuals are not provided in order to protect confidentiality. However, all affected individuals met the Revised Ghent Criteria for MFS (Loeys et al., 2010), including lens dislocation, and their diagnosis had been genetically confirmed. Five of the eight individuals with MFS were employed, employment status was not collected on unaffected family members. None of the participants had a university degree.
Factors Perceived to Cause Marfan Syndrome
Almost all participants (16/18) identified that MFS was heritable. While some stated a possible chromosomal “imbalance,” the majority specifically mentioned genes.
I think it's just dicky genes. ID 6 (AF)
Eye Color
It was also suggested that MFS was associated with eye color, such that those with blue eyes had the condition and those with brown eyes did not.
Like my mum, she knew [Name] was going to have it because he had blue eyes. She knew straight up. She said he's going to have eye problems. ID 7 (UA)
Skipping Generations, Birth Order and/or Gender
A few individuals commented that they had either observed or were concerned about MFS skipping generations.
Um, because of our son and it's something that I want him to be aware of because there's you know there's that chance that it's going to be passed on to his children. And possibly even (my other child) may - I'm not sure if it skips a gene or whatever, yeah I'm not too sure on that. ID 15 (UA)
A few people felt that birth order correlated with the likelihood of having MFS or the severity of the disease, while two suggested that males were more likely to be affected than females.
From what I have seen in, just in my family, the eldest child always has the same features and always has the same little idiosyncrasies…so it seemed to be the first child of everyone, and then after that the severity got worse…especially if it's a boy like… I think it's because the boys live harder or I don't know and they, their bodies grow bigger than ours, I don't know, but they seem to die very young in our family. ID 2 (AF)
Where did the Mutation Originate?
Although most participants acknowledged that the condition was genetic, a third were perplexed about the origin of the
mutation within the family. The responses indicated that the origin of the mutation had clearly been a topic of considerable thought and discussion.
And that dates back to then and we still don’t know where we actually got it. We know we got it off our grandfather but where did he get it from, you know? Like where did it actually come from in the first place? ID 4 (AF)
Some participants did not believe that a mutation was spontaneous, but rather something and/or someone had introduced it into the family. Four people expressed the belief that it was not a condition which originally occurred in Aboriginal Australians but that it was introduced by the English or Europeans when they came to Australia. As this family had a couple of known ancestors from Britain, a few surmised that the condition was introduced into the family via one of these ancestors.
...that’s where I, I think it would come from anyway because we didn’t have these problems in this country before that at all. ID 2 (AF)
Some felt that knowing that it originated with Europeans still did not explain the cause of the original mutation and some believed that environmental factors such as disease or diet triggered the mutagenesis.
Uh well I was just under the belief that it was brought into this country, into my family, through the uh English um who had you know picked up scurvy and syphilis on board and brought it over here and obviously impregnated my lineage along that family tree. ...that were brought to this country via um, the English who had syphilis and that was where they, um, that was you know when the fetus was in there, that was where the chromosomes started breaking down.
It could go back to the parents and grandparents and the breeding. Back to the things uh food that they ate probably years ago. ... You never know back years ago it couldn’t have happened. ... older people hadn’t been hygienic years ago eh? Could have contracted some sort of germ or something. ID 12 (UA)
Factors Influencing Phenotypic Severity
Participants were asked whether they had any theories to explain the clinical variability within the family. About half responded that they didn’t know, while the remainder felt it might be simply the way people were born, lifestyle, or additional genes.
Interviewees were then asked whether and how each of the following factors contributed to phenotypic severity: lifestyle; stress, emotions or worries; germs, pollution or environmental factors; chance or bad luck; spirituality; and other genes.
Born This Way
The most common theory as to why there were phenotypic differences within the family centered on the belief that “everyone’s different” (ID 4, AF) and the acceptance that one is “born with it” (ID 5, UA). This response was volunteered by about half the participants.
I think to me it affects everyone in different ways. Everyone’s different. Like someone can have the heart problem and not have, um, the skeletal feature problems. ...that’s, to me that’s how I think it affects different people, just – it just affects different people in different ways. ID 17 (UA)
Lifestyle
Most participants repeatedly stated that lifestyle affected severity. Lifestyle factors thought to play a role included alcohol, physical activity (overexertion or lack of exercise), drugs, diet, self-care (good self-care or lack thereof) and smoking.
Well I think it’s lifestyle um you know like uh for example I know [name], he died when I was [age], uh he lived pretty hard, he was you know a heavy binge drinker, um he did drugs, he played football you know high-contact sport um and the same with [name] you know, he smoked, you know they all smoked and they played rugby league and you know high contact sport and fighting and stuff like that. ID 2 (AF)
Oh well because – well if I wasn’t drinking now and after having a heart operation like, I’d probably be dead by now because alcohol and Marfan syndrome heart, they just don’t mix. ID 4 (AF)
Physical activity, diet and self-care
Participants seemed aware of the importance of staying healthy and taking care of their bodies when affected with MFS, specifically mentioning the role of exercise, diet, and self-care.
In terms of exercise, half of participants spoke of the challenge of finding the right balance for physical activity. The majority perceived danger from contact and high-impact sports. However, many believed it was important to stay active in order to stay healthy.
Well yeah, um, if you can do – if you do the wrong exercise that can put strain on your body and strain on your heart and make it, you know, and make it worse. ID 17 (AF)
A healthy diet, specifically the avoidance of unhealthy food, was mentioned as important in maintaining physical health and an appropriate weight.
Probably just your diet. You’ve got to um, like I said you’ve got to watch what you eat and then um keep it under control. ... don’t become too overweight and um because I think the more unhealthy and unfit you become, the easier it is to um, the easier it is for these things to take effect. ID 11 (AF)
Taking care of oneself also included adhering to medication and attending medical appointments.
Stress, Emotion or Worries
Although a few participants stated that stress, emotions or worries did not affect the severity of the disease, the majority felt that it did. Those who elaborated, explained that stress put additional strain on the heart and/or body.
Of note, participants articulated that the financial stress associated with the disease or the stress of not being able to work could also be deleterious.
You know they don’t look after themselves very well, they don’t have, you know they’ve got financial stresses and you know all the things that kill normal people like you know not being financially sound and not having full-time jobs and stuff like that, that’s got to be stressful for people and put strain on the heart.
ID 2 (AF)
Other Genes
Most participants thought that other genes could affect severity. One participant knew other individuals affected with a genetic neurodegenerative disorder, and she postulated that there could be an additive effect if one inherited both MFS and this neurological condition.
Um, I do, yeah I do. I do think if someone had another – another genetic disorder, I think it could affect, it could affect Marfan’s in a different way.
ID 17 (AF)
Some participants believed that the genes of the unaffected parent might modulate the severity of the disorder in their offspring, though they emphasized that this was difficult to assess ahead of time.
I suppose you could have genes there that mixed together could cause, like ordinary genes that might react with the Marfan genes. So yeah, and you don’t know what genes you’ve got unless you’re a doctor and you’re looking under a microscope or something.
ID 9 (UA)
Germs, Pollution or Other Environmental Factors
Respondents were divided as to whether germs, pollution and/or environmental factors affected the severity of the illness, with about half believing they did and about half believing that they did not. Pollution and chemicals were seen as potentially disease-causing in general and therefore could be a risk factor for MFS severity specifically.
You know pollution has a large effect on a lot of genetic diseases you know like, I don’t know, cancer and stuff like that . . . you know pollutions and stuff, do release those germs or, you know, make them worse.
ID 15 (UA)
Chance or Bad Luck
The majority of interviewees rejected the idea that chance or bad luck played a role in the severity of the illness.
I don’t think it’s chance or bad luck. If they’re born with it, you’ve got it.
ID 9 (UA)
Paradoxically, while disavowing chance, some participants used language suggestive of the potential role of luck.
Bad luck, hmm. No. I think it’s just one in a million. So you just can’t say it’s a chance or bad luck thing. I think it’s just a one in a million, you know.
ID 7 (UA)
However, a few felt that luck could have played a role. The majority spoke about chance and luck in terms of the probability of inheriting MFS in the first place rather than luck or chance affecting the severity.
Yeah, yeah I reckon because not every single one of us has got it . . . You know what I mean like if it was bad luck everyone would have it . . . Yeah well I reckon that, yeah. Some people have got um other diseases like that runs through their family and we just got – we just unlucky enough to have just Marfan disease that runs through our bloodline.
ID 5 (UA)
One individual did mention luck or chance in conjunction with severity and, in doing so, referred to the gene contribution of the unaffected parent.
If I can say I’m lucky and it’s just by chance that [names] got the worst part of the genes then yeah I think from my mother.
ID 16 (UA)
Spirituality
The majority did not feel that spirituality altered disease severity. A couple of individuals rejected the idea that faith or spirituality could modulate phenotype, as they didn’t think it was fair to implicate God.
But I look at it this way, you can’t blame God for what happens, you know. If you’re dealt something in life you deal with it or you try to, you know?
ID 8 (UA)
I’m not sure, just suppose if you’re born with it you’re born with it I guess but everyone’s. I don’t know. I’m a big believer in God tarnishes everyone with a different brush, you know what I mean, like can’t make us all look the same, everyone’s – people, everyone’s different in the world.
ID 5 (UA)
One individual specifically stated that belief could not remove the disease from the body.
For me, I’m more of a realist. I believe, like I’ve got it um and I don’t think um whether or not you, what your belief is going to make it any better or any worse because it’s one of these little nasty things that once it’s there, unfortunately I don’t think we can get rid of it out of your body.
ID 11 (AF)
However, a few thought that spirituality could play a role and had very different theories. One person believed that the disease itself could have been caused by upsetting the spirits.
Hmm something could have happened years ago, spirits didn’t agree with, or agree with or didn’t you know, never know.
ID 12 (UA)
Another believed that the spirits of deceased relatives were watching over those who were affected today and protecting them.
My grandfather had it, great grandfather. I’d say he’d be, you know, looking over them and stuff, all the family members.
ID 10 (UA)
Another accepted that God sent challenges to everyone, and that MFS was just the challenge in their family.
Yeah just tarred our family with the Marfan . . . But yeah, no, it’s all good. It’s just a hurdle that God threw us, throw at us I reckon.
ID 5 (UA)
Finally, one participant felt that spirituality could affect severity as spiritual people would want to be on the land, which was remotely located and thus far removed from services.
Um maybe you know like cultural and spiritually you know with my family living remotely they want to be connected to land so they want to stay out there... so you know it sort of blocks them off from getting um help and getting services and getting fresh food and lifestyle choices you know having better choices I guess available to them because they want to be connected to the land, to the culture.
**DISCUSSION**
This study is the first to explore causal attributions for a genetic condition in an Aboriginal Australian family. Interviews with eighteen affected and unaffected individuals examined the perceived cause of MFS in the family and found that though the majority believed that the condition was genetic, they were less clear about inheritance, suspecting that the condition could skip generations, and that both birth order and gender affected the probability of having MFS. Environmental factors, such as infections and poor diet, were thought to explain the origin of the initial mutation, which was believed to have been introduced by the British. When asked about the factors which were affecting the severity of the disease in affected individuals, lifestyle and other genes were most commonly discussed possibilities. This family did not ascribe to the belief that God or a spiritual entity caused or moderated the severity of their condition.
Sixteen of the eighteen participants knew that MFS was a genetic condition. This high percentage is unsurprising given that all symptomatic family members were offered genetic testing and counseling through the regional clinical genetics service 5 to 10 years prior. Nonetheless, this is consistent with a previous quantitative study of causal attribution in a MFS support group cohort, which found that 74% selected heredity as a cause (Peters et al., 2001). Causal attributions have been previously explored in genetic conditions generally; and a recurrent theme is the belief that physical features co-segregate with genetic conditions, i.e., a parent would “determine” that a child would or wouldn’t develop a familial condition based on their physical resemblance to another known affected family member (Kessler and Bloch, 1989; Huggins et al., 1992; Emslie et al., 2003; Shaw and Hurst, 2008; Klitzman, 2010). Psychologically, this “preselection” enhances coping by reducing ambiguity and uncertainty (Kessler and Bloch, 1989; Klitzman, 2010).
When the theme of eye color predicting affection status initially emerged from the data, it was tempting to view this as another example of genetic preselection. However, the gene associated with blue eye color, \(_{\text{OCA2}}\) (Duffy et al., 2007) is located at 15q13.1, 30cM proximally to \(_{\text{FBN1}}\) (15q21.1); linkage analysis in this family (not shown here) demonstrated that these loci co-segregate in this family. In Caucasian populations, the \(_{\text{OCA2}}\) locus is autosomal recessive and thus, its co-segregation with a dominant trait would mean that it was unreliable as a marker for disease. We do not have these data for Aboriginal Australians, and it is therefore, possible that these comments indicate that family members have astutely observed that blue eyes are a marker for MFS in this family. Of note, other Marfanoid syndromes, such as Loeps-Dietz, are associated with blue eyes, so modifier loci in related genes in the same pathway may be affecting eye color in affected members of this family.
The concept that genetic conditions “skip” generations is a widely held belief in lay populations (Richards, 1996; Emslie et al., 2003) and may at least in part be attributable to X-linked inheritance, incomplete penetrance, recessive diseases and traits (e.g., red hair), and transmission of translocations within families. Severity of MFS can vary significantly within families, with some people having ocular and/or cardiac involvement at an early age and others being more mildly affected (Dietz et al., 1992). However, it never “skips” generations. There are three possible interpretations for such a belief in this family. Firstly, participants may have been referring to severity of symptoms (i.e., variable expressivity) rather than the presence of the disease (i.e., variable penetrance). Secondly, in one branch of this particular family there are some individuals with isolated dilated aorta in the absence of MFS or an \(_{\text{FBN1}}\) mutation; this would certainly add confusion to this family’s understanding of how MFS is inherited. A third explanation is that a belief that a condition could skip generations allowed people hope when choosing to have children, thus enabling them to engage in “reproductive roulette” (Lippman-Hand and Fraser, 1979).
Birth order does not determine disease status for MFS. This misconception has been reported previously in other conditions (Fanos and Johnson, 1995; Richards, 1996; Hunt et al., 2002). Such beliefs may be a psychological defense mechanism used to cope with the arbitrary nature of an inherited genetic condition (Richards, 1996), given the 50/50 likelihood of any particular allele being passed onto offspring. Although gender \(_{\text{per se}}\) is not associated with MFS status, aortic dissection occurs at an earlier age in males with MFS compared to females (Murdoch et al., 1972).
People were perplexed by the origin of the mutation and speculated that it had been introduced into the family by British ancestors, who may themselves have acquired the mutation through disease or poor diet, particularly in pregnancy. There are a number of psychosocial reasons as to why this Indigenous family might believe that MFS had been introduced by Europeans. British colonization introduced many infectious diseases (including syphilis, tuberculosis, influenza, and measles) to the Indigenous population of Australia (Dowling, 1997). Australian Aboriginal Peoples also believe that cancer was brought to Australia by the “white man” (Prior, 2006, pg 27). Furthermore, as MFS segregates with blue eyes in this family, this may have reinforced the belief that it was originally a “British” disease.
The belief that consumption of certain foods during pregnancy causes congenital conditions is common, especially in certain ethnocultural groups (Cohen et al., 1998). For example, cleft lip is believed to be caused by the consumption of rabbit (hare) (Cheng, 1990), café au lait marks a consequence of fish consumption (Shaw and Hurst, 2008) and ectrodactyly (split hand and foot) a penchant for crab during pregnancy (Abad et al., 2014). Although exposure to certain toxins in pregnancy can undoubtedly result in congenital abnormalities, there is no association with such toxins inducing heritable mutations.
However, holding such beliefs can promote a sense of personal control in subsequent pregnancies (Shiloh, 2006).
When asked to speculate as to factors influencing disease severity, most respondents replied that it was just the way people were born or simply because “everyone is different.” Previous qualitative research on Aboriginal Australians with cancer has shown that fatalism is common (Prior, 2005, 2006, 2009; Shahid et al., 2009). On considering reactions in this family on this point, interviews were closely evaluated but no language or tone was found suggestive of fatalism or resignation. A “what will be, will be” philosophy most accurately describes their reaction. Evaluated against the categories proposed by Shiloh et al. (2002) it appears that this best fits into the “mystical” category, which in turn belongs to the broader category of “hidden causes.” Hidden causes are associated with high levels of uncertainty and low levels of control (Shiloh et al., 2002), which would appear to bode poorly for coping in this family, given that optimal coping traditionally involves limiting uncertainty and regaining control (Leventhal et al., 1997). However, Shiloh also points out that control can be “associated with undesirable feelings of responsibility and self-blame” (Shiloh, 2006, pg 331). Thus participants choosing the “hidden causes” explanation may be absolving themselves of any guilt associated with disease severity. Interestingly, when causal attribution in cancer was qualitatively explored in an Australian Aboriginal sample, the etiology of cancer was often ascribed to payback for the perceived misdeeds of the affected individual, again falling into the “mysterious” category, which researchers felt promoted acceptance and coping (McGrath et al., 2006).
Lifestyle was the second most commonly endorsed modifier of phenotypic severity, with drug and alcohol abuse being the most frequently named. Drug use, particularly amphetamines are known risk factors for aortic aneurysms (Westover and Nakonezny, 2010). The relationship between alcohol consumption and aortic dissection risk has been predominantly studied in abdominal aortic aneurysms and there appears to be a complex relationship, where low to moderate consumption is associated with a reduced risk, as compared to non-drinkers, but higher levels of consumption are associated with increased risk (Wong et al., 2007; Green et al., 2014). Smoking is a well-known risk factor for aortic dissection (Wong et al., 2007), but this was not mentioned as frequently as other lifestyle factors. This may reflect the fact that 56% of Aboriginal Australian Peoples located in remote communities are smokers (Australian Bureau of Statistics, 2015), and suggests that this is less likely to be perceived as a negative health behavior. However, in a survey study Aboriginal Australian smokers held similar views about the negative impacts of smoking compared to smokers in the general Australian population (Nicholson et al., 2015). This family’s concern about the deleterious effect of high impact sports on aortic dilation are consistent with medical advice as this is a known risk factor for aortic dilatation in MFS (Ammash et al., 2008). Finally, diet and moderate exercise were seen as factors which could lower the risk of aortic dissection. Both of these reflect recommendations for individuals with MFS, with the former being found in disease-specific support group materials (Marfan Trust, 2019) and the latter in medical guidelines (Ades and Group, 2007).
This family felt that emotional and financial stress contributed to the severity of aortic dilatation in MFS. Indeed emotional stress has been associated with spontaneous coronary artery dissection, with patients perceiving that it precipitated the event (Saw et al., 2014, 2017). Other studies found that stress and worry were perceived risk factors for genetic diseases generally (Cohen et al., 1998; Parrott et al., 2004; Shaw and Hurst, 2008), though there is currently no scientific evidence which aligns with this belief.
Most participants believed that other genes could be modifying the phenotype in MFS, and some speculated that the genetic fitness of the other parent played a role, though they admitted that identifying a good genetic match would be challenging. Other factors which participants felt could influence severity included environmental factors such as “germs,” chemicals and pollution. The concept that viruses or other infectious diseases could cause genetic conditions has been reported previously (Cohen et al., 1998; Shaw and Hurst, 2008), though usually the infectious agent is blamed for causing the disease, not initiating heritable genetic change per se. The quotations included provide further evidence of people’s ability to hold multiple beliefs (Cohen et al., 1998; Klitzman, 2010) in order to restore their belief that the world is “coherent, cohesive and predictable” (Shiloh et al., 2002, pg 373).
Chance or bad luck were not embraced by this family as possible explanations for phenotypic variability, though some felt it played a role in the inheritance of the FBNI mutation and disease causation. A previous study on causal attribution in MFS found that 54% endorsed chance as having a causative role (Peters et al., 2001). The majority of those people were the first case in the family and thus had a de novo mutation. However, a third of individuals from families with an inherited mutation also thought that chance played a role (Peters et al., 2001), similar to the proportion observed in this study.
The majority of individuals in this family rejected the idea that spirituality moderated the severity of disease. This did not appear to reflect a lack of faith but rather an unwillingness to blame the spirits or higher powers. One person felt that MFS could have been caused by upsetting the spirits, a common causal attribution for genetic diseases and birth defects generally (Cohen et al., 1998; Shaw and Hurst, 2008; Klitzman, 2010; Abad et al., 2014). In qualitative studies of causal attributions for cancer in Indigenous Australians specifically, disease etiology is often associated with the spiritual world of curses (Prior, 2005, 2006, 2009; McGrath et al., 2006); this has also been observed in Native Americans (Burhanstipanov et al., 1999) and African Americans (Cohen et al., 1998). Of note, a causal attribution study undertaken in the Philippines found that MFS was thought to occur when a mythical tree giant impregnated a mother so the offspring had his physical characteristics - i.e., tall stature, long limbs, and arachnodactyly (long fingers) (Abad et al., 2014). The idea that spirits can be protective is not well captured in qualitative research of Australian Aboriginal people. However, when discussing spirituality generally in these interviews, another participant commented on the spirits offering assistance, while another participant points out that the spiritual tie to the land
could compromise access to quality healthcare and services. The psychological and spiritual challenge of leaving home in order to be evaluated and treated has been described by rurally located Indigenous Australians (McGrath et al., 2006; McGrath, 2007; Anderson et al., 2008; Prior, 2009). The desire for a culture-centered outreach approach has been documented (Mincham et al., 2003; Askew et al., 2014) and, to that end, there have been pilot programs initiated in some states in Australia (Hayman et al., 2014; Askew et al., 2016; Davy et al., 2016).
This study has a number of limitations. Qualitative studies do not involve probability sampling, and hence the findings reported are not generalizable to all Australian Aboriginal families with MFS. Members of a single family were assessed, and views may have been more heterogeneous had additional families been interviewed. Also, those who elected not to participate may have had different lived experiences than those who did participate. Furthermore, there was a lack of geographic and educational diversity in this family. A relatively small number of individuals opted to participate in this study. Nonetheless data saturation was achieved, and the small size of the sample is not seen as a limitation of this study (Patton, 1990). The retrospective, reflective nature of the interviews means that there may have been some recall bias. In hindsight, it is possible that the wording of some of the questions was not optimal; for example, it is not clear that all participants fully understood the nuance of factors causing disease as opposed to factors contributing to disease severity.
This qualitative study assessed the range of beliefs in relation to MFS, and future qualitative studies should assess the extent to which these beliefs are endorsed. This is the first study to explore causal attribution in any genetic condition in the Australian Aboriginal population; it would be interesting to conduct further qualitative research in the area of genetics in this population in the future. The concept of events that give rise to mutations is worth further exploration. If sample size permitted, such qualitative research would ideally be combined with quantitative measures (e.g., use of the Revised Illness Perceptions Questionnaire (Moss-Morris et al., 2002), which has not yet been administered to an Aboriginal Australian population). As causal perceptions affect coping, adjustment and many health behaviors, a better understanding in the Indigenous Australian population would be beneficial as genetic testing is increasingly used in a wide variety of healthcare settings.
DATA AVAILABILITY STATEMENT
The datasets for this article are not publicly available due to concerns regarding participant/patient anonymity. Requests to access the datasets should be directed to the corresponding author.
ETHICS STATEMENT
The studies involving human participants were reviewed and approved by University of Queensland Human Research Ethics Committee. The patients/participants provided their written informed consent to participate in this study.
AUTHOR CONTRIBUTIONS
AM-L designed the study, with guidance from BM, ED, and MB. JW and MW provided the introductions to members of the large family with Marfan syndrome and traveled with AM-L to remote towns in Queensland to maximize the chances of participation. AM-L conducted the analysis of the data with oversight from BM and also read and coded a subset of the transcripts in parallel and any discrepancies were discussed and reconciled.
FUNDING
This study was funded by a project grant from the Prince Charles Hospital Foundation (MS 2012-39). AM-L was funded by a National Health and Medical Research Council (NHMRC) Early Career Fellowship (ID 1158111), while BM was funded by an NHMRC Senior Research Fellowship Level B (ID 1078523), and MB by an NHMRC Senior Principal Research Fellowship (ID 1024879). The Translational Research Institute is supported by a grant from the Australian Government.
ACKNOWLEDGMENTS
We are very grateful to the individuals who participated in this study.
REFERENCES
Abad, P. J., Tan, M. L., Balyot, M. M., Villa, A. Q., Talapian, G. L., Reyes, M. E., et al. (2014). Cultural beliefs on disease causation in the Philippines: challenge and implications in genetic counseling. J. Community Genet. 5, 399–407. doi: 10.1007/s12687-014-0193-1
Abramson, L. Y., Seligman, M. E., and Teasdale, J. D. (1978). Learned helplessness in humans: critique and reformulation. J. Abnorm. Psychol. 87, 49–74.
Ades, L., and Group, C. C. G. W. (2007). Guidelines for the diagnosis and management of Marfan syndrome. Heart Lung Circ. 16, 28–30. doi: 10.1016/j.hlc.2006.10.022
Affleck, G., Tennen, H., Pfeiffer, C., and Fifield, J. (1987). Appraisals of control and predictability in adapting to a chronic disease. J. Pers. Soc. Psychol. 53, 273–279. doi: 10.1037/0022-3514.53.2.273
Ammash, N. M., Sundt, T. M., and Connolly, H. M. (2008). Marfan syndrome—diagnosis and management. Curr. Probl. Cardiol. 33, 7–39. doi: 10.1016/j.cpcardiol.2007.10.001
Anderson, K., Cunningham, J., Preece, C., and Cass, A. (2008). "All they said was my kidneys were dead": indigenous Australian patients' understanding of their chronic kidney disease. Med. J. Aust. 189, 499–503.
Askew, D., Brady, J., Brown, A., Cass, A., Davy, C., deVries, J., et al. (2014). Monograph 1: To your Door: Factors that Influence Aboriginal and Torres Strait Islander Peoples Seeking Care. Adelaide, SA: Kanyini Vascular Collaboration. Askew, D. A., Togni, S. J., Schluter, P. J., Rogers, L., Egert, S., Potter, N., et al. (2016). Investigating the feasibility, acceptability and appropriateness of outreach case management in an urban Aboriginal and Torres Strait Islander primary health care service: a mixed methods exploratory study. BMC Health Serv. Res. 16:178. doi: 10.1186/s12913-016-1428-0
Anderson, K., Devitt, J., Cunningham, J., Preece, C., and Cass, A. (2008). “All they said was my kidneys were dead”: indigenous Australian patients’ understanding of their chronic kidney disease. Med. J. Aust. 189, 499–503.
DATA AVAILABILITY STATEMENT
The datasets for this article are not publicly available due to concerns regarding participant/patient anonymity. Requests to access the datasets should be directed to the corresponding author.
ETHICS STATEMENT
The studies involving human participants were reviewed and approved by University of Queensland Human Research Ethics Committee. The patients/participants provided their written informed consent to participate in this study.
AUTHOR CONTRIBUTIONS
AM-L designed the study, with guidance from BM, ED, and MB. JW and MW provided the introductions to members of the large family with Marfan syndrome and traveled with AM-L to remote towns in Queensland to maximize the chances of participation. AM-L conducted the analysis of the data with oversight from BM and also read and coded a subset of the transcripts in parallel and any discrepancies were discussed and reconciled.
FUNDING
This study was funded by a project grant from the Prince Charles Hospital Foundation (MS 2012-39). AM-L was funded by a National Health and Medical Research Council (NHMRC) Early Career Fellowship (ID 1158111), while BM was funded by an NHMRC Senior Research Fellowship Level B (ID 1078523), and MB by an NHMRC Senior Principal Research Fellowship (ID 1024879). The Translational Research Institute is supported by a grant from the Australian Government.
ACKNOWLEDGMENTS
We are very grateful to the individuals who participated in this study.
REFERENCES
Abad, P. J., Tan, M. L., Balyot, M. M., Villa, A. Q., Talapian, G. L., Reyes, M. E., et al. (2014). Cultural beliefs on disease causation in the Philippines: challenge and implications in genetic counseling. J. Community Genet. 5, 399–407. doi: 10.1007/s12687-014-0193-1
Abramson, L. Y., Seligman, M. E., and Teasdale, J. D. (1978). Learned helplessness in humans: critique and reformulation. J. Abnorm. Psychol. 87, 49–74.
Ades, L., and Group, C. C. G. W. (2007). Guidelines for the diagnosis and management of Marfan syndrome. Heart Lung Circ. 16, 28–30. doi: 10.1016/j.hlc.2006.10.022
Affleck, G., Tennen, H., Pfeiffer, C., and Fifield, J. (1987). Appraisals of control and predictability in adapting to a chronic disease. J. Pers. Soc. Psychol. 53, 273–279. doi: 10.1037/0022-3514.53.2.273
Ammash, N. M., Sundt, T. M., and Connolly, H. M. (2008). Marfan syndrome—diagnosis and management. Curr. Probl. Cardiol. 33, 7–39. doi: 10.1016/j.cpcardiol.2007.10.001
Anderson, K., Devitt, J., Cunningham, J., Preece, C., and Cass, A. (2008). “All they said was my kidneys were dead”: indigenous Australian patients’ understanding of their chronic kidney disease. Med. J. Aust. 189, 499–503.
Askew, D., Brady, J., Brown, A., Cass, A., Davy, C., deVries, J., et al. (2014). Monograph 1: To your Door: Factors that Influence Aboriginal and Torres Strait Islander Peoples Seeking Care. Adelaide, SA: Kanyini Vascular Collaboration. Askew, D. A., Togni, S. J., Schluter, P. J., Rogers, L., Egert, S., Potter, N., et al. (2016). Investigating the feasibility, acceptability and appropriateness of outreach case management in an urban Aboriginal and Torres Strait Islander primary health care service: a mixed methods exploratory study. BMC Health Serv. Res. 16:178. doi: 10.1186/s12913-016-1428-0
Anderson, K., Devitt, J., Cunningham, J., Preece, C., and Cass, A. (2008). “All they said was my kidneys were dead”: indigenous Australian patients’ understanding of their chronic kidney disease. Med. J. Aust. 189, 499–503.
Aubart, M., Gross, M. S., Hanna, N., Zabot, M. T., Szajderj, M., Detaint, D., et al. (2015). The clinical presentation of Marfan syndrome is modulated by expression of wild-type FBN1 allele. *Hum. Mol. Genet.* 24, 2764–2770. doi: 10.1093/hmg/ddv037
Australian Bureau of Statistics (2015). *Smoking Prevalence* [Online]. Available online at: https://www.abs.gov.au/ausstats/[email protected]/Lookup/by%20Subject/4737.0-1994?Open=202014-15~Main%20Features~Smoking%20Prevalence~10 (accessed September 23, 2019).
Berkenstadt, M., Shiloh, S., Barkai, G., Maboudi, M., and Goldberg, B. (1999). Perceived personal control (PPC): a new concept in measuring outcome of genetic counseling. *Am. J. Med. Genet.* 82, 53–59. doi: 10.1002/(sici)1096-8628(1999101)82:1<5:aid-aajmg11>3.0.co;2
Brooke, B. S., Habashi, J. P., Judge, D. P., Patel, N., Loeys, B., and Dietz, H. C. III (2008). Angiotensin II blockade and aortic root dilation in Marfan’s syndrome. *N. Engl. J. Med.* 358, 2787–2795. doi: 10.1056/NEJMoa0706585
Burkhanstipanov, L., Lovato, M. P., and Krebs, L. V. (1999). Native American survivors. *Health Care Women Int.* 20, 505–515. doi: 10.1080/073993399245539
Cheng, L. R. (1990). Asian-American cultural perspectives on birth defects: focus on cleft palate. *Cleft Palate J.* 27, 294–300. doi: 10.1597/1545-1569(1990)027<0294:AAAPCP>2.0.CO;2
Coffey, A. J., and Atkinson, P. (1996). *Concepts and Coding. Making Sense of Qualitative Data: Complementary Research Strategies.* Thousand Oaks, CA: Sage Publications.
Cohen, L. H., Fine, B. A., and Persgram, E. (1998). An assessment of ethnocultural beliefs regarding the causes of birth defects and genetic disorders. *J. Genet. Couns.* 7, 15–29. doi: 10.1023/A:102281207511
Davy, C., Cass, A., Brady, J., DeVries, J., Fewquandie, B., Ingram, S., et al. (2016). Facilitating engagement through strong relationships between primary healthcare and aboriginal and Torres Strait Islander peoples. *Aust. N. Z. J. Public Health* 40, 535–541. doi: 10.1111/1753-6405.12553
Denzin, N. K., and Lincoln, Y. S. (1994). *Handbook of Qualitative Research.* London: Sage.
Dietz, H. C., and Pyeritz, R. E. (1995). Mutations in the human gene for fibrillin-1 (FBN1) in the Marfan syndrome and related disorders. *Hum. Mol. Genet.* 4, 1799–1809. doi: 10.1093/hmg/4.suppl_1.1799
Dietz, H. C., Pyeritz, R. E., Puffenberger, E. G., Kendzior, R. J. Jr., Corson, G. M., Maslen, C. L., et al. (1992). Marfan phenotype variability in a family segregating a missense mutation in the epidermal growth factor-like motif of the fibrillin gene. *J. Clin. Invest.* 89, 1674–1680. doi: 10.1172/JCI11576
Dowling, P. J. (1997). “A Great Deal of Sickness” Introduced Diseases among the Aboriginal People of Colonial Southeast Australia 1788-1900. Ph.D. thesis, Australian National University, Canberra.
Duffy, D. L., Montgomery, G. W., Chen, W., Zhao, Z. Z., Le, L., James, M. R., et al. (2007). A chromosomal polymorphism haplotype in intron 1 of OC/24 explains most human eye-color variation. *Am. J. Hum. Genet.* 80, 241–252. doi: 10.1086/501885
Emslie, C., Hunt, K., and Watt, G. (2003). A chip off the old block? Lay understandings of inheritance among men and women in mid-life. *Public Underst. Sci.* 12, 47–65.
Fanos, J. H., and Johnson, J. P. (1995). Perception of carrier status by cystic fibrosis siblings. *Am. J. Hum. Genet.* 57, 431–438.
Fernandes, G. R., Massironi, S. M., and Pereira, L. V. (2016). Identification of loci modulating the cardiovascular and skeletal phenotypes of Marfan syndrome in mice. *Sci. Rep.* 6:22426. doi: 10.1038/srep22426
Green, B. L., Bailey, M. A., Bridge, K. L., and Scott, J. A. (2014). Alcohol consumption as a risk factor for abdominal aortic aneurysm. *J. Visc. Med. Surg.* 2:140. doi: 10.4172/2339-6925.1000140
Hayman, N. E., Askew, D. A., and Spurling, G. K. (2014). From vision to reality: a review of 505 cases. *Archives of Pathology & Laboratory Medicine* 138, 128–135.
Hirst, A. E. Jr., Johns, V. J. Jr., and Kime, S. W. Jr. (1958). Dissecting aneurysm of the aorta: a review of 505 cases. *Medicine (Baltimore)* 37, 217–279. doi: 10.1097/00005195-19580900-00003
Hubbard, E., and Hall, D. (2008). Views of what's wrong: diagnosis and patients’ concepts of illness. *Soc. Sci. Med.* 28, 945–956. doi: 10.1016/0277-9536(89)90324-9
Hutton, S., and Bower, T. (1999). Views of what’s wrong: diagnosis and patients’ concepts of illness. *Soc. Sci. Med.* 28, 945–956. doi: 10.1016/0277-9536(89)90324-9
Leventhal, H., Brissette, I., and Leventhal, E. A. (2003). “The common-sense model of illness representation” and health beliefs regarding the causes of birth defects and genetic disorders. *J. Genet. Couns.* 7, 15–29. doi: 10.1023/A:102281207511
Leventhal, H. (1970). Findings and theory in the study of fear communications. *Adv. Exp. Soc. Psychol.* 5, 119–186.
Leventhal, H., Benyamini, Y., Brownlee, S., Diefenbach, M., Leventhal, E. A., Patrick-Miller, L., et al. (1997). “Illness representations: theoretical foundations,” in *Perceptions of Health and Illness: Current Research and Applications*, eds K. J. Petrie and J. Weinman (Amsterdam: Harwood Academic Publishers), 19–46.
Leventhal, H., Brissette, I., and Leventhal, E. A. (2003). “The common-sense model of self-regulation of health and illness,” in *The Self-regulation of Health and Illness Behaviour*, eds L. D. Cameron and H. Leventhal (London: Routledge). 42–65. doi: 10.1093/rheumatology/kem060
Leventhal, H., Leventhal, A. E., and Contrada, R. J. (1998). Self-regulation, health and behavior: a perceptual-cognitive approach. *Psychol. Health* 13, 717–733.
Lippman-Hand, A., and Fraser, F. C. (1979). Genetic counseling – the postcounseling period: II. Making reproductive choices. *Am. J. Med. Genet.* 4, 73–87. doi: 10.1002/ajmg.1320404109
McGrath, P. (2006b). “The biggest worry.” researching findings on pain management for aboriginal peoples in Northern Territory, Australia. *Rural Remote Health* 6:25. doi: 10.5244/rrh.6.2.3.20692
McGrath, P. (2006a). Exploring aboriginal people's experience of relocation for treatment during end-of-life care. *Int. J. Palliat. Nurs.* 12, 102–108. doi: 10.1177/13561820.2006.12.3.20692
McGrath, P. (2007). ‘I don’t want to be in that big city; this is my country here’: research findings on Aboriginal peoples’ preference to die at home. *Aust. J. Rural Health* 15, 264–268. doi: 10.1111/j.1440-1581.2007.00904.x
McGrath, P., Holecz, H., and Kail-Buckley, S. (2007). “They should come out here”: research findings on lack of local palliative care services for Australian aboriginal people. *Am. J. Hosp. Palliat. Care* 24, 105–113. doi: 10.1177/1049991106297069
McInerney-Leo et al. Causal Attributions in an Australian Aboriginal
McGrath, P., Holewa, H., Ogilvie, K., Rayner, R., and Patton, M. A. (2006). Insights on aboriginal peoples’ views of cancer in Australia. Contemp. Nurse 22, 240–254. doi: 10.5555/contnurse.22.2.240
Miles, M. B., and Huberman, A. M. (1994). Qualitative Data Analysis: An Expanded Sourcebook. London: Sage.
Mincham, C. M., Toussaint, S., Mak, D. B., and Plant, A. J. (2003). Patient views on the management of rheumatic fever and rheumatic heart disease in the Kimberley: a qualitative study. Aust. J. Rural Health 11, 260–265. doi: 10.1111/j.1440-1584.2003.00531.x
Moss-Morris, R., Weinman, J., Petrie, K. J., Horne, R., Cameron, L., and Buick, D. (2002). The revised illness perception questionnaire (IPQ-R). Psychol. Health 17, 1–16.
Murdock, J. L., Walker, B. A., Halpern, B. L., Kuzma, J. W., and McKusick, V. A. (1972). Life expectancy and causes of death in the Marfan syndrome. N. Engl. J. Med. 286, 804–808. doi: 10.1056/NEJM197204132861502
Nicholson, A. K., Borland, R., Bennet, P. T., van der Sterren, A. E., Stevens, M., and Murdoch, J. L., Walker, B. A., Halpern, B. L., Kuzma, J. W., and McKusick, V. A. (1990). Illness perceptions, coping and functioning in patients with rheumatoid arthritis, chronic obstructive pulmonary disease and psoriasis. J. Psychosom. Res. 44, 573–585. doi: 10.1016/0022-3999(97)00254-7
Shahid, S., Finn, L., Bessarab, D., and Thompson, S. C. (2009). Understanding, beliefs and perspectives of aboriginal people in Western Australia about cancer and its impact on access to cancer services. BMC Health Serv. Res. 9:132. doi: 10.1186/1472-6963-9-132
Shaw, A., and Hurst, J. A. (2008). What is this genetics, anyway? Understandings of genetics, illness causality and inheritance among British Pakistani users of genetic services. J. Genet. Couns. 17, 373–383. doi: 10.1007/s10897-008-9156-1
Shiloh, S. (2006). Illness representations, self-regulation, and genetic counseling: a theoretical review. J. Genet. Couns. 15, 325–337. doi: 10.1007/s10897-006-9044-5
Shiloh, S., Rashuk-Rosenthal, D., and Benyamini, Y. (2002). Illness causal attributions: an exploratory study of their structure and associations with other illness cognitions and perceptions of control. J. Behav. Med. 25, 373–394. doi: 10.1023/a:101581832390
Summers, K. M., Xu, D., West, J. A., McGill, J. J., Galbraith, A., Whight, C. M., et al. (2004). An integrated approach to management of Marfan syndrome caused by an FBN1 exon 18 mutation in an Australian aboriginal family. Clin. Genet. 65, 66–69. doi: 10.1111/j.1399-0004.2004.06005.x
Taylor, S. E., Lichtman, R. R., and Wood, J. V. (1988). Attributions, beliefs about control, and adjustment to breast cancer. J. Pers. Soc. Psychol. 46, 489–502. doi: 10.1037//0022-3514.46.3.489
Turnquist, D. C., Harvey, J. H., and Andersen, B. L. (1988). Attributions and adjustment to life-threatening illness. Br. J. Clin. Psychol. 27(Pt 1), 55–65. doi: 10.1111/j.2044-8260.1988.tb00753.x
Westover, A. N., and Nakonezny, P. A. (2010). Aortic dissection in young adults who abuse amphetamines. Am. J. Heart 160, 315–321. doi: 10.1016/j.amjheart.2010.05.021
Wong, D. R., Willett, W. C., and Rimm, E. B. (2007). Smoking, hypertension, alcohol consumption, and risk of abdominal aortic aneurysm in men. Am. J. Epidemiol. 165, 838–845. doi: 10.1093/aje/kw063
Conflict of Interest: BM has a remunerated consultant role with the company AstraZeneca with respect to an unrelated project.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Copyright © 2020 McInerney-Leo, West, Meiser, West, Brown and Duncan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3129,
1
],
[
3129,
9882,
2
],
[
9882,
15802,
3
],
[
15802,
21250,
4
],
[
21250,
26820,
5
],
[
26820,
33551,
6
],
[
33551,
40562,
7
],
[
40562,
50187,
8
],
[
50187,
56920,
9
],
[
56920,
61049,
10
]
]
} |
bba1369b65870974a110e3d079230e93f5bf651f | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 32,
"total-fallback-pages": 0,
"total-input-tokens": 154558,
"total-output-tokens": 40001,
"fieldofstudy": [
"Mathematics"
]
} | Subriemannian Metrics and the Metrizability of Parabolic Geometries
David M. J. Calderbank · Jan Slovák · Vladimír Souček
Received: 29 June 2019 / Published online: 16 November 2019
© Mathematica Josephina, Inc. 2019
Abstract
We present the linearized metrizability problem in the context of parabolic geometries and subriemannian geometry, generalizing the metrizability problem in projective geometry studied by R. Liouville in 1889. We give a general method for linearizability and a classification of all cases with irreducible defining distribution where this method applies. These tools lead to natural subriemannian metrics on generic distributions of interest in geometric control theory.
Keywords Projective metrizability · Subriemannian metrizability · Weyl connections · Cartan geometry · Overdetermined linear PDE · Parabolic geometry · Bernstein–Gelfand–Gelfand resolution
Mathematics Subject Classification Primary 53B15 · 53C17 · Secondary 14M15 · 17B10 · 22E46 · 53C15 · 53C30 · 58A32 · 58J70 · 93C10
1 Introduction
Many areas of geometric analysis and control theory deal with distributions on smooth manifolds, i.e. smooth subbundles of the tangent bundle. Let $\mathcal{H} \subseteq TM$ be such a distribution of rank $n$ on a smooth $m$-dimensional manifold $M$. A smooth curve
c : [a, b] → M (a ≤ b ∈ ℝ) is called horizontal if it is tangent to ℱ at every point, i.e.
for every t ∈ [a, b], the tangent vector c′(t) to c at c(t) ∈ M belongs to ℱ. It is well
known that, at least locally, any two points x, y ∈ M can be connected by a horizontal
curve c if and only if ℱ is bracket-generating in the sense that any tangent vector can be
obtained from iterated Lie brackets of sections of ℱ.
This paper is concerned with bracket-generating distributions arising in parabolic
geometries [5], which are Cartan–Tanaka geometries modelled on homogeneous
spaces G/P where G is a semisimple Lie group and P ⊆ G a parabolic subgroup. On a
manifold M equipped with such a parabolic geometry, each tangent space is modelled
on the P-module g/ℙ, and the socle ℱ of this P-module (the sum of its minimal nonzero
P-submodules) induces a bracket-generating distribution ℱ on M. Simple and well-
known examples include projective geometry and (Levi-nondegenerate) hypersurface
CR geometry: in the former case, g/ℙ is irreducible and so ℱ = TM, but in the latter
case ℱ is the corank one contact distribution of the hypersurface CR structure.
A more prototypical example for this paper is when ℱ ≤ TM is generic of rank n and corank \( \frac{1}{2}n(n - 1) \), i.e. \( m = \frac{1}{2}n(n + 1) = n + \frac{1}{2}n(n - 1) \), and \([Γ(ℱ), Γ(ℱ)] = Γ(TM)\). In this case the Lie bracket on sections of ℱ induces
an isomorphism \( \Lambda^2 ℱ \cong TM/ℱ \) and the distribution is said to be free. Any such
manifold is a parabolic geometry where \( G = SO(V) \) with \( \dim V = 2n + 1 \) and P is
the stabilizer of a maximal (n-dimensional) isotropic subspace U of V [9]. Then g/ℙ
has socle ℱ ∼= U* ⊗ (U⊥/U) with quotient isomorphic to \( \Lambda^2 ℱ \), and ℱ ≤ g/ℙ induces
the distribution ℱ ≤ TM on M.
While parabolic geometry is the main tool for the present work, our motivation is
subriemannian geometry, which concerns the following notion [18].
\textbf{Definition 1.1} Consider an m-dimensional manifold M with a given smooth distribution ℱ ≤ TM of constant rank n. A (pseudo-)Riemannian metric g on ℱ is called a horizontal or subriemannian metric on M.
Horizontal metrics are important in both geometric analysis and control theory. Among the horizontal curves joining two points, it may be important to find those which are optimal in some sense, for example those of shortest length with respect to a
horizontal metric. Horizontal metrics also allow for the definition of a hypo-elliptic
sublaplacian [16], allowing methods of harmonic analysis to be applied. However, this
raises the question: what is a good choice of horizontal metric?
For the distribution ℱ on a parabolic geometry, there is a natural compatibility
condition that can be imposed. Indeed, one of the key features of such a geometry is
that it admits a canonical class of connections, called Weyl connections, which form
an affine space modelled on the space of 1-forms.
\textbf{Definition 1.2} A horizontal metric on the distribution ℱ ≤ TM induced from a
parabolic geometry M is compatible if it is covariantly constant in horizontal direc-
tions with respect to some Weyl connection on M. We say M is (locally) metrizable
if there exists (locally) a compatible horizontal metric.
The metrizability problem has been studied for several classes of parabolic geometry
with ℱ = TM, in particular, the case of real projective. These examples exhibit
several interesting features, which we seek to generalize to all parabolic geometries—in particular to those with $\mathcal{H} \neq TM$.
First, whereas the metrizability condition appears to be highly nonlinear, it linearizes when viewed as a condition on the inverse metric on $\mathcal{H}^*$ multiplied by a suitable power of the horizontal volume form. Secondly, this linear equation is highly overdetermined, with a finite dimensional solution space. Hence parabolic geometries admitting such horizontal metrics are rather special. This has been used to extract detailed information about the structure of the geometry $[1,3,8,10,12,17,21]$.
If $h$ is the socle of $g/p$, it is not generally the case that $S^2 h$ is irreducible—indeed $h$ itself need not be irreducible. In order to generalize the studied examples, we introduce a condition on $P$-submodules $B \leq S^2 h$ containing nondegenerate elements, which we call the algebraic linearization condition (ALC). Our first main result (Theorem 1) justifies this terminology by showing that for parabolic geometries and $P$-submodules $B \leq S^2 h$ satisfying the ALC, there is a bijection between compatible horizontal metrics and nondegenerate solutions of an overdetermined first-order linear differential equation. (In fact, if $h$ is not irreducible we need a technical extra condition, which we call the strong ALC.)
Our second main result (Theorem 2) is a complete classification of all parabolic geometries and all $P$-submodules $B \leq S^2 h$ such that $h$ is irreducible and $B$ satisfies the ALC. The classification exhibits two nicely counterbalancing features. On the one hand, among parabolic geometries with irreducible socle, those admitting $P$-submodules $B \leq S^2 h$ satisfying the ALC are rare. On the other hand, the list of examples is quite long: we state the classification using three tables containing 14 infinite families and 6 exceptional cases. Many of these examples invite further study (see e.g. $[19]$).
The structure of the paper is as follows. In Sect. 2 we briefly outline the main notions and tools of parabolic geometry, referring to $[5]$ for details, but concentrating on examples. We also establish the local metrizability of the homogeneous model. In Sect. 3, we describe the linearization principle and prove Theorem 1. We give examples, and in particular show how explicit formulae can be obtained not only for the homogeneous model, but also for so-called normal solutions. Section 4 is devoted to the main classification result. We conclude by giving examples (Theorem 3) where the socle is not irreducible.
2 Background and Motivating Examples
We work throughout with real smooth manifolds $M$, real Lie groups $P$ and real Lie algebras $\mathfrak{p}$ [e.g. we view $\text{GL}(n, \mathbb{C})$ as a real Lie group and $\text{gl}(n, \mathbb{C})$ as a real Lie algebra].
A (real or complex) $P$-module $W$ is a finite dimensional (real or complex) vector space carrying a representation $\rho_W: P \to \text{GL}(W)$; $W$ is then also a $\mathfrak{p}$-module, where $\mathfrak{p}$ is the Lie algebra of $P$, i.e. it carries a representation $\tilde{\rho}_W: \mathfrak{p} \to \text{gl}(W)$. We write $\xi \cdot w$ for $\tilde{\rho}_W(\xi)(w)$. The nilpotent radical of $\mathfrak{p}$ is the intersection $\mathfrak{n}$ of the kernels of all simple $\mathfrak{p}$-modules. It is an ideal in $\mathfrak{p}$ and the quotient $\mathfrak{p}_0 := \mathfrak{p}/\mathfrak{n}$ is reductive. We let $P_0 := P/\exp \mathfrak{n}$ be the corresponding quotient group with Lie algebra $\mathfrak{p}_0$. Any $P$-module $W$ has a filtration
by \( P \)-submodules, where \( n \cdot W^{(j)} \) is the span of all \( \xi \cdot w \) with \( \xi \in n \) and \( w \in W^{(j)} \).
We let \( \text{gr}(W) := \bigoplus_{1 \leq j \leq k} W^{(j)}/W^{(j-1)} \), which is a \( P_0 \)-module.
### 2.1 Parabolic Geometries and Weyl Structures
Let \( P \leq G \) be a closed Lie subgroup of a Lie group \( G \), whose Lie algebra \( p \leq \mathfrak{g} \) has nilpotent radical \( n \trianglelefteq p \).
**Definition 2.1** A Cartan geometry of type \( G/P \) on a smooth manifold \( M \) is a principal \( P \)-bundle \( \mathcal{G} \rightarrow M \) equipped with a \( P \)-equivariant 1-form \( \theta : T\mathcal{G} \rightarrow \mathfrak{g} \) such that \( \theta_p : T_p\mathcal{G} \rightarrow \mathfrak{g} \) is an isomorphism for all \( p \in \mathcal{G} \), and \( \theta(X_\xi) = \xi \) for all \( \xi \in \mathfrak{p} \), where \( \xi \mapsto X_\xi \) is the infinitesimal \( \mathfrak{p} \)-action on \( \mathcal{G} \). The homogeneous model is the Cartan geometry \( G \rightarrow G/P \) equipped with the Maurer–Cartan form of \( G \).
Any \( P \)-module \( W \) induces a bundle \( \mathcal{W} := \mathcal{G} \times_\mathfrak{p} W \rightarrow M \). A filtration (2.1) of \( W \) induces a bundle filtration \( 0 = \mathcal{W}^{(0)} < \mathcal{W}^{(1)} < \cdots < \mathcal{W}^{(k)} = W \) with \( \text{gr}(\mathcal{W}) := \bigoplus_{k \in \mathbb{N}} \mathcal{W}^{(k)}/\mathcal{W}^{(k+1)} \cong \mathcal{G}_0 \times_{P_0} \text{gr}(W) \) where \( \mathcal{G}_0 := \mathcal{G}/\exp n \) is a principal \( P_0 \)-bundle.
In particular, taking \( W = \mathfrak{g}/\mathfrak{p} \), the projection of \( \theta \) onto \( \mathfrak{g}/\mathfrak{p} \) induces a bundle isomorphism \( TM \rightarrow \mathcal{G} \times_\mathfrak{p} \mathfrak{g}/\mathfrak{p} \). This \( P \)-module has an inductively defined filtration
\[
0 = h^{(0)} < h^{(1)} < \cdots < h^{(k)} = \mathfrak{g}/\mathfrak{p}, \quad \text{where} \quad h^{(j)} := \{x \in \mathfrak{g}/\mathfrak{p} \mid \forall \xi \in n, \ \xi \cdot x \in h^{(j-1)}\}.
\]
In particular \( h := h^{(1)} \) induces a distribution \( \mathcal{H} \leq TM \) on \( M \). We return to this in Sect. 2.3.
We specialize to the case that \( G \) is a semisimple Lie group and \( P \) is a parabolic subgroup of \( G \), meaning that the nilpotent radical of \( \mathfrak{p} \) is its Killing perp \( \mathfrak{p}^\perp \) in \( \mathfrak{g} \). Then Cartan geometries of type \( G/P \) are called parabolic geometries and have several distinctive features which we briefly explain and illustrate in the examples below (see [5] for further details).
First, the Killing form of \( \mathfrak{g} \) induces a duality between \( \mathfrak{p}^\perp \) and \( \mathfrak{g}/\mathfrak{p} \), and hence on any parabolic geometry of type \( G/P \), we have a natural isomorphism \( \mathcal{G} \times_\mathfrak{p} \mathfrak{p}^\perp \cong T^*M \) dual to the isomorphism \( TM \cong \mathcal{G} \times_\mathfrak{p} \mathfrak{g}/\mathfrak{p} \).
Secondly, the principal \( P_0 \)-bundle \( \mathcal{G}_0 \) has a distinguished family of principal connections called Weyl connections. To see this, it is convenient to fix a parabolic subalgebra \( \mathfrak{p}^{op} \) opposite to \( \mathfrak{p} \) in the sense that \( \mathfrak{g} = \mathfrak{p}^\perp \oplus \mathfrak{p}^{op} \). This identifies \( P_0 \) with a subgroup of \( P \), and induces a decomposition of \( P_0 \)-modules
\[
\mathfrak{g} = \mathfrak{m} \oplus \mathfrak{p}_0 \oplus \mathfrak{p}^\perp, \quad (2.2)
\]
where \( \mathfrak{m} \cong \mathfrak{g}/\mathfrak{p} \) is the nilpotent radical of \( \mathfrak{p}^{op} \). A Weyl structure is a \( P_0 \)-equivariant splitting \( \iota : \mathcal{G}_0 \hookrightarrow \mathcal{G} \) of the projection \( \mathcal{G} \rightarrow \mathcal{G}_0 \) (i.e. a reduction of structure group of \( \mathcal{G} \)).
to \( P_0 \); the corresponding Weyl connection is the \( p_0 \)-component of \( \iota^* \theta \). Weyl structures (or connections) form an affine space modelled on the space of 1-forms on \( M \).
**Summary**
A manifold \( M \) with a parabolic geometry of type \( G/P \) comes equipped with a filtration of the tangent bundle \( TM \), a \( G_0 \) structure on \( \text{gr}(TM) \), and a distinguished class of \( G_0 \)-connections (the Weyl connections).
There are general results [4,5] stating that these data are often sufficient to determine the parabolic geometry. Rather than explore this in generality, we turn to examples.
### 2.2 Projective Parabolic Geometries
We begin with some examples in which \( p^\perp \) is abelian, and hence the filtration of \( g/p \) is trivial and (2.2) is a \( \mathbb{Z} \)-grading of \( g \) as a Lie algebra, with \( p_0 \) in degree 0 and \( m, p^\perp \) in degree \( \pm 1 \) (also called a \( |1| \)-grading). There is thus a \( P_0 \)-structure on \( TM \) and an algebraic bracket \( [\cdot, \cdot] \) on \( TM \oplus p_0(M) \oplus T^*M \) where \( p_0(M) \leq \text{gl}(TM) \) is the bundle induced by \( p_0 \). In this case a Weyl connection induces a \( P_0 \)-connection \( \nabla \) on \( TM \) and any other Weyl connection is given (on vector fields \( Y, Z \)) by
\[
\hat{\nabla} Z Y = \nabla Z Y + [[Z, \gamma]], \gamma] = \nabla Z Y + [Z, \gamma] \cdot Y
\]
for some 1-form \( \gamma \), and we write \( \hat{\nabla} = \nabla + \gamma \) for short.
*Projective geometry* in dimension \( m \) may be viewed as a parabolic geometry of type \( G/P \) where \( G = \text{PGL}(m+1, \mathbb{R}) \) and \( P \) is the parabolic subgroup of block lower triangular matrices with blocks of sizes \( m \) and 1. Here \( m = \mathbb{R}^m \), \( p_0 = \text{gl}(m, \mathbb{R}) \), and \( p^\perp = \mathbb{R}^m \), and the homogeneous model \( G/P \) is \( m \)-dimensional real projective space \( \mathbb{R}P^m \).
On a parabolic geometry of this type, the \( G_0 \)-structure carries no information as \( G_0 \cong \text{GL}(m, \mathbb{R}) \), but two Weyl connections \( \nabla \) and \( \hat{\nabla} = \nabla + \gamma \) are related (on vector fields \( Y, Z \)) by
\[
\hat{\nabla} Z Y = \nabla Z Y + \gamma(Z)Y + \gamma(Y)Z.
\]
Using abstract indices we may write this as
\[
\hat{\nabla} a Y^b = \nabla a Y^b + \gamma_a Y^b + \gamma^c Y^c \delta_a^b.
\]
Thus the connections \( \nabla \) and \( \hat{\nabla} \) have the same torsion and the same geodesics (as unparametrized curves). Setting the torsion to zero, we have that the Weyl connections form a projective class \([\nabla]\).
*(Almost) c-projective geometry* is a complex analogue of projective geometry [3,14,15,23] with \( G = \text{PGL}(m+1, \mathbb{C}) \) and \( P \leq G \) block lower triangular as in the projective case, so the homogeneous model \( G/P \) is complex projective space \( \mathbb{C}P^m \) viewed as
a real homogeneous space. A parabolic geometry of this type on a \(2m\)-manifold \(M\) is given by an almost complex structure \(J \in \mathfrak{gl}(TM)\) and a class \([\nabla]\) of connections preserving \(J\) which differ by
\[
\hat{\nabla}_a Y^b = \nabla_a Y^b + \gamma_a Y^b - \gamma_c J^c_a J^d_b Y^d + \gamma_c Y^c \delta^b_a - \gamma_c J^c_a Y^d J^b_d.
\]
This can be obtained from the real projective formula by substituting \((1, 0)\)-forms \(\gamma - i J \gamma\) and \((1, 0)\) vectors into (2.4).
(A)Almost Grassmannian geometries are generalizations of real projective geometry with \(G = \text{PGL}(m + k, \mathbb{R})\) and \(P\) block lower triangular with blocks of size \(m\) and \(k\). The homogeneous model \(G/P\) is the Grassmannian of \(k\)-planes in \(\mathbb{R}^{m+k}\). On a parabolic geometry of this type, the \(G_0\)-structure is given by an identification of the tangent space with the tensor product of two auxiliary vector bundles \(E^*\) and \(F\) of ranks \(k\) and \(m\) (with \(\wedge^k E^* \simeq \wedge^m F\)). In abstract index notation, we write \(e_{A'}\) for a section of \(E\) and \(f_A\) for a section of \(F\), hence \(Y_{A'}\) for a vector field and \(\eta^B_A\) for a one-form.
The Weyl connections are tensor products of connections on \(E^*\) and \(F\) with fixed torsion, and the freedom in their choice is (cf. [5, p. 514])
\[
\hat{\nabla}^A_{A'} Y^B_{B'} = \nabla^A_{A'} Y^B_{B'} + \delta^B_{A'} \nabla^A_{C} Y^C_{B'} + \delta^B_{A'} \nabla^A_{C} Y^C_{B'}.
\]
When \(m = 2\ell\) and \(k = 2\) there is an interesting related geometry obtained by replacing \(\text{PGL}(2\ell + 2, \mathbb{R})\) by another real form of \(\text{PGL}(2\ell + 2, \mathbb{C})\), namely \(\text{PGL}(\ell + 1, \mathbb{H})\). The homogeneous model is then quaternionic projective space \(\mathbb{H}P^\ell\), and a parabolic geometry of this type is an (almost) quaternionic manifold [15].
### 2.3 Parabolic Geometries on Filtered Manifolds
We now turn to the examples of greater interest to us, in which \(\mathcal{H}\) is a proper subbundle of \(TM\). In fact, in these examples, the geometry is often entirely determined by the distribution \(\mathcal{H}\), as we now discuss.
Given a smooth manifold \(M\) of dimension \(m\), equipped with a distribution \(\mathcal{H} = \mathcal{H}^{(1)} \leq TM\) of rank \(n\), the Lie bracket of sections of \(\mathcal{H}\) (as vector fields) defines a bundle map \(\wedge^2 \mathcal{H} \to TM/\mathcal{H}\) called the Levi form of \(\mathcal{H}\). If we assume the image of the Levi form has constant rank, it defines a subbundle \(\mathcal{H}^{(2)} \leq TM\) with \(\mathcal{H}^{(2)}/\mathcal{H}\) equal to the image. We thus inductively define \(\mathcal{H}^{(j)} \leq \mathcal{H}^{(j+1)} \leq TM\) such that \(\mathcal{H}^{(j+1)}/\mathcal{H}^{(j)}\) is the image of the Lie bracket \(\mathcal{H} \otimes \mathcal{H}^{(j)} \to TM/\mathcal{H}^{(j)}\). If we further assume \(\mathcal{H}\) is bracket-generating, i.e. \(\mathcal{H}^{(k)} = TM\) for some \(k \in \mathbb{N}\), then we obtain a filtration
\[
0 = \mathcal{H}^{(0)} < \mathcal{H}^{(1)} < \cdots < \mathcal{H}^{(k)} = TM
\]
such that the Lie bracket of sections of \(\mathcal{H}^{(j)}\) and \(\mathcal{H}^{(j)}\) is a section of \(\mathcal{H}^{(j+1)}\). The associated graded vector bundle \(\text{gr}(TM)\) is, at each \(x \in M\), a graded Lie algebra \(\mathfrak{g}_x\) called the symbol algebra of \(\mathcal{H}\) at \(x\). We finally assume that the Lie algebras \(\mathfrak{g}_x\) are all isomorphic to the same nilpotent radical \(m\) of a fixed parabolic subalgebra \(\mathfrak{p}^\text{op}\) in a
semisimple Lie algebra \( g \). In many cases \( p_0 = p \cap p^{\text{op}}, \) where \( p \) and \( p^{\text{op}} \) are opposite in \( g \), is the full algebra of automorphisms of \( m \) (as a graded Lie algebra), and, as discussed in [4,5], this suffices to equip \( M \) with a parabolic geometry of type \( G/P \).
The decomposition (2.2) of \( g \) is no longer \(|1|\)-graded and this complicates the description of Weyl connections considerably. However, if we work only with horizontal (or partial) connections, i.e. restrict the Weyl connections to covariant derivatives in \( \mathcal{H} \) directions only, then the theory is as simple as in the \(|1|\)-graded case: the Lie bracket between \( m \) and \( p^\perp \) in \( g \) induces a Lie bracket between \( \mathfrak{h} \leq m \) and \( p^\perp/(p^\perp, p^\perp) \cong \mathfrak{h}^* \) with values in \( p_0 \), and hence an algebraic bracket \([\cdot, \cdot]: \mathcal{H} \otimes \mathcal{H}^* \rightarrow p_0(M)\). Any two Weyl connections \( \nabla \) and \( \tilde{\nabla} \) are related by
\[
\tilde{\nabla}_Z v = \nabla_Z v + \|Z, \gamma\| \cdot v,
\]
where \( \gamma \) is a section of \( \mathcal{H}^* \), \( Z \) is a section of \( \mathcal{H} \), and \( v \) is a section of \( G_0 \times p_0 \), \( V \) for any \( G_0 \)-module \( V \). We write \( \tilde{\nabla}|_{\mathcal{H}} = \nabla|_{\mathcal{H}} + \gamma \) for short.
**Free distributions** are parabolic geometries with \( G = SO(n + 1, n) \) and \( P \) block lower triangular with blocks of sizes \( n, 1, n \), where the inner product is defined on the standard basis \( e_0, e_1, \ldots, e_{2n} \) by \( \langle e_i, e_{n+1+i} \rangle = \langle e_n, e_n \rangle = \langle e_{n+1+i}, e_i \rangle = 1 \) for \( 0 \leq i \leq n-1 \) and all other inner products zero, see [9]. The homogeneous model \( G/P \) is the Grassmannian of maximal isotropic subspaces of \( \mathbb{R}^{2n+1} \). Elements of the Lie algebra \( g = so(n + 1, n) \) have the form
\[
\begin{pmatrix}
-A^T & -\xi^T & B \\
-\gamma^T & 0 & \xi \\
C & \gamma & A
\end{pmatrix}
\]
where \( B^T = -B \) and \( C^T = -C \). Here \( A \in \mathfrak{gl}(n, \mathbb{R}) \cong p_0, \xi \in \mathbb{R}^n \cong \mathfrak{h}, \gamma \in \mathbb{R}^{n*} \cong \mathfrak{h}^*, B \in \wedge^2 \mathbb{R}^n \cong \wedge^2 \mathfrak{h} \) and \( C \in \wedge^2 \mathbb{R}^{n*} \cong \wedge^2 \mathfrak{h}^* \).
A parabolic geometry of this type on a manifold \( M \) of dimension \( \frac{1}{2}n(n + 1) \) may be determined by a distribution \( \mathcal{H} \) of rank \( n \) whose Levi form \( \wedge^2 \mathcal{H} \rightarrow TM/\mathcal{H} \) is an isomorphism, hence the term “free distribution”. The \( P_0 \)-structure is no additional data, and Weyl connections may be determined as \( P_0 \)-connections \( \nabla \) such that for any sections \( Y, Z \) of \( \mathcal{H} \), the projection of \( \nabla_Z Y - \nabla_Y Z \) onto \( TM/\mathcal{H} \cong \wedge^2 \mathcal{H} \) is \( X \wedge Y \). If \( \tilde{\nabla}|_{\mathcal{H}} = \nabla|_{\mathcal{H}} + \gamma \) we then compute that
\[
\tilde{\nabla}_Z Y = \nabla_Z Y + \gamma(Y)Z. \tag{2.6}
\]
**Free CR or quaternionic CR distributions** are obtained by replacing \( so(n + 1, n) \) with \( g = su(n + 1, n) \) or \( sp(n + 1, n) \), again with (complex or quaternionic) blocks of sizes \( n, 1, n \), and \( p \) being block lower triangular [20]. Elements of \( g \) now have the form
\[
\begin{pmatrix}
-A^\dagger & -\xi^\dagger & B \\
-\gamma^\dagger & \mu & \xi \\
C & \gamma & A
\end{pmatrix}
\]
where $\dagger$ denotes the (complex or quaternionic) Hermitian conjugate, $B^\dagger = -B$, $C^\dagger = -C$ and $\overline{\mu} = -\mu$. We may thus compute, using matrix commutators
\[
\left[(\xi - \xi^\dagger, \gamma - \gamma^\dagger), \eta - \eta^\dagger\right] = \left(\xi^\gamma \eta + \eta(\gamma^\xi - \xi^\gamma \gamma^\dagger)\right) - \left(\xi^\gamma \eta + \eta(\gamma^\xi - \xi^\gamma \gamma^\dagger)\right)^\dagger.
\]
Note that the order here is important in the quaternionic case.
A parabolic geometry of this type has a complex or quaternionic rank $n$ distribution $\mathcal{H}$ for which the Levi form is complex or quaternionic skew-Hermitian, inducing an isomorphism of $TM/\mathcal{H}$ with such forms on $\mathcal{H}$. If $\hat{\nabla}|_\mathcal{H} = \nabla|_\mathcal{H} + \gamma$ we then have, on sections $Y, Z$ of $\mathcal{H}$,
\[
\hat{\nabla}_Z Y = \nabla_Z Y + Z \gamma(Y) + Y (\gamma(Z) - \overline{\gamma(Z)}).
\]
2.4 First BGG Operators, Local Metrizability of the Homogeneous Model, and Normal Solutions
Let $\mathcal{G} \rightarrow M, \theta$ be a Cartan geometry of type $G/P$. The extension of $\mathcal{G}$ by the left action of $P$ on $G$ is a principal $G$-bundle $\tilde{\mathcal{G}} = \mathcal{G} \times_\rho G$ with $G$-connection $\tilde{\theta} : \tilde{\mathcal{G}} \rightarrow g$, and (by construction) a reduction $\mathcal{G} \subseteq \tilde{\mathcal{G}}$ of structure group to $P$, and this provides an alternative description of the Cartan geometry. It follows that for any $G$-module $V$, there is a canonical induced linear connection on $\mathcal{V} = \mathcal{G} \times_\rho V \cong \tilde{\mathcal{G}} \times_G V$. These bundles are called tractor bundles and their sections tractors.
In the parabolic case, the BGG machinery of [2,6] provides a sequence of invariant linear differential operators between bundles induced by $P$-modules associated to $V$. The first such operator is defined on the bundle $\mathcal{G} \times_\rho V/(p^\perp \cdot V) \cong \mathcal{V}/(T^*M \cdot \mathcal{V})$ and is overdetermined.
When $M = G/P$ is the homogeneous model, the kernel of this first BGG operator is in bijection with the space of parallel sections of the tractor bundle $\mathcal{V}$, and the solutions have an explicit polynomial expression in normal coordinates. In more detail, fix an opposite parabolic subalgebra $p^{\text{op}}$ to $p \leq g$, inducing a decomposition (2.2). Then $\exp m \leq G$ is a unipotent subgroup of $G$ which determines a reduction $\mathcal{G}_0 \cong \mathfrak{p}_0 \exp m \leq G$ of the homogeneous model $G \rightarrow G/P$ to the structure group $\mathfrak{p}_0$ over the image $M$ of $\exp m$ in $G/P$, hence a Weyl connection over $M$, the normal flat Weyl connection.
Now if $\mathcal{V} = \mathcal{G}_0 \times_\rho \mathfrak{p}_0$ is induced by a $\mathfrak{p}_0$-module $V$, the Weyl covariant derivative of sections can be defined as the differentiation of $\mathfrak{p}_0$-equivariant $V$-valued functions on $\mathcal{G}_0$ in the direction of the constant vector fields with respect to the Weyl connection, and the subgroup $\exp m$ is tangent to all such constant vector fields. Thus any constant coordinate function $f : \exp m \rightarrow V$, with $f(x) = f_0$ for all $x \in \exp m$, defines a covariantly constant section with values in $\mathcal{V}$. In particular, choosing any nondegenerate symmetric 2-form $g$ in $S^2m^*$, the metric defined by the constant $g$ in the normal flat coordinates is covariantly constant with respect to the normal flat Weyl connection. Thus the homogeneous model $G/P$ is locally metrizable. By [7], such explicit formulae also apply on general curved geometries to the so-called normal solutions, which are those induced by parallel sections of the corresponding tractor bundle. We discuss this further in Sect. 3.5.
3 Metrizability and the Linearization Principle
3.1 First-Order Operators
In [22], the second and third authors developed a theory of invariant first-order linear operators for parabolic geometries, generalizing work of Fegan [11] in the conformal case (cf. [13, Appendix B]).
We first fix some notation. The Killing form of $\mathfrak{g}$ induces a nondegenerate invariant scalar product on $\mathfrak{p}_0 = \mathfrak{p}/\mathfrak{p}^\perp$, such that the decomposition into the semisimple part $\mathfrak{p}_0^{ss} = [\mathfrak{p}_0, \mathfrak{p}_0]$ and the centre $\mathfrak{z}(\mathfrak{p}_0)$ is orthogonal. Thus any Cartan subalgebra of $\mathfrak{p}_0 \otimes \mathbb{C}$ has an orthogonal decomposition $t = t^\prime \oplus t_0$, where $t^\prime$ is a Cartan subalgebra of $\mathfrak{p}_0^{ss} \otimes \mathbb{C}$ and $t_0 = \mathfrak{z}(\mathfrak{p}_0) \otimes \mathbb{C}$. Further, $t^* = t^* \oplus t_0^*$ is the dual decomposition, hence is orthogonal with respect to the induced scalar product on $t^*$. We write the corresponding decomposition of a weight $\lambda \in t^*$ as $\lambda = \lambda^\prime + \lambda^0$. Let $\Sigma_0$ be the set of simple roots $\alpha$ of $\mathfrak{g}$ whose root space $\mathfrak{g}_\alpha$ is in $\mathfrak{h}^* \otimes \mathbb{C}$. The remaining simple roots have root spaces in $\mathfrak{p}_0 \otimes \mathbb{C}$, and hence belong to $t^* \otimes \mathbb{C}$ (i.e. they vanish on $t_0$). Hence $\alpha^0$, for $\alpha \in \Sigma_0$, form a basis for $t_0^*$ (dual to the basis of $t_0$ formed by the fundamental coweights which belong to $t_0$).
Let $V_\lambda$ be an irreducible complex $\mathfrak{p}_0$-module with highest weight $\lambda = \lambda^\prime + \lambda^0 \in t^*$, let $\alpha = \alpha^\prime + \alpha^0 \in \Sigma_0$, and let $\mu = \mu^\prime + \mu^0$ be the highest weight of a component $V_\mu$ in the tensor product $V_\lambda \otimes V_\alpha$. The key observation from [22, Theorem 4.4] is that there is a first-order invariant operator between sections of the bundles induced by $V_\lambda$ and $V_\mu$ if and only if the scalar expression
$$c_{\lambda, \mu, \alpha} = \frac{1}{2}((\mu - \lambda, \mu + \lambda + 2\rho^\prime) - (\alpha, \alpha + 2\rho^\prime))$$
vanishes, where $\rho^\prime \in t^* \otimes \mathbb{C}$ is half the sum of the positive roots of $\mathfrak{p}_0$. We split this expression into contributions from $t^* \otimes \mathbb{C}$ and $t_0^*$ using the fact that $\mu^0 = \lambda^0 + \alpha^0$. Thus
$$c_{\lambda, \mu, \alpha} = c_{\lambda^\prime, \mu^\prime, \alpha^0} + \frac{1}{2}((\alpha^0, 2\lambda^0 + \alpha^0) - (\alpha^0, \alpha^0))$$
$$= c_{\lambda^\prime, \mu^\prime, \alpha^0} + (\lambda^0, \alpha^0). \quad (3.1)$$
If we fix $\lambda^\prime, \alpha$ and $\mu^\prime$, this decomposition provides one (real) linear equation on the central weight $\lambda^0$. This establishes the existence of many first-order operators [22]. Here we exploit (3.1) in a more specific way.
**Proposition 3.1** Let $\lambda^\prime \in t^*$ be the highest weight of a $\mathfrak{p}_0^{ss}$-module, and for each $\alpha \in \Sigma_0$, let $\mu^\prime_\alpha \in t^*$ be the highest weight of an irreducible component of $V_\mu^\prime \otimes V_\lambda^\prime$. Then there is a unique central weight $\lambda^0 \in t_0^*$ such that for all $\alpha \in \Sigma_0$, there is an invariant linear first-order operator between sections of the bundles induced by $V_\lambda$ and $V_\mu_\alpha$, where $\lambda = \lambda^\prime + \lambda^0$ and $\mu_\alpha = \mu^\prime_\alpha + \lambda^0 + \alpha^0$.
A particular case of this result arises when $\mu^\prime_\alpha = \lambda^\prime + \alpha^0$ so that $\mu_\alpha = \lambda + \alpha$ and $V_\mu_\alpha$ is the Cartan product of $V_\lambda$ and $V_\alpha$. In this case, the unique $\lambda^0$ is such that $(\lambda, \alpha) = 0$ for all $\alpha \in \Sigma_0$, so that $\lambda$ is a dominant weight for $\mathfrak{g}$ and the first-order system is the first BGG operator on the bundle induced by $V_\lambda$.
© Springer
3.2 The Algebraic Linearization Condition
Let \((G \to M, \theta)\) be a parabolic geometry of type \((G, P)\) and let \(\mathfrak{h}\) be the socle of the \(p\)-module \(g/p\), whose central weights form a basis of \(\mathfrak{z}(p_0)^*\). As we have seen, \(G \times_p \mathfrak{h} \subseteq G \times_p g/p \cong TM\) defines a (bracket-generating) “horizontal” distribution \(\mathcal{H} \subseteq TM\). Our aim is to construct compatible subriemannian (or pseudo-Riemannian) metrics, i.e. pseudo-Riemannian metrics \(g\) on \(\mathcal{H}\) for which there exists a horizontal metric Weyl connection (a Weyl connection \(\nabla\)).
Let \(c: \mathfrak{h}^* \otimes S^2\mathfrak{h} \to \mathfrak{h}\) be the natural contraction. We then posit the following.
**Definition 3.2** A nontrivial \(p_0\)-submodule \(B \leq S^2\mathfrak{h}\) satisfies the *algebraic linearization condition* (ALC) if and only if \(B\) has nondegenerate elements, and there exist \(p_0\)-submodules \(\mathfrak{h}_i \leq \mathfrak{h}\) and \(B_i \leq S^2\mathfrak{h}_i\) \((i \in \{1, \ldots, r\})\) with \(\mathfrak{h} = \bigoplus_{i=1}^r \mathfrak{h}_i\) and \(B = \bigoplus_{i=1}^r B_i\) such that for each \(i \in \{1, \ldots, r\}\), \(B_i\) is irreducible, and for any \(\alpha \in \Sigma_0\) and any irreducible component \(W\) of \(B_i \otimes \mathbb{C}\), \((V_\alpha \otimes W) \cap (\ker c \otimes \mathbb{C})\) is irreducible or zero.
**Remark 3.3** Note that \(\eta \in B\) is nondegenerate if and only if the same is true for each component \(\eta_i \in B_i\). The restrictions \(b_i: \mathfrak{h}^* \otimes B_i \to \mathfrak{h}_i\) of \(c\) are then surjective, and so we may write \(\mathfrak{h}^* \otimes B_i = \ker b_i \oplus \zeta_i(\mathfrak{h}_i)\) where \(\zeta_i: \mathfrak{h}_i \to \mathfrak{h}^* \otimes B_i\) is a \(p_0\)-invariant map with \(b_i \circ \zeta_i = id_{\mathfrak{h}_i}\). Since \(B_i\) is irreducible, it must lie in a single weight space of \(t_0\), with weight \(-\alpha^0 - \beta^0\) where \(\alpha, \beta \in \Sigma_0\); hence it is in the image of \(h_\alpha \otimes h_\beta \to S^2\mathfrak{h} \otimes \mathbb{C}\) for the corresponding weight spaces and so \(\mathfrak{h}_i \otimes \mathbb{C}\) has at most two irreducible components.
3.3 The Linearization Principle
Suppose first for simplicity that \(B \leq S^2\mathfrak{h}\) is absolutely irreducible and satisfies the ALC (so \(\mathfrak{h}\) has at most two irreducible components) and let \(\pi = id_{\mathfrak{h}^* \otimes B} - \zeta \circ b\) be the projection onto \(\ker(b: \mathfrak{h}^* \otimes B \to \mathfrak{h})\). The linearization method constructs a (pseudo-Riemannian) metric on \(\mathcal{H}\), i.e. a nondegenerate section \(g\) of \(S^2\mathcal{H}^*\) from a weighted inverse metric, i.e. a section \(\eta\) of \(S^2\mathcal{H} \otimes \mathcal{L}\) for some line bundle \(\mathcal{L}\). For this we suppose \(\eta\) is a section of \(B \otimes \mathcal{L}\), where \(B = G \times_p B\) and \(\mathcal{L}\) is a line bundle induced by a weight of \(\mathfrak{z}(p_0)\). We write \(b, \zeta, \pi\) also for the induced bundle homomorphisms (tensored by the identity on \(\mathcal{L}\)) and choose \(\mathcal{L}\) so that there is an invariant first-order linear operator \(D\) from \(\Gamma(B \otimes \mathcal{L})\) to \(\Gamma(\ker b)\) with \(D = \pi \circ \nabla|\mathcal{H}\) for any Weyl structure \(\nabla\). If \(\dim B = 1\), then \(\ker b = 0\), so \(D\) is the zero operator, and we take \(\mathcal{L}\) to be trivial. Otherwise \(\mathcal{L}, D\) are determined by Proposition 3.1. Due to the ALC, \(\ker b\) is then a sum of Cartan products of summands of \(\mathfrak{h}^*\) and \(B\), and the operator \(D\) is the first BGG operator.
Solutions \(\eta\) of the linear differential equation \(D\eta = 0\) are characterized by the fact that for some (hence any) Weyl structure \(\nabla\), there is a section \(X^\nabla\) of \(\mathcal{H} \otimes \mathcal{L}\) such that
\[\nabla|\mathcal{H}\eta = \zeta(X^\nabla).\]
Now suppose $\hat{\nabla}|_H = \nabla|_H + \gamma$ with $\gamma$ in $H^*$. Then for any $Z \in \Gamma H$, $\hat{\nabla} Z \eta = \nabla Z \eta + [\nabla, \gamma] \cdot \eta$, and $[\cdot, \gamma] \cdot \eta$ is in the image of $\zeta$ by the invariance of $\mathcal{D}$. Hence by Schur’s Lemma and §3.1 (i.e. [22]):
$$[\cdot, \gamma] \cdot \eta = (\zeta \circ b)([\cdot, \gamma] \cdot \eta) = (\zeta \circ b)(\sum_{\alpha \in \Sigma_0} \ell_\alpha \gamma_\alpha \otimes \eta)$$
for nonzero scalars $\ell_\alpha$, where $\gamma = \sum_{\alpha \in \Sigma_0} \gamma_\alpha$ with $\gamma_\alpha \in V_\alpha \subseteq h^* \otimes \mathbb{C}$. If we define $\hat{\nabla}|_H \eta = \hat{\nabla}|_H \eta + \zeta(\hat{\nabla}|_H \eta))$.
Now if $\eta$ is a nondegenerate solution of $\mathcal{D} \eta = 0$, with $\nabla|_H \eta = \zeta(X^\nabla)$ for some Weyl connection $\nabla$ and $X^\nabla \in \Gamma(\mathcal{H} \otimes \mathcal{L})$, we may take $\gamma = -\eta^{-1}(X^\nabla)$ to obtain
$$\hat{\nabla}|_H \eta = \zeta(X^\nabla) + \zeta(\hat{\nabla}|_H \eta) = 0.$$
Hence $\eta$ is (inverse to) a horizontal compatible metric, up to the shift of the weight via the line bundle $\mathcal{L}$. Finally, the nondegenerate weighted metric $\eta$ allows us to build a nonvanishing section $\sigma$ of the line bundle $\wedge^m H \otimes \mathcal{L}^{m/2}$, where $m = \dim h$, with $\hat{\nabla}|_H \sigma = 0$. This line bundle cannot be trivial because the central weight of $B \otimes \mathcal{L}$ is not zero. If $h$ is absolutely irreducible, then $\wedge^m H \otimes \mathcal{L}^{m/2} \cong \mathcal{L}^k$ for some nonzero $k$, and then $\psi = (\sigma^{-1/k} \eta)^{-1}$ is a section of $B^*$ with $\hat{\nabla}|_H \psi = 0$. Otherwise, we need to assume the central weights of $\wedge^m H$ and $\mathcal{L}$ are linearly dependent. The most natural way to achieve this is to suppose that the simple roots $\alpha, \beta$ with $h \otimes \mathbb{C} = h_\alpha \oplus h_\beta$ are related by an automorphism of the Dynkin diagram of $\mathfrak{g}$.
**Definition 3.4** A $p_0$-submodule $B \leq S^2 h$ satisfies the strong algebraic linearization condition (strong ALC) if and only if $B$ satisfies the ALC with respect to $p_0$-submodules $h_i \leq h$ such that whenever $h_i \otimes \mathbb{C} = h_\alpha \oplus h_\beta$ for $\alpha, \beta \in \Sigma_0$, there is an automorphism of the Dynkin diagram of $\mathfrak{g}$ preserving $\Sigma_0$ and interchanging $\alpha$ and $\beta$.
With this definition, the linearization method yields the following result.
**Theorem 1** Let $B \leq S^2 h$ satisfy the strong ALC with respect to $B = \bigoplus_{i=1}^r B_i$ and $h = \bigoplus_{i=1}^r h_i$. Then for all $i \in \{1, \ldots, r\}$ there are induced line bundles $\mathcal{L}_i$ and invariant first-order linear operators $\mathcal{D}_i$ acting on sections of $B_i \otimes \mathcal{L}_i$ such that there is a bijection between nondegenerate solutions $\eta_i : i \in \{1, \ldots, r\}$ of the equations $\mathcal{D}_i(\eta_i) = 0$, and nondegenerate sections $\psi$ of $B^*$ with $\nabla|_H \psi = 0$ for some Weyl connection $\nabla$.
**Proof** Define $b_i, \zeta_i$ as in Remark 3.3 so that $h^* \otimes B_i = \ker b_i \oplus \zeta_i(h_i)$, let $\pi_i = \text{id}_{h^*} \otimes b_i - \zeta_i \circ b_i$ be the projection onto $\ker b_i$, and let $\Sigma_0^l = \{\alpha \in \Sigma_0 : V_\alpha \subseteq h_i^* \otimes \mathbb{C}\}$. We apply the same ideas as in the absolutely irreducible case to each irreducible component $V_{\kappa'}$ of $B_i \otimes \mathbb{C}$. If $\dim V_{\kappa'} \geq 2$ then the ALC implies that $(V_\alpha \otimes V_{\kappa'}) \cap (\ker b_i \otimes \mathbb{C})$ is irreducible for all $\alpha \in \Sigma_0$, and hence Proposition 3.1 provides a
unique $\lambda^0$ so that there is an invariant first-order operator between sections of the bundles induced by $V_\lambda$ and $\ker b_i \otimes \mathbb{C}$. If instead, $\dim V_{\lambda'} = 1$, then $V_\alpha \otimes V_{\lambda'}$ is irreducible, and is contained in $\ker b_i \otimes \mathbb{C}$ unless $\alpha \in \Sigma_i^0$. We thus supplement (3.1) by the equations $(\lambda^0, \alpha^0) = 0$ when $\alpha \in \Sigma_i^0$.
Since $B_i$ is irreducible, $B_i \otimes \mathbb{C}$ is either irreducible or has two reducible components with conjugate weights. Now the system of equations (3.1) and $(\lambda^0, \alpha^0) = 0$ that we impose to find $\lambda^0$ are conjugation invariant. Hence in either case, we obtain a line bundle $\mathcal{L}_i$ and an invariant first-order linear operator $\mathcal{D}_i := \pi_i \circ \nabla \mid_{\mathcal{H}}$ on $B_i \otimes \mathcal{L}_i$, so that any section $\eta_i$ satisfies $\mathcal{D}_i(\eta_i) = 0$ if and only if
$$\nabla \mid_{\mathcal{H}} \eta_i = \zeta_i(X_i^\mathcal{V})$$
for a suitable section $X_i^\mathcal{V}$ of $\mathcal{H}_i \otimes \mathcal{L}_i$. Given such sections $\eta_i$, let $\eta = \sum_{i=1}^r \eta_i$. By construction, the operator $b_i \circ \nabla \mid_{\mathcal{H}}$ is not invariant on $B_i \otimes \mathcal{L}_i$. Hence by Schur’s Lemma, there are nonzero scalars $\ell_\alpha$ such that
$$\langle \cdot, \mathcal{V} \rangle \cdot \eta = (\zeta \circ b)(\langle \cdot, \mathcal{V} \rangle \cdot \eta) = (\zeta \circ b) \left( \sum_{i=1}^r \sum_{\alpha \in \Sigma_0} \ell_\alpha Y_\alpha \otimes \eta_i \right) = \left( \sum_{\alpha \in \Sigma_0} \ell_\alpha Y_\alpha \otimes \eta \right),$$
where $Y = \sum_{\alpha \in \Sigma_0} Y_\alpha$ as before. As before, we define $\sharp_\eta(Y) = \sum_{\alpha \in \Sigma_0} \ell_\alpha b(Y_\alpha \otimes \eta)$, so that if $\hat{\nabla} \mid_{\mathcal{H}} = \nabla \mid_{\mathcal{H}} + Y$ then
$$\hat{\nabla} \mid_{\mathcal{H}} \eta = \nabla \mid_{\mathcal{H}} \eta + \zeta(\sharp_\eta(Y)).$$
If $\eta$ is a nondegenerate then $\sharp_\eta$ is invertible, and so if $\mathcal{D}(\eta) = 0$, i.e. $\mathcal{D}_i(\eta_i) = 0$ for all $i$, then we may set $Y := -\sharp_\eta^{-1}(X^\mathcal{V})$, where $X^\mathcal{V} = \sum_{i=1}^r X_i^\mathcal{V}$ to obtain $\hat{\nabla} \mid_{\mathcal{H}} \eta = 0$.
Finally, taking volume forms of $\eta_i$ on $\mathcal{H}_i$ for each $i$, we obtain nonvanishing sections $\sigma_i$ of $\wedge^{m_i} \mathcal{H}_i \otimes \mathcal{L}_i^{m_i/2}$ with $\hat{\nabla} \mid_{\mathcal{H}} \sigma_i = 0$. The weights of the $\sigma_i$ are linearly independent, and the strong ALC ensures that the central weights of the $\mathcal{L}_i$ are linear combinations of the central weights of $\wedge^{m_j} \mathcal{H}_j$, so for every $i$, we can solve the linear system $\eta_i \otimes \otimes_j \sigma_j^{a_{ij}} \in B_i$, and hence, inverting each component, obtain the section $\psi$ of $B^*$ as required. Since the system is invertible, $\eta$ can be obtained from $\psi$ and its volume forms on each $\mathcal{H}_i$. \hfill $\square$
If only the ALC is assumed, then the proof yields, in place of horizontally parallel metrics on $\mathcal{H}$, horizontally parallel conformal structures on each $\mathcal{H}_i$ and horizontally parallel sections of some line bundles.
### 3.4 Example: Projective Geometry
Let us illustrate the metrizability procedure by showing how the well-known example of projective geometry $[8,10,17,21]$ fits into the general method. Here $g = sl(n + 1, \mathbb{R}) = \mathfrak{h} \oplus \mathfrak{gl}(\mathfrak{h}) \oplus \mathfrak{h}^*$ and $S^2 \mathfrak{h}$ is irreducible. Since $\mathfrak{h}^* \otimes S^2 \mathfrak{h} \cong \mathfrak{h} \oplus (\mathfrak{h}^* \otimes_0 S^2 \mathfrak{h})$, where the second summand is the tracefree part (the Cartan product), $B = S^2 \mathfrak{h}$ satisfies the ALC. The class of covariant derivatives defining the projective structure depends
\[ \mathfrak{h} \] Springer
on an arbitrary 1-form $\Upsilon_a$ and two of them are related by (2.4). Hence on a section $\varphi$ of $B = S^2 TM$, we have
$$[Z, \Upsilon] \cdot \varphi = 2\Upsilon(Z)\varphi + Z \otimes \varphi(\Upsilon, \cdot) + \varphi(\Upsilon, \cdot) \otimes Z$$
for any vector field $Z$ and 1-form $\Upsilon$. If we twist by the line bundle $L$ induced by the $P_0$-module $L$ with highest weight $-2\omega_1$, then for $\eta \in \Gamma(B \otimes L)$ and $\hat{\nabla} = \nabla + \Upsilon$, we have
$$\hat{\nabla}_Z \eta = \nabla Z \eta + b(\Upsilon \otimes \eta) \odot Z$$
where $X \odot Z = X \otimes Z + Z \otimes X$ and $b(\Upsilon \otimes \eta) = \eta(\Upsilon, \cdot)$ is the natural contraction. In abstract indices this contraction of $\Upsilon_c \eta^{ab}$ is $\Upsilon_a \eta^{ab}$ and hence
$$\hat{\nabla}_c \eta^{ab} = \nabla_c \eta^{ab} + \delta^a_c \Upsilon_d \eta^{bd} + \delta^b_c \Upsilon_d \eta^{ad}.$$
We thus have an invariant first-order operator acting on $\eta$ (a first BGG operator) whose solutions satisfy
$$\nabla Z \eta = \langle Z, \xi(X^\nabla) \rangle = \frac{1}{n + 1} X^\nabla \otimes Z$$
for some section $X^\nabla$ of $TM \otimes L$. [Here $\xi(X) = \frac{1}{n + 1} X \otimes id$, or in abstract indices, $\xi(X^a) = \frac{1}{n + 1} (X^a \delta^b_c + X^b \delta^a_c)$ so that $b(\xi(X)) = X$.] Evidently $\eta$ is parallel for $\hat{\nabla}$ provided $b(\Upsilon \otimes \eta) = -\frac{1}{n + 1} X^\nabla$, which we can solve for $\Upsilon$ if $\eta$ is nondegenerate. Direct computation shows that $\det(\eta)$ is a section of $L^{-2}$. So $g^{ab} := \det(\eta) \eta^{ab}$ is a nondegenerate section of $S^2 TM$ and its inverse is parallel with respect to $\hat{\nabla}$. In terms of the general theory herein, if
$$[\cdot, \Upsilon] \cdot \eta = \ell \langle \xi \circ b(\Upsilon \otimes \eta),$$
then
$$b(\Upsilon \otimes \eta) \odot Z = [Z, \Upsilon] \cdot \eta = \ell \langle Z, (\xi \circ b)(\Upsilon \otimes \eta) \rangle = \frac{\ell}{n + 1} b(\Upsilon \otimes \eta) \odot Z,$$
and so $\ell = n + 1$. Hence $\sharp_\eta(\Upsilon) = (n + 1) b(\Upsilon \otimes \eta)$ and the solution is $\Upsilon = -\#^{-1}_\eta(X^\nabla)$.
### 3.5 The Metric Tractor Bundle
As we have seen in Sect. 2.4, the homogeneous model $G/P$ is always locally metrizable, and solutions in the kernel of a given first BGG operator are induced by parallel sections of a corresponding metric tractor bundle. In general, if $M$ has nontrivial curvature, not all solutions to a linearized metrizability problem will correspond to such parallel sections: as discussed in Sect. 2.4, those that do are called normal solutions and exhibit special features. In particular, as shown in [7], they are always of a simple polynomial forms in normal coordinates, exactly as in the homogeneous model. Thus the explicit formulae from the homogeneous case form an Ansatz for solutions in general.
Let us discuss this in the case of free distributions from Sect. 2.3. Here \( g = \mathfrak{so}(n+1, n) = \wedge^2 \mathfrak{h} \oplus \mathfrak{h} \oplus \mathfrak{gl}(\mathfrak{h}) \oplus \mathfrak{h}^* \oplus \wedge^2 \mathfrak{h}^* \), \( B = S^2 \mathfrak{h} \) is irreducible and satisfies the ALC, just as in the case of projective geometry. In this case, however, there is no need to twist by a line bundle, since by (2.6), we already have
\[
\hat{\nabla}_Z \eta = \nabla_Z \eta + b(\mathcal{Y} \otimes \eta) \otimes Z
\]
for any sections \( Z \) of \( \mathcal{H} \) and \( \eta \) of \( S^2 \mathfrak{h} \), where \( \hat{\nabla}|_{\mathcal{H}} = \nabla|_{\mathcal{H}} + \mathcal{Y} \) and \( b \) is the natural contraction. The solution of the linearized metrizability problem then proceeds exactly as in the projective case, so we now consider the form of the normal solutions.
The standard tractor bundle is the bundle associated to the defining representation \( V \) of \( G = \text{SO}(n+1, n) \). Explicitly, using the matrix description in Sect. 2.3, we may write elements of \( V \) as column vectors
\[
v = \begin{pmatrix} \lambda^a \\ \tau \\ \ell_a \end{pmatrix}
\]
on which the action of the nilpotent radical \( m \) of \( \mathfrak{p}^{\text{op}} \) is given by
\[
x \cdot v = \begin{pmatrix} 0 & x^a & y^{ab} \\ 0 & 0 & -x^a \\ 0 & 0 & 0 \end{pmatrix} \begin{pmatrix} \lambda^b \\ \tau \\ \ell_b \end{pmatrix} = \begin{pmatrix} x^a \tau + y^{ab} \ell_b \\ -x^b \ell_b \\ 0 \end{pmatrix}.
\]
The metric tractor bundle in this example is associated to the symmetric tracefree square \( S^2_0 V \) of \( V \). Elements of the symmetric square \( S^2 V \) are given by
\[
\Phi = \begin{pmatrix} \nu^{ab} \\ \sigma^b \\ \kappa | \psi^c_b \\ \xi_b \\ \tau_{bc} \end{pmatrix},
\]
where \( \nu^{ab} \) and \( \tau_{bc} \) are symmetric, and such an element is in \( S^2_0 V \) if \( \kappa = -\psi^c_c \). Our convention is such that \( \Phi = v \otimes \bar{v} \) has components
\[
\nu^{ab} = \lambda^a \bar{\lambda}^b + \bar{\lambda}^a \lambda^b; \quad \sigma^b = \lambda^b \bar{\tau} + \bar{\lambda}^b \tau; \quad \kappa = \tau \bar{\tau}; \quad \psi^c_b = \ell_b \bar{\lambda}^c + \lambda^c \bar{\ell}_b; \\
\xi_b = \ell_b \bar{\tau} + \tau \bar{\ell}_b; \quad \tau_{bc} = \ell_a \bar{\ell}_b + \bar{\ell}_a \ell_b.
\]
The action of the nilpotent radical on the symmetric square is given by
\[
x \cdot \Phi := \begin{pmatrix} 0 & x^a & y^{ab} \\ 0 & 0 & -x^a \\ 0 & 0 & 0 \end{pmatrix} \begin{pmatrix} \nu^{ab} \\ \sigma^b \\ \kappa | \psi^c_b \\ \xi_b \\ \tau_{bc} \end{pmatrix} = \begin{pmatrix} x^a \sigma^b - y^c(a \psi^b_c) \\ x^c \psi^b_c + y^{bc} \xi_c - x^b \kappa \\ -x^b \xi_b | x^c \xi_b \\ -x^a \tau_{ab} \\ 0 \end{pmatrix}.
\]
where \(x^{(a \sigma^b)} = x^a \otimes \sigma^b + \sigma^b \otimes x^a\). The iterated action is therefore given by
\[
x \cdot x \cdot \Phi = \begin{pmatrix}
x^c x^{(a \psi^b)} + 2x^{(a y^b)c} \xi_c - x^a x^b \kappa \\
2x^c x^b \xi_c - \gamma^{bc} x^a \tau_{ac} \\
x^b x^a \tau_{ab} & -x^c x^a \tau_{ab}
\end{pmatrix},
\]
\[
x \cdot x \cdot x \cdot \Phi = \begin{pmatrix}
4x^a x^b x^c \xi_c - 2x^{(a y^b)c} x^d \tau_{dc} \\
-2x^b x^a x^c \tau_{ac} & 0 & 0 & 0
\end{pmatrix},
\]
\[
x \cdot x \cdot x \cdot x \cdot \Phi = \begin{pmatrix}
-4x^a x^b x^c x^d \tau_{cd} \\
0 & 0 & 0 & 0 & 0
\end{pmatrix},
\]
with all further iterates zero. The normal solution is the projection onto \(S^2 \mathcal{H}\) of \(\exp(x) \cdot \Phi\), which is given by
\[
\eta^{ab}(x, y) = \nu^{ab} + x^{(a \sigma^b)} - y^{(a \tau^b)} + \frac{1}{2} x^c x^{(a \psi^b)} + x^{(a y^b)c} \xi_c + \frac{1}{2} x^a x^b \psi_c + \frac{2}{3} x^a x^b x^c \xi_c - \frac{1}{3} x^{(a y^b)c} x^d \tau_{dc} - \frac{1}{6} x^a x^b x^c x^d \tau_{cd}.
\]
4 Classification of Metric Parabolic Geometries with Irreducible \(\mathfrak{h}\)
We have seen that the linearizability problem of the existence of compatible sub-riemannian metrics on parabolic geometries reduces to a purely algebraic question related to the number of components in certain tensor products of the \(p_0\)-modules \(\mathfrak{h}\) and its dual \(\mathfrak{h}^*\). In fact, we are only interested in the actions of the semisimple part of \(p_0 = p/p^\perp\).
In this section, we classify all cases of the ALC where the defining distribution of the parabolic geometry corresponding to \(\mathfrak{h}\) is irreducible. This is the case with all \([1]\)-graded geometries, but many \([2]\)-graded and some more general geometries are involved too. In order to keep the story short, while still providing a complete and simple picture, we use the schematic description of the chosen type of parabolic subalgebra \(p\) of \(g\) by crosses on the Dynkin diagram for \(g\) and we write weights of \(p\)-modules as linear combinations of the fundamental weights for \(g\), depicted as the nonzero coefficients over the nodes of the diagrams, ignoring those over the crossed nodes (see e.g. [5, §3.2] for these conventions). This exactly provides the complete information on the representation of the semisimple part of \(p_0\) in the case of complex algebras and we always add further information on specific real forms of them. Actually for practical reasons (and in accordance with common practice), we rather write the
weights of the dual $p_0$-modules over the Dynkin diagram. Moreover, the displayed diagrams and weights always correspond to the complexified versions and thus we have to keep in mind their meaning for particular real forms.
The classification is given in the following theorem. In the proof we also describe the geometric properties of the metrics in any admissible component $B$, mostly in terms of special structure related to the given parabolic geometry. The classification in Table 1 was also obtained in [19].
**Theorem 2** Let $p$ be a parabolic subalgebra in a real simple Lie algebra $g$ and let $B$ be a $p$-submodule of $S^2\mathfrak{h}$, with $\mathfrak{h} \cong (p^\perp/[p^\perp, p^\perp])^*$ irreducible. Then $B$ satisfies the ALC and admits nondegenerate elements if and only if one of the following holds:
- $g$ is complex and the complexification of $(p, B)$ appears in Table 1;
- $(g, p, B)$ appears as a real form in Table 2 or 3;
- $(g, p, B)$ is (the underlying real Lie algebra of) the complexification of a triple appearing in Table 2.
Table 3 Real geometries with \( \mathfrak{h} \) not absolutely irreducible
| Case | Diagram \( \Delta_\ell \) for \( p, B \) | Real simple \( g \) | Growth |
|---------------|---------------------------------|-----------------------------------|---------------------|
| \( A_2^{2,1} \) | \[
\begin{array}{c}
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\end{array}
\] | \( \text{su}(1, 3), \text{su}(2, 2) \) | \( 4, 5 \) |
| \( A_2^{2,k} \) | \[
\begin{array}{c}
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\end{array}
\] | \( \text{su}(p, q), k \leq p \leq q \) | \( d = 2k(\ell - 2k + 1), \) |
| | \( \ell = p + q - 1 \geq 4 \) | \( n = d + k^2 \) | |
| \( A_2^{2,h} \) | \[
\begin{array}{c}
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\end{array}
\] | \( \text{so}(p, q), 2 \leq p \leq q \) | \( 4(\ell - 3), 4(\ell - 2) \) |
| | \( \ell = p + q - 1 \geq 6 \) | | |
| \( A_2^{2,s} \) | \[
\begin{array}{c}
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\end{array}
\] | \( \text{su}(k, k + 2), \) | \( 4k, 4k + k^2 \) |
| \( A_2^{2,s} \) | \[
\begin{array}{c}
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\end{array}
\] | \( \text{su}(k + 1, k + 1) \) | \( 4k, 4k + k^2 \) |
| \( A_2^{2,k} \) | \[
\begin{array}{c}
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\bullet \quad \quad \quad \bullet \\
\end{array}
\] | \( \text{su}(k, k + 1) \) | \( 2k, 2k + k^2 \) |
| \( D_2^{2,\ell} \) | \[
\begin{array}{c}
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\end{array}
\] | \( \text{so}(\ell - 1, \ell + 1) \) | \( d = 2(\ell - 1), \) |
| \( D_2^{2,h} \) | \[
\begin{array}{c}
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\end{array}
\] | \( \text{so}^*(2\ell), \ell = 2j + 1 \) | \( d + \frac{1}{2}(\ell - 1)(\ell - 2) \) |
| \( D_6^{2,h} \) | \[
\begin{array}{c}
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\end{array}
\] | \( \text{so}(\ell - 1, \ell + 1) \) | \( d = 2(\ell - 1), \) |
| | \( \text{so}^*(2\ell), \ell = 2j + 1 \) | \( d + \frac{1}{2}(\ell - 1)(\ell - 2) \) | |
| \( E_6^{2,h} \) | \[
\begin{array}{c}
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\bullet \quad \quad \bullet \\
\end{array}
\] | \( E_{6(2)} \) | \( 16, 24 \) |
**Outline of Proof** In the gradings of the complex algebras $\mathfrak{g}$ corresponding to parabolic geometries, the number of irreducible components of $\mathfrak{h}^*$ is equal to the number of crosses in the Dynkin diagram describing the chosen parabolic subalgebra. However, in the real forms of $\mathfrak{g}$, there might be complex or quaternionic components giving rise to two components in the complexification. These two complex components have to be either conjugate (in the complex case) or isomorphic (in the quaternionic case).
The latter observation reduces our quest to diagrams with two crosses placed in a symmetric way. Indeed, more than two crosses cannot result in one component, while asymmetric positions of the crosses inevitably yield two complex components which are neither conjugate nor isomorphic. Moreover, having two components in the complexified $\mathfrak{h}$, we may ignore the symmetric products of the individual parts in $S^2\mathfrak{h}$, because there cannot be any nondegenerate metrics there.
We first dispense with the case that $\mathfrak{g}$ is complex but $\mathfrak{B}$ is not, so that $\mathfrak{B} \otimes \mathbb{C}$ is irreducible in $\mathfrak{g} \otimes \mathbb{C} \cong \mathfrak{g} \oplus \mathfrak{g}$ and the diagram for $(\mathfrak{p}, \mathfrak{B})$ is invariant under the automorphism exchanging the two components of the Dynkin diagram. Thus $\mathfrak{B} \otimes \mathbb{C} = \mathfrak{h}_\alpha \otimes \mathfrak{h}_\beta$ where $\mathfrak{h} \otimes \mathbb{C} = \mathfrak{h}_\alpha \oplus \mathfrak{h}_\beta$. Now the ALC is satisfied provided $\mathfrak{h}_\alpha \otimes \mathfrak{h}_\alpha^*$ (and hence also $\mathfrak{h}_\beta \otimes \mathfrak{h}_\beta^*$) has precisely two irreducible components as a representation of a component of $\mathfrak{p}_0 \otimes \mathbb{C}$. Only the (dual) defining representations in type $A$ have this property, and so $\mathfrak{g}$ must have type $A$, $B$ or $G$, where the nodes crossed in $\mathfrak{g} \otimes \mathbb{C}$ are end nodes corresponding to short simple roots. The possibilities are listed in Table 1, covering the following three cases:
**Case 1** ($A^h_\ell$). The c-projective geometries may be equipped with distinguished Hermitian metrics.
**Case 2** ($B^h_\ell$). The almost complex version of a free distribution of rank $\ell$, may be equipped with distinguished Hermitian metrics.
**Case 3** ($G^h_2$). The almost complex version of the $(2, 3, 5)$-distributions may be equipped with distinguished Hermitian metrics.
We analyse the remaining real cases with irreducible $\mathfrak{h}$ by the Dynkin type of $\mathfrak{g}$ in the following sections.
**4.1 Proof of Theorem 2 When $\mathfrak{g}$ Has Type $A_\ell$**
The case $\ell = 1$ is trivial, so we assume $\ell \geq 2$, and first consider the case of a single crossed node. If the crossed node is one of the ends of the Dynkin diagram, the only real $\mathfrak{g}$ is the split form, $\mathfrak{h}$ and $S^2\mathfrak{h}$ are irreducible, and $\mathfrak{B} = S^2\mathfrak{h}$ satisfies the ALC: when $\ell = 2$,
$$\mathfrak{B} \simeq \begin{array}{c} 2 \end{array} \mathfrak{h}^* \otimes \mathfrak{B} \simeq \begin{array}{c} 3 \end{array} \oplus \begin{array}{c} 1 \end{array}$$
and when $\ell \geq 3$,
$$\mathfrak{B} \simeq \begin{array}{c} 2 \end{array} \mathfrak{h}^* \otimes \mathfrak{B} \simeq \begin{array}{c} 1 \end{array} \oplus \begin{array}{c} 2 \end{array} \oplus \begin{array}{c} 1 \end{array} \oplus \begin{array}{c} 1 \end{array}$$
These examples can be summarized in the following statement.
Case 4 \((A^{1,1}_\ell)\). Here \(g = \text{sl}(\ell + 1, \mathbb{R})\), \(\ell \geq 2\), \(\mathfrak{h} \cong \mathbb{R}^\ell\) and \(B = S^2\mathfrak{h}\). This is the most classical case of projective structures on \(\ell\)-dimensional manifolds \(M\), and nondegenerate sections of \(B\) are inverse to arbitrary pseudo-Riemannian metrics on \(M\).
Suppose next that the cross is adjacent to one end of the diagram, with \(\ell \geq 3\). We then have \(S^2\mathfrak{h} = B \oplus B'\), where
\[
\mathfrak{h} \cong \begin{array}{c}
\vdots \\
\vdots \\
1
\end{array} \quad \mathfrak{h}^* \cong \begin{array}{c}
\vdots \\
\vdots \\
1
\end{array} \\
B \cong \begin{array}{c}
\vdots \\
\vdots \\
1
\end{array} (\ell \geq 4) \quad B' \cong \begin{array}{c}
\vdots \\
\vdots \\
2
\end{array}
\]
and \(B\) is trivial for \(\ell = 3\) (when \(\mathfrak{h} \cong \mathfrak{h}^*\)). The tensor product \(\mathfrak{h}^* \otimes B'\) decomposes into four irreducible components, except for the real form \(\text{su}(2, 2)\) when \(\ell = 3\), in which case there are only three components. In any case, \(B'\) does not satisfy the ALC.
In order for \(B\) to have nondegenerate elements, \(\ell\) must be odd, and for \(\ell = 2p + 1 \geq 5\), \(\mathfrak{h}^* \otimes B \cong \begin{array}{c}
\vdots \\
\vdots \\
1
\end{array} \oplus \begin{array}{c}
\vdots \\
\vdots \\
1
\end{array} \); thus the ALC holds for \(B\).
Case 5 \((A^{1,2}_\ell)\). For each \(\ell = 2p + 1 \geq 5\), there are two real forms. When \(g \cong \text{sl}(2p + 2, \mathbb{R})\), the geometries are the almost Grassmannian structures on manifolds \(M\) of dimension \(4p\), modelled on the Grassmannian of 2-planes in \(\mathbb{R}^{2p}\). The tangent bundle \(TM\) is identified with a tensor product \(E \otimes F\), where \(\text{rank} E = 2\), \(\text{rank} F = 2p\), and the nondegenerate metrics in \(B\) are tensor products of area forms on \(E\) and symplectic forms on \(F\). When \(g \cong \text{sl}(p, \mathbb{H})\), the geometries are almost quaternionic geometries, where \(TM\) is a quaternionic vector bundle, and the nondegenerate metrics in \(B\) are the (real parts of) quaternionic Hermitian forms.
When the cross is further from the ends of the diagram, we have \(S^2\mathfrak{h} = B \oplus B'\) with
\[
B \cong \begin{array}{c}
1 \ldots 1 \\
\ldots \\
1
\end{array} \quad B' \cong \begin{array}{c}
2 \ldots 2 \\
\ldots \\
2
\end{array}
\]
and there are too many components in both \(\mathfrak{h}^* \otimes B\) and \(\mathfrak{h}^* \otimes B'\) to satisfy the ALC.
We now turn to cases with two crossed nodes, related by the diagram automorphism of \(A_\ell\). First suppose the crossed nodes are the endpoints. In order to have nontrivial \(B\) we must have \(\ell \geq 3\), in which case \(S^2\mathfrak{h} = B \oplus B' \oplus B''\) where
\[
\mathfrak{h} \cong \begin{array}{c}
\vdots \\
\vdots \\
1
\end{array} \oplus \begin{array}{c}
\vdots \\
\vdots \\
1
\end{array} \cong \mathfrak{h}^* \\
B \cong \begin{array}{c}
2
\end{array} \text{ or } \begin{array}{c}
1
\end{array} \oplus \begin{array}{c}
1
\end{array} \cong \begin{array}{c}
1
\end{array} \quad \text{or } \begin{array}{c}
1
\end{array} \quad \begin{array}{c}
1
\end{array} \quad \begin{array}{c}
1
\end{array} \\
B' \cong \begin{array}{c}
2
\end{array} \oplus \begin{array}{c}
1
\end{array} \quad \begin{array}{c}
1
\end{array} \quad \begin{array}{c}
2
\end{array}
\]
and \(B''\) is trivial. Clearly \(\mathfrak{h}^* \otimes B'\) has too many irreducible components to satisfy the ALC, no matter which real form we consider.
It remains to consider \(B\), first in the case \(\ell = 3\), where the possible real forms (with \(\mathfrak{h}\) irreducible) are \(\text{su}(2, 2)\) and \(\text{su}(1, 3)\). Then
\[
\mathfrak{h}^* \otimes B \cong \begin{array}{c}
\vdots \\
\vdots \\
3
\end{array} \oplus \begin{array}{c}
\vdots \\
\vdots \\
3
\end{array} \oplus \begin{array}{c}
\vdots \\
\vdots \\
1
\end{array} \oplus \begin{array}{c}
\vdots \\
\vdots \\
1
\end{array} \cong \begin{array}{c}
\vdots \\
\vdots \\
3
\end{array} \oplus \begin{array}{c}
\vdots \\
\vdots \\
1
\end{array}
\]
and the ALC is satisfied, since these are complexifications of two complex components for the real form in question. However, for \( \ell \geq 4 \), we find that the product \( h^* \otimes B \) leads to complexifications with three complex components, so the ALC is not satisfied.
**Case 6** \((A_3^{2,1})\). Here \( g \) is \( \text{su}(2, 2) \) or \( \text{su}(1, 3) \), and \( M \) has a CR structure, i.e. a contact distribution \( \mathcal{H} \) equipped with a complex structure. The Levi form induces the class of trivial parallel Hermitian metrics (the Weyl connections corresponding to the contact forms leave parallel both the complex structure and the symplectic form, thus also the associated metric, and the metrizability problem is trivial as in the conformal case). However, we now see that there may also be interesting compatible subriemannian metrics on \( \mathcal{H} \leq TM \) which are Hermitian and tracefree with respect to the Levi form.
Now suppose the crosses are not placed at the ends, say the left one at the \( k \)th position, \( 2 \leq k \). Thus we consider the real forms \( \text{su}(p, q) \) with \( k \leq p \leq q \). We have
\[
\begin{align*}
h &\simeq \underbrace{1 \cdots 1}_{x \cdots x} \oplus \underbrace{1 \cdots 1}_{x \cdots x} \\
h^* &\simeq \underbrace{1 \cdots 1}_{x \cdots x} \oplus \underbrace{1 \cdots 1}_{x \cdots x}
\end{align*}
\]
for \( \ell > 2k \) and
\[
\begin{align*}
h &\simeq \underbrace{1 \cdots 1}_{x \cdots x} \oplus \underbrace{1 \cdots 1}_{x \cdots x} \\
h^* &\simeq \underbrace{1 \cdots 1}_{x \cdots x} \oplus \underbrace{1 \cdots 1}_{x \cdots x}
\end{align*}
\]
for \( \ell = 2k \). In particular, we have \( S^2 \mathcal{H} \supset B \) where
\[
B \simeq \underbrace{1 \cdots 1}_{x \cdots x} \oplus \underbrace{1 \cdots 1}_{x \cdots x}
\]
which admits nondegenerate metrics and satisfies the ALC, with
\[
\begin{align*}
h^* \otimes B &\simeq \left( \underbrace{1 \cdots 1}_{x \cdots x} \oplus \underbrace{1 \cdots 1}_{x \cdots x} \right) \\
&\quad \oplus \left( \underbrace{1 \cdots 1}_{x \cdots x} \oplus \underbrace{1 \cdots 1}_{x \cdots x} \right)
\end{align*}
\]
or
\[
\begin{align*}
h^* \otimes B &\simeq \left( \underbrace{1 \cdots 1}_{x \cdots x} \oplus \underbrace{1 \cdots 1}_{x \cdots x} \right) \\
&\quad \oplus \left( \underbrace{1 \cdots 1}_{x \cdots x} \oplus \underbrace{1 \cdots 1}_{x \cdots x} \right)
\end{align*}
\]
**Case 7** \((A_2^{2,k})\). Here \( g \simeq \text{su}(p, q) \) with nodes \( k \) and \( \ell + 1 - k \) crossed, where \( 2 \leq k \leq p \leq q \), \( p + q = \ell + 1 \). In these geometries, \( \mathcal{H} \cong E \otimes F \), where \( E \) is a complex vector bundle of rank \( k \), and the rank \((\ell - 2k + 1)\) complex vector bundle \( F \) comes with a Hermitian form of signature \((p - k, q - k)\). The corank of \( \mathcal{H} \leq TM \) is \( k^2 \), and the metrics on \( \mathcal{H} \) are the products of Hermitian metrics on \( E \) with the given ones on \( F \). When \( \ell = 2k \) (i.e. \( F \) has rank 1), \( g = \text{su}(k, k+1) \) with the nodes \( k, k+1 \) crossed. These are the free CR geometries with complex structure on \( \mathcal{H} \) studied in \([20]\) (where it is also explained how complex structure arises on \( \mathcal{H} \)).
The remaining components of $S^2\mathfrak{h}$ do not satisfy the ALC, except in special cases $k = 2$, $2k = \ell$ and $2k + 1 = \ell$. In particular, when $k = 2$,
$$B' \simeq \begin{array}{c}
\end{array}$$
satisfies the ALC (and is nontrivial for $\ell \geq 6$).
**Case 8** ($A_2^{2k}$). Here $\mathfrak{g} \simeq \text{su}(p, q)$ with nodes 2 and $\ell - 1$ crossed, where $2 \leq p \leq q$ and $\ell = p + q - 1 \geq 6$. In this geometry, $\mathcal{H} \cong E \otimes F$, where $E$ is a complex vector bundle of rank 2, and $F$ is a complex vector bundle of rank $\ell - 3$. The corank of $\mathcal{H} \subset TM$ is 4. The eligible metrics are the complex symmetric bilinear forms of the form of tensor product of two exterior forms.
When $2k = \ell$, we obtain $S^2\mathfrak{h} = B \oplus B'$ where
$$B' = \begin{array}{c}
\end{array}$$
which admits nondegenerate metrics, and satisfies the ALC, with
$$\mathfrak{h}^* \otimes B' \simeq \begin{array}{c}
\end{array}$$
**Case 9** ($A_2^{2k+1}$). This case is again the free CR geometry, with $\mathfrak{g} = \text{su}(k, k + 1)$, but the eligible metrics are the complex bilinear metrics on $\mathcal{H}$.
Similarly, when $\ell = 2k + 1$ with the $k$th and $(k + 2)$nd nodes crossed,
$$B' \simeq \begin{array}{c}
\end{array}$$
satisfies the ALC.
**Case 10** ($A_2^{2k+2}$). Here $\ell = 2k + 1$, $\mathfrak{g}$ is $\text{su}(k, k + 2)$, or $\text{su}(k + 1, k + 1)$, with nodes $k$ and $k + 2$ crossed. In this geometry, $\mathcal{H} \cong E \otimes F$, where $E$ is a complex vector bundle of rank $k$, and $F$ is a complex vector bundle of rank 2. The codimension of $\mathcal{H} \subset TM$ is $k^2$. The eligible metrics are the complex symmetric bilinear forms of the form of tensor product of two exterior forms.
We have now exhausted all possibilities, completing the proof in type A.
**4.2 Proof of Theorem 2 When $\mathfrak{g}$ Has Type $B_\ell$**
In the type $B$ case, there are no complex or quaternionic modules to consider, so the irreducible cases have one cross only. The unique grading of length one is odd dimensional conformal geometry. In dimension three we then have
$$\mathfrak{h}^* \simeq \begin{array}{c}
\end{array} \simeq \mathfrak{h} \quad S^2\mathfrak{h} \simeq \begin{array}{c}
\end{array}$$
Springer
The trivial representation in $S^2\mathfrak{h}$ corresponds to the trivial case of metrics in the conformal class, which are excluded from our classification, and choosing $B$ to be the other component leads to three components in $B \otimes \mathfrak{h}^*$, so the ALC fails. Similarly, for conformal geometries of dimensions $2\ell - 1 \geq 5$ we obtain
\[ \mathfrak{h}^* \simeq \frac{1}{6} \ldots \simeq \mathfrak{h} \quad S^2\mathfrak{h} \simeq \frac{2}{6} \ldots \oplus \frac{4}{6} \ldots . \]
As before, the trivial summand is excluded, and the other component fails the ALC.
We turn now to Lie contact geometries, with the second node crossed. For $B_3$,
\[ \mathfrak{h}^* \simeq \frac{1}{6} \ldots \frac{2}{6} \simeq \mathfrak{h} \quad S^2\mathfrak{h} = B \oplus B' \oplus B'' \simeq \frac{2}{6} \ldots \oplus \frac{2}{6} \ldots \oplus \frac{2}{6} \ldots . \]
Here, $B \otimes \mathfrak{h}^* = \frac{3}{6} \ldots \frac{2}{6} \oplus \frac{1}{6} \ldots \frac{2}{6}$ and satisfies the ALC. The other choices lead to too many components. For $B_\ell$ with $\ell \geq 4$, we have instead
\[ B \simeq \frac{2}{6} \ldots \mathfrak{h}^* \simeq \frac{1}{6} \ldots \simeq \mathfrak{h} \quad S^2\mathfrak{h} = B \oplus B' \oplus B'' \]
\[ B' \simeq \frac{2}{6} \ldots \mathfrak{h} \simeq \frac{1}{6} \ldots \simeq \mathfrak{h} \quad B'' \simeq \frac{2}{6} \ldots , \]
except that when $\ell = 4$, $B'' = \frac{2}{6} \ldots$. Now we check that $B' \otimes \mathfrak{h}^*$ has six components, $B'' \otimes \mathfrak{h}^*$ has three components, but the ALC is again satisfied by $B$. Lie contact geometries exist for $\mathfrak{g} = \mathfrak{so}(p, q)$ with $2 \leq p \leq q$; $\mathfrak{h}$ is the tensor product of defining representations $\mathbb{R}^2$ of $\mathfrak{sl}(2, \mathbb{R})$ and $\mathbb{R}^{p+q-4}$ of $\mathfrak{so}(p-2, q-2)$, and $B$ is the tensor product of a symmetric form on $\mathbb{R}^2$ and the defining inner product of signature $(p-2, q-2)$ on $\mathbb{R}^{p+q-4}$. See [5, §4.2.5] for more details on these geometries.
Next we consider $B_\ell$ with the cross on $k$th position, $3 \leq k \leq \ell - 1$; the outcome is quite similar to the Lie contact case. For $k \neq \ell - 1$, $S^2\mathfrak{h} = B \oplus B' \oplus B''$, where
\[ \mathfrak{h}^* \simeq \frac{1}{6} \ldots \mathfrak{h} \simeq \frac{1}{6} \ldots \simeq \mathfrak{h} \quad B \simeq \frac{2}{6} \ldots \mathfrak{h} \simeq \frac{1}{6} \ldots \simeq \mathfrak{h} \]
\[ B' \simeq \frac{2}{6} \ldots \mathfrak{h} \simeq \frac{1}{6} \ldots \simeq \mathfrak{h} \quad B'' \simeq \frac{1}{6} \ldots \oplus \frac{1}{6} \ldots \oplus \frac{1}{6} \ldots , \]
so $B$ satisfies the ALC, but $B'$ and $B''$ do not. If $k = \ell - 1$, $S^2\mathfrak{h} = B \oplus B' \oplus B''$ with
\[ \mathfrak{h}^* \simeq \frac{1}{6} \ldots \mathfrak{h} \simeq \frac{2}{6} \ldots \simeq \mathfrak{h} \quad B \simeq \frac{2}{6} \ldots \mathfrak{h} \simeq \frac{1}{6} \ldots \simeq \mathfrak{h} \]
\[ B' \simeq \frac{2}{6} \ldots \mathfrak{h} \simeq \frac{1}{6} \ldots \simeq \mathfrak{h} \quad B'' \simeq \frac{1}{6} \ldots \oplus \frac{1}{6} \ldots \oplus \frac{1}{6} \ldots , \]
and again, $B$ satisfies the ALC, but $B'$ and $B''$ do not. These $|2|$-graded geometries are modelled on the flag variety of isotropic $k$-planes and exist for the real forms $\mathfrak{so}(p, q)$ with $k \leq p \leq q$. We have $\mathfrak{h} \simeq \mathbb{R}^k \otimes \mathbb{R}^{p+q-k}$ and $B$ corresponds to the tensor product of a symmetric form on $\mathbb{R}^k$ with the defining inner product on $\mathbb{R}^{p+q-k}$.
**Case 11** ($B_\ell^{1,k}$). Here $\mathfrak{g} \simeq \mathfrak{so}(p, q)$ with $k \leq p \leq q$ and $p + q = 2\ell + 1$, and the geometries come equipped with the identification of the horizontal distribution $\mathcal{H} \subseteq TM$ with the tensor product $E \otimes F$, where $E$ has rank $k$ and $F$ carries a metric.
of signature \((p - k, q - k)\). The corank of \(H \leq TM\) is \(\frac{1}{2}k(k - 1)\). The metrics in \(B\) are the tensor products of symmetric nondegenerate forms on \(E\) and the given metric on \(F\).
Finally, we arrive at the cross at the very end. For \(B_{\ell}\) with \(\ell \geq 2\), we have
\[
\mathfrak{h}^* \simeq \frac{1}{\cdots} \quad \mathfrak{h} \simeq \frac{1}{\cdots} \quad B = S^2\mathfrak{h} \simeq \frac{2}{\cdots} \\
\mathfrak{h}^* \otimes B \simeq \frac{3}{\cdots} \oplus \frac{1}{\cdots} (\ell = 2) \quad \mathfrak{h}^* \otimes B \simeq \frac{2}{\cdots} \oplus \frac{1}{\cdots} (\ell \geq 3),
\]
and the ALC is satisfied.
**Case 12 \((B_{1, \ell}^{1, \ell})\).** Here \(g\) is the split form \(\mathfrak{so}(\ell, \ell + 1)\). The geometries are the well-known free distributions, cf. [9], with rank \(\ell\) horizontal distribution \(H \leq TM\) of corank \(\frac{1}{2}\ell(\ell - 1)\). The metrics in \(B\) are all nondegenerate metrics on \(H\).
### 4.3 Proof of Theorem 2 When \(g\) Has Type \(C_{\ell}^\ell\)
As with type \(B_{\ell}\), we only have to consider cases with a single crossed node. We begin with the first node crossed, corresponding to the well-known contact projective structures, with
\[
\mathfrak{h}^* \simeq \frac{1}{\cdots} \simeq \mathfrak{h};
\]
we have discussed the lowest dimension three already as the \(B_2\) case, which coincides with the free distribution of rank two. For \(\ell \geq 3\), the picture changes since
\[
B \otimes \mathfrak{h}^* \simeq \frac{3}{\cdots} \oplus \frac{2}{\cdots} \oplus \frac{1}{\cdots} \quad \mathfrak{h}^* \otimes B \simeq \frac{2}{\cdots} \oplus \frac{1}{\cdots} (\ell \geq 3),
\]
and thus the ALC fails.
Moving on to the second node, we obtain another well-known family of examples: the quaternionic contact geometries [for \(g \cong \mathfrak{sp}(p, \ell - p), 1 \leq p \leq \ell/2\)] or their split analogues (for \(g \cong \mathfrak{sp}(2\ell, \mathbb{R})\)]—see [5, §4.3.3]. For \(\ell = 3\), we have
\[
\mathfrak{h}^* \simeq \frac{1}{\cdots} \simeq \mathfrak{h} \quad S^2\mathfrak{h} = B' \oplus B'' \quad \text{with} \quad B' \simeq \frac{2}{\cdots};
\]
and \(B''\) trivial, while for \(\ell \geq 4\), we have
\[
\mathfrak{h}^* \simeq \frac{1}{\cdots} \simeq \mathfrak{h} \quad S^2\mathfrak{h} = B \oplus B' \oplus B''
\]
and \(B''\) trivial. Since \(\mathfrak{h}^* \otimes B'\) decomposes into four components, there are only non-trivial possibilities for \(\ell \geq 4\). For \(\ell = 4\),
\[
\mathfrak{h}^* \otimes B \simeq \frac{1}{\cdots} \oplus \frac{1}{\cdots} \oplus \frac{1}{\cdots} \oplus \frac{1}{\cdots}
\]
and so the ALC holds for $B$, but for $\ell \geq 5$, $\mathfrak{h}^* \otimes B$ has three irreducible components, and the ALC is not satisfied.
**Case 13** ($C^{1,2}_4$). Here the possible real Lie algebras are $\mathfrak{sp}(8, \mathbb{R})$, $\mathfrak{sp}(2, 2)$, or $\mathfrak{sp}(1, 3)$, with the second node crossed. In the first case, the geometries come equipped with the identification of the horizontal distribution $\mathcal{H} \leq TM$ with the tensor product $E \otimes F$, where $E$ is rank 2 and the rank 4 vector bundle $F$ comes with a symplectic form. The eligible metrics in $B$ are the tensor products of a area form on $E$ and the given symplectic form on $F$. In the quaternionic cases, $\mathcal{H}$ is quaternionic and the eligible metrics in $B$ are quaternionic Hermitian forms.
Let us next suppose that the $k$th node is crossed for $3 \leq k \leq \ell - 2$. Then
\[
\begin{align*}
\mathfrak{h}^* &\simeq \begin{array}{c}
1 \\
1
\end{array} \\
S^2\mathfrak{h} &\simeq B \oplus B' \oplus B'' \\
B' &\simeq \begin{array}{c}
2 \\
2
\end{array} \\
B'' &\simeq \begin{array}{c}
1 \\
1
\end{array}
\end{align*}
\]
and so $B$ satisfies the ALC, but the other components do not. The relevant metrics are again tensor products of an exterior form on the rank $k$ auxiliary bundle $E$ and the given symplectic form on $F$ (where the horizontal distribution is identified with $E \otimes F$). These geometries are available for the split form $\mathfrak{sp}(2\ell, \mathbb{R})$ and, if $k$ is even then also for the real forms $\mathfrak{sp}(p, q), k \leq p < q$.
The case with the cross at the last but one node is very similar. Here
\[
\begin{align*}
\mathfrak{h}^* &\simeq \begin{array}{c}
1 \\
1
\end{array} \\
S^2\mathfrak{h} &\simeq B \oplus B' \\
B' &\simeq \begin{array}{c}
2 \\
2
\end{array} \\
B'' &\simeq \begin{array}{c}
1 \\
1
\end{array}
\end{align*}
\]
and so $B$ satisfies the ALC (while $B$ does not).
**Case 14** ($C^{1,k}_\ell$). With the $k$th node crossed for $3 \leq k \leq \ell - 1$, the possible real Lie algebras are $\mathfrak{sp}(2n, \mathbb{R})$, and if $k$ is even, then also $\mathfrak{sp}(p, q), k \leq p < q$. In the split case, the horizontal distribution is a tensor product $\mathcal{H} \simeq E \otimes F$ with $E$ of rank $k$ and $F$ symplectic of rank $2\ell - 2k$, and the eligible metrics are tensor products of antisymmetric forms on $E$ and the given symplectic form on $F$. In the quaternionic cases, $\mathcal{H}$ comes with a quaternionic structure, and the eligible metrics are quaternionic Hermitian forms.
Finally, we consider the cross at the last node of $C_\ell$ with $\ell \geq 3$ ($\ell = 2$ is equivalent to the $B_2$ case with the first node crossed). In this case
\[
\begin{align*}
S^2\mathfrak{h} &\simeq B \oplus B' \\
B' &\simeq \begin{array}{c}
2 \\
2
\end{array} \\
B'' &\simeq \begin{array}{c}
1 \\
1
\end{array}
\end{align*}
\]
and both $B$ and $B'$ have too many components in their tensor products with $\mathfrak{h}^*$ to satisfy the ALC.
4.4 Proof of Theorem 2 When $\mathfrak{g}$ Has Type $D_\ell$
We first consider the cases with one cross on $D_\ell$, $\ell \geq 4$, starting with the first node, i.e. the even dimensional conformal geometries, where $S^2 \mathfrak{h} = B \oplus B'$ with
$$\mathfrak{h}^* \cong \mathfrak{h} \quad B \cong S^2 \mathfrak{h} = B \oplus B_1 \oplus B_2 \oplus B_3 \quad B' \cong \mathfrak{h}^* \cong B_1 \otimes \mathfrak{h}^* \cong B_1 \oplus B_2 \oplus B_3.$$
As in the odd dimensional case (type $B_\ell$), $B$ does not satisfy the ALC, and the trivial summand $B'$ yields metrics in the conformal class, which we exclude.
We turn now to the Lie contact case, with the second node crossed. For $\ell = 4$,
$$\mathfrak{h}^* \cong \mathfrak{h} \quad S^2 \mathfrak{h} = B \oplus B_1 \oplus B_2 \oplus B_3 \quad B_1 \otimes \mathfrak{h}^* \cong B_1 \oplus B_2 \oplus B_3.$$
While $\mathfrak{h}^* \otimes B$ has too many components, $B_1$ satisfies the ALC, as do $B_2$ and $B_3$ by symmetry. The metrics are tensor products of two area forms and a symmetric form on $\mathfrak{h} \cong \mathbb{R}^2 \otimes \mathbb{R}^2 \otimes \mathbb{R}^2$. The geometries exist for the real forms $\mathfrak{so}(4, 4)$, $\mathfrak{so}(3, 5)$ and the quaternionic $\mathfrak{so}^*(8) \cong \mathfrak{so}(2, 6)$. Similarly, for $\ell \geq 5$, we have
$$\mathfrak{h}^* \cong \mathfrak{h} \quad S^2 \mathfrak{h} = B \oplus B' \oplus B'' \quad B \cong \mathfrak{h}^* \otimes B \cong B'' \cong \mathfrak{h}^* \otimes B \cong \mathfrak{h}^* \otimes B$$
where $B$ satisfies the ALC, but $B'$ and $B''$ do not. In addition to the real forms $\mathfrak{so}(p, q)$, $2 \leq p \leq q$, $p + q = 2\ell$, which are analogous to the Lie contact geometries of type $B_\ell$, the real form $\mathfrak{so}^*(2\ell)$ is also possible.
As with type $B_\ell$, the cases where the $k$th node is crossed, with $3 \leq k \leq \ell - 2$, behave in a similar way. If $k \leq \ell - 3$ then
$$\mathfrak{h}^* \cong \mathfrak{h} \quad S^2 \mathfrak{h} = B \oplus B' \oplus B'' \quad B \cong \mathfrak{h}^* \otimes B \cong B'' \cong \mathfrak{h}^* \otimes B$$
so that $B$ satisfies the ALC, while $B'$ and $B''$ do not. The geometries exist for real forms $\mathfrak{so}(p, q)$, $k \leq p \leq q$, $p + q = 2\ell$, and if $k$ is even, then also for $\mathfrak{so}^*(2\ell)$. If $k = \ell - 2$, the computation differs slightly, but the outcome is similar.
where $B$ satisfies the ALC, but the other cases do not. The geometries exist for the real forms $\mathfrak{so}(\ell, \ell)$, $\mathfrak{so}(\ell - 1, \ell + 1)$, and if $\ell$ is even then also $\mathfrak{so}^*(2\ell)$.
**Case 15** ($D_\ell^{1,k}$). Here $\mathfrak{g} \simeq \mathfrak{so}(p, q)$ with $2 \leq k \leq p \leq q$ and $p + q = 2n$, the geometries come equipped with the identification of the horizontal distribution $\mathcal{H} \leq TM$ with the tensor product $E \otimes F$, where $E$ has rank $k$ and $F$ carries a metric of signature $(p - k, q - k)$. The corank of $\mathcal{H} \leq TM$ is $\frac{1}{2}k(k - 1)$. The metrics in $B$ are the tensor products of symmetric nondegenerate forms on $E$ and the given metric on $F$. When $\mathfrak{g} \simeq \mathfrak{so}^*(2n)$ and $k$ is even, the geometries come with the identification of the horizontal distribution $\mathcal{H}$ with the tensor product of a quaternionic rank $k$ bundle $E$ and a quaternionic rank $n - 2k$ bundle $F$ equipped with a quaternionic skew-Hermitian form. The metrics in $B$ are quaternionic Hermitian forms.
The remaining case with one cross is the so-called spinorial geometry with the cross on one of the nodes in the fork. The case of $D_4$ coincides with the 6-dimensional conformal Riemannian geometry. For $\ell \geq 5$, we have $S^2\mathfrak{h} = B \oplus B'$ with
$$
\mathfrak{h}^* \simeq 1_1 \quad \mathfrak{h} \simeq 1_1 \quad B' \simeq 2_1 \\
B \simeq 2_1 \quad B' \simeq 2_1 \\
B_1 \simeq 2_2 \quad B_2 \simeq 1_2 \quad B_3 \simeq 0_1
$$
Now $\mathfrak{h}^* \otimes B$ has three summands, as does $\mathfrak{h}^* \otimes B'$, except for $\ell = 5$, when
$$
\mathfrak{h}^* \otimes B' \simeq 1_1 \oplus 1_1
$$
Here, in the complex setting, $\mathfrak{h} \cong \wedge^2 \mathbb{C}^5$, and $B \cong \mathbb{C}^5 \cong \mathbb{C}^4 \otimes \mathbb{C}^5 \cong S^2 \wedge^2 \mathbb{C}^5$, where $\alpha \in B \cong \mathbb{C}^5$ determines a metric $g_\alpha$ on $\mathfrak{h}^* \cong \wedge^2 \mathbb{C}^5$ by $g_\alpha(\xi, \eta) = \alpha \wedge \xi \wedge \eta \in \wedge^5 \mathbb{C}^5 \cong \mathbb{C}$. Such a metric is never nondegenerate, so this case is excluded.
We next consider $D_\ell$ cases with two crossed nodes. For $\mathfrak{h}$ to be irreducible, the semisimple part of the Levi factor $p/p^\perp$ must be simple. Indeed, working in the complex setting, a direct check reveals that breaking the Dynkin diagram by two crosses into more than one part always leads to nonisomorphic representations for the two components of $\mathfrak{h}$. Furthermore, the only way to obtain isomorphic components is to take the two spinorial nodes of the $D_\ell$ diagram. The only real forms compatible with this geometry are the split form $\mathfrak{so}(\ell, \ell)$, the quasi-split form $\mathfrak{so}(\ell + 1, \ell - 1)$ and the quaternionic form $\mathfrak{so}^*(2\ell)$ with $\ell = 2p + 1$ odd. In the split case, $\mathfrak{h}$ is not irreducible, so this does not fit into our classification. In the quasi-split case, $\mathfrak{h} \cong \mathbb{R}^{\ell-1} \otimes \mathbb{R}$ is complex, while in the quaternionic case, $\mathfrak{h}$ is quaternionic. In $S^2\mathfrak{h} = B \oplus B' \oplus B''$, with
\[\mathfrak{h}^* \simeq 1_1 \quad \mathfrak{h} \simeq 1_1 \quad S^2\mathfrak{h} = B \oplus B_1 \oplus B_2 \oplus B_3\]
where we denote the nonzero weights over the crossed nodes for clarity. Observe that
\[ \mathfrak{h}^* \otimes B \simeq \begin{pmatrix} 2 & 1 & 2 \end{pmatrix} \oplus \begin{pmatrix} 1 \end{pmatrix}, \quad \mathfrak{h}^* \otimes B' \simeq \begin{pmatrix} 1 \end{pmatrix} \oplus \begin{pmatrix} 1 \end{pmatrix}, \quad \mathfrak{h}^* \otimes B'' \simeq \begin{pmatrix} 2 & 2 \end{pmatrix} \oplus \begin{pmatrix} 1 \end{pmatrix}, \]
so that both \( B \) and \( B' \) satisfy the ALC, but \( B'' \) does not (\( \mathfrak{h}^* \otimes B'' \) has eight components).
**Case 16** \((D^2,\mathfrak{s})\). When \( \mathfrak{g} = \mathfrak{s}\mathfrak{o}(\ell - 1, \ell + 1) \), the horizontal distribution \( \mathcal{H} \leq TM \) is a complex vector bundle of complex rank \( \ell - 1 \), and the metrics in \( B \) are complex bilinear. When \( \mathfrak{g} = \mathfrak{s}\mathfrak{o}^*(2\ell) \), with \( \ell = 2p + 1 \) odd, the horizontal distribution \( \mathcal{H} \leq TM \) has a quaternionic structure of quaternionic rank \( p \) and the metrics in \( B \) are quaternionic skew-Hermitian.
**Case 17** \((D^2,\mathfrak{h})\). This case involves the same geometries as in the previous case, with \( \ell = 2p + 1 \) odd, except that when \( \mathfrak{g} = \mathfrak{s}\mathfrak{o}(\ell - 1, \ell + 1) \), the metrics in \( B' \) are Hermitian, while for \( \mathfrak{g} = \mathfrak{s}\mathfrak{o}^*(2\ell) \), the metrics in \( B \) are quaternionic Hermitian.
### 4.5 Proof of Theorem 2 When \( \mathfrak{g} \) Has Exceptional Type
The first case we consider is the Lie algebra \( E_6 \). Let us consider possibilities for parabolic subalgebras with one crossed node. The first possibility is
\[ S^2\mathfrak{h} = B \oplus B' \quad B \simeq \begin{pmatrix} 1 \end{pmatrix}, \quad B' \simeq \begin{pmatrix} 1 \end{pmatrix}. \]
The product \( \mathfrak{h}^* \otimes B \) decomposes [as the product of a spinor and defining representation of \( \text{SO}(10) \)] into the sum of two \( P \)-modules, and hence the ALC is satisfied.
**Case 18** \((E^1,\mathfrak{s})\). This is the \(|1|\)-graded geometry for \( E_6 \) for which the allowed real forms are the split form \( E^0(6) \), or \( E^0(-26) \), and \( \mathcal{H} = TM \) carries the structure of basic spinor representation \( S^+ \) of \( \mathfrak{s}\mathfrak{o}(5, 5) \), or \( \mathfrak{s}\mathfrak{o}(1, 9) \), respectively. The \( P \)-module \( B \) corresponding to the eligible metrics is the defining representation of \( \mathfrak{s}\mathfrak{o}(5, 5) \) or \( \mathfrak{s}\mathfrak{o}(1, 9) \).
Consider next the adjoint variety, with the node on the short leg crossed. We have
\[ S^2\mathfrak{h} = B \oplus B' \quad B \simeq \begin{pmatrix} 2 \end{pmatrix}, \quad B' \simeq \begin{pmatrix} 1 \end{pmatrix}. \]
and find that both \( \mathfrak{h}^* \otimes B \) and \( \mathfrak{h}^* \otimes B' \) have four components.
In the remaining two cases with one crossed node, the semisimple part of $p/p^\perp$ is not simple, and it is easy to see that the ALC cannot be satisfied:
$$\mathfrak{h}^* \cong \begin{array}{c} 1 \\ 2 \end{array} \mathfrak{h} \cong \begin{array}{c} 1 \\ 2 \end{array}$$
$$S^2\mathfrak{h} \cong \begin{array}{c} 1 \\ 1 \\ 2 \end{array} \oplus \begin{array}{c} 2 \\ 1 \\ 1 \end{array}$$
$$\mathfrak{h}^* \cong \begin{array}{c} 1 \\ 2 \end{array} \mathfrak{h} \cong \begin{array}{c} 1 \\ 2 \end{array}$$
$$S^2\mathfrak{h} \cong \begin{array}{c} 1 \\ 1 \\ 2 \end{array} \oplus \begin{array}{c} 2 \\ 1 \\ 1 \end{array} \oplus \begin{array}{c} 2 \\ 1 \\ 1 \end{array}.$$
The only case with two crosses for which $\mathfrak{h}$ could be irreducible is
$$\mathfrak{h}^* \cong \mathfrak{h} \cong \begin{array}{c} 1 \\ 1 \\ 1 \\ 1 \end{array},$$
and indeed, $\mathfrak{h}$ is irreducible for the quasi-split real form $E_6(2)$. For this real form, the nontrivial irreducible summands in $S^2\mathfrak{h}$ are
$$B \cong \begin{array}{c} 1 \\ 1 \end{array} \times B' \cong \begin{array}{c} 1 \\ 1 \end{array} \times B'' \cong \begin{array}{c} 1 \\ 1 \end{array} \times \begin{array}{c} 1 \\ 2 \end{array}.$$
The products $\mathfrak{h}^* \otimes B'$ and $\mathfrak{h}^* \otimes B''$ have too many components but
$$\mathfrak{h}^* \otimes B \cong \begin{array}{c} 1 \\ 1 \\ 1 \\ 1 \end{array} \times \begin{array}{c} 1 \\ 1 \\ 2 \\ 1 \end{array} \times \begin{array}{c} 1 \\ 1 \\ 2 \\ 1 \end{array}$$
and so the ALC is satisfied.
**Case 19** ($E_6^{2,h}$). This is a $[2]$-graded geometry for the quasi-split Lie algebra $E_6(2)$. The horizontal distribution $\mathcal{H}$ carries the structure of the spinor representation $S$ of $\mathfrak{so}(3, 5)$, while the eligible metrics are induced by the defining representation of $\mathfrak{so}(3, 5)$.
For $E_7$ and its real forms, irreducibility of $\mathfrak{h}$ implies that only one node may be crossed, and a similar analysis to the $E_6$ type shows that the cases with the best chance to satisfy the ALC are those with cross over the first or last node, where
$$\mathfrak{h}^* \cong \begin{array}{c} 1 \\ 1 \\ 1 \end{array} \mathfrak{h} \cong \begin{array}{c} 1 \\ 2 \\ 1 \end{array}$$
or
$$\mathfrak{h}^* \cong \mathfrak{h} \cong \begin{array}{c} 1 \\ 2 \\ 1 \end{array}.$$
$$S^2\mathfrak{h} \cong \begin{array}{c} 2 \\ 2 \end{array} \oplus \begin{array}{c} 1 \\ 1 \end{array}$$
It is easy to see that none of these cases satisfy the ALC.
Similarly, for $E_8$, even the most promising candidates
$$\mathfrak{h}^* \cong \mathfrak{h} \cong \begin{array}{c} 1 \\ 2 \\ 2 \end{array}$$
and
$$\mathfrak{h}^* \cong \begin{array}{c} 1 \\ 1 \\ 1 \\ 1 \end{array} \mathfrak{h} \cong \begin{array}{c} 1 \\ 1 \\ 1 \\ 1 \end{array}.$$
fail the ALC. Again there can be no cases with more than one cross.
For $F_4$, the only nonsplit possibility is
$$h^* \simeq h \simeq \begin{array}{l}
\end{array} \quad S^2h = B \oplus B' \quad \text{where} \quad B \simeq \begin{array}{l}
\end{array}$$
and $B'$ is trivial. However, $B$ does not satisfy the ALC.
For the split form, all cases can have only one crossed node. When
$$h^* \simeq \begin{array}{l}
\end{array} \quad S^2h = B \oplus B' \quad B \simeq \begin{array}{l}
\end{array} \quad B' \simeq \begin{array}{l}
\end{array},$$
the elements of $B$ are all degenerate, whereas $h^* \otimes B'$ does not satisfy the ALC.
In the remaining two possibilities for the crossed node,
$$h^* \simeq \begin{array}{l}
\end{array} \quad S^2h \simeq \begin{array}{l}
\end{array} \quad h \simeq \begin{array}{l}
\end{array}$$
and
$$h^* \simeq h \simeq \begin{array}{l}
\end{array} \quad S^2h \simeq \begin{array}{l}
\end{array},$$
the ALC fails in all cases.
Finally, for $G_2$, only the split case is possible, with one crossed node.
$$h^* \simeq \begin{array}{l}
\end{array} \quad S^2h \simeq \begin{array}{l}
\end{array} \quad h \simeq \begin{array}{l}
\end{array}$$
and
$$h^* \simeq h \simeq \begin{array}{l}
\end{array} \quad S^2h = B \simeq \begin{array}{l}
\end{array}$$
and only the last of these satisfies the ALC, with
$$h^* \otimes B \simeq \begin{array}{l}
\end{array}.$$
**Case 20** ($G_{2}^{1,1}$). The real Lie algebra is the split form of $G_2$ and the geometry is given by Cartan’s famous $(2, 3, 5)$ distribution. Hence the horizontal distribution has rank 2 and the $P$-module $B$ corresponding to the eligible metrics is the second symmetric power of the defining representations of $\mathfrak{sl}(2, \mathbb{R})$.
### 5 Examples of Reducible Cases
We now discuss a few cases of geometries with reducible $\mathcal{H}$, where the linearized metrizability procedure works. Actually, we have seen several such examples already, when dealing with real forms with irreducible, but not absolutely irreducible $\mathfrak{h}$ in Theorem 2. We list some of those with irreducible $B$ in the following result.
**Theorem 3** The following real parabolic geometries with the Lie algebra $\mathfrak{g}$ and choice of $B$ satisfy the ALC and the linearized metrizability procedure works.
(i) $B \simeq \begin{array}{l}
\end{array}$, $\mathfrak{g} \simeq \mathfrak{sl}(4, \mathbb{R})$. These are Lagrangian contact structures in dimension 5, where a decomposition $\mathcal{H} = E \oplus F$ of the contact subbundle into a direct
of two Lagrangian subbundles is given. The metrics in \( B \) are the split signature metrics with both \( E \) and \( F \) isotropic.
(ii) \( B \cong \begin{array}{cccccc}
0 & 1 & 0 & -1 & 0 & -5 \cr
& & & & & 2 \cr
\end{array} \oplus \begin{array}{cccccc}
0 & 1 & \cdots & 1 \cr
& & & & & \cdots \cr
\end{array} \), \( g \cong \mathfrak{sl}(n+1, \mathbb{R}) \), \( n \) even, the first cross at the \( k \)th root \((2k < n)\), crosses at symmetric places. These geometries come with \( \mathcal{H} \) identified with the sum of two vector bundles of the form \((E \otimes F^*) \oplus (F \otimes G^*)\), where \( E \) and \( G \) are real vector bundles of rank \( k \), and \( F \) is a real vector bundle of rank \( n-2k+1 \). The metrics are the split signature ones, in the subbundle \( E \otimes G^* \leq E \otimes F^* \otimes F \otimes G^* \).
(iii) \( B \cong \begin{array}{cccccc}
0 & 1 & \cdots & 1 \cr
& & & & & \cdots \cr
\end{array} \), \( g \cong \mathfrak{so}(p, p) \), \( 2p = n \). The horizontal distribution \( \mathcal{H} \leq TM \) is the sum of two rank \( p-1 \) bundles \( E \) and \( F \) coming from the defining representations of \( \mathfrak{sl}(p-1, \mathbb{R}) \) with different weights, and \( B \) stays for general split metrics on \( E \oplus F \).
(iv) \( B \cong \begin{array}{cccccc}
0 & 1 & \cdots & 1 \cr
& & & & & \cdots \cr
\end{array} \), \( g \cong \mathfrak{sl}(n, \mathbb{R}) \), \( n \) even. The geometry is \([2]-\)graded, and the horizontal subspace \( \mathcal{H} \leq TM \) corresponds to the direct sum of two of the three isomorphic defining representations of \( \mathfrak{so}(4, 4) \). The eligible metrics are the generic tracefree split ones and the \( P \)-module \( B \) corresponds to the third defining representation \( \mathbb{R}^8 \), up to the weight.
**Proof** All cases were already treated for different real forms in the previous section, except for the very last case. The computation presented there showed that the ALC is satisfied but the subbundle \( \mathcal{H} \) is not irreducible, but a sum of two subbundles. At the same time, the strong ALC holds, and thus the linearized metrizability procedure works as required. \( \square \)
Our final result illustrates the possibility of finding examples with reducible \( B \), including one in which a trivial one-dimensional component occurs.
**Theorem 4** The following real parabolic geometries with the Lie algebra \( g \) and choice of \( B \) satisfy the ALC.
(i) \( B \cong \begin{array}{cccccc}
0 & 1 & 0 & -1 & 0 & -5 \cr
& & & & & 2 \cr
\end{array} \oplus \begin{array}{cccccc}
0 & 1 & \cdots & 1 \cr
& & & & & \cdots \cr
\end{array} \), the real Lie algebra is the split form of type \( F_4 \) (\([6]-\)graded). The horizontal distribution \( \mathcal{H} \leq TM \) is built of two rank 2 bundles \( E \) and \( F \) coming from the defining representations of \( \mathfrak{sl}(2, \mathbb{R}) \) with different weights in \( \mathfrak{p}_0 \). The first component \( \mathcal{H}_1 \) is a tensor product \( E \otimes F \) with appropriate weight, while \( F \) stays for the other component \( \mathcal{H}_2 \) with another weight. The eligible metrics are the sums of the metrics in \( \Lambda^2 E \otimes \Lambda^2 F \leq \mathcal{S}^2 \mathcal{H}_1 \), and the metrics in \( \mathcal{S}^2 \mathcal{H}_2 \).
(ii) \( B \cong \begin{array}{cccccc}
0 & 1 & \cdots & 1 \cr
& & & & & \cdots \cr
\end{array} \oplus \begin{array}{cccccc}
0 & 1 & \cdots & 1 \cr
& & & & & \cdots \cr
\end{array} \). In this case \( g \cong \mathfrak{sl}(\ell+1) \), \( 5 \leq \ell \), with nodes 2 and \( \ell-1 \) crossed, and \( \mathcal{H} \cong E \otimes F^* \oplus F \otimes G \), where \( E \) is a real vector bundle of rank 2, and \( F \) is a real vector bundle of rank \( \ell-3 \). The corank of \( \mathcal{H} \leq TM \) is 4. The eligible metrics are sums of the symmetric bilinear forms on \( \mathcal{H}_1 \) and \( \mathcal{H}_2 \), both of the form of tensor product of two exterior forms.
(iii) \( B \cong \begin{array}{cccccc}
0 & 1 & \cdots & 1 \cr
& & & & & \cdots \cr
\end{array} \oplus \begin{array}{cccccc}
0 & 1 & \cdots & 1 \cr
& & & & & \cdots \cr
\end{array} \). Similarly to the previous case, \( g \cong \mathfrak{sl}(2k) \), \( 4 \leq k \), with nodes crossed at symmetric positions, and \( \mathcal{H} \cong E \otimes F^* \oplus F \otimes G \), where \( E \) and \( G \) are real vector bundles of rank \( k-1 \), while \( F \) is a real vector bundle of rank 2. The corank of \( \mathcal{H} \leq TM \) is
The eligible metrics are sums of the symmetric bilinear forms on $\mathcal{H}_1$ and $\mathcal{H}_2$, both of the form of tensor product of two exterior forms.
(iv) $B \simeq \begin{array}{c}
\cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \\
\oplus \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots \cdots
\end{array}$, Here $\mathfrak{g} = \mathfrak{sl}(2k + 1)$, the horizontal distribution is the sum of two vector bundles of the same rank $k$, corresponding to the defining representations of the two semisimple components in $\mathfrak{p}_0$. The metrics are sums of metrics on these two parts of $\mathcal{H}$.
Proof (i) Since the strong ALC cannot hold in the case of split forms of the algebras, we work with the complete weights. The form of $\mathfrak{h}$ is seen from the Cartan matrix of type $F$, while the sum and difference of the second and last lines in the inverse Cartan matrix (which corresponds to the crossed nodes in the Dynkin diagram) provide the coefficients $(48116)$ and $(2452)$ expressing two generating weights in the centre of $\mathfrak{p}_0$. With their help, we find
$$\mathfrak{h}^* = \begin{array}{c}
\begin{array}{c}
1 \\
-2 \\
1 \\
0 \\
0 \\
1 \\
-2
\end{array}
\end{array}, \quad \mathfrak{h} = \begin{array}{c}
\begin{array}{c}
1 \\
-1 \\
-1 \\
0 \\
-2 \\
1 \\
1
\end{array}
\end{array}.$$
The part of interest in $S^2\mathfrak{h}$ is
$$B_1 \oplus B_2 = \begin{array}{c}
\begin{array}{c}
0 \\
1 \\
-1 \\
0 \\
-5 \\
2 \\
2
\end{array}
\end{array}.$$
Now, $B_1$ is trivial, while
$$B_2 \otimes \mathfrak{h}^* \simeq \begin{array}{c}
\begin{array}{c}
1 \\
3 \\
1 \\
3
\end{array}
\end{array} \oplus \begin{array}{c}
\begin{array}{c}
1 \\
1 \\
3 \\
1
\end{array}
\end{array} \oplus \begin{array}{c}
\begin{array}{c}
1 \\
1 \\
3 \\
1
\end{array}
\end{array} \oplus \begin{array}{c}
\begin{array}{c}
1 \\
1 \\
3 \\
1
\end{array}
\end{array} \oplus \begin{array}{c}
\begin{array}{c}
1 \\
1 \\
3 \\
1
\end{array}
\end{array}.$$
Hence the kernel of $b$ does not exceed the allowed number of components and the ALC holds. Finally,
$$\Lambda^4 \mathfrak{h}_1 = \begin{array}{c}
\begin{array}{c}
0 \\
2 \\
0 \\
2 \\
0 \\
-2 \\
-2
\end{array}
\end{array}, \quad \Lambda^2 \mathfrak{h}_2 = \begin{array}{c}
\begin{array}{c}
0 \\
-2 \\
0 \\
-3
\end{array}
\end{array}.$$
so that the weight of $L$ can be expressed in terms of them and thus the linearized metrizability procedure can be completed.
(ii)–(iv) All the other cases have been already discussed as the complex versions of some cases in the previous section. The only remaining bit of the proof is the check that the top exterior forms on the individual components provide linearly independent weights and thus may be used to rescale the metrics properly. This can be done exactly as in case (i).
Remark 5.1 Actually, the arguments in the cases (iii) and (iv) above work also in any of the situations where the crosses are either apart by one or next to each other, i.e. without assuming they are placed symmetrically, except if the adjacent crosses appear right at the ends of the diagram. In the latter case of the so-called paths geometries, one of the top degree forms on $\mathfrak{h}_i$ has trivial weight zero and thus the linearized metrizability procedure fails at the stage when we change the weight of the solutions in order to get genuine metrics.
Acknowledgements The authors thank the Czech Grant Agency, Grant Nr. P201/12/G028, for financial support.
References
1. Bryant, R.L., Dunajski, M., Eastwood, M.: Metrisability of two-dimensional projective structures. J. Differ. Geom. 83, 465–499 (2009)
2. Calderbank, D.M.J., Diemer, T.: Differential invariants and curved Bernstein-Gelfand-Gelfand sequences. J. reine angew. Math. 537, 67–103 (2001)
3. Calderbank, D.M.J., Eastwood, M., Matveev, V.S., Neusser, K.: C-projective geometry. Mem. Am. Math. Soc. (to appear). arXiv:1512.04516
4. Čap, A., Schichl, H.: Parabolic geometries and canonical Cartan connections. Hokkaido Math. J. 29, 453–505 (2000)
5. Čap, A., Slovák, J.: Parabolic Geometries I, Background and General Theory. Mathematical Surveys and Monographs, vol 154. American Mathematical Society, Providence (2009)
6. Čap, A., Slovák, J., Souček, V.: Bernstein–Gelfand–Gelfand sequences. Ann. Math. (2) 154(1), 97–113 (2001)
7. Čap, A., Gover, A.R., Hammerl, M.: Normal BGG solutions and polynomials. Int. J. Math. 23, 1250117 (2012)
8. Domashev, V.V., Mikeš, J.: On the theory of holomorphically projective mappings of Kählerian spaces. Math. Notes 23, 160–163 (1978) (translation from Mat. Zametki 23 (1978), 297–304)
9. Doubrov, B., Slovák, J.: Inclusions of parabolic geometries. Pure Appl. Math. Q. 6, 755–780 (2010)
10. Eastwood, M., Matveev, V.: Metric connections in projective differential geometry. In: Eastwood, M., Miller, W. (eds.) Symmetries and Overdetermined Systems of Partial Differential Equations. IMA Volumes in Mathematics and its Applications, vol. 144, pp. 339–350. Springer, New York (2008)
11. Fegan, H.D.: Conformally invariant first order differential operators. Q. J. Math. Oxf. 27, 371–378 (1976)
12. Frost, G.E.: The Projective Parabolic Geometry of Riemannian, Kähler and Quaternion-Kähler Metrics. PhD Thesis, University of Bath (2016). arXiv:1605.04406
13. Gauduchon, P.: Structures de Weyl et théorèmes d’annulation sur une variété conforme autoduale. Ann. Sc. Norm. Sup. Pisa 18, 563–629 (1991)
14. Hrdina, J.: Almost complex projective structures and their morphisms. Arch. Math. 45, 255–264 (2009)
15. Hrdina, J., Slovák, J.: Generalized planar curves and quaternionic geometry. Ann. Glob. Anal. Geom. 29, 349–360 (2006)
16. Jerison, D., Lee, J.M.: The Yamabe problem on CR manifolds. J. Differ. Geom. 25, 167–197 (1987)
17. Liouville, R.: Sur les invariants de certaines équations différentielles et sur leurs applications. J.l’Ecole Polytech. 59, 7–76 (1889)
18. Montgomery, R.: A Tour of Subriemannian Geometries. Their Geodesics and Applications, Mathematical Surveys and Monographs, vol. 91. American Mathematical Society, Providence (2006)
19. Půček, R.: Applications of invariant operators in real parabolic geometries. MSc Thesis, Prague (2016)
20. Schmalz, G., Slovák, J.: Free CR distributions. Cent. Eur. J. Math. 10, 1896–1913 (2012)
21. Sinjukov, N.S.: Geodesic Mappings of Riemannian Spaces. “Nauka”, Moscow (1979) (in Russian)
22. Slovák, J., Souček, V.: First order invariant differential operators for parabolic geometries. In: Proceedings of Conference Analyse harmonique et analyse sur les varietes, 1999, CIRM, Luminy, in Seminaires et Congres, vol 4, pp 249–274. Society of Mathematics France (2000)
23. Yoshimatsu, Y.: H-projective connections and H-projective transformations. Osaka J. Math. 15, 435–459 (1978)
Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
1305,
1
],
[
1305,
4741,
2
],
[
4741,
8378,
3
],
[
8378,
12292,
4
],
[
12292,
15244,
5
],
[
15244,
18892,
6
],
[
18892,
22468,
7
],
[
22468,
26334,
8
],
[
26334,
30401,
9
],
[
30401,
34431,
10
],
[
34431,
38214,
11
],
[
38214,
42217,
12
],
[
42217,
45122,
13
],
[
45122,
47897,
14
],
[
47897,
50444,
15
],
[
50444,
51509,
16
],
[
51509,
55750,
17
],
[
55750,
59355,
18
],
[
59355,
63502,
19
],
[
63502,
66796,
20
],
[
66796,
69097,
21
],
[
69097,
73002,
22
],
[
73002,
75635,
23
],
[
75635,
78666,
24
],
[
78666,
81082,
25
],
[
81082,
84432,
26
],
[
84432,
87335,
27
],
[
87335,
90124,
28
],
[
90124,
92678,
29
],
[
92678,
97237,
30
],
[
97237,
100617,
31
],
[
100617,
104143,
32
]
]
} |
636abf693f9016209bdda21b08d6c989ef752f47 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 12,
"total-fallback-pages": 0,
"total-input-tokens": 40473,
"total-output-tokens": 12908,
"fieldofstudy": [
"Medicine"
]
} | Studying Enhanced Recovery After Surgery (ERAS®) Core Items in Colorectal Surgery: A Causal Model with Latent Variables
Marco Gemma1 · Fulvia Pennoni2 · Marco Braga3
Accepted: 9 December 2020 / Published online: 11 February 2021
© The Author(s) 2021
Abstract
Background Previous Enhanced Recovery After Surgery (ERAS®) studies have not always taken into account that ERAS interventions depend on baseline covariates and that several confounding variables affect the composite outcomes.
Method A causal latent variable model is proposed to analyze data obtained prospectively concerning 1261 patients undergoing elective colorectal surgery within the ERAS protocol. Primary outcomes (composite of any complication, surgical site infection, medical complications, early ready for discharge (TRD), early actual discharge) and secondary outcomes (composite of late bowel function recovery, IV fluid resumption, nasogastric tube replacement, postoperative nausea and vomiting, re-intervention, re-admission, death) are considered along with their multiple dimensions.
Results Concerning the primary outcomes, our results evidence three subpopulations of patients: one with probable good outcome, one with possibly prolonged TRD and discharge without complications, and the other one with probable complications and prolonged TRD and discharge. Epidural anesthesia, waiving surgical drainage, and early ambulation, IV fluid stop and urinary catheter removal act favorably, while preoperative hospital stay and blood transfusion act negatively. Concerning the secondary outcomes our results evidence two subpopulations of patients: one with high probability of good outcome and one with high probability of complications. Epidural anesthesia, waiving surgical drainage, early ambulation and IV fluid stop act favorably, while blood transfusion acts negatively also with respect to these secondary outcomes.
Conclusion The multivariate causal latent class two-parameter logistic model, a modern statistical method overcoming drawbacks of traditional models to estimate the average causal effects on the treated, allows us to disentangle subpopulations of patients and to evaluate ERAS interventions.
Introduction
The ERAS (Enhanced Recovery After Surgery) is a multimodal perioperative care pathway intended to improve and shorten recovery after major surgery through the application of a bundle of interventions [1, 2]. However implementing all of the ERAS items is a hard work for any hospital and it is possible that some ERAS interventions exert a greater impact on outcome than others. Although the final goal is to realize a complete ERAS pathway, concentrating on some possible “core items” in the
Fulvia Pennoni
[email protected]
Marco Gemma
[email protected]
1 Anesthesia and Intensive Care, Fatebenefratelli Hospital, Piazza Principessa Clotilde, 3, 20121 Milan, Italy
2 Department of Statistics and Quantitative Methods, University of Milano-Bicocca, Via degli Arcimboldi, 8, 20126 Milan, Italy
3 Department of Surgery, University of Milano-Bicocca, San Gerardo Hospital, Via Pergolesi G. B., 33, 20900 Monza, Italy
beginning could be prominently facilitating. Much interest is growing about the search for evidence on these core items [3].
Any study on this topic exploited the incomplete compliance with ERAS items, which varyably accompany ERAS databases and provides the necessary variability for addressing the question of the benefit of a single item or of ERAS as a whole. Nevertheless, previous analysis did not adequately consider that non-compliance is scarcely ever independent from other important variables. This raises at least three major methodological issues that have been poorly addressed in previous studies. First, ERAS outcomes and ERAS interventions themselves are affected by several confounding variables. For example, the American Society of Anesthesia score (ASA) status can affect both the outcome and the early postoperative mobilization. Second, ERAS items are themselves inter-related. For example, conservative intraoperative fluid administration is conceivably more applicable in patients who did not receive preoperative bowel preparation. Third, when dealing with ERAS, the outcome measures are composite. For example, the principal outcome measures in ERAS studies, postoperative length of hospital stay and complications, cannot be considered separately since they measure a similar trait.
We retrospectively studied a prospectively collected data of patients undergoing elective colorectal surgery between 2014 and 2018 with an ERAS protocol in 20 Italian hospitals affiliated with the PeriOperative Italian Society (POIS). The aim of our study is to analyze the effects of a number of ERAS items with a statistical model that considers the aforementioned methodological issues, namely a multivariate Latent Class Two-Parameter Logistic (LC-2PL) model [4, 5] formulated within a potential outcome framework [6]. We estimate the Average Causal Effects on the Treated (ATET) adequately weighting each patient through the Inverse-Probability-of-Treatment (IPT, [7]). As recently remarked by [8] propensity score methods allow us to block the association between observed confounding variables and treatments, thus permitting to reduce bias due to pre-treatment imbalances in observational studies. Innovatively we propose to estimate patient’s weights according to the pre-treatment covariates and sequential blocks of treatments. We summarize the responses by means of a multivariate latent variable model suitable to classify patients and to assess the effects of the ERAS interventions on primary and secondary outcomes.
### Materials and methods
#### Study design
Twenty Italian hospitals affiliated with POIS collaborated in collecting data. All centers treated their patients within an ERAS pathway, which was defined with active contribution from the ERAS Society. Before the start of the study, all hospitals had been involved in a pathway implementation program led by POIS consisting in education and audit meetings every six months for a two-year period.
All data were collected prospectively through a standardized electronic spreadsheet, which was used to record 90 variables per patient [9]. Every three months, the center-specific spreadsheets containing data collected in that time period were merged into a web-based password-protected database managed by POIS. Data collected included demographics, patient comorbidities, preoperative and intraoperative variables, adherence to ERAS items, early recovery variables, and short-term postoperative outcomes.
Figure 1 shows the conceptual framework used to define sequentially the treatments, the confounders, the potential outcomes, the latent variables and the observed outcomes. We study 18 items out of the POIS database. For most of these treatments ERAS recommendations are available. Since our purpose is to account for the aforementioned inter-relationship between these treatments, we grouped them according to the phase of the patient’s pathway in which they are applied. Hence, treatments are classified in three consecutive units, namely preoperative, intraoperative, and postoperative, and each unit affects the ones to follow. Four treatments are considered as preoperative, namely preoperative hospital stay (number of days), no bowel preparation, glucidic drink administration, and premedication. Six are considered as intraoperative, namely IV fluid administration (ml/kg/h), epidural anesthesia, antibiotic prophylaxis, maintenance of normothermia, nausea and vomiting (PONV) prophylaxis, and no surgical drainage. Eight treatments are considered as postoperative, namely intravenous fluid administration (ml/kg during POD 1), morphine administration (dichotomous no/yes variable without differentiating between different administration modes—PCA, elastomer, fixed-dose—or different dosing), thromboembolism prophylaxis, prokinetic administration, naso-gastric tube (NGT) removal within POD 0, intravenous fluid stop within POD 2, urinary catheter removal within POD 1, and ambulation within POD 1. The cut-off POD choice for the last four variables has been decided according to the clinical experience.
There are 15 observed confounders. Seven of these potentially affect compliance with all of the treatments:
year of surgery, age, gender, ASA classification of general health status (ASA 1–2 vs ASA > 2), baseline blood hemoglobin (mg/dl), preoperative body weight loss, and preoperative diagnosis of diabetes mellitus. Five confounders potentially affect compliance only with postoperative treatments: intraoperative fluid losses (ml/kg), length of surgery (min), maximum postoperative pain on POD 1–4 [measured on the Numerical Rating Scale (NRS)], laparotomy (vs laparoscopy), and surgical stoma (coded as binary variables). We also consider the following external variables that directly affect the outcomes: type of surgery (colonic or rectal), malignancy of the underlying disease, blood derivatives transfusion.
The main advantage of the proposed causal latent class model is that we jointly account for several outcomes that are distinguished as primary and secondary according to their relative clinical importance in the ERAS framework. The primary outcomes are the following: occurrence of any complication, occurrence of surgical site infection (SSI), occurrence of medical complications unrelated to the surgical site, such as cardiovascular, pulmonary, thromboembolic, or urinary complications, ready for discharge after POD 5, and actual discharge from hospital after POD 6 [10]. Within our proposal, as illustrated in the next section, we are able to consider the outcomes jointly and to account for the fact that they mainly concern two dimensions: the first is made by complications and the second one is made by Time Ready for Discharge (TRD) and actual discharge.
The secondary outcomes that we account jointly are the following: bowel function recovery after POD 1, need for IV fluid resumption after suspension, need for nasogastric tube (NGT) replacement after removal, occurrence of postoperative nausea and vomiting (PONV), surgical re-intervention, hospital re-admission, death within POD 30. In the multivariate model presented below we assume they represent two distinct dimensions: the first made by bowel function, need for IV fluid resumption, need for nasogastric tube, nausea and vomiting and the second made by surgical re-intervention, hospital re-admission and death.
Statistical model
We highlight a proposal for the estimation of multiple treatments within observational studies, since we are interested to disentangle the effects of the ERAS items on primary and secondary outcomes. The treatment is confounded with the patient’s characteristics and to address this problem we follow the potential outcome approach to causal inference as proposed by [11, 12] and we extend the IPT weighted estimator [13, 14] to a multivariate LC-2PL model [15]. The estimation of the probability to be treated blocks of treatments (see Fig. 1).
Differently from previous proposals we suppose the counterfactual outcomes as latent variables denoted as $U_i^{(0)}$ and $U_i^{(1)}$ indicating for each patient $i, i = 1, \ldots, n$, the variable under the non-treated and treated status, respectively. An underlying latent variable $U_i$ is assumed to depend on both $U_i^{(0)}$ and $U_i^{(1)}$ as well as on the treatments (see Fig. 1). We assume local independence [5, 16] meaning that the observed outcomes are conditionally independent given the potential latent variables and the treatments administered sequentially.
In what follows, first we describe the estimation of the probability of treatment exposure given the pre-treatment covariates and each block of ERAS items (according to the arrows depicted in Fig. 1). This estimation is made by a sequence of linear or logistic regressions. Next, we calculate weights for each patient as the inverse of the probability of the observed treatment sequence. Third, we use stabilized weights to estimate a weighted causal LC-2PL model on the primary and secondary outcomes. The estimation is carried out through the maximization of a weighted log-likelihood employing the Expectation–Maximization algorithm (EM [17]). We rely on the Bayesian Information Criterion [BIC 18] through which we choose the suitable number of latent components. Fourth, we estimate the selected model by adding the covariates through a convenient parameterization considering the first latent class as reference since it identifies the subpopulation of patients recovered as expected.
The latent class model [4, 5] first proposed by [19] to classify units within a probabilistic approach is formulated as a finite mixture model [20]. Following some recent proposals in the literature, we formulate a multivariate LC-2PL model [21] to infer causal effects within observational studies [15, 22, 23]. We introduce a novel use of this model by attempting to estimate the effects of various sequential interventions on patients entered in the ERAS project. We model the marginal distribution of the counterfactual variables [6, 24, 25] and as a result the estimated regression coefficient encode the magnitude of the ATET [8]. In this way we mimic an artificial random assignment scenario essential to account for differences among patients. As pointed out by [7] the use of the Inverse-Probability-of-Treatment Weighting [IPTW 26] allow us to disentangle the association between the observed confounding variables and treatments thus permitting to reduce bias due to confounding.
A causal latent variable model
The potential outcomes of the patient are usually referred to as $Y_i^{(1)}$ if the patient is exposed to treatment $Z_i$ and as $Y_i^{(0)}$ if the patient is not exposed with $i, i = 1, \ldots, n$. The treatment effect is given by the difference $Y_i^{(1)} - Y_i^{(0)}$ and the expected value of this difference over the entire population of treated patients is defined as the ATET. We instead postulate the existence of the underlying latent potential variables denoted as $U_i^{(z)}$ and we define the ATET as
$$\text{ATET} = E\left(U_i^{(1)} - U_i^{(0)}|Z_i = 1\right),$$
for $i, i = 1, \ldots, n$, and of the latent variable $U_i$ depending on the treatment through the latent potential variables as follows
$$U_i = (1 - Z_i)U_i^{(0)} + Z_iU_i^{(1)}.$$
For this latent variable we assume a discrete distribution left unspecified with a finite number of support points ranging from 1 to $k$.
Let $Y_{ir}$ be the observed binary response referred to outcome $r, r = 1, \ldots, p$ for each patient $i, i = 1, \ldots, n$, randomly drawn from the population. Following [27, 28] we extend the proposal of [15] to estimate patient specific weights. We formulate the following assumptions: $i)$ conditional exchangeability meaning that the latent potential outcomes are independent on the treatment given the covariates, $ii)$ positivity (ignorable treatment assignment) meaning that there is a positive probability for every patient of receiving any type of treatment, $iii)$ consistency implying that the latent potential variables are well-defined and as a result any observed outcome is the potential variable corresponding to the observed treatment sequence.
Let $x_i$ denotes the covariates for patient $i$ observed prior to the treatment assignment, the weight for this patient corresponds to the inverse of the conditional probability of receiving the treatment. In the case of a binary treatment we use the following logit model to estimate this probability.
\[
p_i = \log \frac{P(Z_i = 1|x_i)}{P(Z_i = 0|x_i)} = \alpha + x'_i \gamma',
\]
where \(\alpha\) and \(\gamma'\) denotes the intercept and the vector of regression coefficients respectively. The weights are estimated sequentially according to the blocks of the ERAS items illustrated in Fig. 1. The arrows reported in this figure pointing from the risk factors into the other blocks indicate that the treatment is confounded and causally endogenous. The overall weight of patient \(i\) is determined as the sum of the product of the estimated inverse-probabilities of the treatments in each block as follows
\[
\hat{w}_i = \sum_{v=1}^{V} \frac{1}{\prod_{j=1}^{J} p_{ij}},
\]
where \(v = 1, \ldots, V\) denotes the block and \(j, j = 1, \ldots, J\) denotes the treatment in each block. The weights are stabilized by trimming them up to certain level to avoid high variability \([23]\).
The dependence of the potential latent variables is modelled through the following multinomial logit model
\[
\log \frac{p(U^{(z)}_i = u)}{p(U^{(z)}_i = u - 1)} = \beta_{0u} + d(z)' \beta_{1u}
\]
where \(u = 2, \ldots, k\), \(\beta_{0u}\) is the intercept specific of each latent class, \(d(z)\) is a vector with elements equal to 1 for treated patients and \(\beta_{1u}\) is the vector of parameters that define the ATET in the distribution of the latent variables.
Another set of parameters is referred to the conditional distribution of the observed outcomes and it is defined as
\[
p(Y_{ir} = 1|U^{(z)}_i = u) = \frac{1}{1 + \exp[-\eta_r(\zeta_i - \delta_r)]},
\]
where \(u = 1, \ldots, k\), \(\eta_r\) is a parameter measuring the discriminating power of the outcome \(r, r = 1, \ldots, p\), \(\delta_r\) is another parameter measuring the difficulty of the outcome, and \(\zeta_i\) indicates the point on the latent continuum where patient \(i, i = 1, \ldots, n\) is located. This is a 2PL model specification for the dichotomously scored outcomes.
The model is estimated through a weighted log-likelihood function by considering a sample of \(n\) independent patients for which we observe the multivariate binary outcomes. The target log-likelihood function is
\[
l(\Theta) = \sum_{i=1}^{n} \hat{w}_i l_i(\Theta),
\]
where \(\Theta\) denotes the overall vector of free parameters arranged in a suitable way. This weighted log-likelihood is maximized through the EM algorithm \([17]\). The latter estimates the model parameters considering as missing data the vector of latent variables. Then E-step of the algorithm computes the conditional expected value of the complete-data log-likelihood given the observed data and the current value of the parameters. The M-step updates the parameters maximizing the expected value of the quantity computed at the E-step. The two steps are alternated repeatedly until a convergence criterion is reached.
In order to choose the best number of latent classes and to discover meaningful groups of patients in the population we apply a model selection strategy. The resulting sub-populations should be internally cohesive and well separated from one another. We rely on the BIC index \([18]\) that is a measure of the relative goodness of fit of the model able to account simultaneously for its accuracy and complexity. It is defined as
\[
\text{BIC} = -2l(\hat{\Theta}) + \log(n) \# \text{par},
\]
where \(l(\hat{\Theta})\) denotes the maximum of the weighted log-likelihood of the model with \(k\) latent classes, \#\text{par} denotes the number of free parameters and \(n\) is the sample size. The model is estimated for an increasing number of latent classes and the best model is selected as that before the BIC starts to increase. Once the suitable number of latent classes is selected every patient is allocated to a latent class according the highest posterior probability. Standard errors for the estimated parameters are obtained as the square root of the diagonal elements of the inverse of the observed or expected information matrix computed through numerical methods.
**Results**
The available observations are related to 1261 patients operated between 2014 and 2018. Table 1 shows some descriptive statistics. The model is estimated by using the open source software \(R\) \([29]\) through the package \(multiLCIRT\) \([30]\). As far as we know there are no other software able to account for the multidimensionality of the responses and latent variables with a discrete distribution. For replicability purposes the code to estimate the proposed model is available from the repository at the following link https://github.com/penful/Eras. The complete results are available from the authors upon request.
**Results for the primary outcomes**
The multivariate causal LC-2PL model is estimated for a number of latent classes ranging from 1 to 5. According to the lowest value of the BIC index we select the model with three latent classes. Table 2 reports on the estimated conditional probabilities of the primary outcomes. The first class, which encompasses 47% of the patients, presents low
The second class, representing 35% of the patients, exhibits a high probability of late TRD and actual discharge although complications are improbable. Patients in the third class, accounting for the remaining 18% of patients, present high probability of complication and late TRD and actual discharge. Since these subpopulations are ordered according to the outcome occurrence we define the first LC as that representing the subpopulation of the best performing patients, the second that of intermediate patients and the third that of worst performing patients.
By looking at the estimated ATETs defined with respect to the first LC chosen as reference due to the fact that it collects patients with the best outcomes, we evaluate the efficacy of each intervention. Table 3 reports on the estimated ATETs of being in the 1st rather than in the 2nd LC. The estimated regression coefficients in the upper part of
the table indicate that no bowel preparation, colon surgery, ambulation within POD 1, IV fluid stop within POD 2, urinary catheter removal within POD 1, epidural anesthesia, and not inserting any surgical drainage significantly favor being in the class of best performing patients (1st LC) rather than in the group of less (intermediate) performing patients (2nd LC).
The estimated regression coefficients in the bottom part of the table indicate that preoperative hospital stay, malignant lesion, and blood transfusion significantly favor being in the class of intermediate performing patients (2nd LC) rather than in the group of worst performing patients (3rd LC) rather than stay in the intermediate group of patients (2nd LC).
**Results for the secondary outcomes**
The multivariate causal LC-2PL model for the secondary outcomes is estimated as the previous model for a number of latent classes ranging from 1 to 5. According to the lowest value of the BIC index we select the model with two latent classes. The estimates of the model parameters for the secondary outcomes are reported in Tables 5 and 6. According to the results showed in Table 5 we notice that both latent classes have a similar probability of bowel function recovery after POD 1 (0.51 and 0.56 respectively) and a null or very low probability of death before POD 30 (0.00 and 0.01 respectively). The first class (78% of patients) exhibits low probability for all the other secondary outcomes. In the second latent class IV fluid resumption, NGT replacement, PONV occurrence and surgical re-intervention are sensibly more probable, while hospital re-admission is only slightly more probable.
Table 4 reports on the estimated ATETs of being in the 2nd rather than in the 3rd LC. The estimated regression coefficients in the upper part of the table indicate that no bowel preparation, colon surgery, ambulation within POD 1, IV fluid stop within POD 2, urinary catheter removal within POD 1, epidural anesthesia, and not inserting any surgical drainage significantly favor being in the class of best performing patients (1st LC) rather than in the group of less (intermediate) performing patients (2nd LC).
thromboembolism prophylaxis, ambulation within POD 1, IV fluid stop within POD 2, NGT removal on POD 0, not inserting any surgical drainage, and epidural anesthesia favor being in the class of best performing patients (1st LC) rather than in the group of the worst performing patients (2nd LC).
The estimated ATETs in the bottom part of the table indicate that blood transfusion, preoperative glucidic drink, and prokinetics significantly favor being in the group of the worst performing patient (2nd LC) with respect to best performing (1st LC).
### Discussion
With respect to our primary outcomes, the model points out three classes of patients. The first one has a high probability of good outcome and represents the majority of our patients. The second class exhibits possibly prolonged TRD and discharge notwithstanding the absence of complications. The third class presents high probability of both complications and prolonged TRD and discharge. Five variables positively affect the outcomes (ambulation within POD 1, IV fluid stop within POD 2, urinary catheter removal within POD 1, epidural anesthesia, and not inserting any surgical drainage). Two variables negatively affect the outcomes (preoperative hospital stay and blood transfusion).
With respect to our secondary outcomes, three of them could not contribute to discrimination between classes of patients, since their probability was uniformly either low (hospital re-admission and death within POD 30) or high (bowel function recovery after POD 1). The model points out two classes of patients: a first class with high probability of good outcome (including the majority of our patients), and a second one with high probability of IV fluid resumption, NGT replacement, PONV occurrence, and surgical re-intervention. Six variables positively affect the secondary outcomes and four of them do the same for the primary outcomes (ambulation within POD 1, IV fluid stop within POD 2, epidural anesthesia, and not inserting any surgical drainage). Four variables negatively affect the secondary outcomes and one of them does the same for the primary outcomes (blood transfusion).
Our results contribute to the ongoing debate about the existence of ERAS “core-items”. In fact, although it is recognized that the complete ERAS protocol is the best way to improve postoperative outcome, the number and relative combination of the ERAS items implemented in
| 2st versus 3rd Latent Class | Estimated coefficient | S.E. | CI |
|-------------------------------------------------------|-----------------------|-------|------------------|
| Intercept | 2.08* | 1.14 | (−0.15, 4.31) |
| Ambulation within POD 1 | −1.51*** | 0.28 | (−2.06, −0.96) |
| Thromboembolism prophylaxis | −1.12 | 0.83 | (−2.75, 0.51) |
| Epidural anesthesia | −0.90*** | 0.27 | (−1.43, −0.37) |
| IV fluid stop within POD 2 | −0.81*** | 0.24 | (−1.28, −0.34) |
| UC removal within POD 1 | −0.68** | 0.26 | (−1.19, −0.17) |
| No surgical drainage | −0.67** | 0.26 | (−1.18, −0.16) |
| No bowel preparation | −0.48 | 0.72 | (−1.89, 0.93) |
| Colon surgery | −0.36 | 0.28 | (−0.91, 0.19) |
| Prokinetics | −0.31 | 0.25 | (−0.80, 0.18) |
| NGT removed on POD 0 | −0.08 | 0.36 | (−0.79, 0.63) |
| Intraoperative fluids | −0.04 | 0.03 | (−0.10, 0.02) |
| IV fluids during POD 1 | 0.02*** | 0.01 | (0.00, 0.04) |
| Preoperative hospital stay | 0.05** | 0.02 | (0.01, 0.09) |
| Preoperative glucidic drink | 0.20 | 0.22 | (−0.23, 0.63) |
| Morphine | 0.24 | 0.21 | (−0.17, 0.65) |
| Malignant lesion | 0.38 | 0.28 | (−0.17, 0.93) |
| PONV prophylaxis | 0.54** | 0.28 | (−0.01, 1.09) |
| Premedication | 0.97*** | 0.33 | (0.32, 1.62) |
| Blood transfusion | 1.90*** | 0.35 | (1.21, 2.59) |
LC: latent class; IV: intravenous; UC: urinary catheter; NGT: nasogastric tube; POD: postoperative day. Significance levels for the test according to the estimated standard errors that each parameter is equal to zero: *significant at 1%; **significant at 5%; ***significant at 10%
previous works varied across studies without substantial differences in postoperative short-term outcomes [31, 32]. Indeed, several studies imply that some ERAS elements may be more significant than others in affecting outcome [33–36] and that simplified protocols could yield comparable results [23, 24]. A systematic review suggests that the number of implemented ERAS items does not affect outcome and that applying only some core-items is sufficient to obtain the benefits of the ERAS program [37]. In contrast, a large cohort of patients from the ERAS Society Registry suggested that the improvement in clinical outcome provided by an ERAS program was directly correlated with the number of implemented items and the degree of compliance to the protocol [38, 39].
In addressing these issues no previous study adequately considered the dependency of multivariate outcomes from confounding variables and non-compliance to ERAS items, the inter-relation between ERAS items themselves, and the composite nature of the primary and secondary outcomes at stake.
The estimation of treatment effects in observational studies when the treatment assignment depends on the sequence of previous assignments and on time-varying confounders is still a matter of debate. The main advantage over the standard simple multinomial logit model is that our counterfactual framework assesses causal associations, corrects for pre-treatment confounders and for multiple treatment conditions in order to reduce the bias due to confounding. Another advantage is that it is a multivariate model-based clustering method and allows us to properly account for suitable groups able to disentangle the heterogeneous population of patients. Moreover, it yields a classification that uses the maximum a-posteriori estimates of the model parameters.
The results of the application provide evidence that waiving bowel preparation increases the probability of good outcome. Recent ERAS guidelines on colonic surgery and a large meta-analysis of more than 21,000 patients agree that mechanical bowel preparation is not associated to any reduction in postoperative complications, mortality, and length of hospital stay when compared with no preparation [40, 41]. Actually, waiving preoperative mechanical bowel preparation reduces the risk of preoperative dehydration and the possibly associated electrolyte disorders. This favors a reduction in fluid administration and the reaching of zero fluid balance. Moreover, a possible increment in Gram-negative bacterial components of the intestinal flora is associated with bowel preparation [42].
Recent studies demonstrate how a balanced intraoperative goal directed therapy reduces mortality, overall morbidity, and the time to first flatus and to oral feeding [43]. The ERAS guidelines clearly state that postoperative IV fluid administration is not necessary if oral intake is possible and that early oral feeding is safe and well tolerated by the majority of patients after colorectal surgery [44]. Early stop of IV fluid infusion and urinary catheter removal, together with good pain control and no surgical drainage, foster patients’ mobilization after surgery. A reduction of postoperative pulmonary and thromboembolic complications and a regain of preoperative functional capacity and muscular strength are strictly related to early postoperative mobilization [45].
The results suggest that epidural analgesia exerts a favorable effect on outcome and therefore they confirm the findings of other studies according to which epidural analgesia after laparotomy provides optimal pain control and reduces the inflammatory stress response. This reduces the incidence of pulmonary and cardiovascular
| Table 5 Estimated probability of each latent class and estimated conditional probabilities of the multivariate causal latent class two-parameter logistic model for the secondary outcomes |
|---------------------------------|----------------|
| **Estimated probability of each Latent Class** | 0.78 | 0.22 |
| **Estimated conditional probability for secondary outcomes** | | |
| Bowel function recovery after POD 1 | 0.51 | 0.56 |
| IV fluid resumption | 0.03 | 0.57 |
| NGT replacement | 0.00 | 0.34 |
| PONV occurrence | 0.08 | 0.41 |
| Surgical re-intervention | 0.01 | 0.16 |
| Hospital re-admission | 0.00 | 0.07 |
| Death within POD 30 | 0.00 | 0.01 |
POD: postoperative day; IV: intravenous; NGT: nasogastric tube; PONV: postoperative nausea and vomiting measure the first dimension and the remaining outcomes measure the second dimension
complications, in particular in frail patient [46, 47]. Nevertheless, the role of thoracic epidural analgesia after laparoscopic procedures is controversial and the importance of multimodal analgesic sparing-opioids strategies is widely accepted in less invasive surgery. Studies addressing purely laparoscopic colonic surgery suggest caution towards epidural analgesia [48]. It should be noted that, although laparoscopy is widely recognized as a predictor of faster recovery, this effect was not apparent in our results.
The negative effect of a prolonged preoperative hospital stay, which we observed, deserves attention. Serious comorbidities requiring longer preoperative hospitalization can make it difficult to optimize patients’ conditions and preserve functional capacity before surgery. It has been recently demonstrated that a decline in preoperative functional capacity, determined by the Duke score activity index, is associated to increased myocardial infarction and death 30 days and one year after major non-cardiac surgery [49]. Prehabilitation, as preoperative optimization of patients’ condition, is a key ERAS concept [50]. However, it is evident that this should not prolong the preoperative hospital stay. Our analysis supports the hypothesis that preoperative hospital stay may even impair prehabilitation. Our results on the adverse effect of blood transfusion on outcome support the evidence that a careful management of preoperative anemia through iron supplemental therapy can improve overall morbidity and mortality by reducing the need for blood transfusion in the perioperative period [39, 51].
A possible limitation of our study is related to conceivable differences between the participating hospitals and through the time span of the data collection, both in the degree of ERAS pathway implementation and in the baseline standard of care. A potential selection bias could be at stake, despite all centers have been invited to submit consecutive elective patients. Nevertheless, the wide mix of ages and comorbidities included indicates a small likelihood of selection bias among patients.
The major strength of our proposal resides in the features of the proposed statistical method. Moreover, we accessed a dedicated ERAS database from which we derived the data used for the analyses and we used a validated indicator of short-term recovery such as the TRD [10].
| Table 6 | Estimated effects of the causal latent class two-parameter logistic model for being in the 1st latent class rather than in the 2nd latent class for the secondary outcomes. Estimated standard errors and asymptotic confidence interval at confidence level of 0.95 |
|-----------------------------------------|-----------------|-----------------|-----------------|
| 1st vs 2nd LC | Estimated coefficient | S.E. | CI |
| Intercept | 1.24 | 1.37 | (−1.45, 3.93) | Favor being in the 1st LC |
| Thromboembolism prophylaxis | − 2.56*** | 0.80 | (− 4.13, − 0.99) |
| Ambulation within POD 1 | − 1.48*** | 0.27 | (− 2.01, − 0.95) |
| IV fluid stop within POD 2 | − 1.45*** | 0.25 | (− 1.94, − 0.96) |
| NGT removed on POD 0 | − 1.21*** | 0.36 | (− 1.92, − 0.50) |
| No bowel preparation | − 0.77 | 0.92 | (− 2.57, 1.03) |
| No surgical drainage | − 0.48* | 0.27 | (− 1.01, 0.05) |
| Epidural anesthesia | − 0.47* | 0.27 | (− 1.00, 0.06) |
| Colon surgery | − 0.23 | 0.28 | (− 0.78, 0.32) |
| PONV prophylaxis | − 0.13 | 0.28 | (− 0.68, 0.42) |
| UC removal within POD 1 | − 0.04 | 0.26 | (− 0.55, 0.47) |
| Intraoperative fluids | − 0.02 | 0.03 | (− 0.08, 0.04) |
| Morphine | − 0.02 | 0.23 | (− 0.47, 0.43) |
| IV fluids during POD 1 | − 0.01 | 0.01 | (− 0.03, 0.01) |
| Preoperative hospital stay | 0.00 | 0.03 | (− 0.06, 0.06) | Favor being in the 2nd LC |
| Malignant lesion | 0.04 | 0.30 | (− 0.55, 0.63) |
| Premedication | 0.46 | 0.38 | (− 0.28, 1.20) |
| Blood transfusion | 0.87*** | 0.33 | (0.22, 1.52) |
| Preoperative glucidic drink | 0.97** | 0.28 | (0.42, 1.52) |
| Prokinetics | 1.22** | 0.20 | (0.75, 1.69) |
LC: latent class; IV: intravenous; UC: urinary catheter; NGT: naso-gastric tube; POD: postoperative day. Significance levels for the test based on the estimated standard errors that each parameter is equal to zero: *significant at 1%; **significant at 5%; ***significant at 10%
Conclusion
We analyze a colonic surgery ERAS database by proposing a multivariate causal latent class two-parameter logistic model. This modern statistical method overcomes several drawbacks of traditional methods to estimate average treatment effects on the treated. Since we are dealing with observational studies we employ a propensity score method. We propose to use a maximum likelihood approach by employing weights estimated through the inverse-probability of receiving the treatments. In this way, we avoid rough comparisons of patients thus producing a valid inference and reducing the possibility of biased treatment effects. As noted by [52] generally a simple covariate adjustment cannot be able to produce unbiased estimates of the model parameters. The proposed method of analysis is able to account for patient heterogeneity and it constitutes a general approach for the analysis of data arising in similar medical contexts.
According to the results early postoperative ambulation and IV fluid stop, epidural anesthesia, and waiving any surgical drainage exert a favorable effect on primary outcomes (time ready for discharge and actual discharge from hospital, and occurrence of any complication, surgical site infection, and medical complications), while prolonged preoperative hospital stay and blood transfusion act unfavorably.
Acknowledgements Turi S. 1 MD, Monzani R. 2 MD, Radrizzani D. 3 MD, Beretta L. 4 MD Scatizzi M. 5 MD, Borghi F. 6 MD, Missana G. 7 MD, Azzola A. 8 MD, Muratore A. 9 MD, Crespi M. 10 MD, Iuliani R. 10 MD, Bima C. 11 MD, Bouzari H. 11 MD, Pisani Ceretti A. 12 MD, Pellegrino L. 13 MD, Casiraghi U. 1 MD, Ficari F. 1 MD
1. Anesthesia and Intensive Care, IRCCS San Raffaele, Milan, Italy, 2. Department of Anesthesia, Humanitas Research Hospital, Rozzano, Milan, Italy, 3. Department of Anesthesia, Legnano Hospital, Milan, Italy, 4. Department of Surgery, Prato Hospital, Prato, Italy, 5. Department of Surgery, Cuneo Hospital, Cuneo, Italy, 6. Department of Surgery, Casa di Cura Città di Udine, Udine, Italy, 7. Department of Surgery, Cantù Hospital, Cantù, Italy, 8. Department of Surgery, Candido Hospital, Turin, Italy, 9. Department of Surgery, Luigi Sacco Hospital, Milan, Italy, 10. Department of Surgery, Cottolengo Hospital Turin, Turin, Italy, 11. Department of Surgery, Mauriziano Hospital Turin, Turin, Italy, 12. Department of Surgery, San Paolo Hospital, Milan, Italy, 13. Department of Surgery, Careggi Hospital, University of Florence, Florence, Italy
Funding Open Access funding provided by Università degli Studi di Milano - Bicocca.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
References
1. Ljungqvist O (2014) ERAS-enhanced recovery after surgery: moving evidence-based perioperative care to practice. J Parenter Enteral Nutr 38:559–566
2. Ljungqvist O, Scott M, Fearon KC (2017) Enhanced recovery after surgery: a review. JAMA Surg 152:292–298
3. Braga M, Scatizzi M, Borghi F et al (2018) Identification of core items in the enhanced recovery pathway. Clin Nutr Espen 25:139–144
4. Lazarsfeld PF, Henry NW (1968) Latent structure analysis. Houghton Mifflin, Boston
5. Pennoni F (2014) Issues on the estimation of latent variable and latent class models. Scholars Press, Saarbücken
6. Rubin DB (2005) Causal inference using potential outcomes: design, modeling, decisions. J Am Stat Assoc 100:322–331
7. Robins JM, Hernan M, Brumback B (2000) Marginal structural models and causal inference in epidemiology. Epidemiology 5:550–560
8. Schulte PJ, Mascha EJ (2018) Propensity score methods; theory and practice for anesthesia research. Anesth Analg 127:1074–1084
9. Braga M, Borghi F, Scatizzi M et al (2017) Impact of laparoscopy on adherence to an enhanced recovery pathway and readiness for discharge in elective colorectal surgery: results from the Perioperative Italian Society registry. Surg Endosc 31:4393–4399
10. Fiore JF, Faragarh IG, Bialocerkowski A et al (2013) Time to readiness for discharge is a valid and reliable measure of short-term recovery after colorectal surgery. World J Surg 37:2927–2934
11. Rubin DB (1974) Estimating causal effects of treatments in randomized and nonrandomized studies. J Educ Psychol 6:688–701
12. Holland PW, Rosenbaum PR (1986) Conditional association and unidimensionality in monotone latent variable models. Ann Statist 4:1523–1543
13. Greenland S, Robins JM (1986) Identifiability, exchangeability, and epidemiological confounding. Int J Epidemiol 15:413–419
14. Cole SR, Hernan MA (2008) Inverse probability weights for marginal structural models. Ann J of Epidemiol 6:656–664
15. Lanza ST, Coffman DL, Xu S (2013) Causal inference in latent class analysis. Struct Equ Model 20:361–383
16. Bartolucci F, Farcomeni A, Pennoni F (2014) Issues on the estimation of latent variable and multiple categorical indicators. Struct Equ Model 21:459–492
17. Dempster AP, Laird NM, Rubin DB (1977) Maximum likelihood from incomplete data via the EM algorithm. J R Stat Soc B 39:1–38
18. Schwarz G (1978) Estimating the dimension of a model. Ann Stat 6:461–464
19. Lazarsfeld PF (1950) The logical and mathematical foundations of latent structure analysis. In: Stouffer SA et al (eds) Measurement and Prediction. Princeton University, N.J, pp 362–412
20. McLachlan GJ, Peel D (2000) Finite mixture models. Wiley, New York
21. Birnbaum A (1968) Some latent trait models and their use in inferring an examinee’s ability. In: Lord FM, Novick MR (eds) Statistical theories of mental test scores. Addison-Wesley, MA, pp 395–479
22. McCaffrey DF, Griffin BA, Almirall D, Slaughter ME, Ramchand R, Burgette LF (2013) A tutorial on propensity score estimation for multiple treatments using generalized boosted models. Stat Med 32:3388–3414
23. Elze MC, Gregson J, Baber U, Williamson E, Sartori S, Mehran R, Nichols M, Stone GW, Pocock SJ (2013) Comparison of propensity score methods and covariate adjustment: evaluation in four cardiovascular studies. J Am Coll Cardiol 69:345–357
24. Hernan MA, Brumback B, Robins JM (2001) Marginal structural models to estimate the joint causal effect of nonrandomized treatments. J Am Stat Assoc 96:440–448
25. Joffe MM, Ten Have TR, Feldman HI, Kimmel SE (2004) Model selection, confounder and marginal structural models: review and new applications. Am Stat 58:272–279
26. Rosenbaum PR (2020) Modern algorithms for matching in observational studies. Ann Rev Stat Appl 7:143–176
27. Robins JM (1986) A new approach to causal inference in mortality studies with sustained exposure periods. Application to control of the healthy worker survivor effect. Math Model 7:1393–1512
28. Robins J (1987) A graphical approach to the identification and estimation of causal parameters in mortality studies with sustained exposure periods. J Chron Dis 40:139–161
29. R Development Team (2020) R: a language and environment for statistical computing. R Found. Stat. Comput, Vienna
30. Bartolucci F, Bacci S, Gnaldi M (2017) MultiLCIRT: Multidimensional Latent Class Item Response Theory Models (R package version 2.11)
31. Ahmed J, Khan S, Lim M et al (2012) Enhanced recovery after surgery protocols–compliance and variations in practice during routine colorectal surgery. Colorectal Dis 14:1045–1051
32. Aarts MA, Oksienec A, Glicksman A et al (2012) Adoption of enhanced recovery after surgery (ERAS) strategies for colorectal surgery at academic teaching hospitals and impact on total length of hospital stay. Surg Endosc 26:442–450
33. Hendry PO, Haugel J, Nygren J et al (2009) Determinants of outcome after colorectal resection within an enhanced recovery programme. Br J Surg 96:197–205
34. Vlug MS, Barteles SA, Wind J et al (2012) Which fast track elements predict early recovery after colon cancer surgery? Colorectal Dis 14:1001–1008
35. Poon JT, Fan JK, Lo OS et al (2011) Enhanced recovery programme in laparoscopic colectomy for cancer. Int J Colorectal Dis 26:71–77
36. da Fonseca LM, Profeta da Luz MM, Lacerda-Filho A et al (2011) A simplified rehabilitation program for patients undergoing elective colonic surgery –randomized controlled clinical trial. Int J Colorectal Dis 26:609–616
37. Gianotti L, Beretta S, Luperto M et al (2014) Enhanced recovery strategies in colorectal surgery: is the compliance with the whole program required to achieve the target? Int J Colorectal Dis 29:329–341
38. ERAS Compliance Group (2015) The impact of enhanced recovery protocol compliance on elective colorectal cancer resection: results from an international registry. Ann Surg 261:1153–1159
39. Gustafsson UO, Hansel J, Thorell A et al (2011) Adherence to the enhanced recovery after surgery protocol and outcomes after colorectal cancer surgery. Arch Surg 146:571–577
40. Gustafsson UO, Scott MJ, Hubner M et al (2019) Guidelines for perioperative care in elective colorectal surgery: enhanced recovery after surgery (ERAS) society recommendations. World J Surg 43:659–695
41. Rollins KE, Javanmard-Emamghissi H, Lobo DN (2018) Impact of mechanical bowel preparation in elective colo-rectal surgery: a meta-analysis. World J Gastroenterol 24:519–536
42. Gaines S, Shao C, Hyman N et al (2018) Gut microbiome influences on anastomotic leak and recurrence rates following colorectal cancer surgery. Br J Surg 105:131–141
43. Sun Y, Chai F, Pan C et al (2017) Effect of peri-operative goal-directed therapy on postoperative recovery following major abdominal surgery- a systematic review and meta-analysis of randomized controlled trials. Crit Care 21:141
44. Zhuang CL, Ye X, Zhang CJ et al (2013) Early versus traditional postoperative oral feeding in patients undergoing elective colorectal surgery: a meta-analysis of randomized clinical trials. Dig Surg 30:225–232
45. Brower RG (2009) Consequences of bed rest. Crit Care Med 37:422–428
46. Block BM, Liu SS, Rowlingson AJ (2003) Efficacy of postoperative epidural analgesia: a meta-analysis. JAMA 290:2455–2463
47. Carli F, Halliday G (1997) Continuous epidural blockade arrests the postoperative decrease in muscle protein fractional synthetic rate in surgical patients. Anesthesiology 86:1033–1040
48. Turi S, Gemma M, Braga M et al (2019) Epidural analgesia vs systemic opioids in patients undergoing laparoscopic colorectal surgery. Int. J. Colorectal Dis. 34:915–921
49. Wijesundera DN, Pearse RM, Shulman MA et al (2018) Assessment of functional capacity before major non-cardiac surgery: an international, prospective cohort study. Lancet 391:2631–2640
50. Scheede-Bergdahl C, Minnella EM, Carli F (2019) Multi-modal prehabilitation: addressing the why, when, what, how, who and where next? Anaesthesia 74:20–26
51. Smilowitz NR, Oberweis BS, Nakula S (2016) Association between anemia, bleeding, and transfusion with long-term mortality following noncardiac surgery. Am J Med 129:315–323
52. Rubin DB (1979) Using multivariate matched sampling and regression adjustment to control bias in observational studies. J Am Stat Assoc 74:318–328
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3149,
1
],
[
3149,
8371,
2
],
[
8371,
10568,
3
],
[
10568,
15697,
4
],
[
15697,
20749,
5
],
[
20749,
21662,
6
],
[
21662,
23846,
7
],
[
23846,
28946,
8
],
[
28946,
33515,
9
],
[
33515,
38667,
10
],
[
38667,
45009,
11
],
[
45009,
50551,
12
]
]
} |
00a655b0c8339d0cff0382b1629eda87df72638b | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 22,
"total-fallback-pages": 0,
"total-input-tokens": 44461,
"total-output-tokens": 11614,
"fieldofstudy": [
"Law",
"Political Science"
]
} | EXTENUATING CIRCUMSTANCES
ADMINISTRATIVE LIABILITY IN THE
LAW ON HANDLING OF ADMINISTRATIVE
VIOLATIONS OF VIETNAM
Nguyen Nhat Khanh
Faculty of Administrative Law and State, Ho Chi Minh City University of Law
[email protected]
Abstract
Extenuating circumstances administrative liability in Vietnamese law are understood as such circumstances related to the determination of sanctioning and the extent of responsibilities towards individuals, organizations which violated administrative law. These circumstances are to alleviate the extent of danger for the society of the violations; therefore, when these circumstances are applied, individuals, organizations violating the administrative law will incur lower legal consequences than normal cases. The paper analyzes the theoretical issues of extenuating circumstances administrative liability in Vietnamese law, points out some shortcomings and provides proposals for improvement.
Keywords
Extenuating circumstances; administrative liability; administrative violation; administrative sanctions; Vietnamese law
1 LLM, PhD Student, Faculty of Administrative Law and State, Ho Chi Minh City University of Law.
02 Nguyen Tat Thanh Street, Ward 12, District 4, Ho Chi Minh City, Vietnam.
CONTENTS
I. Introduction ................................................................. 46
II. The specific characteristics of extenuating circumstances administrative liability ............................................ 48
III. Some practical issues of extenuating circumstances administrative liability in Vietnamese law .......................... 53
3.1. Extenuating circumstances in sanctioning administrative violations ................................................. 53
3.2. Legislative documents that stipulate additional extenuating circumstances that are not in accordance with the provisions of the Law on Handling of Administrative Violations 2012 .......................... 57
3.3. The Law on Handling of Administrative Violations 2012 has not yet developed a general principle to determine the specific level of sanctions in case of administrative violations with extenuating circumstances ........................................... 59
IV. Conclusion ................................................................. 65
References ........................................................................ 66
I. Introduction
Handling administrative violations is not only to punish violators but also to educate them about the awareness to abide by laws and present the humanity of laws. The Law on Handling of Administrative Violations 2012 regulates the sanction principles as follows: “The sanctioning of administrative violations must be based on the nature, seriousness, and consequences of these violations, violators and extenuating as well as aggravating circumstances.” The Law on Handling of Administrative Violations 2012 issues extenuating circumstances to guarantee humanity and to encourage violators to intentionally cooperate with positive attitudes to the problems and declare honestly. Meanwhile, aggravating circumstances are to punish more strictly those who commit crimes in a more dangerous manner, with behaviors that are detrimental to society. Within this paper, the author will focus on analyzing legal provisions on extenuating circumstances applied in how
---
2 Law on the Handling of Administrative Violations in 2012, Art. 3 (1.c).
3 Scientific commentary on the Law on Handling of Administrative Violations in 2012, at 129 (Nguyen Canh Hop ed., Publisher Hong Duc., 2017).
to deal with sanctioning administrative violations, spouting out some shortcomings of the current law and proposing the complete directions.
Based on Article 9 of the Law on Handling of Administrative Violations 2012, the following are deemed as extenuating circumstances administrative liability:
— **An administrative violator has taken an act(s) to prevent or limit consequences of his/her violation or voluntarily remedy consequences and pay damages;**
— **An administrative violator has voluntarily reported his/her violation or has shown sincere repentance for the violation, or has actively assisted functional agencies in detecting or handling administrative violations;**
— **A person commits an administrative violation in the state of being emotionally provoked by an illegal act of another person; or acts beyond the legitimate defense limit or beyond requirements of an emergency circumstance;**
— **A person commits an administrative violation under force or due to his/her material or spiritual dependence on another;**
— **An administrative violator is a pregnant woman, a weak aged person or a person suffering an illness or disability which deprives him/her of the ability to perceive or control his/her acts;**
— **A person commits an administrative violation due to his/her particularly difficult plight which is not attributable to his/her acts;**
— **A person commits an administrative violation due to his/her ignorance;**
— **Other extenuating circumstances stipulated by the Government.**
Compared to the Ordinance on Handling of Administrative Violations 2002 (amended and supplemented in 2007, 2008), the Law on Handling of Administrative Violations 2012 added 3 extenuating circumstances included as follows: i. **An administrative violator has shown sincere repentance for the violation, or has actively assisted functional agencies in detecting or handling administrative violations;** ii. **Committing an administrative violation which is beyond the legitimate defense limit;** iii. **Committing an administrative violation which is beyond the legitimate defense limit;**
which is beyond the requirements of an emergency circumstance. An addition to the circumstances “An administrative violator has shown sincere repentance for the violation; or has actively assisted functional agencies in detecting or handling administrative violations” is essential to encourage violators to correct the violations, cooperate with competent governmental agencies in order to promptly prevent, discover and handle violations. Meanwhile, adding the two circumstances “Committing an administrative violation which is beyond the legitimate defense limit” and “Committing an administrative violation which is beyond the requirements of an emergency circumstance” is considered to be appropriate in actual situations. This illustrates the humanity of laws when the subjects are in particular situations and must protect the legitimate interests of the government and organization as well as legitimate benefits and the rights of themselves and others.
II. The specific characteristics of extenuating circumstances administrative liability
Firstly, the application of extenuating circumstances reduces the level of administrative liability of individuals and organizations that violate administrative violations compared to normal cases.
The values of extenuating circumstances administrative liability are shown when violating subjects in these circumstances will be reduced to administrative liability compared to normal cases (average sanctions). For example, in cases of being fined, Law on Handling of Administrative Violations 2012 issued “if such violation involves an extenuating circumstance(s), the fine may be lower but must not be lower than the minimum level of the fine frame.” For the sanction of deprivation of the right to use licenses or practice certificates for a definite time or suspension of operation for a definite time in road and railway traffic sections, Decree No 46/2016/ND-CP issued:
“the duration of the suspension of a license or practicing certificate or transport business operation as a penalty for a violation specified
---
4 Law on Handling of Administrative Violations in 2012, Art. 23 (4).
in this Decree is the average level of the bracket. The minimum level shall be applied if there is a mitigating factor.”
Through it, it can be seen that the applications of extenuating circumstances are of importance in detail administrative responsibilities, guaranteeing equality in applying laws, violating individuals, and organizations have to take administrative responsibilities which are associated with characteristics, nature, and extent of the violation of each subject.
Secondly, the extenuating circumstances administrative liability with “open-ended” characteristics.
Analyzing the regulations in Article 9 of the Law on Handling of Administrative Violations 2012 to demonstrate that the extenuating circumstances are not in the “closed-in” list but can be expanded and added because along with the extenuating circumstances issued by the National Assembly which are listed in detail from Clause 1 to Clause 7, the Law is also regulated in an open manner in Clause 8 when allowing the Government to issue other extenuating circumstances. The expansion and addition depend on the decision of the Government. This regulation is necessary to create advantages for the authorities for actively making and choosing suitable extenuating circumstances with the variety of administrative violations in each different domain which aims to bring benefits for violators. This is a specific difference between extenuating circumstances and aggravating circumstances because of the policy of aggravating circumstances in a closed list which belongs to the authority of the National Assembly.
---
5 Decree No 46/2016/ND-CP, Art. 77 (2).
6 Article 10 of the Law on Handling of Administrative Violations 2012 stipulates a “closed list” of aggravating circumstances including the following 12:
a) Committing an administrative violation in an organized manner;
b) Repeatedly committing an administrative violation; recidivism;
c) Inciting, enticing or using a minor to commit a violation; forcing one’s materially or spiritually dependent person to commit an administrative violation;
d) Using a person who, with the clear knowledge of the violator, suffers a mental illness or other illness which deprives such person of the ability to perceive or control his/her acts, to commit an administrative violation;
e) Affronting or libeling a person on public duty; committing an administrative violation in a hooligan manner;
It must be noticed that only the government can have this privilege, other agencies which manage the authorities cannot issue regulations related to extenuating circumstances. This is a completely appropriate policy because the National Assembly is not the direct subject to carry out administrative activities, to arise relationships in administrative activities which occur unpredictably so it is needed to vary the form of sanction. Simultaneously, the Government is the highest state administrative body of the Socialist Republic of Vietnam, with the duty to unify and manage from the state to the local area, to enact the Decree in sanctioning administrative violations in all domains. By getting the Government to expand the foundation of extenuating circumstances administrative liability, it helps the competent individual to handle actively towards the variety of circumstances and violators, to enhance democracy in the handling of administrative violations.\textsuperscript{7} However, in the process of enacting the Decree on sanctioning administrative violations in each field, the competent agencies only focus on stipulating the administrative violations and the extent of sanction but not on other extenuating circumstances regulations.\textsuperscript{8}
\begin{itemize}
\item[f)] Abusing one’s position or powers to commit an administrative violation;
\item[g)] Taking advantage of war conditions, a natural calamity, disaster, epidemic, or other special difficulties of the society to commit an administrative violation;
\item[h)] Committing an administrative violation while serving a penalty under a criminal judgment or while executing a decision of application of an administrative violation handling measure;
\item[i)] Continuing to perform an act of administrative violation after being requested by a competent person to stop such an act;
\item[j)] Absconding or concealing an administrative violation after committing such violation;
\item[k)] Committing an administrative violation on a large scale or involving a large quantity of goods or goods of large value;
\item[l)] Committing an administrative violation against many persons, a child, an aged person, a disabled person or a pregnant woman.
\end{itemize}
\textsuperscript{7} Scientific commentary on the Law on Handling of Administrative Violations in 2012, at 169 (Nguyen Canh Hop ed., Publisher Hong Duc, 2017).
\textsuperscript{8} Dao Thi Thu An, Extenuating and Aggravating Circumstances in Handling Administrative Violations — Practice and Issues Raised, 5 Democracy and Law Journal 10 (2007).
Thirdly, the extenuating circumstances administrative liability are applied by the authorized person and must be reflected in the decision to sanction an administrative violation.
Sanctioning of administrative violations is an activity of practicing the state power, through sanctioning, the competent holder in the name of the State’s power to issue sanctions decisions compel the subject to administrative violations. It must be abided by sanctioning forms and remedial measures. Given the nature of the State’s exercise of power, the application of this coercive measure must be sanctioned by the competent subjects in accordance with the provisions of law.
The extenuating circumstances administrative liability are details associated with the violating subject, which reduce the level of administrative responsibility of that subject, so the application of these circumstances must be considered together with the decision to sanction administrative violations. Therefore, if the subject of administrative violations is considered to be applied the extenuating circumstances, these details must be shown in the content of the decision to sanction the administrative violation. On that basis, the subjects with sanctioning competence may apply lower sanctions compared to normal cases. When applying these circumstances, competent subjects must rely on objective truths to comprehensively consider and assess details attached to violating subjects. Therefore, in order to ensure the openness, the strictness of the law and consistency in the sanctioning process, the law stipulates that the application of the extenuating circumstances must be reflected in the sanctioning decision issued by the competent subjects.
Fourthly, the application of the extenuating circumstances administrative liability is “concreteness”.
The “concreteness” of the application of the extenuating circumstances administrative liability firstly has been shown by the extenuating circumstances having to be considered in each violation
---
9 See The Form of Decision on sanctioning administrative violations No. MQD02 in the Appendix “Some forms in handling administrative violations” issued together with the Government’s Decree No 97/2017/ND-CP of August 18, 2017 amending and supplementing a number of articles of the Government’s Decree No 81/2013/ND-CP of July 19, 2013 detailing a number of articles and measures to implement the Law on Handling of Administrative Violations 2012.
case. If an entity commits an administrative violation many times at different periods, the consideration and application of the extenuating circumstances (if any) shall only be applied according to each specific case. Suppose, on August 15, 2017, Mrs. A committed an act of “insulting the honor and dignity of others” as prescribed at Decree No 167/2013/ND-CP to a warning or a fine ranging from VND 100,000 to VND 300,000 shall be imposed for this violation. At the time of performing the act, Mrs. A is 8-month pregnant, this is the time when the woman has unusual emotional, psychological, and mental manifestations, so when considering sanctioning for Mrs. A, the authority may consider adopting the extenuating circumstance as “pregnant woman who has been an administrative violator” to reduce her administrative liability. However, on April 3, 2018, Mrs. A continued to commit acts of “spreading nails on roads”, according to Decree No 46/2016/ND-CP, a fine ranging from VND 6,000,000 to VND 8,000,000 shall be imposed for this act. This time, Mrs. A had given birth, so at the time of carrying out the act, Mrs. A was in a completely normal state of mind and consciousness, so there were no other circumstances to be mitigated and thus, she was sanctioned as in common cases.
In addition, the “concreteness” of the application of the extenuating circumstances in sanctioning administrative violations is reflected by the fact that only those who satisfy the prescribed signs can apply the extenuating circumstances, which shows that clear in the case of multiple people committing an administrative violation. The Law on Handling of Administrative Violations 2012 stipulates that “many people committing the same act of administrative violation shall each be sanctioned for such administrative violation.” However, the level of administrative responsibility of each person may vary depending on the specific circumstances attached to each violator. When conducting the sanctioning, the competent subject will have to base on these facts to decide the sanctioning form and level of sanction for each violator. For example, N and H jointly perform an act of “Stealing property” of other
---
10 Decree No 167/2013/ND-CP, Art. 5 (1.a).
11 Decree No 167/2013/ND-CP, Art. 11 (6.a).
12 Law on the Handling of Administrative Violations in 2012, Art. 3 (1.d).
people (the value of the stolen property is not sufficient to constitute a crime) as prescribed at Decree No 167/2013/ND-CP, this act may be subjected to a fine from VND 1,000,000 to VND 2,000,000.\textsuperscript{13} When discovered by a competent agency, N “voluntarily declared” the violation while H was still resolute not to declare his violation. Therefore, when carrying out the sanction, the competent person will still sanction both N and H for the “Stealing property” of other people. However, only N will be considered for application of the extenuating circumstance as “The violator voluntarily declared the act” as prescribed in Clause 2, Article 9 of the Law on Handling of Administrative Violations 2012.
The practice of implementing the Law on Handling of Administrative Violations 2012 has proved that the application of the administrative violations to individuals and organizations committing administrative violations is effective and practical, effective in personalizing, distributing the level of administrative responsibility. However, the legal regulations on the extenuating circumstances of administrative violations in sanctioning administrative violations still exist many shortcomings, leading to many difficulties and inconsistencies in the application of laws, reducing the effectiveness of sanctions against administrative violations, therefore, there is a need to make proposals for further improvement.
III. Some practical issues of extenuating circumstances administrative liability in Vietnamese law
3.1. Extenuating circumstances in sanctioning administrative violations
The Law on Promulgation of legislative documents 2015 issued: “The language of legislative documents is Vietnamese must be accurate, common, clear, and understandable. Contents of legislative documents must be specific, not vague...”\textsuperscript{14} This is one of the important requirements of the development and promulgation of legal documents to ensure that everyone can read and understand the laws easily and in the spirit of the agency drafting, thereby facilitating the observance
\textsuperscript{13} Decree No 167/2013/ND-CP, Art. 15 (1.a).
\textsuperscript{14} Law on Promulgation of legislative documents in 2015, Art. 8.
and application of legal regulations in an easy and uniform manner in practice.
Basically, most of the bases for using the extenuating circumstances are quite clear and easy to apply in practice. However, there are still some unclear grounds that could make it difficult for the competent entity to apply. For example, Clause 5, Article 9 of the Law on Handling of Administrative Violations 2012 stipulates that “an administrative violator is a pregnant woman, a weak aged person or a person suffering an illness or disability which deprives him/her of the ability to perceive or control his/her acts.” In general, these subjects have certain limitations in terms of awareness, psychology, and health at the time of committing administrative violations, so it is necessary to consider applying the extenuating circumstances, but to apply which one in each regulation is not easy.
Regarding the circumstance that “the administrative violator is a weak old person”, there are currently no regulations guiding this content. According to the provisions of the Law on the Elderly 2009, “the elderly people prescribed in this Law are Vietnamese citizens aged full 60 years or older.”\(^\text{15}\) Meanwhile, according to the guidance of Resolution No 01/2006/NQ-HDTP dated May 15, 2006 of the Council of Judges of the Supreme People’s Court guiding the application of a number of provisions of the Criminal Code, “old people” are defined as people aged 70 or over.\(^\text{16}\) According to Resolution No 01/2007/NQ-HDTP dated October 2, 2007 of the Council of Judges of the Supreme People’s Court guiding the application of a number of provisions of the penal code on the statute of limitations for executing a judgment, exempting from serving penalties or reducing the time limit for serving penalties, there are also guidelines on the subjects “people who are too old and weak” who are 70 years of age or older or people aged 60 or older but often sick.\(^\text{17}\) Thus, it can be seen that although there have been a number of documents mentioned, the subjects specified in the above-mentioned guiding documents are not identical with the subjects of “a weak old
\(^\text{15}\) Law on the Elderly in 2009, Art. 2.
\(^\text{16}\) Resolution No 01/2006/NQ-HDTP, Section 2 (4).
\(^\text{17}\) Resolution No 01/2007/NQ-HDTP, Section 4 (1.a).
person” prescribed in Clause 5, Article 9 of the Law on Handling of Administrative Violations 2012 and, thus, are not applicable in the field of sanctioning of administrative violations but are applied in the field of criminal matters. Therefore, the determination of the circumstance that “administrative violators are weak aged people” when sanctioning in reality now completely depends on the subjective judgment of the person with sanctioning competence. In order to correctly identify this particular subject, the person with sanctioning competence must base on each specific case to assess the health status and age of the violator at the time they commit the administrative violation.
In addition, the Law on Handling of Administrative Violations in 2012 does not provide guidance on the fact that “an administrative violator is a person suffering from an illness or disability which deprives him/her of the ability to perceive or control his/her acts.” These are special subjects that, when committing an administrative violation, they are not fully aware of the dangers to the society of their act and cannot control the act, so the Law on Handling of Administrative Violations 2012 stipulates that this is the extenuating circumstance. However, the question is how to identify a person at the time of committing an administrative violation falling into the above situation. For instance, if a driver of a vehicle with the flu has dizziness or distraction leading to a traffic law violation, it is considered as a disease that limits the ability to perceive or control the behavior. Is it allowed to apply this extenuating circumstance? This requires a thorough medical evaluation to determine, while the majority of sanctioning authority have no specialized knowledge in this field.
In addition, the contents of the extenuating circumstances are when “a person commits an administrative violation due to his/her particularly difficult plight which is not attributable to his/her acts” (Clause 6, Article 9) or “A person commits an administrative violation due to his/her ignorance” (Clause 7, Article 9), but there is no clear explanation. What circumstances are considered to be “particularly difficult”, just based on the presentation of the violator or must be certified by any state agency or not? What criteria to identify an administrative violator with “ignorance”? All of these questions are currently being left out by the Law on Handling of Administrative
Violations in 2012, which implies the risk of inconsistent application of the law, possibly even a fertile area for a competent person to apply these facts at will. Therefore, it is imperative that there is a specific need to explain these grounds for the extenuating circumstances to be applied correctly when the actual sanctions are applied.
Therefore, the Law on Handling of Administrative Violations 2012 and its implementing documents need to supplement regulations explaining terms and guiding specific identification criteria for the extenuating circumstances that have not been clearly defined to create the solid legal framework for the application of these circumstances, avoiding the situation of arbitrary application of the law in practice.
The application of the circumstance “Administrative violators are weak old people”, requires competent sanctioning of persons to prove that violators are both “old” and “weak”. The proof that the violator is an “old person” may be based on their age, but the current law does not specify how old is considered “old”. Meanwhile, to prove that a person is “weak” is even more difficult because it requires thorough supervision and medical examination, so it is difficult to give a common criterion to apply for those different cases. Therefore, according to the author, it is necessary for the Law on Handling of Administrative Violations 2012 to absorb the provisions of the criminal law in dealing with this inadequacy. The previous Criminal Code only used the criteria “the elderly” without the additional “the weak” criteria to apply the mitigating factors of criminal responsibility to offenders, but the application also encountered obstacles due to lack of specified instructions as stated. In order to solve this problem, the Criminal Code 2015 (amended and supplemented in 2017) has been improved when specifying the age of offenders to serve as quantitative criteria for the application of the mitigating factors of criminal responsibility for offenders instead of the previous criteria “old people”. Specifically, this Code stipulates that “offenders who are full 70 years of age or older” are allowed to apply mitigating factors on criminal responsibility. Therefore, the author thinks that the Law on Handling of Administrative Violations
---
18 Criminal Code in 1985, Art. 38 (1.e); Criminal Code in 1999 (amended and supplemented in 2009), Art. 46 (1.m).
19 Criminal Code in 2015 (amended and supplemented in 2017), Art. 51 (1.o).
2012 should also learn from the experience of criminal law and replace the extenuating circumstance “Administrative violators are weak aged people” to “Administrative violators are people full 70 years of age or older.” This provision will make it easier to identify the extenuating circumstances in reality, while also ensuring the humanity of the law because according to many health experts, the 70-year-old landmark is also considered a suitable landmark to evaluate the “old” and “weak” factors in a particular person.20
For other circumstances, such as “A person who commits an administrative violation has a disease or disability that limits his / her ability to perceive or control his/her acts”, “A person commits an administrative violation due to his/her particularly difficult plight which is not attributable to his/her acts” or “A person commits an administrative violation due to his/her ignorance”, the author boldly proposed the Law on Handling of Administrative Violations in 2012 need to consider supplementing the regulations allowing the Government to enact a specific Decree to guide the application of these circumstances. For each extenuating circumstance, the Government should promulgate clear guidelines for each specific criterion so that the authorized sanctioning person can base on that to be considered and applied in practice.
3.2. Legislative documents that stipulate additional extenuating circumstances that are not in accordance with the provisions of the Law on Handling of Administrative Violations 2012
As mentioned, one of the characteristics of the extenuating circumstances in sanctioning administrative violations is “open-ended”, reflected by the Law on Handling of Administrative Violations 2012 which allows the Government to enact the Decrees about how to define new extenuating circumstances in accordance with each specific field of state management to enable citizens to enjoy favorable circumstances
---
20 Diep The Vinh, The definitions of “the elderly, the feeble and elderly, the over feeble and elderly” need to be changed. May 18, 2017. https://kiemsat.vn/can-sua-do-cac-khai-niem-nguoi-gia-nguoi-gia-yeu-va-nguoi-qua-gia-yeu-trong-blhs-47028.html.
to reduce their level of administrative responsibility. However, through the survey, the author found that there are still existing cases where the unauthorized subject entity “arbitrarily” stipulates new extenuating circumstances that are not in accordance with the provisions of the Law on Handling of Administrative Violations 2012.
For example, the issue of sanctioning administrative violations in the field of science-technology and technology transfer is now governed by the Government’s Decree No 64/2013/ND-CP of June 27, 2013 (amended and supplemented by Decree No 93/2014/ND-CP dated October 17, 2014). The content of this Decree does not contain any provisions on the extenuating or aggravating circumstances when sanctioning administrative violations in the field of science — technology and technology transfer. However, the Ministry of Science and Technology’s Circular No 20/2015/TT-BKHCN of November 5, 2015, guiding the implementation of Decree No 93/2014/ND-CP, “arbitrarily” supplemented the extenuating circumstances for violations of the registration of results of performing State budget-funded scientific and technological tasks as follows: “Imposing the extenuating circumstance for cases where the hosting organization has registered the results within 01 year, counting from 30 days after the date of official acceptance of the scientific and technological task until the violation is detected.”
The issue to note here is that the Decree No 64/2013/ND-CP (amended and supplemented by Decree No 93/2014/ND-CP) issued by the Government did not mention this extenuating circumstance. Thus, it can be concluded that in this case the Ministry of Science and Technology has “arbitrarily” provided additional extenuating circumstances, although this agency is not allowed by the Law on Handling of Administrative Violations 2012 to do this.
Conducting a thorough review of legal documents that stipulate the sanctioning of administrative violations, especially the Circulars guiding sanctioning Decrees in the fields to promptly detect and abolish the regulations of “arbitrary” supplementing the extenuating circumstances. It is necessary to abolish the provisions of Article 5 (2.a) of Circular No 20/2015/TT-BKHCN: “Applying the extenuating circumstances
---
21 Circular No 20/2015/TT-BKHCN, Art. 5 (2.a).
when sanctioning for the case where the host organization has registered the results within 01 year, counting from 30 days after the date of official acceptance of the science and technology task until the violation is detected” because of the Ministry of Science and Technology does not have the right to stipulate more new extenuating circumstances.
3.3. The Law on Handling of Administrative Violations 2012 has not yet developed a general principle to determine the specific level of sanctions in case of administrative violations with extenuating circumstances
From the perspective of administrative law, the principle in state management is the overall administration of legal provisions with the main ideas as the basis for organizing the implementation of State management activities.22 The application of the extenuating circumstances in administrative sanctioning activities also needs to adhere to certain principles so that the application of these circumstances will achieve the expected effect. Unfortunately, the Law on Handling of Administrative Violations in 2012 and its implementing documents have not yet developed general principles to apply the extenuating circumstances when sanctioning, thereby creating many legal gaps when applying these facts in sanctioning practice.
The extenuating circumstances are characterized by a reduced level of administrative responsibility of the subject of administrative violations, but these circumstances are only meaningful for those administrative violations subject to sanction forms on the fine bracket (with regulations on the minimum to the maximum level) such as fines, deprivation of the right to use licenses, professional practice certificates for a definite time or suspension of operation for a definite time. For sanctions such as confiscation of material evidence, means of administrative violations or deportation, the application of the sanctions is not meaningful because of the fixedness of these sanctions. For the form of warning sanction, the extenuating circumstances are meant as a condition for subjects of administrative violations for individuals aged
---
22 Hanoi Law University. Vietnamese Administrative Law Curriculum. Publisher People’s Police. 75, 76. (2008).
full 16 years of age or older and organizations violating administrative regulations to apply this form of sanction (non-serious administrative violations with involved extenuating circumstances, and according to regulations subject to this sanctioning form of warning).
Currently, the Law on Handling of Administrative Violations 2012 only stipulates the principle of determining specific levels of fines in case of extenuating circumstances for fines. Meanwhile, for other sanctions such as deprivation of the right to use licenses, professional practice certificates for a definite time or suspension of operation for a definite time, the Law does not have any specific principles for determining the level of sanctions. For fines, the Law on Handling of Administrative Violations 2012 stipulates: “The specific fine level for an administrative violation is the average of the prescribed fine bracket for such behavior; if there are extenuating circumstances, the fine may be reduced but not lower than the minimum of the fine frame; if aggravating circumstances are involved, the fine level may be increased but must not exceed the maximum fine level of the fine bracket.” However, this provision is still general, so it creates a non-uniform application in practice.
For example, Decree No 102/2014/ND-CP stipulates a fine of between VND 5,000,000 and VND 10,000,000 for acts of encroaching on or occupying residential land. According to the guidance of the Law on Handling of Administrative Violations 2012, when individuals commit the above violations with extenuating circumstances, the person with sanctioning competence may reduce fines below the average level (below VND 7,500,000) and the lowest reduction is VND 5,000,000. However, the question is based on the criteria to determine the specific reduction. This issue has not been thoroughly resolved by the Law on Handling of Administrative Violations 2012, resulting in the decision to reduce the amount of the fine depends entirely on the subjective awareness of the person with sanctioning competence without relying on any quantitative criteria.
---
23 Law on Handling of Administrative Violations in 2012, Art. 23 (4).
24 Decree No 102/2014/ND-CP, Art. 10 (3).
In order to solve the above problems, a number of Decrees and Circulars guiding the sanctioning of administrative violations in the fields have developed the principle of determining specific levels of a fine when violating subjects that have extenuating circumstances, through our survey, the following two common ways are summarized:
The first way determines the sanction reduced by a percentage. For example, for administrative violations in the field of tax, Decree No 129/2013/ND-CP stipulates that when a fine is imposed, a specific fine for a tax procedure violation is the average of the frame fines are prescribed for such acts. For acts of violating tax procedures, each extenuating circumstance is entitled to a 20% reduction of the average fine level of the fine bracket. In the field of competition, Decree No 71/2014/ND-CP (amended and supplemented by Decree No 141/2018/ND-CP) provides for administrative violations regarding control of restrictive acts, in case of competition or unfair competition regulations, each extenuating circumstance shall be reduced by 15% of the fine level compared to ordinary violations. Meanwhile, Circular No 07/2014/TT-BTC guiding the sanctioning of administrative violations in the field of management and use of State assets stipulates that for every extenuating circumstance, the fine will be reduced by 20% compared to the average level of the fine bracket which is prescribed for that behavior. Although the determination of reduced fines by percentage makes it easy for the competent entity to determine the specific amount of fines when the extenuating circumstances are available, if this method is applied to the sanctioning of administrative violations in all sectors, it still has certain limitations. Specifically, if the lawmaker stipulates that the percentage (%) is too low (less than 10%), the sanction will not promote the value of the extenuating circumstances because the actual reduction of fines will not be different than a normal violation. Conversely, if the percentage (%) is 10% or 20% or more, there is also the risk of being disabled in many cases. In fact, if the lawmaker has set a percentage (%) of 10%, or 20% or more, only one
25 Decree No 129/2013/ND-CP, Art. 3.
26 Decree No 71/2014/ND-CP (amended and supplemented by Decree No 141/2018/ND-CP), Art. 4 (5), and Art. 5 (4).
27 Circular No 07/2014/TT-BTC Art. 3 (2.b).
extenuating circumstance is required, the fine is equal,\textsuperscript{28} even lower than the lowest fines of fine bracket.\textsuperscript{29} Therefore, if there are two or three extenuating circumstances, it is only possible to apply the minimum fine of the fine frame but not different from the case of having one. This is completely inconsistent with the purpose of sanctioning to educate and deter violators and also not in accordance with the principle of sanction that must be “\textit{based on the nature, seriousness, and consequences of the violation, regarding violations and the world center, aggravating circumstances.”}\textsuperscript{30} It would be absurd if the violating subject has many extenuating circumstances, the fine applied is exactly the same as the violating subject having only one extenuating circumstance.\textsuperscript{31}
\textit{The second way determines the reduction sanction according to the principle of reduction of the average}. In the field of price, fee, fee and invoice management, Decree No 109/2013/ND-CP (amended and supplemented by Decree No 49/2016/ND-CP) stipulates the fine of a violation against regulations on pricing, fee, and invoicing, without aggravating or extenuating circumstances is the average level of the fine bracket for such violation. The average fine is the arithmetic mean of the minimum and maximum fines. An extenuating circumstance shall cause the average fine to decrease. The decrease is the arithmetic mean of the minimum fine and average fine. The minimum fine shall
\textsuperscript{28} Article 6 (2.b) of the Decree No 46/2016/ND-CP sanctioning administrative violations in the field of road traffic and rail transport prescribes a fine of from VND 80,000 to VND 100,000 for the act of “Going three abreast or more”. If the violator has an extenuating circumstance and applies the “10 % reduction of the average fine level of the fine frame”, the fine will be VND 81,000 (VND 90,000 – 10 % × VND 90,000 = VND 81,000). This fine is approximately equal to the minimum fine.
\textsuperscript{29} Article 6 (1.a) of the Decree No 46/2016/ND-CP sanctioning administrative violations in the field of road traffic and rail transport prescribes a fine of from VND 60,000 to VND 80,000 for the act of “Disobeying road signs or road markings”. If the violator has an extenuating circumstance and applies the “20 % reduction of the average fine level of the fine frame”, the fine will be VND 56,000 (VND 70,000 – 20 % × VND 70,000 = VND 56,000). This fine is even lower than the minimum fine.
\textsuperscript{30} Law on the Handling of Administrative Violations in 2012, Art. 3 (1.c).
\textsuperscript{31} Cao Vu Minh, Issues in Need of Amendment under the Law on Handling of Administrative Violations in 2012, \textit{1 State and Law Review} 9 (2019).
apply if two extenuating circumstances are found. An aggravating circumstance shall cancel out an extenuating circumstance.\textsuperscript{32}
In our opinion, this method has many advantages over the method of determining the reduction in percentage and can be applied to all areas, because determining the reduction in this way helped the authority to easily determine the reduction of sanction levels when there is an extenuating circumstance while creating a division of administrative responsibilities between the violation of having a single extenuating circumstance and the violation with many extenuating circumstances.
Regarding the sanctioning form of depriving of the right to use licenses, professional practice certificates for a definite time or suspending operation for a definite time, although the Law on Handling of Administrative Violations 2012 has not specified the principle of determining the reduction level of fines when having extenuating circumstances, but through the survey of Decrees stipulating the sanctioning of administrative violations in the fields, we found that there are a number of Decrees guiding the determination of reduced fines for these sanctions when there are extenuating circumstances as follows:
Decree No 46/2016/ND-CP sanctioning administrative violations in the field of road and rail transport provides: “The duration of the suspension of a license or practicing certificate or transport business operation as a sanction for a violation specified in this Decree is the average level of the bracket. The minimum level shall apply if there is an extenuating circumstance...”\textsuperscript{33} Meanwhile, Decree No 33/2017/ND-CP sanctioning administrative violations in the field of water resources and minerals provides the principle for determining the reduction level of fines for the issuance of extenuating circumstances with sanctioning forms of “deprivation of the right to use licenses, professional practice certificates” or “operation stoppages” are as follows:
“The period of suspension of a license or practicing certificate or suspension of operations as a sanction for a violation specified in this
\textsuperscript{32} Decree No 109/2013/ND-CP (amended and supplemented by Decree No 49/2016/ND-CP), Art. 3 (6).
\textsuperscript{33} Decree No 46/2016/ND-CP, Art. 77 (2).
Decree is the average level of the suspension period bracket applied to such violation. It may be shorter than the average level, but not shorter than the minimum level of the suspension period bracket if there are any extenuating circumstances..."34
However, both of these methods also have certain limitations, so they cannot be used as a common standard to apply to all fields. If being applied according to Decree No 46/2016/ND-CP, it will not make a difference in the level of administrative liability reduction in case the administrative violator has many extenuating circumstances, because for the violators shall be applied the “minimum of the time frame” for deprivation or suspension of operation with only one extenuating circumstance. Meanwhile, if applying under Decree No 33/2017/ND-CP, the specific reduction cannot be determined, but depends on the discretion of the sanctioning authorities. Therefore, it is necessary that lawmakers also need to develop an appropriate general principle to apply to all areas of administrative sanctions for forms of sanctioning “deprivation of the right to use permits and certificates” or “suspend operation” when the violators have the extenuating circumstances.
Thus, lawmakers need to consider adding in the Law on Handling of Administrative Violations 2012 the principle of determining reduced sanction levels for specific administrative violations when the extenuating circumstances are available. As analyzed above, we believe that the lawmakers should not choose the method of determining the reduced sanction levels by the percentage (%). Otherwise, we highly recommend that lawmakers should determine the reduced sanction levels based on the average sanction level of the sanction bracket. The average reduction principle should be considered in accordance with the provisions of Decree No 109/2013/ND-CP (amended and supplemented by Decree No 49/2016/ND-CP). Accordingly, the Law on Handling of Administrative Violations in 2012 may be amended as follows:
Regarding the sanctioning form of fines: "The specific fine level of an act of violation is the average of the fine bracket prescribed for such violation. In case there is an extenuating circumstance, the
34 Decree No 33/2017/ND-CP, Art. 66 (2).
average reduction is applied. The average reduction is determined by halving the total of the minimum and the average. In case there are more than one extenuating circumstances, the minimum rate of the fine bracket shall be applied. In the case of both aggravating circumstances and extenuating circumstances, the principle of an aggravating circumstance is deducted for extenuating circumstances.”
For the sanctioning form of “deprivation of the right to use licenses, professional practice certificates or to suspend operation for a definite time”, the provisions are as follows:
“Time limits for deprivation of the right to use licenses, professional practice certificates or suspension periods only specifically activate for a violation which is the average of the time frame of deprivation or suspension of operation specified in that act. In case there is an extenuating circumstance, the average reduction is applied. The average reduction is determined by halving the total of the minimum and the average. In case there are more than one extenuating circumstances, the minimum of the time limit shall be applied. In the case of both aggravating circumstances and extenuating circumstances, the principle of an aggravating circumstance is deducted for an extenuating circumstance.”
IV. Conclusion
The state of law must uphold the rule-of-law principle. The rule-of-law principle is to recognize the objective existence of the law, to ensure the supremacy of the Constitution and the law. That means state agencies, civil servants and officials are all bound by the law. Not stopping there, the rule of law essentially requires the law to be published publicly, with clear content, no conflict, stability, predictability, feasibility, and general application for all relevant stakeholders reflect the values of social progress such as freedom, dignity, humanity, justice, democracy
---
35 Cao Vu Minh, Issues in Need of Amendment under the Law on Handling of Administrative Violations in 2012, 1 State and Law Review 10, 12 (2019).
36 Tran Thai Duong, Discuss the Concept and Principle of Rule-of-Law, 3 State and Law Review 4 (2017).
and human rights. Therefore, the amendment of the Law on Handling of Administrative Violations 2012 and its guiding documents are related to the establishment and application of the extenuating circumstances is imperative to create a solid legal framework for the application of using these circumstances in the actual sanctioning of administrative violations, thereby creating a division of administrative responsibilities, promoting the educational value of the sanction and showing the law’s humanitarian.
REFERENCES
1. Cao Vu Minh, Issues in Need of Amendment under the Law on Handling of Administrative Violations in 2012, 1 State and Law Review (2019).
2. Dao Thi Thu An, Extenuating and Aggravating Circumstances in Handling Administrative Violations — Practice and Issues Raised, 5 Democracy and Law Journal, 10 (2007).
3. Diep The Vinh, The definitions of “the elderly, the feeble and elderly, the over feeble and elderly” need to be changed. May 18, 2017. https://kiemsat.vn/can-sua-doi-cac-khai-niem-nguoi-gia-nguoi-gia-yeu-va-nguoi-qua-gia-yeu-trong-blhs-47028.html.
4. Hanoi Law University. Vietnamese Administrative Law Curriculum. Publisher People’s Police (2008).
5. Nguyen Duc Minh, Concept and Content of Rule-of-Law Principle, 6 State and Law Review (2018).
6. Scientific commentary on the Law on Handling of Administrative Violations in 2012, at 169 (Nguyen Canh Hop ed., Publisher Hong Duc, 2017).
7. Tran Thai Duong, Discuss the Concept and Principle of Rule-of-Law, 3 State and Law Review (2017).
---
37 Nguyen Duc Minh, Concept and Content of Rule-of-Law Principle, 6 State and Law Review 8 (2018). | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
1242,
1
],
[
1242,
3581,
2
],
[
3581,
5694,
3
],
[
5694,
7839,
4
],
[
7839,
10277,
5
],
[
10277,
12864,
6
],
[
12864,
15338,
7
],
[
15338,
17700,
8
],
[
17700,
19953,
9
],
[
19953,
22299,
10
],
[
22299,
24777,
11
],
[
24777,
27280,
12
],
[
27280,
29491,
13
],
[
29491,
31825,
14
],
[
31825,
34081,
15
],
[
34081,
36315,
16
],
[
36315,
38717,
17
],
[
38717,
41549,
18
],
[
41549,
43893,
19
],
[
43893,
46161,
20
],
[
46161,
48308,
21
],
[
48308,
49940,
22
]
]
} |
19358213e48c8c6edd962b951d87da60164f99d5 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 8,
"total-fallback-pages": 0,
"total-input-tokens": 24733,
"total-output-tokens": 8422,
"fieldofstudy": [
"Medicine"
]
} | Comparison of binocular game and patching in treating mild to moderate anisometropic amblyopia: a study protocol for a randomized controlled trial
Mohammad Etezad Razavi, Marzieh Najjaran, Jaber Mohseni and Shokoufeh Aalaei*
Abstract
Background: Amblyopia, as a neurodevelopmental preventable visual disorder, affects approximately 1.1% in Asia [1]. A binocular approach to treating amblyopia has been recently proposed. Whether the binocular playing game treatment is comparable to patching treatment needs further randomized clinical trials. To address this, the present research, designs, develops, and evaluates a new binocular game to treat amblyopia.
Methods: This study has been designed as a non-inferiority, randomized, two parallel-group, controlled trial. Forty-four patients between 4 and 12 years diagnosed with amblyopia will be randomly assigned to the control and intervention groups. In the intervention group, amblyopia treatment is provided with red-green anaglyphic glasses and a red filter placed in front of the amblyopic eye, along with a game to be played for 30 min twice a day. Those in the control group will receive patch therapy according to amblyopia treatment study protocol. The primary outcome is to change visual acuity in the amblyopic eye from the baseline to 3 months after randomization.
Ethics and dissemination: The Ethics Committee of Mashhad University of Medical sciences’ approval date was February 28, 2018, with a reference code of IR.MUMS.fm.REC.1396.783. Thus far, the recruitment of participants has not been completed and is scheduled to end in September 2021. The results will be disseminated in a peer-reviewed journal.
Trial registration: Iranian Registry of Clinical Trials IRCT2018021703876N1. Registered on 22 April 2019.
Keywords: Amblyopia, Patch therapy, Serious game, Stereopsis, Binocularity, Mobile health
Introduction
Amblyopia, as a neurodevelopmental preventable visual disorder, affects approximately 1.1% in Asia [1]. It follows from inadequate stimulation of the visual system during the critical period of visual development. Amblyopia is mainly associated with visual acuity (VA) reduction and binocular dysfunction. The most common risk factor for unilateral amblyopia is anisometropia [2, 3].
Refractive correction is the first step of amblyopia treatment regardless of the cause of amblyopia (anisometropia, strabismus, or both) [4, 5]. Patching the healthy eye has a long history and is the current standard treatment with at least one to two line improvements in VA in the amblyopic eyes [6–8]; however, it is faced with several limitations. Abnormal binocularity [9–11], recurrence even after a successful treatment in at least 25% of amblyopic children [12], poor compliance and negative outcomes including distress [13], and low self-perception of social acceptance [12] hinder the success of patching treatment.
* Correspondence: [email protected]
© The Author(s). 2021 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Penalizing the fellow eye either with atropine or optically as the alternative options to occlusion therapy along with their effective outcomes needs careful monitoring [14, 15]. Both patching and penalizing methods are monocular treatment approaches, in which by depriving the fellow eye of vision, the use of the amblyopic eye is promoted.
More recently, a novel approach based on the binocular origin of amblyopia has received considerable research interest due to possibly better visual outcomes in treating amblyopia [16, 17].
The relevant hypothesis for the binocular treatment of amblyopia is anti-suppression therapy, according to which the cortical input is suppressed in the amblyopic eye by inhibitory signals from the fellow eye. Thus, through minimizing suppression, the brain learns to see through the amblyopic eye [10]. In this context, the dichoptic presentation stimulus is applied to alleviate suppression in amblyopia.
In dichoptic training, each eye receives different images separately, and to complete a task, both eyes are forced to work together. The fixing eye versus the amblyopic receives the stimuli with the lower contrast. If the task is successfully completed, the contrast in the non-amblyopic eye is slowly increased until that in both eyes is equal.
The association between binocular dysfunction and deficits related to amblyopia has been clarified by Birch et al. [11]. In addition, in children between 3 and 7 years of age, Birch et al. found approximately 1 line improvement of visual acuity in amblyopic eyes with binocular iPad treatment [18]. In another study by Kelly et al. using an adventure binocular game for 1 h a day, they found an improvement of 1.7 lines of VA in 4 weeks [19].
Besides the amblyopic-eye VA improvement with binocular treatment, enhancing binocular functions has been proposed in several studies [5, 20, 21]. Weber et al. reported the improvement of fine motor skills 5 weeks after the binocular treatment in amblyopia [20].
However, despite the promising results of binocular treatment in children with amblyopia [18, 19, 22, 23], randomized clinical trials showed inconsistent results in the amblyopic eye visual acuity improvement [24–26].
Some other research by the Pediatric Eye Disease Investigator Group (PEDIG) did not reveal the priority of binocular game treatment over patching not only to children but also to teenagers with amblyopia [27, 28]. Similarly, Jing Yao et al. showed that although a 40-min daily game played for 3 months could improve VA for 0.18 logMAR, binocular game treatment alone was not more effective than patching in treating children with anisometric amblyopia. The new computer game applied in the study was based on a push-pull method [26]. In this model, stimulating the amblyopic eye while inhibiting the strong eye led to the re-balancing of intraocular interactions [26, 29].
There is also a considerable need for further robust clinical trials to show the possible effectiveness of binocular treatment. The existing differences among several factors in different studies, including the severity and type of amblyopia, type of binocular game, dose of treatment prescribed, and prior amblyopic treatment, make cross-comparison difficult.
More appealing games and more frequent supervision are suggested in studies exploring binocular amblyopia treatment [19, 26, 30]. Furthermore, by increasing the popularity of video games, they can be used for healthcare purposes with a higher interest in research [31]. With this regard, poor compliance as an issue accounting for patching treatment failure could benefit from such an engagement strategy for amblyopia treatment [17, 18, 22].
Therefore, the present study aims to design, develop, and evaluate a new binocular game using dichoptic images to investigate whether binocular game treatment makes any difference in the visual acuity of target patients.
We hypothesized that game with specific, well-defined characteristics designed in a structured way with the participation of a multi-disciplinary team would improve amblyopic-eye visual acuity in patients who received game compared to patients who only received patch therapy. So, a non-inferiority, randomized, two parallel-group, controlled trial is employed to investigate whether the binocular game makes any difference in visual acuity outcome in target patients.
Methods
Study design and setting
This manuscript was written in accordance with the SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) and CONSORT (Consolidated Standards Of Reporting Trials) 2010 guidelines [32, 33] (Additional file 1).
A non-inferiority, randomized, two parallel-group, controlled trial is employed to investigate whether the binocular game makes any difference in visual acuity in target patients.
The study is conducted in Khatam Alanbia Eye Hospital in Mashhad, Khorasan Razavi province, Northeastern Iran. Khatam Alanbia is the only specialized public hospital affiliated with universities providing ophthalmology services in northeastern Iran. The center is affiliated with Mashhad University of Medical Sciences (MUMS).
Participants
The target population consists of untreated patients afflicted with mild to moderate anisometric amblyopia,
who refer to the center and meet the inclusion criteria. Parents or the caretakers of child participants will need to provide written informed consent for participation. Similarly, the child participants will give oral informed consent. An assistant researcher will collaborate to obtain the consent. It will be ensured that they can withdraw from the study any time they want with no effect on their subsequent care. Informed consent has already been evaluated by the Ethics Committee of MUMS (Ethical code: IR.MUMS.fm.REC.1396.783). The eligibility criteria are assessed by one of the investigators (JM) as below.
Inclusion criteria
- Children aged 4 to 12 years with anisometropic amblyopia (amblyopia in the presence of a spherical equivalent $\geq 0.50$ diopter between two eyes or difference in stigmatism in any meridian $\geq 1.50$ diopter) with an interocular difference of at least two lines
- Children not previously treated for amblyopia except for spectacle correction with stable best corrected visual acuity (BCVA)
- Children with mild to moderate anisometropic amblyopia (BCVA $\geq 0.2$)
- Children whose parents or caretakers consented to enter the study
Exclusion criteria
- Patients with amblyopia with other causes (non-refractive)
- Patients with ocular or systemic diseases or any previous intraocular surgery
- Patients with a motor neurological disorder and brain lesions but unable to play the game
- Patients unable or unwilling to provide informed consent or not accessible by the end of the study
Objectives
The primary purpose of the present trial is to investigate whether a 3-month binocular game treatment is comparable to part-time patching in improving amblyopic-eye visual acuity. We hypothesized a 15% change in visual acuity after binocular game treatment. The trial will also assess any change to stereoacuity and suppression; furthermore, the compliance of the binocular game will be addressed.
Treatment arms
Before the randomization, participants with no previous history of using spectacles will have to wear their appropriate optical correction for 16 weeks full-time. After that time, if the visual acuity of the amblyopic eye is stable or a change of 0.1 logMAR or less is observed for another four weeks, and they meet all inclusion criteria, written informed consent will be obtained and randomization will follow the baseline examination. For participants with a history of wearing correction for more than 3 months meeting all eligible criteria, the randomization will follow immediately. Amblyopic children will receive active dichoptic binocular game or patching treatment for 3 months (Fig. 1).
Intervention group
In the intervention group, amblyopia treatment is done with red-green anaglyphic 3D glasses with a red filter placed in front of the amblyopic eye and a game for 30 min twice a day (totally 1 h a day), 5 days a week for 4 weeks. Then, this will continue for 2 days a week for 8 weeks (36 h of total treatment).
Control group
In the control group, according to the amblyopia treatment study protocol, participants will be required to patch their non-amblyopic eye for 2–4 h a day and last for 3 months. They will receive three times in-person clinical assessments between baseline and final assessment as well as the intervention group.
Intervention
Need analysis
To design the game to treat amblyopic patients, the first step was to extensively review the related literature on the games already used in terms of the effectiveness, appropriateness to the target group, specific features of the game, and probable discussions in design. Moreover, the existing games on amblyopia in AppStore were analyzed. Then, in several meetings with eye specialists ($n = 3$), the scientific aspects of the topic and the characteristics of target patients in the game and their compliance to traditional treatments and the necessity of offering an alternative or complementary therapy were analyzed. Consequently, according to the present findings and experts’ comments and patients’ cultural and local tendencies, the scenarios and main features of the game were specified for the design.
Design process
To design the technical features of the game and apply the maximum capabilities of the game production domain, a team of experts was consulted to design and develop the game. The original idea of the game was discussed in meetings shared between the clinical and technical teams. The ambiguities were solved and the technical aspects were discussed. Eventually, considering the fact that the amblyopic eye should be exposed to more complex and dynamic images and static images for
stronger eyes, a dichoptic game (Pivot) is designed to enhance binocularity. The images in this game are designed to be used with red-green anaglyphic 3D glasses to filter out green and red images for amblyopic and fellow eye, respectively. Red dynamic figures were seen only by the amblyopic eye and green static figures only by the fellow eye, without attenuating the contrast of the fellow eye (Fig. 2A).
During the design procedure, a formative evaluation was used and the feedback received from the experts helped remove the existing defects. The major changes made to the game were: adding to the number of constituent levels, adding to the variety of graphical forms applied in different levels and adding specific options to settings and statistics.
**Game features**
The game consists of 30 main levels initially locked except the first level and would be unlocked by playing and winning each level. The user must be able to place the red images in the green frames to move on to the next level and obtain bonus points to open the mini-games. Mini-games were in line with the original game system, and to add more attraction, each level had its properties and challenges.
Overall, there are 200 levels in the game, which can help the user stay in and add to the playing time and thus probably help to improve amblyopia via the binocular game.
Based on game design principles, to add to the game’s attraction and motivate users, various adaptations have
been made according to user’s speed, focus on the least frequent error, duration of daily play and community in gaming, and so on.
An essential part of the Pivot game is the settings and statistics pages. In the settings, you can set a background image and background color, select a color code for static and dynamic figures according to the hexadecimal color coding system, and adjust the color intensity of each ten levels, the ambient brightness, and the game speed (Fig. 2B).
By default, the speed of the game at each level is increased to stimulate the amblyopic eye. Moreover, adjusting the speed enables the physician to make changes to speed according to the patient player’s age, skills, and comments.
In the statistics, there is a graphical presentation of the time, date, and duration of the game, which allows the physician to obtain the required information about the user’s activity at a glance. This feature enables the physician to look for possible reasons for not playing the game and, consequently, recommend alternative treatments if necessary (Fig. 2C).
Moreover, there is an option for sending information in more detail, including demographic data recorded at the beginning of the game, game date, game duration, game level, the frequency of winning and losing in each level, and so on in the form of Excel file online provided to the server.
**Pilot implementation of Pivot**
In the first step, to evaluate the game in terms of the scenarios and appropriateness of goals, the game was evaluated by eye specialists. After applying the expert comments and getting their approval, in the second step, to get patients’ comments in the target group, the game was provided to a limited sample of patients meeting the inclusion criteria (n = 5). After a week, they were questioned about their experience. The focus was on their satisfaction with the game and their willingness to use the game. The feedbacks acquired showed their satisfaction with the game and willingness to continue. Moreover, their suggestions to improve the quality of the game were included.
**Outcome measures**
**Primary outcome**
Primary and secondary outcomes are measured at baseline and 2 weeks, 1 month, 2 months, and 3 months after the baseline.
The primary outcome is the change to visual acuity in the amblyopic eye from baseline to three months after randomization.
**Secondary outcomes**
- Change to visual acuity in the amblyopic eye after 2 weeks and 1, 2, and 3 months of randomization
– Change to two binocular outcomes, including 1) stereoacuity (using Randot Stereotest) and 2) interocular suppression (Worth Four-dot Test at 6 m and 33 cm)
– Compliance with treatment as at least 25% of the recommended time to play the game during the study (9 h) [5]
Data collection tools
Uncorrected and best corrected visual acuity (UCVA and BCVA) will be measured using the Early Treatment Diabetic Retinopathy Study (EDTRS) chart.
For suppression evaluation, the results of the Worth Four-dot Test at 6 m and 33 cm will be interpreted as binocular fusional response, suppression, or diplopia. Stereoacuity will be examined with Randot Stereotest (Stereo Optical Co, Chicago, Illinois, USA).
Objective refraction will be determined with an auto-refractometer (KR-1Auto Kerato-Refractometer, Topcon, Japan). Cycloplegic refraction will be performed in all patients at baseline and the last follow-up examination.
Compliance with treatment will be calculated by the time recorded in the Excel file in the server. The file contents will provide the user’s activity including days, minutes, and stages played, the frequency of winning and losing, etc.
Sample size
Based on the result of Kelly et al. study [19], visual acuity outcome and by using the following formula, at the 5% significance level (two-sided) with 80% power, 20 patients would be required per arm. Considering the 10% dropout rate during the study, our goal is to employ 22 patients in each group (a total number of forty-four participants).
\[
\begin{align*}
\text{Group 1: } & \quad \text{Mean}_1 \pm \text{SD}_1 = 0.15 \pm 0.08 \\
\text{Group 1: } & \quad \text{Mean}_2 \pm \text{SD}_2 = 0.07 \pm 0.08 \\
\end{align*}
\]
\[
n_1 = n_2 = \left[\left(\text{SD}_1\right)^2 + \left(\text{SD}_2\right)^2 \left(Z_{1-\alpha/2} + Z_{1-\beta}\right)^2\right] / (\text{Mean}_2 - \text{Mean}_1)^2
\]
Randomization
The randomization sequence will be created using www.randomization.com. The sequence of the generated random numbers will be transferred to sealed envelopes by an independent researcher (SA), not involved in the data collection or intervention. The corresponding envelopes will be opened only after the target participants signed the informed consent and completed all baseline assessments (JM). Therefore, eligible participants will be randomly divided into two groups to either the patch therapy or binocular game group.
Blinding
Due to the nature of the intervention, the use of red-green anaglyphic glasses does not allow patients to be blind. However, a certified optometrist as the outcome assessor (MN) will be unaware of patient grouping. Moreover, there was no possible access to the database of the results of patients for the certified examiner. Also, the data manager (who generated the randomization sequence, prepared envelopes and maintained a list of enrolled participants) and the data analyzer will be completely blinded to the control and intervention group characteristics.
Data monitoring and management
We formed a management committee (MER, SA, MN) to monitor data quality and also approve any decisions regarding this trial. Data entry and coding will be conducted by people other than the research team and checked by the management committee through range checks for data values.
Statistical analysis
User’s activity in game including days, minutes, and stages played will analyze based on following subgroups:
- Days played (< 30 days, 30 < days < 60, and > 60 days)
- Minutes played (< 500 min, 500 < < 1000, 1000 < < 1500, 1500 < < 2000, > 2000)
- Stages played (< 10, 10 < < 20, > 20)
Also, we collect stereopsis, spherical equivalent in involved and non-involved eyes, and fusional state as the baseline characteristics that have effect on visual acuity. We will use these characteristics in the final analysis.
The normal quantitative variables (based on the Kolmogorov-Smirnov normality test) will be described using mean and standard deviation. The remaining data will be described using the median and interquartile range. To test the mean difference of quantitative variables between the control and intervention groups, if the normality assumption is met, an independent-sample T test will be used. Otherwise, Mann-Whitney U test will be used to compare the data. To test the homogeneity of the qualitative variables in the two groups, the chi-squared test and Fisher’s exact test will be used at the p value of 0.05. Repeated measures ANOVA or non-parametric tests will be applied to continuous outcomes measured repeatedly. All statistical tests will be two-sided at the 5% significance level. Statistical analyses will be performed using SPSS version 22. The analysis will be performed on an intention to treat and also per protocol approaches. No interim analyses are planned.
**Results**
So far, the need analysis, design procedure, and pilot test of intervention are fulfilled. The recruitment of participants has not been completed and is scheduled to end in September 2021. The Pivot game is potentially useful to improve amblyopic-eye visual acuity outcomes.
**Discussion**
This RCT builds upon previous trials requiring further evaluation of the effectiveness of binocular game treatment for children with amblyopia, particularly mild to moderate anisometropic cases without a prior treatment except for refractive correction.
The newly designed binocular dichoptic game in this study consists of specific, well-defined characteristics, including the duration of playing the game, no need to attenuate the contrast of fellow eye to the level of the amblyopic eye, and change in difficulty of the game and mini-games. The results will also assess compliance by recording the time spent playing.
If the binocular playing game treatment works, it possibly manages to reduce the psychological pressure on families to patch their children’s amblyopic eye or may promote further compliance to the treatment or possibly associates with better binocular function outcomes. Moreover, playing binocular games allows the assessment of the user’s activity in the shortest time by the physician. If the results are favorable, children’s habits and interest in video games can be possibly used for healthcare purposes.
**Strengths and limitations**
- The newly designed binocular dichoptic video game in this study was developed in a structured way with the participation of a multi-disciplinary team
- The home-based game consists of specific, well-defined characteristics, including the duration of playing the game, change in difficulty of the game, and mini-games, which distinguishes it from previous similar games
- The design of this study (randomized controlled trial) tends to meet the highest level of evidence
- We cannot blind our patients due to the nature of the intervention, and it is one of our limitations in this study
**Ethics and dissemination**
The Ethics Committee of Mashhad University of Medical sciences’ approval date was February 28, 2018, with a reference code of IR.MUMS.fm.REC.1396.783. This trial is registered in Iran Trial Registrar under the registration number: IRCT20180217038768N and registration date 22 April 2019. The results will be disseminated in a peer-reviewed journal.
The dataset that supports the findings of this study is available from the corresponding author upon reasonable request. Personal information about potential and enrolled participants will be stored on a secure file server research drive at MUMS to ensure confidentiality protection before, during, and after the study.
**Ancillary and post-trial care**
In case of any probable adverse event (e.g., diplopia), the study treatment will cease and appropriate care will be provided by the physicians in the research team for compensation probable harm.
**Trial status**
The study is currently recruiting and enrolling participants according to version 2 of the protocol in July 2020. Recruitment began on May 31, 2019, and the expected recruitment end date will be the end of September 2021.
**Abbreviations**
- ETDRS: Early Treatment for Diabetic Retinopathy Study
- LogMAR: Logarithm of the minimum angle of resolution
- RCT: Randomized controlled trial
- BCVA: Best corrected visual acuity
- UCVA: Uncorrected visual acuity
**Supplementary Information**
The online version contains supplementary material available at https://doi.org/10.1186/s13063-021-05735-2.
**Acknowledgements**
We would like to thank the Vice-Chancellor for Research Affairs of Mashhad University of Medical Sciences for the financial support (grant number 960517). The authors are grateful to all study participants and team members as well as supporting staff.
**Role of the funding source**
The funders do not have any role in study design, data collection, analysis, decision to publish, interpretation, or preparation of the manuscript. The corresponding author will have full access to all the data in the study and will have final responsibility for the decision to submit for publication.
**Authors’ contributions**
MER and SA conceived the study idea and design. SA, MN, and JM designed the plan of RCT implementation. MER and JM conducted the RCT. SA and MN drafted the manuscript. All authors have been involved in critically revising the manuscript. All authors read and approved the final manuscript.
**Funding**
This work is funded by the research deputy of Mashhad University of Medical Sciences, Mashhad, Iran, under grant 960517.
**Availability of data and materials**
The datasets used and/or analyzed during the present study are available from the corresponding author (SA) upon reasonable request.
Declarations
Ethics approval and consent to participate
This study was approved by the Ethics Committee of Mashhad University of Medical Sciences and Medical School (Ethical code: IR.MUMS.FRM.REC.1396.783). Data collection began after the committee approved the study. All participants will sign the informed consent prior to enrollment in the baseline visit.
Consent for publication
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Author details
1. Fu Z, Hong H, Su Z, Lou B, Pan C-W, Liu H. Global prevalence of amblyopia and disease burden projections through 2040: a systematic review and meta-analysis. Br J Ophthalmol. 2019;104(8):1164–70. https://doi.org/10.1136/bjophthalmol-2019-314759.
2. Hashemi H, Pakzad R, Yekta A, Bostamzad P, Aghamirsalim M, Sardari S, et al. Global and regional estimates of prevalence of amblyopia: a systematic review and meta-analysis. Strabismus. 2016;26(4):186–93. https://doi.org/10.1016/j.strab.2016.02.002.
3. Xiao Q, Morgan IG, Ellwein LB, He M. Group RESiCS. Prevalence of amblyopia in school-aged children and variations by age, gender, and ethnicity in a multi-country refractive error study. Ophthalmology. 2015;122(9):1924–31. https://doi.org/10.1016/j.ophtha.2015.05.034.
4. Bloch D, Wick D. Differences between strabismic and anisometropic amblyopia: research findings and impact on management. Prog Optom. 1991;2:376–92.
5. Hess RF, Babu RJ, Clavagnier S, Black J, Bobier W, Thompson B. The i Pod binocular treatment of amblyopia in children. Arch Ophthalmol. 2002;120(3):268–78. https://doi.org/10.1001/archopht.120.3.268.
6. Hess RF, Thompson B. New insights into amblyopia: binocular therapy and occlusion therapy for amblyopia so hard? A qualitative study. Arch Dis Child. 2006;91(6):491–4. https://doi.org/10.1136/adc.2005.09373.
7. Stewart CE, Moseley MJ, Thompson B, Fine motor skills of children with amblyopia improve following binocular treatment. Invest Ophthalmol Vis Sci. 2016;57(11):4713–20. https://doi.org/10.1167/iovs.16-19797.
8. Repka MX, Beck RW, Holmes JM, Lazar EL, Birch EE, Kraker RT, et al. Prolonged periods of binocular stimulation can provide an effective treatment for childhood amblyopia. Invest Ophthalmol Vis Sci. 2012;53(2):817–24. https://doi.org/10.1167/iovs.11-8219.
9. Li S, Jost R, Morale S, Jost RM, Morale SE, De La Cruz A, Stager D, et al. A binocular iPad treatment for amblyopic children. Eye. 2014;28(10):1246–53. https://doi.org/10.1038/eye.2014.165.
10. Reeves M, Tailor VK, Anderson EJ, Bex PJ, Greenwood JA, Dahlmann-Noor A, et al. Binocular therapy for childhood amblyopia improves vision without breaking interocular suppression. Invest Ophthalmol Vis Sci. 2017;58(7): 3031–43. https://doi.org/10.1167/iovs.16-20913.
11. Bossi N, MacKeth D, Vivian A, Fakis A, Ash JM, et al. Randomised controlled trial of video clips and interactive games to improve vision in children with amblyopia using the i-BAT system. Br J Ophthalmol. 2016;100(1):151–6. https://doi.org/10.1136/bjophtha-2015-307798.
12. Holmes JM, Manny RE, Lazar EL, Birch EE, Kelly KR, Summers AI, et al. A randomized trial of binocular Dig Rush game treatment for amblyopia in children aged 7 to 12 years. Ophthalmology. 2019;126(3):456–66. https://doi.org/10.1016/j.ophtha.2018.10.032.
13. Yao J, Moon H-W, Qu X. Binocular game versus part-time patching for treatment of anisometropic amblyopia in Chinese children: a randomised clinical trial. Br J Ophthalmol. 2020;104(3):369–75. https://doi.org/10.1136/bjophtha-2018-313815.
14. Holmes JM, Manh VM, Lazar EL, Beck RW, Birch EE, Kraker RT, et al. Effect of a binocular iPad game versus part-time patching in children aged 5 to 12 years with amblyopia: a randomized clinical trial. JAMA Ophthalmol. 2016;134(12): 1391–400. https://doi.org/10.1001/jamaophthalmol.2016.4262.
15. Holmes JM, Manh VM, Lazar EL, Kraker RT, Wallace DK, Kulp MT, et al. Randomized trial of a binocular iPad game versus part-time patching in children aged 13 to 16 years with amblyopia. Am J Ophthalmol. 2018;186:104–15. https://doi.org/10.1016/j.ajo.2017.11.017.
16. Xu JP, He ZJ, Ooi TL. Effectively reducing sensory eye dominance with a push-pull perceptual learning protocol. Curr Biol. 2010;20(20):1864–8. https://doi.org/10.1016/j.cub.2010.09.043.
17. Li S, Reynaud A, Hess RF, Wang Y-Z, Jost JM, Morale SJ, et al. Dichoptic movie viewing treats childhood amblyopia. J Am Assoc Pediatr Ophthalmol Strabismus. 2015;19(5):401–5. https://doi.org/10.1016/j.jaapos.2015.08.003.
18. Xu CS, Chen JS, Adelman RA. Focus: addiction; video game use in the treatment of amblyopia: weighing the risks of addiction. Yale J Biol Med. 2015;88(3):309–17.
19. Chan A-W, Tetzlaff JM, Altman DG, Laupacis A, Götzsche PC, Krleža-Jerić K, et al. SPIRIT 2013 statement: defining standard protocol items for clinical trials. Br J Ophthalmol. 2014;98(9):1189–93. https://doi.org/10.1136/bjo.2013.308683.
20. Xu CS, Chen JS, Adelman RA. Focus: addiction; video game use in the treatment of amblyopia: weighing the risks of addiction. Yale J Biol Med. 2015;88(3):309–17.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3814,
1
],
[
3814,
9096,
2
],
[
9096,
13752,
3
],
[
13752,
15218,
4
],
[
15218,
17721,
5
],
[
17721,
22528,
6
],
[
22528,
27365,
7
],
[
27365,
32615,
8
]
]
} |
26b917bb7d60db02752836f6b56dc46c484b20aa | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 12,
"total-fallback-pages": 0,
"total-input-tokens": 42860,
"total-output-tokens": 13159,
"fieldofstudy": [
"Biology"
]
} | **Estimation of gene expression at isoform level from mRNA-Seq data by Bayesian hierarchical modeling**
*M. Bhattacharjee, Ravi Gupta, and R. V. Davuluri*
1 Department of Statistics, University of Pune, Pune, India
2 Department of Mathematics and Statistics, University of Hyderabad, Hyderabad, India
3 Center for Systems and Computational Biology, The Wistar Institute, Philadelphia, PA, USA
**INTRODUCTION**
Sequencing technology has advanced at a rapid rate in the past decade. The advent of massive parallel sequencing technologies, such as Illumina Genome Analyzer/Solexa, has revolutionized the genome-wide transcriptome studies leading to multiple applications. One such application is known as mRNA-Seq technology which provides a far more precise measurement of levels of transcripts and their isoforms than microarrays (Nagalakshmi et al., 2008). It is analogous to shotgun sequencing (Staden, 1979) used for whole genome, while it is being applied to transcripts and this method yields copy number of transcripts in a sample. In mRNA-Seq, RNA is isolated from sample; which is then reverse-transcribed to form cDNA, followed by fragmentation. New generation high-throughput sequencers enable to read from these fragments and these reads are then mapped to the genome of interest. Expression measure of any gene is thus available in digitized form through counts of reads mapped to that gene (See Figure 1 for an overview of mRNA-Seq experimental procedure).
With respect to reproducibility, this technique has been shown to be highly reproducible across technical replicates (Mortazavi et al., 2008). On the other hand if microarrays are used to measure expression of genes, it has been shown that the correlation between two replicates is in the range (0.5, 0.95; Draghici et al., 2006).
One of the main advantages of this data is in its quality. The eukaryotic genomes can be roughly thought to be constituted of RNA polymerase II transcribed and non-transcribed regions. In mRNA-Seq data the reads that map to the transcribed region (“exons” of a gene) represent the signal and background, whereas those coming from non-transcribed regions (“introns” or other inter-genic regions) can be assumed to be only background. Often the density of reads falling in the non-transcribed region would be of the order $10^{-4}$ whereas that for the coding region could be $10^{-1}$, which results in a signal to noise ratio of $10^3$.
Apart from being more reliable than the more popular microarrays, there are some very appealing advantages of this technique. For example, it actually produces sequence information of a transcriptome at a single nucleotide base-level, thus enabling us to determine splice junctions and alternative splicing events with high confidence. Often analysis of mRNA-Seq data does not attempt to quantify the expressions at isoform level. In this paper our objective would be use the mRNA-Seq data to infer expression at isoform level, where splicing patterns of a gene is assumed to be known.
A Bayesian latent variable based modeling framework is proposed here, where the parameterization enables us to infer at various levels. For example, expression variability of an isoform across different conditions; the model parameterization also allows us to carry out two-sample comparisons, e.g., using a Bayesian t-test, in addition simple presence or absence of an isoform can also be estimated by the use of the latent variables present in the model. This paper would carry out inference on isoform expression under different normalization techniques, since it has been recently shown that one of the most prominent sources of variation in differential call using mRNA-Seq data is the normalization method used. The statistical framework is developed for multiple isoforms and easily extends to reads mapping to multiple genes. This could be achieved by slight conceptual modifications in definitions of what we consider as a gene and what as an exon. Additionally proposed framework can be extended by appropriate modeling of the design matrix to infer about yet unknown novel transcripts. However such attempts should be made judiciously since the input date used in the proposed model does not use reads from splice junctions.
**Keywords:** mRNA-Seq, isoform expression, Bayesian latent variable modeling, multi-sample comparison, Bayesian t-test, spike-n-slab method
---
**ORIGINAL RESEARCH ARTICLE**
published: 27 November 2012
doi: 10.3389/fgene.2012.00239
www.frontiersin.org
November 2012 | Volume 3 | Article 239 | 1
STUDIES ON ALTERNATIVE SPLICING EVENT
The recent interest in studying alternately spliced gene expression stems from the fact that more than 90% of human genes have been reported to be alternately spliced (Pan et al., 2008; Wang et al., 2008). Furthermore, variants of transcripts produced by same gene have been shown to be involved in wide range of pathways and could perform distinctly different functions (see Davuluri et al., 2008). These phenomena have been under study for decades; however recent advances in biotechnology have enabled us to study these at a genome level simultaneously. With recent techniques such as mRNA-Seq, high-density DNA microarrays, and existing methods like Sanger sequencing of ESTs and cDNAs, genome-wide studies on many species have been reported (high-throughput techniques: Lister et al., 2008; Mortazavi et al., 2008; Pan et al., 2008; Sultan et al., 2008; Wang et al., 2008; Filichkin et al., 2009; and Sanger sequencing: Zhu et al., 2003; Iida et al., 2004; Alexandrov et al., 2006; Campbell et al., 2006; Wang and Brendel, 2006; Chen et al., 2007; Ner-Gaon et al., 2007).
SOURCES OF VARIATION
The statistical challenges in analyzing mRNA-Seq arise from many perspectives. While some sources of error are due to inherent problems with the technology, some are contributed at laboratory or experimental level while remaining are limitations of inference methods used.
As an example of technical limitations, mRNA-Seq data does suffer from non-uniformities, for example it is known to have biases toward certain base compositions (Dohm et al., 2008). Another such known limitation of this technique is that longer genes are more likely to be sequenced and also more likely to be declared differentially expressed (Oshlack and Wakefield, 2009). It has been demonstrated by the same authors that weighting the differential expression statistics by gene length can mitigate this effect.
There are known experimental errors as well during the creation of a sequencing library, seeding and preparation of the flow cell, and synthesis of the sequence reads (sequencing phase of the experiment). Depending on the data these errors may not pose any...
significant problem (see Bullard et al., 2010). They also observed that the well-known control lane used at the base-calling calibration procedure may not have any significant effect on differential expression call.
According to Bullard et al. (2010), the most prominent source of variation in differential call is the normalization or calibration method used on the mRNA-Seq data. In this paper we would carry out inference on isoform expression under different normalization techniques.
Calibration of data becomes essential to remove the effects of such courses of variations. In the following section we have provided detailed descriptions and discussions on the existing methods.
**EXPRESSION ANALYSIS**
Often analysis stops at finding alternately spliced variants only and further research into quantifying their expression is not carried out. In this paper our primary objective would be use the mRNA-Seq data to infer expression at transcript, i.e., isoform level, while utilizing all existing information on splice variants.
In mRNA-Seq data, each transcript is covered by numerous sequence reads. Since these reads are generally short in nature (approximately 35 bases in the data set considered here) these typically do not reflect expression of a single transcript but is often shared by several transcripts. Majority of these would be splice variants; however there are other possibilities as well, which we will address later.
Reads not only from exon regions but also from splice junctions (span two exons) could be available, but they too suffer from similar limitation of being shared by multiple isoforms. In case a junction is unique to a transcript then the reads from splice junction may be used as indicative of the expression. Otherwise in a complex structure of multiple exon sharing between different transcripts, the reads mapping to splice junctions would have to be modeled additionally if they are to be used for inferring expression of the isoforms.
As has been pointed out by Zheng and Chen (2009) comparison of expression at the individual transcript isoform level requires jointly consideration of all sequence reads belonging to the same gene. Thus often it may not be possible to detect isoform level expression in any obvious method and requires statistical techniques that would help us obtain individual isoform expression from this mixture along with estimates of uncertainty.
In the following sections we present review of relevant literature on expression estimation, followed by the description of the proposed Bayesian latent variable model and its application to a real data set.
**MATERIALS AND METHODS**
mRNA-Seq technique provides read count for genomic segments, known as exons. These can then be utilized in various manner depending on the question of interest, more often than not, these are summarized at gene level. However we prefer to quantify them at the level of disjoint exons or parts thereof. The motivation is as followed by Jiang and Wong (2009), wherein isoforms for any gene share exons as a whole. Since in reality this is often not true we achieve this by slightly modifying the known exon constitution of that gene. This is achieved by considering disjoint parts of exons, which are shared and otherwise, individually as pseudo-exons.
**DATA CALIBRATION**
Since number of reads for any genomic region would naturally depend on the length of that region it is a common practice to normalize it using the length information. Quite often it is represented as reads per kilobase per million (RPKM) mapped reads (Mortazavi et al., 2008). Thus the two most influential factors behind any expression summary from the mRNA-Seq data is
1. \(n_j\): total number of mappable reads in the sample for the \(j\)-th region
2. \(l_j\): length of that region,
Then
\[
\text{RPKM}_i = \frac{n_i}{\sum_j n_j} \times 10^9
\]
Although the RPKM thus obtained can then be analyzed similar to methods used for microarrays, in-depth look into the sources of variations is required before inference.
Often in mRNA-Seq data, different lanes of a flowcell would represent different samples (could be biological replicates or could be distinct treatment samples). Each sample would produce different total read counts depending on sequencing depths. Thus, as described above, the common approach is to adjust individual expression counts for each sample by the total count for that sample: e.g., RPKM as described above or a hypergeometric model (Marioni et al., 2008), Bullard et al. (2010) have pointed out that, for the data considered by them, global normalization is heavily affected by a relatively small proportion of highly expressed genes. Since these are often not the same genes across samples this potentially could lead to biased estimates of differential expression over the different samples considered.
A common phenomenon for microarray data analysis lack of robustness of inference (on say differential expression call, or classification) due to change in analysis techniques in previous step of data analysis. Quite often a moderate change in algorithm for normalization can dramatically alter biological conclusions (Bhattacharjee et al., 2004). It has been shown by Bullard et al. (2010) that mRNA-Seq data is no exception from this problem and hence we feel needs extended study before robust conclusions of expressions can be made based on this data.
In literature, so far, the main methods proposed for calibration are
1. Global normalization,
2. Use of house-keeping genes, which is a familiar concept for micro array data,
3. Quantile normalization, which is also a well established method used for other high-throughput data (Irizarry et al., 2003).
It has been reported that the standard total count normalization results in low variation across samples and other technical sources of variation. However does not produce significant gain in terms
of differential expression call, when compared with microarray data (Bullard et al., 2010).
The use of house-keeping genes has a sound biological reasoning. However in practice obtaining a reliable set which would provide consistent expression across samples and replicates might prove difficult. Regarding quantile-based normalization, Bullard et al. (2010) observe that for their data that this technique yields the most robust data without introducing additional noise. Thus for our purpose we would explore global and quantile-based normalization of our data.
MODELING OF EXPRESSION IN CLASSICAL FRAMEWORK
The initial efforts of expression modeling have been to use the counts of reads mapped as the variable itself. This has lead to use of discrete distributions on most of the existing literature. However one should note that with complex normalization techniques being applied to mRNA-Seq data modeling the calibrated data with discrete distributions like Poisson or Negative Binomial may not be feasible. Nevertheless we provide a review of these models and corresponding inference procedures in the following.
Poisson modeling for expression analysis
Let \( n_g \) be the count of reads mapped to gene \( g \). Assuming that the reads are sampled independently and with replacement \( n_g \) follows a Poisson distribution (Sultan et al., 2008), i.e.,
\[
n_g \sim \text{Poisson}(\lambda_g),
\]
where \( \lambda_g \) is the mapping rate and therefore large \( \lambda_g \) means high expression. This can be alternately arrived at by assuming a multinomial model for counts \( n \) of collection of all genes (or parts thereof). If \( p_g \) is the probability of each read being mapped to a gene \( g \) then since \( n \) is large and \( p_g \) is small, approximately \( n_g \sim \text{Poisson}(\lambda_g) \) with \( \lambda_g = n p_g \). It has been shown that more than 95% counts can be modeled well by this Poisson model (Marioni et al., 2008). However, simple Poisson model cannot explain all the variance. One suggestion is to assign a Gamma prior to \( \lambda \). This would lead to the Negative binomial model and there it might be harder to interpret the two parameters.
Poisson modeling for joint calibration and expression analysis
By introducing a normalizing factor \( C_g \) in the Poisson model we can achieve calibration at global level. Then
\[
n_g \sim \text{Poisson}(C_g \lambda_g^*),
\]
where \( \lambda_g^* \) : normalized expression level, \( C_g \): similar to the RPKM described earlier. For differential expression analysis we can treat \( C_g \) as known, and interpret \( \lambda_g^* \).
Generalized linear model for joint calibration and expression analysis
This model was recently proposed by Bullard et al. (2010). Therein a generalized linear model (GLM) framework for the expression modeling was carried out jointly in presence of parameters quantifying other systematic sources of variations, like total read count difference across samples. The Poisson-log GLM proposed by them is as follows:
\[
\log(\mathbb{E}(n_g \mid d_i)) = \log d_i + \lambda_{a(i)g} + \theta_g;
\]
where the natural logarithm of the expected value of the read count \( n_g \) for the \( g \)-th gene in the \( i \)-th sample is modeled as a linear function of the gene’s expression level \( \lambda_{a(i)g} \) for the biological condition \( a(i) \) as reflected in sample \( i \) plus an offset \( (\log d_i) \) and possibly other technical effects \((\theta_g)\). If no technical replicates are carried out, as is often the case, \( a(i) \) will be simply \( i \).
Testing for differential expression
Under the Poisson model, often data is further transformed, e.g., Log-Arcsine (Mortazavi et al., 2008). This is followed by the testing methods developed for microarray methods, e.g., t-test, moderated t-test, to identify significant expression changes. Note that the arcsine-root transformation is suggested for variance stabilization of the per-gene read proportions within each sample (Marioni et al., 2008).
To test \( H_0: \lambda_{g1} = \lambda_{g2} \) following are some examples of existing testing procedures that can be directly applied to this type of data. Following and extending notation introduced earlier,
1. Binomial Test (Ji et al., 2008)
\[
P\left( n_{g1} \mid n_{g1} + n_{g2} \right) \sim \text{Binomial} \left( n_{g1} + n_{g2} , \frac{C_{g1}}{C_{g1} + C_{g2}} \right)
\]
2. Negative Binomial Test (Audic and Claverie, 1997)
\[
P\left( n_{g2} \mid n_{g1} \right) \sim \text{Negative Binomial} \left( n_{g1} + 1 , \frac{C_{g1}}{C_{g1} + C_{g2}} \right)
\]
3. Chi-square goodness-of-fit test for Poisson counts (Mortazavi et al., 2008). Asymptotically,
\[
\sum_{i=1,2} \frac{(m_{g_i} - n_{g_i})^2}{m_{g_i}} \sim \chi_1^2,
\]
where \( m_{g_i} = \frac{C_g(n_{g1} + n_{g2})}{C_{g1} + C_{g2}} \) are the expected observation under null.
For the GLM, three types of methods for differential expression inference have been proposed, viz. Fisher’s exact test statistic, likelihood ratio statistics based on a GLM, and t-statistics based on estimated parameters of the same GLM. All the methods have their merits and demerits, for example, distributions for the GLM-based statistics are derived under asymptotic theory; therefore, might be affected by small numbers of input samples or low counts (depending on which parameter is being tested). On the other hand although Fisher’s exact test does not make any assumptions on the sample size; it only adjusts for global experimental effects, unlike the likelihood ratio statistics adjust for general experimental effects as well as sample covariates. t-statistics on the other hand is severely affected by zero count in even one sample and is unable to detect differential expression in some of the obvious cases.
Adjusting for gene length bias in differential expression
It has been noted by several groups that longer genes contribute more to mapped sequence reads than shorter genes with similar expression. More importantly this may not be removed by scaling with length like RPKM (Mortazavi et al., 2008).
We investigated this feature for our data as following. We distributed the exons, irrespective of their gene, according to lengths into ten approximately equal size bins and calculated average RPKM for each of these bins. The pattern as reported by others seemed to be present for our data as well although not strongly (results not shown). There was a sharp drop at the last bin, which could be due to the fact that this bin contains exceptionally large exons, however there is always a physical limitation of number of reads coming from these exons, thus bringing down the RPKMs.
The effect of length bias has been noted on differential expression call also (Oshlack and Wakefield, 2009; Bullard et al., 2010). One possible explanation is the following. For a particular gene reads coming from these exons, thus bringing down the RPKMs. However, there is always a physical limitation of number of reads coming from these exons, thus bringing down the RPKMs.
The model presented here requires known structure of the splicing patterns of a gene. Although with some advanced modeling we are able to infer when the exiting knowledge of transcription variation is in question (thus leading to randomized-design matrix). Also by extending the design matrix would enable us to infer about yet unknown novel transripts.
Additional advantage of the proposed model is that it is developed for multiple samples comparison unlike Zheng and Chen (2009) and also allows two-sample comparisons. Therefore we do not use latent variables for differential call instead we use latent variable as expression indicator. Thus for a particular gene and a specific isoform thereof we introduce at each sample level a latent variable that indicates whether the isoform is expressed in that particular sample or not.
This technique has twofold advantages over existing methods, at the level of model identifiability and also at the level of pairwise testing. Firstly, when genome level expressions of genes (and exons in this case) are explored the majority of them are not expected to express at every situation. Thus to represent the underlying biology realistically requires the distribution of the expression parameters to be able to take values very close to zero as well as large values. This often poses computational problems and one way to overcome this is to use the "spike-n-slab" type of model (Bhattacharjee and Sillanpää, 2005). By using a latent variable indicating whether an isoform is expressed or not produces the desired biological constraint on the model without introducing computational complexity. The advantage of these indicators for pairwise testing will be discussed later.
PROPOSED MODEL
For each exon \(i\) of a gene \(g\), we model it’s expression (as measured by normalized reads there in) as a linear combination of expressions of the isoforms of that gene which share this particular exon. Thus,
\[
y_{gik} = \mathbf{X}_{gik} \mathbf{\beta}_{gik} + \epsilon_{gik};
\]
where \(y_{gik}\) normalized reads for the \(i\)-th exon in the \(g\)-th gene for the \(l\)-th sample, \(L_{gik}\) latent indicator variable, which is one if the \(k\)-th isoform of the \(g\)-th gene is expressed in the \(l\)-th sample and it is zero otherwise, \(\mathbf{X}_{gik}\) latent variable measuring expression of the \(k\)-th isoform of the \(g\)-th gene in the \(l\)-th sample, \(X_{gik}\) the zero-one matrix indicating whether exon \(i\) is known to form part of the \(k\)-th isoform of the \(g\)-th gene, \(\epsilon_{gik}\); residual error term for the \(i\)-th exon of the \(g\)-th gene in the \(l\)-th sample, \(g\): ranges from 1 to 7737, the number of genes covered in the data, \((7737\) genes from chromosome-1 of the Arabidopsis data), \(i\): ranges over the number of exons for the \(g\)-th gene in the data, these exons could be biologically whole exons or disjoint parts thereof, \((\text{for the current data it ranges from 1 to } 73)\), \(k\): takes values 1 to the known number of isoforms for gene \(g\) (for our data values are between 1 and 10), \(l\): represents different samples (for the data used for illustration there are five distinct treatment conditions with one sample each and a control sample to contrast with are available).
After careful exploration we chose multivariate Normal distribution for the errors; however for individual error a Gamma model
might have been more effective in overcoming some biases in the data. Our choice of distribution was for the ease of computation.
In regard to the choice of error variance, our model assumes a sample \( \times \) gene parameterization. Consider a gene with single isoform with \( m \) exons. Since the gene is known to have only one transcript thus if this transcript it expressed in a sample then all of its constituent exons will be expressed and should be expressed in similar amount. Thus any variation in the measurements \( y_{gkl} \) for \( t = 1, \ldots, m \) would be purely noise in the measurement of expression of the gene at specific sample level. Thus our error modeling is clearly intuitive for single isoform genes which is majority of genes for this particular data (since 6510 genes of the 7737 genes have single isoform in this particular data). Now let’s consider the situation of a gene with multiple known isoforms. There, if a residual error is high that could not only indicate noise in measurement but also incomplete knowledge in all possible isoform formation of that gene, that is we could very well be missing a few isoforms and thus are unable to explain the observed expression measurement for some exons. Thus an assumption of gene \( \times \) sample level variation would be realistic for such genes also.
This helps us in defining the likelihood of the data and to complete the hierarchical structure of the Bayesian model we restrict to known conjugate distributions as priors. In standard notation we define our model for the vector of expression of all exons per-gene \( g \) at specific sample level. Thus our error modeling is clearly intuitive for single isoform genes which is majority of genes for this particular data (since 6510 genes of the 7737 genes have single isoform in this particular data). Now let’s consider the situation of a gene with multiple known isoforms. There, if a residual error is high that could not only indicate noise in measurement but also incomplete knowledge in all possible isoform formation of that gene, that is we could very well be missing a few isoforms and thus are unable to explain the observed expression measurement for some exons. Thus an assumption of gene \( \times \) sample level variation would be realistic for such genes also.
The results of variable selection and regularization against over-fitting.
Inference
One of the main advantages of our model is that it has been developed for multiple conditions, however the parameterization enables us to infer at various level. With availability of data over a range of conditions differential expression call may be one of the many possible questions of interest regarding an isoform’s behavior. Thus instead of including a differential expression call indicator we model expression variability of an isoform across different conditions through the hyper-parameters \( \tau_{gkl} \) in the hierarchical model. This setup allows us to identifying isoforms that are variable across different conditions. The profiles and clusters of isoforms can be studied using the \( \beta_{gkl} \) parameters.
Moreover two-sample comparisons can be carried out with the help of Bayesian \( t \)-test. For Bayesian \( t \)-test, the posterior distribution of a \( t \)-statistic like random variable defined as following is derived for each pairwise comparison of interest. Differential call is made according to whether or not the 95% symmetric posterior probability interval (PPI) contains zero.
As has been reported by Bullard et al. (2010) classical \( t \)-test might suffer from inflated standard error estimate of a gene/isoform is not expressed in one of the two conditions. In our modeling framework we easily overcome this issue by utilizing the latent expression indicator variable. If the two latent variables indicate with high degree of confidence that an isoform is expressed in one condition and not in the other, then irrespective of expression amount it can be safely declared differentially expressed. This is a known knowledge in biology, that in order to play a significant role in a biological process a transcript may not have to be expressed in high degree but differently with high confidence.
Note that these latent indicator variables also serve the purpose of variable selection and regularization against over-fitting.
RESULTS
For illustration purposes we used the RNA-Seq data on *Arabidopsis thaliana* by Filichkin et al. (2009). High-throughput sequencing using the Illumina 1G platform (reviewed in Quail et al., 2008; Shendure and Ji, 2008) were carried out to capture transcriptomic expression information from a range of *Arabidopsis* samples. *Arabidopsis* tissues at different developmental stages and time points of the diurnal cycle were pooled to provide a broad view...
of expression within a specific condition. For five abiotic stress treatment conditions and additionally a control wild-type (WT) sample, RNA-Seq libraries were prepared and sequenced individually. The five conditions are Cold Stress (CS), Drought Stress (DS), heat stress (HS), High-light Stress (LS), and Salt Stress (SS).
The dataset was downloaded from NCBI GEO website (accession ID = SRP000935). Pooled data contains approximately 271 million. The gene information and other necessary resources of Arabidopsis genome were downloaded from TAIR9 database release available at http://www.arabidopsis.org. In house programs were made to extract isoform level information. Bowtie program was used to align the reads to genome and isoforms. We used 30-bases with up to 2-bases mismatch and unique match options to align the reads. The results presented in the subsequent sections will be based on data from Arabidopsis chromosome-1 only for limitation of space.
For technologies like RNA-Seq, sequencing bias toward 3′ end is well-known. However for our data this effect is visible only moderately (see Figure 2 below). This could be due to the reason that we have utilized pooled data from two different primer techniques used in the original experiment. As reported by Filichkin et al. (2009) the 3′ bias is visible in the data where full-length (FL) enriched cDNA libraries were used, however coverage by the randomly primed library was more evenly distributed. Thus combining these two types of data probably have resulted in reduction of this effect on overall data.
As can be seen from the following Figure 3A the data normalized using global calibration produces comparable quantiles for most conditions except for condition HS. We investigated the average expression within each percentile and observed that average expressions across all conditions (except HS) are comparable for all the percentiles using the global calibration. It appears under the HS condition there is a larger proportion of zero tag counts and also very high expression of a small proportion of exons. However in global calibration the total expression has been equated across all samples and thus it has forced all remaining percentiles for
---
**Figure 2** Percentile distributions for first, second, second last, and last exons from transcription start site for single isoform genes. Note that all the figures have been zoom-in for better visibility thus last percentiles have been omitted. (A) Percentiles of lengths of exons. (B) Percentiles of tag counts mapped to exons for wild-type plants. (C) Percentiles of RPKM, i.e., tag counts adjusted by length, for exons for wild-type plants.
**Figure 3** Percentiles of expression data using global and percentile normalization for all six conditions, viz. WT, Wild-type-control; CS, Cold Stress; DS, Draught Stress; HS, Heat Stress; LS, High-light Stress; SS, Salt Stress. (A) Percentiles of expression with global normalization. (B) Percentiles of expression with percentile normalization.
this condition (viz. HS) to be scaled down compared to others. We employ a percentile adjustment using the 10-th to 90-th percentiles of the globally normalized data and the resulting normalized data yielded comparable percentiles for all six conditions (Figure 3B).
However our quantile normalization would differ from that proposed by Bullard et al. (2010) in a number of ways. Firstly, we use ten percentiles and not quantiles. Secondly motivated by what we observe for the stress condition HS we believe that the total need not be scaled after doing the percentile normalization as has been suggested by them. It appears that for the HS condition up to the 90-th percentile data is comparable to other condition. So the exception in the top percentile is probably not an artifact but reflects the plant’s behavior under this stress condition. Thus we believe that it might be inappropriate to scale it down artificially.
Normalizing with house-keeping gene(s) could also be a possibility; however it requires relevant biological information which is not available in our case. Thus we proceed with comparing results based on two different types of normalization, global correction and percentile normalization.
The normalizations continue to have effects on isoform level expressions. In Figure 4 we present percentiles of isoform expressions, estimated by posterior means, under different conditions using the two normalization techniques. The last percentiles, which are large due to few handful isoforms with large expressions, are omitted from plots for better clarity. We also present the average expression within each of these percentiles for the different conditions and different normalization techniques.
Note that in our method of defining expression (at gene or isoform level) we utilize expression at exon-level. For this particular data, 74.2% of the genes (and 78.7% of isoforms) consisted of multiple exons, ranging from 2 to 73 exons for a single transcript. Amongst all isoforms only 12.9% consisted of 50% or more long exons, 28.5% have 50% or more short exons, and remaining had varied lengths of exons. We used 5-th and 95-th percentiles from the exon-length distribution to define short and long exons respectively. Thus in differential calling it is unlikely to have a systematic bias toward any length for these transcripts. However due to RPKM bias toward length the standard error will be affected by the underlying variation in length, weakening our power of detection. Another point to note is that for Arabidopsis, based on present knowledge, nearly 84% of the genes (on chromosome-1) are single isoform genes. However 69.4% of these genes have more than one exon.
Highly expressed isoforms and low/no-expression isoforms for each stress condition and for control sample were identified. A pairwise check for commonality of these when compared to other conditions were carried out (Table 1). We define low/no-expression if the estimated 95-th percentile of the
The entries above diagonal shaded in gray color are low-expressed genes and those below diagonal (not shaded) are highly expressed genes, in both conditions.
(Alternate) We use the PPI of the t-statistics and define those as up or down regulated if the 95% PPI of the t-statistic is in the upper or lower 10 percentile of all (mean) t-statistics thus calculated (Table 2). This method then uses the uncertainty in estimating the t-values and would be applicable to all isoforms irrespective of number of constituent exons. The two methods yield different but non-contradictory results (results not shown).
It is apparent from the results presented so far that the isoform expressions vary under different stress conditions as expected. However many of the genes are not single isoform genes thus we need to explore further whether the observed variability is only at the gene level across conditions or whether alternately spliced transcripts behave differently under different conditions.
Using overall variability parameters we estimate that approximately 4.06% of isoforms with globally normalized data and 2.74% isoforms with quantile normalized data exhibit varied expression under different stress conditions. To infer these we observe that the variability parameters estimated follow approximately Gamma distributions (Figure 5).
Of the total 7737 genes analyzed, 1227 genes have multiple isoforms. Of which 75 genes have at least one isoform not variable (i.e., overall variability parameter estimate is less than the 35-th percentile) and another moderately variable (i.e., variability is over the 65-th percentile). More interestingly 195 genes have at least one highly variable isoform (with variability above than the 90-th percentile) and another in the lower half. Thus for same gene not all isoforms behave in similar manner across all sample conditions. The above findings are based on globally normalized data and similar summary under quantile normalization are respectively 57 and 175 genes.
We compared our findings with two existing methods, namely modeling the exon-expression data with GLMs and secondly by using splice junction data only to infer isoform expression using Bayesian models.
For GLM we used a model similar to one proposed by Bullard et al. (2010). However due to the change in nature of data from discrete to continuous, induced due to normalization the error distribution for modeling in GLM was chosen to be Gaussian. The standard asymptotic theory based inference was carried out for this model. The mean estimates of the transcript level expression (for single transcript genes) were compared and were found
Table 2 | Up and down regulated isoforms for each stress conditions when compared with WT, using alternate definitions of differential expression.
| Stress condition | Based on 95% PPI of t-statistics | Based on posterior mean of t-statistic and appropriate t-critical value |
|------------------|----------------------------------|---------------------------------------------------------------------|
| | Down | Up |
| Normalization | Global Quantile | Global Quantile |
| | Down | Up |
| CS | 100 | 91 | 71 | 39 | 54 | 54 | 42 |
| DS | 90 | 105 | 152 | 129 | 38 | 56 | 102 | 80 |
| HS | 467 | 319 | 55 | 94 | 571 | 364 | 27 | 61 |
| LS | 119 | 134 | 116 | 96 | 84 | 98 | 63 | 51 |
| SS | 78 | 84 | 156 | 149 | 73 | 73 | 51 | 51 |
FIGURE 6 | Comparison of estimates from proposed Bayesian model (Y-axis) and GLM (X-axis). (A) Presents scatter plot of estimated expression levels for single transcript genes under the two methods while (B) presents the estimated standard error in these expression measurements from the same data under the same models.
comparable between the proposed Bayesian method and GLM (Figure 6A). To judge the quality of inference it is not just the point estimates that should be checked but also the confidence on these estimates. The estimated standard error around point estimates provides a well-known means of assessing the quality of the estimated value. For this we firstly note that for single exon genes this can’t be estimated under existing procedures. However Bayesian paradigm does allow posterior inference with singleton data as well and hence an estimate for standard error for such genes too can be obtained. For other transcripts with multiple exons the Bayesian model consistently produces smaller standard error and thus outperforms the standard techniques like GLM (Figure 6B).
The Bayesian modeling framework proposed for exon data can be readily used for splice junction data too. Note that splice junctions would be typically short in length and as discussed earlier mRNA-Seq technology produces number of reads biased toward longer genomic regions. This is expected to cause the reads to have inflated amount zero read counts and underrepresentation the true expression from these regions. Our preliminary exploration of the splice junction data confirmed this hypothesis. We further compared the distributional behavior of the tag counts from the splice junctions to that based on reads from exons and found them to be comparable to those from exons of similar lengths.
For further analysis of the splice junction data, it was first normalized using a similar percentile normalization technique described earlier and then modeled using the proposed Bayesian modeling framework. Figure 7 presents comparison of estimates obtained using this model based on data from exons with those from based on data from splice junctions. The two key outcome of the proposed model is assessing the probability of a transcript being expressed and the corresponding expression (posterior estimates in Figures 7A,B respectively). In addition to plotting the pairs of estimates, regression lines were fitted and corresponding equations (along with $R^2$) were presented in the figures to provide an assessment of linear association between the two sets of estimates.
As was noted from the raw data itself, the splice junction data underestimates transcript presence and thus the estimated probabilities based on junction data are much lower than those obtained using exon data. This relation is curvilinear in nature and possibly is “S-shaped” as has been observed elsewhere, e.g., for microarrays. The $R^2$ measure reflects this low degree of linearity...
between the two estimates. The expression estimates are affected also but the degree of linearity is much improved (with a reasonably high $R^2$), although still continues to under-estimate. We also explored the inherent variability in splice junction data and found it to be higher compared to exon-based-data. This can be assessed easily by comparing measurements on exons and junctions from genes with single transcript only (results not shown).
DISCUSSION
We have presented a Bayesian framework based on mRNA-Seq data to infer expression at isoform level. One of the main advantages of our model is that it has been developed for multiple conditions, however the parameterization enables us to infer at various levels. Although complex experimental scenarios are not uncommon, as can be seen from the real data sued here for illustration, a lot of the focus on methodological development for mRNA-Seq data continues to focus on differential expression (e.g., Wang et al., 2010; Kadota et al., 2012, etc). The exclusion of differentially expressed genes for better calibration is similar to exclusion of the extreme tails in percentile based normalization method used here. The main difference being in presence of more than two condition differential expression may not be a useful criteria however percentiles would still provide a measure of extreme to be excluded from calibration.
The modeling setup allows inference at gene level (in terms of expression or its variability), at transcript level, comparison across samples, comparison between samples, etc.
We have utilized latent indicator variable for expression apart from additional parameters to quantify such expression. This differs significantly from existing approaches, where latent variables are used for differential call. Our method has twofold advantages over existing methods. First advantage is at the level of model identifiability, which is well-known for this type of spike-n-slab models. These variables also serve the purpose of variable selection and regularization against over-fitting.
Secondly, our method has an advantage at the level of pairwise testing. If an isoform is not expressed in one of the samples then this would typically un-stabilize a formal $t$-statistic, however can be captured here using the latent expression indicators. This is a known in biology, that in order to play a significant role in a biological process a transcript may not have to be expressed in high degree but differently with high confidence. Thus if for an isoform the latent expression variables indicate with high confidence expression in one condition and not expressed in another the corresponding isoform can be declared differential, without the help of a $t$-statistics.
We would like to reiterate that focus of this paper has been in developing modeling and inference procedure for expression measurement. Thus several other relevant aspect of mRNA-Seq data has not been addressed. Some of these are, how to use splice junction reads, what to do with multi-reads, how to make better use of unmapped reads. From current literature we see that a considerable effort is being put on these issues [Top-Hat and Cufflink by Trapnell et al. (2012), FX by Hong et al. (2012)], however comparatively less on what to do once we have satisfactorily mapped the reads and have been able to derive the alternate splicing structure.
Our aim has been to make the readers aware that, firstly there should be a modeling setup that allows complex experimental data going beyond differential expression to be analyzed, secondly even for simplest of inference, like expression of a single transcript gene; there could be significant effect of normalization procedure. We believe the proposed model and inference setup here addresses these adequately.
For the particular data used for illustration here, we noted that there could be significant effect of normalization, if not for all, at least in part of the data. For this data an in-depth presentation of the differential and variable expression could not be provided here. However it was known that alternate splicing plays a very important role in functioning of Arabidopsis and for many genes multiple splice variants were found to be active under different conditions.
ACKNOWLEDGMENTS
The work in the Davuluri laboratory is supported by the Commonwealth Universal Research Enhancement (CURE) Research Program, Department of Health, Pennsylvania. R. V. Davuluri holds a Philadelphia Healthcare Trust Endowed Chair Position; research in his laboratory is partially supported by the Philadelphia Healthcare Trust. The use of computational resources in the Centre for Systems and Computational Biology and Bioinformatics Facility of Wistar Cancer Centre (grant # P30 CA010815) are gratefully acknowledged.
REFERENCES
Alexandrov, N. N., Troukhman, M. E., Brover, V. V., Tatarinova, T., Flavel, R. B., and Feldmann, K. A. (2006). Features of Arabidopsis genes and genome discovered using full-length cDNAs. Plant Mol. Biol. 60, 69–85.
Audic, S., and Claverie, J. (1997). The significance of digital gene expression profiles. Genome Res. 7, 986–995.
Bhattacharjee, M., Pritchard, C. C., Nelson, P. S., and Arjas, E. (2004). Bayesian integrated functional analysis of microarray data. Bioinformatics 20, 2943–2953.
Bhattacharjee, M., and Sillanpää, M. J. (2005). Bayesian association-based fine mapping in small chromosomal segments. Genetics 169, 427–439.
Bullard, J. H., Purdom, E., Hansen, K. D., and Dudoit, S. (2010). Evaluation of statistical methods for normalization and differential expression in mRNA-Seq experiments. BMC Bioinformatics 11, 94. doi:10.1186/1471-2105-11-94
Campbell, M. A., Haas, B. J., Hamilt- on, J. P., Mount, S. M., and Buell, C. R. (2006). Computational analysis of alternative splicing in rice and comparative analyses with Arabidopsis. BMC Genomics 7, 327. doi:10.1186/1471-2164-7-327
Chen, F. C., Wang, S. S., Chaw, S. M., Huang, Y. T., and Chuan, T. I. (2007). Plant gene and alternatively spliced variant annotator. A plant genome annotation pipeline for rice and comparative analyses with Arabidopsis. BMC Bioinformatics 8, 2943–2955.
Chen, F. C., Wang, S. S., Chaw, S. M., Huang, Y. T., and Chuan, T. I. (2007). Plant gene and alternatively spliced variant annotator. A plant genome annotation pipeline for rice and comparative analyses with Arabidopsis. BMC Bioinformatics 8, 2943–2955.
Faulkner, G. J., Forrest, A. R. R., Chalk, A. M., Schroder, K., Hayashizaki, Y., Carninci, P., et al. (2008). A rescue strategy for multimappping short sequence tags refines surveys of transcriptional activity by CAGE. Genomics 91, 281–288.
Filichkin, S. A., Priest, H. D., Givan, S. A., Shen, R., Bryant, D. W., Fox, S. E., et al. (2009). Genome-wide mapping of alternative splicing in Arabidopsis thaliana. Genome Res. doi:10.1101/gr.39302.109
Hong, D., Rhee, A., Park, S. S., Lee, J., Ju, Y. S., Kim, S., et al. (2012). FX: an RNA-Seq analysis tool on the cloud. Bioinformatics 28, 721–723.
Iida, K., Seki, M., Sakurai, T., Satou, M., Akiyama, K., Toyoda, T., et al. (2004). Genome-wide analysis of alternative pre-mRNA splicing in Arabidopsis thaliana based on full-length cDNA sequences. Nucleic Acids Res. 32, 5096–5103.
Irrizarry, R. A., Hobbs, B., Collin, F., Beazer-Barclay, Y. D., Antonellis, K. J., Scherf, U., et al. (2003). Explorat ion, normalization, and summarization of high density oligonucl eotide array probe level data. Bio statistics 4, 249–264.
Ji, H., Jiang, H., Ma, W., Johnson, D. S., Myers, R. M., and Wong, W. H. (2008). Statistical inferences for isoform expression in RNA-Seq. Bioinformatics 25, 1026–1032.
Kadota, K., Nishiyama, T., and Shimizu, K. (2012). A normalization strat egy for comparing tag count data. Algorit hms Mol. Biol. 7, 5.
Li, B., Ruotti, V., Stewart, R. M., Thoms son, J. A., and Dewey, C. N. (2010). RNA-Seq gene expression estimation with read mapping uncertainty. Bioinformatics 26, 493–500.
Lister, R., O’Malley, R. C., Tonti Filippini, J., Gregory, D. B., Berry, C. C., Millar, A. H., et al. (2008). Highly integrated single-base resolution maps of the epigenome in Arabidopsis. Cell 133, 523–536.
Marioni, J. C., Mason, C. E., Mane, S. M., Stephens, M., and Gilad, Y. (2008). RNA-seq: an assessment of technical reproducibility and comparison with gene expression arrays. Genome Res. 18, 1509–1517.
Mortazavi, A., Williams, B. A., McCue, K., Schaeffer, L., and Wold, B. (2008). Mapping and quantifying mammalian transcriptomes by RNA-Seq. Nat. Methods 5, 621–628.
Nagalakshmi, U., Wang, Z., Wan, K., Wang, C., Raha, D., Gerstein, M., et al. (2008). The transcriptional landscape of the yeast genome defined by RNA sequencing. Science 320, 1344–1349.
Net-Gaon, H., Leviatan, N., Rubin, E., and Fluhr, R. (2007). Comparative cross-species alternative splicing in plants. Plant Physiol. 144, 1632–1641.
Oshlack, A., and Wakefield, M. J. (2009). Transcript length bias in RNA-Seq data confounds systems biology. Biol. Direct 4, 14.
Pan, Q., Shai, O., Lee, I. J., Frey, B. J., and Blencowe, B. J. (2008). Deep surveying of alternative splicing complexity in the human transcriptome by high-throughput sequencing. Nat. Genet. 40, 1413–1415.
Quail, M. A., Kozarewa, I., Smith, F., Scally, A., Stephens, P. J., Durbin, R., et al. (2008). A large genome center’s improvements to the illumina sequencing system. Nat. Methods 5, 1005–1010.
Shendure, J., and Ji, H. (2008). Next-generation DNA sequencing. Nat. Biotechnol. 26, 1135–1145.
Spiegelhalter, D. J., Thomas, A., and Best, N. G. (1999). WinBUGS Version 1.2 User Manual. Cambridge: MRC Biostatistics Unit.
Staden, R. (1979). A strategy of DNA sequencing employing computer programs. Nucleic Acids Res. 6, 2601–2610.
Sultan, M., Schulz, M. H., Richard, H., Magen, A., Klingenhoff, A., Scherf, M., et al. (2008). A global view of gene activity and alternative splicing by deep sequencing of the human transcriptome. Science 321, 956–960.
Trapnell, C., Roberts, A., Goff, L., Pertea, G., Kim, D., Kelley, D. R., et al. (2012). Differential gene and transcript expression analysis of RNA-seq experiments with TopHat and Cufflinks. Nat. Protocol 7, 562–578.
Wang, B. B., and Brendel, V. (2006). Genomewide comparative analysis of alternative splicing in plants. Proc. Natl. Acad. Sci. U.S.A. 103, 7175–7180.
Wang, E. T., Sandberg, R., Luo, S., Khrebtukova, I., Zhang, L., Mayr, C., et al. (2008). Alternative isoform regulation in human tissue transcriptomes. Nature 456, 470–476.
Wang, L., Feng, Z., Wang, X., and Zhang, X. (2010). DEGseq: An R package for identifying differentially expressed genes from RNA-Seq data. Bioinformatics 26, 136–138.
Zheng, S., and Chen, L. (2009). A hierarchical Bayesian model for comparing transcriptomes at the individual transcript isoform level. Nucleic Acids Res. 37, e75. doi:10.1093/nar/gkp282
Zhu, W., Schlueter, S. D., and Brendel, V. (2003). Refined annotation of the Arabidopsis genome by complete expressed sequence tag mapping. Plant Physiol. 132, 469–484.
Conflict of Interest Statement: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Received: 16 December 2011; accepted: 18 October 2012; published online: 27 November 2012.
Citation: Bhattacharjee M, Gupta R and Davuluri RV (2012) Estimation of gene expression at isofrom level from RNA-Seq data by Bayesian hierarchical modeling. Front. Genet. 3:239. doi: 10.3389/fgene.2012.00239
This article was submitted to Frontiers in Bioinformatics and Computational Biology, a specialty of Frontiers in Genetics. Copyright © 2012 Bhattacharjee, Gupta and Davuluri. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4569,
1
],
[
4569,
6758,
2
],
[
6758,
12691,
3
],
[
12691,
18513,
4
],
[
18513,
23194,
5
],
[
23194,
28039,
6
],
[
28039,
31069,
7
],
[
31069,
34068,
8
],
[
34068,
36725,
9
],
[
36725,
41328,
10
],
[
41328,
45619,
11
],
[
45619,
53390,
12
]
]
} |
dd6bb7276f00f15373911557e08845c330ed8d1e | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 19,
"total-fallback-pages": 0,
"total-input-tokens": 44161,
"total-output-tokens": 12888,
"fieldofstudy": [
"Biology"
]
} | A New Branchipolynoe (Aphroditiformia: Polynoidae) Scale worm from Deep-Sea Hydrothermal Vents in The Onnuri Vent Field on The Northern Central Indian Ridge
Sanglyeol Kim
Korea Institute of Ocean Science & Technology
https://orcid.org/0000-0002-6290-350X
Hyeongwoo Choi
Chung-Ang University
Seong-il Eyun
Chung-Ang University
Dongsung Kim
Korea Institute of Ocean Science & Technology
Ok Hwan Yu (✉️ [email protected])
Korea Maritime University
https://orcid.org/0000-0001-6624-0865
Research
Keywords: Polychaete, Scale worm, Hydrothermal vent, COI, Onnuri vent field (OVF), Indian Ocean, Mitochondrial Genome
DOI: https://doi.org/10.21203/rs.3.rs-674600/v1
License: ☑️ ️ This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License
Abstract
Background: Deep-sea hydrothermal vents are dynamic environments with exotic faunas. In this study, we found a new species of *Branchipolynoe* (Aphroditiformia: Polynoidae) polynoid scale worm living in the recently discovered mussel *Gigantidas vrijenhoeki* in deep-sea hydrothermal vents and methane seeps at depths of 2,014 - 2,023 m. Associations between scale worms and giant mussels are common in hydrothermal ecosystems.
Results: We analyzed the morphology of *Branchipolynoe* from the Onnuri vent field (OVF) on the northern Central Indian Ridge and sequenced the full mitochondrial genome. Based on its morphological traits and mitochondrial genes, we designated the specimens as *Branchipolynoe onnuriensis* n. sp., described herein. This species resembles *B. longqiensis* and *B. tjiasmantoi*, but is distinguished by its notopodial acicular lobe form and the tips of the subacicular neurochaetae. The identity of the new species was well supported by genetic distance and phylogenetic analyses of the mitochondrial *c* oxidase subunit I (COI) gene. Genetically, the new species is closest to the Western Pacific species *B. tjiasmantoi*, phylogenetic analyses support the correlation between Indian Ocean and Western Pacific hydrothermal polychaetes. This study provides a foundation for exploring the evolutionary relationship between scale worms and giant mussels.
Introduction
The first hydrothermal vents were discovered on a sea ridge crest off the Galapagos Islands in 1977 [1]. Many vents continue to be discovered in global mid-ocean ridge systems, back-arc diffusion centers, and off-axis submarine volcanoes [2]. Deep-sea hydrothermal vents are dynamic environments with steep nutrient gradients and physicochemical conditions caused by volcanic and tectonic phenomena [3]. Investigations of the diversity and distribution of the deep-sea hydrothermal vent community are critical due to their unique nature [4]. Vents are commonly associated with dense communities and large biomass of organisms that are patchily distributed on the deep-sea bottom [5, 6]. These communities contribute significantly to the production of chemically synthesized biomass and demonstrate the remarkable adaptability of life in hydrothermal ecosystems [7]. These organisms are maintained in hydrothermal habitats by chemically synthesized energy sources such as hydrogen sulfide and methane from hydrothermal mineral deposits [8].
Since the discovery of a hydrothermal vent community at the Kairei vent field near the Rodriguez Triple Junction in the Indian Ocean in 2000 [9], five hydrothermal vent ecosystems have been found along the northern Central India Ridge (CIR) and Southwest Indian Ridge (SWIR), including the Dodo, Edmond, Longqu, Solitaire, and Onnuri vent fields [10–12]. Therefore, the biogeographical connectivity of vent fauna in the Indian Ocean has become a focus of research [2]. Mid-ocean ridges of the Indian Ocean have been hypothesized as a biogeographical pathway linking vent fauna in the Western Pacific and Atlantic Ridge systems [11]. Thus, the evolutionary affinity of vent invertebrates in the Western Pacific and Indian Ocean vent communities has been proposed to assemble asymmetrically, with a positive bias in the Western Pacific [7].
Polychaetes, in suborder Aphroditiformia (Annelida), are commonly referred to as scale worms due to their dorsal scales (elytra) [12]. Scale worms are classified into 242 genera within seven families (Acoetidae, Aphroditidae, Eulepethidae, Iphionidae, Pholoidae, Polynoidae, and Sigalionidae). Scale worms are found throughout the tropics, polar regions, intertidal zones, and deep seas [13–16]. Branchiate scale worms in the genus *Branchipolynoe* Pettibone, 1984 live within the mantle cavities of bathymodiolin mussels in hydrothermal vents and methane seeps [17]. Notably, these scale worms have very well developed, arborescent branchiae, which are much larger than
those of other branchiate polynoid species [18]. Members of genus *Branchipolynoe* are also unusual for their small scales, as most scale worms have scales that completely cover the dorsum [19]. In the Indian Ocean, *Branchipolynoe* scale worms have been reported from deep-sea hydrothermal vents on the SWIR [20].
*Branchipolynoe* species have been recorded as symbionts of host mussels; however, recent studies have suggested that they may be parasitic [21]. The polynoid *Branchipolynoe seepensis* has been found in mussels that inhabit hydrothermal vents and methane seeps, causing gill filament displacement and minor gill tissue damage [22]. These scale worms eat particulate organic matter filtered by the mussels, and accidentally consume mussel tissues during the feeding process; thus, they are not considered true parasites or commensals but rather kleptoparasites [23]. According to a biogeographical study of *Branchipolynoe* phylogenetics, the most recent common ancestor of eastern Pacific *Branchipolynoe* species may have inhabited methane seeps, and then migrated west to hydrothermal vents and methane seeps of the western Pacific and Indian Oceans [19].
In this study, we examined *Branchipolynoe* specimens collected on a survey of the hydrothermal vents of the Onnuri vent field (OVF) in 2019. We identified and described a new scale worm species and sequenced the full mitochondrial genome. Finally, we described the anatomy of the new species and evaluated its phylogenetic associations within *Branchipolynoe*.
**Materials And Methods**
**Specimen collection and preservation**
Specimens of *G. vrijenhoeki* containing *Branchipolynoe* were collected from deep-sea hydrothermal vents in the OVF on the northern Central Ridge of India (Additional file 1). The OVF is located along the CIR at 11°24.880’ S, 66°25.420’ E. All specimens were obtained at depths of 2,014 and 2,023 m using a video-guided hydraulic grab (television grab) during a Korea Institute of Ocean Science and Technology research cruise in 2019. All specimens were preserved onboard the vessel in 95% (v/v) ethanol solution in a freezer (−20°C) and then transported to a laboratory for morphological examination and phylogenetic analysis. Holotypes and a series of paratypes were deposited in Library of Marine Samples of the Korea Ocean Science & Technology (KIOST) and the National Marine Biodiversity Institute of Korea (MABIK).
**Morphological examination**
We used Leica DMC 4500 cameras mounted on a Leica M205C stereomicroscope for micrography. We used the Helicon Focus v6 software (Helicon Soft Ltd., Kharkiv, Ukraine) to merge images into a stack of pictures, which was then stained with methyl green to observe the parapodial lobes and chaetae in detail. For scanning electron microscope (SEM) observations, several parapodia were isolated from the specimens, rinsed with absolute ethanol, dehydrated, coated with gold, inspected, and photographed using a Hitachi S-4300 SEM. All known *Branchipolynoe* species were tabulated, and several key characteristics were compared including the elytra, filaments, branchiae, dorsal cirri, notochaetae, and neurochaetae.
**DNA extraction, amplification, and sequencing**
The mitochondrial COI (~700 bp) was amplified via polymerase chain reaction (PCR) using the primers polyLCO and polyHCO [24] with the D’Neasy Blood and Tissue Kit (Qiagen, Hilden, Germany) (Additional file 2). PCR amplification was performed in 20 µL reaction volumes containing 10 µL 2X TOPsimpe DyeMIX-
Tenuto (Enzynomics, Korea), 1 µL template DNA (10 pmol/µL), 0.5 µL each primer (20 pmol/µL), and 8 µL distilled water (dH₂O) under the following conditions: 1 cycle of 95°C for 2 min, followed by 35 cycles of 95°C for 30 s, 60°C for 1 min, and 72°C for 1 min, with a final extension of 72°C for 5 min. The PCR products were verified by 1% agarose gel electrophoresis in 1× TAE buffer.
**Genome sequencing and Trimming**
Two libraries (insert size, 550 bp) were constructed using the TruSeq DNA Nano 550bp kit. The libraries were sequenced using the Illumina Novaseq 6000 platform. Low-quality reads (less than Q20) were trimmed using Trim Galore! (ver. 0.6.6) and the reads having shorter than 120 bp or with unknown nucleotides (“N”) were discarded. We obtained a total of 131,865,089 reads.
**Mitochondria genome assembly and annotation**
After the filtering process, *de novo* assembly was performed using the MITOZ [25] and SPAdes (ver. 3.14.0) [26]. Putative mitochondrial contigs generated by both programs were identified and annotated on the MITOS web server [27]. The circular mitogenome was visualized using Circos (ver. 0.69-8).
**Phylogenetic analyses**
To reconstruct phylogenetic relationships, we constructed a dataset including the COI genes from 12 *Branchipolyne* species and 1 outgroup species, *Austrolaenilla antarctica* (Additional file 3), which were downloaded from National Center for Biotechnology Information (NCBI) database. Multiple alignment of the coding genes was performed using MAFFT (ver. 7.475) with the default options. COI distance matrices were implemented using the MEGA X software and the Kimura two-parameter model [28] (Table 1).
We used the Bayesian information criterion (BIC) to select the best model for phylogenetic tree reconstruction obtained using the IQ-TREE software [29]. The TPM2+F+G4 evolutionary model was the best fit for the dataset sequences. Maximum likelihood (ML) phylogeny was reconstructed using the RAxML tool [30]. ML node support was determined from a majority consensus tree constructed using 1,000 bootstrap replicates. Bayesian inference (BI) analysis was performed using MrBayes (ver. 3.2.7a) [31]. Markov chain Monte Carlo (MCMC) searches were run twice for 10⁶ generations with four chains, sampling every 500 generations for each analysis. The phylogenetic tree was visualized using FigTree (ver. 1.4.4).
**Results**
**Systematics**
Family Polynoidae Kinberg, 1856
Subfamily Aphroditiformia Pettibone, 1984
Genus *Branchipolyne* Pettibone, 1984
*B.achipolyne onnuriensis* n. sp.
**Material examined**
Six specimens. Holotype (B_S_MA_00031740) and five paratypes (B_S_MS_00031741-3), collected from the OVF on the northern Central Ridge of India (st. GTV1906 - 11°24.96’ S, 66°25.397’ E, 2064m).
**Etymology and Host**
Named in honor of the discoverer of the OVF, the host is *Gigantidas vrijenhoeki* Jang & Won, 2020.
**Description**
Body long, slightly tapered anteriorly and posteriorly, arched dorsally, and flattened ventrally, with 21 segments, including the first achaetous segment (Fig. 1A, B), and 10 pairs of elytra and elytrophores, on segments 2, 4, 5, 7, 9, 11, 13, 15, 17, and 19. Elytra moderately large, oval, smooth, without border papillae (Fig. 2C-F), covering anterior and posterior dorsal ends, but leaving the middle of the body partially covered by the dorsum (Fig. 1A, B). Non-elytra-bearing segments with short, smooth dorsal cirri on short, cylindrical cirrophores. Dorsal cirri with short slender tips, tapering gradually, longer than anterior and ventral cirri, not extending beyond the tips of the neurochaetae.
Prostomium ellipsoidal, bilobed with almost rounded anterior lobes. Short, conical median antenna inserted between two anterior lobes and pair of short conical palps (Fig. 1C). Median antenna and palps smooth and tapering to the slender tip. Palps extending beyond the prostomium. Prostomium lacking frontal filaments, eyes, and lateral antennae. First segment fused to prostomium with two pairs of short anterior cirri. Anterior cirri smooth and slightly slender, not exceeding the prostomium length (Fig. 1C). Thick, extended muscular pharynx with five pairs of dorsal and ventral small, sac-like terminal papillae surrounding the mouth (Fig. 2A).
Branchiae on segments 3–21 dense and arborescent, with short terminal filaments (Fig. 2B), not extending beyond the elytral border. Branchiae separated into two types, showing dorsal and ventral emergence, respectively. No discernible dorsal tubercles. Branchiae gradually decreasing in size anteriorly and posteriorly.
Parapodia subbiramous. Notopodia smaller than neuropodia, with few notochaetae projecting beyond notopodia (Additional file 4). Neuropodia large, rounded, enclosing numerous neurochaetae with rounded lobes. Notochaetae smooth, stouter, and shorter than neurochaetae (Fig. 3A). Notochaetae few, more abundant on the middle and posterior than anterior segments, slightly tapered with serrated distal part; tip rounded, shaft with inconspicuous rows (Fig. 3B-D). Neurochaetae numerous, more abundant in the middle than anterior and posterior segments, arranged as a vertical fan; tapered, with subdistal swelling and small spines along edge, serrations starting at the midpoint on only one side and extending distally.
Neurochaetae divided into supraacicular and subacicular neuroseatae. Supraacicular neurochaetae long, stout, slender tips, each with a minute hook; serrated distally and flattened on one side. Subacicular neurochaetae with slightly hooked tip, serrated parts shorter than supraacicular neurochaetae (Fig. 4).
Ventral cirri small, smooth, without papillae, and attached to the middle regions of neuropodia; projecting anteriorly (Fig. 1D). Elongated ventral papillae of female on segments 11 and 12; projecting posteriorly, reaching the subsequent segment. Elongated ventral papillae of males on segment 12, not extending beyond half the length of the subsequent segment (Fig. 1D). Pygidium small, with pair of short, thick, tapered conical anal cirri.
**Morphological variation**
The holotype 28 mm long and 13 mm wide, including parapodia. Paratypes vary in size, 23–31 mm long and 9–15 mm wide. All specimens with elongated ventral papillae on segments 11 and 12, suggesting that males were not found.
**Remarks**
Nine species have been described in the genus *Branchipolynoe* [4,13,19,20,32,33]. The diagnostic characteristics of the genus were determined from *Branchipolynoe symmytilida* and amended by Pettibone for *B. seepensis* [13,33]. This latter revision included the first position of the branchiae, the presence of the frontal filament, and form of the parapodium. Subsequently, Zhou (2017) published a description of *B. longqiensis* in the Indian Ocean and Lindgren (2019) published descriptions of five new species from the Pacific Ocean (Additional file 3). Members of the genus have 21 segments and 10 elytra; the elytra partially cover the dorsal region. Only *B. symmytilida* has frontal filaments, and all have a bilobed prostomium, except *B. kajsae*. In *Branchipolynoe* n. sp, the branchiae start at the third segment, and the parapodium is subbiramous, forming a very similar shape to *B. longqiensis* and *B. tjiasmantoi*, but with a short, rounded notopodial acicular lobe, inconspicuous pharynx papillae, and different shapes of the tips of sub-acicular neurochaetae (Additional file 5).
**Phylogenetic analyses**
Among *Branchipolynoe* species, the mitochondrial COI genetic distance ranged from 0.056 to 0.237, with an average of 0.175 (Table 1). The species with the closest COI genetic distance to that of *B. onnuriensis* n. sp. was *B. tjiasmantoi* from the western Pacific, with a value of 0.056. This value is similar to those of *B. halliseyae* and *B. kajsae* (0.059), which are referred to as sister species in previous studies due to their similar morphology [19]. The second closest COI genetic distance belonged to *B. longqiensis*(0.099) found in the Indian Ocean.
The ML tree and BI analyses inferred from *Branchipolynoe* mitochondrial COI sequences produced a single topology for each region (Fig. 5). The phylogenetic trees isolated the *Branchipolynoe* species and supported a clade comprising *B. onnuriensis* n. sp. and *B. tjiasmantoi* (ML: 78%, BI: 1).
**General features of mitochondrial genomes**
The complete mitochondrial genome of *B. onnuriensis* n. sp. was 16,217 bp in length, comprising 15 protein-coding genes (PCGs) (Fig. 6), 7 NADH dehydrogenase subunits (nad1–6 and nad4L), 4 cytochrome oxidase subunits (cob and cox1–3), 2 ATP synthase subunits (atp6 and atp8), and 2 small and large ribosomal RNA genes (rns and rnl). We identified 22 transfer RNA (tRNA) genes and an A+T-rich region. Among these genes, ND5 was the longest (1,525 bp) and atp8 was the shortest (160 bp). The tRNA length ranged from 64 (trnC) to 71 (trnQ), with an average length of 66.41 bp (Table 2). No gene rearrangement was detected (Additional file 6).
**Base composition**
To assess the mitochondrial genome, we calculated its nucleotide composition (A%, C%, G%, T%, A+T%, C+G%), AT skew, and GC skew. AT and GC skew were calculated as follows: AT skew = (A – T%) / (A% + T%) and GC skew = (G – C%) / (G% + C%). The overall nucleotide composition of the complete mitochondrial genome was 28.45% A, 24.44% C, 8.99% G, and 38.12% T. The proportion of AT content (66.57%) was ~1.99 times higher than that of GC content (33.43%) (Additional file 7). Most genes showed positive AT skew, except rns and rnl. All genes also showed negative GC skew, indicating that PCGs in *B. onnuriensis* n. sp. contained a higher percentage of T and C than A and G, except rns and rnl.
Discussion
*Branchipolynoe onnuriensis* n. sp. is the second *Branchipolynoe* species to be found in deep-sea hydrothermal vents in the Indian Ocean. Morphologically, *B. onnuriensis* is very similar to *B. longqiensis* from the Indian Ocean and *B. tjiasmantoi* from the western Pacific. To date, the genus *Branchipolynoe* has been described from only 10 species discovered in hydrothermal vents and methane seeps worldwide, including 5 in the eastern Pacific, 2 in the western Pacific, 1 in the Atlantic, and 2 in India; among these, *B. eliseae, B. halliseyae, B. kajsae, B. meridae*, and *B. tjiasmantoi* were recently described from the eastern and western Pacific [19]. Lindgren (2019) suggested that the most recent common ancestor of *Branchipolynoe* species may have lived in methane seeps in the eastern Pacific. Hydrothermal vents and methane seeps comprise similar basic ecological systems; therefore, it appears that *Branchipolynoe* moved west to colonize vents and seeps of the western Pacific and Indian Oceans. Future studies should elucidate *Branchipolynoe* migration patterns to other regions in the Indian Ocean and their evolution to adapt to these harsh environments.
The species discovered in this study was difficult to distinguish from previously discovered several species using external morphological characteristics, but showed sufficient genetic distance to classify it as a new species. It is increasingly common to describe animal species without obvious morphological differences based on molecular data alone [34]. This technique allows the reliable isolation of polynoid species, which is important for understanding the distribution and geographic patterns of this taxon [35]. Polynoids are the most diverse and widely distributed polychaete group found in hydrothermal vents and seeps, and are a good model for evaluating the biogeographic distribution of deep-sea environmental faunas [4]. Further studies of polynoids are needed to enhance our understanding of deep-sea ecosystems.
*Branchipolynoe* live in hydrothermal vents or methane seeps, and *B. pettiboneae* has been found in both. Two undescribed species have been reported from hydrothermal fields on the CIR and SWIR [36]. *Branchipolynoe* sp. “Dragon,” a commensal scale worm found in bivalve mussels in the Longqi vent field on the SWIR, shows genetic similarity to an undescribed species (*Branchipolynoe* sp. “VG-2002”) that was recorded and sequenced from the Kairei vent field on the CIR. Further research on *Branchipolynoe* species in the Indian Ocean is needed to elucidate the adaptations of animals living in vents of the CIR.
There is no consensus on the minimum COI distance required to classify a species, and each taxon has a different minimum COI distance [37]. Therefore, a minimum interspecies distance greater than the maximum interspecies distance is often applied to identify new species [38]. In polychaetes, a COI distance of $\geq 2\%$ is considered to indicate a different species, and a $\geq 10\%$ difference is used to indicate a different genus. For the newly discovered species, the minimum genetic distance among species was 0.056, representing a difference of $>5\%$ from its most closely related species, *B. tjiasmantoi*. Thus, even without morphological data, the COI distances indicate that *B. onnuriensis* n. sp. is a new species.
The newly discovered polynoid is symbiotic to the recently discovered mussel *G. vrijenhoeki*. Unlike other free-living vent polynoids (e.g., *Branchinotogluma*), *Branchipolynoe* species typically inhabit the inner mantle space of *Bathymodiolus* mussels. Therefore, the lack of a record of symbiotic polynoids in the OVF may be due to the apparent rarity of *Bathymodiolus* in this region [39]. *B. pettiboneae* was first discovered in *Gigantidas platifrons*; however, most specimens have been found in *Bathymodiolus* mussels. Symbiotic species subsist on the byproducts of chemical synthesis by the host, and obtain protection from the external environment from the host shell [40]. Rindren (2019) found no relationship between haplotype and host mussel or depth in a new species.
found in Costa Rica, and no significant within-species genetic distances at different depths or in different hosts. *Branchipolynoe* species are found in a variety of hosts and at a range of depths; thus, it appears that any mussel is capable of inhabiting hydrothermal vents or methane seeps can act as a suitable host. Different host bivalves adapt to different hydrothermal vents or methane seeps, and *Branchipolynoe* has evolved to coexist with these shellfish.
Most members of *Branchipolynoe* live individually inside the mussel mantle pallial cavity [41]. Sometimes two are found, but these are very rare occurrences. Among species found in hydrothermal vents, the polynoid *B. symytilida* was found in the mussel *B. thermophilus*, and the polynoid *B. pettiboneae* was found in the mussel *Bathymodiolus brevior*. The polynoids *B. longqiensis* and *B. tjiasmantois* were found in the mussels *Bathymodiolus marisindicus* and *B. brevior*, respectively. *Branchipolynoe* species have also been newly discovered in cold seeps. *Branchipolynoe* larvae can travel long distances between vents or among numerous ventilation systems, allowing gene flow [19]. Further studies are needed to determine whether *Branchipolynoe* species have host mussel preferences, how they coexist and evolve, and the role that they play in hydrothermal environments.
**Declarations**
**Acknowledgments**
We thank the researcher Sumin Kang for her kind help sorting the specimen. We also thank the cruise members of R/V Isabu for their assistance with the fieldwork.
**Authors’ contributions**
DK and OY: sampled the specimen, HC and SK: methodology, SK, HC, SE, and OY: interpretation, SK, HC, SE, and OY: manuscript writing, OY: supervision
**Funding**
This work was part of the project titled ‘Understanding the deep-sea biosphere on seafloor hydrothermal vents in the Indian Ridge (no. 20170411)’ funded by the Ministry of Oceans and Fisheries, Korea and Korea Institute of Ocean Science and Technology (PM62320).
**Availability of data and materials**
The data presented in this study are available in the main text and associated Supplementary Materials. The complete mitochondrial sequence was submitted to GenBank with an accession number MZ457523.
**Ethics approval and consent to participate**
**Consent for publication**
All authors have read and agreed to the published version of the manuscript.
**Competing interests**
The authors declare no conflict of interest.
1. Lonsdale P. Clustering of suspension-feeding macrobenthos near abyssal hydrothermal vents at oceanic spreading centers. Deep Res. 1977;24.
2. Beedessee G, Watanabe H, Ogura T, Nemoto S, Yahagi T, Nakagawa S, et al. High connectivity of animal populations in deep-sea hydrothermal vent fields in the central indian ridge relevant to its geological setting. PLoS One. 2013;8:e81570.
3. Ryu T, Kim JG, Lee J, Yu OH, Yum S, Kim D, et al. First transcriptome assembly of a newly discovered vent mussel, Gigantidas vrijenhoeki, at Onnuri Vent Field on the northern Central Indian Ridge. Mar Genomics. 2021;57:100819.
4. Wu X, Zhan Z, Xu K. Two new and two rarely known species of Branchinotogluma (Annelida: Polynoidae) from deep-sea hydrothermal vents of the Manus Back-Arc basin, with remarks on the diversity and biogeography of vent polynoids. Deep Res Part I Oceanogr Res Pap. 2019;149:1–20.
5. Rogers AD, Tyler PA, Connelly DP, Copley JT, James R, Larter RD, et al. The discovery of new deep-sea hydrothermal vent communities in the Southern ocean and implications for biogeography. PLoS Biol. 2012;10:e1001234.
6. Van Dover CL. Biogeography of hydrothermal vent communities along seafloor spreading centers. Trends Ecol Evol. 1990;5:242–6.
7. Van Dover CL, Humphris SE, Fornari D, Cavanaugh CM, Collier R, Goffredi SK, et al. Biogeography and ecological setting of Indian Ocean hydrothermal vents. Science. 2001;294:818–23.
8. Dubilier N, Bergin C, Lott C. Symbiotic diversity in marine animals: The art of harnessing chemosynthesis. Nat Rev Microbiol. 2008;6:725–40.
9. Hashimoto J, Ohta S, Gamo T, Chiba H, Yamaguchi T, Tsuchida S, et al. First hydrothermal vent communities from the Indian Ocean discovered. Zoolog Sci. 2001;18:717–21.
10. Kim JG, Lee J. A new species of the genus Smacigastes Ivanenko & Defaye, 2004 (Tegastidae, Harpacticoida, Copepoda) from the Onnuri Vent Field in the Indian Ocean. Zoosystematics Evol. 2020;96:29–45.
11. Mevel C, Tamaki K, German CR, Baker ET, The FST. FUJI ST. Hydrothermal activity along the southwest Indian ridge. Nature. 1998;395:490–93.
12. Rouse GW, Fauchald K. Cladistics and Polychaetes. Zool Scr. 1997;26:139–204.
13. Pettibone M. A new scale-worm commensal with deep-sea mussels in the seep-sites at the Florida escarpment in the eastern Gulf of Mexico (Polychaeta: Polynoidae: Branchipolynoinae). Proc Biol Soc Washingt. 1986;99:444–51.
14. Glasby CJ, Hutchings PA. A new species of Marphysa Quatrefages, 1865 (Polychaeta: Eunicida: Eunicidae) from northern Australia and a review of similar taxa from the Indo-west Pacific, including the genus Nauphanta Kinberg, 1865. Zootaxa. 2010;45:29–45.
15. Wiklund H, Nygren A, Pleijel F, Sundberg P. Phylogeny of Aphroditiformia (Polychaeta) based on molecular and morphological data. Mol Phylogenet Evol. 2005;37:494–502.
16. Norlinder E, Nygren A, Wiklund H, Pleijel F. Phylogeny of scale-worms (Aphroditiformia, Annelida), assessed from 18SrRNA, 28SrRNA, 16SrRNA, mitochondrial cytochrome c oxidase subunit I (COI), and morphology. Mol Phylogenet Evol. 2012;65:490–500.
17. Desbruyères D, Hashimoto J, Fabri MC. Composition and biogeography of hydrothermal vent communities in western pacific back-arc basins. Geophys Monogr Ser. 2006;166:215–34.
18. Hourdez S, Jouin-Toulmond C. Functional anatomy of the respiratory system of Branchipolynoe species (Polychaeta, Polynoidae), commensal with Bathymodiolus species (Bivalvia, Mytilidae) from deep-sea hydrothermal vents. Zoomorphology. 1998;118:225–33.
19. Lindgren J, Hatch AS, Hourdez S, Seid CA, Rouse GW. Phylogeny and biogeography of Branchipolynoe (polynoidae, phyllodocida, acciculata, annelida), with descriptions of five new species from methane seeps and hydrothermal vents. Diversity. 2019;11:1–39.
20. Zhou Y, Zhang D, Lu B, Wang C. Description of a new branchiate scale-worm (Polychaeta: Polynoidae) from the hydrothermal vent on Southwest Indian Ocean Ridge. Zootaxa. 2017;4282:123–34.
21. Bebianno MJ, Cardoso C, Gomes T, Blasco J, Santos RS, Colaço A. Metal interactions between the polychaete Branchipolynoe seepensis and the mussel Bathymodiolus azoricus from Mid-Atlantic-Ridge hydrothermal vent fields. Mar Environ Res. 2018;135:70–81.
22. Ward ME, Shields JD, Van Dover CL. Parasitism in species of Bathymodiolus (Bivalvia: Mytilidae) mussels from deep-sea seep and hydrothermal vents. Dis Aquat Organ. 2004;62:1–16.
23. Britayev TA, Martin D, Krylova EM, Von Cosel R, Aksiuk TS. Life-history traits of the symbiotic scale-worm Branchipolynoe seepensis and its relationships with host mussels of the genus Bathymodiolus from hydrothermal vents. Mar Ecol. 2007;28:36–48.
24. Carr CM, Hardy SM, Brown TM, Macdonald TA, Hebert PDN. A tri-oceanic perspective: DNA barcoding reveals geographic structure and cryptic diversity in Canadian polychaetes. PLoS One. 2011;6:e22232.
25. Meng G, Li Y, Yang C, Liu S. MitoZ: A toolkit for animal mitochondrial genome assembly, annotation and visualization. Nucleic Acids Res. 2019;47:e63.
26. Bankevich A, Nurk S, Antipov D, Gurevich AA, Dvorkin M, Kulikov AS, et al. SPAdes: A new genome assembly algorithm and its applications to single-cell sequencing. J Comput Biol. 2012;19:455–77.
27. Bernt M, Donath A, Jühling F, Externbrink F, Florentz C, Fritzsch G, et al. MITOS: Improved de novo metazoan mitochondrial genome annotation. Mol Phylogenet Evol. 2013;69:313–9.
28. Kumar S, Stecher G, Li M, Knyaz C, Tamura K. MEGA X: Molecular evolutionary genetics analysis across computing platforms. Mol Biol Evol. 2018;35:1547–9.
29. Kalyaanamoorthy S, Minh BQ, Wong TKF, Von Haeseler A, Jermiin LS. ModelFinder: Fast model selection for accurate phylogenetic estimates. Nat Methods. 2017;14:587–9.
30. Kozlov AM, Darriba D, Flouri T, Morel B, Stamatakis A. RAxML-NG: A fast, scalable and user-friendly tool for maximum likelihood phylogenetic inference. Bioinformatics. 2019;35:4453–5.
31. Ronquist F, Huelsenbeck JP. MrBayes 3: Bayesian phylogenetic inference under mixed models. Bioinformatics. 2003;19:1572–4.
32. Miura T, Hashimoto J. Two new branchiate scale-worms (Polynoidae: Polychaeta) from the hydrothermal vent of the Okinawa through and the volcanic seamount off Chichijima Island. Proc Biol Soc Washington.
1991;104:166–74.
33. Pettibone M. A new scale-worm commensal with deepsea mussels on the Galapagos hydrothermal vent (Polychaeta, Polynoidae). Proc Biol Soc Wash. 1984;97:226–39.
34. Halt MN, Kupriyanova EK, Cooper SJB, Rouse GW. Naming species with no morphological indicators: Species status of Galeolaria caespitosa (Annelida:Serpulidae) inferred from nuclear and mitochondrial gene sequences and morphology. Invertebr Syst. 2009;23:205–22.
35. Nygren A. Cryptic polychaete diversity: A review. Zool Scr. 2014;43:172–83.
36. Copley JT, Marsh L, Glover AG, Hühnerbach V, Nye VE, Reid WDK, et al. Ecology and biogeography of megafauna and macrofauna at the first known deep-sea hydrothermal vents on the ultraslow-spreading Southwest Indian Ridge. Sci Rep Nature Publishing Group. 2016;6:1–13.
37. Berriman JS, Ellingson RA, Awbrey JD, Rico DM, Valdés ÁA, Wilson NG, et al. A biting commentary: Integrating tooth characters with molecular data doubles known species diversity in a lineage of sea slugs that consume “killer algae.”. Mol Phylogenet Evol. 2018;126:356–70.
38. Meier R, Zhang G, Ali F. The use of mean instead of smallest interspecific distances exaggerates the size of the “barcoding gap” and leads to misidentification. Syst Biol. 2008;57:809–13.
39. McKiness ZP, McMullin ER, Fisher CR, Cavanaugh CM. A new bathymodioline mussel symbiosis at the Juan de Fuca hydrothermal vents. Mar Biol. 2005;148:109–16.
40. Company R, Serafim A, Cosson R, Fiala-Médioni A, Dixon DR, Bebianno MJ. Adaptation of the antioxidant defence system in hydrothermal-vent mussels (Bathymodiolus azoricus) transplanted between two Mid-Atlantic Ridge sites. Mar Ecol. 2007;28:93–9.
41. Plouviez S, Daguin-Thiébaut C, Hourdez S, Jollivet D. Juvenile and adult scale worms branchipolynoe seepensis in Lucky Strike hydrothermal vent mussels are genetically unrelated. Aquat Biol. 2008;3:79–87.
Tables
Table 1 Kimura two-parameter distance matrix of genus *Branchipolynoe* taxon COI sequences. 1. *Branchipolynoe symmytilida*, 2. *B. seepensis*, 3. *B. pettiboneae*, 4. *B. longqiensis*, 5. *B. eliseae*, 6. *B. halliseyae*, 7. *B. kajsae*, 8. *B. meridae*, 9. *B. tjiasmantoi*, 10. *B. Branchipolynoe* sp. “Dragon,” 11. *Branchipolynoe* sp. “VG-2002,” 12. *B. onnuriensis* n. sp., and 13. *Austrolaenilla antarctica*
| | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 |
|---|-----|-----|-----|-----|-----|-----|-----|-----|-----|-----|-----|-----|
| 1 | | | | | | | | | | | | |
| 2 | 0.204 | | | | | | | | | | | |
| 3 | 0.223 | 0.183 | | | | | | | | | | |
| 4 | 0.223 | 0.191 | 0.068 | | | | | | | | | |
| 5 | 0.102 | 0.185 | 0.189 | 0.209 | | | | | | | | |
| 6 | 0.199 | 0.071 | 0.182 | 0.201 | 0.195 | | | | | | | |
| 7 | 0.197 | 0.066 | 0.184 | 0.199 | 0.197 | 0.059 | | | | | | |
| 8 | 0.096 | 0.205 | 0.216 | 0.237 | 0.087 | 0.214 | 0.193 | | | | | |
| 9 | 0.217 | 0.192 | 0.131 | 0.110 | 0.237 | 0.195 | 0.196 | 0.240 | | | | |
|10 | 0.226 | 0.194 | 0.073 | 0.004 | 0.212 | 0.204 | 0.195 | 0.233 | 0.115 | | | |
|11 | 0.226 | 0.200 | 0.075 | 0.006 | 0.211 | 0.210 | 0.209 | 0.247 | 0.117 | 0.010 | | |
|12 | **0.219** | **0.207** | **0.110** | **0.099** | **0.237** | **0.228** | **0.215** | **0.225** | **0.056** | **0.104** | **0.107** | |
|13 | 0.339 | 0.298 | 0.308 | 0.301 | 0.320 | 0.281 | 0.294 | 0.335 | 0.294 | 0.309 | 0.313 | 0.306 |
**Table 2** Annotation of the *Branchipolynoe onnuriensis* n. sp. mitochondrial genome
| Gene | Strand | Position | Length (bp) | Initiation codon | Stop codon |
|-------|--------|--------------|-------------|------------------|------------|
| trnX | + | 1,099-1,666 | 68 | | |
| nad2 | + | 1,194-2,055 | 862 | ACC | ACC |
| cox1 | + | 2,173-3,694 | 1,522 | CGC | ATT |
| cox2 | + | 3,764-4,430 | 667 | ATG | TGA |
| trnD | + | 4,458-4,524 | 67 | | |
| atp8 | + | 4,524-4,683 | 160 | ATG | TAA |
| trnY | + | 4,681-4,748 | 68 | | |
| cox3 | + | 4,764-5,526 | 763 | TTT | TCT |
| trnQ | + | 5,533-5,603 | 71 | | |
| cob | + | 5,638-6,757 | 1,120 | ATC | ATT |
| trnL2 | + | 6,764-6,831 | 68 | | |
| trnF | + | 6,831-6,898 | 68 | | |
| trnE | + | 6,905-6,969 | 65 | | |
| trnP | + | 6,970-7,034 | 65 | | |
| nad4 | + | 7,223-8,507 | 1,285 | TGA | ATA |
| trnG | + | 8,524-8,589 | 66 | | |
| trnS2 | + | 8,652-8,719 | 68 | | |
| nad1 | + | 8,737-9,628 | 892 | ATT | TTA |
| trnI | + | 9,650-9,718 | 69 | | |
| trnK | + | 9,723-9,790 | 68 | | |
| nad3 | + | 9,776-10,125 | 350 | ATG | TAA |
| trnN | + | 10,149-10,213| 65 | | |
| nad6 | + | 10,274-10,763| 490 | ATG | TAA |
| trnW | + | 10,783-10,847| 65 | | |
| atp6 | + | 10,976-11,537| 562 | AAT | CAC |
| trnR | + | 11,547-11,611| 65 | | |
| trnH | + | 11,612-11,676| 65 | | |
| nad5 | + | 11,742-13,266| 1,525 | TCT | ATT |
| trnT | + | 13,394-13,459| 66 | | |
| nad4L | + | 13,459-13,762| 304 | ATG | TAA |
Figure 1
Branchipolynoe onnuriensis n. sp. holotype, female. A) Dorsal view. B) Ventral view. C) Head and anterior segments in dorsal view. D) Mid-body in ventral view; arrows indicate elongated ventral papillae in segments 11 and 12.
Abbreviations: pa, palps; ma, median antenna; pr, prostomim; e1, first elytron (attached to segment 2); dac, dorsal anterior cirrus; vac, ventral anterior cirrus
Figure 2
Branchipolynoe onnuriensis n. sp. paratype 5. A) Frontal view of proboscis. B) Branchia from segment 10. C) Left 4th elytron from segment 7. D) Left 6th elytron from segment 11. E) Left 8th elytron from segment 15. F) Left 9th elytron from segment 17.
Figure 3
Branchipolynoe onnuriensis n. sp. paratype 5, female, showing left parapodum on segment 4. A) Magnified to show notochaetae view. B–D) Detail of notochaeta tips
Figure 4
Branchipolynoe onnuriensis n. sp. paratype 5, female, showing left parapodium on segment 6. A) Neurochaetae. B–C) Magnified to show supraacicular neurochaetae. D–F) Detail of supraacicular neurochaeta tips. G–H) Detail of subacicular neurochaeta tips
Figure 5
Maximum likelihood (ML) tree inferred from mitochondrial cytochrome c oxidase I (COI) sequences of genus Branchipolynoe. The COI sequence of Austrolaenilla antarctica was used as an outgroup. Numbers at nodes represent ML and Bayesian inference (BI) support values.
Figure 6
Circular map of the complete mitochondrial genome of Branchipolynoe onnuriensis n. sp. Outer to inner circles represent the (i) positions of annotated genes, (ii) sequencing depth, (iii) GC content. The total mtDNA length was 16,217 bp
Supplementary Files
This is a list of supplementary files associated with this preprint. Click to download.
- Additionalfile2.docx
- Additionalfile3.docx
- Additionalfile5.docx
- Additionalfile7.docx
- Additionalfile1.docx
- Additionalfile4.docx
• Additionalfile6.docx | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
800,
1
],
[
800,
4758,
2
],
[
4758,
8288,
3
],
[
8288,
10878,
4
],
[
10878,
14393,
5
],
[
14393,
18028,
6
],
[
18028,
22188,
7
],
[
22188,
24667,
8
],
[
24667,
27333,
9
],
[
27333,
30924,
10
],
[
30924,
33240,
11
],
[
33240,
34698,
12
],
[
34698,
37258,
13
],
[
37258,
37657,
14
],
[
37657,
37919,
15
],
[
37919,
38090,
16
],
[
38090,
38628,
17
],
[
38628,
39123,
18
],
[
39123,
39145,
19
]
]
} |
99ab0bffd8d9eb84c435ed13cfdfc6946dddb211 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 13,
"total-fallback-pages": 0,
"total-input-tokens": 39573,
"total-output-tokens": 12911,
"fieldofstudy": [
"Biology"
]
} | Moonlighting of *Helicobacter pylori* catalase protects against complement-mediated killing by utilising the host molecule vitronectin
Corinna Richter¹, Oindrilla Mukherjee¹, David Ermert², Birendra Singh³, Yu-Ching Su³, Vaibhav Agarwal², Anna M. Blom² & Kristian Riesbeck¹
*Helicobacter pylori* is an important human pathogen and a common cause of peptic ulcers and gastric cancer. Despite *H. pylori* provoking strong innate and adaptive immune responses, the bacterium is able to successfully establish long-term infections. Vitronectin (Vn), a component of both the extracellular matrix and plasma, is involved in many physiological processes, including regulation of the complement system. The aim of this study was to define a receptor in *H. pylori* that binds Vn and determine the significance of the interaction for virulence. Surprisingly, by using proteomics, we found that the hydrogen peroxide-neutralizing enzyme catalase KatA is a major Vn-binding protein. Deletion of the *katA* gene in three different strains resulted in impaired binding of Vn. Recombinant KatA was generated and shown to bind with high affinity to a region between heparin-binding domain 2 and 3 of Vn that differs from previously characterised bacterial binding sites on the molecule. In terms of function, KatA protected *H. pylori* from complement-mediated killing in a Vn-dependent manner. Taken together, the virulence factor KatA is a Vn-binding protein that moonlights on the surface of *H. pylori* to promote bacterial evasion of host innate immunity.
*Helicobacter pylori* is a spiral-shaped Gram negative bacterium that specifically colonises the human stomach¹. It has been estimated that over half of the global population is colonised by this pathogen². Even though infections can remain asymptomatic, colonisation with *H. pylori* will lead to gastritis and peptic ulcers in at least 10% of cases. Approximately 1% of infected individuals develop gastric adenocarcinoma, which represents up to 80% of all gastric cancers¹. Infection with *H. pylori* provokes a strong innate as well as cellular and humoral adaptive response. Despite the immunogenicity of *H. pylori*, however, the host is usually unable to clear the infection. Known mechanisms by which *H. pylori* subverts the host's immune response include avoidance of recognition by Toll-like receptors⁴, survival in macrophages⁵, and inhibition of both T cell activation and memory T cell responses⁶,⁷ to name only a few.
‘Hijacking’ of host molecules is a strategy used by several bacterial pathogens to facilitate adhesion and/or to evade the host's immune response⁸,⁹. In this respect, one particularly interesting host factor is vitronectin (Vn), an abundant component in human plasma and the extracellular matrix (ECM). Vn is a multifunctional glycoprotein (75 kDa single-chain or 65 kDa plus 10 kDa two-chain form), which can be present in a monomeric or an active multimeric state¹⁰. The molecule is composed of the N-terminal somatomedin-B domain followed by an RGD motif, which is recognised by integrins. The central part is dominated by three hemopexin-like domains and a fourth hemopexin-like domain is located at the C-terminus. The heparin binding domain 3 (HBD-3), also located at the C-terminal end, is the major binding site for most studied bacterial Vn-binding proteins⁸. There are two more putative heparin-binding sites (i.e., HBD-1 and HBD-2) located in the central part of the protein, but their relevance for heparin binding is equivocal. Vn is recognised by various cell surface receptors such as
¹Clinical Microbiology, Department of Translational Medicine, Lund University, SE-205 02 Malmö, Sweden.
²Medical Protein Chemistry, Department of Translational Medicine, Lund University, SE-205 02 Malmö, Sweden.
Correspondence and requests for materials should be addressed to K.R. (email: [email protected])
integrins or proteoglycans thereby functioning in cell adhesion, migration, and tissue remodelling. Additionally, Vn is involved in the regulation of the complement system. The complement system is composed of more than 50 soluble and membrane-bound factors and is considered a major player of innate immunity. Activation induces a proteolytic cascade, which eventually leads to the induction of the terminal pathway. Here, Vn interferes with the formation of the membrane attack complex (MAC) by inhibiting the C5b-C7 complex and polymerisation of C9, subsequently preventing cell lysis. Several pathogenic bacteria including Haemophilus influenzae, Neisseria meningitidis, and Pseudomonas aeruginosa have been shown to bind Vn, thereby increasing their complement resistance or ability to adhere to host tissues.
_H. pylori_ binds Vn, but the proteins involved and the importance of this interaction for virulence have not been elucidated. In this study, we identified Vn-binding properties of the _H. pylori_ catalase KatA and characterised the interaction on the molecular level. Furthermore, we showed that Vn binding by _H. pylori_ KatA increases complement resistance and, thus, is an important factor in the evasion of the innate immune response.
Results and Discussion
**_H. pylori_ catalase KatA binds Vn.** The first aim of this study was to identify the major Vn-binding protein in _H. pylori_ since the bacterium has previously been shown to bind Vn with varying binding capacity for different isolates. We therefore tested binding of Vn, which was purified from sera obtained from healthy human volunteers, to 13 clinical strains from our collection (Supplementary Fig. S1 and Supplementary Table 1). For further analysis we chose three _H. pylori_ with strong (CCUG18943; 92% binding), intermediate (KR697; 69%), and weak Vn binding (KR497; 15%). Outer membrane fractions of these three strains were prepared and subjected to 2D-gel electrophoresis followed by far-Western blotting against human serum Vn (Fig. 1a and b). Three Vn-binding protein spots were detected in _H. pylori_ CCUG18943 (Fig. 1a) and KR697 (Fig. 1b). A fourth intense spot corresponded to urease subunit B (UreB). No Vn-binding proteins were detected in the weak Vn-binding strain KR497 (data not shown). Interestingly, subsequent MALDI-TOF MS analysis of proteins in KR697 and CCUG18943 identified spots 1 and 2 as two isoforms of _H. pylori_ urease subunit A (UreA) and spot 3 as catalase KatA. To further verify our findings, we performed a pull-down experiment, in which the outer membrane fraction of _H. pylori_ CCUG18943 was subjected to a Vn-coupled Sepharose column (data not shown). MALDI-TOF MS identified KatA, but neither UreA nor UreB could be detected in this assay, despite the fact that mild conditions (pH 7.4, 140 mM NaCl) were used during the pull-down. It is possible, however, that the conditions chosen in the pull-down experiment were not optimal for urease and we therefore, at this point, cannot exclude that urease binds Vn. However, the fact that _H. pylori_ could successfully be identified as a Vn-binding protein in two different experimental setups suggested a significant interaction between the two proteins and prompted us to focus our attention on the characterisation of KatA in the sequel of this study.
The catalase KatA is an important virulence determinant of _H. pylori_ with designated enzymatic functions, and the identification of KatA as putative Vn-binding partner was at first sight surprising. The phenomenon of additional, unrelated functions in (often highly conserved) proteins has, however, been described previously, and is now known as ‘moonlighting’. Examples of moonlighting proteins are the superoxide dismutase (SOD) of Mycobacterium avium, which exhibits additional functions as an adhesin targeting aldolase, glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and cyclophilin A on epithelial cells, thereby promoting endocytosis or the _H. influenzae_ protein F, an ABC transporter, which also binds laminin and facilitates adhesion.
_H. pylori_ KatA, like other catalases, is a predicted cytoplasmic protein with no signal-sequence for any known transport pathway. However, presence of KatA in the periplasm and even surface location has been proposed, even though the means of translocation remain elusive. Obviously, surface location of KatA would be essential for the interaction with Vn. Thus, we wanted to verify the KatA-dependent Vn binding with intact bacteria and constructed katA deletion mutants in _H. pylori_ strains CCUG18943 and KR697, as well as in the weak Vn-binding strain KR497. All _H. pylori_ ΔkatA strains showed significantly less binding to Vn when compared to the wild type (wt) counterparts in flow cytometry (Fig. 2). Deletion of katA resulted in 21% reduction in binding of Vn to _H. pylori_ CCUG18943, whereas a 50% reduction in Vn binding was observed with the two strains KR697ΔkatA and KR497ΔkatA when compared to their respective wt (Fig. 2). Taken together, our data provided additional evidence for surface localisation of KatA and furthermore showed that KatA is relevant for Vn binding to whole bacteria.
Deletion of _katA_ did not completely abolish Vn binding and the relative impact of the deletion on Vn binding varied between isolates, suggesting that _H. pylori_ has additional Vn-binding proteins. The concept that one bacterial species possesses multiple proteins targeting the same ligand is well established. The concept that one bacterial species possesses multiple proteins targeting the same ligand is well established. The concept that one bacterial species possesses multiple proteins targeting the same ligand is well established.
**Helicobacter pylori** KatA binds Vn using a unique binding site. Next, we wanted to characterise the KatA-Vn interaction at the protein level. We performed an ELISA using full length recombinant KatA and a dilution series (2.5–80 nM) of Vn, which spans most known bacterial binding sites. Binding of KatA and Vn was dose-dependent and saturable, confirming a specific interaction (Fig 3a). To exclude any effects resulting from trace amounts of _Escherichia coli_ proteins, P09011 from _H. influenzae_, which was purified using the same protein expression and purification protocol as for KatA, served as negative control for Vn binding (Fig. 3a, inset). Furthermore, we wanted to reveal whether KatA preferentially targets native (monomeric) Vn or the activated polymeric form. Binding of KatA to either of the two forms of Vn was investigated by ELISA and showed that KatA bound both monomeric and polymeric Vn (Fig 3b). However, binding to the activated, polymeric Vn was...
nearly twice as strong. Given that H. pylori does not typically enter the bloodstream and Vn occurs predominantly in the activated form in tissues and the ECM, targeting mainly polymeric Vn is certainly an advantage for the bacterium. A similar preference for activated Vn has been described for other bacteria including Streptococcus pneumoniae and N. meningitidis. It should be noted that the activated form of Vn undergoes conformational changes and partial unfolding, and thereby exposes potential binding sites, which is likely a reason why the activated form is predominantly targeted by bacteria.
To obtain more detailed information on the binding kinetics, BioLayer Interferometry was performed using Vn as ligand and dilutions of full length KatA as analyte (Fig. 3c,d). We obtained fast association rate constants ($k_{\text{on}} [1/\text{Ms}] \approx 1 \times 10^6$) combined with moderate to slow dissociation rates ($k_{\text{off}} [1/\text{s}] \approx 3 \times 10^{-4}$). Accordingly, steady state analysis revealed a strong affinity ($K_D = 1 \times 10^{-9} \text{M}$), which is higher than the affinities determined for other Vn-binding proteins such as Moraxella catarrhalis ubiquitous surface protein (Usp) A2 ($2.34 \times 10^{-8} \text{M}$) and H. influenzae protein F ($1.28 \times 10^{-8} \text{M}$).
Figure 1. Identification of H. pylori Vn-binding proteins. Outer-membrane fractions of strains CCUG18943 (a) and KR697 (b) were subjected to 2D-analysis. Left panels show Coomassie stained gels, panels to the right show detection of Vn-binding proteins by far-Western blotting. Vn-binding spots were identified by MALDI-TOF as KatA (1) and UreA (2 and 3). Spot 4 corresponds to UreB.
There are two major bacterial binding regions within the Vn molecule. Most pathogens bind to the C-terminal HBD-3. The second binding region, used by *N. meningitidis* Vn-binding proteins Msf and Opc is located at the N-terminal end of the Vn molecule and the importance of sulphated tyrosine residues Y56 and Y59 was demonstrated for Opc binding to Vn26,27. Since KatA bound to Vn80–396, we suspected an interaction with HBD-3 without involvement of the N-terminus. When bacterial Vn-binding proteins utilise the C-terminal HBD-3, binding can be effectively blocked by heparin. Thus, we tested the influence of heparin on the Vn-KatA interaction in an ELISA (Fig. 4a), and pre-incubated Vn with increasing concentrations of heparin (0.1–100 μg/ml). Heparin inhibited binding of KatA to Vn by only 80%, and higher heparin concentrations had no stronger inhibitory effect (data not shown). This result suggests that, in contrast to most other known Vn-binding proteins, KatA does not use HBD-3 as a major binding site but may have additional binding region(s). To narrow down the binding site, binding of KatA to a range of truncated Vn molecules was tested by ELISA (Fig. 4b). KatA was immobilised in 96 well plates and incubated with different Vn-fragments, which were detected using anti-Vn antibodies (Abs). Removal of the C-terminus distal of HBD-3 (Vn 80–373) had no significant effect on binding when compared to Vn80–396. Similarly, Vn80–339, which lacks the whole HBD-3, showed only a minor decrease in binding KatA, which is in agreement with our observation that heparin had only a partial inhibitory effect. Strikingly, binding was substantially impaired with fragment Vn80–229 (Fig. 4b). Reciprocal experiments, i.e., when Vn-fragments were immobilised, incubated with KatA and binding was detected using anti-KatA Abs, gave similar results (data not shown). We observed some residual binding of Vn80–229 and cannot fully exclude involvement of residues located further upstream. However, the primary binding site for KatA is located within amino acids 229 and 339 of the Vn-molecule, a region not involved in interactions with any bacterial Vn-binding proteins described to date. The fact that heparin has an inhibitory effect on the interaction, even though HBD-3 is not the primary binding site, can be explained with the model structures of Vn prepared by two independent research groups28,29. Both models proposed a close proximity of the central domain and the C-terminal HBD resulting in a large inter-domain contact surface, which contains the putative heparin-binding groove. Given the spatial proximity between HBD-3 and our predicted KatA binding region within the central domain, it is feasible that bound heparin sterically disrupts the interaction between KatA and Vn rather than occupying the binding site itself. Docking of heparin to Vn28 supports this hypothesis. The use of an alternative binding site also explains why our kinetic data differed from data obtained for UspA2 and protein F, which both use HBD-3 as binding site22,24. In summary, *H. pylori* KatA uses a not previously characterised bacterial binding site on the Vn-molecule, which allows a high affinity interaction.
**The extended loop of KatA interacts with Vn.** *H. pylori* KatA, like other catalases, forms tetramers30. Each KatA monomer consists of 505 amino acids and exhibits the topology typical for small subunit, clade 3 catalases (Fig. 5a): an N-terminal protruding arm (aa 1–55), involved in formation of the homo-multimer, a central β-barrel domain (aa 56–315), and a C-terminal helical domain (aa 429–500), which is linked to the β-barrel domain by an extended ‘wrapping’ loop (aa 316–428)30. The unique feature of *H. pylori* catalase is a four lysine-motif at the C-terminus (aa 501–505), the function of which is at present unknown. Based on this topology a range of KatA fragments were designed, and their ability to interact with recombinant Vn was investigated by ELISA (Fig. 5b). KatA1–49 did not show any binding to Vn and, accordingly, KatA1–305 was not impaired in binding Vn when compared to the full-length protein (Fig. 5b). These results indicate that the N-terminal arm of KatA is dispensable for the interaction with Vn. This was expected, since the N-terminus is not exposed in the tetramer.
As mentioned above, *H. pylori* catalases comprise a tetra-lysine motif as part of an unstructured C-terminus. It was proposed that this motif could be involved in the transport of KatA to the surface or in anchoring KatA to the membrane; however, neither of these theories has to date been proven. We wanted to know whether this C-terminal motif mediates the interaction with Vn and tested a fragment comprising amino acids 51–488 (Fig. 5b). Indeed, there was a slight reduction in binding to Vn, albeit not statistically significant. Therefore, the C-terminal tetra-lysine motif is, if at all, only marginally involved in Vn binding. Removal of the central β-barrel and the N-terminal part of the wrapping loop resulted in a small, but not statistically significant, reduction in binding (Fig. 5b). Finally, we tested KatA400–488, which additionally lacked most of the wrapping loop. When this KatA fragment was included in our analysis a strong reduction of Vn binding was observed when compared to the full length KatA. Therefore, the region, which binds Vn, most likely is located within the central part of the extended wrapping loop of KatA (Fig. 5b, c). Disordered regions of proteins, which include loops, have a high propensity for involvement in protein–protein interactions, a recent example being the fibronectin protein FNE in *Streptococcus equi*.
KatA surface exposure differs between isolates and is correlated to Vn binding. Having identified the binding site for Vn within a flexible region, which are often less conserved, we wondered whether the differing binding properties of our three *H. pylori* isolates (Supplementary Fig. S1 and Fig. 2) could be attributed to diverging amino acid sequences. Interestingly, alignment of the KatA sequences of our three isolates and the reference strain *H. pylori* 22695 showed no differences in the relevant region between amino acid 350 and 400 (Supplementary Fig. S2). The differences in Vn binding might further be explained by a varying presence of KatA on the bacterial surface. We therefore employed flow cytometry analysis to determine KatA surface exposure in...
are other complement regulators such as Factor H or C4b binding protein (C4BP), that, like Vn, are regularly
Supplementary Fig. S1), Vn binding was not the predominant strategy to avoid complement attack. In fact, there
of both strains (Supplementary Fig. S4). We hypothesised that in this particular "low" Vn-binding strain (Fig. 2,
in serum sensitivity between KR497 wt and its corresponding Δ
katA mutant, and we observed moderate survival of both strains (Supplementary Fig. S4). We hypothesised that in this particular "low" Vn-binding strain (Fig. 2, Supplementary Fig. S1), Vn binding was not the predominant strategy to avoid complement attack. In fact, there are other complement regulators such as Factor H or C4b binding protein (C4BP), that, like Vn, are regularly
**KatA confers increased resistance to complement-mediated attack.** Many pathogens capitalise on components of the host's ECM to enhance adherence or evade the immune response. Vn binds to integrins via its RGD motif and can therefore mediate adherence to host tissues⁵. We tested adherence of *H. pylori* KR697 and CCUG18943 and their isogenic katA mutants to the gastric adenocarcinoma cell line AGS in the presence or absence of Vn (Supplementary Fig. S3). However, there was no significant difference between adherence of wild type and mutant strains irrespective of the addition of Vn, suggesting that the KatA-Vn interaction is irrelevant for adhesion of *H. pylori* to the gastric epithelium. In fact, the *H. pylori* adhesin CagL binds integrins via an RGD motif⁵, which makes the use of Vn as bridging molecule dispensable.
Another possible outcome of Vn binding is increased serum resistance, due to the inhibitory effect of Vn on the terminal pathway of the complement system⁶. Even though *H. pylori* does typically not enter the bloodstream, it is exposed to complement, since both complement factors and regulators, including Vn, are present in the gastric epithelium during *H. pylori* infection⁴,⁵. Furthermore, activation of complement in the presence of *H. pylori* has been demonstrated in vivo and in vitro²¹,²⁶. We obtained normal human serum (NHS) from healthy volunteers and performed serum killing assays. The survival rates of *H. pylori* KR697 and the KatA-deficient KR697ΔkatA in 5% NHS were determined over a time course of 60 min (Fig. 7a). Strikingly, no killing was observed in the wild type, but rather a slight increase in colony forming units (CFU) towards the end of the assay. On the contrary, *H. pylori* ΔkatA showed a marked decrease of 53% already after 15 min, which was, however, not statistically significant. At 30 min CFU counts were significantly reduced to 32% when compared to the wt (p < 0.05). After 60 min, the CFU/ml was reduced to 24% of the starting value (p < 0.001). Heat inactivated control serum, did not kill any of the strains (data not shown).
To corroborate that the observed difference of *H. pylori* wild type and ΔkatA with respect to complement resistance is due to the ability of KatA to interact with Vn, we depleted serum of the same pool from Vn and performed another killing assay (Fig. 7b). Indeed, removal of Vn resulted in killing of the wild type confirming that KatA confers increased complement resistance in a Vn-dependent manner (Fig. 7b).
To elucidate whether previous exposure to *H. pylori*, i.e., the presence of anti-Hp Abs, would influence the outcome of the KatA-Vn mediated complement resistance, we obtained sera from a different pool of donors, and tested them for the presence of anti-*H. pylori* IgG using a commercial whole cell ELISA and pooled the sera according to their positive or negative status. We thereafter investigated the survival of *H. pylori* KR697 wt and ΔkatA in 5% NHS for 60 min (Fig. 7c,d). When using the anti-*H. pylori* IgG negative serum, no killing of *H. pylori* KR697 wt was observed but bacterial counts were significantly reduced for *H. pylori* ΔkatA (Fig. 7c) at all time points (p < 0.001). We obtained similar results for strain CCUG18943 (Supplementary Fig. S4), even though both the wt and mutant were generally more sensitive to serum. Nevertheless, there were statistically significant differences in survival at all time points (p < 0.001). Intriguingly, there was no significant difference in serum sensitivity between KR497 wt and its corresponding ΔkatA mutant, and we observed moderate survival of both strains (Supplementary Fig. S4). We hypothesised that in this particular "low" Vn-binding strain (Fig. 2, Supplementary Fig. S1), Vn binding was not the predominant strategy to avoid complement attack. In fact, there are other complement regulators such as Factor H or C4b binding protein (C4BP), that, like Vn, are regularly
---
**Figure 4. KatA interacts with Vn via an unusual binding site.** (a) Inhibition of the KatA-Vn interaction by heparin was tested by ELISA. KatA (100 nM) was immobilised on plates and Vn⁸₀⁻³⁹₆ was added after being pre-incubated with different concentrations of heparin. (b) The ability of Vn-fragments to bind KatA was tested by ELISA. KatA (100 nM) was immobilised and incubated with different Vn-fragments at a concentration of 20 nM. Data shown are the mean and SD of three independent experiments performed in technical triplicate. Statistically significant differences were determined using one-way ANOVA and Bonferroni’s post-test where (*) equals p < 0.05 and (***)) equals p < 0.001.
11 of our previously tested strains including *H. pylori* CCUG18943, KR697, and KR497 (Supplementary Table 1). When the Vn-binding capacity was plotted against KatA surface expression we found a clear correlation of the two, i.e., overall, high Vn binders were also highly positive for KatA and vice versa (Fig. 6). We attribute the fact that some of our *H. pylori* isolates slightly diverged from this general pattern to the presence of other putative Vn-binding surface proteins in those particular strains. Taken together, our data demonstrate that the differences in KatA-dependent variations of Vn binding are not due to changes in the amino acid sequence of KatA but a result of varying surface exposure in addition to other putative Vn-binding proteins in some strains.
Figure 5. KatA binds vitronectin with its extended wrapping loop. Dimer of *H. pylori* KatA. Each monomer consists of an N-terminal arm (green), a central β-barrel domain (orange), the wrapping-loop (magenta), and the helical domain (blue). The C-terminal amino acids 492–505 are missing in the structure. (b) KatA-fragments were designed based on structural features shown in (a) and the Vn-binding capacity was compared to full length KatA using 20 nM Vn in an ELISA. Data shown are the mean and SD of at three independent experiments performed in technical triplicate. Statistically significant differences were determined using one-way ANOVA and Bonferroni’s post-test where (**) equals *p* < 0.001. (c) Surface structure-model of KatA dimer-surface depicting the area of the Vn-binding site (magenta).
attracted by bacteria to circumvent complement-mediated killing37. If such an additional mechanism was in place for KR497 it might easily mask the effect of the KatA-dependent Vn binding.
Next, we wanted to test survival of bacteria in anti-Hp positive serum. Intriguingly, in the presence of specific IgGs directed against _H. pylori_, we observed killing of both the wild type and ΔkatA strain with no statistically significant differences in the survival rates at any given time point (Fig. 7d). This observation was in agreement with other studies38,39, which observed an overall higher survival rate of _H. pylori_ in serum devoid of detectable anti- _H. pylori_ IgG than in serum, which contained specific antibodies. A possible explanation is the different modes of activation of the complement system. The alternative pathway, which is independent of specific antibodies, is constitutively activated at a low level. Presence of specific IgG or IgM, however, results in strong activation of the complement cascade via the classical pathway11. Thus, we conclude, that the KatA-Vn dependent complement resistance is highly relevant when complement is activated via the alternative pathway, but not sufficient to counteract the stronger activation through the classical pathway that might occur in patients that previously have been exposed to _H. pylori_.
As mentioned above, Vn interferes with formation of the MAC. Consequently, if _H. pylori_ used KatA to bind Vn, MAC deposition should be decreased on _H. pylori_ wt strains compared to those in which katA has been deleted. We therefore conducted an ELISA, in which we determined C9 deposition on wt and ΔkatA of CCUG18943, KR697, and KR497 after incubation with 5% anti-Hp IgG negative NHS (Fig. 8). Indeed, we observed a statistically significant increased MAC deposition on all three ΔkatA strains further confirming a role of KatA in complement resistance. Surprisingly, this finding also included strain KR497, where no significant difference was seen between wt and ΔkatA in the complement killing assays. This discrepancy is best explained with the different nature of the two assays. While the complement killing assay, which is based upon CFU count, gives information on differences in the number of culturable bacteria, the ELISA does not discriminate between viable but not culturable (VBNC) and culturable forms. It is currently impossible to revive coccoid VBNC _H. pylori_ under laboratory conditions, but there are hints that this form is still relevant for pathogenesis in the host40. Hence we cannot exclude a possible effect of a KatA-dependent complement resistance related to Vn for strain KR497, even though it might not be visible in a CFU based assay. Taken together, our data on complement resistance clearly demonstrate the importance of KatA in high and medium Vn-binding strains and even indicate involvement of KatA in Vn-mediated complement resistance in low Vn-binding strains.
This is the first report on identification of a _H. pylori_ Vn-binding protein and the beneficial consequences for the bacterium resulting from this interaction. Moreover, we have identified an alternative function of _H. pylori_ catalase with implications for evasion of the innate immune response and consequently virulence.
Methods
Bacterial strains and culture conditions. Bacterial strains used are listed in supplementary Table S2. _H. pylori_ was grown microaerophilic at 37 °C on GC agar or in Brucella broth supplemented with 1% haemoglobin, 10% horse blood, and 1% Vitox (OXOID). _E. coli_ BL21(DE3) was grown at 37 °C in LB or on LB-agar. Liquid cultures were grown at 200 rpm. Where applicable, 50 μg/ml kanamycin or 1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) was added.
Two-dimensional gel electrophoresis and far-Western blotting. Bacteria outer membrane proteins (OMPs) were prepared as described in Alteri and Mobley41. Two-dimensional (2D)-gel electrophoretic separation
of OMPs was performed essentially as described before42. OMPs (1mg) were incubated with 50 U benzonase at 25 °C for 30 min to digest nucleic acid. Then, samples were solubilized for 1 h in 400 μl of freshly prepared rehydration buffer (5 M urea, 2 M thiourea, 2% (w/v) CHAPS, 65 mM dithiothreitol (DTT), 0.5% (v/v) Pharmalyte pH 3–10, 1 mM phenylmethanesulfonylfluoride (PMSF) and 4 mM -(2-aminoethyl) benzenesulfonyl fluoride hydrochloride). Samples were centrifuged at 8,000 ¥ g for 10 min to remove any precipitates. First dimension separation by isoelectric focusing (IEF) was performed using the following program: 30 V for 0.30 h (15Vh), 500 V for 0.30 h (250 Vh), 1,000 V for 0.30 h (500 Vh), 2,600 V for 0.15 h (650 Vh), 3,500 V (875 Vh) and 5,000 V for 1.00 h (5,000 Vh) resulting in a total voltage of 7291 Vh, with maximum current of 50 μA per strip. Focused strips were equilibrated in equilibration buffer (50 mM Tris-HCl pH8.8, 6 M urea, 30% glycerol, 2% (w/v) SDS, 20 mM DTT, 0.002% bromophenol blue) for 15 min followed by incubation in the same buffer but replacing DTT with 20 mM iodoacetamide. For second dimension separation, strips were placed on 12% (v/v) SDS-polyacrylamide and proteins separated at 70 V for 15 min and 150 V for 40 min. Protein spots were visualized by Coomassie blue staining. Targeted Coomassie stained protein spots were manually excised and identified by Matrix-assisted laser desorption/ionisation-time of flight mass spectrometry (MALDI-TOF-MS) (Protein Analysis Service, Alphalyse). For far-Western blotting immunoassays, proteins were transferred to a polyvinylidenedifluoride membrane.
Figure 7. KatA increases complement resistance in a vitronectin-dependent manner. The resistance to serum complement of KR697 wt and ΔkatA was tested in a series of assays using 5% normal human serum (NHS) of unknown anti- H. pylori antibody status (a), NHS of unknown anti- H. pylori IgG status, which was depleted from vitronectin (VDS) (b), NHS of donors negative for anti- H. pylori IgG (c), and NHS of donors positive for anti- H. pylori IgG (d). Survival rates at 15, 30 and 60 min after addition of serum were determined by counting colony forming units (CFU). Depicted is the reduction in CFU/ml in percent. Results are the mean and SE of at least three independent experiments performed in technical duplicate.
Membranes were blocked in PBS containing 1.5% ovalbumin and probed with 3 μg/ml human Vn. Bound Vn was detected with sheep anti-Vn IgG (BioRad) and HRP-conjugated donkey anti-sheep/goat IgG (AbD Serotec).
Construction of katA deletion mutants. All primers and plasmids used in this study are listed in supplementary Table S3. Deletion mutants of katA were obtained for *H. pylori* strains KR697, KR497 and CCUG18943 by homologous recombination. Constructs containing a kanamycin cassette and 1000 bp flanks were created by overlap extension PCR. All primers used are listed in supplementary Table S2. The kanamycin cassette was amplified from pET26 using primers Kan_F and Kan_R. Downstream flanks of katA were amplified from genomic DNA of the respective strain using primers DF_katA_F2_Kan and DF_katA_R2. Upstream flanks of katA were amplified using primer pair UF_katA_F1 and UF_katA_R6_Kan for KR697 and KR497, and primer pair UF_katA_F1 and UF_katA_R4 for CCUG18943. For the overlap reaction, 10 ng of each PCR product were used with primer pair UF_katA_F1 and DF_katA_R2. Overlap PCR products were transformed into the respective *H. pylori* strain using natural competence. Briefly, freshly grown bacteria were resuspended in PBS to an OD_{600} = 5, spotted on GC agar without selection and incubated for 8 h before adding the 0.2 μg of PCR product. Plates were further incubated for 16 h. Growth spots were scraped of the plates, resuspended in PBS, plated on GC plates with kanamycin and incubated for 72 h. Resulting mutants were confirmed by PCR, Western blotting and loss of catalase function.
Flow cytometry analysis. Clinical isolates of *H. pylori* were grown under microaerophilic conditions in liquid cultures at 37 °C, 200 rpm until mid-log phase. To assess Vn binding, bacteria were incubated with 3 μg of human serum Vn per 10^6 cells for 1 h at 37 °C in PBS 1% BSA to allow Vn to bind bacteria. To remove unbound Vn, bacteria were washed twice in PBS. The amount of bound Vn was measured by flow cytometry (BD Verse) using a sheep anti-human Vn Ab (BioRad) and a fluorescein isothiocyanate (FITC)-conjugated anti-sheep pAb (BioRad) as a secondary layer. To determine KatA surface exposure, bacteria were blocked with PBS 1% BSA and incubated with 1 μg rabbit-anti-KatA for 1 h on ice. Bound anti-KatA was detected by flow cytometry using FITC-conjugated swine anti-rabbit pAb (Dako). Controls without Vn or Abs were included in all experiments to exclude non-specific binding. Data analysis was performed on FACS DIVA.
DNA manipulations and cloning. All primers and plasmids used in this study are listed in supplementary Table S3. Full length katA and katA-fragments were amplified by PCR from CCUG18943 genomic DNA using the following primer combinations: katA_for2 and katA_rev505 or katA_rev49 for full length katA or katA^{1–49}, respectively; katA_for51 and katA_rev505 or katA_rev488 for katA^{51–555} or katA^{51–488}, respectively. Finally, combinations katA_for350 and katA_rev505, and katA_for401 and katA_rev488 were used for katA^{350–505} and katA^{401–488}. Resulting products were digested with NdiI and XhoI and ligated into pET26 previously digested with the same enzymes. All constructs were confirmed by sequencing (MWG Eurofins). Constructs containing Vn-fragments were obtained as described earlier 43.
Purification of proteins and protein fragments. Activated Vn was purified from serum of healthy human volunteers as described in44. Briefly, serum was depleted of other heparin binding components before the heparin binding ability of Vn was activated by treatment with urease. Vn was thereafter purified by affinity chromatography using heparin as matrix. Recombinant full length Vn and Vn fragments were prepared as described previously 43. Briefly, HEK293T cells were grown in three triple flasks (Nunc) to 80% confluence using advanced DMEM supplemented with penicillin (100 U/μl) and streptomycin (100 μg/ml) and 1% FCS at 37 °C with 5% CO₂. Transfected cells were incubated for 3 d at 37 °C with 5% CO₂ followed by harvesting of supernatants. A similar volume of advanced DMEM was once again added to the cells and the procedure was repeated after 3 d. His-tagged Vn was secreted into the medium and purified by Ni-NTA chromatography. Recombinant KatA or
---
**Figure 8. Deposition of MAC is increased in *H. pylori* strains lacking katA.** Deposition of C9 on CCUG18943, KR697, and KR497 wt and ΔkatA strains was analysed by ELISA. Bacteria were coated onto 96-well Maxisorp plates and incubated with 5% NHS negative for anti-Hp IgG. C9 deposition was detected using specific antibodies. Statistically significant differences were determined using one-way ANOVA and Bonferroni’s post-test where (*) equals \( p < 0.05 \) and (**) equals \( p < 0.01 \). Data presented are the mean and SE of three independent biological replicates performed in duplicates.
KatA-fragments were expressed in BL21(DE3). Liquid cultures were grown to an OD_{600} = 0.5, induced with 1 mM IPTG, and grown for further 3 to 4 h before harvest by centrifugation. Cell pellets were resuspended in binding buffer (20 mM NaPO_{4}; 500 mM NaCl; 20 mM imidazole; pH 7.4), lysed by sonication, and centrifuged for 20 min at 4 °C and 23200 x g to separate soluble and insoluble fraction. Proteins were purified by Ni-affinity chromatography on His-Trap FF columns (GE Healthcare) and eluted in binding buffer supplemented with 500 mM imidazole. Fragment KatA^{350–505} was further purified by size exclusion using a Superdex 75 (10/300) column (GE Healthcare). Eluted fractions were analysed by SDS-PAGE and Coomassie staining. Fractions containing the desired protein were dialysed against 50 mM NaPO_{4}; 150 mM NaCl, pH 7.4 and stored at 4 °C until further use. For KatA1–49 the pellet containing the insoluble fraction was dissolved in binding buffer with 8 M urea and proteins were then refolded by dialysis against binding buffer prior to purification. All other proteins were purified from the soluble fraction. Catalase activity of full length KatA as well as KatA fragments containing the active site (KatA^{51–305} and KatA^{51–488}) was confirmed visually using H_{2}O_{2}. Anti-KatA IgG was purified from serum from rabbits immunized with recombinant KatA using KatA-coupled CNBr-activated Sepharose (GE healthcare). Bound anti-KatA polyclonal antibodies (pAb) were eluted with 100 mM glycine, pH 3.0 and immediately neutralized with 1.5 M Tris-HCl, pH 9.0. To reduce unspecific background anti-KatA IgG was absorbed with CCUG18943ΔkatA prior to use.
Assessment of protein interaction by enzyme-linked immunosorbent assay (ELISA). Polysorb 96-well plates (Nunc) were coated with 100 μl of 100 nM full length KatA or KatA fragments dissolved in coating buffer (100 mM Tris-HCl; 150 mM NaCl; pH 9.0) overnight at 4 °C. All following incubation steps were carried out at room temperature for 1 h. Plates were washed 4 times with PBS + 0.05% Tween20 between all incubations. First, plates were blocked with 2.5% BSA in PBS and thereafter incubated with recombinant Vn or Vn fragments at concentrations between 0–80 nM. Binding was detected using sheep anti-human-Vn (BioRad) pAb followed by horseradish-peroxidase (HRP)-conjugated rabbit anti-sheep pAb (AbD Serotec). To assess binding of KatA to monomeric or polymeric Vn, 50 nM Vn was coated to Polysorb plates as described above. After blocking, plates were incubated with 0–80 nM KatA. Bound KatA was detected using rabbit-anti-KatA pAb followed by HRP-conjugated swine anti-rabbit IgG (Dako). Data presented are means of at least three independent experiments performed in technical triplicates. Assays without Vn or KatA and controls consisting of secondary Abs only were included in all experiments.
BioLayer Interferometry. BioLayer Interferometry was carried out on an Octed RED96 platform (Fortebio). Vn^{30–396} was immobilised onto AR2G sensors at pH 5 using the Amine Coupling Reagent kit (Fortebio) according to the manufacturer’s instructions. Binding was tested using a 1 in 3 dilution series of KatA (50–0.055 nM) in PBS, 0.05% Tween20 (Sigma). A buffer only control served as reference well. Data analysis was performed using ForteBio Data Analysis 8.1. and Prism GraphPad version 5. Steady state analysis was performed based on the equilibrium response.
Serum resistance assays. Normal human serum (NHS) was obtained from healthy volunteers of unknown *H. pylori* infection status. Depletion of Vn was accomplished by running serum against anti-Vn-coupled CNBr Sepharose as described previously. Western blotting was used to confirm depletion of Vn and to ensure that levels of serum-antibodies were unaffected by the depletion process. Sera were tested for presence of IgG against *H. pylori* using the whole cell anti-*H. pylori* (ATCC43504) ELISA (Euroimmun) and pooled according to the IgG status. Heat-inactivation was performed for 30 min at 56 °C. To investigate complement-mediated killing of *H. pylori* KR697 or KR697Δ grown in liquid culture were incubated in DGVB (ForteBio). Vn^{80–396} was immobilised onto AR2G sensors at pH 5 using the Amine Coupling Reagent kit and diluted to 2*10^7 CFU/ml in PBS. 1*10^7 CFU in 50 μl bacterial suspension were immobilised over night at 4 °C and Vn proteins were then refolded by dialysis against binding buffer prior to purification. All other proteins were purified from the soluble fraction. Catalase activity of full length KatA as well as KatA fragments containing the active site (KatA^{51–305} and KatA^{51–488}) was confirmed visually using H_{2}O_{2}. Anti-KatA IgG was purified from serum from rabbits immunized with recombinant KatA using KatA-coupled CNBr-activated Sepharose (GE healthcare). Bound anti-KatA polyclonal antibodies (pAb) were eluted with 100 mM glycine, pH 3.0 and immediately neutralized with 1.5 M Tris-HCl, pH 9.0. To reduce unspecific background anti-KatA IgG was absorbed with CCUG18943ΔkatA prior to use.
Measurement of C9 deposition by ELISA. Bacteria were grown to mid-log phase, harvested, washed, and diluted to 2·10^7 CFU/ml in PBS. 1·10^7 CFU in 50 μl bacterial suspension were immobilised over night at 4°C in MaxiSorp 96 well plates (NUNC). Next day, plates were washed thrice using immunowash (20 mM NaPO_{4}, 500 mM NaCl; pH 7.4) and blocked with quench (3% fish gelatine and 0.02% NaN_{3} in immunowash) for 1 h at 37°C. After every incubation step, the plates were washed three times with 250 μl/well immunowash. Five per cent NHS in DGVB + + buffer (72 mM NaCl, 0.9 mM sodium barbital, 1.55 mM barbituric acid, 0.1% gelatine, 1 mM MgCl_{2}, 0.15 mM CaCl_{2}, 2.5% dextrose) with 5% serum at 37 °C, 5% CO_{2} for 60 min. To exclude effects of the incubation conditions on the viability of the bacteria, serum-free controls were run in parallel and CFUs determined at the start and end of the assay (data not shown). Samples were plated 0, 15, 30, and 60 min after serum was added and the colony forming units (CFU/ml) were determined.
Statistical analysis. All statistical analyses were performed using Graph-Pad Prism® version 5.0 (GraphPad Software, La Jolla, CA). To determine statistical differences between obtained data sets, Student’s t-test (flow cytometry analysis of Vn-binding of wt versus ΔkatA strains), one-way ANOVA with Bonferroni’s post-test (ELISA), or two-way ANOVA with Bonferroni’s post-test (serum resistance assays) was performed. Differences between experiments were considered statistically significant for p < 0.05.
40. Segal, E. D., Falkow, S. & Tompkins, L. S. *Helicobacter pylori* attachment to gastric cells induces cytoskeletal rearrangements and tyrosine phosphorylation of host cell proteins. *Proceedings of the National Academy of Sciences of the United States of America* **93**, 1259–1264 (1996).
41. Alteri, C. J. & Mobley, H. L. Quantitative profile of the uropathogenic *Escherichia coli* outer membrane proteome during growth in human urine. *Infection and immunity* **75**, 2679–2688, doi: 10.1128/IAI.00076-06 (2007).
42. Fleury, C. *et al.* Identification of a *Haemophilus influenzae* factor H-binding lipoprotein involved in serum resistance. *J Immunol* **192**, 5913–5923, doi: 10.4049/jimmunol.1303449 (2014).
43. Singh, B. *et al.* *Haemophilus influenzae* protein E recognizes the C-terminal domain of vitronectin and modulates the membrane attack complex. *Molecular microbiology* **81**, 80–98, doi: 10.1111/j.1365-2958.2011.07678.x (2011).
44. Akiyama, S. K. Purification of vitronectin. Current protocols in cell biology/editorial board, Juan S. Bonifacino … [et al.] **60**, Unit 10–16, doi: 10.1002/0471143030.ch1096s60 (2013).
45. Al-Jubair, T. *et al.* *Haemophilus influenzae* Type f Hijacks Vitronectin Using Protein H To Resist Host Innate Immunity and Adhere to Pulmonary Epithelial Cells. *J Immunol*, doi: 10.4049/jimmunol.1501197 (2015).
**Acknowledgements**
We would like to thank Klaudyna Teslar, Marta Brant and Kerstin Norrman for excellent technical support and the other members of the Riesbeck laboratory for helpful discussions during preparation of this manuscript. This work was supported by grants from the Alfred Österlund, Anna and Edwin Berger, and Greta and Johan Kock foundations, the Swedish Medical Research Council (K2012-66X-14928-09-5, http://www.vr.se), the Physiological Society (Forssman’s Foundation), Skåne County Council’s research and development foundation and the Swedish Cancer Foundation (CAN2013-670). We also thank the Clinical Microbiology laboratory at Labmedicin Skåne (Malmö, Sweden) for providing clinical isolates.
**Author Contributions**
C.R. was involved in planning the study, prepared Figs 3–6 and wrote the manuscript. O.M. prepared Figure 2, S1–3. D.E. prepared Figure 8. Y.C.S. prepared Figure 1 and contributed to Figure 6. B.S. contributed to Figures 3–5. V.A. and A.M.B. were involved in planning the study and in experimental work. K.R. planned the study and wrote the manuscript. All authors reviewed the manuscript.
**Additional Information**
**Supplementary information** accompanies this paper at http://www.nature.com/srep
**Competing financial interests:** The authors declare no competing financial interests.
**How to cite this article:** Richter, C. *et al.* Moonlighting of *Helicobacter pylori* catalase protects against complement-mediated killing by utilising the host molecule vitronectin. *Sci. Rep.* 6, 24391; doi: 10.1038/srep24391 (2016).
This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/ | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3903,
1
],
[
3903,
10622,
2
],
[
10622,
12331,
3
],
[
12331,
16653,
4
],
[
16653,
18787,
5
],
[
18787,
25010,
6
],
[
25010,
25817,
7
],
[
25817,
29785,
8
],
[
29785,
32142,
9
],
[
32142,
37057,
10
],
[
37057,
43690,
11
],
[
43690,
43690,
12
],
[
43690,
47121,
13
]
]
} |
2e8b4e7cc883999c3a27f25401e5756ff05556a0 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 7,
"total-fallback-pages": 0,
"total-input-tokens": 23278,
"total-output-tokens": 8830,
"fieldofstudy": [
"Medicine",
"Biology"
]
} | Predictive urinary RNA biomarkers of kidney injury after extracorporeal shock wave lithotripsy
Ahmed Tawfick1 · Marwa Matboli2 · Sara Shamloul2 · Sara H. A. Agwa3 · Maha Saad4 · Hassan Shaker1 · Mohamed M. Yassin Selim1 · Mohamed S. Salim1 · A. Radwan1 · A. A. Shorbagy1 · Waleed Mousa1
Received: 18 December 2021 / Accepted: 18 March 2022 / Published online: 15 April 2022
© The Author(s) 2022
Abstract
Background Extracorporeal shock wave lithotripsy (ESWL) is considered one of the best choices for the treatment of various kinds of urinary tract calculi, although it might cause acute kidney injury.
Objective To measure the urinary long non-coding RNA-messenger RNA (LncRNA-mRNA) panel before and after ESWL to evaluate post-ESWL renal injury in a reliable and non-invasive method.
Patients and methods The study included 60 patients with renal stones treated with ESWL and 30 healthy volunteers. Voided urine samples were obtained before, 2 h, and 1 day after ESWL. We measured the urinary level of LncRNA (SBF2-AS1, FENDRR-19) and mRNA (GBP1, NLRP3) by real-time qPCR and compared the results with serum creatinine and eGFR.
Results LncRNA (SBF2-AS1, FENDRR-19) and mRNA (GBP1, NLRP3) levels were higher in patients with renal stones when compared with healthy volunteers. They showed a statistically significant increase in the level of LncRNA-mRNA panel in baseline and after ESWL treatment.
Conclusion LncRNA (SBF2-AS1, FENDRR-19) and mRNA (GBP1, NLRP3) levels were significantly elevated following ESWL treatment, highlighting the usefulness of urinary biomarkers in identifying patients at higher risk of developing renal injury after ESWL treatment.
Keywords LncRNA · ESWL · SBF2-AS1 · FENDRR-19 · NLRP3 · GBP1
Introduction
The prevalence of urinary stones is around 3–12% and its recurrence rates can reach up to 50% within 10 years [1]. ESWL is considered the first line of treatment together with retrograde intrarenal surgery (RIRS) for stones smaller than 2 cm [2]. ESWL can cause the injury of thin-walled renal vessels, leading to transient haematoma, the release of cytokines/inflammatory cellular mediators and the infiltration of tissue by inflammatory response cells [3].
Serum creatinine and blood urea nitrogen levels, which are routinely used to detect and follow the progression of renal injury, are insensitive, nonspecific, and get higher only after significant kidney damage [4]. Till now, there is no consistent dependable urinary marker to permit the detection of acute kidney injury.
Inflammasome signaling regulates caspase-dependent inflammation and apoptosis. Several inflammasome-linked genes have a crucial role in kidney diseases, especially the NLRP3 (NOD-, LRR- and pyrin domain-containing 3)
which contributes to many acute and chronic renal diseases [5]. Guanylate Binding Proteins (GBP) seem to play a critical role in inflammasome activation. They are expressed in immune cells and the stroma of the lung, kidney, and brain [6].
Circulating urinary long non-coding RNAs (lncRNAs) are fascinating novel biomarkers that reflect intra-nuclear processes noninvasively and may thus provide a better estimate of intracellular processes than currently established biomarkers [7]. lncRNAs are transcripts with a length of more than 200 nucleotides that exhibit tissue-specific expression and are involved in epigenetic regulation [8]. In several studies, circulating lncRNAs have been described as a fascinating new player in pathophysiological studies and the search for novel diagnostic and therapeutic strategies in kidney disease [9–11].
The four selected RNAs gene ontology is not linked only to inflammation, but also extended to different injury response patterns, e.g., renal apoptosis, GTPase activity, and ischemia reperfusion after ESWL-induced kidney injury. Thus, their differential expression level not only reflect the inflammation in kidney injury, but also highlight their role in ischemic renal tissue and apoptotic tissues [12].
In this study, we hypothesized that an RNA panel linked to the inflammasome system and specific to kidney injury could be used as a potential biomarker panel, as combined lncRNA-mRNA panels are more informative than single RNA. We first identified inflammasome-related genes and their epigenetic regulators via in silico data analysis. Then, to confirm this panel, we assessed the differential expression of lncRNA [SBF2-AS1 (SET binding factor 2 antisense RNA1) and FENDRR-19 (Fetal-lethal non-coding developmental regulatory RNA)] and mRNA [NLRP3 (NOD-like receptor and pyrin domain-containing 3) and GBP1 (guanylate-binding protein 1)] in the urine of renal stone patients treated with ESWL, then we compared with healthy volunteers to evaluate their usefulness as diagnostic biomarkers for post-ESWL kidney injury.
Patients and methods
The study was carried out after approval from Ain Shams Faculty’s Medicine Ethical Committee from December 2020 to July 2021. The participants were 60 patients whose gender and ages matched with 30 healthy volunteers with no history of kidney or stone disease. All patients underwent ESWL; for the first time to treat radiopaque stone(s) of 2 cm in diameter or less, located in the kidney. They were recruited from the outpatient clinic of the Urology Department of Ain Shams University hospitals. Informed consent was provided by all participants. The level of kidney disease was staged according to the National Kidney Foundation Kidney Disease Outcomes Quality Initiative Classification [13]. Patients with a history of active urinary tract infection, bleeding disorders, elevated serum creatinine, chronic renal failure (eGFR < 30) and pregnant females were excluded from the study (Supplementary Table 1S).
The procedure was done with 3500 shock waves and the frequency of shock waves was set at 60 shocks per minute.
Each patient provided three urine samples in centrifugal tubes (2 h before, then 2 h and 24 h after ESWL) and blood samples. Urine was then centrifuged at 4000 rpm for 10 min, and the urinary pellet was washed twice with phosphate-buffered saline. The resultant urine pellet was preserved at −80 °C. Sera samples were collected and stored within 15 min at a temperature of −80 °C.
We have selected the NLRP3 mRNA gene, which is highly correlates with inflammation and is essential for proper inflammasome formation and processing. Firstly, we selected an inflammatory pathway closely linked to kidney injury using biosystems available at the NCBI gene database (available at ncbi.nlm.nih.gov/gene) (Supplementary Fig. 1S). Secondly, we selected the NLRP3 mRNA gene, which is closely linked to the inflammasome using the Reactome database (Supplementary Fig. 2S) (available at https://reactome.org/content/detail/R-HSA-844456).
Then, the gene’s ontology was verified (Supplementary Fig. 3S) (available at https://www.ncbi.nlm.nih.gov/gene?Db=Gene&Cmd=DetailsSearch&Term=114548) followed by validating its basal expression in the kidney using the Gene Cards database (available at https://www.genecards.org/cgi-bin/carddisp.pl?gene=NLRP3) (Supplementary Fig. 4S). Thirdly, we selected SBF2-AS1 targeting NLRP3 mRNA using the “lncRNA2target” database (available at 123.59.132.21/lncrna2target/search.jsp) (Supplementary Fig. 5S). The selection of lncRNA is based on how strongly it interacts with mRNA, the novelty in kidney disease, and their basal expression in the kidney (available at https://www.genecards.org/cgi-bin/carddisp.pl?gene=SBF2-AS1&keywords=SBF2%5C-AS1) (Supplementary Fig. 6S).
Fourthly, we retrieved data on the GBP1 mRNA gene, which is involved in cytokine binding and inflammasome signaling, and was correlated with kidney injury (Supplementary Fig. 7S) (available at https://www.ncbi.nlm.nih.gov/gene/2633). We verified its expression in the Genecards database (available at genecards.org/cgi-bin/carddisp.pl?gene=GBP1&keywords=GBP1) (Supplementary Fig. 8S). Finally, we used the “lncRNA2target” database to find a related lncRNA LincFOXF1 (lncRNA-FENDRR:19) (ENSG00000268388) which is supposed to control the expression of the GBP1 gene (Supplementary Fig. 9S), followed by verifying its expression in the kidney (available at 123.59.132.21/lncrna2target/search.jsp) (Supplementary Fig. 10S).
Total RNA was extracted from the urine pellet using an miRNEasy RNA isolation kit (Qiagen, Hilden, Germany) according to the manufacturer’s protocol. Total RNA samples were dissolved in 30 µl of nuclease-free water. Quality and quantity were checked using NanoDrop spectrophotometer. cDNA libraries for mRNAs, LncRNAs and miRNAs were prepared using the miScript II RT Kit (Qiagen, Germany). 4ul of 5 × miScript HiFlex Buffer, 2ul of 10 × miScript Nucleics Mix, 1ul of miScript Reverse Transcriptase Mix and RNase free water were added to 2ug of RNA extract, then incubated at 37 °C for 60 min and at 95 °C for 5 min using the Rotor-Gene thermal cycler (Thermo Electron Waltham, MA).
LncRNA (SBF2-AS1, FENDRR19) and mRNA (NLRP3, GBP1) expressions in urine samples of diseased groups and healthy control groups were quantified by qRT-PCR using QuantiTect SYBR-Green PCR Master Mix (Qiagen, Germany). 4ul of 10 × miScript Nucleics Mix, 1ul of miScript Reverse Transcriptase Mix and RNase free water were added to 2ug of RNA extract, then incubated at 37 °C for 60 min and at 95 °C for 5 min using the Rotor-Gene thermal cycler. 2 − ΔΔCt method. The Rotor-Gene real-time PCR detection system (Qiagen, Hilden, Germany) calculated the threshold cycles (Ct) value of each sample. The value was considered negative if higher than 36 Ct value. The amplification plot curve and melting curve were analyzed to confirm the specificity of the amplicons and Tm values.
The real-time cycler was programmed as follows: initial activation step at 95 °C for 15 min to activate HotStarTaq DNA Polymerase. 40 cycles of PCR were performed under the following conditions: at 94 °C for 15 s, 55 °C for 30 s and 72 °C for 30 s for extension, denaturation and annealing sequentially. Each reaction was carried out three times. Fold change and expression levels were calculated using the 2 − ΔΔCt method. The Rotor-Gene real-time PCR detection system (Qiagen, Hilden, Germany) calculated the threshold cycle (Ct) value of each sample. The value was considered negative if higher than 36 Ct value. The amplification plot curve and melting curve were analyzed to confirm the specificities of the amplicons and Tm values.
The data were statistically presented using SPSS 20. Independent t-test, chi-square test, and Mann Whitney test were used. The (ROC) curve was drawn to characterize the predictive value of the selected RNA-based biomarker panel for post ESWEL kidney injury. The Spearman correlation was carried out to detect the association between clinico-pathological parameters and RNA panel expression. A two-tailed p value of 0.05 or less was supposed to be statistically significant.
Results
In this study, there were no statistically significant differences between the two investigated groups regarding age, sex, body mass index, Serum Creatinine and eGFR and hypertension (p > 0.05).
Details of the demographic and clinical data are shown in (Table 1).
We reported that the positivity rates for urine LncRNA (SBF2-AS1, FENDRR19) and mRNA (GBP1, NLRP3) significantly increased in 24 h post-ESWL. Baseline pre-ESWL voided urine samples collected from patients with renal stones revealed significantly higher positive rates in comparison with the healthy volunteers’ voided urine (p < 0.01). We found that the four RNAs of all patients significantly increased in voided urinary specimens collected 2 and 24 h after ESWL compared with their pre-procedure baseline levels (Table 2, Supplementary Fig. 11S).
Urinary markers continued to rise significantly 1 day after ESWL (p < 0.01). Baseline pre-ESWL voided urine samples collected from patients with renal stones revealed significantly higher RNAs levels when compared to healthy volunteers’ voided urine (p < 0.01) (Supplementary Fig. 12S).
ROC curve analysis and the area under the curve (AUC) values were used to estimate the discriminative power of our selected RNAs between patients with renal stone versus the control group (as illustrated in Supplementary Fig. 13S).
Comparing the diseased groups with healthy control groups shows that the best discriminating cutoff values of LncRNA (SBF2-AS1, FENDRR19) and mRNA (GBP1, NLRP3) were 1.250, 1.250, 1.250 and 1.360, respectively. The sensitivities measured 91.7%, 76.7%, 78.3% and 78.3%, respectively. Accordingly, this result indicates that these thresholds could be used to differentiate/identify diseased patients from healthy subjects (Supplementary Table 2S).
There was a highly significant correlation between all the investigated urine RNAs and serum creatinine. Also, there was a highly significant correlation between serum eGFR and FENDRR19, as well as a significant correlation between serum eGFR and SBF2-AS1.
By studying the correlation between the studied RNA-based biomarkers, we found that there was a highly significant positive correlation between all of them [(LncRNA (SBF2-AS1, FENDRR19) and mRNA (GBP1, NLRP3)] based on fold changes (R.Q.) among all the study groups. Results are shown in Supplementary Table 3S.
No baseline characteristic variables were found to have a significant association by the multilinear regression model. All the tested predictor variables are presented in Supplementary Table 4S. mRNA-NLRP3 and serum creatinine is the most significant predictor of kidney injury (p = 0.000,0.002, respectively).
Discussion
ESWL subjects the renal parenchyma to high levels of energy, leading to a broad spectrum of vascular kidney damage ranging from self-limited hematuria to perinephric/nephric hematomas [11]. In the long term, 11 animal and 13 human studies have suggested that these acute hemorrhagic lesions may progress to scar formation and complete the atrophy of the renal papillae. There is no existing adequate imaging modality available to assess the parenchymal injury, thus creating a need for a potential novel diagnostic test that can reliably detect such renal injuries [14].
We identified that the level of expression of LncRNA (SBF2-AS1, FENDRR19) and mRNA (GBP1, NLRP3)) are highly detected in the urine of post ESWL-procedure patients. This has raised the possibility of using this network as a circulating biomarker for post ESWL renal injury detection. Also, as a part of inflammation, the results of this study revealed that urine NLRP3-mRNA was significantly upregulated in diseased groups compared to normal healthy individuals in the controlled groups (p < 0.01), and it is differentially expressed after overexpression of urine lncRNA SBF2-AS1. Previous reports on cell lines and tissue also indicated that interfering with the process of NLRP3 inflammasome activation can regulate kidney injury [15]. Furthermore, activation of NLRP3 inflammatory corpuscles could promote AKI induced by sepsis. Simultaneously, a renal injury may lead to the production of mitochondrial reactive...
oxygen species (mROS), which may induce the binding of TXNIP to NLRP3 [16].
With regards to GBP1-mRNA, the results of our study revealed that its expression was significantly upregulated in AKI patients compared to healthy individuals (p<0.01) and it is differentially expressed after overexpression of urine lncRNA FENDRR19. Honkala et al. declared that GBP1 governs cellular responses to infection, inflammation, and environmental stressors [17]. At the cellular level, GBP1 activation both restrains proliferation and protects against apoptosis in inflammatory contexts [18].
LncRNAs play a critical role in immunity as they regulate the survival, differentiation and cytokine formation of immune cells [19]. LncRNAs may participate in epigenetic regulation for proinflammatory and anti-inflammatory gene expression in macrophages challenged by inflammatory mediators [20]. Overexpression of lncRNA genes or deficiency has been involved in kidney diseases; they not only function as biomarkers, but also as pathogenic mediators of kidney diseases. Thus, LncRNAs associated with kidney disease identification and characterization may provide new diagnostic and therapeutic opportunities for renal disorders [21].
FENDRR lncRNA (Foxf1 adjacent non-coding developmental regulatory RNA) plays a significant role in heart development [22]. Çekin and his colleagues found that FENDRR expression was lower in coronary artery disease [23]. Interestingly, Munteanu et al.’s results suggested that FENDRR promoted polarization of M1 macrophage and so, targeting FENDRR may act as a potential therapeutic target for the treatment of diseases that occurred with polarization macrophage [24].
| Variables | Pre-procedure No. = 60 | 2 h post-procedure No. = 60 | 24 h post-procedure No. = 60 | Healthy control No. = 30 | p value |
|--------------------------------|------------------------|-----------------------------|-------------------------------|--------------------------|---------|
| Serum Creatinine in mg/dL Mean ± SD | 0.76 ± 0.05 | 0.77 ± 0.06 | 0.89 ± 0.10 | 0.79 ± 0.06 | < 0.001** |
| cGFR in mL/min/1.73m² Mean ± SD | 96.73 ± 19.21 | 97.86 ± 19.89 | 92.17 ± 15.09 | 106.1 ± 10.12 | 0.001** |
| lncRNA-SBF2-AS1 Median (IQR) | 1.64 (1.30–2.57) | 2.36 (1.91–2.42) | 30.89 (16.40–58.85) | 0.97 (0.90–1.00) | < 0.001** |
| lncRNA-FENDRR-19 Median (IQR) | 1.05 (0.40–1.80) | 1.89 (1.61–2.42) | 3.11 (2.50–3.70) | 0.97 (0.90–1.00) | < 0.001** |
| mRNA-GBP1 Median (IQR) | 1.00 (0.80–1.50) | 2.12 (1.42–2.65) | 5.75 (4.15–26.70) | 0.97 (0.90–1.00) | < 0.001** |
| mRNA-NLRP3 Median (IQR) | 1.40 (1.22–1.90) | 3.50 (2.15–4.05) | 6.95 (4.80–9.72) | 0.97 (0.90–1.00) | < 0.001** |
Mann–Whitney test, IQR Inter Quartile Range, (a): One-Way ANOVA test, (b): Kruskal–Wallis test, p: p value, **p < 0.01: Highly Significant, *p < 0.05: Significant, p > 0.05: Non-Significant, n=90
| Variables | Pre-procedure N (%) | 2 h post-procedure N (%) | 24 h post-procedure N (%) | Healthy control N (%) | p value |
|--------------------------------|---------------------|---------------------------|---------------------------|-----------------------|---------|
| Positive Serine creatinine | 27 (45%) | 33 (55%) | 54 (90%) | 15 (50%) | <0.001** |
| Negative Serine creatinine | 33 (55%) | 27 (45%) | 6 (10%) | 15 (50%) | |
| Positive c. GFR | 33 (55%) | 24 (40%) | 27 (45%) | 0 (0%) | <0.001** |
| Negative c. GFR | 27 (45%) | 36 (60%) | 33 (55%) | 30 (100%) | <0.001** |
| Positive lncRNA-SBF2-AS1 | 48 (80%) | 57 (95%) | 60 (100%) | 0 (0%) | |
| Negative lncRNA-SBF2-AS1 | 12 (20%) | 3 (5%) | 0 (0%) | 30 (100%) | |
| Positive lncRNA-FENDRR-19 | 24 (40%) | 54 (90%) | 60 (100%) | 0 (0%) | <0.001** |
| Negative lncRNA-FENDRR-19 | 36 (60%) | 6 (10%) | 0 (0%) | 30 (100%) | <0.001** |
| Positive mRNA-GBP1 | 24 (40%) | 57 (95%) | 60 (100%) | 0 (0%) | |
| Negative mRNA-GBP1 | 36 (60%) | 3 (5%) | 0 (0%) | 30 (100%) | |
| Positive mRNA-NLRP3 | 33 (55%) | 48 (80%) | 60 (100%) | 0 (0%) | 0.001** |
| Negative mRNA-NLRP3 | 27 (45%) | 12 (20%) | 0 (0%) | 30 (100%) | |
Table 2 Positivity rate and mean urinary biomarker concentrations with pre, 2, 24 h post ESWL treatment and healthy control
Of note, lncRNA SBF2-AS1 is located at the chromosome 11p15.1 locus. Several studies showed that lncRNA SBF2-AS1 might act as an essential regulator of tumor progression [25–27]. LncRNA SBF2-AS1 induces hepatocellular carcinoma metastasis by regulating epithelial-mesenchymal transition [28].
The results of our study revealed that urinary LncRNA (SBF2-AS1, FENDRR19) and mRNA (GBP1, NLRP3) were significantly upregulated in AKI patients when compared to healthy control groups. Interestingly, pre ESWL level of urinary SBF2-AS1; and to lesser extent the rest of urinary RNAs measured, is significantly higher than that of healthy control. This can be attributed to kidney injury because of stone obstruction. Followed by sharp rise of the 4 RNA markers 24 h post ESWL compared to pre ESWL level in the patients group.
Accuracy of post ESWL renal injury detection can be improved by measuring urine LncRNA (SBF2-AS1, FENDRR19) and mRNA (GBP1, NLRP3). Pointedly, there was a highly significant positive correlation between urine [LncRNA (SBF2-AS1, FENDRR19) and mRNA (GBP1, NLRP3)] based on fold changes (R.Q.s) among the human study groups.
We have tried to identify baseline characteristic variables that could identify risky patients with significant renal injury after ESWL. We performed a univariate and multivariate analysis. Serum creatinine, LncRNA-FENDRR19 and NLRP3 mRNA were significant independent variables that significantly correlated with renal injury.
Limitations of the study include its inclusion of a relatively small sample size. Moreover, in vitro and in vivo functional analyses are needed to clarify the biological mechanisms of RNA-RNA crosstalk in post ESWL renal injury by assessment of the selected RNAs in rat AKI animal model for further verification.
Conclusion
Urine LncRNA (SBF2-AS1, FENDRR19) and mRNA (GBP1, NLRP3) rise significantly post ESWL. Hence, they may be prospective clinical markers for assessing acute renal injury following ESWL.
Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/s00345-022-03996-3.
Author contributions AT: Conceptualization, data collection, fund acquisition, methodology. MM: Conceptualization, methodology, data curation, formal analysis, funding acquisition, data analysis, manuscript editing. SS: Data analysis, Statics, manuscript writing, investigation. SHAA: Data curation, investigation. MS: Data analysis, investigation. HS: Conceptualization, funding acquisition, methodology. MMYS: Investigation, ESWL procedure. WM: Conceptualization, project development, fund acquisition, manuscript editing, corresponding author.
Funding Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB). This work was supported by Science, Technology and Innovation Funding Agency (STDF), Project No. 28889.
Declarations
Conflict of interest All the authors declare no competing interests.
Ethical approval Written informed consent was obtained from all the participants of this study, which was performed following the Declaration of Helsinki, and was approved by the Research Ethics Committee of Ain Shams University, Faculty of Medicine, Egypt.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
References
1. Preminger GM, Tiselius HG, Assimos DG, Alken P, Buck C, Gallicci M, Knoll T, Lingeman JE, Nakada SY, Pearle MS, Sarica K (2007) guideline for the management of ureteral calculi. J Urol 178(6):2418–2434
2. Türk C, Petrik A, Sarica K, Seitz C, Skolarikos A, Straub M, Knoll T (2016) EAU guidelines on interventional treatment for urolithiasis. Eur Urol 69(3):475–482
3. Connors BA, Evan AP, Blomgren PM, Hsi RS, Harper JD, Sorensen MD, Wang YN, Simon JC, Paun M, Starr F, Cunlitz BW (2014) Comparison of tissue injury from focused ultrasonic propulsion of kidney stones versus extracorporeal shock wave lithotripsy. J Urol 191(1):235–241
4. Bryniarski P, Paradyasz A, Zyczkowski M, Kupilas A, Nowakowski K, Bogacki R (2012) A randomized controlled study to analyze the safety and efficacy of percutaneous nephrolithotripsy and retrograde intrarenal surgery in the management of renal stones more than 2 cm in diameter. J Endourol 26(1):52–57
5. Komada T, Muruve DA (2019) The role of inflammasomes in kidney disease. Nat Rev Nephrol 15(8):501–520
6. Kim BH, Chee JD, Bradfield CJ, Park ES, Kumar P, Mackinger JD (2016) Interferon--induced guanylate-binding proteins in inflammasome activation and host defense. Nat Immunol 17(5):481–489
7. Martens-Uzunova ES, Böttcher R, Croce CM, Jenster G, Visakorpi T, Calin GA (2014) Long noncoding RNA in prostate, bladder, and kidney cancer. Eur Urol 65(6):1140–1151
8. Cabili MN, Trapnell C, Goff L, Koziol M, Tazon-Vega B, Regev A, Rinn JL (2011) Integrative annotation of human large
intergenic noncoding RNAs reveals global properties and specific subclasses. Genes Dev 25(18):1915–1927
9. Wang X, Xu Y, Zhu YC, Wang YK, Li J, Li XY, Ji T, Bai SJ (2019) LncRNA NEAT1 promotes extracellular matrix accumulation and epithelial-to-mesenchymal transition by targeting miR-27b-3p and ZEB1 in diabetic nephropathy. J Cell Physiol 234(8):12926–12933
10. Li Y, Ren D, Xu G (2019) Long noncoding RNA MALAT1 mediates high glucose-induced glomerular endothelial cell injury by epigenetically inhibiting klotho via methyltransferase G9a. IUBMB Life 71(7):873–881
11. Ignarski M, Islam R, Müller RU (2019) Long non-coding RNAs in kidney disease. Int J Mol Sci 20(13):3276
12. Wu YL, Li HF, Chen HH, Lin H (2020) MicroRNAs as biomarkers and therapeutic targets in inflammation-and ischemia-reperfusion-related acute renal injury. Int J Mol Sci 21(18):6738
13. Levey AS, Coresh J, Bolton K, Culleton B, Harvey KS, Islam R, Müller RU (2019) Long non-coding RNAs in kidney disease. Int J Mol Sci 20(13):3276
14. Fahmy N, Sener A, Sabbisetti V, Nott L, Lang RM, Welk BK, Méndez-Probst CE, MacPhee RA, VanEerdewijk S, Cadieux PA, Bonventre JV (2013) Urinary expression of novel tissue markers of kidney injury after ureteroscopy, shockwave lithotripsy, and in normal healthy controls. Am J Kidney Dis 39(2 Suppl 1):i-ii+
15. Xiang H, Zhu F, Xu Z, Xiong J (2020) Role of inflammasomes in kidney diseases via both canonical and non-canonical pathways. Frontiers in cell and developmental biology 8:106
16. Deng H, Chen F, Wang Y, Jiang H, Dong Z, Yuan B, Zhao X (2020) The role of activated NLRP3 inflammatory body in acute kidney injury in rats caused by sepsis and NLRP3-TXNIP signaling pathway. Saudi J Biol Sci 27(5):1251–1259
17. Honkala AT, Tailor D, Malhotra SV (2020) Guanylate-binding protein 1: an emerging target in inflammation and cancer. Front Immunol 10:3139
18. Naschberger E, Bauer M, Stürzl M (2006) Human guanylate binding protein-1 (hGBP-1) characterizes and establishes a non-angiogenic. Adv Enzyme Regul 45:215
19. Chen YG, Satpathy AT, Chang HY (2017) Gene regulation in the immune system by long noncoding RNAs. Nat Immunol 18(9):962–972
20. Liu J, Lao L, Chen J, Li J, Zeng W, Zhu X, Li J, Chen X, Yang L, Xing Y, Chen F (2021) The IRENA lncRNA converts chemotherapy-polarized tumor-suppressing macrophages to tumor-promoting phenotypes in breast cancer. Nat Cancer 2(4):457–473
21. Zhou Q, Chen W, Yu QX (2020) Long non-coding RNAs as novel diagnostic and therapeutic targets in kidney disease. Chronic Dis Transl Med 5(4):252–257. https://doi.org/10.1016/j.cdtm.2019.12.004
22. Grote P, Herrmann BG (2013) The long non-coding RNA Fendrr link epigenetic control mechanisms to gene regulatory networks in mammalian embryogenesis. RNA Biol 10(10):1579–1585
23. Çekin N, Özcan A, Göksel S, Arslan S, Pınarbaşı E, Berkano O (2018) Decreased FENDRR and LincRNA-p21 expression in atherosclerotic plaque. Anatol J Cardiol 19(2):131
24. Munteanu MC, Huang C, Liang Y, Sathiaseelan R, Zeng X, Liu L (2020) Long non-coding RNA FENDRR regulates IFNγ-induced M1 phenotype in macrophages. Sci Rep 10(1):1–12
25. Gao F, Feng J, Yao H, Li Y, Xi J, Yang J (2019) LncRNA SBF2-AS1 promotes the progression of cervical cancer by regulating the miR-361-5p/FOXM1 axis. Artif Cells Nanomed Biotechnol 47(1):776–782
26. Zhao QS, Li L, Zhang L, Meng WX, Li LL, Ge XF, Li ZP (2016) Over-expression of lncRNA SBF2-AS1 is associated with advanced tumor progression and poor prognosis in patients with non-small cell lung cancer. Eur Rev Med Pharmacol Sci 20(14):3031–3034
27. Lv J, Qiu M, Xia W, Liu C, Xu Y, Wang J, Leng X, Huang S, Zhu R, Zhao M, Ji F (2016) High expression of long non-coding RNA SBF2-AS1 promotes proliferation in non-small cell lung cancer. J Exp Clin Cancer Res 35(1):1–13
28. Shi ZG, Sun Y, Wang KS, Jia JD, Yang J, Li YN (2019) Effects of miR-26a/miR-146a/miR-31 on airway inflammation of asthma mice and asthma children. Eur Rev Med Pharmacol Sci 23(12):5432–5440
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. | 2025-03-06T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2747,
1
],
[
2747,
8284,
2
],
[
8284,
13630,
3
],
[
13630,
15132,
4
],
[
15132,
20217,
5
],
[
20217,
25880,
6
],
[
25880,
30001,
7
]
]
} |
6ce3b2a70200168d00aa946d27ea55ef539ffca0 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 7,
"total-fallback-pages": 0,
"total-input-tokens": 20692,
"total-output-tokens": 5462,
"fieldofstudy": [
"Linguistics"
]
} | ABSTRACT
A research program is described in which a particular representational format for meaning is tested as broadly as possible. In this format, developed by the LNR research group at The University of California at San Diego, verbs are represented as interconnected sets of subpredicates. These subpredicates may be thought of as the almost inevitable inferences that a listener makes when a verb is used in a sentence. They confer a meaning structure on the sentence in which the verb is used. To be psychologically valid, these representations should capture (at least):
1. Similarity of meaning
The more similar two verbs seem in meaning to people, the more their representations should overlap.
2. Confusability
The more confusable two verb meanings are, the more their representations should overlap.
3. Memory for sentences containing the verb
The sentence structures set up by the verb's meaning should in part determine the way in which sentences are remembered.
4. Semantic integration
The representations should allow for the integration of information from different sentences into discourse structure
5. Acquisition patterns
The structural partitions in the representations should correspond to the structures children acquire when they are learning the meanings of the verbs.
6. Patterns of extension
The representations should be extendible so as to reflect the ways in which people interpret verb meanings when the verbs are used outside their normal context.
7. Reaction times
The time taken to comprehend a sentence using a given verb should reflect the structural complexity of the verb meaning.
Experiments concerned with predictions 1-5 are described here. The results are promising for a general approach of representation of meaning in terms of interrelated subpredicates, but do not clearly distinguish between several similar representations. For example, to test prediction (2), I read people sentences containing verbs with similar meanings, and asked them to recall the sentences. The degree of overlap in the semantic structures was a good predictor of the number of confusions between sentences. In another sentence-memory experiment (prediction (3)), semantically complex verbs that provided more underlying interconnections between the nouns in a sentence led to better memory for the nouns in the sentence than simple general verbs, or than other complex verbs that did not provide such extra interconnections. To test prediction (5), I tested children's comprehension of a set of possession verbs. Both the order of acquisition among the verbs and the kinds of errors fitted well with an account of the acquisition of verb meaning in terms of interconnected subpredicates.
This research illustrates a breadth-first approach to testing a representation. In the breadth-first approach, many different psychological predictions are made. Each different area of prediction requires a set of process assumptions, and in each case the process assumptions used are those that seem most plausible given previous research in the field. If one representational format can make correct predictions about a number of different kinds of psychological phenomena, then that representation stands a greater chance of being generally useful than one which was tested in only one depth-first way.
This paper describes a program of research that tests a representational format for verb meaning. This research grew out of the LNR (Footnote I) attempt to represent the meanings of words in a psychologically satisfying way. Verb meaning seemed a natural place to start for two reasons: (1) verbs are important: it is arguable that they provide the central organizing semantic structures in sentence meanings and lead verbs used to be at, to remain at, etc. and emotion (to hate, to love, etc.).
In addition to simple stative relationships, verbs can be used to convey changes of state. Following Chafe (1970) I will refer to a change of state as a process. For example, the sentence
Ida received $10.00.
(1) that Ida now has $10.00
(2) that someone else had the $10.00 before;
(3) that a change has taken place from this previous state of possession to the present state.
More commonly, verbs express not simple changes of state but causal changes of state. We seem to be very interested in processes that are volitionally caused by humans and other sentient beings. Figure 2 shows the representation of the sentence:
Ida gives Sam a rose.
An agent may cause a change of state that relates to another object. Or the same person may act on both agent and experiencer of the change of state. The locational verb move can be used in either way, as in the following examples:
a. Ida moved the car.
b. Ida moved to the front seat.
In both these cases the action taken by Ida is unspecified. We often don't care exactly what someone did to cause some process to occur. However, there are also verbs in which the causal action is partially or wholly specified: e.g., walk, snow, meander, stride, run, sprint, race, trot, jog. (See Miller (1972) and Miller & Johnson-Laird (1976) for a more extensive discussion of the verbs of location.)
Thus, this system allows for the representation of verbs as states, changes of state, causal changes of state, simple actions, and complex cases in which specific actions cause changes of state. Further discussion of the LNR system of verb semantics can be found in the articles by Abrahamson, Gentner, Munro, Rumelhart & Levin, and Rumelhart & Norman in the Norman & Rumelhart (1975) volume.
There are certainly gaps in the system, and aspects of verb meaning that are not expressible in this simple vocabulary. Some unresolved issues are discussed later in the paper. However, the system seems plausible at the first level, and allows a fair range of verb meanings to be captured at least roughly.
At this point in the research it seemed appropriate to begin testing the psychological rightness of the system as so far stated before going on to refine it.
Psychological Tests of the Model
One advantage of psychological experimentation (or of computer implementation) is that it forces one to make explicit the assumptions underlying representation and process. At least some of the choices made can then be tested as hypotheses. Some important assumptions are:
1. A verb's representation captures the set of immediate inferences that people normally make when they hear or read a sentence containing the verb.
2. In general, one verb leads to many inferences.
3. These networks of meaning components are accessed during comprehension, by an immediate and largely automatic process.
4. The set of components associated with a given word is reasonably stable across tasks and contexts.
5. Surface memory for exact words fades quite rapidly, so that after a short time, only the representational network remains.
In testing these representations, I took a very literal interpretation of the notion of representation -- namely that the nodes and arrows in a representation correspond to the concepts and relationships that are stored when a person comprehends a sentence containing a verb. The more ferociously literal the interpretation, the better the chances of discovering counter-evidence.
Semantic overlap. One psychological criterion is that the representations should agree with people's intuitive notions of synonymy and similarity in meaning. One straightforward measure of this overlap is the degree to which people rate verbs as similar in meaning. In a study of about 60 selected verbs, I found that people's average rating of the semantic similarity between two verbs agreed very closely with the degree of semantic overlap between their representations.
A more subtle measure of psychological similarity is the degree to which people unconsciously confuse things in memory. People in a sentence-memory experiment probably try to keep their sentence traces clear. But, suppose that within a short time after hearing a verb in a sentence, a person has only the representational network of concepts and relationships, and not the surface verb. Assume further that some pieces of the memory representation may be lost or unaccessible at any time (the "fallibility of human memory" assumption). Then the more two verb representations overlap, the more likely it is that sentences containing the two verbs will be confused in memory, despite people's attempts to keep them straight. In an experiment in sentence memory, using verbs of varying semantic overlap, I found that subjects did indeed confuse the verbs in exactly the way predicted by the theory (Gentner, 1974). The correlation between the number of confusions subjects made between two verbs and the semantic overlap between the verbs, as predicated from the representations, was quite high. In fact, the correlation between representational overlap and number of confusions was slightly higher (though not significantly so) than the correlation between the number of confusions and the rated similarity between the verbs. (The similarity ratings were taken from the first-mentioned study, with a different set of subjects).
Semantic complexity. Semantic complexity refers to the number of underlying subpredicates and interconnections that make up the basic meaning of a verb. More complex meanings correspond to more specific actions or events. For example,
(a) Ida strode across the field.
(b) Ida went across the field.
Various researchers have looked for evidence that semantic complexity may affect comprehensibility, generally on the assumption that more complex semantic structures are harder to process (Kintsch 1974; Thorndyke, 1977). However, the results have been negative. There is no evidence that more complex words lead either to longer reaction-times or to greater processing loads than do simpler words. I believe that it's incorrect to assume across the board that complexity is psychologically hard. Some research of mine suggests that the effects of semantic complexity in memory are more particular.
Semantic Complexity and Connectivity. Although the view that semantic complexity leads to difficulty has not been supported, there is another side to the complexity issue. The additional semantic components in a complex verb may set up additional connections among the nouns in the sentence. In this case, more complex verbs should lead to a richer and more highly interwoven sentence representation, and thus to better memory for the nouns in the sentence.
Notice that this prediction derives from a fanatically literal interpretation of the verb representations: more paths in the representation means more conceptual paths in memory. This prediction is quite specific. It is not simply a question of certain complex versus simple verbs having some overall effect, but rather of complex verbs providing extra connections between the particular nouns in question. This is clearly true for Ida and her tenants in the case of
sell versus give, as can be seen in Fig 3a and 3b.
I tested for this kind of improvement in connectivity in a series of experiments in sentence memory (Gentner, 1977). I read people sentences that differed in the semantic...
connectivity of their verbs, such as the following pair of sentences:
Ida gave her tenants a clock. (simple)
Ida sold her tenants a clock. (complex connective)
Then I gave the people the names of the objects and asked them to recall the sentences. As predicted, they were better able to recall the noun tenants when the complex connective verb sold was used than when the simple verb gave was used. More semantic connections between the two nouns led to stronger memory connections.
To see the specificity of the prediction, consider a complex verb that merely amplifies the simple verb and does not add connections between the key nouns. For example, the verb mail (Fig 3c) adds the information that the object was mailed leads to stronger memory connections. For example, the verb give (Fig 3d) adds the information that the object was given to the agent, Ida, and the recipient, tenants. Therefore, the prediction was that use of such non-connecting verbs would lead to no improvement over use of general verbs in memory between the nouns.
The results were exactly as predicted: The object nouns of complex connective verbs were recalled better than those of general verbs and non-connecting complex verbs. These differences were not traceable to differences in imagery or word-frequency. Thus connectivity is beneficial to sentence memory in a very specific way.
Acquisition. There may be a more direct relationship between complexity and difficulty in children than in adults. Young children often fail to comprehend the full meanings of semantically complex terms (e.g., Bowerman, 1975; Clark, 1973; Gentner, 1975, in press). Working with the verbs of possession, I have observed that children act out the simple verbs give and take correctly before they act out the more complex verbs buy and trade. Still later they learn the yet more complex verbs buy, sell, and spend. The order in which the verbs are learned is exactly the order of increasing semantic complexity. This complexity ordering can be made quite precise, since the verbs are closely related in meaning. The representation of a verb at the nth level of simplicity is properly nested within the representation of a verb at the (n+1)th level. Further, when children around 4-6 years are asked to act out sell (as in "Make Ernie sell Bert a boat") they act out give instead (A boat is transferred from Ernie to Bert). Similarly, buy is acted out as take. They systematically act out complex verbs like simple verbs; and more surprisingly, they choose the appropriate simple verb. My interpretation, consistent with Clark's (1973) semantic features analysis, is that children learn these complex verb meanings gradually, by adding components to their partially correct representations. At any given time, the child comprehends language in terms of the components that he has so far acquired.
Semantic Integration. Another important psychological requirement is combinatoriality. The basic notions of state, change of state, cause, and so on must be combinable into networks larger than the individual sentence. When two verbs share parts of their underlying structure, this redundancy should be utilized to combine the two representations into one discourse structure. How can we test whether this happens? One way is to arrange things so that collapsing the redundancies between two verbs should create the representation of a third verb. Then the prediction is that people should use this third verb in recall.
In a study of semantic integration, I read people short passages and tested their memory by having them fill in blanks (Gentner, 1978). Every passage contained a general verb, such as give. Half the passages also contained additional semantic information, such as the fact that the giver actually owed the money he was giving. According to the representational model, the integration of the representation of give with that of owing should have created the structure of pay. If what people have in their minds after hearing the verbs is the network representations, and if these representations are integrated during discourse comprehension, then people who heard give and owe should end up with the representation of pay. As predicted, subjects hearing the extra material falsely recalled the verb which best fit the composite structure (e.g. pay) rather than the verb actually presented.
Further Issues
I have made the assumption that a verb carries with it a set of inferences that are normally made during comprehension, as well as several supporting assumptions. This view has been fairly well supported by the research presented here, but nevertheless it seems to me an oversimplification. There remain a great many questions, some large and some small.
(1) Where should the line be drawn around a word's meaning? As Clark and Clark (1977) have put it, is word meaning more like a dictionary or an encyclopedia? The extreme of the dictionary approach would be to take a minimal contrast approach, storing with a word only enough to distinguish it from all other words. The extreme of the encyclopedia approach would be to access the entire long-term memory whenever any word is used. The question is, how to define a reasonable middle ground.
(2) What is the process of expansion into a semantic representation during comprehension?
a) Are there invariable inferences? When an incoming word is processed, is there a set of inferences (such as the set I have "called the "almost-inevitable inferences" that is always made during comprehension, or is there variation in which inferences get made?
b) If there is variation, is it quantitative or qualitative? Do context and the person's interests and attention determine which inferences get made, so that there are qualitative differences in what inferences get made? Or is the difference merely quantitative, with the radius of expansion varying with the amount of attention (or energy, or interest) that the person brings to bear?
The notion of at least quantitative variation a seems hard to avoid. It is a fairly strong intuition that we process word meanings with varying degrees of energy. Further, the phenomenon of instantiation (Anderson, R.C., Stevens, K.C., Shifrin, Z., & Osborn, J.; 1977) makes it clear that a model of sentence comprehension must allow for qualitative differences in the final set of inferences stored. For example, compare the sentences
Rover ate his dinner.
Mr. Pritchard ate his dinner.
The verb eat conveys vastly different action sequences when used with different agents, though its causal change-of-state structure remains more-or-less constant. It is possible that this qualitative variation can be accounted for by simple underlying quantitative processes spreading activation. We may have to settle for a more complex model, in which some parts of a verb's meaning are always accessed while other inferences develop out of the interaction of the verb with its context, including its pragmatic context. In Hewitt's (1976) terms, there may be both if-added inferences and if-needed inferences. Where in this model (and whether) we want to draw a line between meaning and knowledge-of-the-world is not at all clear to me. (3) Carrying the notion of variable verb meaning still further, how does metaphorical extension work? Most common verbs can be used in several related ways. For example, consider the range of meanings that give can convey depending on the nouns it is used with:
Ida gave Sam
a rose.
a job.
an heir.
an excuse.
a talking to.
all his best ideas.
the time of his life.
Clearly the subpredicate structure varies across these sentences, so much so that some might want to describe this as a collection of entirely different senses of the same word. This misses the structural similarities. Some kind of metaphorical extension of meaning seems a necessary part of a theory of verb meaning, since it is generally the verb that does most of the adjusting. A series of studies by Albert Stevens and me suggests that people faced with an odd sentence assume that some of the subpredicates normally conveyed by the verb are not meant to apply in the sentence at hand. A current project is to model the rules for which subpredicates apply in different contexts.
(4) I have so far treated nouns as nodes in the semantic representation. Clearly in order to analyze sentence interactions it is necessary to have a representation of noun meaning. Some progress been made with abstract nouns, such as kinship terms. But the truly nounlike nouns ---basic-level nouns--- resist analysis. I believe that these differences in amendability to analysis reflect differences in the kind of meaning that verbs and nouns have, and that a useful representation of concrete noun meaning may be quite different from that used for verbs, prepositions and even abstract nouns.
(5) There are several aspects of the representational scheme that need further thought. To single out one issue, consider the notion of change of state. The LNR representation represents a verb like get as conveying a change from an initial state of possession to a final state of possession. Schank's Conceptual Dependency theory would represent the entire sequence as a primitive act. Many generative semanticists have represented only the inchoative part of the chain (the change to the final state) as belonging to the assertion of the verb, considering the initial state to be more in the nature of a presupposition (e.g. Fillmore, 1966). All these positions seem to me to have merit. The LNR use of change from initial to final state allows a change-of-state verb to hook automatically with relevant state information. The use of acts as primitives captures the psychological wholeness of change. The use of the inchoative captures the intuition that people seem more interested in the results of an event --i.e. in the final state-- than in the setting state. The explicit change-of-state formats (LNR format and inchoative format) have a natural way of capturing some kinds of metaphorical extension: by substituting a different stative while preserving the rest of the verb's structure.
Summary
This work is just beginning. Neither the representations nor the processes that are assumed to operate on them come very close to capturing the subtlety of human language use. Still, the results of the experimental investigation are promising some kind of decompositional model along these lines.
Figure 1. Ida owned a Cadillac from 1970-1977.
Figure 2. Ida gives Sam a rose.
Figure 3a. Ida sold her tenants a clock.
Figure 3b. Ida mailed her tenants a clock.
Footnote
1. The representational format shown here was developed by a group of researchers at the University of California at San Diego: Adele A. Abrahamson, Dedre Gentner, James A. Levin, Stephen E. Palmer, and David E. Rumelhart. The system is explained in detail in Norman & Rumelhart, 1975.
References
Abrahamson, A.A. Experimental analysis of the semantics of movement. In D.A. Norman, & D.E. Rumelhart, (Eds.), Explorations in Cognition. San Francisco: W.H. Freeman & Co., 1975.
Anderson, R.C., Stevens, K.C., Shifrin, Z., & Osborn, J. Instantiation of Word Meanings in Children, May 1977.
Bendix, E.H. Componential analysis of general vocabulary: The semantic structure of a set of verbs in English, Hindi, and Japanese. The Hague; Mouton, 1966.
Bowerman, M. The acquisition of word meaning: An investigation of some current conflicts. Paper presented at the Third International Child Language Symposium, London, September 1975.
Chafe, W.L. Meaning and the structure of language. Chicago: University of Chicago Press, 1970.
Clark, E.V. What's in a word: On the child's acquisition of semantics in his first language. In T.E. Moore (Ed.), Cognitive development and the acquisition of language, New York: Academic Press, 1973.
Clark, H.H. & Clark, E.V. Psychology and Language. New York: Harcourt Brace Jovanovich, Inc., 1977.
Fillmore, C.J. Review of Bendix's Componential analysis of general vocabulary: The semantic structure of a set of verbs in English, Hindi, and Japanese. International Journal of American Linguistics, 1968, 32, Part II, No. 2. Publication 41.
Gentner, D. Towards a psychological theory of the meaning of the possession verbs. Unpublished doctoral dissertation, University of California, San Diego, 1974.
Gentner, D. Evidence for the psychological reality of semantic components: The verbs of possession. In D.A. Norman and D.E. Rumelhart, Explorations in Cognition, San Francisco: W.H. Freeman & Co., 1975.
Gentner, D. On relational meaning: The acquisition of verb meaning. Child Development, in press.
Gentner, D. Semantic integration of word meanings. Bolt Beranek and Newman Inc. Report No. 3826, May 1978. Also to appear as a Center for the Study of Reading Technical Report.
Hewitt, C. Viewing control structures as patterns of passing messages. M.I.T. AI Working Paper 92, 1976.
Lakoff, G. Adverbs and modal operators. Indiana University Linguistics Club Reprint. Bloomington: Indiana University Linguistics Club, 1970(a).
Lakoff, G. Irregularity and syntax. New York: Holt, Rinehart and Winston, 1970(c).
McCawley, J.D. The role of semantics in a grammar. In E. Bach and R.T. Harms (Eds.), Universals in linguistic theory. New York: Holt, Rinehart and Winston, 1968(b). | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3348,
1
],
[
3348,
5890,
2
],
[
5890,
11247,
3
],
[
11247,
16843,
4
],
[
16843,
21719,
5
],
[
21719,
22181,
6
],
[
22181,
24610,
7
]
]
} |
ab4ccba8b1bdff5a09e2a90566362301010b517e | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 8,
"total-fallback-pages": 0,
"total-input-tokens": 28774,
"total-output-tokens": 11582,
"fieldofstudy": [
"Medicine",
"Biology"
]
} | Association of IL-10 and TNF-α polymorphisms with risk and aggressiveness of hepatocellular carcinoma in patients with HCV-related cirrhosis
Ahmed Saleh 1,2*, Ahmed M. Saed 1 and Mostafa Mansour 3
Abstract
Background: Hepatitis C virus (HCV) infection is a significant risk factor for cirrhosis and hepatocellular carcinoma (HCC) that carry a high mortality. The study aims to investigate the effect of tumour necrosis factor (TNF)-α and interleukin (IL)-10 polymorphisms on risk and pattern of HCC in patients with HCV-related cirrhosis.
Results: The mean age of the HCC group was 56.21 ± 4.62 years and 54.27 ± 7.63 years for the cirrhotic group. The GG genotype of TNF-α and TT genotype of IL-10 showed a higher incidence of HCC in comparison to the cirrhotic group with \( P = 0.01 \) and \( P = 0.004 \). On the calculation of the aggressiveness index (AgI), the TT haplotype was significantly associated with more aggressive tumours in contrast to the other haplotypes with \( P < 0.001 \). There is a significant association of portal vein thrombosis, ascites and high AgI with the GG haplotype in contrast to the other haplotypes with \( P = 0.002 \), \( P = 0.029 \) and \( P < 0.001 \), respectively, as regards TNF-α. High AgI (C) was associated with the TT haplotype of IL-10 and GG haplotype of TNF-α.
Conclusion: Our data bring an essential association of IL-10 and TNF polymorphism with the occurrence of HCC in patients with HCV-related liver cirrhosis. The GG haplotype of TNF-α and TT/AT haplotype of IL-10 are associated with the more aggressive pattern of HCC, so those patients must be treated as early as possible.
Keywords: Interleukin, Tumour necrosis factor, Hepatocellular carcinoma
Background
Hepatitis C virus (HCV) and hepatitis B virus (HBV) infections are global health problems as they are associated with high morbidity and mortality. It is estimated that about 200 million people are infected with HCV, among which 170 million are chronically infected [1].
Hepatocellular carcinoma (HCC) is the fifth most common cancer and the third cause of cancer-related deaths worldwide [2]. More than 600,000 people die from HCC each year [3]. The aetiology is still unclear; however, several major risk factors of HCC have been shown to contribute to hepatocarcinogenesis. Chronic inflammation induced by the action of various inflammatory mediators has been recently identified as a cofactor in carcinogenesis [3]. Among these inflammatory mediators, tumour necrosis factor-α (TNF-α) plays an essential role and has been implicated in inflammation-associated tumours [4].
An “HCC Aggressiveness” scoring system was recently described, which incorporates four tumour-related parameters such as maximum tumour diameter (MTD), presence of portal vein thrombosis (PVT), number of lesions and serum alpha-fetoprotein (AFP) levels. The
Aroucha et al. concluded that polymorphisms in TNF-α and interleukin (IL)-10 were associated with increased risk of HCC development in HCV chronically infected patients. The GG genotype of TNF-α and genotypes associated with low/intermediate levels of IL-10 were shown to be associated with increased risk of development of HCC. Moreover, the TT genotype of the IL-10 -819 was significantly correlated with advanced stages of HCC as well as with multiplicity of lesions. These variants were shown to be associated with more inflammation in the liver, mediated by Th1 cytokines and may increase the risk to have HCC and bring an adverse prognosis in these patients [6].
**Aim of the study**
In this study, we tried to assess the association of TNF-α and IL-10 polymorphisms with the risk and aggressiveness of HCC in patients with HCV-related cirrhosis.
**Methods**
**Patients**
This is a case-control study that was conducted on 73 patients attending to early detection at HCC Clinic, Specialized Medical Hospital, Mansoura University, from June 2019 to February 2020 for follow-up of HCC and 85 patients with HCV-related cirrhosis without evidence of hepatic focal lesions proved by post-contrast triphasic computed tomography (CT) scan. Patients with cirrhosis related to other causes than HCV, patients with any tumour outside the liver and patients with ongoing organ failure were excluded from the study. All patients were proved to be HCV infected through the history of previous antiviral treatment for HCV or through testing for HCV Ab or HCV RNA. All patients were exposed to complete history taking with stress on history of smoking or ex-smoker, diabetes, hypertension, history of previous treatment for HCV or bilharziasis and family history of cancer. Complete clinical examination was done for all patients with stress on manifestations of hepatic cellular failure. Baseline clinical characters, including MTD, the number of focal lesions and the presence of portal vein thrombosis, were collected from imaging reports done at Specialized Medical Hospital for the HCC group.
**Laboratory assessment**
Complete blood count, liver function tests with Child-Pugh classification, HBsAg, HCV Ab, serum creatinine and AFP were done for all patients.
**IL-10 and TNF-α polymorphism determination**
Peripheral blood was used to extract genomic DNA using the Wizard Genomic Blood DNA Isolation Kit (Promega, Madison, WI). We stored samples at −80 °C until single nucleotide polymorphism (SNP) genotyping by real-time polymerase chain reaction (PCR) was done. In IL-10 gene, we tested one substitution at position -819 C>T (rs1800871). In TNF-α, the substitution at position -308 G>A (rs1800629) was tested. We used TaqMan SNP Genotyping Assays (Applied Biosystems, Foster City, CA), according to the instructions of the manufacturer.
According to Ventura et al., the aggressiveness index (AgI) score was divided into three categories: a, score − < 4; b, 4 < score ≤ 7; and c, score ≥ 8 (Table 1) [7].
Aggressiveness index was calculated for all patients with HCC. According to the result of genotype frequency distribution of IL-10 -819 (rs1800871) and TNF-α -308 (rs1800629), patients will be divided into groups and compared.
**Statistical analysis**
Data were fed to the computer and analysed using IBM SPSS Corp. (released in 2013, IBM SPSS Statistics for Windows, version 22.0, Armonk, NY: IBM Corp). Qualitative data were described using the number and per cent. Quantitative data were described using median and interquartile range for non-parametric data and mean, the standard deviation for parametric data after testing normality using the Kolmogorov-Smirnov test. The significance of the obtained results was judged at the 0.05 level.
Data analysis was done using the chi-square test and Monte Carlo test for comparison of 2 or more groups of categorical variables as appropriate. Student t test and one-way ANOVA test were used to compare parametric variables. Mann-Whitney U test and Kruskal-Wallis test were used to compare independent groups for non-parametric variables. Spearman’s rank-order correlation is used to determine the strength and direction of a linear relationship between two non-normally distributed continuous variables and/or ordinal variables.
**Results**
**Patient sociodemographic characteristics**
The mean age of the HCC group was 56.21 ± 4.62 years and 54.27 ± 7.63 years for the cirrhotic group. There
| Table 1 Aggressiveness index parameters |
|----------------------------------------|
| **Maximum tumour dimension(CM)** |
| < 4.5 |
| 4.5–9.6 |
| > 9.6 |
| **AFP (ng/ml)** |
| < 100 |
| 100–1000 |
| > 1000 |
| **PVT** |
| No |
| Yes |
| **Tumour nodule (number)** |
| ≤ 3 |
| > 3 |
was no difference in sex distribution between the two groups. In the HCC group, 31 (42.5%) patients had diabetes and 19 (26.0%) hypertensive in comparison to 27 (31.8%) and 23 (27.1%) in the cirrhotic group, respectively. A statistically significant difference between the two groups was found as regards serum albumin, platelets, aspartate aminotransferase (AST) and AFP with $P = 0.003$, $0.001$, $0.001$ and $< 0.001$, respectively (Table 2).
### Genotype testing and distribution
On genotype testing, as regards TNF-α, 35 (47.9%) of patients were GG, 22 (30.1%) were GA and 16 (21.9%) were AA in the HCC group. In the cirrhotic group, 27 (31.8%) of patients were GG, 25 (29.4%) were GA and 33 (38.8%) were AA. As regards IL-10, 29 (39.7%) of patients were TT, 21 (28.8%) were CT and 23 (31.5%) were CC in the HCC group. In the cirrhotic group, 16 (18.8%) of patients were TT, 29 (34.1%) were CT and 40 (47.1%) were CC. From these data, it appears that the GG genotype of TNF-α and TT genotype of IL-10 showed a higher incidence of HCC in comparison to the cirrhotic group with $P = 0.01$ and $0.004$ (Table 3 and Fig. 1).
### Tumour characteristics and aggressiveness concerning genotype polymorphism
As regards IL-10, the portal was found to be thrombosed significantly in the TT haplotype in contrast to the other haplotypes with $P < 0.001$. Similarly, ascites was significantly found in the TT haplotype in contrast to the other haplotypes with $P = 0.006$. On the other hand, there was no significant difference in the size of the spleen, lymph node metastases or site of the lesions between the haplotypes. On the calculation of the AgI, the TT haplotype was significantly associated with more aggressive tumours in contrast to the other haplotypes with $P < 0.001$ (Table 4). Table 5 shows the tumour characteristics and aggressiveness of TNF-α genotypes where there was a significant association of portal vein thrombosis, ascites and high AgI with the GG haplotype in contrast to the other haplotypes with $P = 0.002$, $0.029$
### Table 2 Sociodemographic characteristics of the studied groups
| Parameter | HCC, $n = 73$ | Cirrhotic, $n = 85$ | Test of significance |
|----------------------------|---------------|---------------------|----------------------|
| Age/years, mean ± SD | 56.21 ± 4.62 | 54.27 ± 7.63 | $t = 1.89$ |
| | | | $P = 0.06$ |
| Sex, $N$ (%) | | | |
| Male | 43 (58.9%) | 39 (45.9%) | $\chi^2 = 2.67$ |
| Female | 30 (41.1%) | 46 (54.1%) | $P = 0.102$ |
| DM, $N$ (%) | 31 (42.5%) | 27 (31.8%) | $\chi^2 = 1.94$ |
| | | | $P = 0.164$ |
| Hypertension, $N$ (%) | 19 (26.0%) | 23 (27.1%) | $\chi^2 = 0.021$ |
| | | | $P = 0.884$ |
| Albumin (gm/dl), mean ± SD| 3.29 ± 0.59 | 3.56 ± 0.52 | $t = 2.98$ |
| | | | $P = 0.003^*$ |
| Bilirubin (mg/dl), median (IQR) | 1.2 (0.85–1.7) | 1.2 (0.8–1.64) | $z = 0.821$ |
| | | | $P = 0.412$ |
| WBCS, mean ± SD | 4.78 ± 1.88 | 4.69 ± 1.85 | $t = 0.33$ |
| | | | $P = 0.739$ |
| HB (gm/dl), mean ± SD | 11.33 ± 1.66 | 11.42 ± 1.53 | $t = 0.374$ |
| | | | $P = 0.709$ |
| Platelets * $10^3$, mean ± SD | 101.05 ± 44.17 | 123.92 ± 36.76 | $t = 3.55$ |
| | | | $P = 0.001^*$ |
| ALT (U/l), median (IQR) | 35.0 (28.5–47.0) | 35.0 (29.0–45.0) | $z = 0.452$ |
| | | | $P = 0.65$ |
| AFP (ng/dl), median (IQR) | 89.0 (33.0–230.5) | 5.2 (28–97) | $z = 9.98$ |
| | | | $P < 0.001^*$ |
| AST(U/l), median (IQR) | 68.0 (54.5–89.0) | 40.0(36.0–47.0) | $z = 7.67$ |
| | | | $P < 0.001^*$ |
| INR, mean ± SD | 1.36 ± 0.20 | 1.32 ± 0.19 | $t = 1.09$ |
| | | | $P = 0.275$ |
Parameters described as mean ± SD, median (interquartile range) or number (percentage)
$t$ Student $t$ test, $Z$ Mann-Whitney $U$ test, $\chi^2$ chi-square test
*Statistically significant
Table 3 Genotype distribution among the studied groups
| | HCC, n = 73 (%) | Cirrhotic, n = 85 (%) | Test of significance | Odds ratio (95% CI) |
|----------------|-----------------|-----------------------|----------------------|--------------------|
| IL-10 | | | | |
| TT/AA | 29 (39.7%) | 16 (18.8%) | \(\chi^2 = 8.2, P = 0.004^*\) | 3.15 (1.42–6.99) |
| CT/CA | 21 (28.8%) | 29 (34.1%) | \(\chi^2 = 0.35, P = 0.55\) | 1.26 (0.59–2.69) |
| CC/CC | 23 (31.5%) | 40 (47.1%) | | Reference group |
| Hardy-Weinberg equation | \(P = 0.002^*\) | \(P = 0.02^*\) | | |
| TNF-\(\alpha\) | | | | |
| GG | 35 (47.9%) | 27 (31.8%) | \(\chi^2 = 6.24, P = 0.01^*\) | 2.67 (1.22–5.83) |
| GA | 22 (30.1%) | 25 (29.4%) | | 1.82 (0.79–4.15) |
| AA | 16 (21.9%) | 33 (38.8%) | \(\chi^2 = 2.01, P = 0.16\) | Reference group |
| Hardy-Weinberg equation | \(P = 0.002^*\) | \(P = 0.001^*\) | | |
\(\chi^2\) chi-square test
*Statistically significant
Discussion
HCC represents about 90% or more of primary liver tumours that usually develops in a background of advanced liver disease [8]. Approximately, 10–20% of chronically infected patients with HCV will develop liver cirrhosis, and 1–5% of those patients will develop HCC [9]. The continued cytokines induced hepatocyte damage, and hepatocyte regeneration leads to HCC development. The role of cytokines as IL-1, IL-2, IL-6, IL-10, IL-12 and TNF-\(\alpha\) in hepatocarcinogenesis has been reported. The management of patients with HCC represents a challenge as it is often complicated by the heterogenic pattern of the disease, the state of underlying liver disorders and the need to coordinate a multidisciplinary healthcare team [10].
In our study, the mean age of the HCC group was 56.21 ± 4.62 years and 54.27 ± 7.63 years for the cirrhotic group. There was no difference in sex distribution between the two groups (Table 2). In the HCC group, 31 (42.5%) patients had diabetes and 19 (26.0%) hypertensive in comparison to 27 (31.8%) and 23 (27.1%) in the cirrhotic group, respectively. This runs parallel to many other studies that concluded that advanced age, male gender and DM are well-known risk factors for HCC [11]. Cirrhosis, due to HCV, was found in almost all our patients. This seems logical and in agreement with Seyda et al. who clarified that most of HCC cases develop in a background of cirrhosis [12]. There was no difference in the sociodemographic and clinicolaboratory parameters between the two groups except for serum albumin, platelets, AST and AFP with \(P = 0.003, 0.001, 0.001\) and < 0.001, respectively. This seems logical as platelet count, AST and serum albumin reflect the severity of the liver disease. It also suggests that the two groups are...
matched which is extremely important in the inclusion of our patients to avoid the effect of any factor such as age, sex or comorbidities on risk of HCC development, so the increased incidence of HCC is directly related to the effect of gene polymorphism.
In the present study, on genotype testing, as regards TNF-α, 35 (47.9%) of patients were GG, 22 (30.1%) were GA and 16 (21.9%) were AA. In the cirrhotic group, 27 (31.8%) of patients were GG, 25 (29.4%) were GA and 33 (38.8%) were AA. As regards IL-10, 35 (47.9%) of patients were TT, 21 (28.8%) were CT and 23 (31.5%) were CC. In the HCC group, 16 (18.8%) of patients were TT, 29 (34.1%) were CT and 40 (47.1%) were CC. From these data, it appears that the GG genotype of TNF-α and TT genotype of IL-10 showed a higher incidence of HCC in comparison to the cirrhotic group with \( P = 0.01 \) and \( P = 0.044 \) (Table 3 and Fig. 1). This goes hand in hand with Aroucha et al. who found that the TT/AT haplotypes of IL-10 and GG haplotype of TNF-α were significantly expressed in patients with HCC [6]. Wei et al. and Cheng et al. agreed [13, 14] also with us but not with Zhou et al. who showed that no association was found between HCC and the TNF-α-238 G/A polymorphism [15]. This contrast may be related to the aetiology of the underlying liver disease, ethnicity or number of included patients.
In 2012, Swiatek found that gene polymorphisms may affect the IL-10 level; he showed that IL-10 -819T was associated with significant low IL-10 expression since it is located in transcript factor binding regions [16]. Aroucha et al. observed an increased frequency of IL-10 -819T genotype in patients with HCC [6]. Moreover, they found a significant association between the TT genotype of IL-10 -819 and multiplicity of lesions and terminal stages of HCC.
As regards TNF-α, the results are conflicting. Talaat et al. and Radwan et al. did not find any significant association between HCC and TNF-α -308 polymorphism in HCV-infected Egyptian patients [17, 18]. On the other hand, Baghel et al. and Karimi et al. demonstrated that patients with TNF-α G allele usually show low TNF-α production in vivo and in vitro [19, 20] despite, Vikram et al. failed to confirm this association [21]. It seems that the balance between IL-10 and TNF-α is crucial for the prevention of development of HCC and that low levels lead to progressive damage to the liver tissue and prevention of wound healing.
Similarly, previous studies found elevated levels of circulating TNF-α in patients with HCC. It is reasonable to speculate that in patients with HCC, the high circulating TNF-α levels found may be attributed to its SNPs. Also, TNF-α may stimulate the release of other inflammatory cytokines and induce the release of other fibrogenic factors, such as interleukin-1, interleukin-6 and tumour growth factor-β which can cause or aggravate liver damage [22, 23].
Previously, immunomodulatory cytokines have been described as pre-malignant mediators in different tumour entities in different studies [24]. In HCC, IL-6 promotes multiple stages of tumour development, including initial hepatocyte proliferation, the transformation of hepatocytes into HCC progenitor cells, and progression to HCC nodules and metastases [24]. It seems that the balance between TNF-α and IL-10 is mandatory to the development of HCC, since the shift to Th1 pattern-like cytokines in the liver may lead to more inflammation, necrosis of hepatocytes and subsequent regeneration that leads to mutagenesis and
### Table 4 Tumour characteristics of HCC cases according to IL-10 polymorphism
| Site of lesion | Number of patients | TT/AA | CT/CA | CC/CC | Total, \( n = 73 \) | Genotype Test of significance |
|---------------|--------------------|-------|-------|-------|------------------|-----------------------------|
| Site of lesion | | TT/AA | CT/CA | CC/CC | | |
| Portal vein | Thrombosed | 10 | 10 | 0 | 20 (65.5) | 21 (100.0) | 23 (100.0) | MC \( P < 0.001^* \) |
| | Patent | 63 | 19 | 0 | 39 (61.3) | 8 (12.8) | 14 (21.9) | \( P = 0.154 \) |
| Splenectomy | | 2 | 0 | 2 | 3 (100.0) | 10 (31.2) | 2 (6.2) | MC \( P < 0.001^* \) |
| Splenectomy | | 28 | 19 | 9 | 56 (62.2) | 8 (14.3) | 12 (24.5) | \( P < 0.001^* \) |
| Splenectomy | | 5 | 2 | 3 | 9 (45.0) | 10 (50.0) | 1 (5.0) | \( P = 0.006^* \) |
| Splenectomy | | 12 | 5 | 3 | 20 (66.7) | 8 (26.7) | 6 (20.0) | MC \( P = 0.089 \) |
| Site of lesion | Right lobe | 25 | 10 | 6 | 41 (82.0) | 9 (18.0) | 12 (24.0) | \( P = 0.929 \) |
| Site of lesion | Left lobe | 12 | 5 | 3 | 20 (66.7) | 8 (26.7) | 4 (13.0) | \( P = 0.001^* \) |
| Site of lesion | Multifocal | 36 | 14 | 12 | 62 (55.6) | 21 (18.8) | 23 (20.6) | \( P = 0.001^* \) |
| Site of lesion | AGI | 22 | 6 | 9 | 37 (81.8) | 9 (20.0) | 6 (13.6) | \( P = 0.029 \) |
| Site of lesion | | 23 | 6 | 9 | 48 (66.7) | 12 (16.7) | 12 (16.7) | \( P = 0.006^* \) |
* means statistically significant
activation of protooncogene in the host cells, leading to the development of HCC [25].
A fine tune of the IL-10 and TNF-α balance may exist, and it looks that this balance is controlled by the level of IL-10, where low levels lead to progressive damage to liver tissue and prevention of wound healing. Also, IL-10 can diminish the response to antiviral treatment [26].
Our data showed that the TT haplotype of IL-10 was significantly associated with more aggressive tumours in contrast to the other haplotypes with \( P < 0.001 \). The portal was found to be thrombosed significantly in the TT haplotype in contrast to the other haplotypes with \( P < 0.001 \); they were also associated with the multiplicity of lesions. Similarly, a significant association of portal vein thrombosis, ascites and high AGI with the GG haplotype in contrast to the other haplotypes with \( P = 0.002, 0.029 \) and \( < 0.001 \), respectively. These data are available in the absence of a statistically significant difference between AGI and sociodemographic and clinicolaboratory characteristics except for AFP and genotype distribution. This suggests that the aggressive pattern of the tumour noticed in these haplotypes is related to the direct effect of gene polymorphism. Aroucha et al. agree with us as they observed a significant correlation of advanced stages and multiple lesions of HCC with the TT (AA) genotype of IL-10 -819 (-592) [6]. However, studies covering this sector are relatively rare.
Our study may be limited by some factors such as the limited number of cases in the study, the lack of data about overall survival of patients and lastly we included only patients with HCV-related cirrhosis which may affect hepatocarcinogenesis.
### Table 5: Tumour characteristics of HCC cases according to TNF-α polymorphism
| Genotype | Test of significance |
|----------|----------------------|
| Total, \( n = 73 \) (%) | GG | GA | AA |
| Portal vein | | | |
| Thrombosed | 10 (28.6) | 0 (0.0) | 0 (0.0) | MC | \( P = 0.002^* \) |
| Patent | 63 (71.4) | 22 (100.0) | 16 (100.0) | | |
| Spleen | | | |
| Splenectomy | 2 (0.0) | 2 (9.1) | 0 (0.0) | MC | \( P = 0.313 \) |
| Mild | 26 (40.0) | 8 (36.4) | 4 (25.0) | | |
| Moderate | 39 (51.4) | 12 (54.5) | 9 (56.2) | | |
| Marked | 6 (8.2) | 2 (9.1) | 2 (12.5) | | |
| Ascites | | | |
| Negative | 62 (83.8) | 22 (100.0) | 14 (87.5) | MC | \( P = 0.029^* \) |
| Positive | 11 (16.2) | 0 (0.0) | 2 (12.5) | | |
| Lymph node | | | |
| Negative | 64 (86.3) | 22 (100.0) | 14 (87.5) | MC | \( P = 0.08 \) |
| Positive | 9 (13.7) | 0 (0.0) | 2 (12.5) | | |
| BCLC | | | |
| A | | | |
| B | 24 (59.5) | 15 (68.2) | 7 (43.8) | | |
| C | 24 (59.5) | 2 (9.1) | 3 (18.8) | | |
| D | 14 (33.3) | 0 (0.0) | 0 (0.0) | | |
| Site of lesion | | | |
| Right lobe | 25 (68.9) | 9 (24.3) | 4 (25.0) | MC | \( P = 0.74 \) |
| Left lobe | 12 (29.3) | 2 (9.1) | 3 (18.8) | | |
| Multifocal | 36 (66.7) | 11 (50.0) | 9 (56.2) | | |
| AGI | | | |
| A | 22 (66.7) | 12 (38.9) | 10 (62.5) | MC | \( P < 0.001^* \) |
| B | 23 (69.7) | 8 (24.2) | 6 (37.5) | | |
| C | 28 (82.4) | 2 (9.1) | 0 (0.0) | | |
* means statistically significant
To summarize, specific genotypes of TNF-α and IL-10 may affect the progression of hepatocarcinogenesis in patients with HCV-related liver cirrhosis.
**Conclusion**
Our data bring an essential association of IL-10 and TNF polymorphism with the occurrence of HCC in HCV-related liver cirrhosis. The GG haplotype of TNF-α and TT/AT haplotype of IL-10 are associated with the more aggressive pattern of HCC, so those patients must be treated as early as possible.
**Abbreviations**
HBV: Hepatitis B virus; HCV: Hepatitis C virus; HCC: Hepatocellular carcinoma; TNF-α: Tumour necrosis factor-α; MTD: Maximum tumour diameter; PVT: Portal vein thrombosis; AFP: Alpha-fetoprotein; IL: Interleukin; CT: Computed tomography; PCR: Polymerase chain reaction; SNP: Single nucleotide polymorphism; AgI: Aggressiveness index; AST: Aspartate aminotransferase; ALT: Alanine aminotransferase
---
**Table 6 Association of HCC aggressiveness index with sociodemographic and clinical characteristics and genotype of HCC cases**
| AGI | A | B | C | Test of significance |
|-----|-------|-------|-------|----------------------|
| Age/years, mean ± SD | 54.64 ± 4.44 | 57.74 ± 4.98 | 56.18 ± 4.15 | F = 2.66, P = 0.08 |
| Sex, n (%) | | | | |
| Male | 13 (59.1) | 15 (65.2) | 15 (53.6) | χ² = 0.708, P = 0.702 |
| Female | 9 (40.9) | 8 (34.8) | 13 (46.4) | | |
| Albumin (g/dl), mean ± SD | 3.36 ± 0.62 | 3.23 ± 0.71 | 3.28 ± 0.46 | F = 0.280, P = 0.757 |
| Bilirubin (mg/dl), median (IQR) | 1.1 (0.90–1.7) | 1.68 (0.80–2.4) | 1.20 (0.825–1.5) | KW P = 0.514 |
| WBCS, mean ± SD | 5.25 ± 1.89 | 4.40 ± 1.21 | 4.72 ± 2.28 | F = 1.15, P = 0.32 |
| HB (g/dl), mean ± SD | 11.88 ± 1.43 | 10.91 ± 1.56 | 11.24 ± 1.84 | F = 1.15, P = 0.32 |
| Platelet *10³, mean ± SD | 85.09 ± 29.89 | 116.09 ± 56.5 | 101.25 ± 38.8 | F = 2.92, P = 0.061 |
| ALT (u/l), median (IQR) | 38.0 (27.0–43.0) | 36.0 (32.0–55.0) | 33.0 (29.25–65.75) | KW P = 0.482 |
| AST(u/l), median (IQR) | 56.0 (47.0–88.0) | 69.0 (66.0–89.0) | 71.0 (63.0–98.0) | KW P = 0.09 |
| INR, mean ± SD | 1.36 ± 0.13 | 1.41 ± 0.29 | 1.31 ± 0.15 | F = 1.71, P = 0.189 |
| DM, n (%) | 8 (36.4) | 9 (39.1) | 14 (50.0) | P = 0.58 |
| Hypertension, n (%) | 6 (27.3) | 6 (26.1) | 7 (25.0) | P = 0.98 |
| IL-10 genotypes | | | | |
| TT/AA | 2 (9.1)a | 6 (26.1)b | 21 (75.0)ab | MC P < 0.001* |
| CT/CA | 8 (36.4) | 9 (39.1) | 4 (14.3) | | |
| CC/CC | 12 (54.5) | 8 (34.8) | 3 (10.7) | | |
| TNF-α genotypes | | | | |
| GG | 0 (0.0)ab | 9 (39.1)ac | 26 (92.9)bc | MC P < 0.001* |
| GA | 12 (54.5) | 8 (34.8) | 2 (7.1) | | |
| AA | 10 (45.5) | 6 (26.1) | 0 (0.0) | | |
Acknowledgements
The authors would like to thank all patients who participated in the study.
Authors’ contributions
AA: choosing the idea, patient examination, writing and reviewing. AM: patient examination, writing and reviewing. MW: genotype testing. All authors have read and approved the final manuscript.
Funding
Nil.
Availability of data and materials
Available at the data archive at Specialized Medical Hospital, Faculty of Medicine, Mansoura University, Egypt
Ethics approval and consent to participate
Written consents from patients who participated in the study or from their families were obtained and approved by Mansoura Medical Ethics Committee (MMEC) of Faculty of Medicine. Code: R.19.01.411
Consent for publication
NA
Competing interests
There are no conflicts of interest.
Author details
1Department of Internal Medicine, Hepatology & Gastroenterology unit, Faculty of Medicine, Mansoura University, Mansoura, Egypt. 2Specialized Medical Hospital, Faculty of Medicine, Mansoura University, Mansoura, Egypt. 3Department of Clinical Pathology, Faculty of Medicine, Mansoura University, Mansoura, Egypt.
Received: 8 May 2020 Accepted: 18 August 2020
Published online: 01 September 2020
References
1. Averhoff FM, Glass N, Holtzman D (2012) Global burden of hepatitis C: considerations for health-care providers in the United States. Clin Infect Dis 55(S1):S10–S15
2. Waly Raphael S, Yangde Z, Yuxiang C (2012) Hepatocellular carcinoma: focus on different aspects of management. ISRN Oncol 2012:426763.
3. Yang JD, Roberts LR (2010) Epidemiology and management of hepatocellular carcinoma. Infect Dis Clin N Am 24(4):899–919
4. Yoshimura A (2006) Signal transduction of inflammatory cytokines and tumor development. Cancer Sci 97:439–447
5. Carr BI, Guera V (2016) A hepatocellular carcinoma aggressiveness index and its relationship to liver enzyme levels. Oncology 90(21–220
6. Aroucha DC, Carmo RF, Vasconcelos LR et al (2016) TNF-α and IL-10 polymorphisms increase the risk to hepatocellular carcinoma in HCV infected individuals. J Med Virol 88:1587–1596
7. Ventura N, Carr BI, Kori I, Guerra V, Shibolet O (2018) Analysis of aggressiveness factors in hepatocellular carcinoma patients undergoing transarterial chemoembolization. World J Gastroenterol 24(15):1641–1649
8. EASL-EORTC clinical practice guidelines (2012) management of hepatocellular carcinoma. J Hepatol 56:908–943
9. Shire NJ, Sherman KE (2015) Epidemiology of hepatitis C virus: a battle on new frontiers. Gastroenterol Clin N Am 44:699–716
10. National Comprehensive Cancer Network NCCN clinical practice guidelines in oncology. Hepatobiliary Cancers. Version 3. 2017.
11. Westbrook RH, Dushelko G (2014) Natural history of hepatitis C. J Hepatol 61:S58–S68
12. Seyda Seydel G, Kucukoglu O, Altimbas A, Demir OO, Yilmaz S, Akkiz H, Otani E, Sowa JP, Canbay A (2016) Economic growth leads to increase of obesity and associated hepatocellular carcinoma in developing countries. Ann Hepatol 15:662–672
13. Wei Y et al (2011) Polymorphisms of tumour necrosis factor-alpha and hepatocellular carcinoma risk: a HuGE systematic review and meta-analysis. Dig Dis Sci 56:2227–2236
14. Cheng K et al (2013) Tumour necrosis factor-alpha 238G/A polymorphism and risk of hepatocellular carcinoma: evidence from a meta-analysis. Asian Pac J Cancer Prev 14:3275–3279
15. Zhou P et al (2011) The TNF-alpha-238 polymorphism and cancer risk: a meta-analysis. PLoS One 6:e22092
16. Swiatek BJ (2012) Is interleukin-10 gene polymorphism a predictive marker in HCV infection? Cytokine Growth Factor Rev 23:47–59
17. Talaat RM, Elsmail AA, Elwakil R, Gurgis AA, Nasr MI (2012) Tumour necrosis factor-alpha _308G/A polymorphism and risk of hepatocellular carcinoma in hepatitis C virus-infected patients. Chin J Cancer 31:29–35
18. Radwan MI, Pasha HF, Moharred RH, Hussein HI, El-Khshab MN (2012) Influence of transforming growth-factor-b1 and tumour necrosis factor-α genes polymorphisms on the development of cirrhosis and hepatocellular carcinoma in chronic hepatitis C patients. Cytokine 60:271–276
19. Baghel K, Srivastava RN, Chandra A, Goel SK, Agrawal J, Kazmi HR, Raj S (2014) TNF-a, IL-6, and IL-8 cytokines and their association with TNF-α -308 G/A polymorphism and postoperative sepsis. J Gastrointest Surg 18:1486–1494
20. Karimi M, Goldie LC, Cruickshank MN, Moses EK, Abraham LJ (2009) A critical assessment of the factors affecting reporter gene assays for promoter SNP function: a reassessment of -308 TNF polymorphism function using a novel integrated reporter system. Eur J Hum Genet 17:1454–1462
21. Vikram NK, Bhattacharyya N, Luthra K, Misra A, Poddar PK, Pandey RM, Guleria R (2011) Associations of -308G/A polymorphism of tumor necrosis factor (TNF)-α gene and serum TNF-α levels with measures of obesity, intra-abdominal and subcutaneous abdominal fat, subclinical inflammation and insulin resistance in Asian Indians in north India. Dis Markers 31:39–46
22. Morsi MI et al (2006) Evaluation of tumour necrosis factor-alpha, soluble P-selectin, gamma-glutamyl transferase, glutathioneS-transferase-p and alpha fetoprotein in patients with hepatocellular carcinoma before and during chemotherapy. Br J Biomed Sci 63:74–78
23. Wang YY et al (2003) Increased serum concentrations of tumour necrosis factor-alpha are associated with disease progression and malnutrition in hepatocellular carcinoma. J Chin Med Assoc 66:593–598
24. Schmidt-Annas D, Rose-John S (2016) IL-6 pathway in the liver: from physiology to pathology to therapy. J Hepatol 64:1403–1415
25. Bouzagrou N, Hassen E, Farhat K, Bahri O, Gabbouj S, Maamouri N, Ben Mami N, Safar H, Trabelsi A, Triki H, Chouchane L (2009) Combined analysis of interferon-gamma and interleukin-10 gene polymorphisms and chronic hepatitis C severity. Hum Immunol 70:230–236
26. Coussens LM, Werb Z (2002) Inflammation and cancer. Nature 420:860–867
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Submit your manuscript to a SpringerOpen journal and benefit from:
- Convenient online submission
- Rigorous peer review
- Open access: articles freely available online
- High visibility within the field
- Retaining the copyright to your article
Submit your next manuscript at ► springeropen.com | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2865,
1
],
[
2865,
8019,
2
],
[
8019,
12952,
3
],
[
12952,
16154,
4
],
[
16154,
21938,
5
],
[
21938,
25097,
6
],
[
25097,
27688,
7
],
[
27688,
34014,
8
]
]
} |
ee7101478db6d1e80b788ead65076c19b5f39e42 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 4,
"total-fallback-pages": 0,
"total-input-tokens": 14067,
"total-output-tokens": 3318,
"fieldofstudy": [
"Physics"
]
} | Flow and Bose-Einstein Correlations in Au-Au Collisions at RHIC
Steven Manly for the PHOBOS Collaboration
B.B.Back\(^1\), M.D.Baker\(^2\), D.S.Barton\(^2\), R.R.Betts\(^6\), R.Bindel\(^7\), A.Budzanowski\(^3\), W.Busza\(^4\), A.Carroll\(^2\), M.P.Decowski\(^4\), E.Garcia\(^6\), N.George\(^1\), K.Gulbrandsen\(^4\), S.Gushue\(^2\), C.Halliwell\(^6\), J.Hamblen\(^8\), C.Henderson\(^1\), D.Hofman\(^6\), R.S.Hollis\(^6\), R.Holyński\(^3\), B.Holzman\(^2\), A.Iordanova\(^6\), E.Johnson\(^8\), J.Kane\(^4\), J.Katzy\(^4,6\), N.Khan\(^8\), W.Kucewicz\(^6\), P.Kulinich\(^4\), C.M.Kuo\(^5\), W.T.Lin\(^5\), S.Manly\(^8\), D.McLeod\(^6\), J.Michałowski\(^3\), A.Mignerey\(^7\), R.Nouicer\(^6\), A.Olszewski\(^3\), R.Pak\(^2\), I.C.Park\(^8\), H.Pernegger\(^4\), C.Reed\(^4\), L.P.Remsberg\(^2\), M.Reuter\(^6\), C.Roland\(^4\), G.Roland\(^4\), L.Rosenberg\(^4\), J.Sagerer\(^6\), P.Sarin\(^4\), P.Sawicki\(^3\), W.Skulski\(^8\), S.G.Steadman\(^4\), P.Steinberg\(^2\), G.S.F.Stephans\(^4\), M.Stodulski\(^9\), A.Sukhanov\(^2\), J.-L.Tang\(^5\), R.Teng\(^8\), A.Trzupek\(^3\), C.Vale\(^4\), G.J.van Nieuwenhuizen\(^4\), R.Verdier\(^4\), B.Wadsworth\(^4\), F.L.H.Wolfs\(^8\), B.Wosiek\(^3\), K.Woźniak\(^3\), A.H.Wuosmaa\(^1\), B.Wyslouch\(^4\)
\(^1\) Argonne National Laboratory, \(^2\) Brookhaven National Laboratory, \(^3\) Institute of Nuclear Physics, Kraków, Poland, \(^4\) Massachusetts Institute of Technology, \(^5\) National Central University, Chung-Li, Taiwan, \(^6\) University of Illinois at Chicago, \(^7\) University of Maryland, \(^8\) University of Rochester
Elliptic flow and Bose-Einstein correlations have been measured in Au-Au collisions at \(\sqrt{s_{NN}} = 130\) and 200 GeV using the PHOBOS detector at RHIC. The systematic dependencies of the flow signal on the transverse momentum, pseudorapidity, and centrality of the collision, as well as the beam energy are shown. In addition, results of a 3-dimensional analysis of two-pion correlations in the 200 GeV data are presented.
1. Introduction
The evolution of the space-time structure of the particle emitting source created in heavy ion collisions can be probed by measuring the azimuthal anisotropy (e.g., elliptic flow) and two-particle interferometry in the final state particle distributions. Elliptic flow is thought to provide information on the early stages of the collision, the nuclear equation of state and the degree of equilibration attained during the evolution of the collision \[\square\], while two-particle interferometry provides information on the temporal and spatial extent of the source \[\square\].
The results presented here are based on data taken during the first two RHIC physics runs for Au-Au collisions at \(\sqrt{s_{NN}} = 130\) and 200 GeV. All 200 GeV results presented here are preliminary. The PHOBOS detector employs silicon pad detectors to perform tracking, vertex detection and multiplicity measurements. Details of the setup and the layout of the silicon sensors can be found elsewhere \[\square\]. Event triggering and the determination of
the centrality were based on the information provided by two sets of scintillating paddle counters [4]. The raw data for these analyses came in the form of energy depositions from the passage of charged particles through individual detector pads, known as hits. The hit definition and signal processing procedure used for the flow analysis is described in reference [6]. The position of the primary collision vertex was determined on an event-by-event basis by extrapolating tracks found in the spectrometer arms and/or the vertex detector. The event plane was determined by a standard subevent technique [7] using hits in symmetric and uniform regions of the octagonal multiplicity detector. Charged particle tracks were reconstructed in the spectrometer arms using techniques described previously [8,9]. Pions were identified using the measured momentum and the specific ionization loss observed in the spectrometer silicon detectors.
2. Flow
Flow results from two independent analyses are presented here. In the first, known as the hit-based analysis, the event plane was determined from hits in the single layer Si multiplicity detectors and the second Fourier coefficient of the hit azimuthal angle distribution (also known as the elliptic flow), $v_2$, was evaluated by correlating the event plane to hits in a different region of the multiplicity detectors. In the second flow analysis, known as the track-based analysis, $v_2$ was determined by correlating the event plane to tracks found in the spectrometer arms.
For the hit-based analysis, events were chosen in a fiducial region that maximized the $\eta$ coverage and event plane sensitivity for the analysis. Equal multiplicity subevents were defined in the regions $0.1 < |\eta| < 2$ for the event plane determination and evaluation of the event plane resolution. Details of this analysis are provided in reference [6]. Results from the hit-based flow analysis are shown in Figures 1 and 2. The 1σ statistical errors are shown for both analyses. The 90% confidence level systematic errors are shown as boxes for the 200 GeV data points. As can be seen in these two figures, the flow signal is little changed with the increase in the center-of-mass energy of the collision from 130 to 200 GeV. The unique, and very nearly complete, $\eta$ coverage shown in Figure 2 shows a substantial drop in $v_2$ as a function of $|\eta|$ that is not yet understood [6].
In the track-based analysis, events were chosen in a fiducial region that maximized the tracking efficiency and the reaction plane sensitivity. Subevents were defined roughly in the regions $2 < |\eta| < 3$ for the event plane determination and the evaluation of the reaction plane resolution. Charged tracks reconstructed in the spectrometer arms in the region $0 < \eta < 2.5$ were used to determine the elliptic flow. In detail, the track-based flow analysis is quite different from our previously released flow analysis. Differences include the use of a vertex dependent reaction plane weighting and resolution correction, as well as a larger separation of the subevents in $\eta$. The flow signal is determined as the asymmetry in the track azimuthal angle distribution measured relative to the event plane. The track-based flow results are less dependent on Monte Carlo corrections and less sensitive to background and non-flow correlations as compared to the hit-based flow measurements.
Results from this procedure are shown in Figures 3 and 4. The graphical representation of the errors in these figures is similar to that for the first two figures. Figure 3 shows the
elliptic flow signal as a function of the number of participants for the track-based analysis overlayed with that from the hit-based analysis for 200 GeV data. The two techniques agree very well. This is significant because of the differing sensitivity to background and non-flow correlations. Figure 2 gives the transverse momentum dependence of the elliptic flow of charged hadrons in the 200 GeV data. The saturation observed at a transverse momentum greater than 2 GeV/c is similar to what was observed at 130 GeV, and has been interpreted as evidence for partonic energy loss through gluon radiation in a dense system [10].
3. Bose-Einstein correlations
Pairs of identified same-sign pions were used to calculate the two-particle correlation functions. The results were corrected for the effects of the tracking algorithm and the Coulomb repulsion of the pions. The 3-dimensional analysis of the correlation function using the Bertsch-Pratt parameterization was performed in the LCMS frame in the region of acceptance $0.2 < y < 1.5$ and $150 < k_T < 350$ MeV/c. For the 15% most central events, the preliminary fitted source parameters for $\pi^-\pi^- (\pi^+\pi^+)$ pairs are as follows: $\lambda = 0.54 \pm 0.02(0.57 \pm 0.03)$, $R_{out} = 5.8 \pm 0.2(5.8 \pm 0.2)$ fm, $R_{side} = 5.1 \pm 0.4(4.9 \pm 0.4)$ fm, $R_{long} = 6.8 \pm 0.3(7.3 \pm 0.3)$ fm, and $R_{out-long} = 4.9 \pm 1.7(4.5 \pm 1.9)$ fm. The errors listed are statistical only. In addition, there are systematic errors of $\pm 0.06$ on the values of $\lambda$ and $\pm 1$ fm on the radii. The results reported here are similar to those observed in Au-Au collisions at $\sqrt{s_{NN}} = 130$ GeV [11,12].
4. Summary
Recent PHOBOS measurements of elliptic flow and two-particle correlations in Au-Au collisions at $\sqrt{s_{NN}} = 200$ GeV are very similar to values observed at $\sqrt{s_{NN}} = 130$ GeV.
Figure 3. Elliptic flow as a function of the number of participants for Au-Au collisions at $\sqrt{s_{NN}} = 200$ GeV for the hit-based and track-based analyses.
Figure 4. Elliptic flow as a function of transverse momentum for Au-Au collisions at $\sqrt{s_{NN}} = 200$ GeV.
Notably, the new results at 200 GeV clearly show a saturation of $v_2$ for $p_T > 2$ GeV/c and a dramatic drop of $v_2$ as a function of $|\eta|$.
Acknowledgments: This work was partially supported by US DoE grants DE-AC02-98CH10886, DE-FG02-93ER40802, DE-FC02-94ER40818, DE-FG02-94ER40865, DE-FG02-99ER41099, W-31-109-ENG-38, and NSF grants 9603486, 9722606 and 0072204. The Polish group was partially supported by KBN grant 2-P03B-10323. The NCU group was partially supported by NSC of Taiwan under contract NSC 89-2112-M-008-024.
REFERENCES
1. W. Reisdorf and H. G. Ritter, Ann. Rev. Nucl. Part. Sci. 47 (1997) 663.
2. U. A. Wiedemann and U. Heinz, Phys. Rep. 319 (1999) 145.
3. PHOBOS Collaboration, H. Pernegger et al., Nucl. Inst. Meth. A473 (2001) 197.
4. PHOBOS Collaboration, B. B. Back et al., Nucl. Phys A698 (2002) 416.
5. PHOBOS Collaboration, B. B. Back et al., Phys. Rev. C65 (2002) 031901.
6. PHOBOS Collaboration, B. B. Back et al., nucl-ex/0205021, accepted by Phys. Rev. Lett.
7. A. M. Poskanzer and S. A. Voloshin, Phys. Rev. C58 (1998) 1671.
8. PHOBOS Collaboration, B. B. Back et al., Phys. Rev. Lett. 87, (2001) 102301.
9. PHOBOS Collaboration, B. B. Back et al., nucl-ex/0206012, submitted to Phys. Rev. C.
10. STAR Collaboration, C. Adler et al., nucl-ex/0206006, submitted to Phys. Rev. Lett.
11. STAR Collaboration, C. Adler et al., Phys. Rev. Lett. 87 (2001) 082301.
12. PHENIX Collaboration, K. Adcox et al., Phys. Rev. Lett. 88 (2002) 192302. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3060,
1
],
[
3060,
6664,
2
],
[
6664,
8544,
3
],
[
8544,
10294,
4
]
]
} |
a7feb80fb9aa958e0a360ee17e51aaa1c7fa3eeb | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 12,
"total-fallback-pages": 0,
"total-input-tokens": 38308,
"total-output-tokens": 12048,
"fieldofstudy": [
"Computer Science"
]
} | On the iterative refinement of densely connected representation levels for semantic segmentation
Arantxa Casanova\textsuperscript{1,2}, Guillem Cucurull\textsuperscript{1,2}, Michal Drozdzal\textsuperscript{1,3}, Adriana Romero\textsuperscript{1,3}, Yoshua Bengio\textsuperscript{1}
\textsuperscript{1} Montreal Institute for Learning Algorithms \textsuperscript{2} Computer Vision Center, Barcelona \textsuperscript{3} Facebook AI Research
Abstract
State-of-the-art semantic segmentation approaches increase the receptive field of their models by using either a downsampling path composed of poolings/strided convolutions or successive dilated convolutions. However, it is not clear which operation leads to best results. In this paper, we systematically study the differences introduced by distinct receptive field enlargement methods and their impact on the performance of a novel architecture, called Fully Convolutional DenseResNet (FC-DRN). FC-DRN has a densely connected backbone composed of residual networks. Following standard image segmentation architectures, receptive field enlargement operations that change the representation level are interleaved among residual networks. This allows the model to exploit the benefits of both residual and dense connectivity patterns, namely: gradient flow, iterative refinement of representations, multi-scale feature combination and deep supervision. In order to highlight the potential of our model, we test it on the challenging CamVid urban scene understanding benchmark and make the following observations: 1) downsampling operations outperform dilations when the model is trained from scratch, 2) dilations are useful during the finetuning step of the model, 3) coarser representations require less refinement steps, and 4) ResNets (by model construction) are good regularizers, since they can reduce the model capacity when needed. Finally, we compare our architecture to alternative methods and report state-of-the-art result on the Camvid dataset, with at least twice fewer parameters.
1. Introduction
Convolutional Neural Networks (CNNs) have been extensively studied in the computer vision literature to tackle a variety of tasks, such as image classification \cite{14,17,16}, object detection \cite{12} and semantic segmentation \cite{9,21,6}. Major advances have been driven by novel very deep architectural designs \cite{14,17}, introducing skip connections to facilitate the forward propagation of relevant information to the top of the network, and provide shortcuts for gradient flow. Very deep architectures such as residual networks (ResNets) \cite{14}, densely connected networks (DenseNets) \cite{17} and squeeze-and-excitation networks \cite{16} have exhibited outstanding performance on standard large scale computer vision benchmarks such as ImageNet \cite{39} and MSCOCO \cite{28}.
Among top performing classification networks, ResNets challenge the hierarchical representation learning view of CNNs \cite{26,42,11}. The hierarchical representation view associates the layers in network to different levels of abstraction. However, contrary to previous architectures such as \cite{40}, dropping or permuting almost any layer in a ResNet has shown to only minimally affect their overall performance \cite{42}, suggesting that the operations applied by a single layer are only a small modification to the identity operation. Significant effort has been devoted to analyzing and understanding these findings. On one hand, it has been argued that ResNets behave as an ensemble of shallow networks, averaging exponentially many subnetworks, which use different subsets of layers \cite{42}. On the other hand, it has been suggested that ResNets engage in an unrolled iterative estimation of representations, that refine upon their input \cite{11}. These arguments have been exploited in \cite{8} to learn normalized inputs for iterative estimation, highlighting the importance of having transformations prior to the residual blocks.
Fully Convolutional Networks (FCNs) were presented in \cite{30,38} as an extension of CNNs to address per pixel prediction problems, by endowing standard CNNs with an upsampling path to recover the input spatial resolution at their output. In the recent years, FCN counterparts and enhanced versions of top performing classification networks have been successfully introduced in the semantic segmentation literature. Fully Convolutional ResNets (FC-ResNets) were presented and analyzed in \cite{9} in the context of medical image segmentation. Moreover, Fully Convolutional DenseNets (FC-DenseNets) \cite{21} were proposed to build low capacity networks for semantic segmentation, taking advantage of iterative concatenation of features maps.
In this paper, we further exploit the iterative refinement properties of ResNets to build densely connected residual networks for semantic segmentation, which we call Fully Convolutional DenseResNets (FC-DRNs). Contrary to FC-DenseNets \cite{21}, where the convolution layers are densely connected, FC-DRN apply dense connectivity to ResNets models. Thus, our model performs iterative refinement at each representation level (in a single ResNet) and uses dense connectivity to obtain refined multi-scale feature representations (from multiple ResNets) in the pre-softmax layer. We demonstrate the potential of our architecture on the challenging CamVid \cite{4} urban scene understanding benchmark and report state-of-the-art results. To compare and contrast with common pipelines based on top performing classification CNNs, we perform an in depth analysis on different downsampling operations used in the context of semantic segmentation: dilated convolution, strided convolution and pooling. Although dilated convolutions have been well adopted in the semantic segmentation literature, we show that such operations seem to be beneficial only when used to finetune a pre-trained network that applies downsampling operations (e.g. pooling or strided convolution). When trained from scratch, dilation-based models are outperformed by their pooling and strided convolutions-based counterparts, highlighting the generalization capabilities of downsampling operations.
The contributions of our paper can be summarized as:
- We combine FC-DenseNets and FC-ResNets into a single model (FC-DRN) that fuses the benefits of both architectures: gradient flow and iterative refinement from FC-ResNets as well as multi-scale feature representation and deep supervision from FC-DenseNets.
- We show that FC-DRN model achieves state-of-the-art performance on CamVid dataset \cite{4}. Moreover, FC-DRN outperform FC-DenseNets, while keeping the number of trainable parameters small.
- We provide an analysis on different operations enlarging the receptive field of a network, namely poolings, strided and dilated convolutions. We inspect FC-DRN by dropping ResNets from trained models as well as by visualizing the norms of the weights of different layers. Our experiments suggest that the benefits of dilated convolutions only apply when combined with pre-trained networks that contain downsampling operations. Moreover, we show that ResNets (by model construction) are good regularizers, since they can reduce the model capacity at different representation levels when needed, and adapt the refinement steps.
2. Related work
In the recent years, FCNs have become the de facto standard for semantic segmentation. Top performing classification networks have been successfully extended to perform semantic segmentation \cite{43,34,9,44,21}.
In order to overcome the spatial resolution loss induced by successive downsampling operations of classification networks, several alternatives have been introduced in the literature; the most popular ones being long skip connections in encoder-decoder architectures \cite{30, 2, 38, 19} and dilated convolutions \cite{46, 7}. Long skip connections help recover the spatial information by merging features skipped from different resolutions on the contracting path, whereas dilated convolutions enlarge the receptive field without downsizing the feature maps.
Another line of research seeks to endow segmentation pipelines with the ability to enforce structure consistency to their outputs. The contributions in this direction include Conditional Random Fields (CRFs) and its variants (which remain a popular choice) \cite{24, 7, 49}, CRFs as Recurrent Neural Networks \cite{49}, iterative inference denoising autoencoders \cite{37}, convolutional pseudo-priors \cite{45}, as well as graph-cuts, watersheds and spatio-temporal regularization \cite{3, 44, 25}.
Alternative solutions to improve the performance of segmentation models are based on combining features at different levels of abstraction. Efforts in this direction include iterative concatenation of feature maps \cite{17, 21}; fusing upsampled feature maps with different receptive fields prior to the softmax classifier \cite{5}, along the network \cite{27, 1} or by means of two interacting processing streams operating at different resolutions \cite{33}; gating skip connections between encoder and decoder to control the information to recover \cite{19}; and using a pyramid pooling module with different spatial resolutions for context aggregation \cite{48}. Moreover, incorporating global features has long shown to improve semantic segmentation performance \cite{10, 29}.
Finally, semantic segmentation performance has also been improved by training with synthetic data \cite{35}, propagating information through video data \cite{20}, or modeling uncertainties in the model \cite{23}.
3. Fully Convolutional DenseResNet
In this section, we briefly review both ResNets and DenseNets, and introduce the FC-DRN architecture.
3.1. Background
Let us denote the feature map representation of the $l$-th layer of the model as $x_l$. Traditionally, in CNNs, the feature map $x_l$ is obtained by applying a transformation $H$, composed of a convolution followed by a non-linearity, to the $l-1$-th feature map $x_{l-1}$ as $x_l = H(x_{l-1})$. CNNs are built by stacking together multiple such transformations. However, due to the non-linearity operation, optimization of such networks becomes harder with growing depth. Architectural solutions to this problem have been proposed in
ResNets [14] and DenseNets [17].
In ResNets, the representation of \( l \)-th feature map is obtained by learning the residual transformation \( H \) of the input feature map \( x_{l-1} \) and summing it with the input \( x_{l-1} \). Thus, the \( l \)-th feature map representation can be computed as follows: \( x_l = H(x_{l-1}) + x_{l-1} \). This simple modification in network’s connectivity introduces a path that has no non-linearities, allowing to successfully train networks that have hundreds (or thousands) of layers. Moreover, lesion studies performed on ResNets have opened the door to research directions that try to better understand how these networks work. Following these lines, it has been suggested that ResNets layers learn small modifications of their input (close to the identity operation), engaging in an iterative refinement of their input.
In DenseNets, the \( l \)-th feature map is obtained by applying a transformation \( H \) to all previously obtained feature maps such that \( x_l = H([x_0, x_1, \ldots, x_{l-1}]) \), where \([\cdot, \cdot]\) denotes the concatenation operation. One can easily notice that when following the dense connectivity pattern in DenseNets, the pre-softmax layer receives the concatenation of all previous feature maps. Thus, DenseNets introduce deep feature supervision by means of their model construction. It has been shown that using the deep connectivity pattern one can train very deep models that outperform ResNets [17]. Moreover, it is worth mentioning that combining information at different representation levels has shown to be beneficial in the context of semantic segmentation [10, 21].
### 3.2. FC-DRN model
FC-DRNs extend the FC-DenseNets architecture of [21] and incorporate a dense connectivity pattern over multiple ResNets [14]. Thus, FC-DRNs combine the benefits of both architectures: FC-DRNs perform iterative estimation at each abstraction level (by using ResNets) and combine different abstraction levels while obtaining deep supervision (by means of DenseNets connections).
The connectivity pattern of FC-DRN is visualized in Figure 1. First, the input is processed with an Initial Downsampling Block (IDB) composed of a single convolution followed by \( 2 \times 2 \) pooling operation and two \( 3 \times 3 \) convolutions. Then, the output is fed to a dense block (the densely connected part of the model), which is composed of ResNets, transformations and concatenations, forming a downsampling path followed by an upsampling path.
In our model, there are 9 ResNets, motivated by the standard number of downsampling and upsampling operations in the FCN literature. Each ResNet is composed of 7 basic blocks, computing twice the following operations: batch normalization, ReLU activation, dropout and \( 3 \times 3 \) convolution. After each ResNet, we apply a transformation with the goal of changing the representation level. This transformation is different in the downsampling and upsampling paths: in the downsampling path, it can be either a pooling, a strided convolution or a dilated convolution; whereas in the upsampling path, it can be either an upsampling to compensate for pooling/strided convolution or a \( 1 \times 1 \) convolution in case of dilated convolutions, to keep models with roughly the same capacity. Following [7, 6], transformations in the dilation-based model adopt a multi-grid pattern (for more details see Figure 1 in the supplementary material).
The outputs of the transformations are concatenated such that the input to the subsequent ResNet incorporates information from all the previous ResNets. Concatenations are
\[\text{Note that in [21] the dense connectivity pattern is over convolutional operations.}\]
performed over channel dimensions and, if needed, the resolution of the feature maps is adjusted using transformations that are applied independently to each concatenation input\(^2\). After each concatenation, there is a \(1 \times 1\) convolution to mix the features. Finally, the output of the dense block is fed to a Final Upsampling Block (FUB) that adapts the spatial resolution and the number of channels in the model output. A detailed description of the architecture is available in Table 1 of the supplementary material.
4. Analysis and Results
In this section, we assess the influence of applying different kinds of transformations between different ResNets and report our final results.
All experiments are conducted on the CamVid\(^4\) dataset, which contains images of urban scene. Each image has a resolution of \(360 \times 480\) pixels and is densely labeled with 11 semantic classes. The dataset consists of 367, 101 and 233 frames for training, validation and test, respectively. In order to compare different architectures, we report results on the validation set with two metrics: mean intersection over union (mean IoU) and global accuracy.
All networks were trained following the same procedure. The weights were initialized with HeUniform\(^1\), and the networks were trained with RMSProp optimizer\(^41\), with a learning rate of \(1e^{-3}\) and an exponential decay of 0.995 after each epoch. We used a weight decay of \(1e^{-4}\) and dropout rate of 0.2. The dataset was augmented with horizontal flipping and crops of 324x324. We used early stopping on the validation mean IoU metric to stop the training, with a patience of 200 epochs.
4.1. FC-DRN transformation variants
State-of-the-art classification networks downsample their feature maps’ resolution by successively applying pooling (or strided convolution) operations. In order to mitigate the spatial resolution loss induced by such subsampling layers, many segmentation models only allow for a number of subsampling operations and change the remaining ones by dilated convolutions\(^7\)\(^46\)\(^47\). However, in some other cases\(^2\)\(^3\)\(^1\)\(^3\)\(^8\), the number of downsampling operations is preserved, recovering fine-grained information via long skip connections. Therefore, we aim to analyze and compare the influence of pooling/upsampling operations versus dilated convolutions. To that aim, we build sister architectures, which have an initial downsampling block, followed by a dense block, and a final upsampling block, as described in Section 3.2, and only differ in the transformation operations applied within their respective dense blocks.
Max-Pooling architecture (FC-DRN-P): This architecture interleaves ResNets with four max-pooling operations (downsampling path) and four nearest neighbor upsamplings followed by \(3x3\) convolutions to smooth the output (upsampling path).
Strided convolution architecture (FC-DRN-S): This architecture interleaves ResNets with four strided convolution operations (downsampling path) and four nearest neighbor upsamplings followed by \(3x3\) convolutions to smooth the output (upsampling path).
Dilated architecture (FC-DRN-D): This architecture interleaves ResNets with four multi-grid dilated convolution operations of increasing dilation factor (2, 4, 16 and 32\(^2\)) and standard convolutions to emulate the upsampling operations. Note that the dense block of this architecture does not change the resolution of its feature maps.
FC-DRN-P finetuned with dilations (FC-DRN-P-D): This architecture seeks to mimic state-of-the-art models based on top performing classification networks, which replace the final subsampling operations with dilated convolutions\(^7\)\(^46\)\(^47\). More precisely, we substitute the last two pooling operations of FC-DRN-P by dilated convolutions of dilation rate 4 and 8, respectively. Following the spirit of FC-DRN-D, the first two upsampling operations become standard convolutions. We initialize our dilated convolutions to the identity, as suggested in\(^46\).
FC-DRN-S finetuned with dilations (FC-DRN-S-D): Following FC-DRN-P-D, we substitute the last two strided convolution operations of FC-DRN-S by dilated convolutions (rates 4 & 8), whereas the first two upsampling operations become standard convolutions. In this case, we initialize the weights of the dilated convolutions with the weights of the corresponding pre-trained strided convolutions.
Table 1 reports the validation results for the described architectures. Among the networks trained from scratch, FC-DRN-P achieves the best performance in terms of mean IoU by a margin of 0.8% and 3.7% w.r.t. FC-DRN-S and FC-DRN-D, respectively. When finetuning the pooling and strided convolution architectures with dilations, we further improve the results to 81.7% and 81.1%, respectively. It is worth noting that we also tried training FC-DRN-P-D and FC-DRN-S-D from scratch, which yielded worse results, highlighting the benefits of pre-training with poolings/strided convolutions, which capture larger contiguous contextual information.
Figure 2 presents qualitative results from all architectures. As illustrated in the figure, FC-DRN-P prediction seems to better capture the global information, when compared to FC-DRN-D. This can be observed on the left part of the predictions, where dilated convolutions predict different classes for isolated pixels. Max-poolings understand better
\(^2\)Note that in order to maintain the number of transformations when comparing different models (e.g. pooling-based vs. dilation-based model), we apply a convolution even when concatenating same resolution feature maps.
\(^3\)We tested many different variants of dilation factors and found out that this multi-grid structure gives the best results.
we finetuned it by using soft targets [15, 36] (while only using 13% of its parameters. The model still preserves the global information of FC-DRN-P, reducing some errors that were present on the left part of the image in the FC-DRN-D prediction. If we take a look at the FC-DRN-S prediction, we can see that it recovers the pedestrian on the right successfully, but fails to correctly segment the sidewalk and the pedestrians on the left. Finetuning this architecture with dilations (FC-DRN-S-D) helps capture a lost pedestrian on the left, but still lacks the ability to sharply sketch the right column pole. Furthermore, there are some artifacts on the top part of the image in both strided convolution-based architectures.
### 4.2. Results
Following the comparison in Table 1, we report our final results on the FC-DRN-P-D architecture. Recall that this architecture is a pre-trained FC-DRN with a dense block of 4 max pooling operations (downsampling path) and 4 repeat and convolve operations (upsampling path) and finetuned by substituting the last two max poolings and the first two upsamplings by dilated convolutions, on the same data.
Table 2 compares the performance of our model to state-of-the-art models. As shown in the table, our FC-DRN-P-D exhibits state-of-the-art performance when compared to previous methods, especially when it comes to segmenting under-represented classes such as column poles, pedestrians and cyclists. It is worth noting that our architecture improves upon pre-trained models with 10 times more parameters. When compared to FC-DenseNets, FC-DRN outperforms both FC-DenseNet67 (with a comparable number of parameters) and FC-DenseNet103 (with only 41.5% of its parameters) by 2.5% and 1.4% mean IoU, respectively. Moreover, it also exceeds the performance of more recent methods such as G-FRNet, which uses gated skip connections between encoder and decoder in a FCN architecture, while only using 13% of its parameters.
In order to further boost the performance of our network, we finetuned it by using soft targets [13, 36] (0.9 and 0.01, instead of 1 and 0 in the target representation), improving generalization and obtaining a final score of 69.4% mean IoU and 91.6% global accuracy. Using soft targets allows the network to become more accurate in predicting classes such as pedestrian, fence, column pole, sign, building and sidewalk when compared to the original version. FC-DRN recovers slim objects such as column poles and pedestrians much better than other architectures presented in the literature, while maintaining a good performance on classes composed of larger objects.
It is worth mentioning that, unlike most of current state-of-the-art methods, FC-DRN have not been pre-trained on large datasets such as ImageNet [39]. Moreover, there are other methods in the literature that exploit virtual images to augment the training data [15] or that leverage temporal information to improve performance [20]. Note that those enhancements complement each other and FC-DRN could most likely benefit from them to boost their final performance as well. However, we leave those as future work.
Figure 3 shows some FC-DRN segmentation maps (right) compared to their respective ground truths (middle). We can observe that the segmentations have good quality, aligned with the quantitative results we obtained. The column poles and pedestrians are sharply segmented, but some difficulties arise when trying to distinguish between sidewalks and roads or in the presence of small road signs.
#### 5. Delving deeper into FC-DRN transformations
In this section, we provide an in depth analysis of the variants of the trained FC-DRN architectures to compare different transformation operations: pooling, dilation and strided convolution. We start by dropping ResNets from a FC-DRN and then look into the weight’s norms of all ResNets in the models. We end the section with a discussion exploiting the observations from the network inspection.
We follow the strategy of dropping layers introduced in [42, 18], and drop all residual blocks of a ResNet (we only keep the first residual block that adjusts the depth of the feature maps) with the goal of analyzing the implications of using different transformation operations. The results of the experiment are shown in Figure 4. On one hand, Figure 4(a) reports the performance drops in percentage of mean IoU for each ResNet in the initial networks (i.e. FC-DRN-P, FC-DRN-D and FC-DRN-S). Surprisingly, dropping ResNets 3 to 8 barely affects the performance of FC-DRN-D. However, both pooling and strided convolution models suffer from the loss of almost any ResNet. On the other hand, Figure 4(b) presents the results of dropping ResNets in the finetuned models (i.e. FC-DRN-P-D and FC-DRN-P-S). Finetuning the pooling network with dilations makes ResNets 5 and 6 slightly less necessary, while ResNet 8 becomes the most relevant one. Finetuning the strided convolution net-
Figure 2: Qualitative results on the test set: (a) input image, (b) ground truth, (c) FC-DRN-P prediction, (d) FC-DRN-D prediction, (e) FC-DRN-P-D prediction, (f) FC-DRN-S prediction and (g) FC-DRN-S-D prediction. Main differences are highlighted with white boxes.
Table 2: Results on CamVid test set, reported as IoU per class, mean IoU and global accuracy, compared to state-of-the-art.
| Model | # params [M] | Ext. Data | Building | Tree | Sky | Car | Sign | Road | Pedestrian | Fence | Column pole | Sidewalk | Cyclist | mean IoU [%] | Gl. acc. [%] |
|------------------------|--------------|-----------|----------|------|-----|-----|------|------|------------|-------|------------|----------|---------|--------------|-------------|
| SegNet [2] | 29.5 | Yes | 68.7 | 52.0 | 87.0| 58.5| 13.4 | 86.2 | 25.3 | 17.9 | 16.0 | 60.5 | 24.8 | 46.4 | 62.5 |
| DeconvNet [31] | 252 | Yes | n/a | n/a | n/a| n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a | 48.9 | 85.9 |
| FCN8 [30] | 134.5 | Yes | 77.8 | 71.0 | 88.7| 76.1| 32.7 | 91.2 | 41.7 | 24.4 | 19.9 | 72.7 | 31.0 | 57.0 | 88.0 |
| Visin et al. [43] | 32.3 | Yes | n/a | n/a | n/a| n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a | 58.8 | 88.7 |
| DeepLab-LFOV [57] | 37.3 | Yes | 81.5 | 74.6 | 89.0| 82.2| 42.3 | 92.2 | 48.4 | 27.2 | 14.3 | 75.4 | 50.1 | 61.6 | - |
| Bayesian SegNet [22] | 29.5 | Yes | n/a | n/a | n/a| n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a | 63.1 | 86.9 |
| Dilation8 [46] | 140.8 | Yes | 82.6 | 76.2 | 89.0| 84.0| 46.9 | 92.2 | 56.3 | 35.8 | 23.4 | 75.3 | 55.5 | 65.3 | 79.0 |
| FC-DenseNet67 [21] | 3.5 | No | 80.2 | 75.4 | 93.0| 78.2| 40.9 | 94.7 | 58.4 | 30.7 | 8.4 | 81.9 | 52.1 | 65.8 | 90.8 |
| Dilation8 + FSO [25] | 130 | Yes | 84.0 | 77.2 | 91.3| 85.6| 49.9 | 92.5 | 59.1 | 37.6 | 16.9 | 76.0 | 57.2 | 66.1 | 88.3 |
| FC-DenseNet103 [42] | 9.4 | No | 83.0 | 77.3 | 93.0| 77.3| 43.9 | 94.5 | 59.6 | 37.1 | 37.8 | 82.2 | 50.5 | 66.9 | 91.5 |
| G-FRNet [19] | 30 | Yes | 82.5 | 76.8 | 92.1| 81.8| 43.0 | 94.5 | 54.6 | 47.1 | 33.4 | 82.3 | 59.4 | 68.0 | 90.8 |
| FC-DRN-P-D | 3.9 | No | 82.6 | 75.7 | 92.6| 79.9| 42.3 | 94.1 | 61.2 | 36.9 | 42.6 | 81.2 | 61.8 | 68.3 | 91.4 |
| FC-DRN-P-D + ST | 3.9 | No | 83.5 | 75.6 | 92.1| 78.5| 46.6 | 93.9 | 62.7 | 44.3 | 43.1 | 82.2 | 60.8 | 69.4 | 91.6 |
Table 2: Results on CamVid test set, reported as IoU per class, mean IoU and global accuracy, compared to state-of-the-art.
Figure 3: Qualitative results on the CamVid test set. Left: images, middle: ground truths, right: FC-DRN predictions.
Figure 4: Results of dropping ResNets from a trained FC-DRN reported for the validation set: y-axis represents the loss in mean IoU when comparing to the model with all ResNets, x-axis represents the ID of the dropped ResNet.
Figure 5: Results of dropping ResNets from a trained FC-DRN reported for the validation set: y-axis represents the loss in mean IoU when comparing to the model with all ResNets, x-axis represents the ID of the dropped ResNet.
It is important to note that the structure of ResNets, due to the usage of the residual block, allows the model to self-adjust its capacity when needed, forcing the residual transformation of the residual block to be close to 0 and using the identity connection to forward the information. We hypothesize that this behavior of residuals is observed for some layers in our model (as it is shown in Figure 5). To test our hypothesis, we decided to reduce the capacity of trained FC-DRN by removing layers from the residuals of ResNets for which the norm of weights is small and to monitor the performance of the compressed FC-DRN model. If our hypothesis is true, then removing the residuals in the layers where the norm is close to 0 should not affect strongly the model performance. We choose to drop the layers whose weight norms are close enough to zero, based on visual inspection of Figure 5, thus allowing each representation level to have different number of refinement steps. The results of this trained model compression experiment are reported in Table 3. We can see that after removing 8% of the parameters from FC-DRN-P and FC-DRN-S, there is a drop in mean IoU on validation set of $-1.6$ and $-5.4$, respectively. Interestingly, we were able to remove 38% of weights from FC-DRN-D model while only experiencing a drop of $-0.8$ in mean IoU. Both finetuned models can be compressed by removing 15% of the capacity with slight performance drops of $-1$ and $-1.7$ for FC-DRN-P-D and FC-DRN-S-D, respectively. In general, it seems that finetuning the models with dilations not only improves the segmentation results but also makes the model more compressible.
Finally, we test if the optimization process of the high capacity FC-DRN reaches better local minima, due to self-adjustment of ResNets’ capacity, than if we train a low capacity FC-DRN from scratch. To this end, we trained from scratch the reduced capacity FC-DRN-D model and compared the results to the numbers reported in Table 3. The re-trained model obtained the mean validation IoU of 76.6. This is 0.8% below the result reported for the high capacity FC-DRN-D. We hypothesize that the model capacity reduction during the optimization process helps in reach-
6. Conclusions
In this paper, we combined two standard image segmentation architectures (FC-DenseNets and FC-ResNets) into a single network that we called Fully Convolutional DenseResNet. Our FC-DRN fuses the benefits of both models: gradient flow and iterative refinement from FC-ResNets as well as multi-scale feature representation and deep supervision from FC-DenseNets. We demonstrated the potential of our model on the challenging CamVid urban scene understanding benchmark and reported state-of-the-art results, with at least 2x fewer parameters than concurrent models.
Additionally, we analyzed different downsampling operations used in the context of semantic segmentation: dilated convolution, strided convolution and pooling. We inspected the FC-DRN by dropping ResNets from the trained models as well as by visualizing the weight norms of different layers and showed that ResNets (by model construction) are good regularizers, since they can reduce the model capacity when needed. In this direction, we observed that coarser representations seem to benefit from less refinement steps. Moreover, our results comparing different transformations suggest that pooling offers the best generalization capabilities, while the benefits of dilated convolutions only apply when combined with pre-trained networks that contain downsampling operations.
Acknowledgments
The authors would like to thank the developers of PyTorch [32]. We acknowledge the support of the following agencies for research funding and computing support: CI-
FAR, Canada Research Chairs, Compute Canada and Calcul Québec, as well as NVIDIA for the generous GPU support.
References
[1] I. Ardiyanto and T. B. Adj. Deep residual coalesced convolutional network for efficient semantic road segmentation. MVA, 2017.
[2] V. Badrinarayanan, A. Kendall, and R. Cipolla. Segnet: A deep convolutional encoder-decoder architecture for image segmentation. CoRR, 2015.
[3] T. Beier, B. Andres, U. Köthe, and F. A. Hamprecht. An efficient fusion move algorithm for the minimum cost lifted multicut problem. In Lecture Notes in Computer Science. 2016.
[4] G. J. Brostow, J. Shotton, J. Fauqueur, and R. Cipolla. Segmentation and recognition using structure from motion point clouds. In ECCV, 2008.
[5] H. Chen, X. Qi, J.-z. Cheng, and P.-a. Heng. Deep contextual networks for neuronal structure segmentation. AAAI, 2016.
[6] L. Chen, G. Papandreou, F. Schroff, and H. Adam. Re-thinking atrous convolution for semantic image segmentation. CoRR, 2017.
[7] L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille. DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs. arXiv:1606.00915, 2016.
[8] M. Drozdzal, G. Chartrand, E. Vorontsov, M. Shakeri, L. Di Jorio, A. Tang, A. Romero, Y. Bengio, C. Pal, and S. Kadoury. Learning normalized inputs for iterative estimation in medical image segmentation. Medical image analysis, 2018.
[9] M. Drozdzal, E. Vorontsov, G. Chartrand, S. Kadoury, and C. Pal. The importance of skip connections in biomedical image segmentation. Deep Learning and Data Labeling for Medical Applications, 2016.
[10] C. Gatta, A. Romero, and J. van de Veijer. Unrolling Loopy Top-Down Semantic Feedback in Convolutional Deep Networks. In CVPRW, 2014.
[11] K. Greff, R. K. Srivastava, and J. Schmidhuber. Highway and residual networks learn unrolled iterative estimation. ICLR, 2017.
[12] K. He, G. Gkioxari, P. Dollár, and R. Girshick. Mask r-cnn. arXiv preprint arXiv:1703.06870, 2017.
[13] K. He, X. Zhang, S. Ren, and J. Sun. Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification. ICCV, 2015.
[14] K. He, X. Zhang, S. Ren, and J. Sun. Deep Residual Learning for Image Recognition.CVPR, 2016.
[15] G. Hinton, O. Vinyals, and J. Dean. Distilling the Knowledge in a Neural Network. Deep Learning workshop at NIPS, 2014.
[16] J. Hu, L. Shen, and G. Sun. Squeeze-and-excitation networks. arXiv preprint arXiv:1709.01507, 2017.
[17] G. Huang, Z. Liu, and K. Q. Weinberger. Densely Connected Convolutional Networks. CVPR, 2017.
[18] G. Huang, Y. Sun, Z. Liu, D. Sedra, and K. Q. Weinberger. Deep networks with stochastic depth. CoRR, 2016.
[19] M. A. Islam, M. Rochan, N. D. Bruce, and Y. Wang. Gated feedback refinement network for dense image labeling. In CVPR, 2017.
[20] V. Jampani, R. Gadde, and P. V. Gehler. Video Propagation Networks. CVPR, 2017.
[21] S. Jégou, M. Drozdzal, D. Vazquez, A. Romero, and Y. Bengio. The one hundred layers tiramisu: Fully convolutional densefet for semantic segmentation. In CVVT, CVPRW, 2017.
[22] A. Kendall, V. Badrinarayanan, and R. Cipolla. Bayesian segnet: Model uncertainty in deep convolutional encoder-decoder architectures for scene understanding. CoRR, 2015.
[23] A. Kendall and Y. Gal. What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision? NIPS, 2017.
[24] P. Krähenbühl and V. Koltun. Efficient inference in fully connected crfs with gaussian edge potentials. In NIPS, 2011.
[25] A. Kundu, V. Vineet, and V. Koltun. Feature space optimization for semantic video segmentation. In CVPR, 2016.
[26] Q. Liao and T. A. Poggio. Bridging the gaps between residual learning, recurrent neural networks and visual cortex. CoRR, 2016.
[27] G. Lin, A. Milan, C. Shen, and I. Reid. RefineNet: Multi-path refinement networks for high-resolution semantic segmentation. CVPR, 2017.
[28] T.-Y. Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Dollár, and C. L. Zitnick. Microsoft coco: Common objects in context. In ECCV, 2014.
[29] W. Liu, A. Rabinovich, and A. C. Berg. ParseNet: Looking Wider to See Better. arXiv:1506.04579, jun 2015.
[30] J. Long, E. Shelhamer, and T. Darrell. Fully Convolutional Networks for Semantic Segmentation. CVPR, 2015.
[31] H. Noh, S. Hong, and B. Han. Learning deconvolution network for semantic segmentation. arXiv preprint arXiv:1505.04366, 2015.
[32] A. Paszke, S. Gross, S. Chintala, G. Chanan, E. Yang, Z. DeVito, Z. Lin, A. Desmaison, L. Antiga, and A. Lerer. Automatic differentiation in pytorch. In NIPS-W, 2017.
[33] T. Pohlen, A. Hermans, M. Mathias, and B. Leibe. Full-resolution residual networks for semantic segmentation in street scenes. CVPR, 2017.
[34] T. M. Quan, D. G. Hilderbrand, and W.-K. Jeong. Fusionnet: A deep fully residual convolutional neural network for image segmentation in connectomics. arXiv preprint arXiv:1612.05360, 2016.
[35] S. R. Richter, V. Vineet, S. Roth, and V. Koltun. Playing for data: Ground truth from computer games. ECCV, 2016.
[36] A. Romero, N. Ballas, S. E. Kahou, A. Chassang, C. Gatta, and Y. Bengio. Fitnets: Hints for thin deep nets. In ICLR, 2015.
[37] A. Romero, M. Drozdzal, A. Erraqabi, S. Jégou, and Y. Bengio. Image segmentation by iterative inference from conditional score estimation. arXiv preprint arXiv:1705.07450, 2017.
[38] O. Ronneberger, P. Fischer, and T. Brox. U-net: Convolutional networks for biomedical image segmentation. CoRR, 2015.
[39] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, A. C. Berg, and L. Fei-Fei. ImageNet Large Scale Visual Recognition Challenge. *International Journal of Computer Vision (IJCV)*, 115(3):211–252, 2015.
[40] K. Simonyan and A. Zisserman. Very Deep Convolutional Networks for Large-Scale Image Recognition. *ICLR*, 2015.
[41] T. Tieleman and G. Hinton. rmsprop adaptive learning. In *COURSERA: Neural Networks for Machine Learning*, 2012.
[42] A. Veit, M. Wilber, and S. Belongie. Residual Networks Behave Like Ensembles of Relatively Shallow Networks. *NIPS*, 2016.
[43] F. Visin, M. Ciccone, A. Romero, K. Kastner, K. Cho, Y. Bengio, M. Matteucci, and A. Courville. ReSeg: A Recurrent Neural Network-based Model for Semantic Segmentation. *CVPR workshop*, 2016.
[44] Z. Wu, C. Shen, and A. van den Hengel. Wider or Deeper: Revisiting the ResNet Model for Visual Recognition. *arXiv:1611.10080*, nov 2016.
[45] S. Xie, X. Huang, and Z. Tu. *Top-Down Learning for Structured Labeling with Convolutional Pseudoprior*, pages 302–317. Springer International Publishing, Cham, 2016.
[46] F. Yu and V. Koltun. Multi-Scale Context Aggregation by Dilated Convolutions. *ICLR*, 2016.
[47] F. Yu, V. Koltun, and T. Funkhouser. Dilated residual networks. *CVPR*, 2017.
[48] H. Zhao, J. Shi, X. Qi, X. Wang, and J. Jia. Pyramid scene parsing network. *CVPR*, 2017.
[49] S. Zheng, S. Jayasumana, B. Romera-Paredes, V. Vineet, Z. Su, D. Du, C. Huang, and P. Torr. Conditional random fields as recurrent neural networks. *ICCV*, 2015.
A. Supplementary Material
We present the architecture details in Table 4, agnostic to the type of transformation used in between ResNets.
The outputs of transformation blocks are reused when needed. In the case of dilations, we still maintain all transformations to keep the number of parameters roughly constant. The right column in the table indicates the number of feature channels after applying each operation.
The detailed composition of the ResNet block and the multi-grid dilation block used in our architecture is presented in Figure 6.
Additionally, we also present some additional output segmentations for FC-DRN-P-D, the max-pooling architecture finetuned with some dilated convolutions and trained with soft targets. Predictions are shown in Figure 7.
**Figure 6**: ResNet block and multi-grid dilation block used in our architecture. For multi-grid dilation block, we use \( r \) to represent dilation factor.
| Operation | Out |
|-----------|-----|
| IDB: 3 × 3 conv, max pool, 2 3 × 3 conv | 50 |
| \( R1 \) | 30 |
| \([TF_d(IDB), TF_d(R1)]\) mixing block | 80 |
| \( R2 \) | 40 |
| \([TF_d^2(IDB), TF_d^2(R1), TF_d(R2)]\) mixing block | 120 |
| \( R3 \) | 40 |
| \([TF_d^3(IDB), TF_d^3(R1), TF_d^2(R2), TF_d(R3)]\) mixing block | 160 |
| \( R4 \) | 40 |
| \([TF_d^4(IDB), TF_d^4(R1), TF_d^3(R2), TF_d^2(R3), TF_d(R4)]\) mixing block | 200 |
| \( R5 \) | 50 |
| \([TF_d^5(IDB), TF_d^5(R1), TF_d^4(R2), TF_d^3(R3), TF_d^2(R4), TF_d(R5)]\) mixing block | 200 |
| \( R6 \) | 40 |
| \([TF_d^6(IDB), TF_d^6(R1), TF_d^5(R2), TF_d^4(R3), TF_d^3(R4), TF_d^2(R5), TF_d(R6)]\) mixing block | 240 |
| \( R7 \) | 40 |
| \([TF_d^7(IDB), TF_d^7(R1), TF_d^6(R2), TF_d^5(R3), TF_d^4(R4), TF_d^3(R5), TF_d^2(R6), TF_d(R7)]\) mixing block | 280 |
| \( R8 \) | 40 |
| \([TF_d(R1), TF_u(R2), TF_u^2(R3), TF_u^3(R4), TF_u^4(R5), TF_u^3(R6), TF_u^2(R7), TF_u(R8)]\) mixing block | 320 |
| \( R9 \) | 30 |
| \([TF_u(R1), TF_u(R2), TF_u^2(R3), TF_u^3(R4), TF_u^4(R5), TF_u^3(R6), TF_u^2(R7), TF_u(R8), R9]\) mixing block | 350 |
| FUB: 2x2 repeat upsampling, 3 × 3 conv | 50 |
| Linear classifier: 1 × 1 conv | 11 |
Table 4: Architecture details. ResNets are indicated as \( R \), the initial downsampling block as \( IDB \) and the final upsampling Block as \( FUB \). We use \( TF_d^i \) and \( TF_u^i \) to denote transformation blocks in the downsampling path and in the upsampling path, respectively. The superscript \( i \) is the number of cascaded transformations applied to their input.
Figure 7: Additional qualitative results on CamVid test set. Test images are shown on the left, ground truth on the middle and FC-DRN predictions on the right. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4752,
1
],
[
4752,
10332,
2
],
[
10332,
14072,
3
],
[
14072,
19873,
4
],
[
19873,
24846,
5
],
[
24846,
28088,
6
],
[
28088,
30897,
7
],
[
30897,
32435,
8
],
[
32435,
37902,
9
],
[
37902,
39497,
10
],
[
39497,
42016,
11
],
[
42016,
42175,
12
]
]
} |
74370f84f36065b15b49f57c5c95112e205820d7 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 10,
"total-fallback-pages": 0,
"total-input-tokens": 30569,
"total-output-tokens": 9895,
"fieldofstudy": [
"Medicine"
]
} | The clinical effect of rehabilitation following arthroscopic rotator cuff repair
A meta-analysis of early versus delayed passive motion
Shuxiang Li, MD, Han Sun, MD, Xiaomin Luo, MD, Kun Wang, MD, Guofeng Wu, MD, Jian Zhou, MD, Peng Wang, MD, Xiaomin Luo, MD, Kun Wang, MD, Guofeng Wu, MD, Xiaoliang Sun, MD
Abstract
Background: The argument on the recommended rehabilitation protocol following arthroscopic rotator cuff repair remains to be resolved. So this meta-analysis was presented to evaluate the differences of clinical effects between the 2 distinct rehabilitation protocols after arthroscopic rotator cuff repair.
Methods: The PubMed, Cochrane Library, Web of Science, and EMBASE were systematically searched. Only randomized controlled trials (RCTs) published up to July 25, 2017, comparing early passive motion (EPM) versus delayed passive motion (DPM) rehabilitation protocols following arthroscopic rotator cuff repair were identified. The primary outcomes included range of motion and healing rate, while the secondary outcomes were Constant score, American Shoulder and Elbow Society (ASES) score, and Simple Shoulder Test (SST) score. The exclusion criteria contained biochemical trials, reviews, case reports, retrospective studies, without mention about passive motion exercise, no assessment of outcomes mentioned above, and no comparison of EPM and DPM rehabilitation protocols.
Results: Eight RCTs with 671 patients were enrolled in this study. The EPM resulted in improved shoulder forward flexion at short-term, mid-term, and long-term follow-ups. The EPM group was superior to the DPM group in terms of external rotation (ER) at short-term and mid-term follow-ups. However, the DPM performed better long-term ASES score. These 2 protocols were equivalent in terms of ER at long term, ASES score at mid-term, SST score, Constant score, and healing rate. After excluding 2 RCTs that examined only small- and medium-sized tears, the pooled results of healing rate decreased from 82.4% to 76.6% in the EPM and 86.9% to 85.9% in the DPM.
Conclusion: The meta-analysis suggests that the EPM protocol results in superior ROM recovery after arthroscopic rotator cuff repair but may adversely affect the shoulder function, which should be supported by further research. The healing rate at long-term follow-up is not clearly affected by the type of rehabilitation, but the EPM protocol might result in lower rates of tendon healing in the shoulder with large-sized tendon tears.
Abbreviations: ASES = American Shoulder and Elbow Society, DPM = delayed passive motion, EPM = early passive motion, ER = external rotation, RCTs = randomized controlled trials, SST = Simple Shoulder Test.
Keywords: arthroscopy rotator cuff repair, early passive motion, meta-analysis, rehabilitation
1. Introduction
With constant developments and advances in surgical instruments and technique, open techniques are slowly being replaced by arthroscopic repairs that allow faster recovery and good cosmetic results in rotator cuff repair. A partial or full-thickness tear, which can produce symptoms that interfere with the normal functioning of patients and has no response to conservative treatment, is an indication for arthroscopic repair of a rotator cuff tear.[1] However, the rate of anatomical failure after arthroscopic rotator cuff repair still remains at 20% to 90% despite the significant advances and refinements in the arthroscopic techniques.[2,3] Shoulder stiffness, which is the most common complication of rotator cuff repair, can be a source of pain, functional limitation, and frustration for patients.[4]
In recent years, controversy still exists regarding the influence of early passive motion (EPM) versus delayed passive motion (DPM) on the stiffness and healing rate after rotator cuff repair. Traditionally, the EPM protocol refers to the shoulder range of motion (ROM) that begins on day 1 postoperatively, whereas the DPM regimen requires rigorous sling immobilization within the first 4 to 6 weeks after surgery. In theory, the EPM rehabilitation prevents postoperative stiffness, fatty infiltration, and muscle atrophy, but may also decrease the possibility of tendon healing.[5,6] Most of the animal studies have shown that early
ROM exercise deteriorates tendon healing, but an artificial tendon injury may not have the usual degenerative tear patterns in human rotator cuffs.\[7,8\] Furthermore, recent studies have shown that most recurrent rotator cuff tears occur within 3 to 6 months after surgery, which further supports the DPM protocol.\[9,10\] However, delayed motion exercise may increase the risk of shoulder stiffness, and then delay the recovery of shoulder function.
As far as we know, several previous systematic reviews\[11-13\] and meta-analyses\[14-16\] have been published comparing the EPM and DPM protocols after arthroscopic rotator cuff repair. However, there was discordance in the conclusions of these published studies, and the argument on the recommended postoperative protocol remains to be resolved. In 2015 and 2017, 3 new randomized controlled trials (RCTs)\[17-19\] were published. Some important information may be obtained if these 3 studies are analyzed. Thus, it is important to conduct a new meta-analysis on these studies to make a relatively more credible and overall assessment about which rehabilitation protocol after arthroscopic rotator cuff repair is the best choice.
2. Materials and methods
2.1. Search strategy
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guidelines\[20\] and the recommendations of the Cochrane Collaboration\[21\] to conduct this meta-analysis. The detailed guidelines can be found at www.prisma-statement.org. Reviewers searched the PubMed, Cochrane Library, Web of Science, and EMBASE online databases using the key phrases “early passive motion exercise,” “delayed motion,” “rehabilitation,” “immobilization,” “early physical therapy,” “stiffness,” and “rotator cuff repair” for all English-language RCTs published up to July 25, 2017. Ethical approval was not necessary because the present meta-analysis was performed on the basis of previous published studies.
2.2. Inclusion and exclusion criteria
The studies on RCT focusing on comparing EPM and DPM rehabilitation exercise following arthroscopic rotator cuff repair were included in our meta-analysis. At least 1 of the following outcomes should have been measured: Constant score, American Shoulder and Elbow Society (ASES) score, Simple Shoulder Test (SST) score, ROM, and healing rate of rotator cuff. EPM required passive shoulder ROM exercises conducted within the first 2 weeks after arthroscopic rotator cuff repair. The exclusion criteria contained biochemical trials, reviews, case reports, retrospective studies, without mention about passive motion exercise, no assessment of outcomes mentioned above, and no comparison of EPM and DPM rehabilitation protocols.
2.3. Study selection
Two independent authors (S.X.L. and H.S.) followed the unified search strategy to screen the titles and abstracts of potentially relevant studies. Any inconsistencies between reviewers were resolved through discussion and consensus. If a consensus could not be reached, a senior author (X.L.S.) was consulted for a final decision.
2.4. Data extraction
Data were extracted from the included studies by 2 independent reviewers (S.X.L. and X.M.L.). Relevant data extracted from the RCTs included patient characteristics, technical categories of arthroscopic repair, details of rehabilitation protocols, duration of follow-up, and outcome measurements (Table 1).\[22-27\] The primary outcome measures of the study included ROM and healing rate, whereas the secondary outcomes were functional scores, including Constant score, ASES score, and SST score. Short-term follow-up was defined as within 3 months, mid-term was defined as 3 to 6 months, and long-term was defined as more than 6 months. If the data could not be extracted directly, we contacted the authors for more information. Otherwise, we extracted them from figures or calculated them with the guideline of Cochrane Handbook for Systematic Reviews of Interventions 5.1.0.\[28\]
2.5. Data analysis
The present meta-analysis was performed using the Review Manager Software (RevMan Version 5.3, The Cochrane Collaboration, Copenhagen, Denmark). Risk ratios (RRs) with a 95% confidence interval (CI) or mean difference (MD) with 95% CI were assessed for dichotomous outcomes or continuous outcomes, respectively. \( P < .05 \) was set as the level of significance. It was also considered as statistically significant if \( I^2 \) was not included in the 95% CI of RR or \( 0 \) was not included in the 95% CI of MD. The heterogeneity was assessed by using the Q test and \( I^2 \) statistic. If \( P > .1 \) and \( I^2 < 50\% \), no significant heterogeneity was noted and the fixed effect model was used. On the contrary, if \( P \leq .1 \) or \( I^2 \geq 50\% \), a random effects model was used for the heterogeneity. The source of heterogeneity was investigated using the sensitivity analysis.
2.6. Assessment of methodological quality and evidence synthesis
On the basis of Cochrane Handbook for Systematic Reviews of Interventions 5.1.0\[28\] the risk of bias of the included studies was assessed by 2 independent authors (S.X.L. and X.M.L.) with the application of the “Cochrane collaboration’s tool for assessing the risk of bias.” The publication bias and funnel plots were not reliable due to the limited number of studies. Evidence grade of outcome was evaluated in accordance with the Grading of Recommendations Assessment, Development, and Evaluation (GRADE).\[29\] Any disagreement was resolved by discussing with a third reviewer (X.L.S.).
3. Results
3.1. Search results
On the basis of the key phrases mentioned above, a total of 338 citations were identified from the following databases: 28 from PubMed, 171 from Web of science, 130 from EMBASE, and 9 from Cochrane library. A screen of the 10 RCTs\[17-19,22-27,30\] was conducted for eligibility and the full text read after excluding duplicate, irrelevant, and nonrandomized clinical studies. One trial\[30\] comparing the clinical outcomes of passive self-assisted ROM exercise with those associated with the use of continuous passive motion in patients after arthroscopic rotator cuff repair was excluded. Another RCT,\[27\] which compared the differences of slow and accelerated rehabilitation protocols
### Table 1: Study characteristics.
| Arthroscopic technique | Protocol in EPM group | Protocol in DPM group | Relevant outcomes |
|------------------------|-----------------------|-----------------------|-------------------|
| | | | |
| Arthroscopic | | | |
| Follow-up, mo | | | |
| Tear size, cm | | | |
| Age (years) | | | |
| Age (EPM/DPM) | | | |
| Arthroscopic technique | | | |
| Arthroscopic | | | |
| Follow-up, mo | | | |
| Tear size, cm | | | |
| Age (years) | | | |
| Age (EPM/DPM) | | | |
After arthroscopic rotator cuff repair, was also excluded because it only documented the Disabilities of the Arm, Shoulder, and Hand score in patients. Finally, of the 338 studies, only 8 RCTs were included in our meta-analysis (Fig. 1).
### 3.2. Quality assessment of included RCTs
The detailed information of the characteristics of included studies can be seen in Table 1. A standardized assessment of the risk of bias in the 8 RCTs is summarized in Fig. 2A. There was no blinding of the participants and personnel in all 8 studies. In the study of Arndt et al and Duzgun et al, the details of randomization and outcome assessments were not described, and there was no evidence of allocation concealment. On the contrary, Kim et al demonstrated the use of a randomization technique; however, the study failed to report allocation concealment and provided incomplete outcome data and could not blind outcome assessors to the rehabilitation protocol. All of these 3 studies represented a high risk of bias in terms of methodologic quality. Lee et al did not describe their randomization method; although allocation concealment was adequate, treating surgeons who performed outcome assessments were not blinded. A reasonable methodology was used in the study of Cuff et al, but this RCT had incomplete data. The study of Roo mentioned appropriate randomization measures and provided complete statistical data; however, the allocations were not concealed. Finally, the studies by Mazzocca et al and Keener et al used appropriate randomization, detailed allocation concealment, and blinded outcome assessments, representing a methodologic quality with a low risk of bias. Each risk of the bias item was expressed in terms of the percentage across all the included studies, which indicated the proportion of risk levels for each item bias (Fig. 2B).
### 3.3. The primary outcome measurements
#### 3.3.1. Range of motion
ROM data could be extracted from all of the included RCTs. It was evaluated in terms of forward flexion (FF) and external rotation (ER) at short-term, mid-term, and long-term follow-up.
The FF and ER on short-term follow-up were shown in 6 studies. The result of the meta-analysis revealed a significant difference in the FF at short-term follow-up between the EPM and DPM protocols (MD, 10.31; 95% CI, 5.02–15.61; P = .0001; I² = 63%, a random effect model was used) (Fig. 3A). The sensitivity analysis presented that the study of Arndt et al contributed to the heterogeneity, and a statistically significant difference still existed after excluding it. In addition, data pooled from these studies indicated a significant difference in the ER at short-term follow-up between the 2 groups (MD, 8.28; 95% CI, 3.52–13.04; P = .0007; I² = 67%) (Fig. 3B). Sensitivity analysis reported that the study of Mazzocca et al was the main source of the heterogeneity, and a statistically significant difference was also found when it was excluded.
All 8 studies reported the outcomes of FF and ER at mid-term follow-up in 671 patients. On analysis of the pooled data from the studies, the EPM group showed a significantly better FF at mid-term follow-up than did the DPM group (MD, 3.01; 95% CI, 0.31–5.72; P = .03; I² = 56%) (Fig. 4A). The study conducted by Cuff et al caused the heterogeneity, and the EPM group was also superior to the DPM group in terms of FF after excluding this study. Similarly, the summarized results showed
that ER was better in the EPM group at mid-term follow-up (MD, 2.00; 95% CI, 0.94–3.05; \( P = .0002; I^2 = 48\%\)) (Fig. 4B).
Six studies\(^{18,22–26}\) with 503 patients reported the FF and ER during long-term follow-up. According to our analysis, long-term FF is better with EPM rehabilitation than with DPM rehabilitation (MD, 1.24; 95% CI, 0.25–2.23; \( P = .01; I^2 = 0\%\)) (Fig. 5A). However, no statistically significant was noted in the differences in the long-term ER between the 2 groups (MD, 2.24; 95% CI, –2.72 to 7.19; \( P = .38; I^2 = 57\%\)) (Fig. 5B). The sensitivity analysis revealed no statistically significant difference between the 2 groups on eliminating study of Arndt et al.\(^{22}\)
### 3.3.2. Tendon healing
Tendon healing was compared in 6 studies\(^{18,22–26}\) at long-term postoperative follow-up, including a total of 493 patients (EPM, \( n = 256\); DPM, \( n = 237\)). Keener et al.\(^{24}\) and Cuff et al.\(^{23}\) assessed the anatomic outcome using the ultrasound; CT arthrography was used by Arndt et al.\(^ {22}\) and Kim et al.\(^ {25}\); magnetic resonance imaging (MRI) was used in the remaining 2 studies.\(^ {18,26}\) The summarized results showed that tendon healing was observed in 211 out of 256 patients (82.4%) in the EPM group, which revealed that the 2 rehabilitation groups were comparable in tendon healing (RR, 0.95; 95% CI, 0.88–1.02; \( P = .16\)) (Fig. 6). Then, we excluded the study of Keener et al.\(^ {24}\) and Kim et al.\(^ {25}\) that examined only small and medium-sized rotator cuff tears to perform a sensitivity analysis. The pooled results changed to 105 out of 137 (76.6%) in the EPM group and 116 out of 135 (85.9%) in the DPM group, which demonstrated that the DPM rehabilitation protocol performed better in tendon healing. However, there was no statistically significant difference in healing rate of rotator cuff between the 2 groups (RR, 0.90; 95% CI, 0.80–1.01; \( P = .06\)).
### 3.4. The secondary outcome measurements
#### 3.4.1. Medium functional scores
The Constant score at mid-term follow-up was measured in 4 studies\(^ {17,18,24,25}\) consisting of 407 patients. One study\(^ {25}\) did not present standard deviation, so we imputed it depending on the \( P \) value. Data pooled from these studies showed no significant difference between both groups (MD, 0.87; 95% CI, –1.97 to 3.71; \( P = .55; I^2 = 0\%\)) (Fig. 7A). Similarly, the EPM and DPM groups revealed little difference in the ASES scores at mid-term follow-up in 3 studies (MD, 0.19; 95% CI, –6.66 to 7.03; \( P = .96; I^2 = 55\%\)) (Fig. 7B).\(^ {18,24,25}\) The study by Mazzocca et al.\(^ {18}\) contributed to the heterogeneity, and no difference was found when it was rejected. Four studies\(^ {17,18,24,25}\) reported the mid-term SST score. The pooled result revealed that the 2 groups were comparable in terms of the SST score (MD, 0.47; 95% CI, –0.08 to 1.02; \( P = .09; I^2 = 17\%\)) (Fig. 7C).
#### 3.4.2. Long-term functional scores
The long-term Constant score was reported in 4 studies\(^ {18,22,24,25}\) with 371 patients. Meta-analysis presented a similar long-term Constant score between the 2 groups (MD, 1.90; 95% CI, –1.62 to 5.41;
The sensitivity analysis showed that the study of Arndt et al.\(^{[22]}\) was the main reason for the heterogeneity; no significant difference was found after excluding it. However, the meta-analysis result for the long-term ASES score indicated that the DPM group had a significantly higher score than the EPM group (MD, −1.66; 95% CI, −2.76 to −0.55; \(P = .003\); \(I^2 = 25\%\)) (Fig. 8A). The difference in long-term SST score was not statistically significant according to our analysis (MD, 0.07; 95% CI, −0.26 to 0.40; \(P = .68\); \(I^2 = 0\)) (Fig. 8C).
3.5. Quality of evidence
The GRADE system was used to assess the quality of evidence across the various outcomes in our study. In our final assessments, none of the outcomes showed high quality of evidence, while short-term ER, long-term FF, and long-term ASES score revealed moderate quality. The evidence for short-term or mid-term FF, mid-term ER, mid-term Constant score, mid-term or long-term SST score, and healing rate was low. Evaluation of the results of long-term ER, long-term Constant score, and mid-term ASES score revealed that the evidence was very low. The details of the results are summarized in Table 2.
4. Discussion
Currently, arthroscopic repair has been increasingly used in the treatment of rotator cuff tears. However, there has been debate on the timing of shoulder passive ROM postoperatively, with proponents of the DPM rehabilitation protocol submitting the potential for increased rate of tendon healing by minimizing micromotion and improved shoulder functional outcomes.\(^{[31]}\) Advocates of the EPM rehabilitation protocol suggest that it may increase shoulder ROM, which could ultimately decrease shoulder stiffness and muscle atrophy. To our knowledge, several systematic reviews\(^{[11]}\) and meta-analyses\(^ { [14]} \) have been published to compare the effect of the EPM and DPM rehabilitation protocols after arthroscopic rotator cuff repair. However, the number of included RCTs among the studies was small and discordance existed in the conclusions of these studies. With this, the argument on the recommended postoperative protocol remains to be resolved. Therefore, we performed a meta-analysis of RCTs to compare the EPM and DPM rehabilitation protocol in terms of ROM, healing rate, and shoulder function scores and to provide an evidence-based recommendation of the best rehabilitation after arthroscopic rotator cuff repair.
The present meta-analysis indicated that these 2 rehabilitation protocols were equivalent in terms of long-term ER, mid-term ASES score, SST score, Constant score, and healing rate. However, there was a significant difference between the 2 protocols for FF, short-term or mid-term ER, and long-term ASES score according to an accurate analysis. The sensitivity analysis indicated that the patients with large-sized tears preoperatively who underwent the EPM rehabilitation had slow tendon healing, although there was no significant difference compared with DPM rehabilitation (RR 0.90; 95% CI, 0.80–1.01; \( P = 0.06 \)).
EPM rehabilitation was often opposed by many biochemical trials. Peltz et al\(^{[32]}\) suggested that an EPM performed postoperatively could increase scar formation and extracellular tissue in the subacromial space in a rat model and lead to...
decreased ROM and increased joint stiffness. On the basis of largest number of available RCTs, our pooled analysis provided the most stable and reasonable evidence that the EPM protocol is beneficial in terms of FF and ER. However, unlike that for FF, the advantage of ER could not be extended to long-term follow-up. It was speculated that the inconsistent results between FF and ER were generated by the initial ROM limit given by the shoulder motion planes. To avoid excessive loading on the sutured supraspinatus tendons, the ER angle was restricted to 30° and the FF angle was allowed to be more than 90° in the EPM protocol of the most trials. The ROM difference between the 2 protocols in 3 difference periods showed a downward trend at 1 year postoperatively regardless of the FF and ER. In addition, a previous retrospective cohort study suggested that the DPM protocol would not lead to long-term stiffness.\[33\] Evaluation of the permanent ROM defects in the DPM protocol was not performed because majority of the data in the long-term follow-up were extracted at 1 year postoperatively. Thus, further RCTs need to assess and compare the outcomes of the 2 rehabilitation protocols at longer-term follow-up.
As far as we know, the possibility that EPM reduced the probability of tendon healing has been the principal focus of the debate. On the basis of the currently available evidence, the present meta-analysis showed that the EPM and DPM rehabilitation protocols led to statistically equivalent tendon healing at long-term follow-up (P = 0.16). Consequently, sensitivity analysis also revealed no statistically significant difference in tendon healing of the rotator cuff between the 2 protocols (P = .06) after excluding 2 studies\[24,25\] that only enrolled patients with small and medium-sized rotator cuff tears. However, the factors associated with rotator cuff healing include tear size, surgical techniques, patient’s age, tendon quality and number, and fatty muscle degeneration and atrophy. The diversity of factors in our included trials tended to mitigate the statistical significance of the tendon healing. Therefore, after excluding the study of Keener et al\[24\] and Kim et al.\[25\] which examined only small- and medium-sized tears, the pooled results of healing rate decreased from 82.4% to 76.6% in the EPM group and 86.9% to 85.9% in the DPM group. Our discovery might indicate that the large-sized tears (3–5 cm) might benefit from delayed motion. Thus, it was warranted that the comparison of outcomes in the 2 protocols be focused on larger tear size in future research.
A previous study\[24\] indicated that most functional scores were at the plateau after 6 to 12 months postoperatively. Our analysis revealed that there was no statistically significant difference in shoulder function outcomes between the 2 protocols, with the exception of the long-term ASES score, which was higher in the DPM protocol. The ASES score includes pain assessment, instability scales, and daily-life questionnaires.\[34\] It has been validated and widely used for the evaluation of shoulder function after arthroscopic rotator cuff repair. According to our comprehensive and detailed analysis, the quality of evidence of the long-term ASES score was moderate, which represented a relatively credible level. In contrast, the quality of evidence for the
Figure 8. (A) A forest plot diagram showing Constant score at long term after surgery. (B) A forest plot diagram showing ASES score at long term after surgery. (C) A forest plot diagram showing SST score at long term after surgery.
Table 2
Quality of the evidence.
| Outcomes | No. of participants (studies) follow-up | Quality of the evidence (GRADE) | Anticipated absolute effects |
|---------------------------|-----------------------------------------|---------------------------------|------------------------------|
| Short-term FF | 470 (6 studies) 3 mo | LOW $^{1,3,4,5}$ due to risk of bias, large effect, inconsistency, imprecision | MD 10.31 higher (5.02–15.61 higher) |
| Mid-term FF | 671 (8 studies) 4–6 mo | LOW $^{1,3,4,5}$ due to risk of bias, large effect, inconsistency, imprecision | MD 3.01 higher (0.31–5.72 higher) |
| Long-term FF | 503 (6 studies) 12–24 mo | MODERATE $^{1,3,4,5}$ due to risk of bias, large effect, imprecision | MD 1.24 higher (0.25–2.23 higher) |
| Short-term ER | 415 (5 studies) 3 mo | MODERATE $^{1,3,4,5}$ due to risk of bias, large effect, imprecision | MD 10.22 higher (6.46–13.98 higher) |
| Mid-term ER | 671 (8 studies) 4–6 mo | LOW $^{1,3,4,5}$ due to risk of bias, large effect, inconsistency, imprecision | MD 2.00 higher (0.94–3.05 higher) |
| Long-term ER | 503 (6 studies) 12–24 mo | VERY LOW $^{1,3,4,5}$ due to risk of bias, inconsistency, imprecision | MD 2.24 higher (2.72 lower to 7.19 higher) |
| Mid-term Constant score | 407 (6 studies) 4–6 mo | LOW $^{1,3,4}$ due to risk of bias, imprecision | MD 0.67 higher (1.97 lower to 3.71 higher) |
| Long-term Constant score | 371 (4 studies) 12–24 mo | VERY LOW $^{1,3,4,5}$ due to risk of bias, inconsistency, imprecision | MD 1.90 higher (1.62 lower to 5.41 higher) |
| Mid-term ASES score | 277 (3 studies) 6 mo | VERY LOW $^{1,3,4,5}$ due to risk of bias, inconsistency, imprecision | MD 0.19 higher (6.66 lower to 7.03 higher) |
| Long-term ASES score | 347 (4 studies) 12–24 mo | MODERATE $^{1,3,4,5}$ due to risk of bias, large effect, imprecision | MD 1.66 lower (2.76 to 0.55 lower) |
| Mid-term SST score | 407 (4 studies) 4–6 mo | LOW $^{1,3,4,5}$ due to risk of bias, imprecision | MD 0.47 higher (0.08 lower to 1.02 higher) |
| Long-term SST score | 347 (4 studies) 12–24 mo | LOW $^{1,3,4,5}$ due to risk of bias, imprecision | MD 0.07 higher (0.26 lower to 0.40 higher) |
| Healing rate | 493 (6 studies) 6–24 mo | LOW $^{1,3,4,5}$ due to risk of bias, imprecision | RR 0.95 (0.88–1.02) |
ER = external rotation, FF = forward flexion, MD = mean difference, RR = risk ratio.
1 No details of randomization.
2 No concealment.
3 Effect is really stable.
4 Result is inconsistent.
5 Indirect data.
6 Inconsistent follow-up time point.
7 Limited sample size.
Constant and SST scores was low or very low. Therefore, our findings implied that the EPM protocol might adversely affect the shoulder function compared with the delay protocol.
There are some limitations in the current systematic review and meta-analysis: First, although the present study included the largest number of RCTs, the number of trials was still relatively small, and more large-scale prospective studies were needed to produce more convincing conclusions. Second, there was no high quality of evidence in all outcomes of our study. Most of the included studies provided only Level II data due to incomplete or inaccurate protocol reports or due to small sample size. The most common defect in these studies was that outcome assessors were not blinded to rehabilitation protocol. Inconsistencies in the observation time for some of the outcomes could have a negative influence on the reliability of results. Third, the standard deviation is not provided in some included studies, so we had to calculate them with the guideline of the Cochrane Handbook for Systematic Reviews of Interventions 5.1.0. Fourth, publication bias is unavoidable because only English trials were included.
5. Conclusion
On the basis of the largest number of available RCTs, the meta-analysis suggests that the EPM protocol results in superior ROM to rehabilitation protocol. Inconsistencies in the observation time in some included studies was that outcome assessors were not blinded to rehabilitation protocol. Inconsistencies in the observation time for some of the outcomes could have a negative influence on the reliability of results. Third, the standard deviation is not provided in some included studies, so we had to calculate them with the guideline of the Cochrane Handbook for Systematic Reviews of Interventions 5.1.0. Fourth, publication bias is unavoidable because only English trials were included.
References
[1] Gartsman GM. Arthroscopic management of rotator cuff disease. J Am Acad Orthop Surg 1998;6:259-66.
[2] Boileau P, Brassart N, Watkinson DJ, et al. Arthroscopic repair of full-thickness tears of the supraspinatus: does the tendon really heal? J Bone Joint Surg Am 2005;87:1229-30.
[3] Galatz LM, Ball CM, Teefey SA, et al. The outcome and repair integrity of completely arthroscopically repaired large and massive rotator cuff tears. J Bone Joint Surg Am 2004;86:219-24.
[4] Namdari S, Green A. Range of motion limitation after rotator cuff repair. J Shoulder Elbow Surg 2010;19:290-6.
[5] Huberty DP, Schoolfield JD, Brady PC, et al. Incidence and treatment of postoperative stiffness following arthroscopic rotator cuff repair. Arthroscopy 2009;25:880-90.
[6] Papalia R, Franceschi F, Vasta S, et al. Shoulder stiffness and rotator cuff repair. Br Med Bull 2012;104:163-74.
[7] Peltz CD, Sarver JJ, Dourte LM, et al. Exercise following a short immobilization period is detrimental to tendon properties and joint mechanics in a rat rotator cuff injury model. J Orthop Res 2010;28:841-5.
[8] Zhang S, Li H, Tao H, et al. Delayed early passive motion is harmless to shoulder rotator cuff healing in a rabbit model. Am J Sport Med 2013;41:1885-92.
[9] Kluger R, Bock P, Mittbock M, et al. Long-term survivorship of rotator cuff repairs using ultrasound and magnetic resonance imaging analysis. Am J Sport Med 2011;39:2071-81.
[10] Miller BS, Downie BK, Kohen RB, et al. When do rotator cuff repairs fail? Serial ultrasound examination after arthroscopic repair of large and massive rotator cuff tears. Am J Sport Med 2011;39:2064-70.
[11] Saltzman BM, Zuke WA, Go B, et al. Does early motion lead to a higher failure rate or better outcomes after arthroscopic rotator cuff repair? A systematic review of overlapping meta-analyses. J Shoulder Elbow Surg 2017;26:1681-91.
[12] Gallagher BP, Bishop ME, Tjoumakaris FP, et al. Early versus delayed rehabilitation following arthroscopic rotator cuff repair. Physician Sportmed 2015;43:178-87.
[13] Shen C, Tang ZH, Hu JZ, et al. Does immobilization after arthroscopic rotator cuff repair increase tendon healing? A systematic review and meta-analysis. Orthop Traumatol-Sur 2014;14:1279-83.
[14] Chan K, MacDermid JC, Hoppé DJ, et al. Delayed versus early motion after arthroscopic rotator cuff repair: a meta-analysis. J Shoulder Elbow Surg 2014;23:1631-9.
[15] Chang KV, Hung CY, Han DS, et al. Early versus delayed passive range of motion exercise for arthroscopic rotator cuff repair. Am J Sport Med 2013;41:1265-73.
[16] Riboh JC, Garrigues GE. Early passive motion versus immobilization after arthroscopic rotator cuff repair. Arthroscopy 2014;30:997-1005.
[17] Rood PD. Passive mobilization after arthroscopic rotator cuff repair is not detrimental in the early postoperative period. Acta Orthop 2015;86:1485-92.
[18] Mazzocca AD, Arciero RA, Shea KP, et al. The effect of early range of motion on quality of life, clinical outcome, and repair integrity after arthroscopic rotator cuff repair. Arthroscopy 2017;33:1138-48.
[19] Düzgün I, Baltaci G, Turgut E, et al. Effects of slow and accelerated rehabilitation protocols on range of motion after arthroscopic rotator cuff repair. Acta Orthop Traumatol Turc 2014;48:642-8.
[20] Moher D, Liberati A, Tetzlaff J, et al. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. BMJ 2009;339:b2535.
[21] Bero L, Rennie D. The Cochrane Collaboration. Preparing, maintaining, and disseminating systematic reviews of the effects of health care, JAMA 1995;274:1935-8.
[22] Arndt J, Clavert P, Mielcarek P, et al. Immediate passive motion versus immobilization after endoscopic supraspinatus tendon repair: a prospective randomized study. Orthop Traumatol-Sur 2012;98:S313-8.
[23] Cuff DJ, Papello DR. Prospective randomized study of arthroscopic rotator cuff repair using an early versus delayed postoperative physical therapy protocol. J Shoulder Elbow Surg 2012;21:1430-5.
[24] Koemer JD, Galatz LM, Stoobbs-Cooki G, et al. Rehabilitation following arthroscopic rotator cuff repair: a prospective randomized trial of immobilization compared with early motion. J Bone Joint Surg Am 2014;96:11-9.
[25] Kim YS, Chung SW, Kim JY, et al. Is early passive motion exercise necessary after arthroscopic rotator cuff repair? Am J Sport Med 2012;40:815-21.
[26] Lee BG, Cho NS, Rhee YG. Effect of two rehabilitation protocols on range of motion and healing rates after arthroscopic rotator cuff repair: aggressive versus limited early passive exercises. Arthroscopy 2012;28:34-42.
[27] Düzgün I, Baltaci G, Atay O. Comparison of slow and accelerated rehabilitation protocol after arthroscopic rotator cuff repair: pain and functional activity. Acta Orthop Traumatol Turc 2011;45:23-33.
[28] Higgins J, Green S. Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 [updated March 2011]. The Cochrane Collaboration 2011. Available at: http://handbook-5.1.cochrane.org. Accessed October 3, 2017.
[29] Atkins D, Briss PA, Eccles M, et al. Grading quality of evidence and strength of recommendations. BMJ 2004;328:1490.
[30] Garofalo R, Conti M, Notarnicola A, et al. Effects of one-month continuous passive motion after arthroscopic rotator cuff repair: results at 1-year follow-up of a prospective randomized study. Musculoskelet Surg 2011;94:579-83.
[31] van der Meijden OA, Westgard P, Chandler Z, et al. Rehabilitation after arthroscopic rotator cuff repair: current concepts review and evidence-based guidelines. Int J Sports Phys Ther 2012;7:197-218.
[32] Peltz CD, Dourte LM, Kuntz AF, et al. The effect of postoperative passive motion on rotator cuff healing in a rat model. J Bone Joint Surg Am 2009;91:2421-9.
[33] Parsons BO, Gruson KI, Chen DD, et al. Does slower rehabilitation after arthroscopic rotator cuff repair lead to long-term stiffness? J Shoulder Elbow Surg 2010;19:1034-9.
[34] Richards RR, An KN, Ridgani LU, et al. A standardized method for the assessment of shoulder function. J Shoulder Elbow Surg 1994;3:347-52. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4265,
1
],
[
4265,
10559,
2
],
[
10559,
15374,
3
],
[
15374,
18589,
4
],
[
18589,
19152,
5
],
[
19152,
19776,
6
],
[
19776,
21900,
7
],
[
21900,
25276,
8
],
[
25276,
28411,
9
],
[
28411,
36405,
10
]
]
} |
146ec3309fa483fa0d8b8227db50cc5c5523b9ed | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 16,
"total-fallback-pages": 0,
"total-input-tokens": 57027,
"total-output-tokens": 20383,
"fieldofstudy": [
"Biology",
"Medicine"
]
} | A molecular signature for delayed graft function
Dagmara McGuinness1 | Suhaid Mohammed1 | Laura Monaghan1 | Paul A. Wilson2 | David B. Kingsmore3 | Oliver Shapter1,3 | Karen S. Stevenson3 | Shana M. Coley4 | Luke Devey5 | Robert B. Kirkpatrick6 | Paul G. Shiels1
1Wolfson Wohl Translational Research Centre, Institute of Cancer Sciences, College of Medical, Veterinary & Life Sciences, University of Glasgow, Glasgow, UK
2Computational Biology, GlaxoSmithKline Medicines Research Centre, Stevenage, UK
3Renal Transplant Unit, NHS Greater Glasgow and Clyde, South Glasgow University Hospital, Glasgow, UK
4Research Institute of Infection Immunity and Inflammation, College of Medical, Veterinary & Life Sciences, University of Glasgow, Glasgow, UK
5Metabolic Pathways Cardio Therapy Area Unit, GlaxoSmithKline, King of Prussia, Pennsylvania
6The Pipeline Futures Group, GlaxoSmithKline, Collegeville, Pennsylvania
Correspondence
Paul G. Shiels, Wolfson Wohl Translational Research Centre, Institute of Cancer Sciences, University of Glasgow, Garscube Estate, Switchback Road, Glasgow G61 1QH, UK.
Email: [email protected]
Present address
Dagmara McGuinness, Wellcome Centre for Molecular Parasitology, Institute of Infection, Immunity and Inflammation, College of Medical, Veterinary & Life Sciences, University of Glasgow, Glasgow, UK.
Abstract
Chronic kidney disease and associated comorbidities (diabetes, cardiovascular diseases) manifest with an accelerated ageing phenotype, leading ultimately to organ failure and renal replacement therapy. This process can be modulated by epigenetic and environmental factors which promote loss of physiological function and resilience to stress earlier, linking biological age with adverse outcomes post-transplantation including delayed graft function (DGF). The molecular features underpinning this have yet to be fully elucidated. We have determined a molecular signature for loss of resilience and impaired physiological function, via a synchronous genome, transcriptome and proteome snapshot, using human renal allografts as a source of healthy tissue as an in vivo model of ageing in humans. This comprises 42 specific transcripts, related through IFN signalling, which in allografts displaying clinically impaired physiological function (DGF) exhibited a greater magnitude of change in transcriptional amplitude and elevated expression of noncoding RNAs and pseudogenes, consistent with increased allostatic load. This was accompanied by increased DNA methylation within the promoter and intragenic regions of the DGF panel in preperfusion allografts with immediate graft function. Pathway analysis indicated that an inability to sufficiently resolve inflammatory responses was enabled by decreased resilience to stress and resulted in impaired physiological function in biologically older allografts. Cross-comparison with publically available data sets for renal pathologies identified significant transcriptional commonality for over 20 DGF transcripts. Our data are clinically relevant and important, as they provide a clear molecular signature for the burden of “wear and tear” within the kidney and thus age-related physiological capability and resilience.
INTRODUCTION
The changing demographics of age in human society is anticipated to result in a major burden of age-related morbidities, as improvements in health span have failed to match the increase in average global lifespan. Notably, deaths due to chronic kidney disease have increased globally, despite significant decline in other aetiologies (Christopher & Murray, 2015). This is pertinent to researchers investigating ageing, as accelerated cellular and physiological ageing are underlying components of renal dysfunction (Kooman, Kotanko, Schols, Schild, & Stenvinkel, 2014; McGlynn et al., 2009; Schnitt, Susnik, & Melk, 2015), where systemic differences are layered on top of the dysregulated ageing process and patients show a higher incidence of mortality in comparison to healthy chronologically age-matched individuals. As the prevalence of CKD parallels an increased prevalence in type 2 diabetes, obesity and a sedentary lifestyle (Stengel, Tarver-Carr, Powe, Eberhardt, & Brancati, 2003), an allostatic outcome reflecting the “burden of life style” may be present next to the renal dysfunction.
Allostatic load can be defined as a composite indicator of accumulated biological stress over the life course, which predisposes to morbidity in the face of chronic or repeated stress exposure (Rubin, 2016). It is reflective of the biological age of a tissue organ or organism, as it directly impacts on age-related physiological function.
We have previously developed the use of renal allografts as an in vivo model to study healthy tissue ageing in humans, whose physiological function can be tracked longitudinally to demonstrate that allograft biological age is more important than chronological age in prognostication of post-transplant allograft performance (Gingell-Littlejohn et al., 2013; McGuinness et al., 2016).
One testable prediction following this demonstration is that organs with increased biological age should reflect the cumulative burden of “wear and tear” and thus be less resilient to transplant-related stresses and display reduced physiological function as a consequence. Assessing the related changes in molecular biology in these renal allografts is not straightforward (Shiels, McGuinness, Eriksson, Kooman, & Stenvinkel, 2017). To do so, we have used an analysis of a notable clinically relevant allograft phenotype displaying impaired physiological function, termed delayed graft function (DGF), to determine whether organs showing DGF are less resilient than those undergoing immediate graft function (IGF). Additionally, we have determined whether features associated with lack of physiological resilience were reflected in the pretransplant transcriptomes of respective organs.
A higher incidence of DGF has been associated with the use of allografts from older extended criteria donors (ECD, age >60, or >50 with two of the following: a history of high blood pressure, a creatinine ≥1.5, or death resulting from a stroke), donation after cardiac death donors (DCD) and increased allograft biological age (Mallon, Summers, Bradley, & Pettigrew, 2015; McGuinness et al., 2016; Menke, Sollinger, Schamberger, Heemann, & Lutz, 2014; Mundt, Yard, Kramer, Benck, & Schnulle, 2015; Schroppel & Legendre, 2014). The extent to which donor and recipient-related characteristics influence the magnitude of IRI and/or DGF occurrence, beyond accepted clinical risk factors for DGF, remains to be proven (Menke et al., 2014; Mundt et al., 2015; Schroppel & Legendre, 2014), particularly in the context of allograft repair, or regeneration pathways, activated in response to IRI. Increased demand for organ donation, coupled with increasing chronological age and associated comorbidities in the donor population, has necessitated the use of organs that have been previously deemed as marginal for clinical use (Morrissey & Monaco, 2014; Nagaraja et al., 2015).
DGF has also proven refractory to modelling both in vitro and in preclinical model organism studies. This study aimed to identify a human-specific molecular signature associated with DGF and to enable adjustment for the effects of IRI-related molecular changes, in the absence of model systems for analysis of DGF mechanisms, as well as providing direct insight into its manifestation. Additionally, this strategy was designed, to enable the validation of any DGF-associated signature, by comparison with existing publically available renal data sets, in order to elucidate whether there were common underpinning molecular processes in their manifestation and whether these reflected the burden of “wear and tear” in the kidney. We selected a very closely matched clinical cohort based on age, gender, length of ischaemic time and low HLA mismatch. These were divided into two groups based on recovery of organ function after transplantation, with emphasis on extreme functional differences, namely either DGF or IGF (Table 1). An extreme DGF phenotype was defined as (a) the need for dialysis within 7 days of
| Variable | DGF (n = 11) | IGF (n = 12) |
|----------------------------------------------|------------------------------------------------------------------------------|------------------------------------------------------------------------------|
| Donor gender (males/females) | 6/5 | 5/7 |
| Donor age (years) | 55 (40–74) | 53.2 (40–77) |
| Donor serum creatinine at retrieval (µmol/L) | 73.25 (44–125) | 78.64 (41–125) |
| Donor type | | |
| DBD/DCD | 7/4 | 9/3 |
| ECD | 6 | 4 |
| DBD-ECD | 4 | 3 |
| DCD-ECD | 2 | 1 |
| Cause of death | | |
| Intracranial haemorrhage | 7 | 8 |
| Hypoxic brain injury | 0 | 2 |
| Trauma | 0 | 2 |
| Cardiac arrest | 1 | 0 |
| Intracranial thrombus | 1 | 0 |
| Respiratory failure | 1 | 0 |
| Meningitis | 1 | 0 |
| Recipient gender (males/females) | 8/3 | 8/4 |
| Recipient age (years) | 57.9 (42–72) | 51.1 (35–70) |
| Previous transplantation | 0 | 0 |
| HLA Mismatch | | |
| HLA-A (0/1/2) | 4/5/2 | 4/6/1 |
| HLA-B (0/1/2) | 2/9/0 | 3/7/1 |
| HLA-DR (0/1/2) | 4/7/0 | 3/7/0 |
| Cold ischaemic time (hr) | 12.1 (9–17) | 11.1 (6–20) |
| Warm ischaemic time (min) | 29.3 (21–40) | 31.3 (22–40) |
| T1/2 | from 4 to 21 days | less than 3 days |
| Serum creatinine level at 6 months (µmol/L) | 126.7 (89–158) | 104.4 (84–166) |
**Combined patient cohort (N = 55)**
| Variable | Mean (Min-Max)/Proportion | Standard deviation (if applicable) |
|----------------------------------------------|---------------------------|-----------------------------------|
| Donor gender (males/females) | 33/22 | |
| Donor age (years) | 50.5 (11–77) | 16.4 |
| Donor serum creatinine at retrieval (µmol/L) | 87.1 (41–195) | 35.7 |
| Ethnicity (Caucasian: Asian) | 54:1 | |
| Donor type | | |
| DBD/DCD | 37/11 | |
| ECD | 34 | |
| DBD-ECD | 16 | |
| DCD-ECD | 5 | |
| Cause of death | | |
| Intracranial haemorrhage | 35 | |
| Hypoxic brain injury | 10 | |
| Trauma | 7 | |
| Cardiac arrest | 1 | |
| Intracranial thrombus | 1 | |
| Respiratory failure | 1 | |
(Continues)
transplantation, with the exception of hyperkalaemia on the first postoperative day, and (b) the failure of serum creatinine to reduce by 50% within the first week, which is indicative of poor recovery of renal function (Figure 1a; Aitken et al., 2015). IGF was defined as reduction in serum creatinine by 50% in less than 3 days post-transplantation. Additionally, we undertook a retrospective analysis of paired allograft biopsies from this cohort, obtained at two time points: preimplantation during the preparatory phase (preperfusion) and after allograft reperfusion when circulation was restored (post-perfusion).
### TABLE 1 (Continued)
| Combined patient cohort (N = 55) | Mean (Min-Max)/Proportion | Standard deviation (if applicable) |
|----------------------------------|---------------------------|-----------------------------------|
| Meningitis | 0 | |
| Donor hypotension (N) | 39 | |
| Donor hypertension (N) | 23 | |
| Recipient gender (males/females) | 32/23 | |
| Recipient age (years) | 54.2 (27–73) | 10.1 |
| Aetiology of Renal Failure | | |
| IgA nephropathy | 9 | |
| Glomerulonephritis | 6 | |
| APKD | 12 | |
| Reflux/obstructive uropathy | 7 | |
| Hypertension | 4 | |
| Diabetes mellitus | 2 | |
| Unknown | 8 | |
| Goodpasture’s syndrome | 1 | |
| Anatomical anomalies | 2 | |
| Drug related | 1 | |
| Stone disease | 1 | |
| HUS | 1 | |
| FSGS | 1 | |
| Previous Transplant | | |
| 0 | 45 | |
| 1 | 7 | |
| 2 | 2 | |
| HLA Mismatch | | |
| HLA-A (0/1/2) | 9/18/10 | |
| HLA-B (0/1/2) | 10/23/5 | |
| HLA-DR (0/1/2) | 24/10/0 | |
| Cold ischaemic time (hr) | 12.8 (6–21) | 3.9 |
| Immunosuppression | | |
| Induction | | |
| Basiliximab/campath/ATG | 42/6/6 | |
| Maintenance | | |
| Tacrolimus/Sirol/Cyc | 54/0/1 | |
| Prednisolone | 51 | |
| Mycophenolatefetil | 55 | |
| BPAR | 1 | |
| DGF | 19 | |
| Serum creatinine level at 6 months (μmol/L) | 131 (61–523) | 74.4 |
| MDRD4 at 6 months (ml min⁻¹ 1.73 m⁻²) | 57.4 (9–102) | 22.3 |
| Serum creatinine level at 12 months (μmol/L) | 137.6 (67–377) | 62.8 |
| MDRD4 at 12 months (ml min⁻¹ 1.73 m⁻²) | 51.1 (10–89) | 20.8 |
Note. Continuous variables are expressed as mean with standard deviation, whereas categorical variables are expressed as proportions.
We have used this approach to test the hypothesis that DGF is a manifestation of organ “wear and tear” (i.e. its allostatic load as a function of its biological age) and that the impaired physiological capability can be defined using a specific set of molecular features, independently of allograft damage acquired during the peritransplantation period.
2 | RESULTS
2.1 | RNAseq cohort characteristics are related to differences associated with DGF and response to reperfusion injury
Differential gene expression associated with perfusion, or DGF status, was assessed alone, or stratified by donor gender (Figure 1b). Significant variation (28%) was observed between donor genders. This was more pronounced than the pre-/postperfusion or DGF/IGF variance, each of which grouped into distinct gender clusters (Supporting Information Figure S1 in Data S1). The magnitude of change in pre- versus postperfusion biopsies was more pronounced than in a comparison of DGF versus IGF biopsies (Figure 1b, Supporting Information Figure S2 in Data S1). Hierarchical clustering of the samples is presented in Figure 1c. 3,893 of 4,052 transcripts were differentially expressed between preperfusion and postperfusion samples after adjustment for donor gender. Fifty-five transcripts correlated uniquely with DGF outcome, and only six of these were significantly different between DGF and IGF after adjustment for donor gender: REG1B, GABBR1, UBD, DAZ1, ABCA7 and BTN3A2.
Comparison of whole transcriptional profiles (TOM1) with different clinical risk factors for DGF including ECD/SCD, DCD/DBD, cold (CIT) and warm ischaemia time (WIT) did not reveal any overt transcriptional changes. However, each factor was independently associated with unique transcript expression (Figure 1d). CDKN2A/p16ink4a expression correlated with DGF occurrence and long-term allograft function post-transplantation, when used as a composite BioAge (pre-transplant donor risk classification system) (Gingell-Littlejohn et al., 2013; McGuinness et al., 2016). No common targets were associated with BioAge and DGF risk factors. Eight hundred and eighty-one transcripts were differentially expressed between samples with low versus high CDKN2A/p16ink4a expression (below/above median for sequenced sample set) with 349 being unique for BioAge. Stratification by BioAge revealed 58 common targets for DBD/DCD, ECD/SCD and BioAge (Figure 1d).
Further analysis, in conjunction with (a) perfusion status changes (preperfusion vs. postperfusion, TOM2) and (b) differences between DGF and IGF transcriptomes (TOM3), established a molecular signature for DGF adjusted for the effect of reperfusion (DGF-specific signature; TOM4).
Forty-nine transcripts were identified as markers of DGF (TOM3). However, adjustment for perfusion status (TOM4) reduced this to 42 differentially expressed transcripts (Supporting Information Table S1 in Data S2, Figure 2). DGF outcome and donor gender were related to the DGF-specific signature, with male and female donors forming distinct clusters indicative of a donor gender-driven DGF phenotype (Figure 2a). Further analysis of DGF-specific transcripts revealed that overall expression changes in response to reperfusion occurred along a similar trajectory in both DGF and IGF, but the magnitude of this change was greater for those exhibiting DGF. This suggests that the degree of response to reperfusion injury is significant in post-transplant outcome (Supporting Information Figure S2 in Data S1). Further analysis of the transcriptome for DGF-specific signatures independent of IRI, but stratified by BioAge, revealed the presence of only 22 DGF-specific targets (Figure 2b). Analysis of the DGF signature across the age groups (<50; 50–60 and >60) revealed significant changes for NLR4C, IL7R and GRIN3B.
Additional analyses, involving comparison of the allograft response to IRI (both for DGF and IGF status), using the respective RNAseq data sets, with publically available data sets for other renal pathologies, were undertaken to identify additional transcripts. These were included in further validation testing (Supporting Information Data S1 and S3).
2.2 | DGF, immune response generation and senescence pathways
Initial analysis of the effect of perfusion status on the pathways associated with senescence revealed 22 transcripts that had been significantly affected, these included BM11, CDK2, CDKN1A, ETS2, RB1, RBL2, p53/Rb signalling (CDKN2B, CITED2, ING1, MYC, PCNA, PIK3C, SERPINE1, SIRT1, SPARC), interferon-related (CDKN1A, EGR1, INFG, IRF7, RB1), IGF1, MAP2K3, PCNA, as well as cell adhesion molecules affected by senescence (COL1A1, COL3A1).
Ingenuity Pathway Analysis (IPA®; QIAGEN Redwood City; USA) indicated that DGF was allied with activation of innate immune responses, including GABA signalling, TREM1 signalling, pattern recognition receptors and B-cell development (Supporting Information Tables S2–S5 in Data S2). The top ranked networks comprised immune system activation, cell death/survival, cellular fitness and cell–cell communication, renal and urological system development and function. An upstream regulator analysis algorithm also identified activation of the immune system and cell-mediated immune responses in the development of DGF. The main “drivers” within these processes, typically related to the response to pathogens (bacterial and viral), or sterile inflammation and induction of the secondary INF-γ-associated responses to dsRNA, which have been linked to mechanisms underpinning autoimmune diseases (Bencicova & Diebold, 2013; Nellimarla & Mossman, 2014; Pollard, Cauvi, Toomey, Morris, & Kono, 2013) and cellular ageing (Gorbunova, Boeke, Helfand, & Sedivy, 2014).
Further analysis of our data sets, either individually, or in direct comparison with publicly available data sets, demonstrated overt links to inflammatory responses, inherent in renal pathologies, and highlighted the importance of interaction between lymphoid and nonlymphoid cells in the context of renal function (Supporting Information Figures S1–S4 in Data S3). Pathway analysis of differentially expressed targets relative to perfusion status highlighted pathways...
overlapping with processes inherent in ageing, including eIF2 (eukaryotic initiation factor 2) signalling, protein ubiquitination, eukaryotic initiation factor 4 (eIF4) and ribosomal protein S6 kinase beta-1 (p70S6K) signalling and interleukin-17 (IL-17)-mediated cytokine regulation (Supporting Information Tables S1–S3 in Data S5).
These data suggest that DGF outcome may be related to a differential capacity to restore physiological homeostasis following IRI. Further analysis of an additional model (DGF pre- vs. postperfusion and IGF pre- vs. postperfusion, Model 2) revealed that the transcriptional response to reperfusion injury was similar for allografts, irrespective of their post-transplant outcome. Resolution of transplant-related stresses and restoration of physiological homeostasis were, however, exacerbated in those that manifested DGF (Supporting Information Tables S5–S10 in Data S5) suggesting, that donor-organ resilience to stress may be a key determinant of DGF, which is congruent with recent findings linking organ function to biological age (Gingell-Littlejohn et al., 2013; McGuinness et al., 2016). Significantly, this is supported by the observation that the transcriptional amplitude of change in DGF signature genes following reperfusion was significantly larger than in IGF. This is consistent with there being a greater degree of allostatic load in organs developing DGF and an inability to restore transcriptional and physiological processes to function within normal physiological parameters as quickly as organs with IGF.
2.3 Epigenetic status is linked to DGF and perfusion status
The effect of IRI on epigenetic status associated with DGF outcome was analysed in connection with changes in global DNA methylation...
and frequency of alternative splicing (AS) events. Alternative splicing is a common post-transcriptional modification that enables cells to increase protein diversity from a single copy of a gene by generation of unique coding transcripts or regulatory noncoding RNAs. These can be affected by changes in DNA methylation status and GC content at the intron–exon boundaries, ultimately affecting both splicing outcomes, alternative splicing networks and affecting writing and/or maintenance of epigenetic marks and changes in chromatin status (Francisco & Baralle, 2017; Shiran Naftelberg, Ast, & Kornblihtt, 2015). Transcripts associated with chromatin remodelling were significantly affected by the perfusion status in our cohort and included polycomb group genes (BMI1, SUZ12, TRIM27); chromobox/HP1 homologs (CBX4, CBX8); bromodomain proteins (BRD2, WDR11); ING family members (ING1, ING2, ING4) and PHF21B.
DGF-specific transcripts revealed differential promoter methylation status dependent upon perfusion state and DGF occurrence. An increase in DNA methylation within the promoter and intragenic regions of the respective DGF-associated genes was observed for IGF compared to DGF in preperfusion samples. This relationship was lost, or reversed after reperfusion, suggesting that reperfusion can directly and immediately affect epigenetic status in tissues. These observations are supported by animal model studies where 30-min IRI was sufficient to affect global DNA methylation levels (Endres et al., 2000; Meller, Pearson, & Simon, 2015).
A representative comparison between epigenetic status, RNAseq and qPCR data is summarized in Supporting Information Figures S1 and S2 in Data S4. Interestingly, an increased incidence of AS events was associated with reperfusion injury, but not DGF occurrence. No AS events were detected in the DGF-specific transcript set. The top 100 differentially expressed transcripts between perfusion states were analysed, with only 42 displaying AS events, as indicated by differential exon expression (Figure 3a,b, Supporting Information Data S6). A representative example of an AS event has been illustrated using interferon regulatory factor 1 (IRF1) (Figure 3c,d). This is suggestive of a complex relationship between methylation status and alternative splicing associated with IRI injury that is not observed in organs displaying DGF.
2.4 Validation of DGF-specific transcripts
Transcripts related to DGF, selected based on their ranking by statistical significance of observed expression change and diversity of signalling pathway involvement in relation to publically available data sets, were further validated in 19 paired biopsies from the RNAseq cohort and an independent cohort of 32 pairs of samples. Three samples were excluded from further analysis due to post-transplant complications. Nineteen genes were validated as markers of DGF.
Transcripts were further analysed in relation to donor characteristics, ischaemic time and estimated glomerular filtration rate (eGFR/MDRD4), as a measure of physiological function, at 3, 6 and 12 months post-transplantation (Table 2). Interestingly, most of the DGF-specific transcripts that correlated positively with donor age were negatively correlated with allograft performance after transplant, including PTPRC, SEMA3A, KLRB1, CD52, CCL19,
LNRC4, GABBR1 and UBD. KLRB1 was a common denominator for DGF, ECD status, DCD status and allograft performance, while SEMA3A was differentially expressed in relation to DGF, ECD status; Cr T1/2 and allograft function post-transplant. FCGR1C was related to DGF outcome, DCD status and allograft function.
Additionally, RNAscope was performed on 18 preperfusion biopsies from the RNAseq cohort for which tissue sections were available, to confirm the presence of five selected transcripts namely CD45, INFγ, ACKR3, SEMA3A, REG1B and PPIB as a positive control.
2.5 | Validation of DGF gene expression at the protein level
The expression of DGF-specific genes at the protein level was undertaken using western blotting in a subset of samples to determine whether changes at the protein level could be related to DGF and/or reperfusion injury (Supporting Information Data S7). Samples were isolated sequentially from the same tissue specimen after RNA and genomic DNA isolation and equal amount of total protein was used in the further analysis. Eight DGF-associated proteins were detected. Four (FCGR1C, FCGR2C, CD52 and PTPRC) were not detectable. Five genes associated with DGF were further validated in the tissue biopsies to verify both their expression and localization at the protein level in 18 preperfusion biopsies. Morphological and histological analyses of these biopsies were undertaken for 17 histological variables associated with renal pathology (including CKD and AKI) and assessed using several published histological scoring systems for kidney quality (Maryland Aggregate Pathology Index, Chronic Allograft Damage Index, Banff Score and Remuzzi Score).
These revealed no obvious correlation between overall quality score and the presence or absence of DGF. The expression of CD45, IFNγ, REG1B or SEMA3A showed no significant differences in signal intensity or location between biopsies from donor kidneys that would experience IGF compared with those demonstrating DGF. ACKR3 expression was negative overall with the exception of two samples.
2.6 | DGF and perfusion status are associated with cellular senescence
The expression of the CDKN2 locus in relation to the DGF outcome and perfusion status at the transcript level, including CDKN2A/p16INK4, ARF/p14, CDKN2B, as well as other senescence-associated markers (TP53, CDKN1A/p21 and CDKN1B/p27) was determined.
Significantly, higher expression of CDKN2B/p15 and CDKN1A was noted in postperfusion biopsies compared to the preperfusion biopsies (p = 0.0003 and p < 0.0001, respectively).
Furthermore, a positive relationship between CDKN2A/p16INK4 and TP53 in preperfusion biopsies (cc = 0.305; p = 0.039) was observed; this correlation was lost after reperfusion. CDKN2A/p16INK4 was positively correlated with ARF/p14 (cc = 0.48, p = 0.000) and CDKN2B/p15 (cc = 0.405, p = 0.004) in preperfusion biopsies, whereas ARF/p14 was correlated with CDKN2B (cc = 0.397, p = 0.004). CDKN2A/p16INK4 was correlated with donor age (cc = 0.489; p = 0.000 and cc = 0.419, p = 0.002) and MDRD4 at 3 months (cc = −0.473; p = 0.001 and cc = −0.310, p = 0.030) and 6 months (cc = −0.471; p = 0.001 and cc = −0.397, p = 0.005) post-transplant regardless of the perfusion status. However, only preperfusion CDKN2A/p16INK4 was negatively correlated with MDRD4.
## Table 2 Summary of validated DGF-specific targets in the RNAseq cohort, the validation cohort (directionality of changes indicated by arrows)
| RNAseq cohort (IGF = 11, extreme DGF = 8) | Validation cohort (IGF = 7, DGF = 9) | Combined cohort (Overall DGF = 17; no DGF = 34) | DCD vs. DBD | ECD vs. SCD | Correlations (correlation coefficient; p-value) |
|------------------------------------------|--------------------------------------|-----------------------------------------------|--------------|--------------|-----------------------------------------------|
| ACKR3 Pre (p = 0.02)‡ | C1QB Pre (p = 0.029) | ABCA8 Pre (p = 0.017)† | REG1A Pre (p = 0.026)† | CXCL9 Pre (p = 0.009)† | Donor age (MDRD4 at 3 months) (0.46; 0.001) TAGAP Pre (0.45; 0.001) PTTPC Pre (0.35; 0.000) PTTPC Pre (0.45; 0.001) |
| BTN3A2 Post (p = 0.041)† | FCGR1C Pre (p = 0.01)† | ACKR3 Pre (p = 0.07)§ | CHGB Post (p = 0.015)† | UBD pre (p = 0.036)† | MDRD4 at 6 months (0.45; 0.02) PTTPC Pre (0.34; 0.014) PTTPC Post (0.34; 0.020) ACKR3 Post (0.37; 0.019) AKR3 Pre (0.35; 0.010) |
| CCL19 Post (p = 0.041)† | FCGR1C Pre (p = 0.001)† | BTN3A2 Post (p = 0.031)† | CHGB Pre (p = 0.021)† | TAGAP Pre (p = 0.001)† | MDRD4 at 12 months (0.33; 0.028) NLRC4 Pre (0.32; 0.024) NLRC4 Pre (0.34; 0.014) NLRC4 Pre (0.48; 0.000) |
| CD52 Pre (p = 0.021)† | KLRB1 Pre (p = 0.029)† | CD52 Pre (p = 0.029)† | CORIN Post (p = 0.001)† | PTTPC Pre (p = 0.001)† | CIT (0.33; 0.000) WIT (0.33; 0.025) CHGB Post (0.35; 0.006) |
| CHGB Post (p = 0.008)† | NLRC4 Pre (p = 0.043)† | CHGB Post (p = 0.012)† | REG1B Post (p = 0.026)† | KLRB1 Pre (p = 0.028)† | Cr T1/2 (0.01; 0.01) |
| CHGB Pre (p = 0.01)† | OAS2 Pre (p = 0.045)† | CXCL10 Post (p = 0.039)† | FCGR1C Pre (p = 0.001)† | CD69 Pre (p = 0.008)† | Cr T1/2 (0.01; 0.01) |
| CXCL10 Post (p = 0.015)† | PTTPC Post (p = 0.029)† | FAM34A Pre (p = 0.012)† | TRIM Pre (p = 0.031)† | CD52 Pre (p = 0.014)† | Cr T1/2 (0.01; 0.01) |
| CORIN Pre (p = 0.017)↓ | PTTPC Pre (p = 0.020)† | FCGR1C Post (p = 0.020)† | TRIM Post (p = 0.003)† | SEMA3A Pre (p = 0.041; 0.007) | Cr T1/2 (0.01; 0.01) |
| DISP2 Post (p = 0.013)† | SEMA3A Post (p = 0.018)† | FCGR1C Pre (p = 0.009)† | RSPO1 Pre (p = 0.013)† | C1QB Pre (p = 0.033)† | Cr T1/2 (0.01; 0.01) |
| DISP2 Pre (p = 0.02)↑ | GABBR1 Post (p = 0.037)† | KLRB1 Post (p = 0.007)† | ISG20 Pre (p = 0.045)† | UBD Pre (p = 0.020)† | SEMA3A Post (0.33; 0.025) |
| FAM34A Pre (p = 0.043)↓ | KLRB1 Pre (p = 0.026)† | FAM34A Pre (p = 0.028)† | LLCOB Post (p = 0.040)† | SEMA3A Post (0.36; 0.016) | SEMA3A Pre (0.36; 0.014) |
(Continues)
| RNAseq cohort (IGF = 11, extreme DGF = 8) | Validation cohort (IGF = 7, no DGF = 34) | Combined cohort (Overall DGF = 17; no DGF = 34) | Correlations (correlation coefficient; p-value) |
|------------------------------------------|------------------------------------------|-----------------------------------------------|-----------------------------------------------|
| **Donor age** | **MDRD4 at 3 months** | **MDRD4 at 6 months** | **MDRD4 at 12 months** | **CIT** | **WIT** | **Cr T1/2** |
| **FCGR1C Post** | | | | | | | |
| $(p = 0.041)\dagger$ | | | | | | | |
| **FCGR2C Post** | | | | | | | |
| $(p = 0.033)\dagger$ | | | | | | | |
| **FCRL3 Pre** | | | | | | | |
| $(p = 0.013)\dagger$ | | | | | | | |
| **FLG Pre** | | | | | | | |
| $(p = 0.042)\dagger$ | | | | | | | |
| **KLRB1 Pre** | | | | | | | |
| $(p = 0.026)\dagger$ | | | | | | | |
| **OAS Post** | | | | | | | |
| $(p = 0.027)\dagger$ | | | | | | | |
| **REG1B Post** | | | | | | | |
| $(p = 0.043)\dagger$ | | | | | | | |
| **REG1B Pre** | | | | | | | |
| $(p = 0.034)\dagger$ | | | | | | | |
| **NLRP2 Pre** | | | | | | | |
| $(p = 0.013)\dagger$ | | | | | | | |
| **UBD Post** | | | | | | | |
| $(p = 0.006)\dagger$ | | | | | | | |
| **UBD Pre** | | | | | | | |
| $(p = 0.013)\dagger$ | | | | | | | |
| **SEMA3A Post** | | | | | | | |
| $(p = 0.004)\dagger$ | | | | | | | |
| **TAGAP Post** | | | | | | | |
| $(p = 0.021)\dagger$ | | | | | | | |
| **ZNF676 Pre** | | | | | | | |
| $(p < 0.001)\dagger$ | | | | | | | |
Notes. The Mann–Whitney U test was used to compare variables in relation to different categories analysed. The FDR adjustment for multiple comparisons was performed and all targets retained significance ($p < 0.05$). Spearman correlation was used to analyse DGF-specific targets with donor age, CIT, WIT, Cr T1/2 and MDRD4 at 3, 6 and 12 months post-transplant. The table presents unadjusted $p$-values.
CIT: cold ischaemic time; Cr T1/2: time of reduction in serum creatinine by 50%; DBD: donation after brain death; DCD: donation after cardiac death; ECD: extended criteria donor; IGF: immediate graft function no DGF includes allografts with primary function; MDRD4: estimated glomerular filtration rate modification of diet in renal disease (MDRD) equation; SCD: standard criteria donor; WIT: warm ischaemic time.
compared to younger donors. The expression of preperfusion CDKN1A was positively correlated with CIT (cc = 0.347, p = 0.017), whereas serum creatinine at 12 months post-transplant was positively correlated with postperfusion expression of CDKN1B (cc = 0.333, p = 0.033) and CDKN1A (cc = 0.344, p = 0.022). Additionally, preperfusion expression of CDKN1B was associated with serum creatinine (cc = 0.432, p = 0.003) and MDRD4 (cc = −0.346, p = 0.020) at 6 months post-transplant.
Four preperfusion biopsies from the RNAseq cohort were selected to visualize histological differences in extreme DGF and IGF outcomes between young (age 40–42) and old (age 74) donors and were stained for CDKN2a/p16INK4 and γH2AFX (Figure 4).
Overall CDKN2a/p16INK4 positivity appears to be predominately cytoplasmic, with greater positivity within distal tubular epithelium versus proximal tubular epithelium across all samples and with proximal tubular epithelium showing greater positivity in older donors compared to younger donors.
Overall, regardless of DGF/IGF status, positivity for γH2AFX appeared to be predominately nuclear in tubular epithelial cells, at least mild and more often in the distal tubules than the proximal tubules. Interestingly, the subcellular localization of the γH2AFX signal within arterial myocytes appeared to associate with donor age more so than with immediate function status after transplantation, in that biopsies from kidneys from younger donors showed more cytoplasmic than nuclear γH2AFX, with older kidneys showing greater nuclear than cytoplasmic positivity in arterial myocytes.
3 | DISCUSSION
To our knowledge, this work is the first study in human subjects that demonstrates a unique molecular signature for impaired physiological resilience (DGF), encompassing epigenetic and transcriptomic data sets. Critically, at a translational level, it also provides a platform for the development of a universal IRI signature and the ability to relate it to post-transplant outcomes. This is also the first study linking DNA methylation status to reperfusion injury and DGF outcome, in the context of immune system status, overall dysregulation of cellular homeostasis and its consequences for allograft performance. These data, together with the validation of DGF-associated gene products at the protein level, provide a unique and synchronous genome, transcriptome and proteome snapshot.
Our study provides strong evidence that biological age in combination with physiological stress, resulting from immune system activation and generation of inflammatory responses, plays a major role in DGF occurrence and the physiological manifestations of IRI. From a clinical perspective, this also suggests that these effects are driven by donor characteristics, which may therefore be even more discriminating than reperfusion injury itself. Correspondingly, BioAge in the preperfusion biopsy analyses appeared to be a significant determinant of post-transplant allograft function, in keeping with previous observations centring on the CDKN2 and CDKN1 loci (Gingell-Littlejohn et al., 2013; McGuinness et al., 2016). Furthermore, it suggests strongly that increased allograft biological age is contributory to less successful outcomes in renal transplantation and poorer post-transplant performance.
Activation of the immune system may be a prerequisite driver for DGF occurrence, with the predicted top ranked upstream regulators being associated with innate immune system activity. Upregulation of transcription for UBD, NLRC4, IFNγ and IFNγ-inducible targets OAS2 and CXCL10 is congruent with a model of inflammation-some activation leading to the recruitment of corresponding IFNγ effector cells, represented by elevated transcriptional levels observed for CD52, CD45, FCGR1C and KLRB1. These data suggest that therapeutic intervention to modulate the innate immune pathway activation and related events would be expected to mitigate the effect of IRI and reduce acute kidney injury during the peritransplantation period. Furthermore, these observations are consistent with the thesis of inflammaging, whereby increased chronic inflammation may hyperinflrate biological ageing processes, thus causing more rapid deterioration of organ function.
Three of the top five ranked DGF-specific transcripts locate to the major histocompatibility class 1 (HLA) locus. GABBR1 (6p22.1) and ubiquitin D (UBD; 6p21.3) are located very close to the region coding for HLA-F (Fan et al., 1996), suggesting that this locus may
FIGURE 4 Immunohistochemical staining for γH2A (top panel) and CDKN2a/p16INK4 (bottom panel) on kidney biopsies from young and old patients with IGF (immediate graft function) and DGF (delayed graft function) outcomes, respectively. Sections were scanned at 20x magnification. NC: negative control.
be of importance for the development of DGF and impairment of physiological resilience. The third transcript, BTN3A2, originates from the gene located in the juxta-telomeric region of the HLA locus and is involved in the adaptive immune response and inhibition of INFγ release. GABBR1 has been associated with proteasome activation and cytoskeleton remodelling, with differential responses to IRI, dependent upon severity of insult (Caldeira, Salazar, Curcio, Canzoniero, & Duarte, 2014). UBD has also previously been implicated in renal pathology, via involvement in inflammatory-mediated signalling and immune system modulation. UBD directs its substrates for 26S proteasomal degradation (irreversible proteolysis) and accelerates autophagy in nutrient-deprived conditions (Gong et al., 2010; Schmidtke, Aichem, & Groettrup, 2014). This can be supported by significant transcriptome changes associated with mitochondria and mitochondrial energy metabolism (PMAIP1, SLC25A25, BBC3, SH3GL1, NEFL, SLC25A2, HSP90AA1, CPT1B, GADD45B, DNAJB1, LRP5L, ARRDC3, HSPA1B, HSPA1A, EDN1) and autophagy (MAP1LC3B, CXCR4, DAPK1, EIF2AK3, INFG, IGF1, PTEN, TNFα, RB1, DRAM2, HSP90AA1).
The activation of the recipient immune system and downstream signalling pathways may be superseded by the subsequent impact of repair/regeneration pathways. It is possible that this inability to sufficiently resolve inflammatory responses manifests as DGF and could be attributed to a differential response to reperfusion injury by biologically older allografts and decreased resilience to stress (Gingell-Littlejohn et al., 2013; McGuinness et al., 2016; O’Neill et al., 2015; Salvadori, Rosso, & Bertoni, 2015). Such a scenario is supported by the reperfusion signature associated with cellular events (metabolic shift, autophagy, RNA metabolism, ribosomal biogenesis, protein synthesis) activated in response to environmental stress (ischaemic damage, nutrient deprivation and hypoxia), which facilitates cellular repair after insult, or induces apoptosis if the damage is too severe. Recent evidence has indicated that these prosurvival and repair pathways associated with ageing are conserved across taxa, and include the mTOR, AKT and p38 pathways, suggesting that insufficient resolution of the response to peritransplant stresses is associated with dysregulation of cellular homeostasis (Gingell-Littlejohn et al., 2013; Kennedy & Lamming, 2016; Figure 5). These pathways are critical regulators of cellular metabolism allowing cells to sense and adapt to environmental factors, with some being involved in the regulation of lifespan in model organisms, including mTOR signalling. Notably, sirolimus, a mTOR inhibitor, is already used as an immunosuppressant following renal transplantation. While use of sirolimus is associated with reduced risk of malignancy in transplant recipients, it also correlates with an increased risk of death when the allograft has originated from a cadaveric donor. No such correlation, however, has been observed from living donor allografts, supporting the central role of mTOR signalling and associated pathways for the regulation of cellular metabolism and health span (Knoll et al., 2014).
Our results are in keeping with recent studies linking organ biological age, as opposed to chronological age, to allograft performance post-transplant and overall kidney function (Gingell-Littlejohn et al., 2013; McGuinness et al., 2016). Increased biological age of renal allografts can be associated with reduced functional/repair capacity as a result of a greater degree of allostatic load/overload. This is thought to contribute to the DGF phenotype, as a result of an inability to restore physiological homeostasis in the face of peritransplant stress and subsequent recipient immune challenge (Kooman et al.,...
Our data are in keeping with such a scenario, with organs displaying DGF also exhibiting a greater change in transcriptional amplitude in DGF signature transcripts following transplantation and requiring a longer period to restore physiological homeostasis, which may be related to deficient proteostasis. These data indicate that allografts exhibiting DGF may therefore be displaying features of allostatic overload at a transcriptional level whose effects are extrapolated across the organ as a whole, resulting in functional impairment (Kooman et al., 2014).
Analysis of transcript expression observed solely in DGF, but not IGF, in relation to perfusion status, indicated a transcript biotype shift (Supporting Information Table S1 in Data S8), including an increase in antisense, pseudogenes, noncoding and coding RNAs, for example immunoglobulin gene (IgV and IgV pseudogenes) and T-cell receptor (TR J) transcripts. These changes in the transcriptome biotypes further support the hypothesis that the response to IRI, both in magnitude and context, are dependent upon donor characteristics and organ response/resilience to stress and may also reflect deregulation of alternative splicing networks and epigenome status overall. The biotype changes observed may reflect a burst of “transcriptional noise” in DGF allografts in response to IRI, as a direct result of changes in the methylation status of promoters and intragenic regions (Huh, Zeng, Park, & Yi, 2013). These observations are also consistent with the derepression of LINE elements in ageing cells (De Cecco et al., 2013).
Immunological response to “danger signals” may lead to excessive activation of proinflammatory cytokines and chemokines, consequently leading to organ damage over time, as observed in the case of autoimmune diseases, where there is a loss in ability to downregulate/attenuate proinflammatory signalling (de Jesus, Canna, Liu, & Mansky, 2015). This suggests that donor characteristics are important for DGF occurrence and may be linked to organismal/organ stress levels in relation to the type of organ donation (Bon et al., 2012; Morrissey & Monaco, 2014). Ultimately, allograft quality will be related to organ resilience to stress, and this by itself provides an opportunity for the development of new therapeutic interventions aimed at exploiting this phenomenon. Recent studies focusing on the progression of chronic kidney disease (CKD) have exemplified the importance of epigenetic changes and loss of kidney function (Smyth, McKay, Maxwell, & McKnight, 2014; Wing et al., 2014). Our data have indicated rapid changes in the epigenome during perfusion, suggesting that the effect of IRI on long-term allograft function may be more pronounced than originally anticipated.
The lack of available models for DGF has meant that direct validation of targets and any related mechanism has not been possible. To mitigate the impact of this shortfall, we have therefore used our DGF transcriptomic signature to identify any commonality with other renal/urological transcriptomic expression profiles derived from publicly available data sets. These included renal interstitial fibrosis, kidney transplant failure and rejection, kidney disease, nephrotic syndrome, cystic disease of kidney and renal tubular disorder (Supporting Information Data S3). Notably, the overlap identified encompassed transcripts involved in immune system activation, both supporting the importance of interaction between lymphoid and nonlymphoid cells in the context of renal function and highlighting the biological plausibility of the DGF-related findings.
Overall, our data suggest that allografts exhibiting DGF present with an impaired ability to restore physiological homeostasis in response to stress, consistent with their biological age and associated allostatic load. This is reflected in changes in epigenome, transcriptome and dysregulation of RNA metabolism. The magnitude of change in transcriptional amplitude in response to physiological stress, along with elevated expression of noncoding RNAs and pseudogenes, raises the possibility that reduction in available cellular resources for activation of damage repair mechanisms slows down physiological and cellular repair processes, resulting in long-term damage to the allograft (Figure 5).
4 | EXPERIMENTAL PROCEDURES
4.1 | Clinical cohort characteristics
Fifty-five paired preperfusion and postperfusion renal biopsies collected from deceased donors were included in this study, and all kidneys were subsequently transplanted with no occurrence of primary nonfunction. Detailed patient characteristics and follow-up markers are described in Table 1. This study and consent procedure was approved by the Regional Ethics Committee of North Glasgow NHS Trust (GN10SU334, 10/50704/42, NHS GG&C Biorepository −276 and 348). Donors from the national pool donated their organs for transplantation. The recipient of the organ provided preoperative written informed consent and records are stored at Queen Elizabeth University Hospital (QEUH). From this cohort, 24 pairs of samples defined as the extreme DGF phenotype or immediate graft function were selected for further analysis (Figure 1a).
4.2 | Biopsy processing
Total RNA, genomic DNA and protein were sequentially isolated from the same tissue biopsies using TRI®Reagent according to the manufacturer's instructions (Invitrogen, UK).
4.2.1 | RNA isolation
Total RNA from kidney biopsies was extracted using TRI®Reagent according to the manufacturer’s instructions, DNase treated (RNA Clean & Concentration, #R1015, Zymo Research, USA) and stored at −80°C for further analysis. RNA underwent spectral analysis (A260/280 nm) and determination of RIN number (RNA Nano kit and 2100 BioAnalyzer (#5067–1511, Agilent Technologies, Inc. USA). For further analysis, samples with RIN>6.0 were used.
4.2.2 | RNaseq and data analysis
Libraries, from 400 ng total RNA, were created using ribosomal depletion (n = 48, 24 paired biopsies; TruSeq Stranded Total RNA
were sequenced on a NextSeq500 (Illumina) using a paired-end 75 × 75 bp run. The raw sequence reads in FASTQ format were further analysed using the following pipeline: Initial QC for RNAseq output was analysed using FastQC v.0.11.2 (https://www.bioinformatics.bbsrc.ac.uk/projects/fastqc), adapters were removed using trim_galore v.0.3.7 (https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/). Three bases from the 3’ end of all paired-reads were trimmed to avoid any biases related to adapter sequence or basecall quality. Poor quality bases (phred score <20) were removed. Two samples 39B1 and 106B were excluded from further analyses as they failed QA and QC checks. The trimmed reads were mapped against the reference human genome (GRCh38 from ensemble) using TopHat v.2.0.13 (Kim et al., 2013). The annotation file (GRCh38 release 78 from Ensemble) was used for mapping of transcript annotations. Mapped fragment counts were summarized using featureCounts with ensemble gene-ids using subread v1.4.6 (Liao, Smyth, & Shi, 2014). Differential gene expression (DGE) analysis was performed using DESeq2 v1.6.3 (Love, Huber, & Anders, 2014). Raw count data were transformed to log2 scale to normalize expression counts. Multiple testing correction was performed using the Benjamin–Hochberg approach to control false discovery rate (FDR) at 10% (FDR ≤ 0.1 was considered significant). Differentially expressed gene targets were analysed using Ingenuity®Pathway Analysis (IPA®, QIAGEN's, USA) and NextBio Research (Illumina, USA).
4.2.3 Alternative splicing analysis
Alternative splicing events were investigated in the RNAseq cohort (top 100 differentially expressed transcripts, pre vs. post) and DGF-specific transcripts identified by RNAseq using DESeq2 v1.16.10 (Anders, Reyes, & Huber, 2012). Differential exon usage (DEU), specific transcripts identified by RNAseq using DEXSeqv1.16.10 (IPA®, QIAGEN's, USA) and NextBio Research (Illumina, USA).
4.2.4 Whole genome bisulphite sequencing and analysis
Genomic DNA was isolated from the same biopsy, using TRIzol®Reagent, after separation of RNA into aqueous fraction and further purified (Genomic DNA Clean & Concentrator, Zymo Research, USA). 100 ng of genomic DNA was bisulphite converted using EZ DNA Methylation-Gold™ Kit (Zymo Research). Bisulphite-converted genomic DNA was used to generate indexed libraries using EpiGnome™-Methyl-Seq kit (Epigenome®Illumina) according to the manufacturer’s instruction. Library quality and quantity were assessed using DNA High Sensitivity kit (Agilent Technologies, Inc. USA). Sample selection for WGBS included 20 samples (10 matched pairs with five being DGF and five IGF). The resulting 20 libraries were sequenced on a NextSeq500 (Illumina) with 30× coverage, paired-end run (2 × 150 bp). The raw sequence reads in FASTQ format underwent QC as previously described above. Sample 180b1 was excluded from further analysis as it failed QA and QC.
The human reference genome (GRCh38) was in silico bisulphite converted before aligning trimmed reads with Bismark v0.10.1 and bowtie2 (v2.1.0) as previously described(Krueger & Andrews, 2011; Langmead & Salzberg, 2012). PCR bias was removed by a deduplication step. The methylation content was measured on CpG context sites of DGF-specific targets. The promoter and intragenic regions were extracted from biomaRT (BiomaRT; Durinck, Spellman, Birney, & Huber, 2009), and differences within the methylated CpG sites were further analysed using Kruskal–Wallis test. FDR correction for multiple comparison was applied for all analyses. Adjusted p-value below 0.05 was considered as statistically significant.
4.2.5 QPCR and data validation
For each individual RT reaction, 150 ng of total RNA from each sample was used and reverse transcription was performed using SuperScript™II Reverse Transcriptase (# Life Technologies Inc., UK) and then qPCR was performed. Gene expression was analysed using TaqMan®-gene expression assays, or custom design assays using Roche UPL (Supporting Information Data S9), which were normalized against HPRT1 and 18S rRNA control primer sets. Taqman® assays, including standards, were performed using the manufacturers recommended qPCR protocols and TaqMan®Master Mix (#4370074, Life Technologies, UK). For UPL probes, primers were used at final concentration 360 nM while probes were used at final concentration of 100 nM. The comparative threshold cycle method (ΔΔCT) was used to quantify relative gene expression, and the obtained quantification was transformed to exponential value 2^–ΔΔCT. Commercially available RNA was used as a calibrator (#AM7976, Life Technologies, Inc.). Further testing involved Spearman correlations and Kruskal–Wallis test. FDR correction for multiple comparison was applied for all analyses. Adjusted p-value below 0.05 was considered as statistically significant.
4.2.6 RNAscope
In situ hybridization detection for CD45 (601998), REG1B (312058), ING7 (310508), SEMA3A (416568), ACKR3 (441458) and PP1B (313908) mRNA was performed using RNAscope 2.5 LS (Brown) detection kit (Advanced Cell Diagnostics, Hayward, CA) on a Leica Bond Rx autostainer strictly according to the manufacturer’s instructions. The analysis was performed using an established method at the Histology Core in the Beatson Institute for Cancer Research, Glasgow, UK.
4.2.7 Western blot
Protein fractions were isolated from the phenol–ethanol fraction after removal of genomic DNA (TRIzol®Reagent, Invitrogen, UK). Protein concentration was estimated using DC® Protein assay (BioRad, UK), and 7.5 µg of total protein was loaded per well. Samples were resolved in
4%–12% or 12% SDS-PAGE in MOPS buffer using NuPAGE® System (Life Technologies Inc., UK) and transferred onto the polyvinylidene fluoride (PVDF) membrane. After immunoblotting, membranes were washed in Tris-buffered saline (TBS), blocked in 5% nonfat dry milk in TBS-0.1% Tween 20 (TBST) for 1 hr at room temperature, followed by the incubation with primary antibodies in blocking solution overnight at 4°C. After incubation, membranes were washed in TBST and incubated with the goat antirabbit IgG (1:10,000; Cell Signalling, #7074S) or goat antitmouse IgG (1:5,000, Cell Signalling, #7074) horse-radish peroxidase (HRP)-conjugated for 1 hr in room temperature. The reaction was developed using Enhanced Chemiluminescence (ECL) System (Life Technologies Inc.). The following primary antibodies from Abcam (UK) were used: CHGB (1:4,000; ab151568), REG1B (1:1,000; ab87205), Corin (1:500; ab56158), UBDB (1:500, ab134077), KLKB1 (1:500, ab197979), SEMA3A (1:1,000, ab23393), ACKR3 (1:2,000), ZNF676 (1:500, ab179754), TAGAP (1:1,000, ab187664), HPRT1 (1:10,000, ab109021), FCGR1C (ab119843), FCGR2C (1:2,000), ZNF676 (1:500, ab179754), TAGAP (1:1,000, ab187664), HPRT1 (1:10,000, ab109021), FCGRC (ab119843), FCGRC1 (ab125013), CD52 (ab194860) and PTPRC (ab23393), ACKR3 (ab81299, 1:1,000; Abcam, UK) antibodies were stained manually.
4.2.8 Immunohistochemistry
The analysis was performed using an established automated method at the Histology Core in the Beatson Institute for Cancer Research, Glasgow, UK. Heat-induced antigen retrieval was performed using sodium citrate retrieval buffer (pH = 6, Thermo, UK) at 98°C for 25 min followed by peroxidase block (Dako, UK) for 5 min. The previously optimized primary antibodies from Abcam: ACKR3 (ab72100, 1:500), INFγ (ab9657, 1:500), SEMA3A (ab23393, 1:500) and CD45 (Dako, M0701, 1:1,000), REG1B (MyBioSource, MBS2025956, 1:500) and an appropriate secondary antibody (EnVision, Dako, UK) were used. Staining and counterstaining were performed using 3,3′-diaminobenzidine (DAB) and haematoxylin, respectively. Previously validated CDKN2A/p16 (M-156-sc759; 1:250; Santa Cruz Biotechnology Inc, USA) and γH2AFX (ab81299, 1:1,000; Abcam, UK) antibodies were stained manually.
4.3 Data and materials availability
RNAseq and WGBS data have been deposited at GEO repository with reference no: GSE90865 for publication, and this includes subseries GSE90861 and GSE90863. All relevant data are available upon request. Data from this study are available from the senior author (PGS).
CONFLICT OF INTEREST
No conflict of interest.
AUTHORS’ CONTRIBUTION
PGS, RBK, LD, KS and DBK designed and coordinated study. KS, DBK and OS collected clinical samples and compiled clinical data. DMcG, LM and OS performed laboratory work. SM and PAW generated computational data sets. PGS, DMcG, SM, PAW, KS, LD SC and DBK compiled and analysed data. PGS, DMcG and KS drafted manuscript. All authors read and commented on manuscript drafts.
ORCID
Dagmara McGuinness http://orcid.org/0000-0001-9057-3504
Shana M. Coley http://orcid.org/0000-0002-2152-5469
Paul G. Shiels http://orcid.org/0000-0002-7577-9843
REFERENCES
Aitken, E., Cooper, C., Dempster, N., McDermott, M., Ceresa, C., & Kingsmore, D. (2015). Delayed graft function is a syndrome rather than a diagnosis. Experimental and Clinical Transplantation, 13, 19–25.
Anders, S., Reyes, A., & Huber, W. (2012). Detecting differential usage of exons from RNA-seq data. Genome Research, 22, 2008–2017. https://doi.org/10.1101/gr.133744.111
Bon, D., Chataret, N., Giraud, S., Thuillier, R., Favreau, F., & Hauet, T. (2012). New strategies to optimize kidney recovery and preservation in transplantation. Nature Reviews Nephrology, 8, 339–347. https://doi.org/10.1038/nrneph.2012.83
Brenicova, E., & Diebold, S. S. (2013). Nucleic acids and endosomal pattern recognition: How to tell friend from foe? Frontiers in Cellular and Infection Microbiology, 3, 37.
Caldeira, M. V., Salazar, I. L., Curcio, M., Canzonero, L. M., & Duarte, C. B. (2014). Role of the ubiquitin-proteasome system in brain ischemia: Friend or foe? Progress in Neurobiology, 112, 50–69.
Christopher, P., & Murray, J. L. Global, regional, and national disability-adjusted life-years (DALYS) for 315 diseases and injuries and healthy life expectancy (HALE), 1990–2015: a systematic analysis for the Global Burden of Disease Study 2015. www.thelancet.com Lancetthe lancet.com 388, 1603–58 (2016).
De Cecco, M., Criscione, S. W., Peterson, A. L., Neretti, N., Sedivy, J. M., & Kreiling, J. A. (2013). Transposable elements become active and mobile in the genomes of aging mammalian somatic tissues. Aging, 5, 867–883. https://doi.org/10.18632/aging.100621
de Jesus, A. A., Canna, S. W., Liu, Y., & Goldbach-Mansky, R. (2015). Molecular mechanisms in genetically defined autoinflammatory diseases: Disorders of amplified danger signaling. Annual Review of Immunology, 33, 823–874. https://doi.org/10.1146/annurev-immunol-032414-112227
Durinck, S., Spellman, P. T., Birney, E., & Huber, W. (2009). Mapping identifiers for the integration of genomic datasets with the R/Bioconductor package biomaRt. NatureProtocols, 4, 1184–1191. https://doi.org/10.1038/nprot.2009.97
Endres, M., Meisel, A., Biniszewicz, D., Namura, S., Prass, K., Rüscher, K., … Dimagl, U. (2000). DNA methyltransferase contributes to delayed ischemic brain injury. The Journal of Neuroscience: the Official Journal of the Society for Neuroscience, 20, 3175–3181. https://doi.org/10.1523/JNEUROSCI.20-09-03175.2000
Fan, W., Cai, W., Parimo, S., Schwarz, D. C., Lennon, G. G., & Weissman, S. M. (1996). Identification of seven new human MHC class I region genes around the HLA-F locus. Immunogenetics, 44, 97–103. https://doi.org/10.1007/BF02660056
Francisco, E., & Baralle, J. G. (2017). Alternative splicing as a regulator of development and tissue identity. Nature Reviews Molecular Cell Biol, 18, 437–451. https://doi.org/10.1038/nrm.2017.27
Gingell-Littlejohn, M., McGuinness, D., McGlynn, L. M., Kingsmore, D., Stevenson, K. S., Koppelstaetter, C., … Shiels, P. G. (2013). Pre-transplant CDKN2A expression in kidney biopsies predicts renal function and is a future component of donor scoring criteria. PLoS One, 8, e68133. https://doi.org/10.1371/journal.pone.0068133
Gong, P., Canaan, A., Wang, B., Leventhal, J., Snyder, A., Nair, V., ... Ross, M. J. (2010). The ubiquitin-like protein FAT10 mediates NF-kappaB activation. Journal of the American Society of Nephrology: JASN, 21, 316–326.
Gorbunova, V., Boeke, J. D., Helfand, S. L., & Sedivy, J. M. (2014). Human genomics. Sleeping dogs of the genome. Science, 346, 1187–1188.
Huh, I., Zeng, J., Park, T., & Yi, S. V. (2013). DNA methylation and transcriptional noise. Epigenetics & Chromatin, 6, 9. https://doi.org/10.1186/1756-8935-6-9
Kennedy, B. K., & Lamming, D. W. (2016). The mechanistic target of rapamycin: The grand conductor of metabolism and aging. Cell Metabolism, 23, 990–1003.
Kim, D., Pettea, G., Trapnell, C., Pimentel, H., Kelley, R., & Salzberg, S. L. (2013). TopHat2: Accurate alignment of transcriptomes in the presence of insertions, deletions and gene fusions. Genome Biology, 14, R36. https://doi.org/10.1186/gb-2013-14-4-r36
Knoll, G. A., Kokolo, M. B., Mallick, R., Beck, A., Buenaventura, C. D., Ducharme, R., ... Fergusson, D. A. (2014). Effect of sirolimus on malignancy and survival after kidney transplantation: Systematic review and meta-analysis of individual patient data. BMJ, 349, g6679. https://doi.org/10.1136/bmj.g6679
Kooman, J. P., Kotanko, P., Schols, A. M., Shiels, P. G., & Stenvinkel, P. (2014). Chronic kidney disease and premature ageing. Nature Reviews Nephrology, 10, 732–742. https://doi.org/10.1038/nrneph.2014.185
Krueger, F., & Andrews, S. R. (2011). Bismark: A flexible aligner and methylation caller for Bisulfite-Seq applications. Bioinformatics, 27, 1571–1572.
Langmead, B., & Salzberg, S. L. (2012). Fast gapped-read alignment with Bowtie 2. Nature Methods, 9, 357–359.
Liao, Y., Smyth, G. K., & Shi, W. (2014). featureCounts: An efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics, 30, 923–930. https://doi.org/10.1093/bioinformatics/btt656
Love, M. I., Huber, W., & Anders, S. (2014). Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biology, 15, 550. https://doi.org/10.1186/s13059-014-0550-8
Mallon, D. H., Summers, D. M., Bradley, J. A., & Pettigrew, G. J. (2015). Defining delayed graft function after renal transplantation: Simplest is best. Transplantation, 96, 885–889.
McGlynn, L. M., Stevenson, K., Lamb, K., Zino, S., Brown, M., Prina, A., ... Shiels, P. G. (2009). Cellular senescence in pretransplant renal biopsies predicts postoperative organ function. Aging Cell, 8, 45–51. https://doi.org/10.1111/j.1474-9726.2008.00447.x
McGuinness, D., Leierer, J., Shapter, O., Mohammed, S., Gingell-Littlejohn, M., Kingsmore, D. B., ... Shiels, P. G. (2016). Identification of molecular markers of delayed graft function based on the regulation of biological ageing. PLoS One, 11, e0146378. https://doi.org/10.1371/journal.pone.0146378
Meller, R., Pearson, A., & Simon, R. P. (2015). Dynamic changes in DNA methylation in ischemic tolerance. Frontiers in Neurology, 6, 102. https://doi.org/10.3389/fneur.2015.01012
Menke, J., Sollinger, D., Schamberger, B., Heemann, U., & Lutz, J. (2014). The effect of ischemia/reperfusion on the kidney graft. Current Opinion in Organ Transplantation, 19, 395–400. https://doi.org/10.1097/MOT.0000000000000909
Morrissy, P. E., & Monaco, A. P. (2014). Donation after circulatory death: Current practices, ongoing challenges, and potential improvements. Transplantation, 97, 258–264.
Mundt, H. M., Yard, B. A., Kramer, B. K., Bench, U., & Schnell, P. (2015). Optimized donor management and organ preservation before kidney transplantation. Transplant International, 29(9), 974–984.
Nagaraja, P., Roberts, G. W., Stephens, M., Horvath, S., Kaposzta, Z., Chavez, R., & Asderakis, A. (2015). Impact of expanded criteria variables on outcomes of kidney transplantation from donors after cardiac death. Transplantation, 99, 226–231. https://doi.org/10.1097/TP.0000000000003034
Nellimara, S., & Mossman, K. L. (2014). Extracellular dsRNA: Its function and mechanism of cellular uptake. Journal of Interferon & Cytokine Research: the Official Journal of the International Society for Interferon and Cytokine Research, 34, 419–426.
O’Neill, S., Gallagher, K., Hughes, J., Wigmore, S. J., Ross, J. A., & Harrison, E. M. (2015). Challenges in early clinical drug development for ischemia-reperfusion injury in kidney transplantation. Expert Opinion on Drug Discovery, 10, 753–762. https://doi.org/10.1517/17460441.2015.1044967
Pollard, K. M., Cauvi, D. M., Toomey, C. B., Morris, K. V., & Kono, D. H. (2013). Interferon-gamma and systemic autoimmunity. Discovery Medicine, 16, 123–131.
Rubin, L. P. (2016). Maternal and pediatric health and disease: Integrating biopsychosocial models and epigenetics. Pediatric Research, 79, 127–135. https://doi.org/10.1038/pr.2015.203
Salvadori, M., Rosso, G., & Bertoni, E. (2015). Update on ischemia-reperfusion injury in kidney transplantation: Pathogenesis and treatment. World Journal of Transplantation, 5, 52–67. https://doi.org/10.5550/wjt.v5.i2.s2
Schmidtke, G., Aichem, A., & Grootenrup, M. (2014). FAT1Oylation as a signal for pro-inflammatory degradation. Biochimica Et Biophysica Acta, 1843, 97–102. https://doi.org/10.1016/j.bbagrm.2013.01.009
Schmitt, R., Susnik, N., & Melk, A. (2015). Molecular aspects of renal senescence. Current Opinion in Organ Transplantation, 20, 412–416. https://doi.org/10.1016/j.mato.2015.05.002
Schroppel, B., & Legendre, C. (2014). Delayed kidney graft function: From mechanism to translation. Kidney International, 86, 251–258.
Shiels, P. G., McGuinness, D., Eriksson, M., Kooman, J. P., & Stenvinkel, P. (2017). The role of epigenetics in renal ageing. Nature Reviews Nephrology, 13, 471–482. https://doi.org/10.1038/nrneph.2017.78
Shiran Naffarelberg, I. E. S., Ast, G., & Kornblihtt, A. R. (2015). Regulation of alternative splicing through coupling with transcription and chromatin structure. Annual Review of Biochemistry, 84, 165–198. https://doi.org/10.1146/annurev-biochem-060614-034242
Smyth, L. J., McKay, G. J., Maxwell, A. P., & McKnight, A. J. (2014). DNA hypermethylation and DNA hypomethylation is present at different loci in chronic kidney disease. Epigenetics, 9, 366–376. https://doi.org/10.4161/epi.27161
Stengel, B., Tarver-Carr, M. E., Pove, N. R., Eberhardt, M. S., & Brancati, F. L. (2003). Lifestyle factors, obesity and the risk of chronic kidney disease. Epidemiology, 14, 479–487. https://doi.org/10.1097/01.EDE.00007141.55296.c4
Wing, M. R., Devaney, J. M., Joffe, M. M., Xie, D., Feldman, H. I., Dominic, E. A., ... Raj, D. S. (2014). DNA methylation profile associated with rapid decline in kidney function: Findings from the CRIC study. Nephrology (Carlton)Dialysis, Transplantation: Official Publication of the European Dialysis and Transplant Association - European Renal Association, 29, 864–872. https://doi.org/10.1093/ndt/gft537
SUPPORTING INFORMATION
Additional supporting information may be found online in the Supporting Information section at the end of this article. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3225,
1
],
[
3225,
8233,
2
],
[
8233,
16360,
3
],
[
16360,
21473,
4
],
[
21473,
27675,
5
],
[
27675,
29435,
6
],
[
29435,
32778,
7
],
[
32778,
36095,
8
],
[
36095,
39207,
9
],
[
39207,
41522,
10
],
[
41522,
46355,
11
],
[
46355,
50180,
12
],
[
50180,
56222,
13
],
[
56222,
61879,
14
],
[
61879,
68169,
15
],
[
68169,
75251,
16
]
]
} |
c304e86e001d084f9e2ae2375027f68adfbdf728 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 7,
"total-fallback-pages": 0,
"total-input-tokens": 20028,
"total-output-tokens": 6198,
"fieldofstudy": [
"Medicine"
]
} | Introduction
Tetanus is caused by tetanospasmin, also known as tetanus toxin, an exotoxin produced by Clostridium tetani. Neonatal tetanus (NT), also known as tetanus neonatorum, is a chief cause of infant mortality in developing countries. It is estimated that the entity led to 787,000 and 200,000 deaths worldwide in 1988 and in 2000, respectively. The latest World Health Organization estimate shows is 25,000 deaths from NT in 2018, which means an 88% reduction from the situation in 2000. Maternal tetanus and NT have not yet been eliminated by July 2019 in 12 countries. Unlike in low and middle-income countries, NT is rare and was reported in only 2 newborns from 2009 to 2015 in the United States.
Key words: Immunization, Passive; Infant, Newborn; Seizures; Tetanus; Vaccination
terminal and drumstick in appearance, and withstand heat or chemical disinfection. However, the vegetative forms can be killed by chemical disinfectants (glutaraldehyde, iodine, and hydrogen peroxide), heat, and some antimicrobials. Spores are present in soil and the gastrointestinal tract of mammals; they are nonpathogenic in earth or contaminated animal tissues until conditions are suitable for alteration to the vegetative (pathogenic) forms. Such conditions emerge when oxygen concentration decreases in the devitalized tissue by either foreign body, injury (primarily crush) or suppuration.
Although C. tetani itself does not invade the tissue, this bacterium induces illness through production of tetanospasmin. Toxigenic C. tetani strains encode tetanospasmin on a plasmid, and the toxin is produced by the proliferating C. tetani at the location of infection. The toxin acts in the central nervous system to prevent the discharge of inhibitory neurons, thereby disinhibiting the motor neurons. Unlike botulinum toxins, it has no action at the neuromuscular junctions. Its action is as a metalloprotease preventing the release of glycine and γ-aminobutyric acid. The neuromuscular endplates and motor nuclei of the central nervous system are stimulated by tetanospasmin, thus generating muscle spasm and convulsions.
The absorption and transport of tetanospasmin have 2 mechanisms. When a large amount of the toxin is produced, it spreads to the neurons via the circulation and lymphatics, causing spasm at distant sites and initially affecting the muscles with shortest neural path. The toxin undergoes transcytosis and interneural transfer, resulting in effects distant from the toxin production sites. Lockjaw and risus sardonicus are manifestations of this fast-developing illness. The inception of muscular involvement relates to the neural distance from the toxin production sites.
Nidogen-1 and nidogen-2, extracellular matrix proteins, are the receptors of tetanospasmin that reach the nervous system through the neuromuscular junctions. These proteins spread backward transsynaptically, shielded from neutralizing antitoxins, to the inhibitory synapses where the proteins bind and permanently prevent the secretion of acetylcholine. The lower motor neurons become uninhibited, increasing the tone of agonist and antagonist muscles. This increase in muscle tone produces the typical confined spasm and rigidity.
Tetanospasmin also can generate paralysis by stopping transmission at the neuromuscular junctions. Some clinical manifestations indicate the involvement of the sympathetic nervous system. The lasting effect of the toxin may be reduced by the neuronal generation of additional branches and the creation of new links among the surviving neurons.
2. Clinical manifestations
NT is a generalized tetanus that usually occurs only in infants delivered from unimmunized or inadequately immunized mothers. Causes of this entity include infection of the umbilical stump, inadequate obstetric care, delivery away from medical facilities, poor obstetrical care following delivery, and cultural routines such as application of cow stool or soil ing over the umbilical stump. The incidence of NT may be reduced by the immunization of adolescent and adult women, and local use of antimicrobials on the stump.
Weakness and failure to suck are the most frequent initial manifestations that usually occur in 5–7 days (range, 3–24 days) after birth. These symptoms continue to develop from an initial trismus and sucking difficulty to risus sardonicus within hours, and subsequently to generalized tetanic spasm, rigidity, and opisthotonus (Fig. 1). Affected infants stay awake despite the spasm.
NT has a poor prognosis with a mortality rate over 90%. This outcome can be caused by the hyper sympathetic state. The most frequent causes of mortality are apnea during the first 7 days and septicemia in 7–14 days often due to infection starting at the umbilical stump. Other complications include pneumonia, atelectasis, hemorrhage from the lung or central nervous system, renal failure, electrolyte imbalance, and laryngospasm. Younger age and lower
weight are associated with death\textsuperscript{10}. Earlier onset, leukocytosis, and longer time between first symptom and admission are also indicators of mortality and longer hospital stay\textsuperscript{10}.
Improvement is heralded by defervescence (usually within 3–7 days), reduction in spasm episodes, and improvement of muscular rigidity. Resolution may take up to 45 days. Survivors often exhibit developmental disabilities. Newborns with severe umbilical infection are prone to bacteremia due to Enterobacteriaceae, gram-negative anaerobic bacilli (e.g., \textit{Bacteroides fragilis} group), and \textit{Staphylococcus aureus}\textsuperscript{11}.
3. Diagnosis
The diagnosis is made on clinical grounds in the settings of increased risk, and by ruling out other potential reasons for tetanic spasm in newborns. The isolation of \textit{C. tetani} from the stump may not be related to production of tetanospasmin. NT can be suspected once the aforementioned initial manifestations emerge. Laboratory tests are only helpful in excluding other causes. Thus, management of NT cannot be delayed pending the laboratory findings\textsuperscript{6}.
The differential diagnosis includes pseudo–tetanus, birth injury, rabies, hypocalcemic tetany, seizure, meningitis, encephalitis, dystonic reactions to antipsychotics or other central dopamine antagonists, and strychnine poisoning\textsuperscript{12,13}.
4. Management and outcome
The managing goals are to neutralize its toxin, eradicate \textit{C. tetani}, care for wound, and offer supportive care, such as mechanical ventilation, parenteral nutrition, sedation, neuromuscular blockade, and management of autonomic dysfunction\textsuperscript{10}. Most data on this topic are based on observation or studies in adults. Survival rests on the various measures for supportive care with emphasis on intubation, administration of neuromuscular blocking agents, and providing assisted ventilation\textsuperscript{17,18}. It is difficult but necessary to administer forceful, but calculated, and supportive care that eliminates any stimulus. Affected infants to an adequate setting should be transferred early in the course of illness.
1) Neutralization of tetanospasmin
The newborn should receive human tetanus immunoglobulin (TIG) without delay\textsuperscript{15,18}. TIG should be given as a single intramuscular (IM) dose of 500 U to bind the circulating toxin. However, the exact dose has not been determined, and total doses as high as 3,000–6,000 U are also recommended. Part of the dose can be infiltrated around the identified wound. If TIG is unavailable, equine tetanus antitoxin can be administered. When this antitoxin is given, immediate and delayed reactions to the antitoxin may take place. Although intravenous (IV) immunoglobulin offers passive immunization, there is insufficient familiarity with its dose and efficacy. Since tetanus itself does not confer an antibody response, those who underwent NT should be immu–
nized with the series of diphtheria and tetanus toxoids and acellular pertussis vaccine (DTaP), following complete recovery from the active illness\(^{1,18}\).
Intrathecal administration of TIG or tetanus antitoxin enables higher concentration of TIG or antitoxin within the nervous system than IM administration. Despite the conflicting evidence on superiority of intrathecal over IM administration, a meta-analysis using 12 clinical trials involving 942 patients suggested that intrathecal administration is more beneficial than the latter\(^{19}\). A randomized controlled trial shows a reduced mortality in neonates who received intrathecal lyophilized human immunoglobulin while also receiving IM tetanus antitoxin\(^{20}\).
2) Eradication of *C. tetani*
*C. tetani* is susceptible to several antimicrobials: penicillins, cephalosporins, metronidazole, macrolides, tetracyclines, and imipenem\(^{21-23}\). It is recommended that specific antimicrobial therapy should be given with one of these modalities: per os or IV metronidazole (≤ 7 days of age, 15 mg/kg/day in divided doses; > 7 days, 30 mg/kg/day in divided doses, for 10–14 days); or IV penicillin G (100,000 U/kg/day administered at 6-hour intervals, for 10–14 days).
3) Wound care
In the presence of an umbilical stump infection, it may be necessary to obtain the adequate antimicrobial coverage against polymicrobial aerobic–anaerobic pathogens\(^{11}\). Topical application of antimicrobials or disinfectants to the umbilical cord can serve as an effective preventive measure\(^{16}\). It is important to clean the wound by removing dirt, foreign object, and the dead tissue.
4) Sedation and neuromuscular blockade
Clinical experience has demonstrated that mortality rate was lowest in those treated with combined therapy of diazepam, phenobarbital sodium, and/or chlorpromazine\(^{20}\). Omphalecetomy has also been used to eliminate the toxin production\(^{20}\).
Sedation and muscle relaxation in mild cases are achievable with diazepam. An initial dose of 0.1–0.2 mg/kg IV is given to release an acute spasm, followed by a continuous IV infusion of 15–40 mg/kg/day, titrated to control the spasm. After 5–7 days, it can be decreased by 5–10 mg/day and given orally or nasogastric\(^{10}\).
Concomitant phenobarbital is given with a loading dose of 20 mg/kg and maintenance dose of 5 mg/kg/day to reach a serum phenobarbital concentration of 30–50 mg/dL. Substantial apnea occurs in about 10% of treated patients and respiratory support should be given immediately. Assisted ventilation is vital at higher doses. Other benzodiazepines, such as lorazepam and midazolam, are also acceptable. Further sedation with phenothiazines may be needed. If spasm cannot be suppressed, therapeutic paralysis is required\(^{10}\).
Curariform agents, such as vecuronium, can induce neuromuscular blockade. Since these agents are often administered for extended period of time, the doses should be reduced slowly to prevent withdrawal symptoms\(^{10}\).
5) Management of autonomic dysfunction
Decreasing production of catecholamines can reduce the autonomic dysfunction. Labetalol (0.25–1.0 mg/minute) that possesses both α- and β-blockade can be given. Providing only β-blockade (e.g., propranolol) is not advised because it may lead to sudden death\(^{18}\). Morphine sulfate (0.5–1.0 mg/kg/hour, continuous IV infusion) is frequently used to sedate and control the autonomic dysfunction by diminishing cardiac and vascular sympathetic tone. This drug adjusts cardiac instability without compromising\(^{10}\). Only magnesium sulfate has been evaluated in a randomized clinical trial of severe tetanus\(^{25}\), and in several series for managing spasm including NT\(^{26-29}\). Thus, magnesium sulfate is given as a first-line agent in the management of NT. It relaxes the vascular smooth muscles and decreases catecholamine release from the adrenal medulla\(^{30}\) and the adrenergic nerve endings\(^{30}\). Other medications to control the autonomic dysfunction are atropine, clonidine, and...
epidural bupivacaine\textsuperscript{22,33}.
6) Other measures for supportive care
Nutritional support is important, and should be provided early to satisfy the neonates’ high caloric and protein demands\textsuperscript{34}. Enteral feeding is rarely utilized. Instead, a percutaneous endoscopic gastrotomy tube is frequently used in decreasing gastroesophageal reflux than a nasogastric tube. Central venous hyperalimentation may also be used. Heparins or other anticoagulants are given to prevent thromboembolism. Flotation bed or frequent position changes can prevent skin breakdown and peroneal nerve palsies\textsuperscript{26,27}.
7) Outcome
Many survivors from NT experience developmental handicaps. Only a few studies evaluated the long-term effects of this entity. Furthermore, actual rates of sequelae are likely to be greater. A Kenyan study shows that 20\%–40\% of survivors sustained permanent brain damage, manifested as microcephaly, and neurological, developmental, and behavioral difficulties\textsuperscript{35}. A Nigerian case series shows cerebral palsy, cognitive–developmental delay, and deafness in 20\% of survivors\textsuperscript{36}. These complications might have been caused by the hypoxia and hypoglycemia that are frequent throughout the illness.
5. Prevention
The World Health Organization recommends the administration of 6 doses (3 primary plus 3 booster ones) of tetanus toxoid–containing vaccines to each child. The 3-dose primary series should start at the age of 6 weeks, followed by 3 boosters given 4 weeks apart\textsuperscript{37}. It is recommended that the 3 boosters are given in the second year of life, and additional single booster doses at 4–7 years and at 9–15 years. Ideally, there should be at least 4 years between booster doses.
To prevent the occurrence of NT, pregnant women should receive 1 dose of reduced diphtheria and pertussis toxoids (Tdap) during each pregnancy, preferably during 27–36 weeks of gestation\textsuperscript{37,38}. Previously unvaccinated pregnant women should receive 2 doses of tetanus toxoid (at least 1 dose of Tdap) given 1 month apart, with the first dose as early as possible in pregnancy\textsuperscript{1,37}. Further doses should be given in subsequent pregnancies or at intervals of at least 1 year, up to a total of 5 doses that is considered sufficient for the lifelong protection\textsuperscript{7}.
Conclusion
NT is one of the most common life-threatening consequences of unhygienic deliveries and umbilical cord management. This entity is an indicator of inadequate availability of immunization and other maternal and neonatal care. NT–related mortality rates are high mainly when inadequate health care is provided as often occurs in developing countries. The managing goals are neutralization of toxin, eradication of \textit{C. tetani}, wound care, and specific supportive care. Mortality can be averted easily by practicing sanitary delivery and adequate cord care or by vaccinating children and women with tetanus toxoid–containing vaccines.
ORCID
Itzhak Brook (https://orcid.org/0000-0001-6068-1475)
Conflicts of interest
No potential conflicts of interest relevant to this article were reported.
Acknowledgements
No funding source relevant to this article was reported.
References
1. Centers for Disease Control and Prevention (CDC). Summary of notifiable disease–United States, 2010. MMWR Morb Mortal Wkly Rep 2012;59:1-111.
2. Thwaites CL, Beeching NJ, Newton CR. Maternal and neonatal tetanus. Lancet 2015;385:362-70.
3. World Health Organization (WHO). Maternal and neonatal tetanus elimination (MNTE) [Internet]. Geneva: WHO; c2021 [cited 2021 Jan 29]. Available from: https://www.who.int/immunization/diseases/MNTE_initiative/en/.
4. Jorgensen JH, Pfaller MA, Carroll KC; American Society for Microbiology, editors. Manual of clinical microbiology. 11th ed. Washington, DC: ASM Press; 2015. p. 940-66.
5. Dong M, Masuyer G, Stenmark P. Botulinum and tetanus neurotoxins. Annu Rev Biochem 2019;88:811-37.
6. Yen LM, Thwaites CL. Tetanus. Lancet 2019;393:1657-68.
7. Chen S. Clostridial neurotoxins: mode of substrate recognition and novel therapy development. Curr Protein Pept Sci 2014;15:490-503.
8. Hubbard K, Beske P, Lyman M, McNutt P. Functional evaluation of biological neurotoxins in networked cultures of stem cell-derived central nervous system neurons. J Vis Exp 2015;96:52361.
9. Vannini E, Caleo M, Chillemi S, Di Garbo A. Dynamical properties of LFPs from mice with unilateral injection of TeNT. Biosystems 2017;161:57-66.
10. Lam PK, Trieu HT, Lubis IN, Loan HT, Thuy TT, Wills B, et al. Prognosis of neonatal tetanus in the modern management era: an observational study in 107 Vietnamese infants. Int J Infect Dis 2015;33:7-11.
11. Brook I. Anaerobic bacteria in omphalitis. Pediatr Infect Dis 1985;4:704.
12. Pileggi DJ, Cook AM. Neuroleptic malignant syndrome. Ann Pharmacother 2016;50:973-81.
13. Hur MH, Havalad V, Clardy C. Strychnine: old remedy, silent killer. Pediatr Ann 2019;48:e205-07.
14. Vuralli D. Clinical approach to hypocalcemia in newborn period and infancy: who should be treated? Int J Pediatr 2019;2019:4318075.
15. Mine J, Taketani T, Yoshida K, Yokochi F, Kobayashi J, Maruyama K, et al. Clinical and genetic investigation of 17 Japanese patients with hyperekplexia. Dev Med Child Neurol 2015;57:372-7.
16. Amare Y. Umbilical cord care in Ethiopia and implications for behavioral change: a qualitative study. BMC Int Health Hum Rights 2014;14:12.
17. Rodrigo C, Fernando D, Rajapakse S. Pharmacological management of tetanus: an evidence-based review. Crit Care 2014;18:217.
18. American Academy of Pediatrics. Tetanus. In: Committee on Infectious Diseases; American Academy of Pediatrics; Kimberlin DW, Brady MT, Jackson MA, Long SS, editors. Red Book. 31st ed. Itasca (IL): American Academy of Pediatrics; 2018. p. 793-8.
19. Kabura L, Ilibagiza D, Menten J, Van den Ende J. Intrathecal vs. intramuscular administration of human antitetanus immunoglobulin or equine tetanus antitoxin in the treatment of tetanus: a meta-analysis. Trop Med Int Health 2006;11:1075-81.
20. Ahmad A, Qaisar I, Naeem M, Mazhar AU, Ashfaq M. Intrathecal anti-tetanus human immunoglobulin in the treatment of neonatal tetanus. J Coll Physicians Surg Pak 2011;21:539-41.
21. Wang X, Yu R, Shang X, Li J, Gu L, Rao R, et al. Multicenter study of tetanus patients in Fujian Province of China: a retrospective review of 95 cases. Biomed Res Int 2020;2020:8508547.
22. Hanif H, Anjum A, Ali N, Jamal A, Imran M, Ahmad B, et al. Isolation and antibiogram of Clostridium tetani from clinically diagnosed tetanus patients. Am J Trop Med Hyg 2015;93:752-6.
23. Onuki T, Nihonyanagi S, Nakamura M, Ide T, Hattori J, Kanoh Y, et al. Clostridium tetani isolated from patients with systemic tetanus. Kansenshogaku Zasshi 2013;87:33-8.
24. Tane N, Okuda N, Imanaka H, Nishimura M. Neuurally adjusted ventilatory assist improves patient-ventilator synchrony in a patient with tetanus and unstable diaphragmatic electrical activity. Respir Care 2015;60:e76-9.
25. Thwaites CL, Yen LM, Loan HT, Thuy TT, Thwaites GE, Stepniewska K, et al. Magnesium sulphate for treatment of severe tetanus: a randomised controlled trial. Lancet 2006;368:1436-43.
26. Trieu HT, Lubis IN, Qui PT, Yen LM, Wills B, Trieu CL, et al. Neonatal tetanus in Vietnam: comprehensive intensive care support improves mortality. J Pediatr Infect Dis Soc 2016;5:227-30.
27. Wen SC, Webb C, Miles F, Wilson E. Tetanus in New Zealand children: intensive care management of a vaccine preventable disease. J Paediatric Child Health 2016;52:1070-4.
28. Shanbag P, Mauskar A, Masavkar S. Intravenous magnesium sulphate infusion as first-line therapy in the control of spasm and muscular rigidity in childhood tetanus. Paediatr Int Child Health 2019;39:201-7.
29. Puliyel MM, Pillai R, Korula S. Intravenous magnesium sulphate infusion in the management of very severe tetanus in a child: a descriptive case report. J Trop Pediatr 2009;55:58-9.
30. Lishajko F, von Euler US. Intravenous magnesium sulphate infusion in the management of very severe tetanus in a child: a descriptive case report. J Trop Pediatr 2009;55:58-9.
31. Lishajko F. Effects of Mg2+ and Ca2+ on catecholamines, ATP, and protein from chromaffin cell granules. Acta Physiol Scand 1970;79:575-84.
noradrenaline release and uptake in adrenergic nerve granules in differential media. Acta Physiol Scand 1973;89:415-22.
32. Brook I. Current concepts in the management of Clostridium tetani infection. Expert Rev Anti Infect Ther 2008;6:327-36.
33. Rhinesmith E, Fu L. Tetanus disease, treatment, management. Pediatr Rev 2018;39:430-2.
34. Sun C, Zhao H, Lu Y, Wang Z, Xue W, Lu S. et al. Prognostic factors for generalized tetanus in adults: a retrospective study in a Chinese hospital. Am J Emerg Med 2019;37:254-9.
35. Barlow JL, Mung’Ala-Odera V, Gona J, Newton CR. Brain damage after neonatal tetanus in a rural Kenyan hospital. Trop Med Int Health 2001;6:305-8.
36. Mchil Ugwu GI, Okolugbo NE. Neonatal tetanus in Warri Niger Delta: a ten year retrospective study. Cont J Med Res 2010;4:3-7.
37. Havers FP, Moro PL, Hunter P, Hariri S, Bernstein H. Use of tetanus toxoid, reduced diphtheria toxoid, and acellular pertussis vaccines: updated recommendations of the Advisory Committee on Immunization Practices - United States, 2019. MMWR Morb Mortal Wkly Rep 2020;69:77-83.
38. Schleiss MR. Tetanus (Clostridium tetani). In: Kliegman RM, St Geme JW 3rd, Blum NJ, Shah SS, Tasker RC, Wilson KM, editors. Nelson textbook of pediatrics. 21th ed. Philadelphia (PA): Elsevier: 2015. p. 1549-51. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
792,
1
],
[
792,
4961,
2
],
[
4961,
7958,
3
],
[
7958,
12025,
4
],
[
12025,
15313,
5
],
[
15313,
20379,
6
],
[
20379,
21679,
7
]
]
} |
b2f6eb6ff6f31749c799a566519c49d8372653b1 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 19,
"total-fallback-pages": 0,
"total-input-tokens": 78270,
"total-output-tokens": 15856,
"fieldofstudy": [
"Computer Science"
]
} | Generative Adversarial Neural Operators
Md Ashiqur Rahman
Department of Computer Science
Purdue University
Manuel A. Florez
Seismological Laboratory
California Institute of Technology
Anima Anandkumar
Computing + Mathematical Sciences
California Institute of Technology
Zachary E. Ross
Seismological Laboratory
California Institute of Technology
Kamyar Azizzadenesheli
NVIDIA Corporation
Reviewed on OpenReview: https://openreview.net/forum?id=X1VzbBU6aZ
Abstract
We propose the generative adversarial neural operator (GANO), a generative model paradigm for learning probabilities on infinite-dimensional function spaces. The natural sciences and engineering are known to have many types of data that are sampled from infinite-dimensional function spaces, where classical finite-dimensional deep generative adversarial networks (GANs) may not be directly applicable. GANO generalizes the GAN framework and allows for the sampling of functions by learning push-forward operator maps in infinite-dimensional spaces. GANO consists of two main components, a generator neural operator and a discriminator neural functional. The inputs to the generator are samples of functions from a user-specified probability measure, e.g., Gaussian random field (GRF), and the generator outputs are synthetic data functions. The input to the discriminator is either a real or synthetic data function. In this work, we instantiate GANO using the Wasserstein criterion and show how the Wasserstein loss can be computed in infinite-dimensional spaces. We empirically study GANO in controlled cases where both input and output functions are samples from GRFs and compare its performance to the finite-dimensional counterpart GAN.
We empirically study the efficacy of GANO on real-world function data of volcanic activities and show its superior performance over GAN.
1 Introduction
Generative models are one of the most prominent paradigms in machine learning for analyzing unsupervised data. To date, there has been considerable success in developing deep generative models for finite-dimensional data (Goodfellow et al., 2014; Kingma & Welling, 2013; Dinh et al., 2014; Radford et al., 2015). Generative adversarial networks (GANs) are among the most successful generative models with rich theoretical and empirical developments (Arjovsky et al., 2017; Liu et al., 2017). The empirical success of GANs has been mainly within finite-dimensional data regimes; there has been relatively little progress on developing generative models for infinite-dimensional spaces—and importantly—function spaces. This is the case despite the fact that many fields of science and engineering, including seismology, computational fluid dynamics, aerodynamics,
Applications of functional data are abundant in seismology. For example, for a specific region on Earth, the base stations record data/seismograms on the surface of Earth. These receiver centers are located on irregular grids, e.g., point clouds (e.g., more stations closer to faults). The point cloud configuration also is different from region to region (Tokyo and Osaka). Moreover, due to measurement and local noise, some of the receivers are on and off in time. It means that we are dealing with functional data that are observed on irregular grids both in time and space. Moreover, when studying these functions, we aim to evaluate and query them at any spatiotemporal point. Since the governing equations are partial differential wave equations, access to the temporal and spatial derivatives reveals information about the dynamics of physical phenomena. Generative models for the mentioned wave functions allow for sampling many potential seismic behaviors of each region of Earth, facilitating the hazard study. Similarly, in weather forecasts, many recording stations on the surface of Earth are located on irregular grids (fewer stations on oceans than on lands) with different fidelity and frequency of observation. For these applications, scientists represent the weather condition as a function on a 2D sphere. This allows for evaluating weather conditions at any point on the surface of Earth and computing the gradient and momentum of the fluid dynamics. The function representation with a learned generative model allows for accurate sampling of weather forecasts and future events.
In this paper, we study the problem of generative models in function spaces. We propose generative adversarial neural operator (GANO), a deep learning-based approach that enables the learning of probabilities on function
We construct a series of controlled empirical study to assess the performance of GANs. GANs generalize the GAN paradigm to function spaces, and in particular, to separable Polish and Banach spaces. GANO, unlike traditional kernel density estimation methods, is computationally tractable, works on general spaces, and does not require the existence of a density nor the assumption of defined underlying measures for density [Rosenblatt, 1956; Parzen, 1962; Craswell, 1965]. Another line of work proposes to use neural stochastic differential equation (SDE) solver (Tzen & Raginsky, 2019) to generate temporal signal function with finite-dimensional co-domain [Kidger et al., 2021]. However, while the generated signals are infinite dimensional objects, the loss construction in the mentioned work is still for finite-dimensional spaces, for grid evaluation points, therefore, making the learned generative model implicitly yet for finite-dimensional domains, and ergo, a special case of GAN setting. Such generative models require an underlying SDE solver to solve the temporal equation and are only designed for temporal data. The same SDE structure is also used for the discriminator models, resulting in causal discriminators, for which the optimality even for GAN setting is an open problem.
GANO consists of two main components, a generator neural operator and a discriminator neural functional. GANO architecture is empowered by neural operators, which are maps between function spaces [Li et al., 2020b]. The generator neural operator receives a function sampled from a Gaussian random field (GRF) and outputs a function sample. This is in contrast to GAN, where the input is a sample from a finite-dimensional multivariate random variable and the output is a finite-dimensional object. The efficiency of traditional sampling methods from GRFs enables GANO to be considered as a computationally efficient generative model. The discriminator neural functional consists of a neural operator followed by an integral function. The discriminator receives either synthetic or real data as input and outputs a scalar. For the architecture choices in the generator, we use the efficient implementation of U-shaped neural operators (U-NO) (Rahman et al., 2022) and use Fourier integration layers, termed Fourier neural operator (FNO) (Li et al., 2020a) layers to construct push-forward maps from GRFs to the desired probability over function data. We use a similar architecture for the discriminator neural functional and use a three-layered neural network to implement the integral functional layer. For the adversarial min-max game, in particular, we instantiate the GANO framework by generalizing Wasserstein GAN (Arjovsky et al., 2017) setting to infinite dimension space. For the Wasserstein formulation, the discriminator neural functional is constrained to have a bounded norm in the infinite-dimensional space in terms of the Fréchet derivative operator. We propose how to impose this constraint in infinite dimensional space, which is invariant to the discretization. The discretization invariance property introduces one of the main differences to GAN setting where imposing the norm constraint requires hyperparameter tuning for each resolution and discretization.
The generator in GANO is a neural operator, a type of deep learning model that is resolution and discretization invariant [Kovachki et al., 2021]. It means that, the input function to the generator can be expressed with an arbitrary discretization or basis representation, yet the generated output is a function, which can be queried at any resolution or point. Similarly, the discriminator is a neural functional with input functions that can be expressed in any resolution or basis representation. These properties follow the recent advancements in operator learning that generalize neural networks that only operate on a fixed resolution [Li et al., 2020b; Kovachki et al., 2021].
The effective dimension of the output function space can be controlled by restricting the effective dimension of the GRF, e.g., by increasing the length scale of the defining covariance function. This is in contrast to GANs where the dimension of the input space controls the dimension of the output manifold. Table 1 compares the settings of GANOs and GANs. Since finite-dimensional spaces are special cases of infinite-dimensional spaces, and multi-variate Gaussian is a reduction of GRFs, then, GAN is a special case of GANO when applied on fixed grids.
We construct a series of controlled empirical study to assess the performance of GANO. To maintain full control of the data characteristics and complexity of the task at hand, we generate the data itself using GRFs of varying complexities. We show that GANO can learn probability measures on function spaces. One important example is when the data is generated from a mixture of GRFs; GANO reliably recovers the measure, while GAN collapses to a mode. In this work, we use the Wasserstein version of GAN for
---
1In finite dimensional spaces, it is conventional and standard to define density with respect to Lebesgue measures. However, in the infinite dimensional cases considered in this paper, Lebesgue measures do not exist and a density, if exists, needs to be defined with respect to a user-defined measure that the users need to argue for its relevance.
Table 1: GANs and GANOs
| Models | GANO | GAN |
|--------|------|-----|
| Input/output spaces | Function Spaces | Euclidean spaces |
| Input measure | Gaussian Random Fields | Multivariate random variables |
| Controls | length scale, variance, energy, etc. | dimension, variance, etc. |
We show that as the roughness/noisiness of the input GRF is increased, GANO properly learns to generate functions from the underlying data probability, while if the input GRF generates smooth or nearly fixed-value functions, the trained models lose the ability to properly capture the data measure.
We extend our empirical study to satellite remote sensing observations of an active volcano, where each data point is the phase of a complex-valued function defined on a 2D domain \cite{Rosen2012}. This is a real world function dataset in which each data point represents ~ millimeter-scale changes in the surface of a volcano at a spatial resolution of ~ 70 meters, measured every 12 days. This dataset constitutes a noisy and challenging function dataset for GANO and GAN training. We show that GANO learns to generate functions on par with the real dataset while GAN fails in generating these volcanic phase functions.
We release the code to generate the data sets in the first part of the empirical study. For the purpose of benchmarking, we also release the processed volcano dataset, which is ready to be deployed in future studies. We also release the implementation code along with the training procedure.
2 Related Works
The original GAN formulation can be interpreted as an adversarial game procedure in which the Jensen–Shannon divergence between a synthetic distribution, implicitly defined by a generator model, and a real data distribution is minimized \cite{Goodfellow2014}. However, models trained with a Jensen-Shannon objective function require substantial tuning, suffer from stability issues, and are notoriously difficult to scale \cite{Radford2015}. Considerable work has therefore been devoted to developing novel architectures, improving the formulation, and enhancing the theoretical understanding. In particular, the Wasserstein version of GAN allows for a more stable training scheme, is less sensitive to hyperparameter and architectural choices, and provides a loss function that correlates with output quality \cite{Arjovsky2017}. The Wasserstein formulation is often understood as an attempt to minimize the Wasserstein or Earth Mover’s distance between the synthetic and real data distributions. In \cite{Adler2018}, a rigorous theoretical extension of WGANs along with theoretically grounded choices of hyperparameters are presented, which the present paper follows. For the comparison study, we choose the Wasserstein version of GAN.
There has been limited previous work on learning densities over function spaces. These works have mainly focused on non-parametric density estimation with δ-sequences on separable Banach spaces and topological groups \cite{Rao2010,Craswell1965}. Heuristic kernel density estimation for infinite-dimensional spaces was also developed \cite{Dabo-Niang2004}. Such methods assume the existence of a density with respect to (sometimes unspecified) base measures (Lebesgue measures are undefined for infinite-dimensional spaces) and impose strong assumptions on the metric and similarity of the output spaces. Moreover, learning the density does not provide matching algorithmic sampling methods from such infinite-dimensional spaces. Since pure memorization using δ-sequences does not exploit the data structure and does not constitute a particularly appealing approach, we do not consider it an appropriate baseline for this study. For this study, we choose the GAN framework mainly due to its proximity to GANO, its vast success in many machine learning domains, and the lack of suitable methods for learning generative models in infinite dimensional spaces.
Pioneering work by \cite{Li2020b} generalized the notion of neural networks to infinite-dimensional spaces and introduced the concept of neural operators, a novel composable architecture that is able to learn mappings between functions spaces. \cite{Li2020a} showed that neural operators could be efficiently implemented as a series of convolutions performed in the Fourier domain of the input function. It has also been shown that any complex operator can be approximated by neural operators, which are compositions of linear integral operators and non-linear activation functions \cite{Kovachki2021}. Neural operators have been successfully
used for learning the solution spaces of Partial Differential Equations (PDE). FNOs have been used to learn
the solutions to the Acoustic Wave-equation in two spatial dimensions \cite{Yang et al., 2021}. Operator learning
has transformed the field of physics-informed machine learning, \cite{Li et al., 2021, 2020a} and improvements
in the underlying architecture have allowed neural operators to learn complex solutions to multiphase flow
problems \cite{Wen et al., 2021}.
A set of earlier attempts are made to develop learning methods to generate function samples using point
cloud and point-wise evaluation of sampled data functions. Along these, neural process \cite{Garnelo et al., 2018}
is motivated as a Bayesian framework to generate function samples. While motivated as a generative model
for underlying function distributions, the proposed amortized variational method aims at generating values
of point sets, rather than functions. Therefore, it may come with a few limitations to be considered as a
generative model for the underlying function distribution. This approach consists of an encoder model that
given the point evaluation data, generates a finite-dimensional vector $z$ of noise which is aimed to be close to
a prior multivariate Gaussian random variable. The noise $z$ is used as an input to an implicit neural network,
i.e., the decoder. For any $z$, the implicit neural network represents a function sample that can be queried at
any point on the domain. However, this early attempt does not learn the data distribution over functions
and comes with a few limitations.
The proposed neural process method has limitation in the way the data is perceived, lacks expressively, and
may not learn the underlying function distribution. As pointed out in the prior works \cite{Dupont et al., 2021},
the proposed neural process approach perceives the point cloud data as a set of values (no metric between
points), therefore it ignores the presence of the metric space which is noted as a crucial limitation in prior
works \cite{Dupont et al., 2021}. Furthermore, the proposed model maps a finite-dimensional $z$ vector to an
infinite dimensional space of functions. Due to this limited input dimension, the generated functions can only
cover a finite-dimensional manifold in function spaces, ergo, this method may lack the required approximation
theoretic expressively to learn generating data functions. The last limitation is the fundamental issue with
the formalism of the proposed neural process that prevents this method from being a generative model for
the data function distribution. The proposed approach learns a model to maximize the probability of observing
the values on the set of points rather than learning to generate function data. This is a major limitation of
the proposed method and therefore, undesirable for the setting of learning function distribution. For example,
consider a dataset consisting of many functions with a very low resolution (a few point evaluations) and
only one function in the dataset with super high resolution (orders of magnitude more point evaluations,
e.g., infinity). The objective of the neural process ignores the presence of all the function samples except
the high-resolution one since it aims at maximizing the probability of point evaluation rather than the
function samples. Moreover, since the formulation is Bayesian for points rather than functions, for a fixed
number of function data samples, as we increase the number of point evaluations (e.g., to infinity), the prior
will be ignored, and at the inference time, since the $z$ is drawn from the prior, even the generated points
sample would not match the data. In the appendix, we show these limitations in a set of empirical tests,
appendix A.1.
Another line of an attempt to learn generative models in function spaces proposes to accomplish the learning
task in the space of implicit neural network parameterize of the given function space \cite{Dupont et al., 2021}.
This approach proposes to train implicit neural networks to fit each data point in the data set. Ergo, for
each data point, there will be a trained implicit neural network approximating it. Then a GAN model is
trained to map input random vector $z$, e.g., drawn from a multi-variate Gaussian to the parameters of the
implicit neural network. Ergo, for each draw of $z$, this approach computes the parameters of an implicit
neural network, resulting in a function that can be queried at any point on the domain. This method requires
extensive computation due to fitting many implicit neural networks and needs extensive memory to store
these models. Furthermore, this model in the end is a map from finite-dimensional $z$ to infinite dimensional
space, limiting its cover to the space of functions. In general, the proposed approach is a generic idea that has
many favorable points as opposed to the prior works and does not have the fundamental and mathematical
limitation of point samplings in prior works of neural processes \cite{Garnelo et al., 2018}. However, the current
setting proposed in the prior work \cite{Dupont et al., 2021} comes with a few limitations that prevent this
approach to be considered as generative models of underlying function distribution.
The definition of the discriminator and the gradient penalty introduces fundamental mathematical limitations that prevent the model from learning the data distribution on function space. Given the construction of the discriminator, as the function resolution increases, e.g., the resolution goes to infinity, the proposed discriminator reduces to trivial maps, outputs a function instead of a number, and lacks the discrimination power desired for the learning task. The discriminator is implemented such that for an input function \( u \), it computes \( \sum_i W(x_i - x)u(x_i) \) where the summation is on the nearest neighbors and \( W \) is a learnable multi layered neural network. As the resolution goes to infinity, i.e., on a uniform grid, \( \sum_i W(x_i - x)u(x_i) \rightarrow W'(x)u(x) \). Therefore, repeating such pointwise operating layers many times results in a function with values at any \( x \) equal to \( W(x)u(x) \), here \( W \) is the multiplication of \( W \)'s at all the layers. This pointwise architecture lacks expressive discriminating power. Moreover, the output of the discriminator is a function instead of a single number, which is undesirable since the discriminator is expected to output a number. The second issue is that, as the resolution of the function increases, e.g., goes to infinity, the gradient penalty merges to infinity and the training process misses learning the data distribution. A similar limitation is also observed in period works that use the stochastic differential equations (Kidger et al., 2021) to generate causal in-time data. This limitation makes the resulting models to be generative models for finite-dimensional spaces.
## 3 Generative Models in Function Spaces
One of the requirements to develop a stable model that maps an input probability measure to a general probability measure defined on infinite dimensional spaces is to have an infinite-dimensional input space. In this section, we describe the setting of such maps and propose GANO, a deep learning approach for learning generative models in infinite-dimensional function spaces. We propose GANO by extending the Wasserstein GAN formulation (Gulrajani et al., 2017) with a gradient penalty term applied to an infinite-dimensional training procedure. In
3.1 GANO
Let \( A \) and \( U \) denote Polish function spaces, such that for any \( a \in A \), \( a : D_A \rightarrow \mathbb{R}^{d_A} \), and for \( u \in U \), \( u : D_U \rightarrow \mathbb{R}^{d_U} \). Let \( G \) denote a space of operators and for any operator \( G \in G \), we have \( G : A \rightarrow U \), an operator map from \( A \) to \( U \). Let \( L \) denote a space of functionals such that for any functional \( d \in L \), we have \( d : U \rightarrow \mathbb{R} \), a functional map from \( U \) to \( \mathbb{R} \).
Let \((A,\sigma(A),P_A)\) denote a probability space induced by a GRF on the function space \( A \). Following the construction of GRF, \( P_A \) is a probability measure such that for any sample \( a \sim P_A \) we have that for any finite collection of points \( \{x_i\} \) in the domain \( D_A \), the joint probability of collection \( \{a(x_i)\} \) follows a Gaussian probability. Furthermore, let \((U,\sigma(U),P_U)\) denote the probability space on the function spaces \( U \) that the real data is generated from. For a given function space \( U \), let \( \bar{U} \) denote the dual space of \( U \). When \( U \) is also a Banach space, and \( G \) is Fréchet differentiable, we define \( \partial G \) as the Fréchet derivative of \( G \). For the measure \( P_U \) and the pushforward measure of \( P_A \) under map \( G \), i.e., \( G\#P_A \), we define the Wasserstein distance as follows,
\[
W(P_U, G\#P_A) = \sup_{d: d \in L, \|\partial d(u)\|_{\bar{U}} \leq 1} E_{P_U}[d] - E_{G\#P_A}[d]
\]
For the dual space \( \bar{U} \), we have that \( Lip(d) \leq 1 \Leftrightarrow \|\partial d(u)\|_{\bar{U}} \leq 1, \forall u \in U \). Therefore, we write the constraint in the form of an extra penalty part in the objective function, i.e.,
\[
\inf_{G \in G} \sup_{d \in L} E_{P_U}[d(u)] - E_{G\#P_A}[d(u)] + \lambda \mathbb{E}$$\|\partial d\|_{\bar{U}} - 1)^2$$
\]
This relaxation is similar to the relaxation proposed in improved Wasserstein GAN (Gulrajani et al., 2017) for finite dimensional spaces which recently have been shown to be equivalent to congestion transport (Milne & Nachmani, 2022). In this objective, the constraint is induced as a soft penalty. The \( P_U \) is an uniform mixture of the data and generated data measures, i.e., \( P_U := \gamma P_P + (1-\gamma) P_U \) for \( \gamma \sim U[0,1] \), where \( U[0,1] \) is the uniform distribution in the interval \( [0,1] \). Note that, while the cost functional in Eq. (2) is well defined, showing that the learned measure is indeed an approximation of \( P_U \) remains an open problem. We address this issue empirically and perform a set of experiments that demonstrate that GANO produces diverse outputs from the data probability measure. To this end, algorithm 14 summarizes the GANO training procedure. In
When the input is provided on a regular and uniform grid, this operation can be accomplished using fast Fourier transforms. We explain neural operators architecture as maps between function spaces. We describe the input and discriminator architectures in GANO to learn maps between function spaces, and the space of reals.
Neural operators are deep learning models that are the building blocks of the generator and discriminator architectures in GANO to learn maps between function spaces, and the space of reals. Given an input function \( a \) to a neural operator \( \mathcal{G} \), we first apply a pointwise operator \( \mathcal{P} \), parameterized with a neural network \( P \), to compute \( \nu_0 \), i.e., \( \nu_0(x) = P(a(x)) \ \forall x \in \mathcal{D} \). Let \( \mathcal{D}_0 \) denote the domain functions for which \( \nu_0 \) is defined in. Given the application of \( \mathcal{P} \), we have \( \mathcal{D}_0 = \mathcal{D} \). This point-wise operator layer is followed by \( L \) integral layers. For any layer \( i \), we have,
\[
\nu_{i+1}(y) = \int_{\mathcal{D}_i} \kappa_i(x,y) \nu_i(x) d\mu_i(x) + W_i \nu_i(y) + b_i(y), \quad \forall y \in \mathcal{D}_{i+1}
\]
where \( \kappa_i \) is the kernel function, \( d\mu_i \) is the measure in the \( i \) th layer, \( W_i \) is a pointwise operator, and \( b \) is the bias function. This operation is followed by a pointwise non-linearity. The role of the pointwise operator \( W_i \), aside from decomposing the linear operator to local and global terms, is similar to the residual connection in residual neural networks [He et al., 2016]. We deploy convolution theorem to compute this integral as proposed in Fourier neural operator layer [Li et al., 2020a]. In particular, we write the \( \kappa_i(x - y) \) and compute the first part of the integral operation in the Fourier domain. Let \( \mathcal{F} \) denote the Fourier transform and \( \mathcal{F}^{-1} \) the inverse Fourier transform operations. Given a periodic function \( \mu_i \) (periodicity can be achieved by padding, a common practice in convolutional neural networks), the output of the layer is as follows,
\[
\nu_{i+1} = \mathcal{F}^{-1}(R_i \cdot (\mathcal{F} \nu_i)) + W_i \nu_i + b_i
\]
where \( R \) is the Fourier transform of \( \kappa \), and for each Fourier mode \( k \), \( R_i(k) \) is a matrix of learnable parameters. To improve computation complexity, after the Fourier transforms at each layer \( i \), we keep Fourier modes up to \( k^{\text{max}}_i \). This allows for an efficient implementation of the layer and the presence of the residual connection \( W_i \) makes sure all the Fourier components are passed to the next layer. This step, along with the residual connection \( W_i \), allows the resulting Fourier neural operators to take into account all the Fourier components at each layer.
After \( L \) above mentioned integral layers and computing \( \nu_L \) defined on the domain \( \mathcal{D}_L = \mathcal{D} \), we apply the final pointwise operator \( \mathcal{Q} \), parameterized with a neural network \( Q \). It is such that for any \( x \in \mathcal{D} \), we have \( u(x) = Q(\nu_L(x)) \). When the input function is provided on a discretized domain, e.g., on a grid, we use the Riemannian approximation of the Fourier transform to compute the Fourier modes at each layer. When the input is provided on a regular and uniform grid, this operation can be accomplished using fast and memory-efficient methods such as fast Fourier transform, resulting in efficient implementation of the corresponding neural operators.
Neural operators output functions that can be queried at any point. Furthermore, they can be applied on input functions presented in many forms, e.g., presented as weighted sum of basis functions, or presented on a discrete set of points that includes regular and irregular grids. This property of neural operators is known as discretization invariance [Kovachki et al., 2021]. In the following, we provide the definition of discretization invariance. Let \( \mathcal{D}_j \) denote a discretization (e.g., point cloud) of size \( j \) in \( \mathcal{D}_A \). We call a sequence of nested sets \( \mathcal{D}_1 \subset \mathcal{D}_2 \subset \cdots \subset \mathcal{D}_A \) a discrete refinement of \( \mathcal{D}_A \) and each \( \mathcal{D}_j \) a discretization of \( \mathcal{D}_A \) if for any \( \epsilon > 0 \),
there exists a number \( j \in \mathbb{N} \) such that,
\[
D \subseteq \bigcup_{x \in D_j} \{ y \in D_A : \| y - x \|_2 < \epsilon \}.
\]
**Definition 3.1 (Discretization Insurance)** For an operator \( G : A \rightarrow U \), where \( A \) is a set of \( m \)-valued functions, let \( D_L \in \mathbb{R}^d \) be a \( L \)-point discretization of \( D_A \), and for any \( \theta \in \Theta \), a finite dimensional parameter space, \( \hat{G} : \mathbb{R}^{Ld} \times \mathbb{R}^{Lm} \times \Theta \rightarrow U \) some map. We define the discretized uniform risk as,
\[
R(G, \hat{G}, D_L, \Theta) = \sup_{a \in A} \| \hat{G}(D_L, a|D_L) - G(a) \|_U.
\]
Given a discrete refinement \((D_j)_{j=1}^{\infty}\) of the domain \( D_A \) we say \( G \) is discretization-invariant if there exists a sequence of maps \( \hat{G}_1, \hat{G}_2, \ldots \) where \( \hat{G}_L : \mathbb{R}^{Ld} \times \mathbb{R}^{Lm} \times \Theta \rightarrow U \) such that, for any \( \theta \in \Theta \),
\[
\lim_{L \rightarrow \infty} R(G(\cdot, \theta), \hat{G}_L(\cdot, \cdot, \theta), D_L) = 0.
\]
This definition implies that, as the discretization used to present the input function becomes finer, the approximate error in the approximate operator vanishes. It has been proven that neural operators are discretization invariant deep learning models and traditional neural networks fall short in this desirable property.
**Generator** We implement the generator operator \( G \) using an eight-layered neural operator model. The \( P \) point-wise operator consists of a one-layered neural network. The \( Q \) point-wise operator consists of a two-layer neural network. The parameter vector of the generator model is denoted by \( \theta_G \). The inputs to the \( G \) model are samples generated from a GRF defined on the 2D domain of \( D = [0, 1]^2 \). The output of \( G \), and \( u \)'s are sample functions that are defined on a 2D domain. In this work, we utilize the \( U-NO \) architecture (Rahman et al., 2022) for its efficiency, stability, and robustness to the choice of hyperparameters. This architecture uses skip connections between layers and increases the dimensionality of the co-dimensions of the functions in the intermediate layers. Moreover, \( U-NO \) allows for highly parameterized models, a favorable property missing in the earlier Fourier neural operator models. In GANO, the generator neural operator model \( G \) outputs a function \( u \) given a sample function \( a \), i.e., \( u = G(a) \). Therefore, \( G \) pushes the GRF measure to a measure on the data space.
**Discriminator** The discriminator is a neural functional that consists of an eight-layer neural operator followed by an integral functional that maps the output function of the neural operator to a number in \( \mathbb{R} \). In other words, we feed an input function \( u \in U \) to the neural operator part of the discriminator to compute the intermediate function \( h \) and the output of the discriminator \( r \in \mathbb{R} \) is computed as,
\[
r := d(u) = \int \kappa_d(x)h(x)dx
\]
where the function \( \kappa_d \) is parameterized as a 3-layered fully connected neural network. Note that \( h \) is the output of the inner neural operator with \( u \) as an input, therefore, \( h \) directly depends on \( u \). The parameter vector of the discriminator model is denoted by \( \theta_d \). The function \( k_d \) constitutes the integral functional \( \int \kappa_d(x) \) which acts point-wise on its input function. This linear integral functional as the last layer is the direct generalization of the last layer of discriminators in GAN models to map a function to a number. In many GAN models, the last layer maps a high dimensional vector to a number. This step is accomplished by a vector-vector inner product. Such a product, in continuum, resembles function-function inner product, i.e., the act of linear integral functional. This also directly follows the Riesz representation theorem (Walter, 1974) stating that, under suitable construction, a linear functional (map from infinite dimension to the space of reals) can be written as a linear integral functional. Fig. 1 demonstrate the architecture of the generator \( G \) and the discriminator \( d \).
Gradient penalty In this paper, we consider the case where $D_u$ is a subset of a Euclidean space and to define the function space $U$, we consider a measure $\mu$ on $D_u$. We often use Lebesgue measure for $\mu$ in this paper. The construction of the dual space of $U$ and the computation of the Fréchet derivative used in the penalty term Eq. 2 follow after defining $\mu$. We represent the input function on a grid of $m_1 \times m_2$ (in general, it can be on any point cloud or basis function representation and following derivation follows). It allows us to use auto-grad to compute the gradient penalty for the Wasserstein loss. Following the function space definitions, the gradient penalty using the auto-grad call of $\nabla d(u)$ is implemented as $E_{x \sim \mathcal{P}} (||\nabla d(u)||_{1(x_i)}^{m_1m_2} - 1/\sqrt{m_1m_2})^2$ which is different than the finite-dimensional view in GAN. The choice of $\sqrt{m_1m_2}$ arises from the fact that we use the Lebesgue measure on $D_u$ to define the space $U$. This ratio resembles that the basis functions deployed to represent the function $u$ on the grid of $m_1 \times m_2$ need to be chosen and scaled according to the fact that the basis functions are functions with unit norms in the metric space $U$ with $L^2$ as the metric. It is important to note that since we compute the gradient with the consideration of the underlying metric space, the gradient computation using auto-grad is resolution invariant. Ergo, any resolution of choice can be used to train GANO models, fulfilling the premise of learning in infinite-dimensional spaces. Note that, for irregular grids where measures other than the deployed Lebesgue measures are used, this calculation should be adapted properly.
Algorithm 1 GANO
1: **Input:** Gradient penalty weight $\lambda$, number of discriminator updates per iteration $n_d$, number of generator updates per iteration $n_g$, number of samples per update $m$.
2: **Init:** Initialize generator parameters $\theta_G$, discriminator parameters $\theta_D$, and optimizers $Opt_D, Opt_G$.
3: for each iteration $t = 1, \ldots$ do
4: for $\tau = 1, \ldots, n_d$ do
5: Sample $\{a_i\}_1^n$ from $\mathcal{P}_A$, $\{u_i\}_1^n$ from $\mathcal{P}_U$, and $\{\gamma_i\}_1^n$ from $U[0, 1]$
6: Compute loss $L := \frac{1}{m} \sum_i^n (d(u_i) - d(G(a_i)) + \lambda(||\nabla d(u)||^2_{\gamma(\gamma_i) + (1-\gamma_i)u_i} - 1)^2)$
7: Update $\theta_D$ via a call to $Opt_D(L, \theta_D)$
8: end for
9: for $\tau = 1, \ldots, n_g$ do
10: Sample $\{a_i\}_1^n$ from $\mathcal{P}_A$
11: Compute loss $L := \frac{1}{m} \sum_i^n -d(G(a_i))$
12: Update $\theta_G$ via a call to $Opt_G(L, \theta_G)$
13: end for
14: end for
4 Experiments
In this section, we study the performance of GANO when the data is generated from a GRF. We compare the performance of GANO against GAN in this setting. The models in GANO consist of eight-layer neural operators following the architecture in [Rahman et al. 2022]. The initial lifting dimension, i.e., co-dimension is set to 16 and the number of modes is set to 20. To implement the GAN baseline model, we deploy convolutional neural networks, consisting of ten layers for the generator and half the size discriminator, and use Wasserstein loss for the training. For both models, we kept the number of parameters of the generative models roughly the same ($20M$). To train GAN models, we use GAN loss with the gradient penalty provided in the prior section. This choice is made to avoid otherwise required parameter turning for GAN loss for any resolution. We use the same grid representation of the input and output functions for the GAN and GANO studies. For training, we use Adam optimizer [Kingma & Ba 2014] and choose a 2D domain of $[0, 1]^2$ to be the domain where both input and output functions are defined on. For the empirical study, GAN is trained, optimized, and tested on a given discretization. Despite the fact that GANO can be trained and tested on any discretization, to make the comparison on par with GAN, we limit GANO experiment to the same discretization as GAN. It is worth noting that, GANO generates samples of functions that can be queried at any point and the GAN setting does not allow for it. Since GAN is not resolution invariant and does not generate function samples, it fails to be applicable to the general setting of function spaces.
We then study the effect of the roughness and smoothness of the input GRF on the quality of learning probability measures on function spaces for which we use the resolution of $\sim 64$ for each dimension. Lastly,
we study the performance of GANO on a real-world remote sensing dataset of an active volcano for which we use the resolution of \( \sim 128 \) for each dimension. This is a challenging dataset with often times very low signal-to-noise ratio. For the choice of \( \text{GRF} \), we choose the efficient implementation of Matérn based Gaussian process \cite{Vere2020} parameterized with \( \nu \), the inverse length scale.
**GRF data.** For the setting where data is generated by sampling from a GRF, we use a dataset of random functions drawn from a GRF with length scale \( \nu = 1 \) (somewhat smooth functions). We use GAN and GANO approaches to learn the data GRF. We train the generative models using the inputs sampled from the same GRF. Fig. 2a demonstrate the sample data. Subsequently, Figs. 2b and 2c demonstrate the generated samples of GAN and GANO models respectively. To analyze the quality of the generated functions, we compare the auto-correlation and histogram of point-wise function values of the generated data and the true data, Fig. 2. The \( x \)-axis in the histogram plots denote the values the functions take. The \( x \)-axis in the auto-correlation plots denote the positional distance of the points on the domain \( D_\nu \) for which we compute the auto-correlation. We observe that GANO properly recovers the statistics of the data GRF in terms of auto-correlation, Fig. 2d, 2e, and the histogram of the generated function values, Figs. 2f and 2g. We observe that, while the GAN approach provides smoother-looking functions, the functional statistics fail to be exact.
**Mixture of GRFs data.** For this experiment, we aim to learn to generate data from a mixture of GRFs. The training data is generated with an equal chance from either a GRF with a fixed mean function of \( 1 \) or \(-1\), and both with \( \nu = 1 \). We use GAN and GANO approaches to learn the data probability measure, where the input functions are sampled from a mean zero GRF with \( \nu = 1 \), Fig. 3. Fig. 2a demonstrate the sample data. Subsequently, Figs. 2b and 2c demonstrate the generated samples of GAN and GANO models respectively. The auto-correlation and histogram of point-wise function values of generated data and the true data are provided in Figs. 2d, 2e, 2f, and 2g. As we observe, GANO properly recovers the statistics of the data GRF in terms of functional statistics of auto-correlation and histogram. Similar to the previous experiment, we observe that the GAN approach provides smoother-looking functions, but in terms of the functional statistics, it drastically underperforms GANO.
In the previous two experiments, we observed that GANO enables us to learn measures on function spaces and generate samples that match the functional statistics of the underlying data. In the following, we examine the importance of the choice of input GRF on the performance of GANO.
**GANO and the length scale of the input GRF.** In GANO, when the GRF input to the generative model is very smooth (compared with the output GRF), we expect the generator to fail to learn a proper map. We expect this to be the case because the input lacks sufficient high-frequency components, and this smoothness
Figure 3: The input function sample is GRF and the data is generated from a mixture of GRFs. (a) The samples of data from a mixture of GRFs. (b) The samples of generated data from GAN model. (c) The samples of generated data from GANO model. (d) GAN Auto correlation. (e) GAN histogram. (f) GANO Auto correlation. (g) GANO histogram
Figure 4: GANO trained on smooth data ($\tau = 5$) with rougher input GRF ($\tau = 7$)
prevents the generator from generating high-frequency and rough output functions. On the contrary, we expect that when the input GRF is much rougher than the data GRF and contains many high-frequency components, the generator would have an easier task to generate output functions. Therefore, the length scale and smoothness of the input GRF can play a role in regularizing GANO model, a very similar role that the dimension of the input multivariate Gaussian plays in the GAN approach.
We first show that when the input GRF is rougher ($\tau = 7$) and contains more high frequency components than the output GRF ($\tau = 5$), GANO successfully learns to generate samples with similar statistic of data GRF, Fig. 4.
When the output and input GRF are identical measures ($\tau = 5$), GANO still successfully learns to generate samples with similar statistics of the data GRF, Fig. However, this setting requires more delicate hyper parameter tuning and requires more training epochs to converge. It is worth noting that, with proper choices of the spaces, an identity map may also be a solution.
Lastly, when the input GRF is smoother ($\tau = 3$) than the functions samples in the output data GRF ($\tau = 5$), the generative model fails to recover higher order statistics, including the auto correlation Fig. In this experiment the input function is much simpler than the output functions. This study suggest that, when the real function data is very complex, very noisy, contains varying high frequency components, and poses high
entropy, it is crucial to provide the generator with on par GRF. On the contrary, when the function data at hand poses smoother behavior, a smooth GRF suffices for training a generator.
**Inputs and outputs of the generator in GANO are functions** The GANO framework is based on neural operators that are discretization invariant maps between functions spaces $[3, 11]$. The inputs to the generator neural operator model in GANO are functions and following the discretization invariance property of such models, these input functions can be provided to the generator in any discretization, and in particular in any mesh, grid, and resolution. In addition, the generator outputs functions, therefore, by definition, the outputs can be queried at any point in the domain. In the following empirical study, we demonstrate these properties of neural operators. We train the GANO models on one resolution and test the trained generator in another resolution. In particular, we consider a setting where for the training, the input GRF samples (with $\tau = 5$) are presented on a $64 \times 64$ grid on the two-dimensional domain. Moreover, the training data functions are draws from a GRF, with the same parameter as the input GRF, and samples are represented on the same $64 \times 64$ grid.
After training, we assess the above-mentioned properties of neural operators. We double the resolution of the input GRF samples and present them in a $128 \times 128$ grid. We provide these higher-resolution inputs to the generator to generate output functions. We evaluate the generated functions on a finer grid of $128 \times 128$. Figure 7 demonstrates the result of the study. Figure (c) represents high-resolution data, and figure (e) represents the generated samples on the higher-resolution input and query points. Figures (g) and (i) demonstrate the histogram and auto-correlation of the higher-resolution data and higher-resolution generated samples. This study expresses that neural operators can take inputs at any resolution and the
Figure 7: We train GANO on a function data set of resolution $64 \times 64$. The data samples are generated using a GRF ($\tau = 5$). The input GRF ($\tau = 5$) sample functions are also represented on a $64 \times 64$ grid. The generator neural operator takes a function as an input and outputs a function. To demonstrate this fact, we test the trained generative model on a different resolution. We change the resolution of the input function to a higher resolution of $128 \times 128$ and query the generated function samples on a higher resolution of $128 \times 128$. Figure (c) represents high-resolution data, and figure (e) represents the generated samples on the higher-resolution input and query points. Figures (g) and (i) demonstrate the histogram and auto-correlation of the higher-resolution data and higher-resolution generated samples. This study expresses that neural operators can take inputs at any resolution and the output function can be queried at any point in the domain. Furthermore, despite the fact that the model has never seen high-resolution data during the training, it can generate statistically matching samples of high resolution.
output function can be queried at any point in the domain. Furthermore, despite the fact that the model has never seen high-resolution data during the training, it can generate statistically matching samples of high resolution. These are desirable properties of the GANO framework, as the first generative model on function spaces. Please note that, for this empirical study, we use smaller models in GANO in order to fit the high-resolution data to the present GPU machines. In particular, we reduce the number of layers to 5, the co-dimension to 8, and the number of modes to 10. These choices for the smaller model did not alter the performance of the trained generator.
Volcano deformation signals in InSAR data. Interferometric Synthetic Aperture Radar (InSAR) is a remote sensing technology used to measure deformation of Earth’s surface, often in response to volcanic eruptions, earthquakes, or subsidence due to excessive groundwater extraction. In InSAR, a radar signal is emitted from satellites or various types of aircraft and echoes are recorded. Changes in these echoes over time (as measured by repeat flyovers) can be used to precisely measure the amount that a point on the surface moves between repeats. The most common form of InSAR data is the interferogram, which
We produce a dataset of interferograms, each in a grid of 128 × 128, from the Sentinel-1 satellites covering the Long Valley Caldera, which is an active volcano near Mammoth Lakes, California. We processed the InSAR functions/images, publicly provided by the European Space Agency, from 2014-Nov to 2022-Mar, covering an area around Long Valley Caldera (approximately 250 by 160 km wide) using the InSAR Scientific Computing Environment [Rosen et al. 2012]. The stack of SAR functions is co-registered with pure geometry (precise orbits and digital elevation model) and the network-based enhanced spectral diversity approach. Then, we pair each function (277 in total) with its three nearest neighbors in time to form 783 initial interferograms with pixel spacing of 70 m. Finally, we subset each interferogram into six windows non-overlapping windows of 128 × 128 grid. Examples of real samples are shown in Fig. 8a.
We train GANO on the entire dataset of 4096 interferograms. Generated samples are shown in Fig. 8c, where it is clear that many of the complexities of this dataset have been learned. One of the types of noise in interferograms results from decorrelation of the radar signal between repeat flyovers, and in the most extreme case, can lead to a stochastic process that is random uniform on [−π, π] that covers part or all of the image. GANO is able to learn an effective operator that approximates this complex behavior. We quantitatively
evaluate the quality of the learned samples using circular statistics, which is necessary since these functions are angular-valued. Analogously to traditional random variables, there are moments of circular random variables. For a collection of $N$ random angular variables, $\{\theta_i\}_{i=1}^N$, define $z_p = \sum_{j} e^{ip\theta_j}$, where $i = \sqrt{-1}$. Then, $R_p = |z_p|/N$ and $\phi_p = \arg(z_p)$. The circular variance is then given by $\sigma = 1 - R_1$, and the circular skewness is given by, $s = \frac{R_2 \sin(\phi_2 - 2\phi_1)}{(1-R_1)^{1/2}}$. Figs. 8d and 8e show the performance of GANO w.r.t circular variance and circular skewness. These results demonstrate the suitability of GANO framework on learning complex probabilities on function spaces and emphasizes the data efficiency of this framework.
For the comparison study, we train a GAN model on the same data set. Despite extensive hyperparameter tuning, the GAN model fails to learn to generate proper samples of functions. Fig. 8b demonstrates samples generated using a trained GAN model. The generated samples do not resemble the true samples, neither perceptually nor with respect to the circular variance and skewness Fig. 8d,8e. This study establishes the importance of learning the generative model directly in function spaces using global kernel integration instead of local kernels.
5 Conclusion
We propose GANO, a generative adversarial learning approach for learning probabilities on function spaces and generating samples of functions. GANO generalizes GAN, an established and powerful method for learning generative models on finite-dimensional samples. GANO framework consists of two models, a generator operator and a discriminator functional. We use the neural operator framework to directly model the generator and deploy the ideas from neural operators, and propose a new deep learning paradigm, namely neural functional, for the discriminator. We empirically show that the GANO framework is suitable for dealing with function spaces. We show that the input to the generative model can be chosen to be a GRF for which the length scale controls the diversity of the pushed measure. We release the code, package, datasets, and the results of this study for future reproducibility.
Acknowledgments
The authors would like to thank the TMLR reviewers and the action editor for their constructive comments. Part of this research is developed when K. Azizzadenesheli was with the Purdue University. A. Anandkumar is supported in part by Bren endowed chair.
References
Jonas Adler and Sebastian Lunz. Banach wasserstein gan. Advances in Neural Information Processing Systems, 31, 2018.
Martin Arjovsky, Soumith Chintala, and Léon Bottou. Wasserstein generative adversarial networks. In International conference on machine learning, pp. 214–223. PMLR, 2017.
KJ Craswell. Density estimation in a topological group. The Annals of Mathematical Statistics, 36(3):1047–1048, 1965.
Sophie Dabo-Niang. Kernel density estimator in an infinite-dimensional space with a rate of convergence in the case of diffusion process. Applied mathematics letters, 17(4):381–386, 2004.
Laurent Dinh, David Krueger, and Yoshua Bengio. Nice: Non-linear independent components estimation. arXiv preprint arXiv:1410.8516, 2014.
Emilien Dupont, Yee Whye Teh, and Arnaud Doucet. Generative models as distributions of functions. arXiv preprint arXiv:2102.04776, 2021.
Marta Garnelo, Jonathan Schwarz, Dan Rosenbaum, Fabio Viola, Danilo J Rezende, SM Eslami, and Yee Whye Teh. Neural processes. arXiv preprint arXiv:1807.01622, 2018.
Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. Advances in neural information processing systems, 27, 2014.
Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron C Courville. Improved training of wasserstein gans. Advances in neural information processing systems, 30, 2017.
Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016.
Patrick Kidger, James Foster, Xuechen Li, and Terry J Lyons. Neural sdes as infinite-dimensional gans. In International Conference on Machine Learning, pp. 5453–5463. PMLR, 2021.
Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014.
Diederik P Kingma and Max Welling. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013.
Nikola Kovachki, Zongyi Li, Burigede Liu, Kamyar Azizzadenesheli, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Neural operator: Learning maps between function spaces. arXiv preprint arXiv:2108.08481, 2021.
Zongyi Li, Nikola Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Fourier neural operator for parametric partial differential equations. arXiv preprint arXiv:2010.08895, 2020a.
Zongyi Li, Nikola Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Neural operator: Graph kernel network for partial differential equations. arXiv preprint arXiv:2003.03485, 2020b.
Zongyi Li, Hongkai Zheng, Nikola Kovachki, David Jin, Haoxuan Chen, Burigede Liu, Kamyar Azizzadenesheli, and Anima Anandkumar. Physics-informed neural operator for learning partial differential equations. arXiv preprint arXiv:2111.03794, 2021.
Shuang Liu, Olivier Bousquet, and Kamalika Chaudhuri. Approximation and convergence properties of generative adversarial learning. Advances in Neural Information Processing Systems, 30, 2017.
Tristan Milne and Adrian I Nachman. Wasserstein gans with gradient penalty compute congested transport. In Conference on Learning Theory, pp. 103–129. PMLR, 2022.
Nicholas H Nelsen and Andrew M Stuart. The random feature model for input-output maps between banach spaces. SIAM Journal on Scientific Computing, 43(5):A3212–A3243, 2021.
Emanuel Parzen. On estimation of a probability density function and mode. The annals of mathematical statistics, 33(3):1065–1076, 1962.
Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks. arXiv preprint arXiv:1511.06434, November 2015. doi: 10.48550/arXiv.1511.06434. URL https://arxiv.org/abs/1511.06434v2
Md Ashiqur Rahman, Zachary E Ross, and Kamyar Azizzadenesheli. U-no: U-shaped neural operators. arXiv preprint arXiv:2204.11127, 2022.
BLS Prakasa Rao. Nonparametric density estimation for functional data by delta sequences. Brazilian Journal of Probability and Statistics, 24(3):468–478, 2010.
Paul A Rosen, Eric Gurrola, Gian Franco Sacco, and Howard Zebker. The insar scientific computing environment. In EUSAR 2012; 9th European conference on synthetic aperture radar, pp. 730–733. VDE, 2012.
Murray Rosenblatt. Remarks on Some Nonparametric Estimates of a Density Function. The Annals of Mathematical Statistics, 27(3):832 – 837, 1956. doi: 10.1214/aoms/1177728190. URL https://doi.org/10.1214/aoms/1177728190
Belinda Tzen and Maxim Raginsky. Neural stochastic differential equations: Deep latent gaussian models in the diffusion limit. arXiv preprint arXiv:1905.09883, 2019.
Rudin Walter. Real and complex analysis, 1974.
Gege Wen, Zongyi Li, Kamyar Azizzadenesheli, Anima Anandkumar, and Sally M Benson. U-fno—an enhanced fourier neural operator based-deep learning model for multiphase flow. arXiv preprint arXiv:2109.03697, 2021.
Yan Yang, Angela F Gao, Jorge C Castellanos, Zachary E Ross, Kamyar Azizzadenesheli, and Robert W Clayton. Seismic wave propagation and inversion with neural operators. The Seismic Record, 1(3):126–134, 2021.
A Appendix
A.1 Neural Process and generative functions
In this section, we expand the discussion on neural process approach (Garnelo et al., 2018) and show a few problems it may have in learning generative models for data function distribution. For a given function sample \( u \), let \( (x_i, y_i)_{i=1}^n \) denote its point evaluation representation, where for any \( x_i \), a collocation point, \( y_i \) is the point evaluation of the function at point \( x_i \). For this construction, the following is the evidence lower-bound objective function proposed in neural process,
\[
\log p (\{y_i \} | \{x_i \}) \geq E_{q(z|x_i,y_i)} \left[ \sum_i \log p(y_i | z, x_i) + \log \frac{p(z)}{q(z | \{x_i, y_i\}_{i=1}^n)} \right]
\]
where the method learns the encoder \( q \) and a decoder map from \((z, x)\) to mean and variance of \( p(y_i | z, x) \). This objective function maximize the probability \( y_i \)'s, and does not give a formulation to learn the data function distribution.
Let’s consider a trivial setting where the function distribution is a Dirac, meaning that, the data set consists of many repetitions of a single function. For example, consider the function \( u(x) = 0.5 \) on the interval
The dataset consists of many sample functions, all are the mentioned \( u \). Following the Bayesian and variation form of this objective, when \( n \) is small, e.g., \( n = 100 \), training this model results in a function distribution around the input function but does not collapse on the data function, Figure 9b.
For a reasonable generative model, we expect that, if we increase the function resolution, e.g., taking it to infinity, the function distribution learning approach to get better at learning a sensible generative model. However, in the heuristic neural process approach, as we increase the function resolution, the first term in the objective function dominates (goes to negative infinity), and \( q(z|\{x_i, y_i\}_n^i) \) no longer can be replaced by \( p(z) \) in the inference time. Figure 9c shows data function with 10000 point evaluations, and when neural process is trained on 10000 resolution input function, Figure 9d shows that the generated sample become more off and do not capture much about the function distribution.
Figure 9: The sample generation in the neural process does not collapse to the data samples. As a Bayesian approach for point evaluations rather functions, as the number of point evaluations increases, the training process ignores the prior, resulting in an inconsistent model in the inference time.
To elaborate more on the inconsistency of the proposed heuristic neural processes model and its lack of foundations on learning function data distribution, we construct the following additional toy example. Consider a similar setting as previous example with function distribution as a mixture of two Diracs on \( u(x) = 0.5 \) and \( u(x) = 0.7 \). In this setting, the data set consists of repetitions of \( u(x) = 0.5 \) and \( u(x) = 0.7 \) functions. A sensible function distribution learning method should be able to learn this mixture. Let us consider the setting where the resolution of \( u(x) = 0.7 \) is 2 (2 point evaluations), and the resolution of \( u(x) = 0.5 \) is 100. Per our above discussion on the lack of motivation behind the objection function proposed in neural process approach and the fact that this approach aims to capture point evaluation distribution, training on such mixture of data results in model that totally ignores the data samples of \( u(x) = 0.7 \). Figure 10a shows the data set and Figure 10b shows the learned model totally ignores the function samples \( u(x) = 0.7 \), only because they contain fewer point evaluations. We expect that, as the resolution of \( u(x) = 0.5 \) increases, neural process...
approach even misses learning \( u(x) = 0.5 \). Figure 10c shows the data sets where \( u(x) = 0.5 \) has a resolution of 10000 and Figure 10d shows training on such data set does not learn the function distribution.
To this end, we concluded that the heuristic approach proposed in neural process does not learn function distribution, rather attempts to learn point evaluation, and it is not clear how this approach can be helpful to learn distribution of function data.\(^2\)
\(^2\)For the empirical study on neural processes, we used the implementation provided in https://github.com/EmilienDupont/neural-processes | 2025-03-07T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2732,
1
],
[
2732,
4554,
2
],
[
4554,
9935,
3
],
[
9935,
14497,
4
],
[
14497,
19772,
5
],
[
19772,
24893,
6
],
[
24893,
29315,
7
],
[
29315,
33588,
8
],
[
33588,
38214,
9
],
[
38214,
41428,
10
],
[
41428,
43385,
11
],
[
43385,
45421,
12
],
[
45421,
47873,
13
],
[
47873,
49329,
14
],
[
49329,
51880,
15
],
[
51880,
55084,
16
],
[
55084,
58392,
17
],
[
58392,
60995,
18
],
[
60995,
61614,
19
]
]
} |
3ccd2f6aacd2d6e14d65192f024a6802f3b56ea8 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 6,
"total-fallback-pages": 0,
"total-input-tokens": 18751,
"total-output-tokens": 7433,
"fieldofstudy": [
"Environmental Science",
"Medicine"
]
} | A Systematic Review of Radon Investigations Related to Public Exposure in Iran
Meghdad Pirsaheb, Farid Najafi, Touba Khosravi, Lida Hemati
1Environmental Health Engineering Department, Kermanshah University of Medical Sciences, Kermanshah, IR Iran
2Epidemiology Research Center, Kermanshah University of Medical Sciences, Kermanshah, IR Iran
3Environmental Epidemiology research center, Public Health Faculty, Kermanshah University of Medical Sciences, Kermanshah, Kermanshah, IR Iran
4Environmental Engineering Department, Public Health Faculty, Kermanshah University of Medical Sciences, Kermanshah, Kermanshah, Iran
*Corresponding Author: Lida Hemati, Environmental Health Engineering Department, Kermanshah University of Medical Sciences, Kermanshah, IR Iran. Tel: +98-9187240367, Fax: +98-8318263048, E-mail: [email protected]
Received: January 13, 2013; Revised: May 5, 2013; Accepted: July 1, 2013
Background: The main sources of radiation exposure of all living organisms including humans are natural. In fact, radon and its decay products are the cause of 50% of the total dose that is derived from natural sources. Because of the significant health hazards of radon gas, its levels are widely monitored throughout the world. Accordingly, considerable researches have also been carried out in Iran.
Objectives: The aim of this research is a systematic review of the most recent studies associated with evaluation of radon gas levels in Iran. The main emphasis of this study was on public exposure to radon gas.
Materials and Methods: The most important route of exposure to such radiation is indoor places. In this investigation measurement of radon in water resources, tap water, indoor places and exhalation of radon from building material, the major sources of indoor radon gas emission, were considered.
Results: Significantly high levels of radon gas were found mostly in water and residential buildings.
Conclusions: It conclusion with regard to the study of building materials, granite stone and adobe coverings cannot be recommended for construction purposes.
Keywords: Radon; Public exposure; Water supply; Building material; Iran
1. Background
Natural radiation is the main source of radiation in the surrounding environment. It is estimated that the contribution of natural sources of radiation exposure to humans is approximately 90% (1). Accordingly, radon and its decay products are the cause of 50% of the total dose originating from natural sources (2).
Radon (Radon-222), a trace element, is colorless, odorless and tasteless. It is a natural radioactive gas that is derived from uranium decay present in rocks and soil (2). Radon is highly soluble in the water, thus the radon gas present in the underlying rocky bed can easily pass through the soil and rocks, inevitably entering underground water sources. Therefore, soil and, various types of rock in the earth’s crust and underground water are the main sources of radon gas propagation (3). The concentration of radon in the outdoors is much lower than in the indoor places, where subsequent radioactivity has been found to increase. The significant aspect of radon at high concentrations can be dangerous for humans, mostly leading to lung cancer (2, 4). The alpha-emitting particles of radon gas that enter tissues through water, food and inhalation can have negative biological effects on such organs (5-8).
Because of the significant health hazards associated with radon, its concentrations are widely monitored throughout the world. Considerable research has been carried out in Iran, mainly in the northern region, which has high background radioactivity. As reported by the United Nations Scientific Committee on Atomic Radiation the coastal city of Ramsar has been shown to have the highest levels of radioactivity when compared to other inhabited areas of the world (2). The studies that have been conducted on radon have been ineffective. Hence an investigation and review of all researches that have been carried out so far will be highly valuable to any future studies on radon gas in Iran.
2. Objectives
This paper presents a review of the studies that have been carried out on the measurement radon in water, indoor places and building materials used in Iran, all of which have a critical and important role in the exposure of the general population to radiation.
3. Materials and Methods
Initially, an investigation of the available research regarding the measurement of radon was carried out through an online literature search using the Pub-Med, Science Direct, ISI Web of Knowledge Scopus, Medlib, SID, Iranian Research Institute for Information Science and Technology (IRANDOC) and IRANMEDEX databases. The keywords “radon, Iran” or “radon measurement in Iran”, used in this search. All studies associated with the measurement of radon until August 2012, in both the English and Persian languages, were collected and investigated. The relevance of articles have been screened by two independent reviewer initially. In total number of 1455 investigated papers, 176 titles and abstracts were recognized as potentially appropriate. Papers which contained other natural radioactivity were excluded. The main purpose of this study was on public exposure to radon gas. The most important route of exposure to such radiation is indoor places.
Table 1. Concentration Values of Radon in Water Resources Reported in Literature
| S. No | Reference | Technique Used | Location | Results |
|-------|-----------|----------------|----------|---------|
| 1 | Sohrabi et al. (9) | Liquid scintillation counting technique | Domestic water supplies, including ground and surface waters, in 23 provincial centers | The minimum and maximum mean concentrations of $^{222}$Rn in ground water were, respectively, 7.9 ± 4.5 kBq m$^{-3}$ in Sanandaj and 46.5 ± 11.5 kBq m$^{-3}$ in Tehran with an overall national mean value of 21 ± 8.3 kBq m$^{-3}$. The $^{222}$Rn concentrations in surface waters ranged from less than 1 to 7 kBq m$^{-3}$ with a mean value of 3.9 ± 1.9 kBq m$^{-3}$. The mean concentration of $^{222}$Rn in tap water in different parts of Tehran is 3.8 ± 1.1 kBq m$^{-3}$ |
| 2 | Alirezazadeh (10) | Liquid scintillation counting technique | 71 water samples, including 49 groundwater, 10 surface water, and 12 tap water samples in Tehran | The mean $^{222}$Rn concentrations in groundwater and surface water supplies were 46.40±11.50 and 2.50±1.20 Bq/L, respectively. The mean radon concentration in tap water was 3.70±0.94 Bq/L. The annual total effective dose to adults due to waterborne radon in Tehran was estimated to be about 10 µSv |
| 3 | Beitollahi et al. (11) | Liquid scintillation counting technique | Five hot springs called `Abe-garm-e-Mahallat`, located in the central part of Iran | $^{222}$Rn concentrations ranged from 145±37 to 273±98 Bq/L |
| 4 | Mowlavi et al. (12) | PRASSI system | 14 drinking water sources in the Ramsar region | All of the water supplies have radon concentrations greater than 10 kBq/L as normal level |
| 5 | Binesh et al. (13) | PRASSI system | 8 springs, flume and rivers water sources of Kelardasht-Kalenov region | 775 Samples Have Radon Concentration Gather Than 10 Bq/L |
| 6 | Binesh et al. (14) | PRASSI system | 15 drinkable water sources in Shirvan region | The results show that 33.3 samples have radon concentration higher than 10 kBq/m$^3$ as normal level, and radon in 3 samples are near normal level(15) |
| 7 | Binesh et al. (15) | PRASSI system | 120 samples of drinking water, river & spring water of Zoshik, Abdreh & Slandiz regions (Mashhad) | 315.83 samples have radon concentration gather than 10 Bq/L |
| 8 | Binesh et al. (16) | PRASSI system | 120 water samples of Water sources of 3 northern regions(Ramsar, Sadatshar and Javaherdeh regions) | 9.17 samples have radon concentration higher than 11Bq/L as normal level. radon induced the total annual effective dose greater than 0.1 mSv/y in 31.7 samples |
| 9 | Forozani Gh and Soori Gh. (17) | PRASSI system | 15 Drinking water sources in the Toyskeran region | %33.3 samples have radon concentration higher than 10 Bq/L as normal level |
| 10 | Binesh and Arabshahi (18) | PRASSI system | 120 samples of drinking, springs and rivers water sources of northwest regions of Mashhad city | The average value of radon concentration was 30.2±5.1 Bq/m$^3$. The dose rate due to radon, radium and their progenies received by the population in the studied location between 0.1-0.5 mSv y$^{-1}$.314.67 samples have radon concentration higher than 11 Bq/L as normal level |
| 11 | Pourhabib et al. (19) | PRASSI system | 43 water samples of the Sadatshar and Javaherdeh regions | %9.3 samples have radon concentration higher than 11Bq/L as normal level |
| 12 | Karimdust & Ardehili (20) | RAD7 Radon detector | Hot springs of Sarein | Radon concentrations in water varied from 212 Bq/m$^3$to 3890 Bq/m$^3$ |
### Table 2. Indoor Radon Concentration Values Reported in Literature
| S. No | Reference | Technique Used | Location | Results |
|-------|-----------|----------------|----------|---------|
| 1 | Sohrabi and Solaymanian (21) | The AEOI passive radon diffusion dosimeters | 206 randomly selected houses in some regions of Iran including Ramsar, Tehran, Babolsar and Gonabad | The mean radon levels in Ramsar, Tehran, Babolsar and Gonabad were determined to be respectively 578, 80, 88 and 84 Bq.m⁻³, leading to average effective dose equivalents of 17.6, 2.44, 2.68, 2.56 mSv/y. |
| 2 | Karamdoust et al. (22) | Passive radon measurement method | Dwellings (mostly guest-houses) around the hot springs in the north-west of Iran | The measurements were carried out during winter for a period of 2.5 months. The radon levels in the majority of dwellings have been higher than 100 Bq/m³ and in two cases exceeded the limitation value recommended by ICRP for future homes (i.e. 200 Bq/m³). |
| 3 | Sohrabi and Babapouran (23) | AEOI passive radon diffusion chambers | 500 houses in 12 regions of Ramsar | The annual mean effective equivalent dose (Ē) in different regions due to ²²²Rn ranges from 2.48 to 71.74 mSv with maximum levels up to 640 mSv determined in one house in Talesh Mahalleh. |
| 4 | Hadad K et al.,(24) | Solid state nuclear track detectors (SSNTD) with CR-39 polycarbonate and PRASSI Portable radon Gas Surveyor | A total of 1124 samplers in Lahijan, Ardabil, Sar-Ein and Namin | The average radon concentration during the year in Lahijan, Ardabil, Sar-Ein and Namin were 163, 240, 168, 124 and 133 Bq/m³, respectively. These concentrations give rise to annual effective doses of 1.43 mSv/y for Lahijan and 5.00 mSv/y for Ardabil. The maximum recorded concentration was 2386 Bq/m³ during winter in Ardabil and the minimum concentration was 55 Bq/m³ during spring in Lahijan. |
| 5 | Bouzarjomehri and Ehrampoosh (25) | A portable radon gas surveyor | 84 dwellings basement from various regions of Yazd | Radon concentrations of the basements were between 5.55 to 747.4 Bq/m³ with mean of 137.36 Bq/m³. More than 30% of the basements had radon concentration more than 148 Bq/m³ (EPA guideline). |
| 6 | Ranjbar et al. (26) | Radon working level meter, based on the Environmental Protection Agency (EPA) conditions | 68 houses, which cover 0.23% of the total houses in Rafsanjan city | The concentration in 51.2% of the houses is more than the acceptable value. |
| 7 | Binesh et al. (27) | PRASSI system | 40 apartments in Mashhad city | The result demonstrates about 35% of the apartments have a radon level lower than the normal level (148 Bq/m³) and more than 65% have high radon concentration. |
| 8 | Gillmore and Jabarivasal (28) | CR-39 alpha track-etch detectors | 30 Dwellings in Hamadan, western Iran, significantly, built on permeable alluvial fan deposits | The indoor radon levels varied from 4 (i.e. Below the lower limit of detection for the method) to 364 Bq/m³ with a mean value of 86 Bq/m³. The effective dose equivalent to the population in Hamadan estimates from this study to be in the region of 2.7 mSv/y which is above the guidelines for dose to a member of the public of 1 mSv/y suggested by the International Commission on Radiological Protection (ICRP) in 1993. |
| 9 | Hadadi (29) | Radon diffusion dosimeters | 196 Tabriz houses | This study showed that the average radon concentration were 39 Bq/m³ in the houses. At different floors & different construction material the average effective dose equivalent of lung tissue was 0.97 mSv/y. |
| 10 | Mowlavi et al. (30) | PRASSI system | 150 apartments in Mashhad city | About 94.7% of the apartments had radon concentration less than 100 Bq/m³. |
| 11 | Hadad et al. (31) | Solid State Nuclear Track Detectors (SSNTD), CR-39 polycarbonate films | Dwelling of Shiraz | Annual average indoor radon concentration for the survey period was 94 ± 52 Bq/m³. The calculated mean annual effective doses in basements and different floors were less than the lower limit recommended action level by ICRP. |
In this investigation, the measurement of radon in water resources, tap water, indoor places, and exhalation of radon from building material, as the major sources of radon gas emission, was considered. The review articles were included as potentially appropriate. The results of multiple publications were considered only once, and the newest reference of it is reported. All of papers were monitored for inclusion by two reviewers separately. Disagreement was resolved during discussion. Due to inclusion and exclusion criteria, the full text of total 27 papers are included as potentially appropriate.
(12 studies on radon measurement of water resources, 11 studies on indoor radon measurement and 4 studies on radon exhalation from building material) were screened in this systematic review. All investigated articles in this review were cross-sectional studies. Accordingly, information on the methods measurement, sample size and sampling locations and the subsequent results of the selected studies are indicated in Tables 1 and 2. Radon concentration values in the water resources, indoor places and the amount of radon exhaled from construction materials are classified in above mentioned tables. In this paper, the studies involving the mining industry workers and occupational exposure to radon were not reviewed.
4. Results
A review of researches involving the evaluation of radon levels in Iran indicated that of the studies measured radon in water resources, especially in areas with high radiation levels and dwellings.
All investigate studies in this paper were description and outcome has not been reported as OR or RR. Therefore, the publication bias could not be prove or disprove by the Statistical and graphical methods.
The general overview of research regarding measurement of radon levels in the water resources in Iran (Table 1) showed that the concentration of radon was higher than the values recommended by USA Environmental Protection Agency (8). Thus such, high levels of radon in the water supplies must be reduced before reaching the consumers. In fact, a comparison of radon levels in surface and ground water sources indicated that the concentration of radon in the groundwater sources and hot springs were much higher than those of surface water sources. The highest levels of radon, in the range of 145 ± 37 to 2731 (Bq/L) were reported from the Mahallat hot spring in northern region. Therefore, using certain methods such as mixing groundwater with surface water in large reservoirs can reduce the radon activity to acceptable limits. In regions where only groundwater sources are used, aerating the water before consumption lead to a dramatic reduction of radon levels in drinking water.
A review of households radon studies (Table 2) also shows that the radon concentrations, especially in areas with high background radiation, and basements during the cold seasons, were not at desirable levels, thus emphasizing the need for improving ventilation. The highest indoor radon concentration level was reported in a house in the Tallesh Mahalleh of Ramsar, that received doses of 640 mSv/y. Studies also indicated that the indoor radon levels in apartments were higher than those in houses, due mainly to the presence of unsuitable ventilation and air conditioning in the former.
A review of radon activity in building materials, demonstrated that the use of local granite and stone in areas with high background radiation areas should not be recommended. Investigation of Iranian compressed granite depicted that the amount of $^{222}$Ra present in this material was $1.605 \pm 0.055$ kBq/m$^3$, which is 4 times higher than the level recommended by the International commission on radiological protection (ICRP) (200 - 600 Bq/m$^3$). According to the conclusion drawn by this review, the levels of uranium and radium present in granite are high, and can thus significantly increase radon levels in areas where it has been used. Hence, the use of such stones in buildings can become health hazard necessitating the need for a solution (32). Furthermore, a study on 10 pieces of granite stone used in building construction showed that $^{226}$Ra and $^{232}$Th cause emission of radon from granite stones (33). Other investigations also showed that the rate of radon release from building materials used in Ramsar areas with high background radiation. A linear correlation coefficient between the emission of radon and radium concentration was estimated as 0.90. The results showed that the radon exhalation rate and radium content in a number of local stones used in the basement were at high intensity levels. These were the main sources of radon and gamma emission from uranium (34). A study of different covering materials has indicated that individuals resident in houses covered with materials such as plaster and wallpaper received average annual doses less than those living in houses with walls covered with wood and plastic paint, ($P = 0.05$). Therefore, the use of wallpaper and plaster to cover the parts of residential buildings is recommended. In addition, a comparison the type of materials used in buildings presented a significant difference ($P = 0.05$) in radon gas concentration levels between buildings, that were constructed with adobe and concrete. However, there were no significant differences between buildings constructed with concrete and brick material, and those that were built with adobe and brick buildings (31). In another study, results showed that houses constructed with adobe emitted radon gas more than buildings made of concrete, brick and plaster ($P < 0.05$), while plaster walls emitted the lowest levels of radon gas ($P < 0.01$) (25). According to other researchers, adobe has a highly porous structure, thus scattering high levels of radon gas.
5. Discussion
It should be noted that the most studies in Iran are in high background radiation areas, which it could be lead to overestimation of results presentation. On the other the only published articles were examined in this study also the number of investigated studies were limited, probably many studies are published and it is so far a study was part of gray literature.
In this paper, studies involving the evaluation of radon levels in Iran with an emphasis on public exposure rather than occupational were reviewed. Most of the reported was associated with measurement of radon in water supplies; especially in high background radiation areas, such as Ramsar, and also indoor places with health hazards.
Consequently, the highest values of radon gas in the water resources were reported in the hot springs of Mahallat and a house in the Talesh Mahalleh area of Ramsar. Most of the studies have proposed methods to reduce radon levels in water supplies, prior to its consumption by humans. In many of the studies involving the measurement of radon in residential buildings, the calculated dose received by humans was higher than the recommended amount. As for studies involving assessment of radiation levels in buildings, construction materials such as granite stone and adobe are thus not recommended. However, it should be noted that the evaluation of radon levels carried out in Iran, took place in a period of less than one year, where seasonal changes were not reported.
Acknowledgments
The authors wish to acknowledge the invaluable cooperation and support from the Deputy of the School of Public Health, Kermanshah University of Medical Sciences for facilitating the issue of this paper.
Authors’ Contribution
None declared
Financial Disclosure
Authors declare no conflict of interests.
Funding/Support
None declared.
References
1. Ahmad N, Khatibeh AJ. Comparative studies of indoor radon concentration levels in Jordan using CR-39 based bag and cup dosimeters. Health Phys. 1998;75(1):60–2.
2. Unscarce Sources. Effects of Ionizing Radiation. United Nations, New York. 2000;453–487.
3. Nakhlí A. Environment and Radon Gas. J. Nucl. Energy. 1997;39:19–29.
4. Khan AJ. A study of indoor radon levels in Indian dwellings, influencing factors and lung cancer risks. Radiation Measurements. 2000;32(2):87–92.
5. Introduction. Annals of the ICRP. 1993;23(3):1–5.
6. Baykara Oktay, Dogru Mahmut. Measurements of radon and uranium concentration in water and soil samples from East Anatolian Active Fault Systems (Turkey). Radiation Measurements. 2006;41(3):362–367.
7. Sohrabi M, Alireza Zadeh N, Tajik Ahmadi H. A survey of 222Rn concentrations in domestic water supplies of Iran. Health Phys. 1998;74(4):417–21.
8. Alireza Zadeh N. Radon concentrations in public water supplies in Tehran and evaluation of radiation dose. Iran J Radiat Res. 2005;3(2):79–83.
9. Bettollahi M, Ghassemi Nejad M, Esmaeili A, Dunker R. Radiological studies in the hot spring studies in the hot spring region of Mahallat, Central Iran. Radiation Prot Dosimetry. 2007;123(4):505–8.
10. Mowlavi AA, Shabbahrami A, Binesh A. Dose evaluation and measurement of radon concentration in some drinking water sources of the Ramsar region in Iran. Isotopes Environ Health Stud. 2009;45(3):269–72.
11. Binesh Alireza, Mowlavi Ali Asghar, Shabbahrami Amrolah, Najafabadi Reza Izadi. Determination of radon and radium content of drinking water supplies of Kelarash-t-Kalavan region in Iran. Asian Journal of Experimental Sciences. 2010;23(1):385–388.
12. Arabshahi Hadi. Radon and Radium Measurement in Drinkable Water Supplies of Shirvan Region in Iran by Prassi System. Archives of Applied Science Research. 2010;2.
13. Binesh A, Mohammad S, Mowlavi AA, Parvareh P. Measuring Radon and Radium Concentrations in 120 Samples of Drinking Water Sources, Springs and Rivers of Shandiz, Zoshik and Abdereh Regions. Environmental Research. 2011;15(3):5–8.
14. Arabshahi Hadi. Radioactivity and dose assessment of heavy radioactive pollution, radon and radium from water sources of 3 northern regions in Iran. International Journal of Physical science. 2016;6.
15. Forozeni Ghasem, Soori Ghasem. Study on radon and radium concentrations in drinking water in west region of Iran. arXiv preprint arXiv:1006.3646. 2011.
16. Binesh A, Arabshahi H. Radon and radium concentrations in 120 samples of drinking springs and rivers water sources of northwest regions of Mashhad. Int. J. Water Resour. Environ. Eng. 2012;3(2):16–21.
17. Binesh A, Pourhabib Z, Arabshahi H, Mohammad S. Radium and Radium Measurement in Water Supplies of Sadatshahr and Java-herdeh Regions in Iran.
18. Karimoudast S, Ardeshiri L. The environmental impact of radon emitted from hot springs of Sarein (a touristic city northwestern Iran), World Appl Sci J. 2010;10(8):930–935.
19. Sohrabi M, Solaymanian AR. Indoor radon level measurements in some regions of Iran. International Journal of Radiation Applications and Instrumentation. Part D. Nuclear Tracks and Radiation Measurements. 1988;8(1–4):1–5.
20. Karimoudast NA, Afarideh H, Hatami P. Radon measurement in dwellings around the hot spring in the north west of Iran. Nuclear Tracks and Radiation Measurements. 1993;22(1–4):351–353.
21. Sohrabi Mehdi, Babapourou Moghzan, editors. New public dose assessment from internal and external exposures in low-and elevated-level natural radiation areas of Ramsar, Iran; International Congress Series; Elsevier; pp. 169–174.
22. Hadad Kamal, Douladzar R, Mehdizadeh S. Indoor radon monitoring in Northern Iran using passive and active measurements. J Environ Radiat. 2007;99(1):39–52.
23. Bouzarijomehri F, Ehrampoosh MH. Radon level in dwellings basement of Yazd-Iran. Iranian Journal of Radiation Research. 2008;6(6):134–144.
24. Askari Hassan Ranjbar, Rahimi Mojtaba, Negarestani Ali. The investigation and measurement of the radon gas working level inside buildings in Rasanjan. International Journal of Low Radiation. 2008;5(2):928–931.
25. Binesh A, Mowlavi AA, Mohammad S, Parvareh P, editors. Indoor radon measurement in some apartments of Mashhad city (Iran); World Congress on Medical Physics and Biomedical Engineering, September 7–12, 2009, Munich, Germany; 2009; Springer; pp. 189–190.
26. Gillmore GK, Jabbarzadeh H. A reconnaissance study of radon concentrations in Hamadan city, Iran. Nat. Hazards Earth Syst. Sci. 2010;10(4):835–863.
27. Haddadi Gholamhassan, Assessment of Radon level in dwellings of Tabriz. Journal of Fasa University of Medical Sciences. 2012;13(3):19-31.
28. Mowlavi Ali Asghar, Fornasier Maria Rosa, de Denaro Mario. Indoor radon measurement and effective dose assessment of 150 apartments in Mashhad, Iran. Environmental monitoring and assessment. 2012;184(2):1085–1098.
29. Hadad H, Kazimdavoud MR, HASEM TALEHNOORI M. Evaluation of a radon survey in Shiraz-Iran using developed passive measurement method. IRANIAN JOURNAL OF RADIATION RESEARCH (IJRR). 2011.
30. Ranjbar AH, Barghordar H. Activity concentration of 222Rn released from crushed granite stones of Iranian origin. Journal of Radioanalytical and Nuclear Chemistry. 2006;260(1):43–46.
31. Nassiri P, Ebrahimi H, Jafari Shalkouhi P. Evaluation of radon exhalation rate from granite stone. J Sci Ind Res. 2011;70(2):230–231.
32. Bavaregine F, Fatubahdi N, Vahabi Moghaddam M, Vasheghani Farahani M, Moradi M, Babakhani A. Radon exhalation rate and
natural radionuclide content in building materials of high background areas of Ramsar, Iran. J Environ Radioact. 2013;117:36-40.
33. Baradaran-Ghaflarokhi MILAD, Faghihi R, Karami M, Siavashpour Z, Owji H. The Effects of Internal Wall Covering Materials on Hazards of Indoor Radon Concentrations in Houses of Iran. IRANIAN RED CRESCENT MEDICAL JOURNAL (IRCMJ). 2011;13(9):637-640.
34. Qiuju Guo, Bo Chen, Quanfu Sun., editors. A pilot survey on indoor radon and thoron progeny in Yangjiang, China.; International Congress Series.; 2005; Elsevier; pp. 313-314. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4376,
1
],
[
4376,
8977,
2
],
[
8977,
13709,
3
],
[
13709,
19658,
4
],
[
19658,
26335,
5
],
[
26335,
26896,
6
]
]
} |
dd66db02910dc91eba7f6c0d6d2fdcdfe0c2b6b5 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 10,
"total-fallback-pages": 0,
"total-input-tokens": 36494,
"total-output-tokens": 16886,
"fieldofstudy": [
"Environmental Science"
]
} | Competition and Predation in Soil Fungivorous Microarthropods Using Stable Isotope Ratio Mass Spectrometry
Felicity V. Crotty† and Sina M. Adl*
Department of Soil Science, University of Saskatchewan, Saskatoon, SK, Canada
The soil food web is often described as having three main energy channels: root, bacterial and fungal. Here we provide quantitative data using a sensitive stable isotope ratio mass spectrometry procedure with microcosms on species interactions in the fungal pathway. We measured $^{15}$N and $^{13}$C enrichment in microarthropods through grazing rare isotope enriched fungal mycelia. Experimental treatments were various combinations of 1, 2, 3, 4 microarthropods species. We used three fungivores (the collembolan Lepidocyrtus curvicollis, the Astigmata Tyrophagus putrescentiae, the Oribatida Oribatula tibialis), and the Mesostigmata predator Hypoaspis acquilifer. We collected individuals of each species separately, as well as their feces, and molt where available. All three fungivorous microarthropods consumed significantly more than their own body weight per day. The three fungivores differed in their consumption of the mycelium as it was not equally palatable to each. The Mesostigmata predator Hypoaspis also differed in its microarthropod prey preference. In multiple species combinations microarthropod behavioral interactions modified consumption and predation rates. Our selection of mites of different sizes, with varied preference for the mycelium, combined with differing predation rates on each mite, demonstrate that even three trophic level interactions with only five interacting species are not predictably simple. The interpretation of the stable isotope results and consumed-excreted weights indicate that: (a) behavior and microscopic observations should not be ignored in competition-predation interactions, and (b) functional guilds can take advantage of more diverse food opportunities. The reality of mixed diets complicates functional guild assignments that are reflected in $^{15}$N and $^{13}$C isotope levels at natural abundances in the environment. Microcosm experiments with this sensitive technique can help decipher the interpretation of rare isotope natural abundance values, as well as providing measured consumption, growth, and excretion rate values for modeling soil food web interactions.
Keywords: fungivory, microarthropods, nutrient cycling, stable isotopes, trophic interactions
INTRODUCTION
Soil food web models typically recognize three main routes energy can flow through the soil – roots, the fungal or bacterial pathway (Moore et al., 2005). An early study of soil food web structure found that both top-down control and bottom-up control existed simultaneously and were necessary for ecosystem stability (de Ruiter et al., 1995). Increased predation rate on prey, or of grazing consumption (e.g., on bacteria lawn, a mycelium, or pasture) decreases the biomass and abundance of the species consumed, while increasing the abundance of the consumers over time. In food webs, the increased abundance or biomass of the consumer trophic level, relative to the consumed trophic level is called a trophic cascade. It makes allusion to the appearance in graphs of successive reduced and increased biomass in the trophic levels in the system. These increasing and decreasing consumption rates oscillate through time and ripple through generations and are one cause of population dynamics. An interesting and useful parameter for ecologists is to find out at what relative consumption rates trophic cascades occur, for they cannot occur under elevated food resources (Polis et al., 2000). It has been postulated that the microarthropods, particularly the Collembola and Oribatida, have a more significant role in the fungal pathway (Faber, 1991; Schneider and Maraun, 2005; Jonas et al., 2007). Several additional papers discussed fungivory by soil microarthropods (e.g., Maraun et al., 2003; Schneider et al., 2004; Pollierer et al., 2009; Thiele-Bruhn et al., 2012), but few tried to quantify rates of hyphae consumption except for individual species, such as Collembola (Jonas et al., 2007), while others focused on feeding preferences (e.g., Ruess et al., 2005; Koukol et al., 2009; Semenina and Tiunov, 2011). However, little is known regarding the quantities of fungal mycelium consumed and assimilated by microarthropods species when in monoculture, in competition, or with predation pressure. Likewise there is little insight into the microarthropods foraging behavior within the soil itself, or their response to the distribution and quality of food resources (Adl, 2003; Hassall et al., 2006). Considering there are 10^3–10^5 m of hyphae in one gram of fertile soil (Leake et al., 2003), this is a large biomass contribution to soil nutrient turnover that ought to be studied more. The difficulty has been in designing microcosms with compatible species, and to develop a technique with sufficient sensitivity to work with the small number of individuals in a microcosm experiment.
One of the main methods of differentiating between feeding preferences and measuring consumption rates is the use of stable isotopes (Tiunov, 2007; Semenina and Tiunov, 2011), as this can provide a time-integrated measure of the trophic position of soil animals (Pollierer et al., 2009). An organisms’ tissues exhibit a fixed isotopic enrichment in relation to their diet; therefore stable isotopes can be measured to assess the assimilation rates over the long-term (Peterson and Fry, 1987). Introducing an enriched food source as a pulse provides a mechanism to accurately monitor consumption in a controlled environment (Bradford et al., 2012). The addition of a substrate with a distinct isotopic signature can be traced into newly synthesized compounds and tissues (Elfrstrand et al., 2008). The introduction of a food source which is enriched above natural abundance levels with stable isotopes into the soil food web can elucidate feeding interactions as they are happening. In ecology, shifts in ^13C indicate change in diet, and the ^13C natural abundance value indicates an equilibrium average of the various food sources. In contrast ^15N values have been used to indicate position in trophic level in a food web – the higher the δ^15N value, the higher the trophic level. It also shows categorically that the food source has been consumed and assimilated.
However, Maraun et al. (2011) advised there was a need to investigate the individual feeding channels, but there is a difficulty differentiating basal resources using stable isotopes at natural abundance. The problem with the interpretation of natural abundance studies with the rare stable isotopes ^13C and ^15N is that it provides an equilibrium and average value of all the sources of food consumed (Moore et al., 2004; Schneider et al., 2004; Pollierer et al., 2009; Schneider and Maraun, 2009). These studies indicated there was sometimes poor resolution between some trophic compartments without more detailed studies. To address these concerns with natural abundance field data, it is necessary to turn toward microcosms to make specific measurements. In these microcosm studies, it is important to consider both competition and predation and not just single species consumption rates, as single species results might not be representative of field data (Wardle and Yeates, 1993). A number of studies employing stable isotopes focused on quantifying the bacterial energy pathway (e.g., Murray et al., 2009; Crotty et al., 2011b, 2012a), and here this study focuses on the fungal pathway. Tracking the energy flux through the fungal-feeding channel within soil food webs is more complicated than the bacterial channel (Crotty et al., 2012b). Fungi are easily shredded during sampling, and are not easily re-introduced in situ, highlighting a need to validate the use of enriched isotope tracers in microcosm settings to study fungal energy pathways.
Energy flux dynamics in multitrophic interactions food webs provide the data to understand biodiversity-ecosystem functioning. Yet, there are few studies portraying this, despite the crucial link these studies would provide in linking trophic interactions with ecosystem function (Barnes et al., 2018). The overall aim of the experiments presented here, were to study the effects of species interactions (competition and predation) on the rate of consumption and assimilation as determined by isotope ratio mass spectrometry (IRMS). To do this the rate of consumption of saprotrophic fungal biomass by soil microarthropods needed to be determined. We tested the hypothesis that mycelium assimilation by the fungivores, measured as body mass enrichment, will be the same for each of the species used. We tested the hypothesis that in the presence of a predator, there is evidence of a trophic cascade, so that consumption of the mycelium is reduced. We also tested the hypothesis that the δ^13C and δ^15N isotopic signatures of the predator would be affected by its prey’s isotopic composition.
MATERIALS AND METHODS
General Methods
Soil cores were obtained from a wooded area of Point Pleasant Park in Halifax, Nova Scotia, Canada (latitude 44.621876277, longitude −63.5711053). The microarthropods were extracted from the soil using a Tullgren funnel heat extraction (5 mm mesh) into prepared microcosms. The microcosms were plastic vessels (54 mm diameter, 60 mm height) with a ~10 mm layer of plaster of Paris mixed with activated charcoal (5:1 by weight, 2 parts powder to 1 part water). Numerous species were found but only some were successfully cultured on fungal mycelia or baker’s yeast in the necessary quantities for the microcosm experiments. Laboratory cultures were established of Lepidocyrtus curvicollis (Collembola: Entomobryidae), Oribatula tibialis (Acari: Oribatidae), Tyrophagus putrescentiae (Acari: Astigmata: Acaridae), and Hypoaspis acquisfer (Acari: Mesostigmata: Laelapidae). Microarthropods were grown on a mixed diet of baker’s yeast (added every few days) and ground dried nettle as organic matter (added weekly), in darkness at room temperature (about 24–26°C).
Fungal cultures were obtained from the same soil as the microarthropods. Saprotrophic fungi were initially isolated by placing peds of soil on potato dextrose agar (1.5%) with Rose Bengal (0.1%) to suppress bacterial contamination. Sub-cultures of the hyphae, which grew out of the soil, were taken in a series to obtain single fungal species in culture that was identified as an ascomycete by microscopy. The minimal medium was composed of 7 g Na₂HPO₄, 3 g KH₂PO₄, 0.12 g MgSO₄, 0.011 g CaCl₂, 0.5 g NaCl per liter deionised water, with 1.2% w/v technical agar (number 3), which were combined and sterilized (autoclaved at 121°C for 15 min), before the addition of 2.5 and 1 g, respectively, of filter sterilized ¹³C-glucose (C₆H₁₂O₆) and ¹⁵N-ammonium chloride (NH₄Cl) until the agar substrate was enriched to 99.9% (Crotty et al., 2011a). Reference control cultures were also grown prepared in the same way but the glucose and ammonium chloride were at natural abundance of C and N isotopes. Nitrocellulose disks (45 mm dia. Bio-Rad, Ontario, Canada) were placed on top of the agar and inoculated with the mycelium. Fungal hyphae grow without penetrating the agar to ease the transfer of mycelium to microcosms.
The feeding experiments were conducted in Petri dish (diameter 6 cm) with a <5 mm layer of plaster of Paris and activated charcoal at the bottom. In preparation for each experiment, microarthropods were transferred from culture to a new Petri dish without food, and starved for 48 h before the feeding experiments commenced. Fungal biomass was harvested by lifting the nitrocellulose disks from the agar, and each placed in the center of a fresh Petri dish microcosms. Each microcosm received about 20 mg wet weight of mycelium (weight measured and recorded for stable isotope analysis). We verified in controls the fungal mycelium transferred to the experiment Petri plate microcosms did not grow, as there was no substrate.
In preliminary trials to determine optimal experimental conditions, we established that 20 mg of mycelium in the Petri dish area (approximately a quarter of the nitrocellulose disk) could maintain a stable environment for about 20 Lepidocyrtus or Tyrophagus, with up to 10 Oribatula; more than 10 individuals of the predator Hypoaspis was too crowded in one dish, especially without prey, but 5–10 with prey provided stable microcosms. Oribatula was slow-growing on the cultured fungus, and this species clearly was not a preferred food source. We kept this species so as to provide a range of grazing fungus for the competition and predation experiments. These preliminary growth experiments on labeled and un-labeled mycelium were conducted for up to 21 days to establish optimal durations. We established that optimal multispecies microcosm incubations were 5–7 days; shorter incubations (<5 days) showed more variability and lower enrichment; longer periods of up to 2–3 weeks allow recycling nutrients by coprophagy, losing too many prey to the predator, and over-grazing the mycelium. We calculated the amount of consumption using the Petersen and Luxton (1982) equation, Assimilation = consumption – feces.
The experiments were terminated by removing the microarthropods from each dish, grouped by species, killed by submersion in 100% ethanol for <1 min, and samples were processed immediately for analysis by IRMS. The microarthropod filled capsules were dried at 65°C for 48 h, re-weighed, and ¹³C/¹²C and ¹⁵N/¹⁴N ratios measured using Costech ECS4040 elemental analyzer coupled to a Delta V Advantage mass spectrometer (Thermo Fisher Scientific, Finnegan, Germany) with a Conflo IV interface (Thermo Fisher Scientific, Finnegan, Germany), following the methods described in Crotty et al. (2013) for low mass samples.
The fecal pellets and exoskeletons remaining in the microcosms were also collected, counted, added to pre-weighed tin capsules, dried at 65°C for 48 h, re-weighed, and ¹³C/¹²C and ¹⁵N/¹⁴N ratios measured as described above. The amount of fungi consumed by the microarthropods over the incubation period was calculated and expressed as dry weight per individual per day.
Competition Experiments With and Without a Predator
This experiment focused on the effect competition or predation had on feeding rates, using the level of isotope enrichment acquired by each species as a proxy for consumption. Stable isotope enriched mycelium was added to ten different treatments (each treatment replicated five times) in microcosms. The treatments were, four single species monocultures (Lepidocyrtus, Tyrophagus, Oribatula, and Hypoaspis); one with two fungivores in competition (Lepidocyrtus and Oribatula); one with a fungivore and a predator (Lepidocyrtus and Hypoaspis); one with three fungivores (Lepidocyrtus, Tyrophagus, and Oribatula); two microcosm treatments with two fungivores and a predator (either Lepidocyrtus, Oribatula and Hypoaspis or Lepidocyrtus, Tyrophagus, and Hypoaspis); and the final microcosm treatment which had all faunal species in competition / predation with each other (Lepidocyrtus, Tyrophagus, Oribatula, and Hypoaspis) Microcosms were incubated for 7 days, as established above. The number of individuals at the end of each treatment were counted. Following incubation, the Petri dishes were prepared for analysis by mass spectrometry as described above.
Statistical Analysis
Microarthropod effects on mycelium consumption, fecal pellet biomass and isotopic composition where assessed by a general analysis of variance. Where applicable multiple comparison within tables of means were made using the Student-Newman-Keuls test (Sokal and Rohlf, 1995), with the experiment-wise type-1 error rate set at 5%. In order for the mycelium samples to be within the enrichment thresholds of the mass spectrometer a companion reference sample was used to dilute the isotope signal, consequently the isotopic results were back-calculated using the equations of Hauck and Bremner (1976) to ascertain original enrichment levels. All data were analyzed using GenStat (14th Edition, Payne et al., 2011) and are presented as mean ± standard error, or SED (standard error of the difference), unless otherwise stated.
RESULTS
Consumption of Mycelium by Microarthropods
The mean ingestion (consumption rate) and fecal pellet excretion rates over a 21 day single microarthropod incubation with unlabelled mycelium and with $^{13}$C and $^{15}$N enriched mycelium (except without Oribatula as they were too few) were compared (Table 1). There were no significant differences in microarthropod weight when comparing unlabelled mycelium with highly enriched mycelium between the two experiments after 21 days ($P = 0.073$). There were no significant differences between the consumption rate of Lepidocyrtus, Oribatula, and Tyrophagus ($P = 0.352$). The weight of each fecal pellet produced per organism was: Oribatula ($0.04 ± 0.012 \mu g$), Tyrophagus ($0.33 ± 0.032 \mu g$), and for Lepidocyrtus ($0.10 ± 0.020 \mu g$). There were no significant differences in the fecal pellet weight excreted per organism per day in the isotope enriched cultures compared to the natural abundance measurements ($P = 0.15$). There was a significant difference between the fungivorous organisms, with Tyrophagus excreting significantly greater quantities than Lepidocyrtus and Oribatula ($P = 0.011$). The feces of both species showed a lower level of enrichment in $^{13}$C than the mycelium at the end of the experiment (Lepidocyrtus APE 3.23 ($± 0.506$) $P = 0.033$; Tyrophagus APE 2.15 ($± 0.475$) $P = 0.010$) compared to mycelium APE 5.08 ($± 0.573$), although not in APE $^{15}$N [Lepidocyrtus APE 18.54 ($± 2.780$) $P = 0.375$; Tyrophagus APE 23.99 ($± 3.054$) $P = 0.438$] compared to mycelium APE 22.15 ($± 3.449$). These results taken together (Table 1) indicate that the three species of fungivores do not assimilate the same amount of the mycelium consumed, and that differences accumulate over time between the enrichment of fecal pellets and body mass among the fungivores.
The mesostigmatid predator Hynoaspis on its own with mycelium but without prey lost weight and did not excrete fecal pellets. The three fungivorous microarthropods consumed mycelium at about the same rate $\sim 5 \mu g$ individual$^{-1}$ day$^{-1}$, and the differences were not statistically significant. The oribatid Oribatula was the least efficient on this fungus mycelium, excreting $\sim 31$% of the ingested fungus while consuming more than three times its body-weight daily. The astigmatid Tyrophagus was the most efficient, excreting only about $11\%$ of the ingested mycelium, but it also consumed about three times its body-weight. The collembolan ingested about $1\%$ times its body weight daily and excreted about $16\%$ of the ingested mycelium. Lepidocyrtus was the only organism to produce exoskeletons in enough numbers to collect and weigh. Each collembolan exoskeleton weighed $1.10 \mu g$ ($± 0.244$) and equated to each individual collembolan shedding its exoskeleton $1.17$ ($± 0.096$) times within a 21 day period.
Analyzing the organisms, their fecal pellets, and the mycelium that remained at the end of the incubations, there were no significant differences between the mycelium $^{13}$C and $^{15}$N atom% levels and those in the body mass of Lepidocyrtus, Oribatula, or Tyrophagus $P > 0.1$ for all fungivores.
Survival
In competition experiments with and without the predator, the number of surviving individuals at the end of the incubation in each microcosm is informative (Table 2). Individuals are lost to predation and poor adaptation to the incubation conditions. The predator Hypoaspis survival was about 75% in monoculture without prey. Over half of the Lepidocyrtus and Tyrophagus survived in monoculture on the mycelium. Surprisingly, even though the mycelium did not provide a good food source to Oribatula, there was only a 12% loss of individuals in monoculture (Table 2). The highest survival rates of predators was when there were two or more prey species, although only some combinations are statistically significant. The collembolan Lepidocyrtus did not ameliorate survival of the predator. The Oribatula losses were greatest when competing for the mycelium with Lepidocyrtus, with indiscernible predation effect. Tyrophagus losses were least with all three fungivores present and the predator. The Lepidocyrtus microcosms show a distinct increase in losses in the presence of the predator, but this observation is nuanced by the observation (above) that the collembolan did not improve survival of the predator. Visually, at the microscope, collembolan were better at escaping, and it was more effort for the predator to capture one.
Species Interactions Effect on Consumption
The results of the competition-predation microcosm incubation experiments with stable isotope enrichment are summarized in Figure 1. For all these graphs the microarthropod $^{13}$C and $^{15}$N value at the beginning of each experiment is at the 0 mark of both axis, and we plot the enrichment transferred to the microarthropods in atom percent enrichment (APE). It does not represent the natural abundance value in nature, as these organisms were cultivated in the laboratory for about 2 years.
Lepidocyrtus (Figure 1A) consumption of mycelium differed significantly depending on competition/predation for both isotopes (APE $^{13}$C $P = 0.010$ and APE $^{15}$N $P = 0.018$).
Consumption was numerically greatest when incubated as a single microcosm. When *Lepidocyrtus* was incubated solely with the predator *Hypoaspis* (black triangle, Figure 1A) there was very little change in $^{13}$C or $^{15}$N value, indicating consumption did not decrease significantly, although survival reduced to 33% (Table 2). When in competition with *Oribatula* (open circle, Figure 1A) or when the microcosms contained two competitors, the value is not statistically different than when it was on its own, indicating that *Oribatula* (and *Tyrophagus*) in fact did not pose a competition threat to *Lepidocyrtus*. When *Lepidocyrtus* is incubated with the predator *Hypoaspis* the $^{13}$C value is significantly reduced (less consumption) but not the $^{15}$N value (no change in trophic level, and as results above). When both competitors are present (clear triangle), the three fungivores interfere with each other’s grazing so as to reduce the $^{15}$N value and the $^{13}$C value compared to *Lepidocyrtus* alone or with *Oribatula*. In the presence of the predator and *Oribatula* as the only competitor, *Lepidocyrtus* (shaded square) values are not statistically different than with *Oribatula* alone, indicating it is relatively unbothered by predation in this combination. In the absence of the predator but both competitors (clear triangle), there is an intermediate amount of grazing. It clearly shows the difference in competition imposed on *Lepidocyrtus* by *Tyrophagus* but not by *Oribatula*. Without *Oribatula* (clear circle), but with competition from *Tyrophagus* and predation from *Hypoaspis* (clear square), the $^{13}$C value is reduced but not the $^{15}$N value when compared to *Lepidocyrtus* alone, indicating reduced mycelium consumption. Curiously, when the predator and both competitors are present with *Lepidocyrtus* (shaded diamond), the value overlaps that of competition with *Oribatula* alone (clear circle). This value is statistically the same as *Lepidocyrtus* with the predator alone for $^{15}$N, and for *Lepidocyrtus* alone for $^{13}$C. However, when the microcosm included *Tyrophagus* as a competitor and predators (open square, black diamond, Figure 1A), significantly less consumption occurred. At the microscope in these microcosms, *Lepidocyrtus* seemed unbothered or less-so by the predator, which preferred to chase after the two other fungivores, and thus alleviating competition from the two other fungivores. Feces enrichment levels were also measured for all the different treatments containing *Lepidocyrtus* and there were no significant differences in enrichment levels dependent on species interaction ($P = 0.634$ for $^{13}$C and $P = 0.498$ for $^{15}$N).
The amount of isotopic enrichment was relatively lower in *Oribatula* (Figure 1B), as this fungus was not a preferred food source for this species. However, there were no significant differences between treatments in enrichment levels for either $^{13}$C or $^{15}$N ($P = 0.194$ and $P = 0.276$, respectively) for *Oribatula*. Comparison of enrichment through grazing between *Lepidocyrtus* and *Oribatula* across the different treatments showed significant differences in enrichment levels in both $^{13}$C and $^{15}$N ($P < 0.001$ for both).
*Tyrophagus* showed a different pattern of response to grazing competition and predation (Figure 1C). The level of enrichment from the mycelium between *Lepidocyrtus* and
### Table 1: Weight of microarthropods and mycelium consumed in daily dry-weight per individual ($n = 5$).
| Treatment | Hypoaspis (µg) | Oribatula (µg) | Tyrophagus (µg) | Lepidocyrtus (µg) | Probability (SED) |
|-------------------------------|----------------|----------------|-----------------|-------------------|------------------|
| Monoculture | 18.87 (± 1.017)<sup>a</sup> | 1.76 (± 0.128)<sup>b</sup> | 1.77 (± 0.126)<sup>b</sup> | 3.16 (± 0.105)<sup>b</sup> | $P < 0.001$ (0.734) |
| Two competitors | -0.82 (± 6.461)<sup>a</sup> | 5.29 (± 1.376)<sup>b</sup> | 5.11 (± 1.106)<sup>b</sup> | 4.70 (± 1.279)<sup>b</sup> | $P < 0.013$ (3.77) |
| Three competitors | 0.00 | 1.65 (± 0.012)<sup>a</sup> | 0.58 (± 0.032)<sup>b</sup> | 0.74 (± 0.020)<sup>b</sup> | $P < 0.011$ (0.706) |
| One fungivore and predator | - | 31.2% | 11.4% | 15.8% | |
| Two competitors and predator | - | 31.2% | 11.4% | 15.8% | |
| Three competitors and predator| - | 31.2% | 11.4% | 15.8% | |
| Probability | $P = 0.001$ | $P = 0.017$ | $P = 0.183$ | $P = 0.011$ | |
| SED | 11.46 | 11.51 | 16.68 | 8.12 | |
### Table 2: Loss (%) of individuals in each treatment.
| Treatment | Lepidocyrtus | Oribatula | Tyrophagus | Hypoaspis |
|-------------------------------|--------------|-----------|------------|-----------|
| Monoculture | 38 (± 9.8)<sup>abc</sup> | 12 (± 5.8)<sup>a</sup> | 46 (± 11.0) | 24 (± 2.4)<sup>ab</sup> |
| Two competitors | 28 (± 6.2)<sup>ab</sup> | 44 (± 8.6)<sup>ab</sup> | - | - |
| Three competitors | 18 (± 8.0)<sup>a</sup> | 18 (± 4.9)<sup>ab</sup> | 41 (± 6.8) | - |
| One fungivore and predator | 67 (± 7.7)<sup>c</sup> | - | - | 32 (± 8.0)<sup>b</sup> |
| Two competitors and predator | 68 (± 10.7)<sup>cd</sup> | - | 48 (± 14.9) | 6 (± 6.0)<sup>a</sup> |
| Two competitors and predator | 56 (± 9.3)<sup>cd</sup> | 42 (± 8.6)<sup>b</sup> | - | 16 (± 8.7)<sup>ab</sup> |
| Three competitors and predator| 71 (± 5.1)<sup>cd</sup> | 48 (± 10.2)<sup>b</sup> | 13 (± 6.0) | 2 (± 2.0)<sup>a</sup> |
| Probability | $P = 0.001$ | $P = 0.017$ | $P = 0.183$ | $P = 0.011$ |
| SED | 11.46 | 11.51 | 16.68 | 8.12 |
**Notes:**
Numbers indicate mean ± SE ($n = 5$) % loss of organisms at the end of competition/predation experiments. General analysis of variance found fauna to have significantly different losses dependent on species ($P < 0.001$); general ANOVA was performed, on individual faunal groups to assess differences between treatments (monoculture, competition with or without predator). Student Newman-Keuls test was also performed to define where the significant differences were between treatments; where there are significant differences between treatments within faunal groups different letters indicate significance.
Tyrophagus were the same \( (P < 0.001 \text{ for both isotopes}) \). As a single species microcosm and in the presence of two competitors, the three species microcosm had significantly greater for both \(^{13}\text{C}\) and \(^{15}\text{N}\) when compared to Tyrophagus with predation \( (P < 0.001 \text{ for both}) \). The APE \(^{15}\text{N}\) result was significantly highest when two competitors were present (open circle, Figure 1C), this could be through coprophagy on fecal pellets of one or both competitors. In the presence of a predator and either Lepidocyrtus or Lepidocyrtus and Oribatula as competitors, the amount of grazing significantly decreased and the isotope APE were greatly reduced for both isotopes. In this composition of the organisms (Figure 1C, black/open triangle) grazing on the mycelium is interfered with through effective grazing competition, combined with organisms trying to escape predation, and by the predation itself.
The level of enrichment obtained by the predator Hypoaspis was significantly lower than all other species \( (P < 0.001 \text{ for both } ^{13}\text{C} \text{ and } ^{15}\text{N}) \) by one order of magnitude. This is to be expected with the predator being one trophic
level above the fungi consumers. When *Hypoaspis* was in a single species monoculture, negligible amounts of enrichment occurred for both $^{13}$C and $^{15}$N, however, significant quantities of enrichment were obtained when in microcosms with prey species ($P = 0.007$ for $^{13}$C and $P < 0.001$ for $^{15}$N) (**Fig. 1D**). There is a trend for greater levels of $^{15}$N enrichment when in a microcosm with *Tyrophagus* in comparison to *Lepidocyrtus* or *Oribatula* (open triangle / black square).
Comparing the microcosms with all four organisms together (**Fig. 1**) shows the variety of consumption preferences detected between the predator (Mesostigmata) on each of its prey, and of each of the fungivores (Collembola, Astigmata, Oribatida) on the mycelium. The predator (**Table 2** and **Fig. 1D**) far preferred the Astigmata *Tyrophagus* as prey, compared to the oribatid *Oribatula*, or the collembolan *Lepidocyrtus*. This is consistent with our results above, and with our visual observations. In the combination with all four species present, *Lepidocyrtus* consumption of mycelium (**Fig. 1A**) was not affected, as the other fungivores were preferred alternative prey for the predator, and *Oribatula* was not an effective competitor for mycelium consumption. *Tyrophagus* was an effective competitor for *Lepidocyrtus* as both consumed the mycelium at the same rate, but it could barely eat (**Fig. 1C**) because of interference from *Lepidocyrtus*, and preferential predation by *Hypoaspis*. In this combination, *Oribatula* was a secondary prey for *Hypoaspis* but nonetheless preyed upon. In addition, it was in competition with both *Lepidocyrtus* and *Tyrophagus* that both preferred the mycelium, which was less palatable to *Oribatula*. Thus, *Oribatula* and *Tyrophagus* were less enriched in this four species combination.
**DISCUSSION**
We selected an ascomycete culture that would sustain three fungivorous microarthropods, and one predatory microarthropod that would prey on the fungivores. We quantified rates of hyphae consumption using $^{13}$C and $^{15}$N stable isotopes enriched mycelium. We compared microcosms with and without predation, and with or without competition for changes in hyphae consumption and predation rates.
**Precision and Sensitivity**
We showed that standard mass spectrometry instruments could perform at much higher sensitivity by modifying the chemistry of the combustion stage in the sample chamber (Crotty et al., 2013). The IRMS results in that paper showed that 1–10 individual microarthropods were sufficient for reliably accurate measurements. The microcosms in this study consisted of 10–60 individuals at the start of treatments. The resolution of the experiments (**Fig. 1**) are the mean values with five replications. The exception was the *Oribatula* (**Fig. 1B**), where low consumption showed the pattern but without statistical validity. However, the spread of the estimate of error in these experiments (**Fig. 1**) are reasonably low for IRMS microcosms.
The objective was to obtain measurements of fungal consumption, assimilation and foraging behavior of each microarthropod when affected by competition and/or predation. We understand that in these microcosms we have a flat surface that does not mimic the soil three dimensional structure. Within the complexity of the soil matrix there would be a diversity of more or less palatable species to choose from, and behavior of escape and competition interference could be different. Conceivably, in the microcosm the *Forca* (jumping organ, unique to *Collembola*), may have reduced the time needed to avoid predation (Hopkin, 2002) and resume fungal consumption; and in contrast in the soil matrix, *Collembola* could be harder to find but less able to escape. In addition, the fungivore consumers were fed a single species of fungus continuously that does not represent a more mixed diet in the soil. Our 7 days incubation period (from trials of up to 21 days) was chosen to provide sufficient levels of consumption to occur for stable isotope enrichment in 2nd and 3rd trophic levels, while providing the number of individuals necessary for isotopic analysis.
**Differential Consumption-Predation**
In natural abundance stable isotope studies, due to isotope fractionation during metabolism, $^{13}$C values indicate the source of food and the amount of consumption while the $^{15}$N values indicate the amount consumed and the trophic level of the species (Tiunov, 2007; Heijboer et al., 2018). In environmental samples with natural abundances, the rare stable isotopes ratio to common isotopes ($^{13}$C and $^{15}$N to $^{12}$C and $^{14}$N) reflect equilibrium values of the isotopes from diverse food sources. That is not necessarily the case in microcosms with species in culture and then in experimental treatments, even though the microarthropods were kept on the same mycelium for many months prior to the treatment. The natural abundance ratios of the three fungivores growing on the same mycelium for several months showed variations in isotope levels between the different microarthropods. If all three species had the same diet and consumed the same amount of the mycelium, then *Lepidocyrtus*, *Oribatula* and *Tyrophagus* would show the same isotope ratios, but that is not the case. The differences are due to probably an array of reasons, such as differences in development stage or exoskeleton - coprophagy / fecal pellet consumption, and selective metabolism (Semenina and Tiunov, 2011; Maraun et al., 2011; Hatch, 2012; Potapov et al., 2014).
The $^{13}$C values of each fungivore in monoculture (**Figures 1A–C**, dark circles) reflects the equilibrium amount and value of $^{13}$C when undisturbed. During isotopic enrichment treatment incubation, the rare stable isotope is accumulated and the graphs show enrichment for both isotopes compared to the starting point. However, with increasing disturbance from competition and predation in mixed cultures, the $^{13}$C values show less enrichment ($^{13}$C APE) moving closer to the zero value (the equilibrium value at the start).
There was a visible trend that *Tyrophagus* was a preferred prey for *Hypoaspis*, as those microcosms provided a greater amount of enrichment to the predator. It is often stated that oribatida live relatively sheltered from predation due to their size and sclerotisation (Peschel et al., 2006; Schneider and Maraun, 2009) which are reasonable assumptions based on the low predation rate on *Oribatula* (Figures 1B,D) although the greatest losses of *Oribatula* were in the microcosms that contained predators (Table 2). Moreover, *Oribatula tibialis* is known to have a defensive cyanogenic aromatic ester (mandelonitrile hexanoate) which would further explain the low predation rate (Brückner et al., 2017). These experiments provide evidence of differential consumption and differential prey preference, so that in the soil optimal foraging would contribute to the behavior of both the fungivores and predators (Hassall et al., 2006; Schneider and Maraun, 2009).
The natural abundance $^{15}$N levels were similar between the predatory mite *Hypoaspis* and two fungivores *Lepidocyrtus* and *Oribatula* before commencement of the competition/predation experiment. However, when *Hypoaspis* was in microcosms with a prey species, enrichment from fungal mycelium was traceable to a third trophic level. This exemplifies the caution that is needed when assessing isotope levels in relation to trophic level. Based on natural abundance values from field data increases of 3.4% (Post, 2002) or 1–2% (McCutchan et al., 2003; Illig et al., 2005) have been promoted as indicators of trophic level differences in food webs across ecosystems (see also Moore et al., 2004). Thus, we cannot ignore that these organisms could have consumed eggs, exoskeletons and fecal pellets, engaged in cannibalism, or even predation on juveniles as alternative sources of food in the microcosms, thus appearing to be at a higher trophic level. Some studies have even found Mesostigmata to act as more than just predators by engaging in some opportunistic omnivory (Beaulieu, 2012) which would lower its apparent trophic level in natural abundance data.
The literature is rich with examples of microarthropods having a more mixed diet in the field than typically construed, based on casual laboratory observations and from observations in microcosms (Chahartaghi et al., 2005; Erdmann et al., 2007). Other modes of feeding such as coprophagy using exoskeletons and fecal pellets, on eggs, by cannibalism or predation on juveniles have been observed elsewhere (Ponge, 1991; Briones et al., 1999; Ladygina et al., 2008; Endlweber et al., 2009; Fiera, 2014; Feng et al., 2019). For example, *Lepidocyrtus* has a hemimetabolous lifecycle compared to *Oribatula* and *Tyrophagus* which are holometabolous, so that different levels of fractionation may have occurred (Spence and Rosenheim, 2005). *Tyrophagus* is also known to be somewhat omnivorous (Behan-Pelletier, 1999) and therefore, could have potentially found an alternative food source within the microcosms, consuming eggs or immature hatchlings. Not implying that all species are opportunistically or a little omnivorous, we must nonetheless recognize that not all are as specialized as the literature might suggest. These additional sources of consumption that must be considered would affect trophic level interpretations in field data, as they do in microcosms. Microcosm observations then become useful in identifying these cases, and in quantifying energy fluxes in interactions food webs.
### Top-Down Control and Competition Interference
We have shown the fungivorous microarthropods investigated had a voracious appetite, consuming more than their own body weight per day (Table 1). In our preliminary microcosm runs, with incubation of up to 21 days, to avoid the fungivores grazing down the mycelium we provided sufficient non-growing mycelium so as to avoid bottom-up resource limitation on the microcosm. We observed several parameters of optimal foraging theory to be at play in our resource competition interference and predation food webs. One was the interference by competition for the mycelium by other fungivores (Figure 1, and see section “Species Interactions Effect on Consumption”), another was a differential preference for, consumption and excretion of the consumed mycelium biomass. There were more complicated behavioral effects, such as the mycelium consumption of *Tyrophagus* (Figure 1C) reduced by competition from *Lepidocyrtus* especially in the presence of the predator *Hypoaspis* as *Tyrophagus* was a preferred prey. Last, we speculated that the detailed interpretation of the stable isotope results (Figure 1) suggest a more mixed diet that included fecal pellets, exoskeleton, or even juveniles. Excretion by Mesostigmata is mainly fluid with no fecal pellets being formed (Koehler, 1999), but Collembola and Oribatida fecal pellets are important in soil aggregate formation, as a food source for other soil invertebrates, and sometimes contribute significantly to soil structure (Behan-Pelletier, 2003; Maas et al., 2015).
Although the three-dimensional nature of the soil pore space is dissimilar to the Petri dish microcosms, our results convey the pivotal role fungivores must play in nutrient cycling, when considering the amount of fungal mycelium consumed and excreted over time, as both groups represent major trophic and functional components of the soil environment (Leake et al., 2003; Crowther and A’Bear, 2012). The interactive effects of biotic (microarthropods and mycelium) and abiotic (habitat) are central in resolving the functional and ecological responses of the soil food web and this study provides quantified values for use in global models. However, in the natural environment abiotic factors or resource accessibility may affect populations more than competition. Similar studies by others will provide additional comparative data to better understand the complex soil food web interactions, including quantifying the fungivory pathway in detrital food webs.
Here, we have shown methods to ascertain the levels of consumption, competition and predation within the soil food web, knowledge gaps that have been highlighted historically, (Moore et al., 1988; Polis, 1994) but until now have not been quantified. Our results suggest the importance of trophic cascades, these have been modeled, and observed in low productivity systems (de Ruiter et al., 1995; Moore and De Ruiter, 2000; Moore, 2018) but have not been shown empirically until now. Studies such as this set of microcosm studies described here with stable isotope tracers, provide both
the required resolution and microscope observations that allow collection of the necessary data for predictive model simulations to be developed.
DATA AVAILABILITY
The datasets generated for this study are available on request to the corresponding author.
AUTHOR CONTRIBUTIONS
This manuscript was carried out by FC during a postdoctoral year with SA. SA wrote the research proposal, obtained the grant, and the overall design of the experiments. FC obtained the measurements and conducted the analysis and calculations. FC wrote the first version of this manuscript. It has since been rewritten, with additional statistical analysis by SA.
REFERENCES
Adl, M. S. (2003). The Ecology of Soil Decomposition. Wallingford: CABI. 336.
Barnes, A. D., Jochem, M., Lefcheck, J. S., Eisenhauer, N., Scherer, C., O’Connor, M. I., et al. (2018). Energy flux: the link between multitr richness and ecosytem functioning. Trends Ecol. Evol. 33, 186–197. doi: 10.1016/j.tree.2017.12.007
Beaulieu, F. (2012). Saproxyly in predatory mites? Mesostigmata in decaying log habitats versus litter in a wet eucalypt forest, Tasmania, Australia. Int. J. Acarol. 38, 313–323. doi: 10.1080/01647975.2011.647072
Behan-Pelletier, V. M. (1999). Oribatid mite biodiversity in agroecosystems: role for biodiscion. Agric. Ecosyst. Environ. 74, 411–423. doi: 10.1016/b978-0-444-50019-9.50023-6
Behan-Pelletier, V. M. (2003). Acari and Collembola biodiversity in Canadian agricultural soils. Can. J. Soil. Sci. 83, 279–288. doi: 10.4141/s01-063
Bradford, M. A., Strickland, M. S., Devore, J. L., and Maerz, J. C. (2012). Root carbon flow from an invasive plant to belowground foodwebs. Plant Soil 359, 233–244. doi: 10.1007/s11104-012-1210-y
Briones, M. J. I., Ineson, P., and Sleep, D. (1999). Use of 13C to determine food selection in collembolan species. Soil Biol. Biochem. 31, 957–940. doi: 10.1016/s0038-0717(98)00179-5
Brückner, A., Raspotnig, G., Wehner, K., Meusinger, R., Norton, R. A., and Heethoff, M. (2017). Storage and release of hydrogen cyanide in a chelicerate (Oribatula tibialis). Proc. Natl. Acad. Sci. U.S.A. 114, 3469–3472. doi: 10.1073/pnas.1618271114
Chahartaghi, M., Langel, R., Scheu, S., and Russ, L. (2005). Feeding guilds in Collombola based on nitrogen stable isotope ratios. Soil Biol. Biochem. 37, 1718–1725. doi: 10.1016/j.soilbio.2005.02.006
Crotty, F. V., Adl, S. M., Blackshaw, R. P., and Murray, P. J. (2012a). Protozoan pulses unveil their pivotal position within the soil food web. Microb. Ecol. 63, 905–918. doi: 10.1007/s00248-011-9956-y
Crotty, F. V., Adl, S. M., Blackshaw, R. P., and Murray, P. J. (2012b). Review: using 13C/15N values to differentiate trophic feeding channels within soil food webs. Soil Biol. Biochem. 44, 404–411. doi: 10.1016/j.soilbio.2011.07.006
Crowther, T. W., and A’Bear, A. D. (2012). Impacts of grazing soil fauna on decomposer fungi are species-specific and density-dependent. Fungal Ecol. 5, 277–281. doi: 10.1016/j.fusco.2011.07.006
de Ruitter, P. C., Neutel, A. M., and Moore, J. C. (1995). Energetics, patterns of interaction strength, and stability in real ecosystems. Science 269, 1257–1260. doi: 10.1126/science.269.5228.1257
Elsstrånd, S., Lagerløf, J., Hedlund, K., and Mårtensson, A. (2008). Carbon routes from decomposing plant residues and living roots into soil food webs assessed with 13C labelling. Soil Biol. Biochem. 40, 2530–2539. doi: 10.1016/j.soilbio.2008.06.013
Endlweber, K., Russ, L., and Scheu, S. (2009). Collembola switch diet in presence of plant roots thereby functioning as herbivores. Soil Biol. Biochem. 41, 1151–1154. doi: 10.1016/j.soilbio.2009.02.022
Erdmann, G., Otte, V., Langel, R., Scheu, S., and Marszun, M. (2007). The trophic structure of bark-living oribatid mite communities analysed with stable isotopes (N-15, C-13) indicates strong niche differentiation. Exp. Appl. Acarol. 41, 1–10. doi: 10.1007/s10993-007-9060-7
Faber, J. H. (1991). functional classification of soil fauna - a new approach. Oikos 62, 110–117.
Feng, L., Chang, L., Zhang, S., Zhu, X., Adl, S., and Wu, D. (2019). What is the carcass-usage mode of the collombola? A case study of Entomobrya proxima in the laboratory. Insects 10:67. doi: 10.3390/insects10030067
Fiera, C. (2014). Application of stable isotopes and lipid analysis to understand trophic interactions in springtails. North West. J. Zool. 10, 227–235.
Hassall, M., Adl, S., Berg, M., Griffiths, B., and Scheu, S. (2006). Soil fauna-microbe interactions: towards a conceptual framework for research. Eur. J. Soil. Biol. 42, S54–S60.
Hatch, K. A. (2012). “The use and application of stable isolate analysis to the study of starvation, fasting, and nutritional stress in animals,” in Comparative Physiology of Fasting, Starvation, and Food Limitation. ed. M. McCue (Berlin: Springer), 337–364. doi: 10.1007/978-3-642-29056-5_20
Hauck, R. D., and Bremner, J. M. (1976). Use of tracers for soil and fertilizer nitrogen research. Adv. Agron. 28, 219–266. doi: 10.1016/s0065-2113(08)60556-8
Heijbroek, A., Russ, L., Traugott, M., Jousset, A., and de Ruiter, P. C. (2018). “Empirical methods of identifying and quantifying trophic interactions for constructing soil food-web models,” in Adaptive Food Webs: Stability and Transitions of Real and Model Ecosystems. eds J. C. Moore, P. C. de Ruiter, K. S. McCann, and V. Wolters (Cambridge: Cambridge Univ Press), 257–280.
Hopkin, S. P. (2002). “Collembola,” in Encyclopedia of Soil Science. ed. R. Lal (New York: Marcel Dekker), 207–210.
Illig, J., Langel, R., Norton, R. A., Scheu, S., and Maraun, M. (2005). Where are the decomposers? Uncovering the soil food web of a tropical montane rain forest in southern Ecuador using stable isotopes (N-15). J. Trop. Ecol. 21, 589–593. doi: 10.1017/s0266467405002646
Jonas, J. L., Wilson, G. W. T., White, P. M., and Joern, A. (2007). Consumption of mycorrhizal and saprophytic fungi by Collembola in grassland soils. Soil Biol. Biochem. 39, 2594–2602. doi: 10.1016/j.soilbio.2007.05.004
Koehler, H. H. (1999). Predatory mites (Gamasina, Mesostigmata). Agric. Ecosyst. Environ. 74, 395–410. doi: 10.1016/b978-0-444-50019-9.50022-4
FUNDING
This work was supported through NSERC grant 249889 to SA.
ACKNOWLEDGMENTS
The authors would like to thank Myles Stocki for operating the mass spectrometer and maintaining its accuracy and precision to the extent necessary for this work with low biomass samples.
Koukol, O., Mourek, J., Janovsky, Z., and Cerna, K. (2009). Do orbbitid mites (Acaris: Orbitioda) show a higher preference for ubiquitous vs. specialized saprotrrophic fungi from pine litter? Soil Biol. Biochem. 41, 1124–1131. doi: 10.1016/j.soilbio.2009.02.018
Ladygina, N., Caruso, T., and Hedlund, K. (2008). Dietary switching of collombola in grassland soil food web. Soil Biol. Biochem. 40, 2898–2903. doi: 10.1016/j.soilbio.2008.08.012
Leake, J. R., Johnson, D., Donnelly, D. P., Muckle, G. E., Boddy, L., Read, D. J., et al. (2003). “Networks of power and influence: the role of mycorrhizal mycelium in controlling plant communities and agroecosystem functioning,” in Proceedings of the 4th International Conference on Mycorrhiza (ICOM 2003). (Montreal), 1016–1045. doi: 10.1139/b04-060
Maas, S., Caruso, T., and Rillig, M. (2015). Functional role of microarthropods in soil aggregation. Pedobiologia 58, 59–63. doi: 10.1016/j.pedobi.2015.03.001
Maraun, M., Marsden, A., Fischer, B. M., Pollierer, M. M., Norton, R. A., Schneider, K., et al. (2011). Stable isotopes revisited: their use and limits for orbbitid mite trophic ecology. Soil Biol. Biochem. 43, 877–882. doi: 10.1016/j.soilbio.2011.01.003
Maraun, M., Martens, H., Migge, S., Theenhaus, A., and Scheu, S. (2003). Adding to ‘the enigma of soil animal diversity’: fungal feeders and saprophagous soil invertebrates prefer similar food substrates. Eur. J. Soil Biol. 39, 85–95. doi: 10.1016/s1164-5536(03)00006-2
MucUeuan, J. H., Lewis, W. M., Kendall, C., and Mcgrath, C. C. (2003). Variation in trophic shift for stable isotope ratios of carbon, nitrogen, and sulfur. Oikos 102, 378–390. doi: 10.1371/journal.pone.0140946
Moore, J. C. (2018). Predicting tipping-points in complex environmental systems. Proc. Natl. Acad. Sci. U.S.A. 115, 635–636. doi: 10.1073/pnas.1721206115
Moore, J. C., Berlow, E. L., Coleman, D. C., de Ruiter, P. C., Dong, Q., Hastings, A., et al. (2004). Detritus, trophic dynamics and biodiversity. Ecol. Lett. 7, 584–600. doi: 10.1111/j.1461-0248.2004.00606.x
Moore, J. C., and De Ruiter, P. C. (2000). “Invertebrates in detrital food web along gradients of productivity,” in Invertebrates as Webmasters in Ecosystems, ed D. C. Coleman and P. Hendrix (Wallingford: CAB), 161–184. doi: 10.1079/9780851993426.0161
Moore, J. C., McCann, K., and de Ruiter, P. C. (2005). Modeling trophic pathways, nutrient cycling, and dynamic stability in soils. Pedobiologia 49, 499–510. doi: 10.1016/j.pedobi.2005.05.008
Moore, J. C., Walter, D. E., and Hunt, H. W. (1988). Arthropod regulation of micro- and mesobiota in below-ground detrital food webs. Ann. Rev. Entomol. 33, 419–439. doi: 10.1146/annurev.en.33.1.419
Murray, P. J., Clegg, C. D., Crotty, F. V., De La Fuente Martinez, N., Williams, J. K., and Blackshaw, R. P. (2009). Dissipation of bacterially derived C and N through and mesobiota in below-ground detrital food webs. Annu. Rev. Ecol. Syst. 40, 2898–2903. doi: 10.1146/annurev.es.18.110187.001453
Polis, G. A. (1994). Food webs, trophic cascades, and community structure. Aust. J. Ecol. 19, 121–136. doi: 10.1111/j.1442-9993.1994.tb00475.x
Polis, G. A., Sears, A. L., Huxel, G. R., Strong, D. R., and Maron, J. (2000). When is a trophic cascade a trophic cascade? Trends Ecol. Evol. 15, 473–475. doi: 10.1016/s0169-5347(00)01971-6
Pollierer, M. M., Langel, R., Scheu, S., and Maraun, M. (2009). Compartmentalization of the soil animal food web as indicated by dual analysis of stable isotope ratios (15N/14N and 13C/12C). Soil Biol. Biochem. 41, 1221–1226. doi: 10.1016/j.soilbio.2009.03.002
Ponge, J. F. (1991). Food Resources and diets of soil animals in a small area of Scots pine litter. Geoderma 49, 33–62. doi: 10.1016/0016-7061(91)90009-g
Post, D. M. (2002). Using stable isotopes to estimate trophic position: models, methods, and assumptions. Ecology 83, 703–718. doi: 10.1890/0016-6565(2002)083[0703:ustipe]2.0.co;2
Potapov, A. M., Semenyuk, I. L., and Tiunov, A. V. (2014). Seasonal and age-related changes in the stable isotope composition (15N/14N and 13C/12C) of millipedes and collembolans in a temperate forest soil. Pedobiologia 57, 215–222. doi: 10.1016/j.pedobi.2014.09.005
Russ, L., Tiunov, A., Haubert, D., Richnow, H. H., Hagglom, M. M., and Scheu, S. (2005). Carbon stable isotope fractionation and trophic transfer of fatty acids in fungal based soil food chains. Soil Biol. Biochem. 37, 945–953. doi: 10.1016/j.soilbio.2004.09.015
Schneider, K., and Maraun, M. (2005). Feeding preferences among dark pigmented fungal taxa (“Dematiaceae”) indicate limited trophic niche differentiation of orbbitid mites (Orbitida, Acari). Pedobiologia 49, 61–67. doi: 10.1016/j.pedobi.2004.07.010
Schneider, K., and Maraun, M. (2009). Top-down control of soil microarthropods - evidence from a laboratory experiment. Biol. Soc. Biochem. 41, 170–175. doi: 10.1016/j.soilbio.2008.10.013
Schneider, K., Migge, S., Norton, R. A., Scheu, S., Langel, R., Reneking, A., et al. (2004). Trophic niche differentiation in soil microarthropods (Orbitida, Acari): evidence from stable isotope ratios (15N/14N). Soil Biol. Biochem. 36, 1769–1774. doi: 10.1016/j.soilbio.2004.04.033
Semenina, E. E., and Tiunov, A. V. (2011). Trophic fractionation ([Delta]15N) in Collembola depends on nutritional status: a laboratory experiment and mini-review. Pedobiologia 54, 101–109. doi: 10.1016/j.pedobi.2010.10.004
Sokal, R. R., and Rohlf, F. J. (1995). Biometry - The Principles and Practice of Statistics in Biological Research. New York, NY: Freeman and Company.
Spence, K. O., and Rosenheim, J. A. (2005). Isotopic enrichment in herbivorous insects: a comparative field-based study of variation. Oecologia 146, 89–97. doi: 10.1007/s00442-005-0170-9
Thiele-Bruhn, S., Bloem, J., De Vries, F. T., Kalbitz, K., and Wagg, C. (2012). Linking soil biodiversity and agricultural soil management. Curr. Opin. Environ. Sustain. 4, 523–528. doi: 10.1016/j.cosust.2012.06.004
Tiunov, A. V. (2007). Stable isotopes of carbon and nitrogen in soil ecological studies. Biol. Bull. 34, 395–407. doi: 10.1134/S0006235507070412
Wardle, D., and Yeates, G. W. (1993). The dual importance of competition and predation as regulatory forces in terrestrial ecosystems: evidence from decomposer food-webs. Oecologia 93, 303–306. doi: 10.1007/BF00317685
Conflict of Interest Statement: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2457,
1
],
[
2457,
9089,
2
],
[
9089,
15546,
3
],
[
15546,
21656,
4
],
[
21656,
28446,
5
],
[
28446,
29674,
6
],
[
29674,
35824,
7
],
[
35824,
42471,
8
],
[
42471,
48881,
9
],
[
48881,
55426,
10
]
]
} |
a2e5b48be015d4a40e904bfaea9041f208959a0b | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 6,
"total-fallback-pages": 0,
"total-input-tokens": 17286,
"total-output-tokens": 3993,
"fieldofstudy": [
"Medicine",
"Psychology"
]
} | RESPONSE TO REVIEWERS
ACADEMIC EDITOR:
1. **Please ensure that your manuscript meets PLOS ONE's style requirements.** including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf
**Answer:** We have checked and corrected the manuscript according to PLOS ONE's style requirements.
2. **Please note that according to our submission guidelines outmoded terms and potentially stigmatizing labels should be changed to more current, acceptable terminology.** For example: “drug addict” should be changed to “person with alcohol use disorder”, or “person with substance use disorder”.
**Answer:** Thank you for the comment. We have changed problematic terms for acceptable terminology.
The line numbering given in the answers refers to the version of the text subjected to review (first submission).
REVIEWER #1:
According to general comment that “the manuscript would also benefit from thorough editing for word choice and flow” our paper has been checked and corrected by a highly experienced native proofreader.
1. **Authors make an assumption that craving is a “universal experience” amongst those with a substance use disorder when it is fact widely recognized that it is not.**
**Answer:** Thank you for drawing attention to this issue. It seems to us that our choice of the word “universal” was misleading. We meant to indicate that craving is an experience which occurs independently of substance type. Therefore we have changed lines 16-17 to the following:
> “Based on the assumption that the experience of craving is independent of substance type, the Polish version of the PACS was modified to measure drug craving, thus creating the Penn Drug Craving Scale (PDCS)”.
A similar correction applies to the sentence in lines 73-74.
> “Based on the assumption that the experience of craving is independent of substance type, the team modified the PACS for drug craving measurement”.
2. **The paper seems to contract its own definition of craving (paragraph at top of pg. 3)**
**Answer:** Thank you for the suggestion. In the article we did not attempt to create our own definition. The whole paragraph (lines 46-54) was based on a literature review of the craving phenomenon. The issue has been clarified.
3. The introduction reads as somewhat meandering by discussing a host of topics related to craving. It would benefit from being streamlined to the topic at hand.
Answer: Thank you for the suggestion. We have carefully re-read the introduction and changed the structure of this part of the paper. The entire fragment (lines 55-63) about the correlation between craving and other variables has been removed from the Introduction. This description is a theoretical explanation for choosing variables constituting comparative criteria (see answer #7). Therefore we have used it at the beginning of the Criterion Validity section. Moreover, the fragment about the PACS characteristics (lines 77-85) has been moved to a new subsection titled Instruments. In our opinion, all the changes made have streamlined the Introduction.
4. A. The description of when and how the participants were ascertained is confusing (pg. 5, paragraph beginning “The research was conducted from…”).
What is meant by starting “therapy”? Also, the manuscript says data was collected longitudinal, but the following sentence implies that different patients were assessed at different points in therapy.
Answer: Thank you for drawing attention to this issue. Two paragraphs (lines 101-114) have been clarified:
“The research was carried out in 14 inpatient and 13 outpatient randomly selected facilities for the psychosocial treatment of people with SUD in Poland. In most facilities participating in the research, the treatment was based on cognitive-behavioural psychotherapy, modified mostly by combining it with methods such as motivational interviewing, therapeutic community or solution-focused brief therapy. Average therapy duration was 6 months, with a range from 2 to 12 months. The research was conducted from June 2018 until July 2019. Once records with missing data regarding responses to the PDCS questions were removed, the analyses were conducted on 282 cases. The data collection process was developed during two studies. In Study 1, data were collected from 111 patients at different stages of the therapy. Study 2 was longitudinal and consisted of measurements at two time points (T1 and T2). T1 was conducted among patients at the beginning of the therapy (where the beginning of therapy means that the patients had been under treatment in a particular facility for no longer than two weeks). 171 patients were surveyed at T1. At T2, data were collected from 70 out of these patients who had completed their therapy (the rest failed to complete the therapy).”
B. Was this psychosocial treatment only?
Answer: Yes, it was psychosocial treatment only, based on cognitive-behavioural psychotherapy modified mostly by combining it with methods such as motivational interviewing, therapeutic community or solution-focused brief therapy. This information had already been included in lines 103-105. Moreover, we have added the word “psychosocial” before “treatment”.
5. A. In the “primary drug used” of Table 1, please provide what is being abbreviated by “NPS”.
Answer: Thanks for the suggestion. In Table 1 the full form of the abbreviation “NPS” has been added (New Psychoactive Substances).
B. Does primary drug use correspond to the substance use disorder (SUD) diagnoses? Would it be possible to provide SUD diagnosis as well as drug of choice?
Answer: The wording “Primary drug used” in Table 1 could be misleading. We have replaced the row header with “The most frequently used drug”. These data only concern the self-reported most frequently used substances. It is highly probable that they correspond to SUD diagnoses; however, we did not conduct analyses of diagnostic documents. So it is not possible to provide an SUD diagnosis as well as drug of choice.
C. Mean & SD of length of treatment would also be appropriate to include in this table.
Answer: Thank you for your suggestion. We have added this information in Table 1.
6. A. Given the length of treatment was highly variable (between 2-12 months), would this impact the LMI results being compared for T1 & T2?
Answer: No, the variable length of treatment would not impact the LMI results. Despite the diversity of the data, the measures of model fit were very good. In addition, data were always collected for the LMI analyses – regardless of the length of therapy – at the beginning and at the end of therapy, so each patient completed the entire treatment program. Conclusions regarding this issue were included in the Discussion section.
B. Were the mid-therapy assessments not used in any analyses?
Answer: The description of the study procedure may have led to the misconception that data from the mid-therapy assessment (study 1) were collected longitudinally, which in fact was not the case. Only the data in study 2 – measurements T1 and T2 – were collected longitudinally (see answer #4), and these were used in the LMI analyses. On the other hand, data from patients during therapy (study 1) were used for: CFA, estimation of reliability coefficients, criterion validity and normalisation. We hope that the changes in the Research procedures section (lines 101-114) have clarified this issue for the reader.
7. A. The measures used to assess criterion validity are not discussed in the methods section.
Answer: Thank you for drawing attention to this omission. Guided by the reviewer's advice, we have added a separate Instruments subsection. It contains information about the instruments used to assess criterion validity.
B. Besides the SPN, which seems to be another measure of craving, how were the other measures picked? How do they demonstrate criterion validity? There is no rationale provided as to why those measures were chosen. I wouldn’t necessarily expect there to be strong correlation between PDCS and the majority of the measures presented in Table 7.
Answer: The measures used to assess criterion validity were picked on the basis of the relapse prevention model. In this model, different variables, including craving, determine a risk of relapse. Therefore their mutual interactions are also assumed. This allowed us to treat the selected variables as the comparative
criteria. Based on this theoretical concept, we did not expect strong correlations either. We assumed that all these variables – jointly determining the risk of relapse – should correlate with each other statistically significantly, although weakly or moderately.
In response to this comment, we have modified the Criterion validity section, making use of a fragment from the Introduction (see answer #3), in the following way:
“The assessment of criterion validity is always based on an analysis of relations. In publications addressing the issue of a correlation between craving and predicting a relapse, craving is shown as a co-determining factor, alongside other intra- and interpersonal variables such as self-efficacy, motivation, negative affect (aggression, self-aggression – self-injury, impulsiveness) and social relations (sense of loneliness, social support). Most of the reported relations are incorporated in the cognitive-behavioural model of relapse. All of the listed factors contribute to a relapse; therefore their mutual interactions are also assumed. Based on this assumption, these variables were considered comparative criteria. The criterion validity assessment involved the measurement of the relations of the PDCS with instruments testing criterion variables and other scale assessing craving”.
In response to the comment on the strength of correlation, we have amended the passage in the text (lines 288-291).
8. In the discussion, further commentary on the strengths/limitations of the study is warranted.
Answer: Thank you for the suggestion. We have modified the Discussion section, highlighting the issue of strengths and limitations of the study.
9. The findings of the study are somewhat overstated in the conclusion section. While the PDCS may be used clinically, this study does not “confirm its clinical utility” nor investigate whether this measure improves treatment planning or predicts relapse.
Answer: Thank you for focusing attention on this overstatement. We have deleted this problematic fragment.
REVIEWER #2:
1. The Polish Drug Craving Scale (PDCS) should be mentioned in the title instead of PACS.
Answer: Thank you for the suggestion, but we have decided not to change the title of the manuscript. Our decision is based on previous arrangements with the Research Society on Alcoholism (RSA), which has exclusive copyrights to the PACS. Such wording of the title best protects the copyrights of the authors of the PACS and has been accepted by the RSA. The title expresses that the paper is about assessing the psychometric properties of the Polish version of the "PACS", modified for drug craving measurement.
2. On page 4, some specific research questions can be developed to guide the reader to better understand the study. These questions can follow “Aims of the analysis”. In addition, “Aims of the analysis” can be renamed “Aims of This Study”.
Answer: Thank you for the suggestion.
We have changed the title of the section Aims of the analysis to Aims of the study. With regard to specific research questions, we have decided not to add them. In our opinion, developing the research questions – where two hypotheses are included – could not improve understanding of the study.
3. In the methods, the authors missed the description of the PDCS (Table 3). How many items? How many points? Likert scale?
Answer: Thank you for pointing out this deficiency. In the Methods section, we have added an Instruments subsection. A detailed description of the PACS and the PDCS is provided in this subsection.
4. Line 148 on page 7, MPLUS needs a citation.
Answer: Thank you for drawing attention to this omission. It has been corrected.
5. Lines 155-156 on page 7, please specifically indicate what tests need the p-values?
Answer: Thank you for your comment. We have changed the sentence in lines 155-156 in relation to this issue:
“For all analyses involving a probability value, 0.05 was assumed as the threshold for statistical significance. In the presentation of the results of analyses in which a p-value was needed, it was reported each time”.
Such a change seems sufficient to us.
6. Lines 157-158 on page 8, please specifically indicate what analyses using MPLUS and LAVAAN, respectively
Answer: Thank you for this comment. In order to convey this specific information we have changed the sentence from lines 157-158 in the following way:
“The modelling was performed with Mplus 8.3. The reliability and criterion validity analysis were conducted using RStudio 1.2.5. with the application of the lavaan package. Furthermore, Jasp 0.12.2 statistical software was used for other analyses”.
7. Some statistical analyses in the results were not mentioned in the methods. The authors reported criterion validity and percentile norms of the scale. But I could not find any statements regarding these two results in the methods.
Answer: Indeed, originally in the Data analysis subsection, there was no information about the normalisation method. We have corrected this issue by adding the following statements:
“A normalisation of the PDCS results – due to the skewed character of their distribution – was prepared using a tercile scale. A tercile scale does not reflect the shape of the raw score distribution; the distribution of its values is always uniform. This means there is the same probability of the occurrence of all values of a variable”.
Regarding criterion validity, in the initial version of the manuscript (subsection Data analysis – lines 152-154), the following information had already been included:
“A criterion validity analysis was also conducted by determining the value of the r-Pearson correlation coefficient between the PDCS result and the results from other tools, constituting the comparative criteria”.
8. For criterion validity, I am not sure if the authors used latent scores or observed scores to correlate with criterion variables.
Answer: For criterion validity, observed scores were used. The r-Pearson correlation coefficient was calculated between the observed general score of the PDCS and the observed scores of the other scales constituting the comparative criteria. In order to address the issue more specifically in the manuscript, we have changed the sentence from lines 265-267 in the following way:
This is indicated by statistically significant correlations between the observed general score of the scale and the observed results from other tools used.
9. On page 8, since all the values are reported in the text, there is no need to present this table (Table 2). Please add 95% RMSEA in the text as well.
Answer: Thank you for the suggestion. Table 2 has been deleted and values of the 95% CI for RMSEA have been added in the text.
10. The longitudinal measurement invariance should be reported before the descriptive statistics at T1 and T2.
11. LMI results should be reported before the descriptive statistics. Because as the authors mentioned that “it is reasonable to compare latent variable means (drug craving) obtained during consecutive measurements” due to LMI of the scale.
Answer: Thank you for the suggestion, but we have decided not to change the sequence of the text. The arrangement of content in the text reflects the sequence of undertaken research and analysis activities. Before deciding to examine LMI, we analysed descriptive statistics for data from T1 and T2. Consequently, it was the results obtained, supplemented by the Wilcoxon signed-rank test, that provoked the question of whether the observed differences in craving levels were the effect of therapy or due to the lack of reliability over time of the PDCS. LMI was chosen as the statistical method to answer this question. Moreover, it seems to us that this sequence of content presentation allows readers – particularly readers not familiar with this method – to better understand the importance of the LMI in the context of assessing the reliability over time of a tool.
12. Lines 239-240, the statistical values for chi-square with df and p-value can be reported here.
Answer: Thank you for the suggestion. The mentioned statistical values have been reported.
13. Some statistics reports violate the APA style (p < .05 or p < .01 or p < .001).
Answer: Thank you for the comment. We checked our way of reporting statistics carefully. Consequently, we have changed the style of presentation of 95% CI (lines 32-33; 163-164). However, we have decided not to make changes in p-value reporting. In our opinion, the method of reporting these results used in the manuscript complies with the community standards contained in the PLOS ONE Submission Guidelines. Also, we have verified that the used style of presentation of the p-value (i.e. p<0.05 or p<0.01 or p<0.001) prevails in papers published in PLOS ONE. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2437,
1
],
[
2437,
5623,
2
],
[
5623,
8604,
3
],
[
8604,
11848,
4
],
[
11848,
14925,
5
],
[
14925,
17426,
6
]
]
} |
9c222f4f36d969d79ea000ad4ce3a05537a0e057 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 2,
"total-fallback-pages": 0,
"total-input-tokens": 5406,
"total-output-tokens": 1522,
"fieldofstudy": [
"Medicine"
]
} | Endovascular Repair of Blunt Traumatic Thoracic Aortic Injuries
Mehrdad Vahedian, Somayeh Rastinia, and Masumeh Taghizadeh
1Department of General Surgery, Bahonar Hospital, Kerman University of Medical Sciences, Kerman, IR Iran
Keywords: Thoracic Aorta, Stent Graft, Thoracic Injury, Thoracic Endovascular Aortic Repair
Dear Editor,
During the last decade, endovascular (EV) repair has replaced open surgical repair (OSR) as the preferred method of treatment of blunt traumatic thoracic aortic injuries (BTAIs) at many trauma centers. The thoracic endovascular aortic repair (TEVAR) means percutaneous replacement of a stent within the descending thoracic or thoracoabdominal aorta to treat aortic aneurysm (1). The thoracic endovascular aortic repair has the advantage of being a less invasive technique compared to the open surgical method of repairing, owing to the unique ability to insert the stent through a small incision. Because the incision is small, the patients who are operated by the TEVAR technique have minimal amounts of blood loss. Through the TEVAR method, prolonged cross clamping of aorta is not necessary. As a result, the incidence of renal, visceral and spinal ischemia is fewer than that in the standard open surgical repair technique (2).
Studies show that the TEVAR reduces early mortality and paraplegia compared with the open surgical management. The risk of dependency to the mechanical respiratory ventilation is lower in the thoracic endovascular aorticrepair. Unlike the traditional aortic repair, standard recovery after the TEVAR is remarkably straightforward. Patients, who have undergone the TEVAR, typically spend one night in hospital to be monitored although it has been suggested that the TEVAR can be performed as a same-day procedure. In 2005, the United States Foods and Drugs Administration (FDA) approved the pivotal trial of the TEVAR for treating the patients with thoracic aorta aneurysm (2).
Certainly, blunt traumatic thoracic aortic injuries are among the most dangerous and fatal emergency situations. The typical mechanism causing these injuries is blunt deceleration, usually from motor vehicle collisions, falls, and crashes with significant amounts of force. So, the blunt traumatic thoracic aortic injuries occur in young patients with multiple traumas (3). Since the last decade, so important advancements have been developed in medical and surgical management of blunt traumatic thoracic aortic injuries. Endovascular stent graft technologies are employed increasingly as an off label emergency treatment of these kinds of aortic injuries (4, 5). Finally, as experiments, we present two cases of 23 and 25 year-old men, who were admitted to our hospital because of blunt traumatic thoracic aortic injuries from car accidents. They were operated by the TEVAR technique. The management of both patients was the same. First of all, a chest X-ray was performed. The widening of mediastinum was obvious on the X-ray (Figure 1).
In the next step, Computed Tomography (CT) angiography and an aortography were done.
Figure 1. Preoperation CXR-widening of mediastinum is obvious in preoperation CXR.
This study confirms the dissection of descending thoracic aorta. The patients were underwent the repair of thoracic aorta by the EV stent insertion approach. At the end of surgery, after repairing the injured part of aorta, the patients were referred to Intensive Care Unit (ICU) in the surgery ward. They were closely observed for 3 days. After three days of hospitalization, they were discharged with acceptable conditions.
As it is obvious, this minimally invasive EV approach results in several advantages for the patients compared with the open surgical repair. Therefore, it is highly recommended that the TEVAR be used instead of the open repair of blunt traumatic injury of the aorta.
Footnotes
Authors' Contribution: Mehrdad Vahedian developed the original idea and protocol and wrote the manuscript. Masumeh Taghizadeh collected the data and wrote the manuscript. Somayeh Rastinnia collected, abstracted, and analyzed data, wrote the manuscript, and was a guarantor.
Funding/Support: This study was supported by the surgery ward of Kerman University of Medical Sciences.
References
1. Cannon RM, Trivedi JR, Pagni S, Dwivedi A, Bland JN, Slaughter MS, et al. Open repair of blunt thoracic aortic injury remains relevant in the endovascular era. J Am Coll Surg. 2012;214(6):943–9. doi: 10.1016/j.jamcollsurg.2012.03.003. [PubMed: 22541985]
2. Azizzadeh A, Keyhani K, Miller C3, Coogan SM, Safi HJ, Estrella AL. Blunt traumatic aortic injury: initial experience with endovascular repair. J Vasc Surg. 2009;49(6):1403–8. doi: 10.1016/j.jvs.2009.02.234. [PubMed: 19497498]
3. Miller LE. Potential long-term complications of endovascular stent grafting for blunt thoracic aortic injury. ScientificWorldJournal. 2012;2012:897489. doi: 10.1100/2012/897489. [PubMed: 22547999]
4. Rousseau H, Eiaassar O, Marcheix B, Cron C, Chabbert V, Combelles S, et al. The role of stent-grafts in the management of aortic trauma. Cardiovasc Intervent Radiol. 2012;35(1):2–14. doi: 10.1007/s00270-011-0135-9. [PubMed: 21442377]
5. Borsa JJ, Hoffer EK, Karmy-Jones R, Fontaine AB, Bloch RD, Yoon JK, et al. Angiographic description of blunt traumatic injuries to the thoracic aorta with specific relevance to endograft repair. J Endovasc Ther. 2002;9 Suppl 2:S84–91. [PubMed: 12166847] | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3160,
1
],
[
3160,
5438,
2
]
]
} |
de8bde1ec1331cfd22e883f91713bb6b565ce921 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 10,
"total-fallback-pages": 0,
"total-input-tokens": 34711,
"total-output-tokens": 12159,
"fieldofstudy": [
"Biology"
]
} | Isometric artifacts from polymerase chain reaction-massively parallel sequencing analysis of short tandem repeat loci: An emerging issue from a new technology?
Irena Zupanič Pajnič 1 | Carlo Previderè 2 | Tomaž Zupanc 1 | Martina Zanon 3 | Paolo Fattorini 3
Abstract
The recent introduction of polymerase chain reaction (PCR)-massively parallel sequencing (MPS) technologies in forensics has changed the approach to allelic short tandem repeat (STR) typing because sequencing cloned PCR fragments enables alleles with identical molecular weights to be distinguished based on their nucleotide sequences. Therefore, because PCR fidelity mainly depends on template integrity, new technical issues could arise in the interpretation of the results obtained from the degraded samples. In this work, a set of DNA samples degraded in vitro was used to investigate whether PCR-MPS could generate “isometric drop-ins” (IDIs; i.e., molecular products having the same length as the original allele but with a different nucleotide sequence within the repeated units). The Precision ID GlobalFiler NGS STR panel kit was used to analyze 0.5 and 1 ng of mock samples in duplicate tests (for a total of 16 PCR-MPS analyses). As expected, several well-known PCR artifacts (such as allelic dropout, stutters above the threshold) were scored; 95 IDIs with an average occurrence of 5.9 IDIs per test (min: 1, max: 11) were scored as well. In total, IDIs represented one of the most frequent artifacts. The coverage of these IDIs reached up to 981 reads (median: 239 reads), and the ratios with the coverage of the original allele ranged from 0.069 to 7.285 (median: 0.221). In addition, approximately 5.2% of the IDIs showed coverage higher than that of the original allele. Molecular analysis of these artifacts showed that they were generated in 96.8% of cases through a single nucleotide change event, with the C > T transition being the most frequent (85.7%). Thus, in a forensic evaluation of evidence, IDIs may represent an actual
Abbreviations: ADO, allelic dropout; AI, allelic imbalance; HDI, heterometric drop-in; IDIs, isometric drop-ins; LDO, locus dropout; MPS, massively parallel sequencing; ST, stutter product.
1 | INTRODUCTION
Autosomal DNA testing is usually performed for human identification and kinship analysis, with polymerase chain reaction followed by capillary electrophoresis (PCR-CE) of short tandem repeat (STR) markers as the gold standard [1]. In the last decade, however, new technologies, such as massively parallel sequencing (MPS), have increased the potential of forensic laboratories by enabling high-throughput acquisition of large amounts of genetic information from a single experiment [2, 3]. In particular, MPS allows the determination of sequence variability within the STR motif and single nucleotide polymorphism (SNP) variability in their flanking regions [4–7]. More recently, several kits that allow PCR-MPS of forensically relevant STR markers have been made commercially available and validated [3]. Owing to the intrinsic properties of MPS technology, its discriminatory power has been shown to be outstanding, and this approach has therefore been proposed as an ideal tool for both mixture DNA analysis and degraded samples [8].
From a technical point of view, PCR is the first step in MPS [2, 3]. Thus, MPS may reveal the presence of well-known PCR artifacts, such as allelic imbalance (AI), allelic dropout (ADO), stutter (ST) products, and allelic drop-ins [9–11]. In addition, background noise sequences (i.e., molecular products showing at least one nucleotide substitution within the STR motif) are described as occurring at very low coverage, even in the analysis of high-molecular-weight samples [12].
A recent study performed on 75-year-old bone samples using the Precision ID GlobalFiler NGS STR panel kit [13] observed the stochastic occurrence of highly covered allelic drop-ins that were named “isometric” because they had the same length as the allele that they were presumably generated from, albeit with a different nucleotide sequence. Therefore, because these drop-ins were generated from degraded templates, they were assumed to have arisen from DNA degradation itself. However, contamination issues could not be fully excluded [13].
Thus, in this study, we aimed to test whether high levels of DNA degradation could promote the synthesis of these artifacts using damaged samples produced in vitro. The samples were then analyzed using the Precision ID GlobalFiler NGS STR panel kit. This study might provide valuable insights into handling a new class of artifacts otherwise not detected by capillary electrophoresis technology.
2 | MATERIALS AND METHODS
2.1 | DNA samples
Four DNA samples (samples A, B, FM, and TS) extracted from the blood of living men were used. Informed consent was obtained before blood collection, and the samples were anonymized. Two of the samples (TS and FM) had already been applied in other validation studies [14, 15], whereas the remaining two samples (A and B) were prepared for this study. For DNA extraction, we used the protocol described by Cigliero et al. [16], with minor modifications. Briefly, the DNA was extracted by incubation at 55°C for 4 h in 0.2 M Na–acetate (pH 7.4), 0.5% sodium dodecyl sulfate, and 100 µg/ml Proteinase K. After phenol/chloroform/isoamyl alcohol (25/24/1) purification, the samples were precipitated with ethanol (2.5 volumes), washed twice in 70% ethanol, and resuspended in double-distilled water. A NanoDrop-1000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA) was used to quantify the extracts. Replicate assessments of a 1-µl sample were performed according to the manufacturer’s user guide [17]. The final concentration of the samples was adjusted to 70 ng/µl using double-distilled water.
2.2 | DNA degradation and quantification
An amount of 20 µg of each sample was incubated at 70°C as described elsewhere [18]. For all but sample A, incubation was performed for 8 and 24 h (Table 1). After incubation, the samples were purified through a 3K Amicon column (Merck KGaA, Darmstadt, Germany) and resuspended in a low-TE buffer (1-mM Tris [pH 7.4]
Table 1. Samples employed in this study
| Sample | Incubation (h) | MW | UV (ng/µl) | Auto (ng/µl) | Deg (ng/µl) | Auto/Deg | UV/Auto | PCR-MPS |
|--------|----------------|----|------------|-------------|-------------|-----------|---------|---------|
| A | 0 | +++| 412 | 4.111 (on 1:100) | 4.107 (on 1:100) | 1.0 | 1.0 | 1 |
| A8 | 8 | ++ | 205 | 8.401 | 0.012 | 700 | 24.4 | 2 |
| B | 0 | +++| 586 | 5.951 (on 1:100) | 5.061 (on 1:100) | 1.2 | 1.0 | 1 |
| B8 | 8 | ++ | 223 | 19.302 | 0.013 | 1,485 | 11.6 | 2 |
| B24 | 24 | + | 187 | 0.123 | <LOQ | n.c. | 1,520 | 2 |
| FM | 0 | +++| 582 | 5.731 (on 1:100) | 5.619 (on 1:100) | 1.0 | 1.0 | 2 |
| FM8 | 8 | ++ | 446 | 26.111 | 0.460 | 56.8 | 17.1 | 2 |
| FM24 | 24 | + | 322 | 0.049 | <LOQ | n.c. | 6,571 | 4 |
| TS | 0 | +++| 492 | 4.503 (on 1:100) | 4.908 (on 1:100) | 0.9 | 1.1 | 1 |
| TS8 | 8 | ++ | 554 | 13.702 | 0.039 | 721 | 40.4 | 2 |
| TS24 | 24 | + | 443 | 0.033 | <LOQ | n.c. | 13,424 | 2 |
Incubation: length of the incubation at 70°C; MW: molecular weight as assessed by agarose gel electrophoresis (see Section 2 for an explanation of the scores); UV: results of NanoDrop analysis; Auto and Deg refer to the results obtained using the PowerQuant System (Promega) Auto and Deg probes, respectively; Auto/Deg: ratio between the Auto and Deg values (n.c.: not calculable); UV/Auto: ratio between the quantification data in NanoDrop analysis and the Auto probe; PCR-MPS: number of PCR-MPS tests performed for each sample. Untreated control samples A, B, FM, and TS were diluted 1:100 for the qPCR assay; LOQ (limit of quantification): from 50 ng/µl to 3.2 pg/µl.
and 0.1-mM Na₂EDTA (pH 8.0). No template degradation controls (NTDCs) were used.
Degradation was assessed by electrophoresis on 1.8% agarose gels (containing 5-ng/ml EtBr) in the presence of molecular weight markers. Estimation of the molecular weight of the DNA samples was visually performed by considering the migration of the brightest point (BP) of the smear [14], and the following scores were arbitrarily assigned: BP > 23.1 kb: ++++; BP from 2 to 23.1 kb: ++++; BP from 1 to 2 kb: +++; BP from 0.25 to 1 kb: +; BP < 0.25 kb: . For DNA quantification, both NanoDrop (Thermo Fisher Scientific) analysis and quantitative PCR-based assays were performed. The PowerQuant System kit (Promega, Madison, WI, USA) was used under the suggested conditions for each sample in duplicate [19]. Raw data were obtained using an ABI 7500 Real-Time PCR System (Applied Biosystems, Foster City, CA, USA). The raw data were converted into Excel files using PowerQuant Analysis Software (Promega). Negative template controls and NTDCs were analyzed to verify the sterility of laboratory plastics and reagents.
2.3 STR typing
The Precision ID GlobalFiler™ NGS STR panel kit version 2 (Thermo Fisher Scientific) was used in this study. The DNA libraries and template preparations were run automatically on the Ion Chef System (Thermo Fisher Scientific), and an Ion S5 System (Thermo Fisher Scientific) was used for sequencing. As shown in Table S1, this method was used for duplicate analyses of 0.5- and 1-ng DNA, as assessed by the Auto probe of the PowerQuant System (Promega). Seven degraded DNA samples and four untreated samples (Table S2) were amplified using 24 cycles of PCR (for a total of 16 and 5 PCR-MPS tests, respectively). Three no-template (NT) controls were run in the same PCR runs. Fully automated library preparation was performed using the Precision ID DL8 Kit for Chef, and barcoded libraries were pooled (50 pM) and loaded onto an Ion 530 chip according to the manufacturer’s user guide [20].
Ion Torrent Suite Software 5.6 (Thermo Fisher Scientific) and Converge Software version 2.0 (Thermo Fisher Scientific) were used for MPS analysis of STR markers. The manufacturer’s default relative settings were used (0.05 was applied for both the analytical and stochastic thresholds) [21], with the exceptions reported in Table S3 [22]. Default ST ratios were also applied (Table S3). The AI threshold was set at a default value of 0.35. Coverage analysis was carried out using the Coverage Analysis v 5.6.0.1 plugin. Information about mapped reads, on-target percentage, mean depth, and uniformity of coverage were downloaded for each sample library (Barcode Summary Report file). The resulting Excel files were then used for the data analysis.
2.4 Data analysis and genotyping
The relative depth of coverage (rDoC) of the markers was calculated for each sample as the ratio between the mapped reads for a specific marker and the total mapped reads of the sample [13]; only the autosomal markers
were considered for this analysis. To assess repeatability between duplicates, the rDoC values were compared using $r^2$ tests. The sequencing data for six high-molecular-weight DNAs, run on an Ion 520 Chip during a training test performed before this study [13], were also used as controls (therefore, our sequencing control was represented by 11 tests in total; Table S2). The average molecular weight (mw) of each of the autosomal STR markers was computed as follows: (mw of the shortest amplicon + mw of the longest amplicon)/2.
The minimum depth of coverage to assign a genotype depends on the MPS technology and the aim of the study [2]. In the current study, we set a conservative fixed value of 100× coverage as a threshold for locus call and genotype assignment. Below this cut-off value, each specific locus was classified as “locus dropout” (LDO). This approach aims to limit the number of potentially mistyped loci [2, 3, 23, 24]. The correctness of the genetic typing was confirmed by two operators independently by comparison with the genotyping data of the corresponding untreated sample. For each sample, the occurrence of the following artifacts was scored: LDO, ADO, AI, ST, and allelic drop-in. The frequencies of all artifacts were computed after normalization of the data (e.g., the frequencies of ADO and AI were computed based on the number of heterozygous markers having at least 100× coverage). Consistent with the aim of this study, amplicons genotyped by the software and showing a −1 or +1 repeats with respect to the original allele were scored as STs if above the ratio in Table S3.
The allelic drop-ins were further divided into heterometric drop-ins (HDIs) and isometric drop-ins (IDIs). The IDIs comprised molecular products with the same length as the original allele with at least one nucleotide change within the STR motif, whereas the HDIs comprised length artifacts different from those classified as STs. The nucleotide sequences of the IDIs were compared with the published sequences of the STR alleles as catalogued in the STRSeq database [7] hosted at the NCBI BioProjects (https://www.ncbi.nlm.nih.gov/bioproject/380127; accessed: April 25, 2021). The typing of SNPs in the flanking regions was also checked.
Finally, the STR data of each duplicated test were used to build the composite and consensus profiles. Composite profiles were created by combining DNA profiling information from duplicate tests [25], whereas consensus profiles contained the genetic information confirmed in both duplicate tests [26]. To test the concordance, the resulting profiles were compared with the genotyping data of the corresponding untreated samples. After this task, the following four categories of results were identified: correct typing, incorrect typing, no typing, and profiles with more than two alleles.
2.5 | Calculations and graphs
Microsoft Excel 2007, version 3.0.1 (Palo Alto, CA, USA) was used for calculations and graphs. The main sequencing parameters (mapped reads, on-target percentage, mean depth, and uniformity of coverage) of the degraded samples were compared with the same parameters of the control samples using two-tailed t-tests (significance was assumed with $p$ values < 0.05).
2.6 | Comparison with IDIs found in naturally degraded samples
The main goal of the current work was to test whether in vitro degraded samples produced IDIs similar to the 75 IDIs first found in Second World War skeletal remains [13]. For both artificially degraded and naturally degraded samples, the following data were considered: coverage of the IDI, ratio with the coverage of the original allele, and availability of the sequence within the STRSeq database [23]. The same threshold of 100× was applied for locus calls as well.
3 | RESULTS AND DISCUSSION
In this study, seven degraded DNA samples were produced in vitro (Table 1) and then tested with the Precision ID GlobalFiler NGS STR panel kit in replicated analyses (for a total of 16 tests; Tables 1 and 2). In addition, a comparison with the IDIs found in naturally degraded samples [13] was performed.
3.1 | DNA degradation and quantification
A standard hydrolytic procedure [18] was applied to the four DNA samples, allowing the production of the seven samples listed in Table 1. In agreement with our expectations, all samples exhibited severe levels of degradation, related to the length of incubation at 70°C, as assessed by agarose gel electrophoresis (Figure S1) [14] and the ultraviolet (UV)/Auto ratio [14, 18], which is the ratio between the UV-spectrophotometric quantification and the molecular human DNA quantification as assessed using the PowerQuant Autosomal probe (84-bp long). The Auto/Deg ratio (the ratio between the quantification values of the Auto and Deg probes of the PowerQuant kit) [19] could be calculated only for samples treated for 8 h, whereas the lack of amplification of the 249-bp target (Deg amplicon) in all samples exposed for 24 h did not allow the calculation of the Auto/Deg ratio for these samples.
TABLE 2 Main features of the IDIs scored in the in vitro degraded samples
| | Control samples | In vitro degraded samples | Second World War bones |
|-----------------------|-----------------|---------------------------|------------------------|
| DNA samples | 10 | 7 | 16 |
| PCR-MPS | 11 | 16 | 32 |
| DNA amount | | | |
| Auto/Deg | | | |
| PCR cycles | 24 | 24 | 24 |
| Libraries (pM) | 50 | 50 | 50 |
| Threshold | 100× | 100× | 100× |
| IDIs | 0 | 95 (1) | 75 (1) |
| IDIs/test (average) | | 5.9 | 2.3 |
| Coverage | / | Average = 272; | Average = 204; |
| | | median = 239; min = 19; | median = 145; min = 10;|
| | | max = 981 | max = 1.615 |
| Ratio IDI versus original allele | / | Average = 0.389; | Average = 0.350; |
| | | median = 0.221; | median = 0.245; |
| | | min = 0.069; max = 7.285 | min = 0.053; max = 2.833|
| Single nucleotide change | / | 92/95 (96.8 %) | 64/75 (85.3 %) |
| C > T | / | 84/98 (85.7 %) | 72/89 (80.9 %) |
For comparison, data for the IDIs found in naturally degraded samples [13] are reported in the last column together with data for the control (undegraded) samples (see Table S2 for details). DNA samples: number of DNA samples; PCR-MPS: total number of PCR-MPS tests; DNA amount: amount of template (in nanograms) as assessed by the Auto probe in the PowerQuant System; Auto/Deg: Auto/Deg ratio as assessed by the PowerQuant System (n.c.: number of samples for which the ratio was not calculable); PCR cycles: number of PCR cycles; Libraries (pM): concentration (in picomoles) of the pooled libraries; Threshold: threshold used for the locus call; IDIs: number of IDIs scored (in brackets, the number of IDIs corresponding to true alleles as catalogued in the STRSeq database [7]); IDIs/test (average): number of IDIs scored in each PCR-MPS test; Coverage: coverage (in reads) of the IDIs; Ratio IDI versus original allele: ratio between the reads of the IDI and the reads of the original allele; Single nucleotide changes: number (and percentage) of single nucleotide changes scored as the source of the IDIs; C > T: number (and percentage) of C to T transitions out of the total number of nucleotide changes.
Abbreviation: IDIs, isometric drop-ins.
NTDCs and NT samples contained no quantifiable products. Therefore, the samples were not processed further.
The degradation method employed in this study caused the hydrolysis of the phosphodiester bond of the DNA [27], enriched the molecule in apurinic–apyrimidinic sites [28], and promoted the deamination of C to U [29], which is the most common DNA lesion found in ancient DNA [30]. Although it is debatable whether our approach could mimic what spontaneously occurs on DNA in a natural environment as those based on UV exposure [31], sonication [32], and DNase I digestion [33], our approach represents a unique model for understanding the molecular mechanisms of PCR artifacts and their frequency in real casework samples.
3.2 Sequencing data
For the Ion 530 chip [20] used in this study, out of the addressable wells, 47.3% showed ion sphere particles (ISPs), with more than 99.1% represented by the libraries. The final library ISP percentage was 35.5, with 3.2% adapter dimers. Overall, these data are expected when sequencing degraded samples [13, 20, 23].
The PCR-MPS of 0.5 and 1ng degraded samples could be summarized as follows. When compared with the 11 untreated test samples shown in Table S2, the degraded samples yielded, on average, fewer mapped reads (168896 vs. 391872, respectively; p value = 5.9 × 10⁻⁶), lower mean depth of coverage (3 610 vs. 9 612, respectively; p value = 3.3 × 10⁻⁶), lower percentage of on-target reads (74.3% vs. 88.9%, respectively), and lower uniformity of coverage (90.5% vs. 97.6%; p values ≤ 0.008). However, the degraded samples showed good replicability, as indicated by the r² values computed from the eight duplicates (average r² value: 0.594 ± 0.390; median: 0.641). This result is likely due to the sufficient amount of template used for PCR amplification.
The rDoC of each of the 31 autosomal STR for degraded samples and untreated controls is shown in Figure S2. As
already observed for heat-degraded samples tested with other PCR-MPS panels [15, 24, 34], the coverage of a few markers showed anomalously high values in the degraded samples. For example, the high-molecular-weight FGA marker showed an rDoC of 0.116 in the degraded samples (0.024 in the control). In agreement with Amosova et al. [35], the most likely explanation is that some sequences may be more resistant to DNA depurination on the basis of their nucleotide sequences; therefore, they are more prone to be amplified through PCR.
### 3.3 Genotyping
The genotyping data for each of the 16 degraded samples analyzed in this study are reported in Table S4, which contains the Excel files provided by Converge Software version 2.0 [21]. A comparison with the corresponding untreated sample enabled the identification of the artifacts reported in Table S1, and Figure 1 summarizes these results. On average, the frequency of LDO was approximately 4.5%, whereas AI and ADO affected approximately 15.5% and 20.0% of the heterozygous STR markers, respectively. In addition, as shown in Figure S3A (top), the occurrence of these artifacts seemed to be related to the molecular weight of the amplicons, in agreement with the model of PCR fidelity [9–11].
Among all typed markers, 30 ST products (above the threshold) were scored, along with 29 HDIs and 95 IDIs, corresponding to frequencies of 5.5%, 5.0%, and 16.6%, respectively. None of the artifacts cited earlier were scored in the untreated samples (n = 11) used as a control. No genotypes were obtained from the three NT controls. Thus, our current data showed that IDIs were generated from the PCR-MPS of severely degraded samples as one of the most frequent artifacts (on average 5.9 IDIs per sample; min: 1, max: 11), and because of a high number of IDIs scored, further detailed data were acquired (see Table 2).
Regarding the molecular mechanism that generated the IDIs, a single nucleotide change was scored in 92 of 95 cases (96.8%), whereas double change events were scored in the remaining three IDIs. In total, among all nucleotide changes, 85.7% were C > T transitions, well-known PCR artifacts [30, 36–38], mediated by the deamination of C to U [29]. C > T transitions are described as the most common errors in sequencing ancient samples [30]. As a result, these artifactual alleles usually showed more complex sequences than those of the original alleles, and as shown in Table S5, even different IDIs could arise from the same original allele in the duplicates (e.g., sample B24 at the TPOX locus, which yielded two different IDIs of allele 9). In addition, even a double IDI could arise from the original allele, as was observed for sample TS24 at locus D19S433, which yielded the original alleles 13 and 14 plus two different IDIs of allele 14. In addition, both original alleles could generate independent IDIs (e.g., sample TS8, which yielded the multi-allelic pattern 12,15,15,15 at locus D2S1338). Interestingly, sample A8 yielded profile 12,12,15,15,15 at locus D3S4529 (original genotype: 12,15; Figure S4). Finally, only one of the 95 IDIs showed a molecular sequence corresponding to the true allelic variants cataloged in the STRSeq database [7] hosted at the NCBI BioProjects (https://www.ncbi.nlm.nih.gov/bioproject/380127; accessed: April 25, 2021). For example, allele [AATG]8 of the TPOX locus yielded three different “allele 8s” ([AATG]7 [AATA]1, [AATG]6 [AATA]1 [AATG], and [AATG]1 [AATA]1 [AATG]6) that were not cataloged. Even the sex-specific markers DYS391, SRY, and Y-InDel showed sequence artifacts. The coverage of these artifacts ranged from 19× to 981× (average: 271 ± 190; median: 239), with 44 observations (46.3% of the total) in the range of 101× to 300× (Figure 2A, top). In addition, the ratios between the coverage of the IDI and the coverage of the original allele ranged from 0.069 to 7.285 (average: 0.289 ± 0.808; median: 0.221), indicating that in 5.2% of cases, the coverage of the spurious amplicons was even higher than that of the original amplicons (Figure 2B, bottom). As shown in Figure S3B (bottom), these artifacts originated in certain loci, such as D2S1338, D21S11, D6S474, and TPOX, suggesting that the STR motif could play a role in their synthesis. However, because six loci with the same [AGAT]n core motif sequence showed IDIs with wide frequencies ranging from 7% to 47% (Table S6), it is likely that other factors are involved as well. We speculate that these findings could depend on the level of molecular damage in the template [35] and/or the amplification conditions [9, 10], for example, primer binding sequences and annealing temperatures. Additionally, the
SNPs of the flanking regions were checked for concordance. The software identified spurious reads that were mislabeled as SNPs in 17 markers in the degraded samples (Figure S4D). Taken together, these artifacts showed very low coverage, reaching no more than 10% of the reads of the original allele and could easily be identified by comparison with the corresponding untreated sample.
Because the data for duplicate tests were available, both consensus [25] and composite [26] profiles were generated. As shown in Figure 3, the frequency of correct typing was higher for the consensus profile than for the composite profile (82.6% vs. 59.3%), which also showed a slightly higher frequency of genotyping errors (8.3% vs. 6.6%). For the consensus profile, mistyping was always related to the same ADO phenomenon occurring twice, whereas for the composite profile, mistyping was related to ADOs (nine cases), allelic drop-ins (seven cases), and a combination of these two phenomena (five cases). Interestingly, the composite profile of approximately 32% of the markers was composed of more than two alleles. As expected, these 90 multi-allelic profiles were mainly found in loci exhibiting higher frequencies of IDIs (Figure S5). Therefore, the presence of IDIs represents a real issue, even when generating a composite profile from duplicate tests performed on single-donor source samples, such as those used in this study.
3.4 Comparison with IDIs found in naturally degraded samples
Although the sample size was limited, the results presented in this paper provide an experimental explanation for the results obtained in the 75-year-old bone samples analyzed using the same PCR-MPS method [13]. In particular, as shown in Table 2, the in vitro degraded samples were able to reproduce the principal features of the IDIs found in the naturally degraded samples, even at a higher frequency (5.9 IDIs per sample vs. 2.3 IDIs per sample). In both sets of samples, single nucleotide changes (85.3% in aged bones and 96.8% in mock samples) within the repeat arrays caused a drop in artifactual alleles. In addition, among all the nucleotide changes, the C > T transition was the most frequent in both mock samples (85.7%) and in aged bones (80.9%). However, it is likely that the main features of these artifacts (e.g., frequency, coverage) were derived from both the amount of template DNA and the DNA degradation level.
4 CONCLUDING REMARKS
In this study, seven samples were produced in vitro to test whether severe levels of DNA degradation promoted the synthesis of IDIs [13]. The PCR-MPS results for 0.5 and 1 ng of DNA showed that IDIs were detectable only in the degraded samples, as were several other well-characterized PCR artifacts [1, 3, 10, 11], consistent with the model of PCR fidelity [9, 39]. In addition, among the different PCR artifacts, IDIs were some of the most frequent...
Degraded samples are often subjected to forensic investigation by STR analysis [1, 10, 11], and PCR artifacts are known to occur in such cases. The results presented in this study supported the conclusion that a new type of drop-in artifact, based on variations in the nucleotide sequence (IDI), could be highlighted in MPS in addition to length artifacts (HDI), which have been well characterized by PCR-CE analysis of STR markers. The occurrence of IDIs should be considered when PCR-MPS of STR markers is performed on aged forensic samples because these IDIs can represent actual issues, particularly if DNA mixtures need to be interpreted. The high number of IDIs that can appear in a single test (up to 11 in this study) could mislead the operator with regard to the number of contributors (Figure S4). By contrast, the artifactual origin of the IDIs should be suggested, in real casework, based on the stochastic manner in which they appear [40–42]. In addition, the unusual sequence of the IDI should also alert the operator to its spurious origin. However, this implies that duplicate tests are a reliable method for identifying these artifacts.
The molecular features of IDIs make capillary electrophoresis an unsuitable tool for identification because IDIs have the same molecular length as the original allele. By contrast, sequencing is an ideal tool for both identification and characterization. Thus, some of the potential offered by PCR-MPS technologies could be counteracted by the occurrence of these artifactual PCR products, which are undetected by the gold standard of CE. The Ion Torrent sequencing technology employed in this study is known to be prone to insertion/deletion artifacts [43], whereas the Illumina technology is mainly subjected to misinsertions [44]. Therefore, since each platform offers its own advantages and disadvantages in STR sequencing [6, 7, 12, 13, 15, 42, 45–51], it would be beneficial to compare the outcomes of the same heavily degraded samples across different platforms. In fact, when the ForenSeq kit was used in Illumina platforms to type degraded samples [15, 46–49], no IDI was scored, which could be because of the different levels of DNA degradation or the sequencing technology used. Moreover, because the data from this study suggested that IDIs were generated during the first PCR cycles, it may be interesting to investigate whether alternative kit designs containing unique molecular indices [52] can mitigate the occurrence of these artifacts.
In conclusion, although more complex assessments of larger sets of degraded samples are necessary, the results of this work provide further evidence that IDIs can be detected at measurable levels in heavily degraded samples after PCR-MPS on the Ion Torrent platform.
ACKNOWLEDGMENTS
This study was financially supported by the Slovenian Research Agency (project: Inferring ancestry from DNA for human identification, J3-3080). The authors wish to thank the anonymous reviewers for their helpful comments during the revision of the manuscript.
Open Access Funding provided by Università degli Studi di Trieste within the CRUI-CARE Agreement.
CONFLICT OF INTEREST
The authors have declared no conflict of interest.
DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available in the supplementary material of this article.
ORCID
Paolo Fattorini https://orcid.org/0000-0002-3416-1684
REFERENCES
1. McCord BR, Gauthier Q, Cho S, Roig MN, Gibson-Daw GC, Young B, et al. Forensic DNA analysis. Anal Chem. 2019;91:673–88.
2. Goodwin S, McPherson JD, McCombie WR. Coming of age: ten years of next-generation sequencing technologies. Nat Rev Genet. 2016;17:333–51.
3. Bruijns B, Tiggelaar R, Gardeniers H. Massively parallel sequencing techniques for forensics: a review. Electrophoresis. 2018;39:2642–54.
4. Willems T, Gymrek M, Highnam G, 1000 Genomes Project Consortium, Mittelman D, Erlich Y. The landscape of human STR variation. Genome Res. 2014;24:1894–904.
5. Phillips C, Devesse L, Ballard D, van Weert L, de la Puente M, Melis S, et al. Global patterns of STR sequence variation: sequencing the CEPH human genome diversity panel for 58 forensic STRs using the Illumina ForenSeq DNA Signature Prep Kit. Electrophoresis. 2018;39:2708–24.
6. Parson W, Ballard D, Budowle B, Butler JM, Gettings KB, Gill P, et al. Massively parallel sequencing of forensic STRs: considerations of the DNA commission of the International Society for Forensic Genetics (ISFG) on minimal nomenclature requirements. Forensic Sci Int Genet. 2016;22:54–63.
7. Gettings KB, Borsuk LA, Ballard D, Bodner M, Budowle B, Devesse L, et al. STRSeq: a catalog of sequence diversity at human identification short tandem repeat loci. Forensic Sci Int Genet. 2017;31:111–7.
8. Ballard D, Winkler-Galicki J, Wesoly J. Massive parallel sequencing in forensics: advantages, issues, technicalities, and prospects. Int J Legal Med. 2020;134:1291–303.
9. Eckert KA, Kunkel TA. DNA polymerase fidelity and the polymerase chain reaction. Genome Res. 1991;1:17–24.
10. Alaeddini R, Walsh SJ, Abbas A. Forensic implications of genetic analyses from degraded DNA: a review. Forensic Sci Int Genet. 2010;4:148–57.
11. Gill P, Haned H, Bleka O, Hansson O, Dörum G, Egeland T. Genotyping and interpretation of STR-DNA: low-template,
mixtures and database matches-twenty years of research and development. Forensic Sci Int Genet. 2015;18:100–17.
12. Young B, King JL, Budowle B, Armgard L. A technique for setting analytical thresholds in massively parallel sequencing-based forensic DNA analysis. PLoS One. 2017;12:e0178005.
13. Zupanič Pajnič I, Fattorini P. Strategy for STR typing of bones from the Second World War combining CE and NGS technology: a pilot study. Forensic Sci Int Genet. 2021;50:102401.
14. Fattorini P, Previdere C, Sorçaburu-Ciglierò S, Marrubini G, Alò M, Barbaro AM, et al. The molecular characterization of a depurinated DNA sample can be a model to understand the reliability of the results in forensic genetics. Electrophoresis. 2014;35:3134–45.
15. Fattorini P, Previdere C, Carboni I, Marrubini G, Sorçaburu-Ciglierò S, Grignani P, et al. Performance of the ForenSeq™ DNA Signature Prep kit on highly degraded samples. Electrophoresis. 2017;38:1163–74.
16. Sorçaburu Ciglierò S, Edalucci E, Fattorini P. In: Stanta G., editor. Guidelines for molecular analysis in archive tissues. Berlin: Springer-Verlag; 2011. pp. 45–54.
17. Thermo Fisher Scientific, NanoDrop 1000 Spectrophotometer V3.8 user’s manual. Wilmington, DE: Thermo Fisher Scientific; September 2011.
18. Fattorini P, Marrubini G, Bonin S, Bertoglio B, Grignani P, Recchia E, et al. Producing standard damaged DNA samples by heating: pitfalls and suggestions. Anal Biochem. 2018;549:107–12.
19. Promega Corporation. PowerQuant System technical manual. Madison, WI: Promega Corporation; 2019.
20. Thermo Fisher Scientific. Precision ID GlobalFiler™ NGS STR Panel v2 with the Ion S5™ System: application guide. Waltham, MA: Thermo Fisher Scientific; 2017.
21. Thermo Fisher Scientific. Converge Software: setup and reference guide. Waltham, MA: Thermo Fisher Scientific; November 2017.
22. Thermo Fisher Scientific. Technical note: performance of the Precision ID GlobalFiler™ NGS STR Panel v2: artifacts, thresholds and chip loading. Waltham, MA: Thermo Fisher Scientific; 06 October 2019.
23. Eduardoff M, Santos C, de la Puente M, Gross TE, Fondevila M, Strobl C, et al. Inter-laboratory evaluation of SNP-based forensic identification by massively parallel sequencing using the Ion PGM™. Forensic Sci Int Genet. 2015;17:110–21.
24. Turchi C, Previdére C, Bini C, Carnevali E, Grignani P, Manfredi A, et al. Assessment of the Precision ID Identity Panel kit on challenging forensic samples. Forensic Sci Int Genet. 2020;49:102400.
25. Bright J-A, Gill P, Buckleton J. Composite profiles in DNA analysis. Forensic Sci Int Genet. 2012;6:317–21.
26. Taberlet P, Griffin S, Goossens B, Questiau S, Manceau V, Escaravage N, et al. Reliable genotyping of samples with very low DNA quantities using PCR. Nucleic Acids Res. 1996;24:3189–94.
27. Lindahl T, Andersson A. Rate of chain breakage at apurinic sites in double-stranded deoxyribonucleic acid. Biochemistry. 1972;11:3618–23.
28. Lindahl T, Nyberg B. Rate of depurination of native deoxyribonucleic acid. Biochemistry. 1972;11:3610–8.
29. Lindahl T, Nyberg B. Heat-induced deamination of cytosine residues in deoxyribonucleic acid. Biochemistry. 1974;13:3405–10.
30. Stiller M, Green RE, Ronan M, Simons JF, Du L, He W, et al. Patterns of nucleotide misincorporations during enzymatic amplification and direct large-scale sequencing of ancient DNA. Proc Natl Acad Sci USA. 2006;103:13578–84.
31. Breslin K, Wills B, Rall A, Ventayol Garcia M, Kukla-Bartoszek M, Possieh E, et al. HRisPlex-S system for eye, hair, and skin color prediction from DNA: massively parallel sequencing solutions for two common forensically used platforms. Forensic Sci Int Genet. 2019;43:102152.
32. Gettings KB, Kiesler KM, Vallone PM. Performance of a next generation sequencing SNP assay on degraded DNA. Forensic Sci Int Genet. 2015;19:1–9.
33. Dixon LA, Dobbins AE, Pulker HK, Butler JM, Vallone PM, Coble MD, et al. Analysis of artificially degraded DNA using STRs and SNPs: results of a collaborative European (EDNAP) exercise. Forensic Sci Int. 2006;164:33–44.
34. Salata E, Agostino A, Ciuna I, Wootton S, Ripani L, Berti A. Revealing the challenges of low template DNA analysis with the prototype Ion AmpliSeq™ Identity panel v2.3 on the PGM™ Sequencer. Forensic Sci Int Genet. 2016;22:25–36.
35. Amosova O, Coulter R, Fresco JR. Self-catalyzed site-specific depurination of guanine residues within gene sequences. Proc Natl Acad Sci USA. 2006;103:4392–7.
36. Dadney J, Meyer M, Pålbo S. Ancient DNA damage. Cold Spring Harb Perspect Biol. 2013;5:a012567.
37. Chen G, Mosier S, Gocke CD, Lin MT, Eshleman JR. Cytosine deamination is a major cause of baseline noise in next-generation sequencing. Mol Diagn Ther. 2014;18:587–93.
38. Wong SQ, Li J, Tan AY, Vedururu R, Pang JM, Do H, et al. CANCER 2015 Cohort. Sequence artefacts in a prospective series of formalin-fixed tumours tested for mutations in hotspot regions by massively parallel sequencing. BMC Med Genomics. 2014;7:23.
39. Acinas SG, Sarma-Rupavtarm R, Klepac-Ceraj V, Polz MF. PCR-induced sequence artefacts and bias: insights from comparison of two 16S rRNA clone libraries constructed from the same sample. Appl Environ Microbiol. 2005;71:8966–9.
40. Nix DA, Hellwig S, Conley C, Thomas A, Fuertes CL, Hamil CL, et al. The stochastic nature of errors in next-generation sequencing of circulating cell-free DNA. PLoS One. 2020;15:e0229063.
41. Singh RR. Next-generation sequencing in high-sensitive detection of mutations in tumors: challenges, advances, and applications. J Mol Diagn. 2020;22:994–1007.
42. Müller P, Alonso A, Barrio PA, Berger B, Bodner M, Martin P, DNASEQEX Consortium, et al. Systematic evaluation of the early access applied biosystems precision ID Globalfiler mixture ID and Globalfiler NGS STR panels for the iOn S5 system. Forensic Sci Int Genet. 2018;36:95–103.
43. Bragg LM, Stone G, Butler MK, Hugenholtz P, Tyson GW. Shining a light on dark sequencing: characterising errors in Ion Torrent PGM data. PLoS Comput Biol. 2013;9:e1003031.
44. Schirmer M, Ijaz UZ, D’Amore R, Hall N, Sloan WT, Quince C. Insight into biases and sequencing errors for ampiclon sequencing with the Illumina MiSeq platform. Nucleic Acids Res. 2015;43:e37.
45. Xavier C, Parson W. Evaluation of the Illumina ForenSeq™ DNA Signature Prep Kit – MPS forensic application for the MiSeq FGx™ benchtop sequencer. Forensic Sci Int Genet. 2017;28:188–94.
46. Müller P, Sell C, Hadrys T, Hedman J, Bredemeyer S, Laurent FX, et al. SeqForSTR-Consortium. Inter-laboratory study on standardized MPS libraries: evaluation of performance, concordance, and sensitivity using mixtures and degraded DNA. Int J Legal Med. 2020;134:185–98.
47. Churchill JD, Schmedes SE, King JI, Budowle B. Evaluation of the Illumina(®) Beta Version ForenSeq™ DNA Signature Prep Kit for use in genetic profiling. Forensic Sci Int Genet. 2016;20:20–9.
48. Carrasco P, Inostroza C, Didier M, Godoy M, Holt CL, Tabak J, et al. Optimizing DNA recovery and forensic typing of degraded blood and dental remains using a specialized extraction method, comprehensive qPCR sample characterization, and massively parallel sequencing. Int J Legal Med. 2020;134:79–91.
49. Zavala EI, Rajagopal S, Perry GH, Kruzic I, Bašić Ž, Parsons TJ, et al. Impact of DNA degradation on massively parallel sequencing-based autosomal STR, iiSNP, and mitochondrial DNA typing systems. Int J Legal Med. 2019;133:1369–80.
50. Wang Z, Zhou D, Wang H, Jia Z, Liu J, Qian X, et al. Massively parallel sequencing of 32 forensic markers using the Precision ID GlobalFiler™ NGS STR Panel and the Ion PGM™ System. Forensic Sci Int Genet. 2017;31:126–34.
51. Tao R, Qi W, Chen C, Zhang J, Yang Z, Song W, et al. Pilot study for forensic evaluations of the Precision ID GlobalFiler™ NGS STR Panel v2 with the Ion S5™ system. Forensic Sci Int Genet. 2019;43:102147.
52. Smith T, Heger A, Sudbery I. UMI-tools: modeling sequencing errors in Unique Molecular Identifiers to improve quantification accuracy. Genome Res. 2017;27:491–9.
SUPPORTING INFORMATION
Additional supporting information may be found in the online version of the article at the publisher’s website.
How to cite this article: Zupanič Pajnič I, Previdere C, Zupanc T, Zanon M, Fattorini P. Isometric artifacts from polymerase chain reaction-massively parallel sequencing analysis of short tandem repeat loci: An emerging issue from a new technology?. Electrophoresis. 2022;43:1521–1530. https://doi.org/10.1002/elps.202100143 | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2209,
1
],
[
2209,
6199,
2
],
[
6199,
11338,
3
],
[
11338,
16389,
4
],
[
16389,
21481,
5
],
[
21481,
26172,
6
],
[
26172,
29067,
7
],
[
29067,
34401,
8
],
[
34401,
40620,
9
],
[
40620,
42888,
10
]
]
} |
2f5d25f725af630bcc037d8a901d52af6a988301 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 14,
"total-fallback-pages": 0,
"total-input-tokens": 36823,
"total-output-tokens": 10206,
"fieldofstudy": [
"Education"
]
} | Radical Flexibility and Relationality as Responses to Education in Times of Crisis
George Veletsianos1 • Shandell Houlden1
Accepted: 17 September 2020 / Published online: 2 October 2020
© The Author(s) 2020
Abstract
As educational institutions negotiate numerous challenges resulting from the current pandemic, many are beginning to wonder what the future of education may look like. We contribute to this conversation by arguing for flexible education and considering how it can support better—more equitable, just, accessible, empowering, imaginative—educational futures. At a time of historical disorder and uncertainty, we argue that what we need is a sort of radical flexibility as a way to create life-sustaining education, not just for some, but for all, and not just for now, but far into the future. We argue that such an approach is relational, and centers justice and trust. Furthermore, we note that radical flexibility is systemic and hopeful, and requires wide-ranging changes in practices in addition to the application of new technologies.
Keywords Radical flexibility • Flexible learning • Online learning • Education in crisis
Introduction
We are living in times of multiple and multiplying crises, some apparently slow and later, and maybe abstract, others fast and tangible and now. In the immediacy of yesterday and today, the novel coronavirus moves quickly through some of our communities, unevenly striking folks down, disproportionately killing elderly kin, essential workers, and Black people, Indigenous people, and people of color (BIPOC). In the future, or in a place that feels like somewhere else, somewhere at least a little bit distant for now, until it does not, the crisis of the climate emergency, of biodiversity loss and extinction, of people displaced due to climate catastrophe, and of ecological collapse, moving at a pace that for many somehow registers as never too late to address—or else, something yet to be addressed, no doubt just in time.
These two crises—the current pandemic and the climate emergency—appear to many to be distinct from each other, to be separate disasters moving at different speeds in different places. It seems that one can be addressed now, and the other later. But there is no coronavirus pandemic, at least not as we are seeing it today, without the same activities at the root of ongoing and increasingly dire climate disasters (Kolinjivadi 2020), those anthropogenically induced planetary changes so vast as to necessitate their own geologic category of the Anthropocene (Crutzen 2002). The Anthropocene—the epoch characterized by the significance of human impact on our planet—is everywhere, including in our educational institutions, which are negatively impacted by the disasters that now characterize our world, while these same institutions simultaneously fail in fundamental ways to adequately grapple with their role in its perpetuation. What this moment demands of higher education researchers, practitioners, and administrators is meaningful and just response. That response must be manifested in a willingness to navigate and adapt to unpredictable and shifting circumstances that impact people in profoundly uneven ways. It must be about imagining, and then enacting, better futures—meaning imaginative, equitable, accessible, sustainable, and decolonial—for higher education. Such a response needs to be deeply flexible, and flexible across social, cultural, and material differences.
This need has been made painfully clear throughout the far-reaching disruptions flowing from the coronavirus pandemic. The topic of flexible education—that is education that is responsive to learner and societal needs, available in multiple formats, through multiple delivery modes, in multiple timeframes and locations—has perhaps never been so salient, so immediately tangible to the lives of so many people. As various degrees of lockdown in places across the world have closed brick and mortar doors at educational institutions of all levels and kinds, and educational institutions negotiate numerous challenges, many researchers and commentators are turning their attention to the future of education, pondering what it might look like (e.g., Selwyn and Jandrić 2020; Walsh 2020; and Witze 2020 among many others). In this paper, we contribute to the conversation about the future of education at this particular moment, by arguing for radical flexibility and considering how it can support better—more equitable, just, accessible, empowering, imaginative—educational futures. To examine these issues with particular attention to flexible digital education at a time of historical disorder and uncertainty, we draw on our shared knowledges and experiences as educators and non-Indigenous researchers located in the settler colonial context of North America, where one of us is well established in his field of education and learning technologies, and one of us has recently completed her doctoral degree in cultural studies. We proceed by describing the context in which we situate our analysis. Next, we present a theoretical framework to guide our analysis of radical flexibility, and finally, we discuss how radical flexibility may look like in practice, drawing attention to issues of trust and relationality.
**Context**
Prior to the pandemic, the anticipated or predicted future of education was already often described as flexible, or as needing to be flexible (Barnett 2014; Gordon
This question of flexibility has developed in itself as a field of inquiry, with increasing attention paid to it in the previous 10 years (Houlden and Veletsianos 2019, 2020; Selwyn 2011; Sheail 2018), though related research has been ongoing for decades (e.g., Daniel 1998; Edwards 1997; Evans 2000; Veletsianos and Houlden 2019). No longer thought simply in terms of access to education at ‘anytime’ from ‘anywhere’, the breadth of how we might make education more flexible has expanded in scope to include everything from shifts in entrance and completion requirements (e.g., flexible admissions policies via prior-learning assessment), and multiple modes of access that provide learners with hybrid choices between in-person and online learning, as well as choice in curriculum and assessment better suited to learners’ needs, for example. In other words, beyond the conventional interpretation of flexibility as being about time and space, flexibility has come to be understood as about making many educational practices malleable and responsive to students and markets (Naidu 2017). Today, the overall ideals of flexible education are to increase the student-centered and empowering aspects of education, thereby improving not just access, but also equity, diversity, inclusion, retention, completion, and satisfaction (Houlden and Veletsianos 2019).
Still, flexible and online education is not without its critics and cautions, and this too has become more apparent, and more widely discussed, currently as the effects of the pandemic reverberate through every aspect of our education systems. Many practitioners for example have recently argued that online learning comes with a whole host of drawbacks, including concerns around accessibility, security, and quality (Fain 2019; Herman 2020; Xie et al. 2020), while researchers have often noted that inequities and technological determinism beleaguer the field of digital education (e.g., Reich in press; Veletsianos 2020). But what is clear to many scholars studying online education, especially those who have been studying it prior to the pandemic, is that a distinction needs to be drawn between the education that was delivered in the spring and summer of 2020, or what has aptly been called emergency remote teaching, and the skillful and well-researched methods of online education (Hodges et al. 2020). What we are bearing witness to now—not just the rapid transition in February and March of 2020, but also the use of online and hybrid options for Fall 2020 and beyond—is flexible digital education deployed in haste, driven by an immediate need to adapt to rapid changes in delivery, namely as suddenly other than face-to-face, all amidst the threat and uncertainty of a widely circulating, poorly understood pathogen.
It is in this specific ongoing context, set against the backdrop of the existential crisis of the climate emergency, that a certain kind of flexible education should emerge, one capable of addressing the crisis at hand and those on the horizon. In this paper therefore, we ask:
- In what do we need to ground this flexibility such that it is capable of responding to the circumstances in which we find ourselves, circumstances which dissolve our false sense of a stable, secure, and reliable future, and underscore the precarity of our globalized infrastructures and networks?
- How to do so without minimizing the under-critiqued and underthought tendencies and mechanisms of flexible education?
Theoretical Framework
Rather than begin with a set of ostensibly flexible solutions (e.g., new platforms or tools) that may either prove unviable or in fact make education more inflexible or less effective for whatever purpose it is intended to serve, institutions require the adoption of a radical approach to flexibility. Radical flexibility is not just about making the logistics of education practices easier or more flexible (e.g., providing students with a menu of assignments to choose from), but means taking seriously the nature and purpose of learning itself at the fundamental level of human life, where human life is understood to be enmeshed relationally with all that goes on around, with, and through it. In other words, radical flexibility is a backdoor into thinking not just about how to deliver education equitably, but to ask what kind of education, what kind of university, do we want—which is in turn to ask, what kind of life, what kind of future do we want, and for whom? These are the kinds of questions that education theorists worthy of the crises of the pandemic, climate change, and global racial and colonial injustice are asking (Allen et al. 2020; Bozkurt et al. 2020; Costello et al. 2020). This means that rather than proposing solutions to a series of complex problems, radical flexibility is an invitation to imagine and turn to the tools, mechanisms, and systems needed in order to create life-sustaining education, not just for some, but all, and not just for now, but far into the future. Which is to say that radical flexibility is not a structure but is an orientation, one defined by its openness, to how we think about the problems made legible by the pandemic.
To imagine life-sustaining education means beginning with a more just paradigm of who the learner is and can be, or, in other words, that to be flexible is to begin by interrogating assumptions about who the learner is and what tools and capacities they have at their disposable. Elsewhere, we argue (Houlden and Veletsianos 2020) that conventional forms of flexible education that are sometimes reducible to ‘anytime, anywhere’ discourses (i.e., where flexibility is seen through the lens of things such as flexible pacing and the capacity to work from anywhere) are often limited by a structurally implicit orientation to an ideal learning subject, what McMillan Cottom (2015) calls the ‘roaming autodidact’. This is the learner who has the wherewithal to make or access the capital, time, and space for learning in spite of all the other obligations that they have. McMillan Cottom argues that such an orientation inevitably favors white, able-bodied male learners of particular socio-economic status by virtue of the significant privileges that often come with occupying that identity space. Research into the challenges of flexible education for female and BIPOC students, for example, supports this thesis (McMillan Cottom 2017; Selwyn 2011; Simon et al. 2014). In contrast, the imagined ideal learner is the learner as a good liberal humanist subject (Houlden and Veletsianos 2019), he who is independent and above all has fully internalized responsibility as being entirely located in and oriented to the individual (Houlden and Veletsianos 2020). This is the model learner: the one who is self-directed, can command resources, skills, space, and time; the one who has choices and options, and faces far fewer systemic obstacles. This is a learner divorced from the messy and cacophonous reality that the majority of the world faces.
Taking the latter version of the learner to be how we orient our educational systems is to reinforce anti-relational capitalist ideologies and systems that tend to enforce a hierarchy of life according to cis-heteropatriarchic and racial logics.
(Melamed 2015; Gilmore 2002). It is also to perpetuate structures that disavow the relational nature of life and subjectivity, which is to what radical flexibility has the potential to respond. In other words, instead of developing education for the so-called roaming autodidact, or the learning subject of neoliberal market economics, radical flexibility begins with the principle of the relational nature of all things, a perspective which has a rich and varied theoretical history in a number of critical traditions, including posthumanist thought (e.g., Braidotti 2013; Haraway 2016; Wolfe 2009) and ecofeminist thought (e.g., Gaard 2017; Plumwood 1991), and which was long preceded by Indigenous thought and cultural systems (e.g., Atleo 2011; Todd 2020; Wilson 2008). What unifies some aspects of these diverse perspectives is an ethical orientation guided by an understanding of the relational nature of existence, where all of us are reliant upon and thus responsible to the beings and the worlds in which we live.
With this relational ethical frame in mind, radically flexible education is grounded on the recognition that all learners are embedded in multiple communities and webs of obligations and shared responsibilities that figure deeply into any learning such an individual can do. This kind of education takes its learners to be rich, and complex beings, with deep inter-generational histories of both joy and suffering that impact how and what they both desire and need to learn. To be clear, the responsibility that shapes radical flexibility is in distinct contrast to neoliberal or biopolitical forms of responsibility to which the roaming autodidact is normatively oriented. This latter form of responsibility, seen through Foucault’s (2007) insight into the ways in which governable subjects are taught to internalize their circumstances as wholly their own responsibility, alludes to the learner as being responsibilized. The responsibilized learner is the individual who accepts their need for growth and education as a responsibility they have to perform, a duty even, and in a way that meets the narrow parameters of their own already circumscribed desires (Peters 2005). In doing so, they sustain dehumanizing neoliberal logics that assert that the individual is a distinct, self-determining unit, and that their social and economic status is strictly determined through their own actions rather than through systemic factors that constrain or support their activity (Houlden and Veletsianos 2020).
In contrast, radical flexibility approaches responsibility in a far more holistic sense, where responsibility is moved first by the capacity to respond in a life-sustaining and life-supporting way, whether that be to respond to one’s own fundamental needs and desires, the needs and desires of one’s community and broader ecological environment, and even the needs and desires of one’s ancestral and future kin, human and otherwise. This is responsibility understood as a tending of relations. As Potowatomi scholar Whyte (2013: 518) explains:
to be in a relationship is to have responsibilities toward the others in the relationship. Responsibilities refer to the reciprocal (though not necessarily equal) attitudes and patterns of behavior that are expected by and of various parties by virtue of the different roles that each may be understood to play in a relationship.
Systems that inhibit capacities to respond to relationships in this way are actually antithetical to radically flexible education. What’s more, radically flexible education takes its learners to have capacities that shift in meaningful ways throughout the
duration of their institutional learning according to the ebbs and flows of everything from their very bodies, to their home lives, to their access to resources, to the effects of violence sustained by broader social and cultural systems that, for example, over-police Black and racialized people, or celebrate white nationalism, or deny the lasting impacts of colonial genocide, or disproportionately harm working class or elderly people, as has so often been the cases with the coronavirus pandemic (Center for Disease Control 2020; Eldeib et al. 2020; García de Müeller et al. 2020). In other words, radical flexibility in education begins with the recognition that learners are relational beings and must be honored and collaborated with as such.
This also means that radically flexible education accounts for present materialities, i.e., it is responsive to the circumstances people live with on a day-to-day basis, why people are doing the work of learning and developing new skills, and who they are doing the work for and with. To begin with, learners understood in this way means flexibility becomes a value or principle that shapes educational infrastructure and pedagogical practices. Here, education is guided by adaptability, suitability, responsiveness, and creativity, all of which fall under the umbrella of justice, or as hooks (1994) calls it, education as the practice of freedom. While some educational technologies may prove beneficial in support of this, they are not the solution. They are the means by which flexibility is mobilized and enacted, or how education is made more responsive and more relational. If this core value is obscured—the relational nature of justice—then there is a serious risk of relying on solutions that create more problems than they purport to solve. For example, in the name of permitting students to take exams at home, institutions might insist that they use test-proctoring technology that relies on invasive forms of surveillance, thereby formalizing distrust and deepening dehumanization in our pedagogical methods (Flaherty 2020; Swauger 2020). Those in positions to make decisions—especially administrators such as deans, directors of centers of teaching and learning, and many others in positions of institutional power, and even faculty with power over the technologies they use and lobby for at their institutions—would do well to seriously and continuously consider what problems are being addressed by educational technology interventions, and what values are inherent in the solutions being offered.
**Practice**
Key shifts are needed in order to enact this kind of radical flexibility. Bayley (2018: 245) argues that in crises, ‘[w]e need to find practices to stay with the trouble stirred up by late capitalism in the anthropocene moment – a moment where “scholarship committed to the refusal if not the undoing of a world riven by new kinds of warcraft, injustice and exploitation” requires the courage of action’. Such practices are not inseparable from the theory of relationality articulated above, because as hooks (1991): 2) observes:
[w]hen our lived experience of theorizing is fundamentally linked to processes of self-recovery, of collective liberation, no gap exists between theory and practice. Indeed, what such experience makes more evident is the bond between the two – that ultimately reciprocal process wherein one enables the other.
Fundamentally, because radically flexibility is grounded in relationality, it is a process of self-recovery and collective liberation. Perhaps to the dismay of some (e.g., educational technology advocates and those who have much to gain from the expanded use of technology in education, ranging from Silicon Valley startups to educational technology consultants), this process, and the practices that come with them, does not by necessity involve inserting new technology into education (though, once again, carefully vetted new technology may prove helpful in cultivating and supporting the approaches we outline here). The shifts radical flexibility may require, however, are not dependent upon anything so facile. What follows are some suggestions into radically flexible education (though no doubt there are many more) which center relationality, both in theoretical and practical terms.
**Trust**
To engage with students as relational beings, designers, administrators, and practitioners could consider eliminating the mechanisms and ideologies that reinforce the institution’s and learner-educator’s suspicion of the learner (Fawns and Ross 2020). To ground education on the notion that learners must prove themselves is potentially dehumanizing and reduces all the complexities addressed above to a footnote to how learners are expected to participate in their learning, rather than as the very means by which they arrive to their learning.
In practice, this means to trust learners—which, in abstract terms is something that many can agree upon, but in practical terms may confound or elicit resistance. In practical terms, trust could mean no more doctor’s notes, no more demand for proof that a family member died or that a learner has actually been diagnosed with Covid-19. It could mean accepting digital copies of reference letters and transcripts while building the digital systems to maintain the privacy and security of such documents (which is where educational technology can actually be useful). It could mean developing sustainable and holistic assessment practices, practices such as having students write critical reflections of their own work, which as Stommel (2018) notes, require releasing ‘attachment to accuracy’ and objectivity to ‘give way to a dialogue – one that is necessarily emergent and subjective’. It means avoiding or abandoning technologies that engender distrust (Ross and Macleod 2018), such as plagiarism detection tools. It may also mean considering that certain foundational elements of established practice may be antithetical to trust, and potentially begin the process of reconsidering and rejecting them. There are too many elements to list here, but to illustrate they may include various principles of instructional design, such as designing instruction grounded on predetermined performance objectives or evaluating outcomes around criterion-referenced assessments. This process may be difficult, not only due to sedimentation around practices that have long been recognized as ‘good’, but also because in the face of crisis, relying on familiar tools/approaches can provide comfort and a sense of stability in the face of uncertainty.
Central here is the recognition that dialog is intrinsically relational, and relationality reinforces trust, which is to say that trust is itself an emergent practice. Trust is not something one gives, but something one does, and the reciprocal nature of it means that it works both ways, that both institution and faculty, as well as learners, can practice trust. Trust is not something that can be granted through statements or declarations...
without meaningful action, which partially explains why so many individuals in the academy are so skeptical of both pandemic reopening plans that do not have realistic attitudes toward health and safety (e.g., Welch 2020), and of equity and diversity statements in the wake of ongoing efforts to dismantle colonization (Doharty et al. 2020) and ongoing anti-police and Black Lives Matter protests (Howard 2020; Melaku and Beeman 2020).
Counterarguments to the educational practices of trust, such as the argument that some of these adjustments mentioned above are unfair to other students in a class, return to the zero-sum scenario in which justice only looks one way, and thus lose track of the relational nature of education. Importantly, relational approaches are not approaches that disavow accountability. Accountability in relational settings multiplies and manifests in non-prescriptive ways, which is to say, ways that are actually accountable to the complex moving parts of education—the learner, the learner-educator, and the broad ecologies and networks in which both of these beings are embedded. Accountability, for example, might be better enacted in terms of collaborative roles learners occupy and are responsible for together, or by emphasizing a learner’s education in relation to their community responsibilities, but always specific to what their role and responsibility are to their specific community.
Trust also means listening to and responding to the needs of learners, based on their experiences as relational beings enacting, but not reducible to, the role of learners. This could mean, for example, building in accessibility through universal design and the understanding that disability is not something to be overcome or to be treated as a deficit as ableist structures would have it, but is instead ‘a valued part of identity’ (Ban 2020). It also reflects another opportunity to do away with the frame of suspicion that demands proof of disability in the cases of less readily apparent disabilities like chronic or mental illness, for example, and instead ‘views students through a holistic lens and trusts students as people who are experts on their own lives rather than assigning expertise to a third party with medical authority’ (Evans et al. 2017: 365). The effect of this will be to reduce the risk inherent to disclosure of disability, as well as to reduce the labor learners are required to put in with respect to being seen and responded to as their needs dictate, which in turn will permit them to put their labor into their learning.
This same shift away from suspicion to trust also needs to occur for educators, faculty and staff. For example, does a faculty member, adjunct instructor, or graduate student working as a teaching assistant prefer or need to teach online rather than face to face during a pandemic? Demanding they provide narrow forms of evidence of immunocompromise for themselves or members of their household, or urging and requesting them to teach face to face as has been the case for many in the USA, is a failure to respect not only the privacy, expertise, and labor of an individual, but also their relational nature. In practice, what radically flexible education may look like is better support for all academic workers—many of whom are far more unfavorably resourced and precariously positioned than others, irrespective of location—which includes everything from reasonable and sustainable working hours, support in technical skills and pedagogy development, support for parental leave, and adequate care during times of illness and disability, for example. More broadly still, it means actively dismantling the institutional forces that contribute to illness and disability, like racism, sexism, and transphobia, and given the lack of supports for anti-oppressive pedagogies and practices (e.g., Valcarlos et al. 2020), expanding supports for them, specifically in the context of
postdigital efforts. In the case of disability, for example, this means refiguring the ways by which ‘excellence’ is anchored in individualistic notions of self-reliance and independence, given that too often disabled scholars are expected to perform such a circumscribed form of excellence in spite of their disabilities (Merchant et al. 2019). This logic of excellence as the purview of the individual, rather than as being a collaborative way of being, is exemplary of the norms of suspicion within anti-relational systems, as excellence here is defined by the notion that one does it alone.
What follows from this is that radical flexibility is a systemic approach. It does not arise solely through the application of new technologies, partnerships with big tech companies, or semi-nouveau ideas like ‘openness’ or ‘upskilling’ or ‘MOOCs’ or ‘learning analytics’ or ‘learning dashboards’. Rather, it means that those with power, namely privileged faculty and administrators—and the institutions they work with—cannot treat one group in a system relationally while managing another as cogs in a machine. This creates a divide between those who are treated as human beings and those who are not, thereby undermining attending to relationality itself.
**Conclusion**
What this pandemic makes abundantly clear is the pressing need not just to build resilient and adaptable ways of designing, developing, and delivering education, but also to subvert the marriage of capitalism and postdigital education in order for education to become a place for the practice of freedom. This is where flexible education becomes radical: it is simultaneously practice and politics, even if education has always been both of those things. What this amounts to is an educational environment in which the people participating and supporting education are understood to be and thus treated as holistic beings, and the digital tools used are meant to facilitate the process of enabling and encouraging the complex relationality of each individual learner and their life. Doing so means attending to the reality of larger circumstances in which we can no longer disavow late capitalism’s racist and imperialist environmental impacts (Heglar 2019; Holthaus 2020; McKibben 2020; Nixon 2013), especially as they are bound up with the effects of anthropogenic climate change (e.g., Alexander 2020; Randall and Gray 2019). Consequently, radically flexible education as an orientation to relationality needs to be far more accountable to the history of education itself than conventional education currently is. By ‘history’ here, we are not alluding to the history of the use of technology in education and the lessons embedded within it. Instead, we are referring to the histories of violence in which a vast majority of Western education systems and institutions are imbricated. Such histories include legacies of slavery and white supremacy (Crawley 2018), which still shape and impact access to education along racialized divides (Reece and O’Connell 2016), with white supremacy, which advantages white people while disadvantaging BIPOC, remaining a structural issue across Western academic institutions (Gillborn 2005; Tate and Bagguley 2016). This also includes attending to histories of Indigenous genocide and colonization, and specifically in settler colonial North America, the role of the land grant system enacted through the Morrill Act of 1862, which ‘turned Indigenous land into college endowments’, and which to this day materially sustains many major academic institutions in North America (Lee and Ahtone 2020; Stein 2020). Such histories are important if we are to meaningfully respond to the ongoing legacies of
colonization that currently amount to significant inequality for Indigenous and racialized faculty and students in higher education (Henry et al. 2017) and in racial divides within education more broadly. Without centering these histories in our education institutions, a relational orientation to the future, one predicated on the sustainability of life itself on a finitely resourced planet is nearly impossible, given the direct relationship between the history of unsustainability and the academy (Carp 2013), and the relationship between histories of violence and the production of academic knowledge. Pertinent examples of this are readily visible across the academy: consider the discipline of geography, whose extended engagement with militarism has been argued to be directly tied to settler colonialism and white supremacy (Inwood and Bonds 2016), or the discipline of English which has a long colonial history tied to language and canon (Ngũgĩ 1986; Said 1979), or the history of instructional design and technology which is tied to militarism and war (Reiser 2001). This is to say that if radical flexibility begins with the premise of life as relational, that relationality must extend to the awareness that all are materially bound to the earth and the resources drawn from the earth, as well as to each other, and as such, sustainable futures are inherently connected to that reality and the histories and legacies that shape its future.
We are doubtful that the university as it existed before the pandemic was capable of enacting the kind of radically flexible education outlined above in a robust way. With respect to the climate emergency, Carp (2013: 229) questions whether it is even possible for the academy today to become ecologically sustainable, even though he notes the inevitability of change, that ‘we will either help to shape it and learn to ride it, or we will be inundated by it’. But abrupt and likely permanent change has already arrived in the form of the pandemic, a crisis which Hall (2020: 6) argues will not find its solutions in academia, as ‘[t]he capitalist University cannot save us, because it is driven by short-term economic interests, rather than the long-term conditions of life’. But perhaps it is within this crisis that those of us willing to might make something more out of our circumstances, that especially in this darkness and uncertainty, that we might find hope and the strength to change, to reimagine, and collectively bring into being something new in a way that has long been necessary. Solnit (2020) urges us to remember that ‘[o]rdinary life before the pandemic was already a catastrophe of desperation and exclusion for too many human beings, an environmental and climate catastrophe, an obscenity of inequality’, and this was in many ways as true in the halls of education as anywhere else. But she further reminds us that hope ‘offers us clarity that, amid the uncertainty ahead, there will be conflicts worth joining and the possibility of winning some of them’. If, out of this struggle, we ground our hope in attention to the relational nature of the many worlds in which we all live together, then perhaps we can achieve the radical flexibility truly liberatory education deserves.
Funding This research was undertaken, in part, thanks to funding from the Canada Research Chairs Program and the Commonwealth of Learning Research Chairs program.
Compliance with Ethical Standards
Conflict of Interest The authors declare that they have no conflict of interest.
Code Availability Not applicable.
**Open Access** This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
To view a copy of this licence, visit [http://creativecommons.org/licenses/by/4.0/](http://creativecommons.org/licenses/by/4.0/).
**References**
Alexander, B. (2020). One example of climate change directly impacting higher education. 12 June. [https://bryanalexander.org/climatechange/one-example-of-climate-change-directly-impacting-higher-education/](https://bryanalexander.org/climatechange/one-example-of-climate-change-directly-impacting-higher-education/). Accessed 17 June 2020.
Allen, J., Rowan, L., & Singh, P. (2020). Teaching and teacher education in the time of COVID-19. *Asia-Pacific Journal of Teacher Education, 48*(3), 233–236. [https://doi.org/10.1080/1359866X.2020.1752051](https://doi.org/10.1080/1359866X.2020.1752051).
Atleo, R. E. (2011). *Principles of Tsawalk: An Indigenous approach to global crisis*. Vancouver, BC: University of British Columbia Press.
Ban, S. (2020). Review of *Disability in higher education: A social justice approach*. N. J. Evans, E. M. Broido, K. R. Brown, & A.K. Wilke, eds. *Disability Studies Quarterly, 40*(1). [https://dsq-sds.org/article/view/7410/5501](https://dsq-sds.org/article/view/7410/5501). Accessed 15 September 2020.
Barnett, R. (2014). Conditions of flexibility: Securing a more responsive higher education system. York, UK: Higher Education Academy. [https://www.advance-he.ac.uk/knowledge-hub/conditions-flexibility-securing-more-responsive-higher-education-system](https://www.advance-he.ac.uk/knowledge-hub/conditions-flexibility-securing-more-responsive-higher-education-system). Accessed 22 September 2020.
Bayley, A. (2018). Posthumanism, decoloniality, and re-imagining pedagogy. *Parallax, 24*(3), 242–253. [https://doi.org/10.1080/13534645.2018.1496576](https://doi.org/10.1080/13534645.2018.1496576).
Bozkurt, A., Jung, I., Xiao, J., Vladimirschi, V., Schuwer, R., Egorov, G., et al. (2020). A global outlook to the interruption of education due to COVID-19 pandemic: Navigating in a time of uncertainty and crisis. *Asian Journal of Distance Education, 15*(1), 1–126. [https://asianjde.org/ojs/index.php/AsianJDE/article/view/462](https://asianjde.org/ojs/index.php/AsianJDE/article/view/462). Accessed 22 September 2020.
Braidotti, R. (2013). *The posthuman*. Boston, MA: Polity.
Carp, R. M. (2013). Toward a resilient academy. In L. F. Johnston (Ed.), *Higher education for sustainability: Cases, challenges, and opportunities from across the curriculum* (pp. 223–237). New York, NY: Routledge.
Centre for Disease Control. (2020). COVID-19 in racial and ethnic minority groups. Center for Disease Control and Prevention. [https://www.cdc.gov/coronavirus/2019-ncov/need-extra-precautions/racial-ethnic-minorities.html](https://www.cdc.gov/coronavirus/2019-ncov/need-extra-precautions/racial-ethnic-minorities.html). Accessed 20 June 2020.
Costello, E., Brown, M., Donlon, E., & Girme, P. (2020). ‘The pandemic will not be on zoom’: A retrospective from the year 2050. *Postdigital Science and Education*. [https://doi.org/10.1007/s42438-020-00150-3](https://doi.org/10.1007/s42438-020-00150-3).
Crawley, A. (2018). Introduction to the Academy and what can be done? *Critical Ethnic Studies, 4*(1), 4–19. [https://doi.org/10.5749/critethnstud.4.1.0004](https://doi.org/10.5749/critethnstud.4.1.0004).
Crutzen, P. J. (2002). Geology of mankind. *Nature, 415*, 23. [https://doi.org/10.1038/415023a](https://doi.org/10.1038/415023a).
Daniel, J. (1998). Can you get my hard nose in focus? Universities, mass education and appropriate technology. In M. Eisenstadt & T. Vincent (Eds.), *The knowledge web: Learning and collaborating on the net* (pp. 21–29). London: Kogan Page.
Doherty, N., Madriaga, A., Johnson, A., Waldman, A., Martin, N., Buford, T., & Briscoe, T. (2020). The first 100: COVID-19 took black lives first. It didn’t have to. Propublica, 9 May. [https://features.propublica.org/chicago-first-deaths/covid-coronavirus-took-black-lives-first/](https://features.propublica.org/chicago-first-deaths/covid-coronavirus-took-black-lives-first/). Accessed 17 June 2020.
Stein, S. (2020). A colonial history of the higher education present: rethinking land-grant institutions through processes of accumulation and relations of conquest. *Critical Studies in Education, 61*(2), 212–228. https://doi.org/10.1080/17508487.2017.1409646
Stommel, J. (2018). How to ungrade. https://www.jessestommel.com/how-to-ungrade/. Accessed 18 June 2020.
Swauger, S. (2020). Our bodies encoded: Algorithmic test proctoring in higher education. Hybrid pedagogy, 2 April. https://hybridpedagogy.org/our-bodies-encoded-algorithmic-test-proctoring-in-higher-education/. Accessed 31 July 2020.
Tate, S. A., & Bagguley, P. (2016). Building the anti-racist university: Next steps. *Race, Ethnicity & Education, 3*, 289–299. https://doi.org/10.1080/17508487.2016.1260227.
Todd, Z. (2020). Relationships. In C. Howe & A. Pandian (Eds.), *Anthropocene unseen: A lexicon* (pp. 381–384). Santa Barbara, CA: Punctum Books. https://doi.org/10.21983/P3.0265.1.63.
Valcarlos, M. M., Wolgemuth, J., Haraf, S., & Fisk, N. (2020). Anti-oppressive pedagogies in online learning: A critical review. *Distance Education, 41*(3), 345–360. https://doi.org/10.1080/01587919.2020.1763783.
Veletsianos, G. (2020). *Learning online: The student experience*. Baltimore: Johns Hopkins University Press.
Veletsianos, G., & Houlden, S. (2019). An analysis of flexible learning and flexibility over the last 40 years of *Distance Education*. *Distance Education, 40*(4), 454–468. https://doi.org/10.1080/01587919.2019.1681893.
Walsh, J. D. (2020). The coming disruption: Scott Galloway predicts a handful of elite cyborg universities will soon monopolize higher education. Intelligencer, 11 May. https://nymag.com/intelligencer/2020/05/scott-galloway-future-of-college.html. Accessed 12 September 2020.
Welch, N. (2020). A semester to die for: Dispatches from the frontlines of care. Spectre, 6 July. https://spectrejournal.com/a-semester-to-die-for/. Accessed 6 August 2020.
Whyte, K. P. (2013). Justice forward: Tribes, climate adaptation and responsibility. *Climatic Change, 120*, 517–530. https://doi.org/10.1007/s10584-013-0743-2.
Wilson, S. (2008). *Research is ceremony: Indigenous research methods*. Black Point, NS: Fernwood Publishing.
Witze, A. (2020). Universities will never be the same after coronavirus crisis. *Nature, 582*, 162–164. https://doi.org/10.1038/d41586-020-01518-y.
Wolfe, C. (2009). *What is posthumanism?* Minneapolis, MN: University of Minnesota Press.
Xie, R., Blume, H., & Money, L., (2020). USC, school districts getting ‘Zoom-bombed’ with racist taunts, porn as they transition to online meetings. Los Angeles Times, 25 March. https://www.latimes.com/california/story/2020-03-25/zoombombing-usc-classes-interrupted-racist-remarks. Accessed 6 August 2020. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
1993,
1
],
[
1993,
5475,
2
],
[
5475,
8962,
3
],
[
8962,
12759,
4
],
[
12759,
16420,
5
],
[
16420,
19843,
6
],
[
19843,
23481,
7
],
[
23481,
27438,
8
],
[
27438,
31147,
9
],
[
31147,
34718,
10
],
[
34718,
39561,
11
],
[
39561,
39561,
12
],
[
39561,
39561,
13
],
[
39561,
42343,
14
]
]
} |
4a29871f166a2a783f267839a6f7190d8fdf7a73 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 14,
"total-fallback-pages": 0,
"total-input-tokens": 40979,
"total-output-tokens": 12640,
"fieldofstudy": [
"Environmental Science"
]
} | RESEARCH ARTICLE
No increasing risk of a limnic eruption at Lake Kivu: Intercomparison study reveals gas concentrations close to steady state
Fabian Bärenbold1*, Bertram Boehrer2, Roberto Grilli3, Ange Mugisha4, Wolf von Tümpling2, Augusta Umutoni4, Martin Schmid1
1 Eawag, Swiss Federal Institute of Aquatic Science and Technology, Surface Waters—Research and Management, Kastanienbaum, Switzerland, 2 Helmholtz-Centre for Environmental Research–UFZ, Magdeburg, Germany, 3 CNRS, Université Grenoble Alpes, IRD, Grenoble INP, Institut des Géosciences de l’environnement, Grenoble, France, 4 Lake Kivu Monitoring Programme LKMP, Gisenyi, Rwanda
* [email protected]
Abstract
Lake Kivu, East Africa, is well known for its huge reservoir of dissolved methane (CH4) and carbon dioxide (CO2) in the stratified deep waters (below 250 m). The methane concentrations of up to ~ 20 mmol/l are sufficiently high for commercial gas extraction and power production. In view of the projected extraction capacity of up to several hundred MW in the next decades, reliable and accurate gas measurement techniques are required to closely monitor the evolution of gas concentrations. For this purpose, an intercomparison campaign for dissolved gas measurements was planned and conducted in March 2018. The applied measurement techniques included on-site mass spectrometry of continuously pumped sample water, gas chromatography of in-situ filled gas bags, an in-situ membrane inlet laser spectrometer sensor and a prototype sensor for total dissolved gas pressure (TDGP). We present the results of three datasets for CH4, two for CO2 and one for TDGP. The resulting methane profiles show a good agreement within a range of around 5–10% in the deep water. We also observe that TDGP measurements in the deep waters are systematically around 5 to 10% lower than TDGP computed from gas concentrations. Part of this difference may be attributed to the non-trivial conversion of concentration to partial pressure in gas-rich Lake Kivu. When comparing our data to past measurements, we cannot verify the previously suggested increase in methane concentrations since 1974. We therefore conclude that the methane and carbon dioxide concentrations in Lake Kivu are currently close to a steady state.
Introduction
Lake Kivu, with a surface area of 2386 km² and a maximum depth of 485 m, is situated on the border between Rwanda and the Democratic Republic of the Congo (DRC). Along with other African great lakes Tanganyika and Malawi, Lake Kivu is part of the East African Rift System (EARS). To the north, Lake Kivu borders on the Virunga volcano chain, while to the South it drains into Lake Tanganyika via the Ruzizi River. Lake Kivu is fed by numerous small streams
from the European Community’s Seventh Framework Programme ERC-2015-PoC under grant agreement no. 713619 (ERC OCEAN-IDs) and from the Agence Nationale de la Recherche (ANR) under grant agreement ANR-18-CE04-0003-01. Authors AU and AM are staff of the unit LKMP (Lake Kivu Monitoring Programme) of EDCL (Energy Development Corporation Limited) which is a subsidiary of the Rwandan Energy Group Limited (REG). EDCL/LKMP supported the involved research teams with local transport and help with field work. In addition, they paid the flight to Rwanda and accommodation in Rwanda for the participants of UFZ (authors BB and WvT) and CNRS (author RG). The specific roles of these authors are articulated in the ‘author contributions’ section. The funders had no other role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have read the journal’s policy and the authors of this manuscript have the following competing interests: Authors AU and AM are affiliated to EDCL (Energy Development Corporation Limited) which is a subsidiary of the Rwandan Energy Group Limited (REG). The latter is a government-owned company that is responsible for the import, export, procurement, generation, transmission, distribution and sale of electricity in Rwanda. Within EDCL, AU is leading the Lake Kivu Monitoring Programme (LKMP), which is responsible for monitoring the impacts of methane gas extraction on Lake Kivu. Because of their interest in high-accuracy gas measurements, the LKMP/EDCL supported the involved research teams with local transport and help with field work. In addition, they paid the flight to Rwanda and accommodation in Rwanda for the participants of UFZ (authors BB and WvT) and CNRS (author RG). There are no patents, products in development or marketed products to declare. This does not alter our adherence to PLOS ONE policies on sharing data and materials.
[1] and by subaquatic groundwater sources [2] with the latter contributing about 45% of the total inflow. The groundwater sources mainly enter the lake at the northern shore and can be divided into two categories: two cool and fresh sources above a depth of 260 m and several warm, saline and carbon dioxide (CO₂)-rich sources below 260 m. This has two main consequences, namely a very stable density stratification due to the salinity gradient, which prevents annual mixing below a depth of 50 to 60 m and the accumulation of dissolved CO₂ over long time scales. In addition to CO₂, biogenic methane (CH₄) is present in the deep waters in large amounts due to decomposition of organic matter at the lake bottom and CO₂ reduction [3, 4].
Gas concentrations in Lake Kivu were first recorded by Damas in 1935 [5] who measured CO₂ and H₂S. However, Damas only analyzed the sample water after degassing, thus losing more than half of the CO₂ to the atmosphere. Between 1952 and 1954, Schmitz and Kufferath carried out the first CH₄ measurements and additionally determined CO₂ concentrations. However, they only analyzed the gas that outgassed under atmospheric conditions, neglecting the gas remaining dissolved in the water [6]. In 1974, Tietze performed the first comprehensive survey of dissolved gas concentrations, including CH₄ and CO₂ from both the gas exsolved under atmospheric conditions and the remaining part in the sample water [7]. Tietze concluded that about 300 km³ STP (gas volume at 0°C and 1 atm) of CO₂ and 60 km³ STP of CH₄ were stored in the permanently stratified deep waters (below ~ 60 m) of Lake Kivu [7]. Subsequently, based on new measurements from M. Halbwachs and J.-C. Tochon in 2003 (published in [8]), Schmid et al., 2005 suggested that CH₄ concentrations had increased by 15% since 1974 and that they could possibly reach saturation within the 21st century. With the examples of deadly limnic eruptions due to high gas loads in Lakes Nyos [9] and Monoun [10], it was clear that the gas concentrations of Lake Kivu needed to be monitored. Besides the threat to the local population, the gas content in Lake Kivu also represents a valuable resource: In December 2015, a 26 MW gas power plant was connected to the Rwandan grid and several hundred MW could follow according to projections [11].
In 2017, a gas intercomparison campaign was initiated by the Lake Kivu Monitoring Programme (LKMP) with the goal of 1) accurately determining CH₄ and CO₂ concentrations using different methodologies and 2) finding an appropriate technique to regularly monitor the gas concentrations in the future. However, gas sampling in highly outgassing environments is challenging and thus, the measurement methods had to be adapted accordingly. In this work, we describe the methodologies for three research teams involved in the campaign: The Swiss Federal Institute of Aquatic Science and Technology (Eawag), the Helmholtz Centre for Environmental Research (UFZ) and the National Center of Scientific Research in France (CNRS). Subsequently, we present the results of each group and compare them to the previous measurements of Tietze in 1974 [7] and Halbwachs and Tochon in 2003 and Schmid in 2004 (both published in [8]). Finally, we reevaluate the gas (CH₄ and CO₂) content in Lake Kivu and its potential change in time.
Materials and methods
The intercomparison campaign took place close to Gisenyi/Rubavu, Rwanda (1.74087°S / 29.22602°E) from 9 to 13 March 2018 and involved research teams from Eawag, UFZ, CNRS and from the power plant operator KivuWatt. Eawag prolonged its measurement period until 18 March and UFZ also included earlier measurements from 2017. The campaign was planned and organized by LKMP and therefore, no special permit was necessary to perform measurements on the lake. Further details on the results of the campaign can also be found in a report to LKMP [12]. Note that while the report includes the measurements of KivuWatt, the latter decided to not be part of this publication.
The measurements taken by each research team are summarized in Table 1. In this publication, only the approach of Eawag is explained more comprehensively, while further details on the methods of UFZ and CNRS can be found elsewhere ([13, 14]). In the following, we will first present the methodology of Eawag and then shortly summarize the approaches of UFZ and CNRS.
Measurement method used by Eawag
The measurement setup was built around “miniRuedi”, a gas-equilibrium portable membrane-inlet mass spectrometric system (GE-MIMS) which allows on-site quantification of different dissolved gases in water (i.e. N₂, O₂, CO₂, CH₄, He, Ar, see [15]). The continuous sampling water flow (~ 1 L/min) required to maintain gas equilibrium at the MS inlet was provided by a submersible pump (0.75 kW, Lechner Pumpen) and 250 m long, 6 mm inner diameter polyamide (PA) tubing. The pump was used only above 250 m and yielded a flow of ~ 1.6 L/min. Below 250 m, TDGP increases drastically and, following initiation of the flow by a suction pump at the surface, the buoyancy due to bubble formation within the tube was sufficient to lift the water to the surface. 10 mm PA tubing was used in this case and the resulting flow was ~ 1 L/min (except between 270 and 310 m where it was ~ 0.5 L/min only). The water-gas mixture was subsequently dispersed through a nozzle into a custom-made cylindrical equilibration chamber (12.3 cm diameter, 38 cm height, see Fig 1 in S1 Appendix). While the degassed sample water accumulated and discharged at the bottom of the chamber, the gas phase stayed above and left the chamber through a tube at the top. The gas content in the gas phase and the water phase (via a headspace created by the membrane contactor Liqui-Cel G542, 260 cm³ external volume) was analyzed by the “miniRuedi”. Finally, gas and water flow rates were recorded to compute in-situ gas concentrations in the lake as sketched in Fig 1 in S1 Appendix. The overall analytical accuracy (i.e. the maximum deviation from the true value) of the setup was deduced from the accuracies of its individual components and estimated to around ± 5% for CO₂ and ± 10% for CH₄ in the deep water (see S1 Appendix for more details). Measurements were done at a resolution of 20 m starting from 10 m depth down to 450 m (430 m was omitted due to time constraints). Between 90 and 130 m, the gas flow was too low to be quantified but still substantial enough to have an effect on gas results. Therefore, results for this depth range are not reported.
The mass spectrometer was calibrated using two gas standards (80% CO₂ + 20% CH₄ and 60% CO₂ + 30% CH₄ + 10% air) with partial pressures similar to the average gas composition of water gassing out from Lake Kivu deep water. One of the gas standards additionally contained atmospheric air for potential calibration of N₂ and O₂. However, in the special setting of Lake Kivu, the determination of N₂ at mass/charge = 28 proved to be difficult because of the presence of a large peak of CO from the fragmentation of CO₂ during ionization in the mass spectrometer. The interference of the CO fragment accounted for more than 95% of the intensity at mass 28. Therefore N₂ could not be determined reliably and hence was not included in this publication.
Table 1. Summary of gas measurements performed by the different research teams of Eawag, UFZ and CNRS.
| | CH₄ 0–150 m | CH₄ 150–450 m | CO₂ | TDGP |
|----------|-------------|---------------|----------|------|
| Eawag | - | + | + | - |
| UFZ | - | + | + | + |
| CNRS | + | - | - | - |
The “+” indicates which measurements were performed by which groups.
https://doi.org/10.1371/journal.pone.0237836.t001
Measurement method used by UFZ
The measurement method used by UFZ had previously been used in highly gas charged mine pit lakes (for CO$_2$ see [16], for CH$_4$ see [17]) and was modified for the conditions of Lake Kivu by [13]. Water was sampled using gas-tight bags, which were lowered to the appropriate depth together with a small pump and an automatic pump controller. The pump partially filled the bags while leaving enough space for the gas phase, which forms once the bags are retrieved. At the surface, the water and gas phases in the bags were equilibrated over night and the composition of the gas phase was analyzed using a gas chromatograph. Subsequently, the remaining amount of gas in the water phase was deduced by assuming equilibrium between gas and water phase. In order to compute in-situ gas concentrations, the gas and water volumes in the bag were determined using a syringe and a laboratory scale respectively. Total uncertainties for CH$_4$ (CO$_2$) concentrations were determined as ±5 (±6) % below and ±7 (±8) % above 250 m. Note again that these uncertainties should be interpreted as maximum deviation from the true value. The UFZ group also measured total dissolved gas pressure (TDGP) using a prototype probe from Pro Oceanus with an accuracy of ± 0.04 bar according to the manufacturer.
Measurement method used by CNRS
The measurement method applied by CNRS is fully described elsewhere [14]. In short, an in-situ membrane-inlet laser spectrometer (MILS), called SubOcean, was deployed for continuous dissolved CH$_4$ measurements. The instrument is based on a patented membrane extraction system [18] coupled to an optical spectrometer for trace gas sensing based on an optical feedback cavity enhanced absorption spectroscopy (OFCEAS) technique [19, 20]. The extraction system does not rely on gas equilibration across the membrane, but the dry side of the membrane is maintained at low pressure while continuously flushing it with dry zero air. This allows achieving fast response times < 30 sec, making the technique adapted for fast 3D mapping of water masses [21]. The accuracy (standard deviation at the three sigma level) of the measurements was quantified to be ±33% from repeated measurements at the same depth. Note that the uncertainty is given at the three sigma level, which we judge roughly equivalent to the concept of “maximum deviation from the true value” used for the other two methods. In addition to the CH$_4$ signal, this method needs external TDGP and dissolved CO$_2$ measurements in order to compute CH$_4$ partial pressure in the lake water, which is finally converted to concentrations using the conversion method presented below.
Conversion from concentration to partial pressure
Eawag and UFZ measured gas concentrations while the MILS sensor used by CNRS provided partial gas pressure. In order to compare these results, conversion from concentration to partial pressure and vice-versa is required. However, this conversion is not straightforward as the gas-water partition coefficients (Henry coefficients) depend on temperature, salinity and hydrostatic pressure [22]. In addition, the fugacity effect cannot be calculated separately for each gas since it depends on the relative mixture of the involved gas species [23]. We thus express concentration $C_i$ as a function of partial pressure $p_i$ by the following equation:
$$C_i = K_i(T, S, P)p_i\phi_i(T, P)$$
with $\phi_i$ the fugacity coefficient i.e. the ratio between the fugacity of a gas and its partial pressure at temperature $T$ and pressure $P$, and $K_i$ the solubility coefficient, i.e. the ratio between the dissolved concentration of a gas and its fugacity. The solubility coefficient $K_i(T, S)$ is computed as a function of temperature and salinity according to [22] for CO$_2$, [24] for CH$_4$ and [25] for N$_2$.
The salinity terms of these equations were originally derived for sea salt and not for Lake Kivu, where salinity is dominated by bicarbonates. We accounted for this by assuming that the salinity effect mainly depends on the ionic strength of the solution. More details about the salinity correction are provided in S2 Appendix.
According to [22], the dependence of the solubility coefficient $K_i$ on local total pressure (hydrostatic plus atmospheric pressure) can be written as
$$K_i(T, S, P) = K_i(T, S) e^{-\frac{P - P_0}{R T}}. \quad (2)$$
Here, $R = 83.1446 \text{ cm}^3 \text{ bar K}^{-1} \text{ mol}^{-1}$ is the gas constant, and $\nu_i$ are the partial molar volumes (cm$^3$ mol$^{-1}$). The partial molar volumes of CO$_2$ and N$_2$ were assumed constant at 32.3 cm$^3$ mol$^{-1}$ [22] and 35.7 cm$^3$ mol$^{-1}$ [26] respectively, while for CH$_4$ it was calculated according to [27].
The resulting pressure correction factors $K_i(T, S, P)/K_i(T, S)$ range between 1.00 at atmospheric pressure and 0.93 (CO$_2$) or 0.94 (CH$_4$) at 50 bar (485 m depth), i.e. the in-situ pressure reduces the solubility coefficient of the gases by 6 to 7% in the lowest layers of Lake Kivu.
The fugacity coefficients $\phi_i(T, P)$ of CO$_2$, CH$_4$ and N$_2$, including the interaction between the gases, are computed according to [23] (Octave scripts available in S3 Appendix).
Results and discussion
Eawag results
The resulting CH$_4$ and CO$_2$ concentrations using the Eawag mass spectrometer setup are depicted in Fig 1, along with temperature and salinity profiles. We can identify the well-mixed epilimnion with constant salinity (above ~ 60 m) and the main chemocline at ~ 255 m. As expected, gas concentrations correlate well with salinity and temperature due to the common hydrothermal origin of CO$_2$, dissolved solids and heat [8]. The detailed CH$_4$ and CO$_2$ results can be found in S1 Table.
Contamination with atmospheric air is a major source of measurement errors in gas content analysis. Thus, we use our O$_2$ results to estimate the maximum atmospheric contamination affecting our measurements in the completely anoxic deep waters of Lake Kivu. We find that the mixing ratio of O$_2$ in the sampled gas phase is always less than 1% (maximum between 270 and 310 m due to lower water flow at these depths). Most likely, this O$_2$ signal indicates a small contamination with atmospheric air. However, it could stem from gas fragmentation during ionization in the mass spectrometer. Independently of its origin, the signal is small enough to not significantly affect the CH$_4$ and CO$_2$ results.
Intercomparison of CH$_4$ and CO$_2$ using past and new measurements
The results of the dissolved gas measurements (CH$_4$ and CO$_2$) of Eawag, UFZ [13] and CNRS [29] are shown in Fig 2. For both CH$_4$ and CO$_2$, the measurements agree well within the uncertainties of the different approaches. The profile from Eawag shows higher CH$_4$ concentrations (up to 10%) than UFZ between 250 and 350 m depth, whereas UFZ measured higher CH$_4$ and CO$_2$ concentrations (up to 5%) below 400 m. In particular, the UFZ profile indicates further increasing CH$_4$ and CO$_2$ concentrations with depth below 400 m, while the Eawag profile levels off or even decreases. However, note that the comparison below 400 m is based on very few measurement points. The results of Eawag, UFZ and CNRS show a good agreement at their junction at 150 m, thus validating the conversion method under moderate hydrostatic pressure.
In order to estimate the total gas content in the lake, we need to derive quasi-continuous depth profiles for CH$_4$ and CO$_2$ from the measurements depicted in Fig 2. We chose to interpolate the discrete profiles by fitting them to an electric conductivity profile (corrected to 25˚C), because i) conductivity is most probably closely related to gas concentrations due to the long residence time in the lake and because similar transport processes affect both dissolved solids and gases [30] and ii) it can be easily measured at a high resolution.
The following procedure was applied to derive high-resolution curves for the CH$_4$ and CO$_2$ concentrations measured by UFZ and Eawag: The conductivity profile from Fig 1 was extended down to 480 m depth using the background conductivity profile published by [2]. The latter was corrected with the mean difference between both profiles in their lowest common 20 m. From this profile, we then extracted the conductivity values at the depths of the gas measurements of UFZ and Eawag. Then, a 6$^{th}$ order polynomial function was fitted ($R^2 > 0.995$ for all four profiles) with conductivity as the independent and gas concentrations as the dependent variables. The regression was used to compute the gas concentration as a function of conductivity and to relate it to depth. The resulting curves are shown in Fig 3 (at 0.5 m
resolution), along with previous CH$_4$ and CO$_2$ measurements. The uncertainties of the previous measurements were assumed to be ±5% for Tietze [7], ±4% for Halbwachs and Tochon [8] and ±10% for Schmid ([8] and pers. comm. M. Schmid).
In general, previous and current measurements are in good agreement. For both CH$_4$ and CO$_2$, the measurements of 1974 are at the lower end of the spectrum, and those of 2003 at the higher end. From this fact, [8] concluded that CH$_4$ concentrations had increased by 15% from 1974 to 2003 and that they could possibly reach saturation within the 21st century. Pasche et al. [4] later determined an upper bound for the CH$_4$ increase of around half this value based on carbon cycle analysis. However, our new measurements show no measurable increase of CH$_4$ (and CO$_2$) concentrations within the last 45 years. This implies that the measured differences
were largely due to measurement uncertainty and that the CH$_4$ and CO$_2$ concentrations in the lake are currently close to a steady state.
**Risk assessment using total dissolved gas pressure (TDGP)**
In order to assess the danger of a potential gas eruption associated with the high gas concentrations in Lake Kivu, it is helpful to look at gas pressure saturation within the lake. CH$_4$ and CO$_2$ concentrations thus need to be converted to partial pressure using the conversion method presented in the previous section. Besides CH$_4$ and CO$_2$, dissolved nitrogen (N$_2$) is the only gas present in sufficient amounts to influence gas pressure. As no N$_2$ data is available for Lake Kivu,
Kivu, we estimated its contribution assuming that it mimics the profile of atmospheric noble gases which show concentrations close to air saturated water (ASW) at the lake surface and a decrease of ~50% in the deep water [31]. The derived N\textsubscript{2} profile was subsequently included in the conversion algorithm, which includes the effect of gas mixture between CH\textsubscript{4}, CO\textsubscript{2} and N\textsubscript{2} on the fugacity coefficients.
The accuracy of the calculated TDGP is estimated from the accuracies of the gas/water flow measurements and the CH\textsubscript{4} concentration in the gas phase. The contributions to the accuracy from CO\textsubscript{2} and N\textsubscript{2} are negligible in comparison to CH\textsubscript{4} because of the high solubility of CO\textsubscript{2} and low concentration of N\textsubscript{2}.
Fig 4 shows that calculated TDGP of Eawag and UFZ and direct TDGP measurements using the Pro Oceanus sensor are in good agreement, usually well within the uncertainties of the respective methods. Still, below 250m depth, TDGP calculated from Eawag and UFZ data is slightly higher than the measured TDGP. The mean difference between calculated and measured TDGP is 5.9% (maximum of 8.7% at 290 m) and 4.4% (maximum of 8.8% at 417 m) for Eawag and UFZ respectively. This discrepancy could be due to a bias in the conversion of concentrations to partial pressures, to an overestimation of concentrations by both Eawag and UFZ or to a problem of calibration of the TDGP sensor at high pressure.
Gas concentrations are very high in the deep water of Lake Kivu. If this gas was released to the atmosphere, it would cause a large catastrophe by suffocating humans and animals in the surrounding area, qualitatively similar to the events at Lake Nyos in 1986 [9]. Currently, total dissolved gas pressure (TDGP) is well below absolute pressure (hydrostatic plus atmospheric pressure) at all depths in Lake Kivu. The maximum gas saturation in terms of pressure is reached at 320 m and amounts to ~ 50% (or a maximum of 57% if we take the upper limit of the uncertainty range of the Eawag data). This means that the gas concentrations are still far
away from the point of spontaneous ebullition (i.e. close to 100% saturation). Nevertheless, volcanic structures on the lake floor indicate frequent volcanic activity within the lake in the geologically recent past [32]. We cannot exclude that similar volcanic activity could trigger a gas eruption from the lake in the future, even though TDGP is far away from saturation. Therefore, in spite of no measurable increase of gas concentrations during the last 45 years, artificial degassing is still beneficial to reduce the danger of a potential natural disaster.
### Update of Lake Kivu gas reserves
We estimated the gas content of Lake Kivu by multiplying our interpolated gas profiles (Fig 3) by the lake area at each depth and subsequent integration over the lake depth at a resolution of 0.5 m. The lake areas were deduced from the bathymetry of Lake Kivu by K.A. Ross from the blended bathymetric data of [32, 33]. Tables 2 and 3 show the gas content in different depth ranges for CH$_4$ and CO$_2$ respectively. The average CH$_4$ estimate from the 2018 campaign
Table 2. CH$_4$ content in Lake Kivu in km$^3$ STP for different depth ranges.
| Depth range [m] | Eawag 2018 | UFZ 2018 | CNRS 2018 | Average | Halbwachs 2003 |
|-----------------|------------|----------|-----------|---------|----------------|
| 0–70 | - | - | 0.1 ± 0.3 | 0.1 | 0.1 |
| 70–150 | - | - | 6.5 ± 2.1 | 6.5 | 6.5 |
| 150–200 | 5.8 ± 1.6 | 5.7 ± 0.4 | - | 5.7 | 8.5 |
| 200–260 | 8.5 ± 1.3 | 8.2 ± 0.5 | - | 8.3 | 8.3 |
| 260–300 | 12.5 ± 1.7 | 12.1 ± 0.6 | - | 12.3 | 12.3 |
| 300–350 | 14.7 ± 1.6 | 13.8 ± 0.7 | - | 14.2 | 14.2 |
| 350–400 | 9.5 ± 0.9 | 9.4 ± 0.5 | - | 9.5 | 9.5 |
| 400–480 | 5.5 ± 0.5 | 5.7 ± 0.3 | - | 5.6 | 5.6 |
| Resource zone (260–480) | 42.2 ± 4.8 | 40.9 ± 2.0 | - | 41.5 | 44.7 |
| Upper resource zone (260–310) | 15.6 ± 2.1 | 15.0 ± 0.7 | - | 15.3 | 15.3 |
| Lower resource zone (310–480) | 26.6 ± 2.6 | 26.0 ± 1.3 | - | 26.3 | 26.3 |
| Entire lake | 62.2 ± 6.9 | 65.1 ± 4.8 | - | 65.1 | 65.1 |
The reference values (Halbwachs 2003) were calculated in [34] based on the data of M. Halbwachs and J.-C. Tochon in [8]. The resource zones are defined as in [11], but including half of the bordering gradients.
https://doi.org/10.1371/journal.pone.0237836.t002
Table 3. CO$_2$ content in Lake Kivu in km$^3$ STP for different depth ranges.
| Depth range [m] | Eawag 2018 | UFZ 2018 | Average | Halbwachs 2003 |
|-----------------|------------|----------|---------|----------------|
| 0–150 | 24.7 ± 2.1 | - | 24.7 | 24.7 |
| 150–200 | 23.4 ± 1.5 | 21.7 ± 1.5 | 22.5 | 22.5 |
| 200–260 | 37.0 ± 2.3 | 37.0 ± 2.4 | 37.0 | 38 |
| 260–300 | 56.2 ± 4.2 | 55.2 ± 2.8 | 55.7 | 55.7 |
| 300–350 | 68.7 ± 4.0 | 66.8 ± 3.4 | 67.7 | 67.7 |
| 350–400 | 47.2 ± 2.4 | 48.2 ± 2.4 | 47.7 | 47.7 |
| 400–480 | 28.1 ± 1.4 | 30.1 ± 1.5 | 29.1 | 29.1 |
| Resource zone (260–480) | 200.2 ± 12.1 | 200.3 ± 10.0 | 200.2 | 214 |
| Upper resource zone (260–310) | 69.9 ± 5.1 | 68.2 ± 3.4 | 69.1 | 69.1 |
| Lower resource zone (310–480) | 130.3 ± 6.9 | 132.1 ± 6.6 | 131.2 | 131.2 |
| Entire lake | 285.3 ± 18.0 | 284.5 ± 22.0 | 284.5 | 284.5 |
The reference values (Halbwachs 2003) were calculated in [34] based on the data of M. Halbwachs and J.-C. Tochon in [8]. The resource zones are defined as in [11], but including half of the bordering gradients.
https://doi.org/10.1371/journal.pone.0237836.t003
shows slightly lower values in the resource zone (7%) and in the entire lake (4.5%) than calculated from the data of Halbwachs and Tochon 2003 (published in [8]). Similarly, the CO\textsubscript{2} content measured in 2018 is 6.5% lower in the resource zone and 3% lower for the entire lake. We do not think that this apparent decrease in gas concentrations since 2003 reflects the real gas dynamics in Lake Kivu because i) the total CH\textsubscript{4} extracted by the existing power plant was less than 0.2 km\textsuperscript{3} until March 2018 [11] and therefore not measurable by current methods, ii) the residence time of gases in the deep water is on the order of 1000 years [34] and iii) to our knowledge, there is no process that would consume either CO\textsubscript{2} or CH\textsubscript{4} under the conditions present in the deep water of Lake Kivu (i.e. below 70 m [35]).
Schmid et al. [8] suggested a CH\textsubscript{4} production rate of 120 g C/m\textsuperscript{2}/year (grams of carbon in CH\textsubscript{4} per sediment area per year) in order to explain the difference between the CH\textsubscript{4} profiles of K. Tietze in 1974 and M. Halbwachs and J.-C. Tochon in 2003. This rate would lead to a CH\textsubscript{4} increase of about 5–10% since 2003 (i.e. 3–6 km\textsuperscript{3}) and thus can be excluded based on our data. Similarly, we can rule out a production rate of 93 g C/m\textsuperscript{2}/year in the deep water as proposed by [4]. Based on our data from 2018, we suggest that the actual production rate of CH\textsubscript{4} is probably close to the steady state rates of 32 and 35 g C/m\textsuperscript{2}/year calculated by [4, 8] respectively.
We conclude that the variability of gas concentrations measured in the last 45 years is due to the uncertainties of the applied methods. In contrast to previous work [8], our data suggests that the lake gas content is currently close to a steady state with no or small net recharge rate. Consequently, the risk of a gas eruption does not seem to be increasing over time. Additionally, our findings question whether the methane in Lake Kivu is replenished fast enough to be used as a long-term energy source, once the current methane storage has been exploited.
The CH\textsubscript{4} content amounts to around 41.5 km\textsuperscript{3} STP in the resource zone (between 260 and 480 m) and 62.2 km\textsuperscript{3} in the whole lake. Furthermore, the results of the two methodologies suitable for deep water gas analysis (Eawag and UFZ) agree within the expected accuracy of 5–10% for both CH\textsubscript{4} and CO\textsubscript{2}. For regular gas monitoring in view of increased industrial gas extraction, the method of UFZ is easier to apply due to the use of relatively simple equipment. The prototype MILS sensor used by CNRS was able to record gas concentrations down to a depth of 150 m and provides the advantage of in-situ, fast and high-resolution data. However, the technique was not yet adapted to the high CH\textsubscript{4} concentration below 150 m in Lake Kivu. In addition, the sensor requires total dissolved gas pressure (TDGP) and dissolved CO\textsubscript{2} profiles to determine CH\textsubscript{4} partial pressure. The direct measurement of TDGP may be the most appropriate measurable quantity to monitor the risk of a spontaneous ebullition in the lake in the future. It also has the advantage of offering simple, reproducible and high-precision measurements for further monitoring purposes (see for example [36]).
**Supporting information**
S1 Table. Detailed results of Eawag measurement method. (DOCX)
S2 Table. Temperature data from a Sea and Sun CTD. (TXT)
S3 Table. Salinity data computed using conductivity data. (TXT)
S1 Appendix. Detailed description of Eawag measurement method and calculations. (DOCX)
S2 Appendix. Calculation of salinity effect of Lake Kivu dissolved solids.
(DOCX)
S3 Appendix. Octave scripts for conversion of concentration to partial pressure.
(DOCX)
Acknowledgments
Michael Plüss and Reto Brit for help with the development of sampling equipment and fieldwork. Ivo Beck, Maximilian Schmidt (Heidelberg University) and the team of LKMP (Lake Kivu Monitoring Programme), especially Epaphrodite Havugimana and Pierre Simbizi for help with fieldwork. Matthias Brennwald for assistance with the “miniRuedi” mass spectrometer. Serge Robert and Patrick Kathriner for helpful discussions about gas measurements. Nic Spycher (Lawrence Berkeley National Laboratory) for discussing and reviewing the pressure-concentration conversion method and Zaman Ziabakhsh-Ganj (TU Delft) for providing maple scripts for computing fugacity coefficients.
Author Contributions
Conceptualization: Fabian Bärenbold, Bertram Boehrer, Roberto Grilli, Ange Mugisha, Wolf von Tümpling, Augusta Umutoni, Martin Schmid.
Funding acquisition: Martin Schmid.
Investigation: Fabian Bärenbold, Bertram Boehrer, Roberto Grilli, Ange Mugisha, Wolf von Tümpling.
Methodology: Fabian Bärenbold, Bertram Boehrer, Roberto Grilli, Martin Schmid.
Project administration: Ange Mugisha, Augusta Umutoni.
Supervision: Martin Schmid.
Writing – original draft: Fabian Bärenbold, Martin Schmid.
Writing – review & editing: Fabian Bärenbold, Bertram Boehrer, Roberto Grilli, Ange Mugisha, Wolf von Tümpling, Augusta Umutoni, Martin Schmid.
References
1. Muvundza FA, Wüest A, Isumbisho M, Kaningini MB, Pasche N, Rinta P, et al. Modelling Lake Kivu water level variations over the last seven decades. Limnologica-Ecology and Management of Inland Waters. 2014; 47:21–33.
2. Ross KA, Gashugi E, Gafasi A, Wüest A, Schmid M. Characterisation of the subaqueous groundwater discharge that maintains the permanent stratification within Lake Kivu; East Africa. PloS one. 2015; 10 (3) e0121217. https://doi.org/10.1371/journal.pone.0121217 PMID: 25799098
3. Schoell M, Tietze K, Schoberth SM. Origin of methane in Lake Kivu (east-central Africa). Chemical Geology. 1988; 71(1–3):257–265.
4. Pasche N, Schmid M, Vazquez F, Schubert CJ, Wüest A, Kessler JD, et al. Methane sources and sinks in Lake Kivu. Journal of Geophysical Research: Biogeosciences, 2011; 116(G3):G001690.
5. Damas H. La stratification thermique et chimique des lacs Kivu Édouard et Ndalaga (Congo Belge). Internationale Vereinigung für theoretische und angewandte Limnologie: Verhandlungen, 1938; 8 (3):51–68.
6. Schmitz DM, Kufferath J. Problèmes posés par la présence de gaz dissous dans les eaux profondes du lac Kivu. Bulletin des Séances de l’Académie Royale des Sciences Coloniales. 1955; 1:326–356.
7. Tietze K. Geophysikalische Untersuchung des Kivusees und seiner ungewöhnlichen Methangaslagerstätte-Schichtung, Dynamik und Gasgehalt des Seewassers [dissertation]. Christian-Albrechts-Universität zu Kiel; 1978.
8. Schmid M, Haltbwachs M, Wehrli B, Wüest A. Weak mixing in Lake Kivu: new insights indicate increasing risk of uncontrolled gas eruption. Geochemistry, Geophysics, Geosystems. 2005; 6(7), Q07009.
9. Kling GW, Clark MA, Compton HR, Devine JD, Evans WC, Humphrey AM, et al. The 1986 Lake Nyos gas disaster in Cameroon, West-Africa. Science. 1987; 236(4798): 169–175 https://doi.org/10.1126/science.236.4798.169 PMID: 17789781
10. Sigurdsson H, Devine JD, Tchoua FM, Presser TS, Pringle MKW, Evans WC. Origin of the lethal gas burst from Lake Monoun. Cameroun, Journal of Volcanology and Geothermal Research. 1987; 31(1–2):1–16.
11. Expert Working Group on Lake Kivu Gas Extraction. Management prescriptions for the development of Lake Kivu gas resources. Report prepared for the Ministry of Infrastructure of the Republic of Rwanda and the Ministry of Hydrocarbons of the Democratic Republic of the Congo, 2009 June.
12. Schmid M, Bärenbold F, Bohrer B, Darchambeau F, Grilli R, Triest J, et al. Intercalibration Campaign for Gas Concentration Measurements in Lake Kivu, Report prepared for the Lake Kivu Monitoring Programme (LKMP) of the Energy Development Corporation Limited (EDCL), Kigali, Rwanda, 2018.
13. Bohrer B, von Tümpeling W, Mugisha A, Rogemont C, Umutooni A. Reliable reference for the methane concentrations in Lake Kivu at the beginning of industrial exploitation. Hydrol Earth Syst Sci. 2019; 23: 4707–4716
14. Grilli R, Triest J, Chappellaz J, Calzas M, Desbois T, Jansson P, et al. SUB-OCEAN: subsea dissolved methane measurements using an embedded laser spectrometer technology. Environmental Science & Technology. 2018; 52(18): 10543–10551.
15. Brennwald MS, Schmid M, Oser J, Kipfer R. A portable and autonomous mass spectrometric system for on-site environmental gas analysis. Environmental Science & Technology. 2016; 50(24):13455–13463.
16. Bohrer B, Yusta I, Magin K, Sanchez-Espaňa J. Quantifying, assessing and removing the extreme gas load from meromictic Guadiana pit lake, Southwest Spain. Science of the Total Environment. 2016; 563–564:486–477. https://doi.org/10.1016/j.scitotenv.2016.04.124
17. Horn C, Metzler P, Ultrich K, Koschorreck M, Bohrer B. Methane storage and ebullition in monimolimnetic waters of polluted mine pit lake Vollert-Sued, Germany. Science of the Total Environment. 2017; 584–585:1–10. https://doi.org/10.1016/j.scitotenv.2017.01.151 PMID: 28131935
18. Triest J, Chappellaz J, Grilli R. Patent 08276–01: System for fast and in-situ sampling of dissolved gases in the ocean, CNRS, Grenoble, France. 2017.
19. Morville J, Romanini D, Chenevier M. Patent WO03031949, Universite J. Fourier, Grenoble, France. 2003.
20. Morville J, Romanini D, Kerstel E. Cavity enhanced absorption spectroscopy with optical feedback. In: Gagliardi G, Loock HP, editors. Cavity-Enhanced Spectroscopy and Sensing. Berlin: Springer;2014. p. 163–209.
21. Jansson P, Triest J, Grilli R, Ferrés B, Silyakova A, Mienert J, et al. High-resolution under-water laser spectrometer sensing provides new insights to methane distribution at an Arctic seepage site. Ocean Sci. 2019; 15:1055–1069.
22. Weiss RF. Carbon dioxide in water and seawater: the solubility of a non-ideal gas. Marine Chemistry. 1974; 2:203–215.
23. Ziabakhsh-Ganji Z, Kooi H. An Equation of State for thermodynamic equilibrium of gas mixtures and brines to allow simulation of the effects of impurities in subsurface CO$_2$ storage. International Journal of Greenhouse Gas Control. 2012; 11S:S21–S34.
24. Wiesenburg DA, Guinasso NL. Equilibrium solubilities of methane, carbon monoxide and hydrogen in water and seawater. Journal of Chemical and Engineering Data. 1979; 24(4): 357–360.
25. Weiss RF. The solubility of nitrogen, oxygen and argon in water and seawater. Deep Sea Research and Oceanographic Abstracts. 1970; 17(4): 721–735.
26. Moore JC, Battino R, Rettich TR, Handa YP, Wilhelm E. Partial molar volumes of gases at infinite dilution in water at 298.15 K. Journal of Chemical and Engineering Data. 1982; 27(1):22–24.
27. Rettich TR, Handa P, Battino R, Wilhelm E. Solubility of gases in liquids. 13. High-precision determination of Henry’s constants for methane and ethane in liquid water at 275 to 328 K. Journal of Physical Chemistry. 1981; 85(22):3230–3237.
28. Wüest A, Piepke G, Halfman JD. Combined effects of dissolved solids and temperature on the density stratification of Lake Malawi. In: Johnson TC, Oda da EO, editors. The Limnology, Climatology and Paleoclimatology of the East African Lakes. New York: gordon and Breach;1996. p. 183–202.
29. Grilli R, Darchambeau F, Chappellaz J, Mugisha A, Triest J, Umutooni A. Continuous in situ measurement of dissolved methane in Lake Kivu using a membrane inlet laser spectrometer. Geoscientific Instrumentation, Methods and Data Systems. 2020; 9(1),141–151.
30. Schmid M, Wüst A. Stratification, mixing and transport processes in Lake Kivu. In: Descy J, Darchambeau F, Schmid M, editors. Lake Kivu: Limnology and biogeochemistry of a unique tropical lake. Dordrecht: Springer, Aquatic Ecology Series. 2012. P. 13–29.
31. Bärenbold F, Schmid M, Brennwald MS, Kipfer R. Missing atmospheric noble gases in a large tropical lake: The case of Lake Kivu, East-Africa. Chemical Geology 532 (2020): 119374.
32. Ross KA, Smets B, De Batist M, Hilbe M, Schmid M, Anselmetti FS. Lake-level rise in the late Pleistocene and active subaqua tic volcanism since the Holocene in Lake Kivu, East African Rift. Geomorphology. 2014; 221: 274–285.
33. Lahmeyer O. Bathymetric survey of Lake Kivu. Final report. Kigali: Republic of Rwanda, Ministry of Public Work, Directory of Energy and Hydrocarbons; 1998.
34. Wüst A, Jarc L, Bürgmann H, Pasche N, Schmid M. Methane formation and future extraction in Lake Kivu. In: Descy J, Darchambeau F, Schmid M, editors. Lake Kivu: Limnology and biogeochemistry of a unique tropical lake. Dordrecht: Springer, Aquatic Ecology Series. 2012. P. 13–29.
35. Morana C, Borges AV, Roland FAE, Darchambeau F, Descy JP, Bouillon S. Methanotrophy within the water column of a large meromictic tropical lake (Lake Kivu, East Africa). Biogeosciences. 2015; 12(7): 2077–2088.
36. Sánchez-España J, Yusta I, Boehrer B. Degassing Pit Lakes: Technical Issues and Lessons Learnt from the HERCO 2 Project in the Guadiana Open Pit (Herrerías Mine, SW Spain). Mine Water and the Environment. 2020. https://doi.org/10.1007/s10230-020-00654-1 | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2758,
1
],
[
2758,
8741,
2
],
[
8741,
12546,
3
],
[
12546,
16406,
4
],
[
16406,
19942,
5
],
[
19942,
21321,
6
],
[
21321,
22220,
7
],
[
22220,
22921,
8
],
[
22921,
25266,
9
],
[
25266,
29224,
10
],
[
29224,
33071,
11
],
[
33071,
36043,
12
],
[
36043,
40888,
13
],
[
40888,
42477,
14
]
]
} |
5323673f038892c6fd5f389cf6a4eb8a4719bc9b | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 10,
"total-fallback-pages": 0,
"total-input-tokens": 41311,
"total-output-tokens": 12359,
"fieldofstudy": [
"Materials Science"
]
} | 1,3-PROPANEDIOL: STATISTICAL OPTIMIZATION OF MEDIUM TO IMPROVE PRODUCTION BY Clostridium beijerinckii DSM 791
Daiana Wischral¹, Carolina Araújo Barcelos¹, Nei Pereira Jr¹, Fernando L. Pellegrini Pessoa²
¹Laboratories of Bioprocess Development, School of Chemistry, Federal University of Rio de Janeiro, 21949-900
²Chemical Engineering Department, School of Chemistry, Federal University of Rio de Janeiro, 21949-900
ABSTRACT
Biodiesel is a promising alternative biofuel, and as its production increases, so does production of the principal co-product, glycerol. The present study is aimed at maximizing the glycerol consumption and 1,3-propanediol (1,3-PDO) production by C. beijerinckii DSM 791, using sequential experimental design methodology for optimization of nutrients concentration for the fermentation medium. Three sequential experimental designs were performed: Plackett-Burman, fractional factorial and central composite rotational. The following nutrients of the medium were evaluated: urea, yeast extract, C₂H₃NaO₂, KH₂PO₄, K₂HPO₄, MgSO₄.7H₂O, MnSO₄.H₂O, FeSO₄.7H₂O, glycerol, butyric acid, glucose and CaCO₃. Urea, KH₂PO₄, MgSO₄.7H₂O, MnSO₄.H₂O, FeSO₄.7H₂O, butyric acid, glucose and CaCO₃ were removed from the medium optimization since they had an insignificant statistical effect on 1,3-PDO production. The optimal concentrations of yeast extract, C₂H₃NaO₂, KH₂PO₄ and glycerol predicted by the optimization were as follows (g/l): 0.5, 0.005, 5 and 8, respectively, which were validated experimentally. Desirability function allows the maximization of the 1,3-PDO production and percentage glycerol consumption simultaneously, which resulted in yield of 0.58 mol/mol of 1,3-PDO and 100 % of glycerol consumption. The results showed that the use of sequential experimental design and the use of the desirability function led to the optimization of the 1,3-PDO fermentation by C. beijerinckii DSM 791.
Indexing terms/Keywords
1,3-propanediol; Glycerol; Clostridium beijerinckii DSM 791; Fractional factorial; Central composite rotatable design.
Academic Discipline And Sub-Disciplines
Biotechnology, Bioprocess Engineering, Design of experiments.
SUBJECT CLASSIFICATION
Biotechnology, Biochemistry, Bioprocess Engineering, Design of experiments, Statistical analysis.
TYPE (METHOD/APPROACH)
Biotechnology, Biochemistry, Bioprocess Engineering, Anaerobic fermentation, Design of experiments, Statistical analysis.
INTRODUCTION
In recent years, concern over the biological production of commercially important metabolites has been growing. This is mainly attributed to escalating global energy and environmental problems, which have led researchers worldwide to devise methods for producing almost everything in a “green” way [1,2]. Of these, the production of biofuels has attracted a great deal of attention [1]. Biodiesel is an alternative fuel that reduces net greenhouse effects and its use has become mandatory in many countries. The main by-product of the biodiesel plant is glycerol, a chemical compound of importance as an end product and as a starting material for other useful products [3].
The production of waste glycerol follows increasing biodiesel production; since the stoichiometry of the reaction dictates that for each 100 kg of biodiesel approximately 10 kg of glycerol is produced [4, 5]. The surplus of glycerol coming from biodiesel fabrication has increased enormously, representing about 65 % of the world’s glycerol production [3]. Compared to direct application and chemical transformation, microbial conversion of waste glycerol is a viable alternative without certain disadvantages, such as low product specificity, high energy input (pressure/temperature), intensive pretreatment requirements and/or release of environmentally toxic intermediate compounds [2, 4, 5]. In addition, compared to conventional biorefinery substrates, such as glucose and sucrose, waste glycerol presents a class of substrates that are inexpensive, sustainable, and not considered a suitable human food source [4, 5]. Moreover, bioprocess should be periodically reviewed and incorporate to technological innovations in order to increase their performance and lucrativeness [6].
Glycerol is the natural substrate for microbial production of 1,3-propanediol (1,3-PDO). First the glycerol is dehydrated to 3-hydroxypropionaldehyde (3-HPA) by glycerol dehydratase. Then the product of dehydration reaction, 3-HPA, is reduced to 1,3-PDO by an NAD-dependent oxidoreductase [7, 8]. Since 50% of the entire cost of microbial production of 1,3-PDO is due to the price of raw materials, waste glycerol from biodiesel production processes may be an interesting renewable carbon source for microorganisms that produce 1,3-PDO [9]. 1,3-PDO can be successfully produced fermentatively from glycerol by bacteria of Enterobacteriaceae and Clostridiaceae families, mainly Klebsiella spp., Clostridium spp. [5, 7, 10] and E. coli modified [11].
1,3-PDO is a bifunctional organic compound that could potentially be used for many synthesis reactions, in particular as a monomer for polycondensations to produce polyesters, polyethers and polyurethanes [7]. Products obtained by 1,3-PDO polymerization are characterized by good biodegradability, better specificity and higher industrial safety, in addition to being cheaper than those based on 1,2-propanediol, ethylene glycol or butanediol [4]. It has a number of other interesting applications in addition to that of polymer constituent, for example, synthesis of polytrimethylene terephthalate (PTT) that in turn can be used to make carpets (Corterra W, Shell), special textile fibers (Sorona W, DuPont), monofilaments, films, and nonwoven fabrics [12]. Furthermore, 1,3-PDO can improve the properties of solvents, adhesives, laminates, resins, detergents and cosmetics [13]. Therefore, this work is aimed at optimizing the fermentation medium to improve glycerol consumption and 1,3-PDO production by C. beijerinckii DSM 791 using sequential experimental design methodology.
MATERIALS AND METHODS
Microorganism and culture medium
C. beijerinckii DSM 791 was obtained from the Leibniz-Institut DSMZ – German Collection of Microorganisms and Cell Cultures and was maintained in microtubes at +80 °C on Reinforced Clostridial Medium (RCM) [14] with 30 % (v/v) of glycerol. Cellular growth was performed in two steps. First, the cells were cultured anaerobically in a rotary shaker (New Brunswick Scientific – Edison N. J., USA) at 37°C, initial pH 6.5 and 80 rpm. The medium used for pre-culture and inoculum contained (g/l): peptone, 10; beef extract, 10; yeast extract, 3; NaCl, 3; cysteine, 0.5; C₂H₄NaO₂, 3; agar, 0.5; and glycerol, 5. The medium was reduced with nitrogen gas and sterilized before inoculation. Then the pre-culture, inoculum and fermentation medium were grown anaerobically in 100 mL flasks, containing 70 mL of medium, at 37 °C and 80 rpm. The pre-culture and inoculum times were respectively 16 h and 9 h, determine in previous experiments. The inoculum culture was initialized with 10 % (v/v) from pre-culture medium and the fermentation medium with 20 % (v/v) from inoculum culture.
Optimization of the fermentation medium for 1,3-PDO production
A strategy of three sequential experimental designs was adopted to optimize the fermentation medium to improve 1,3-PDO production and glycerol consumption. The design experiment and the statistical treatment of the results were performed by the software package STATISTICA, version 6.0 (StatSoft, Inc.) including ANOVA, to obtain the impact and significance of each term and the interactions between the process variables and response. The fit quality of the polynomial model was expressed via the determination coefficient $R^2$, and its statistical significance was verified with the F-test using the same software program, considering a confidence level of 95 % or p-value less than 0.05 [15, 16].
In the first experimental design, a Plackett-Burman design was chosen to screen and identify variables that had significant influence. This design is based on the first order polynomial model [17]:
$$Y = \beta_0 + \sum \beta_i X_i$$
where $Y$ is the response, $\beta_0$ is the model intercept and $\beta_i$ is the coefficient of linear equation, and $X_i$ is the level of independent variable.
Twelve components were selected for analysis: urea, yeast extract, \( \text{C}_2\text{H}_5\text{NaO}_2 \), \( \text{KH}_2\text{PO}_4 \), \( \text{K}_2\text{HPO}_4 \), \( \text{MgSO}_4\cdot\text{7H}_2\text{O} \), \( \text{MnSO}_4\cdot\text{H}_2\text{O} \), \( \text{FeSO}_4\cdot\text{7H}_2\text{O} \), glycerol, butyric acid, glucose and \( \text{CaCO}_3 \); in two levels, minimum and maximum, coded as “−1” and “+1”, respectively, and three replicates of the center point, coded as “0” (Table 1) resulting in 19 runs performed at 37 °C and 80 rpm.
Table 1. Factors and levels used in the Plackett-Burman design for optimization of fermentation medium
| Factor (g/l) | Code | Minimum | Center Point | Maximum |
|-------------|------|---------|--------------|---------|
| Urea | x1 | 0.50 | 1.50 | 3.00 |
| Yeast extract | x2 | 0.50 | 1.50 | 3.00 |
| \( \text{C}_2\text{H}_5\text{NaO}_2 \) | x3 | 0.005 | 0.01 | 0.05 |
| \( \text{KH}_2\text{PO}_4 \) | x4 | 0.10 | 0.50 | 1.50 |
| \( \text{K}_2\text{HPO}_4 \) | x5 | 0.20 | 2.00 | 4.00 |
| \( \text{MgSO}_4\cdot\text{7H}_2\text{O} \) | x6 | 0.10 | 0.20 | 0.40 |
| \( \text{MnSO}_4\cdot\text{H}_2\text{O} \) | x7 | 0.005 | 0.01 | 0.02 |
| \( \text{FeSO}_4\cdot\text{7H}_2\text{O} \) | x8 | 0.01 | 0.30 | 0.50 |
| Glycerol | x9 | 5.00 | 35.00 | 70.00 |
| Butyric acid | x10 | 0.20 | 0.50 | 2.00 |
| Glucose | x11 | 0.00 | 6.00 | 12.00 |
| \( \text{CaCO}_3 \) | x12 | 0.30 | 1.00 | 2.00 |
After the first screening experiments using Plackett-Burman design, a fractional factorial design was used to find significant factors affecting 1,3-PDO production. This design was used because it is accurate for estimating the main effects and interaction, with a reduced number of experiments if compared to a complete factorial design [18]. Yeast extract, \( \text{C}_2\text{H}_5\text{NaO}_2 \), \( \text{K}_2\text{HPO}_4 \), urea, \( \text{MgSO}_4\cdot\text{7H}_2\text{O} \) and glycerol were used as independent variables and the 1,3-PDO production was used as a dependent variable. Therefore, to analyze these six components the second experimental fractional factorial design \((2^6-2)\) was performed with two levels, minimum and maximum, coded as “−1” and “+1”, respectively, and three replicates of the center point (Table 2) resulting in 19 runs performed at 37 °C and 80 rpm.
Table 2. Factors and levels used in the fractional factorial \((2^6-2)\) design for optimization of fermentation medium
| Factor (g/l) | Code | Minimum | Center Point | Maximum |
|-------------|------|---------|--------------|---------|
| Urea | x1 | 0.00 | 0.50 | 1.00 |
| Yeast extract | x2 | 0.50 | 1.50 | 3.00 |
| \( \text{C}_2\text{H}_5\text{NaO}_2 \) | x3 | 0.005 | 0.01 | 0.05 |
| \( \text{K}_2\text{HPO}_4 \) | x4 | 0.20 | 2.00 | 4.00 |
| \( \text{MgSO}_4\cdot\text{7H}_2\text{O} \) | x5 | 0.00 | 0.10 | 0.20 |
| Glycerol | x6 | 5.00 | 10.00 | 20.00 |
Finally, a central composite rotatable design (CCRD) was carried out to optimize 1,3-PDO production using the response surface methodology. The concentrations of \( \text{K}_2\text{HPO}_4 \) and glycerol, ranging from 0.5 to 5.0 g/L and 2.0 to 12.0 g/L, respectively, were chosen as independent variables (Table 3). The experiment comprised 12 runs (3 replicate runs at the center point and 2 axial levels). The concentration of yeast extract and \( \text{C}_2\text{H}_5\text{NaO}_2 \) were maintained at 0.5 and 0.005 g/L, respectively, in accordance with the results obtained from the fractional factorial design performed at 37 °C and 80 rpm. 1,3-PDO production and percentage reduction of substrate (PRS) were used as response variables. The fundamental of this method, for quantitative variables, involves fitting first-order (linear) or second-order (quadratic) functions of the predictors to one or more response variables, and then examining the characteristics of the fitted surface to decide the appropriated action [18]. A quadratic polynomial equation was proposed to describe the mathematical relationship between the independent variables and response variables under the current conditions [15]:
\[
Y = \beta_0 + \beta_1 x_1 + \beta_2 x_2^2 + \beta_3 x_1 x_2 + \beta_4 x_2 + \beta_5 x_1 x_2
\]
where Y is the response, βᵣ are the coefficients and xᵣ the independent variables.
### Table 3. Factors and levels used in the experimental CCRD for optimization of fermentation medium
| Factor (g/l) | Code | Axial | Minimum | Center Point | Maximum | Axial |
|-------------|------|-------|---------|--------------|---------|-------|
| K₂HPO₄ | x₁ | -1.41 | 0.50 | 0.90 | 2.60 | 4.30 | 5.00 |
| Glycerol | x₂ | 2.00 | 3.50 | 7.00 | 10.50 | 12.00 |
The global desirability function (D) was performed to maximize the 1,3-PDO production and percentage reduction of substrate (PRS) simultaneously. This function consists in converting each response into a single desirability function (di) that ranges from 0 to 1 ≤ d ≤ 1. For a function with two independent variables, a plot of composite desirability can be constructed as a function of filler alone, using the equation [19]:
\[
D = (d_1d_2)^{1/2}
\]
The predicted values were validated experimentally in batch fermentation (three replicates), carried out in predicted conditions during 24 h.
### Analytical methods
The concentrations of 1,3-PDO, glycerol and organic acids were analyzed by high-performance liquid chromatography (HPLC) equipped with a refractive index detector and UV detector. Samples were first centrifuged at 10.000 rpm for 10 minutes at 4 °C (Sigma Laborzentrifugen 2K15). The supernatants were filtered (Millex-HV, PVDF membrane, 0.2 μm pore size, 13 mm diameter - Millipore) for measurements of 1,3-PDO, glycerol and acids (lactic, acetic and butyric). The column used for separation was Hi Plex H, 300 x 7.7 mm (Agilent Technologies) at 45 °C. Analyses performed at a flow rate of 0.6 mL/min at a constant temperature of 45 °C, H₂SO₄ (0.5 mM) were the mobile phase and a wavelength of 210 nm. External standards were applied for identification and quantification of peak areas. The cell concentration (g/l) was determined using a linear equation derived from the relationship of cell dry weight (90 °C until constant weight) and the optical density (OD) at 600 nm.
### RESULTS AND DISCUSSION
The Plackett-Burman design was the first step in the sequential strategy to select the factors to optimize the medium. The 1,3-PDO concentrations ranged from 0.34 g/l to 7.23 g/l (Table 4). In accordance with ANOVA (analysis of variance), using 95 % of confidence level (p ≤ 0.05) and standard error, the r squared was shown to be a good fit to the experimental results of 0.98 and the r squared adjusted 0.93. These results allowed made it possible to plot the Pareto chart (Fig 1), which provided important data about the statistical relevance of the factors. From this chart, one can see that urea, yeast extract, C₂H₂NaO₄, K₂HPO₄, MgSO₄·7H₂O, glycerol and glucose had significant relevance for 1,3-PDO production. On the contrary, KH₂PO₄, MnSO₄·H₂O, FeSO₄·7H₂O, butyric acid and CaCO₃ presented insignificant effect for 1,3-PDO production and then were removed from medium composition.
The glucose was eliminated from the medium composition since it presented the significant but negative effect for 1,3-PDO production and the interest of this work is to use glycerol. Even though the glycerol had a significant positive effect, the analysis range was decreased based on the residual glycerol results shown in Table 4, since the amount for residual glycerol reached was 68.18 g/l. The range for urea and MgSO₄·7H₂O were reduced, since these nutrients presented a significant but negative effect. The other significant factors that presented a positive effect had the analysis range maintained and the significant factors with a negative effect had the analysis range decreased.
Nonetheless, the 1,3-PDO production can still be enhanced, as the center points are still lower than the other points, showing an increasing tendency in the direction of the maximum point. According to the results of the second experimental design, the fractional factorial (2⁴−2), which are shown in Table 5, the 1,3-PDO production ranged from 2.32 g/l to 4.55 g/l. The ANOVA, using 95 % of confidence interval (p ≤ 0.05), of fractional factorial design results with replicated center points, the curvature in the tested region was significant with a p-value of less than 0.05, which suggests that a response surface study with a quadratic model is required to optimize 1,3-PDO production with the selected significant variables in this design. These results (Table 5) allowed made it possible the plotting of the Pareto chart (Fig 2), which provided important data about the statistical relevance of the factors, as well as of their interactions. From this chart, one can see K₂HPO₄ and glycerol had significant relevance for 1,3-PDO production. The K₂HPO₄ that presented a positive effect had the analysis range increased.
Even though the glycerol had a significant positive effect, the analysis range was decreased based on residual glycerol results shown in Table 5, since the amount for residual glycerol reached was 16.09 g/l. The urea and MgSO₄·7H₂O was eliminated from the medium, since the minimum concentration analyzed was zero and these factors had a negative effect on 1,3-PDO production. The C₂H₂NaO₄ and yeast extract factors presented an insignificant effect on 1,3-PDO production, so these concentrations were kept to a minimum. The analysis of the Pareto chart also depicted significant curvature denoting that there exists a point of maximum 1,3-PDO productivity. Moreover, the values of the center points are close to the maximum obtained in the studied range, indicating the need to add axial points to the following experiment to optimize the 1,3-PDO production.
### Table 4. Matrix of experimental Plackett-Burman design and their corresponding results of 1,3-PDO production and residual glycerol (Res. gly.)
| Run | x1 | x2 | x3 | x4 | x5 | x6 | x7 | x8 | x9 | x10 | x11 | x12 | 1,3-PDO (g/l) | Res. gly. (g/l) |
|-----|----|----|----|----|----|----|----|----|----|-----|-----|-----|-------------|--------------|
| 1 | +1 | -1 | -1 | -1 | +1 | -1 | -1 | +1 | -1 | +1 | -1 | 0.51 | 67.72 |
| 2 | +1 | +1 | -1 | -1 | +1 | -1 | -1 | +1 | -1 | +1 | -1 | 2.59 | 64.82 |
| 3 | +1 | +1 | +1 | -1 | +1 | -1 | -1 | +1 | -1 | +1 | -1 | 0.34 | 5.45 |
| 4 | +1 | +1 | +1 | +1 | +1 | -1 | -1 | +1 | -1 | +1 | -1 | 0.42 | 4.93 |
| 5 | +1 | +1 | +1 | +1 | -1 | -1 | -1 | +1 | -1 | +1 | -1 | 7.23 | 55.43 |
| 6 | +1 | -1 | +1 | +1 | +1 | -1 | -1 | +1 | -1 | +1 | -1 | 3.05 | 0.81 |
| 7 | +1 | -1 | +1 | +1 | +1 | +1 | -1 | +1 | -1 | -1 | -1 | 0.37 | 4.87 |
| 8 | +1 | -1 | -1 | +1 | +1 | +1 | +1 | +1 | -1 | -1 | -1 | 2.87 | 0.00 |
| 9 | +1 | -1 | +1 | -1 | +1 | +1 | +1 | +1 | -1 | -1 | -1 | 3.57 | 60.94 |
| 10 | +1 | +1 | -1 | -1 | +1 | +1 | +1 | +1 | -1 | -1 | -1 | 5.68 | 59.00 |
| 11 | +1 | +1 | +1 | -1 | +1 | +1 | +1 | +1 | -1 | -1 | -1 | 0.41 | 68.04 |
| 12 | +1 | +1 | +1 | -1 | +1 | +1 | +1 | +1 | -1 | -1 | -1 | 0.69 | 67.47 |
| 13 | +1 | -1 | +1 | +1 | +1 | +1 | +1 | +1 | -1 | -1 | -1 | 0.46 | 5.33 |
| 14 | +1 | -1 | -1 | +1 | +1 | +1 | -1 | -1 | +1 | +1 | +1 | 0.77 | 68.18 |
| 15 | +1 | -1 | +1 | -1 | +1 | +1 | +1 | +1 | -1 | -1 | -1 | 2.80 | 0.34 |
| 16 | +1 | -1 | +1 | +1 | -1 | -1 | -1 | -1 | -1 | -1 | -1 | 2.96 | 0.15 |
| 17 (PC) | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0.62 | 34.69 |
| 18 (PC) | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0.64 | 36.90 |
| 19 (PC) | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0.64 | 36.42 |
*Fig 1: Pareto chart for 1,3-PDO production in the Plackett-Burman design*
Table 5. Matrix of experimental fractional factorial ($2^{6-2}$) design and their corresponding results of 1,3-PDO production and residual glycerol (Res. gly)
| Run | x1 | x2 | x3 | x4 | x5 | x6 | 1,3-PDO (g/l) | Res. gly. (g/l) |
|-----|----|----|----|----|----|----|---------------|----------------|
| 1 | -1 | -1 | -1 | -1 | -1 | -1 | 2.46 | 0.44 |
| 2 | 1 | -1 | -1 | 1 | -1 | -1 | 2.57 | 0.27 |
| 3 | -1 | 1 | -1 | 1 | 1 | 1 | 3.35 | 13.37 |
| 4 | 1 | 1 | -1 | -1 | -1 | 1 | 2.98 | 12.78 |
| 5 | -1 | -1 | 1 | -1 | 1 | 1 | 2.64 | 14.24 |
| 6 | 1 | -1 | 1 | -1 | -1 | 1 | 2.32 | 16.09 |
| 7 | -1 | 1 | 1 | -1 | -1 | 1 | 2.62 | 0.28 |
| 8 | 1 | 1 | 1 | -1 | -1 | 1 | 2.76 | 0.00 |
| 9 | -1 | -1 | 1 | 1 | -1 | 1 | 4.27 | 13.21 |
| 10 | 1 | -1 | -1 | 1 | 1 | 1 | 4.14 | 12.38 |
| 11 | -1 | 1 | -1 | 1 | 1 | -1 | 2.89 | 0.10 |
| 12 | 1 | 1 | -1 | 1 | -1 | -1 | 2.86 | 0.00 |
| 13 | -1 | -1 | 1 | 1 | 1 | -1 | 2.86 | 0.00 |
| 14 | 1 | -1 | 1 | 1 | -1 | -1 | 2.65 | 0.00 |
| 15 | -1 | 1 | 1 | 1 | -1 | 1 | 5.23 | 10.71 |
| 16 | 1 | 1 | 1 | 1 | 1 | 1 | 5.31 | 9.97 |
| 17 | PC | 0 | 0 | 0 | 0 | 0 | 4.41 | 6.11 |
| 18 | PC | 0 | 0 | 0 | 0 | 0 | 4.51 | 5.70 |
| 19 | PC | 0 | 0 | 0 | 0 | 0 | 4.55 | 5.57 |
Fig 2: Pareto chart for 1,3-PDO production in the fractional factorial ($2^{6-2}$) design
Thus, the next step was to optimize the $K_2HPO_4$ and glycerol concentrations, evaluating the 1,3-PDO production and percentage reduction of substrate using the Central Composite Rotational Design (DCCR), since it allows the predicting of...
the response surface with curvature plots and visualize of the maximum values of the dependent variables. The ANOVA of CCRD results were presented in Table 6 and 7, using 95 % of confidence interval (p<0.05) and pure error, shows in both K$_2$HPO$_4$ and glycerol concentrations significant linear effects; this means the factors presented mutual interference in 1,3-PDO production and PRS (percentage reduction of substrate). The CCRD showed quadratic effect for glycerol (p<0.05) on 1,3-PDO production and also for PRS, thus denoting a curvilinear plan, as can be observed in ANOVA.
Table 6. Analysis of variance (ANOVA) of factors for 1,3-PDO production
| Factor | Sum of squares | Degrees of freedom | Mean Square | F | p-value |
|--------------|----------------|--------------------|-------------|---------|---------|
| K$_2$HPO$_4$ (L) | 0.407 | 1 | 0.407 | 85.906 | 0.003 |
| Glycerol (L) | 4.907 | 1 | 4.907 | 1035.532| 0.001 |
| Glycerol (Q) | 1.358 | 1 | 1.358 | 286.589 | 0.001 |
| 1L by 2L | 0.125 | 1 | 0.125 | 26.312 | 0.014 |
| Lack of fit | 0.020 | 4 | 0.005 | 1.080 | 0.495 |
| Pure error | 0.014 | 3 | 0.005 | | |
| Total SS | 6.831 | 11 | | | |
Table 7. Analysis of variance (ANOVA) of factors for PRS
| Factor | Sum of squares | Degrees of freedom | Mean Square | F | p-value |
|--------------|----------------|--------------------|-------------|---------|---------|
| K$_2$HPO$_4$ (L) | 276.294 | 1 | 276.294 | 60.345 | 0.004 |
| Glycerol (L) | 2059.883 | 1 | 2059.883 | 449.901 | 0.001 |
| Glycerol (Q) | 310.569 | 1 | 310.569 | 67.832 | 0.004 |
| 1L by 2L | 92.719 | 1 | 92.719 | 20.251 | 0.020 |
| Lack of fit | 10.998 | 4 | 2.749 | 0.600 | 0.690 |
| Pure error | 13.736 | 3 | 4.579 | | |
| Total SS | 2764.199 | 11 | | | |
Analyzing the data from Tables 6 and 7, the model adequately adjusts to the experimental points, representing the confidence of the results for both dependent variables analyzed. Considering the confidence level of 95 %, the models obtained were significant and were able to explain 99 % of variance (R$^2$), since the determination coefficient R$^2$ values (99 % to 1,3-PDO product and PRS), R adjust values (99 % to 1,3-PDO product and PRS), the lack of fit was insignificant (above 0.05) combined with the significant F and p-values. By means of ANOVA, for both response factors, significant effects quadratic and linear were found for the glycerol, while only a linear effect was found for K$_2$HPO$_4$, the quadratic term of K$_2$HPO$_4$ was removed from models because it presented insignificant effect. This means the glycerol concentration has more influence than K$_2$HPO$_4$ for both response factors analyzed (1,3-PDO production and PRS).
The response surface of the third experimental design is presented in Fig 3. The surfaces present the relation between dependent factors (Fig 3a: 1,3-PDO production and Fig 3b: PRS) and independent factors (glycerol and K$_2$HPO$_4$ concentrations). At the range studied, the higher production of 1,3-PDO was obtained at the higher level of glycerol and K$_2$HPO$_4$, presenting the maximum results of around 4 g/l to 1,3-PDO. The analysis of the PRS surface made it possible to observe the higher percentage of PRS was obtained at lower level of glycerol, while in the lower level of glycerol the influence of K$_2$HPO$_4$ level was the same.
The equations representing 1,3-PDO production and PRS, where $x_1$ is the K$_2$HPO$_4$ concentration (g/l) and $x_2$ is the glycerol concentration (g/l), follow:
$$1,3\text{-PDO (g/l)} = 3.43 - 0.45x_1^2 + 0.22x_1 + 0.78x_2 + 0.18x_1x_2$$
$$\text{PRS (±)} = 90.88 - 6.82x_1^2 + 5.88x_1 - 16.05x_2 + 4.81x_1x_2$$
The global desirability value to reach the optimum for the two factors simultaneously was 0.93, meaning that the optimization by this function fulfills 93 % of the maximum obtainable for each dependent variable (response variable). The optimum concentration for K$_2$HPO$_4$ was 5 g/l and for glycerol was 8 g/l. The predicted values using the global desirability function and the experimental validation of the results in optimized conditions for 1,3-PDO production and PRS are presented in Table 8. It is observed that the experimental values are within the confidence limits ±95 % and +95 %, showing that these experimental findings were in close agreement with the model prediction.
According to the results of Table 8 and Fig 4, the yield obtained was 0.58 mol/mol of 1,3-PDO per consumed glycerol in 12 h of fermentation without pH control. The efficiency of the glycerol fermentation by *C. beijerinckii* DSM 791 with the medium optimized in this study was 81 %, since Biebl et al. (1999) conclude the maximum yield for 1,3-PDO from glycerol by fermentation is 0.72 mol/mol of 1,3-PDO per consumed glycerol. It was demonstrated that *C. beijerinckii* DSM 791 was perfectly able to grow and produce 1,3-PDO on glycerol as the sole source of carbon and energy (Fig 4).
### Table 8. Validation of the fermentation medium for improve glycerol consumption and 1,3-PDO production
| Factor | Limits | Predict value | Experimental result |
|--------|----------|---------------|---------------------|
| | ±95 % | ±95 % | |
| 1,3-PDO (g/l) | 3.85 | 4.14 | 3.99 | 3.85 ± 0.06 |
| PRS (%) | 91.70 | 100 | 96.18 | 100 ± 0.00 |
The results of this study are comparable with those obtained by Gungormusler et al. (2010a), who obtained the yield of 0.58 mol/mol of 1,3-PDO per consumed glycerol in 24 h by \textit{C. beijerinckii}. In another study, Gungormusler et al. (2010b) compared the 1,3-PDO production potential, without pH control, by different \textit{Clostridium} spp. (\textit{C. saccharobutylicum}, \textit{C. acetobutylicum}, \textit{C. pasteurianum}, \textit{C. beijerinckii} and \textit{C. butyricum}). Yields varied between 0.37 mol/mol to 0.54 mol/mol of 1,3-PDO per consumed glycerol while the PRS varied between 38 % and 93 % in 24 h of fermentation. Moon et al. (2011) optimized the medium for 1,3-PDO production from glycerol without pH control, by \textit{C. pasteurianum} and obtained yield of 0.36 mol/mol of 1,3-PDO per consumed glycerol in 24 h.
The pH certainly influenced the metabolism of \textit{Clostridium} spp. and some authors obtained higher yields of 1,3-PDO by controlling the pH during the glycerol fermentation. For example, Biebl et al. (1992) obtained 0.62 mol/mol of 1,3-PDO per consumed glycerol by \textit{C. butyricum} and Otte et al. (2009) obtained 0.64 mol/mol of 1,3-PDO per consumed glycerol by \textit{C. diolis}, controlling this important process variable.
**CONCLUSIONS**
The sequential experimental design strategy combined with the use of the desirability function for the optimization of the composition medium was shown to be an important tool for maximizing 1,3-PDO production and glycerol consumption by \textit{C. beijerinckii} DSM 791, resulting in the yield of 0.58 mol/mol of 1,3-PDO per consumed glycerol. Among the evaluated factors, glycerol and K$_2$HPO$_4$ are of greater importance to 1,3-PDO production and percentage reduction of substrate by \textit{C. beijerinckii} DSM 791. An optimum condition was successfully found and validated in this interval of study. The optimal concentrations for the medium compounds are 0.5 g/l of yeast extract, 0.005 g/l of C$_6$H$_{12}$NaO$_4$, 5 g/l of K$_2$HPO$_4$ and 8 g/l of glycerol.
**Acknowledgements**
The authors acknowledge the Human Resources Program 13 of the National Petroleum Agency (ANP - PRH 13) for scholarship.
**Conflict of Interest**
The authors declare that they have no conflict of interest.
**Ethical statement**
This article does not contain any studies with human participants or animals performed by any of the authors. The authors confirm that principles of ethical and professional conduct have been followed in this research and in the preparation of this article.
**REFERENCES**
[1] Kaur G., Srivastava A. K., Chand S. Advances in biotechnological production of 1,3-propanediol. Biochemical Engineering Journal. 64 (2012), 106–18.
[2] Yang F. X., Hanna M. A., Sun R. C. Value-added uses for crude glycerol — A byproduct of biodiesel production. Biotechnol Biofuels. 5 (2012), 13.
[3] Pinto B. P., Mota C. J. A. (2014), in Developments in glycerol byproduct-based biorefineries. Advances in Biorefineries: Biomass and Waste Supply Chain Exploitation (Waldron K., ed.), Woodhead publishing, Sawston, Cambridge, UK, pp. 364–385.
[4] Silva G. P., Mack M., Contiero J. Glycerol: A promising and abundant carbon source for industrial microbiology. Biotechnology Advances. 27 (2009), 30–39.
[5] Yazdani S. S., Gonzalez R. Anaerobic fermentation of glycerol: a path to economic viability for the biofuels industry. Current Opinion in Biotechnology. 18 (2007), 213–219.
[6] Pereira Jr N., Bon E. P.S., Ferrara M. A. (2008). in Séries em Biotecnologia, vol. 1: Tecnologia de Bioprocessos, Biblioteca Nacional, Rio de Janeiro, RJ, pp. 56.
[7] Biebl H., Menzel K., Zeng A. P., Deckwer W. D. Microbial production of 1,3-propanediol. Appl Microbiol Biotechnol. 52 (1999), 289–297.
[8] Venkataramanan K. P., Boatman J. J., Kurniawan Y., Bothun G. D., Scholz C. Impact of impurities in biodiesel-derived crude glycerol on the fermentation by \textit{Clostridium pasteurianum} ATCC 6013. Appl Microbiol Biotechnol. 93 (2011), 1325–1335.
[9] Gonzalez-Pajuelo M., Andrade J. C., Vasconcelos I. Production of 1,3-propanediol by \textit{Clostridium butyricum} VPI 3266 using a synthetic medium and raw glycerol. Journal of Industrial Microbiology & Biotechnology. 9 (2004), 442–446.
[10] Celinski E. Debottlenecking the 1,3-propanediol pathway by metabolic engineering. Biotechnology Advances. 28 (2010), 519–530.
[11] Hartlep M., Hussman W., Prayitno N., Meynial-Salles I., Zeng A. P. Study of two-stage processes for the microbial production of 1,3-propanediol from glycerol. Appl Microbiol Biotechnol. 60 (2002), 60–66.
[12] Li C., Lesnik K. L., Liu H. Microbial conversion of waste glycerol from biodiesel production into value-added products. Energies. 6 (2013), 4739-4768.
[13] Zeng A. P., Biebl H. (2002). in Bulk-Chemicals from Biotechnology: The Case of 1,3-Propanediol Production and the New Trends, vol. 47: Advances in Biochemical Engineering/Biotechnology (Schepers T., ed.), Springer-Verlag, Berlin, Heidelberg, pp. 239–259.
[14] Powalowska D. S. 1,3-Propanediol production from crude glycerol by Clostridium butyricum DSP1 in repeated batch. Electronic Journal of Biotechnology. 17 (2014), 322–328.
[15] Rodrigues M. I., Iemma A. F. (2009). Planejamento de Experimentos e Otimização de Processos, 2nd ed., Casa do Pão, Campinas, SP.
[16] Maeda R. N., Silva M. M. P., Santa Anna L. M. M., Pereira Jr N. Nitrogen source optimization for cellulase production by Penicillium funiculosum, using a sequential experimental design methodology and the desirability function. Appl Biochem Biotechnol. 161 (2009), 411–422.
[17] Plackett R. L., Burman J. P. The design of optimum multifactorial experiments. Biometrika. 33 (4) (1946), 305–325.
[18] Calado V., Montgomery D. C. (2003). Planejamento de experimentos usando o Statistica, 1nd ed., E-papers, Rio de Janeiro, RJ.
[19] A Balancing Act: Optimizing a Product’s Properties 1994. Available from: www.statease.com/pubs/derringer.pdf. Accessed June 15, 2015.
[20] Gungormusler M., Gonen C., Azbar N. Comparative evaluation of Clostridium beijerinckii (NRRL B-593) and Klebsiella pneumoniae for 1,3-propanediol production. Journal of Biotechnology. 150S (2010a), S210–S211.
[21] Gungormusler M., Gonen C., Ozdemir G., Azbar N. 1,3-propanediol production potential of Clostridium saccharobutylicum NRRL B-643. New Biotechnology. 27 (6) (2010b), 782–788.
[22] Moon C., Lee C. H., Sang B., Um Y. Optimization of medium compositions favoring butanol and 1,3-propanediol production from glycerol by Clostridium pasteurianum. Bioresource Technology. 102 (2011), 10561–10568.
[23] Biebl H., Marten S., Hippe H., Deckwer W. D. Glycerol conversion to 1,3-propanediol by newly isolated clostridia. Appl Microbiol Biotechnol. 36 (1992), 592–597.
[24] Otte B., Grunwaldt E., Mahmoud O., Jennewein S. Genome shuffling in Clostridium diolis DSM 15410 for improved 1,3-propanediol production. Appl Environ Microbiol. 75 (24) (2009), 7610–7616. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2436,
1
],
[
2436,
8338,
2
],
[
8338,
12837,
3
],
[
12837,
18524,
4
],
[
18524,
20554,
5
],
[
20554,
22541,
6
],
[
22541,
27480,
7
],
[
27480,
28718,
8
],
[
28718,
33350,
9
],
[
33350,
35723,
10
]
]
} |
cebd6195cb0585450fde4265ec405613e154e4ad | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 17,
"total-fallback-pages": 0,
"total-input-tokens": 61617,
"total-output-tokens": 20336,
"fieldofstudy": [
"Medicine"
]
} | Systematic review of dietary trans-fat reduction interventions
Lirije Hyseni, Helen Bromley, Chris Kypridemos, Martin O’Flaherty, Ffion Lloyd-Williams, Maria Guzman-Castillo, Jonathan Pearson-Stuttard & Simon Capewell
Objective To systematically review published studies of interventions to reduce people’s intake of dietary trans-fatty acids (TFAs).
Methods We searched online databases (CINAH, the CRD Wider Public Health database, Cochrane Database of Systematic Reviews, Ovid, MEDLINE, Science Citation Index and Scopus) for studies evaluating TFA interventions between 1986 and 2017. Absolute decrease in TFA consumption (g/day) was the main outcome measure. We excluded studies reporting only on the TFA content in food products without a link to intake. We included trials, observational studies, meta-analyses and modelling studies. We conducted a narrative synthesis to interpret the data, grouping studies on a continuum ranging from interventions targeting individuals to population-wide, structural changes.
Results After screening 1084 candidate papers, we included 23 papers: 12 empirical and 11 modelling studies. Multiple interventions in Denmark achieved a reduction in TFA consumption from 4.5 g/day in 1976 to 1.5 g/day in 1995 and then virtual elimination after legislation banning TFAs in manufactured food in 2004. Elsewhere, regulations mandating reformulation of food reduced TFA content by about 2.4 g/day. Worksite interventions achieved reductions averaging 1.2 g/day. Food labelling and individual dietary counselling both showed reductions of around 0.8 g/day.
Conclusion Multicomponent interventions including legislation to eliminate TFAs from food products were the most effective strategy. Reformulation of food products and other multicomponent interventions also achieved useful reductions in TFA intake. By contrast, interventions targeted at individuals consistently achieved smaller reductions. Future prevention strategies should consider this effectiveness hierarchy to achieve the largest reductions in TFA consumption.
Abstracts in العربية, 中文, Français, Русский and Español at the end of each article.
Introduction
Over two-thirds of the global burden of disability and death is attributable to noncommunicable diseases. Diseases such as cardiovascular diseases, common cancers, dementia, diabetes and respiratory disorders kill over 35 million people annually. The World Health Organization (WHO) priority areas for reducing noncommunicable diseases include tobacco, alcohol, physical inactivity and poor diet. Poor diet generates a larger burden of disease than the other three risk factors combined. It accounts for an estimated 11.3 million deaths annually, compared with 2.1 million for low physical activity, 6.1 million for tobacco smoke and 3.1 million for alcohol and drug use. The problem predominantly reflects an unhealthy global food environment dominated by processed foods high in sugar, salt, saturated fats and, crucially, industrial trans-fatty acids (TFAs).
TFAs are found naturally in small amounts in some meat and dairy products produced by bacterial action in the stomach of ruminant animals. However, the majority of TFAs are industrial, being manufactured by partial hydrogenation of edible vegetable oils, such as palm oil, cottonseed oil, soybean oil or canola oil. Industrial TFAs are then added to processed or packaged food, mainly to prolong shelf life and enhance taste and texture at a low cost. Since the 1990s, research evidence has accumulated demonstrating that TFA consumption substantially increases people’s risk of coronary heart disease. It does this mainly by elevating harmful low-density lipoprotein cholesterol levels and decreasing protective high-density lipoprotein cholesterol. TFAs may also increase the risk of Alzheimer’s disease and certain cancers, and worsen insulin sensitivity, thereby increasing the risk of type 2 diabetes.
A reduction in people’s intake of industrial TFA is thus a WHO policy priority. However, TFA intake in most countries still exceeds the WHO target of 2 g/day, mainly reflecting consumption of industrial TFAs in processed food. Furthermore, even as overall TFA consumption falls, intake is likely to remain higher in poorer populations, who are more likely to eat processed food products. Different strategies and policy options, targeting different groups, have been proposed to meet these targets. These can be described as upstream or downstream interventions. Downstream interventions generally target individuals and involve behavioural approaches. Intermediate interventions target subgroups in worksites, schools or communities. Both downstream and intermediate interventions are dependent on the individual to respond. By contrast, upstream interventions take place at the population level and typically involve regulatory approaches, taxes or subsidies. By creating a healthier environment, they avoid any dependence on an individual response. In alcohol and tobacco control policies, for instance, an effectiveness hierarchy of preventive interventions has been observed, whereby population-wide policy interventions appear to be more powerful than interventions targeting individuals.
Policy interventions to remove industrial TFAs from foods have therefore been suggested as the most effective public health approach for reducing TFA intake and decreasing the burden of noncommunicable diseases. Some countries have demonstrated that this is feasible. In Denmark, for example, multicomponent interventions progressively reduced the population’s TFA intake and a subsequent legislative ban virtually eliminated TFAs in margarines and vegetable shortenings. However, that success required substantial political will sustained over a decade. Most other countries only have
---
* Department of Public Health and Policy, Whelan Building, University of Liverpool, Liverpool L69 3GB, England.
* School of Public Health, Imperial College London, London, England.
Correspondence to Lirije Hyseni (email: [email protected]).
Submitted: 14 December 2016 – Revised version received: 11 September 2017 – Accepted: 12 September 2017 – Published online: 19 October 2017
doi: http://dx.doi.org/10.2471/BLT.16.189795
Bull World Health Organ 2017;95:821–830G
achieved voluntary TFAs limits, reflecting concerns about political feasibility and generally lower levels of public pressure for change.11
The evidence supporting the most effective policies for reducing TFA intake remains unclear. To inform future prevention strategies we conducted a systematic review of the evidence on the effectiveness of interventions to reduce people’s TFA intake. We also hypothesized that an effectiveness hierarchy might exist.
Methods
Data sources and searches
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines.14 We carried out a systematic search of online databases (CINAHL, the Centre for Reviews and Dissemination Wider Public Health database, Cochrane Database of Systematic Reviews, Ovid®, MEDLINE®, Science Citation Index and Scopus) for studies of interventions to reduce people’s TFA intake, published between 1985 and 24 August 2017. A combination of relevant keywords, identified from exemplar papers (including a systematic review of policies for reducing dietary TFA),15 was used to construct the search strategy (Box 1). All identified papers were imported into a web-based data management software (Zotero, version 4.0.29, Roy Rosenzweig Center for History and New Media, Fairfax, United States of America).
One author conducted the searches and removed the duplicates. Two authors then independently screened titles and abstracts for eligibility; if they deemed papers eligible, the full text was retrieved and again screened independently. We also scanned reference lists of included studies for potential relevant papers. Any differences in screening outcomes were resolved either by consensus, or by involving the senior author.
Study selection
We included a wide range of study designs including trials, observational studies, meta-analyses and modelling studies. Modelling studies added value by allowing the evaluation of certain interventions using different scenarios where empirical data are impractical or lacking (e.g. food labelling). However, we analysed them separately from the empirical studies. To be eligible, studies had to include the effectiveness of specific interventions on dietary TFA intake and have quantitative outcomes. Only studies published in English language were included. We assessed retrieved studies by using the population, intervention, comparison and outcomes study design approach (Box 2).14
The primary outcome was dietary TFA intake in a population, reported as g/day. For studies reporting TFA intake as a percentage of total energy intake (E%), we converted the data to g/day using the conventional formula below (1).16
\[
\text{Trans fat intake (g/day)} = \frac{\text{Trans fat intake E%}}{9} \times \frac{100}{\text{day}}
\]
Box 1. Search strategy used in the systematic review of dietary trans-fatty acid reduction policies
The following keywords were used in a search of CINAHL, the Centre for Reviews and Dissemination Wider Public Health database, Cochrane Database of Systematic Reviews, Ovid®, MEDLINE®, Science Citation Index and Scopus:
("(Trans-fats" OR "trans-fats" OR "Trans-fat" OR "Diary trans-fat" OR "Industrial trans-fat") AND ("Refomulation" OR "Regulation" OR "Self-regulation" OR "Labelling" OR "Limits" OR "Ban" OR "Elimination" OR "Legislation" OR "Agreements" OR "Campaigns" OR "Tax" OR "Health promotion" OR "Nutrition education" OR "Marketing control")
For Scopus, additional terms were used to narrow down the search:
("(Intake" OR "Consumption" OR "Composition" OR "Content" OR "Effect" OR "Effectiveness" OR "Cons-effictiveness") AND ("Public policy" OR "Nutrition policy" OR "Health policy" OR "Policies" OR "Interventions" OR "Strategies" OR "Initiatives" OR "Policy options" OR "Actions")
Box 2. Inclusion and exclusion criteria for selecting studies for the systematic review of dietary trans-fatty acid reduction policies
Participants
We included studies covering all age groups from all populations, from high-, middle- and low-income countries. We excluded studies on pregnant women, animals and cells.
Interventions
We included primary studies and systematic reviews evaluating the effects of actions to promote TFA reduction by government policy or adopted in specific real or experimental settings. We excluded studies evaluating the effect of a general or specific diet.
Comparators
We included studies where actions to promote TFA reduction were evaluated or compared. We excluded studies with no evaluation or comparison of actions to promote TFA reduction.
Outcomes
We included primary outcome of interest was change in dietary TFA intake, reported as g/day. When TFA intake was reported as percentage of energy from total fat, values were converted to g/day. Secondary outcomes included changes in clinical or physiological indicators related to noncommunicable diseases and behaviours associated with a healthy diet.
We excluded process evaluations reporting on implementation of interventions or policies without any quantitative outcome data; feasibility or acceptability studies without an assessment or primary outcomes (intake); studies on individuals as opposed to populations; data on cost only; reduction in product content only; body mass index studies.
Study design
We included primary studies, randomized controlled trials, systematic reviews, empirical observational studies, natural experiments, secondary analysis, before and after interventions, and modelling studies.
We excluded commentary or opinion articles; purely qualitative evaluations with no quantitative assessment.
TFA: trans-fatty acid.
1 An empirical study whereby the exposure to the experimental and control groups are determined by nature or other factors outside the control of the researchers.
Note: Based on the population, intervention, comparison, outcome and study design14 inclusion and exclusion criteria.
Trans fat intake(\(\frac{g}{\text{day}}\)) = \(\frac{\text{Trans fat intake E%}}{9} \times \frac{100}{\text{day}}\) (1)
Papers only reporting on the reduction of TFA content in food products without a link to intake were excluded.
Secondary outcomes included changes in clinical or physiological indicators related to noncommunicable diseases (e.g. coronary heart disease deaths) and behaviours associated with a healthy diet.
**Data extraction and quality assessment**
We used pre-designed and pre-piloted data extraction forms to extract data from all included studies. One author conducted the data extraction, which was independently checked by other authors for the empirical and modelling studies.
We used the National Heart, Lung and Blood Institute quality assessment tools to assess the quality of empirical studies. Two authors independently assessed the methodological quality of each study as poor, fair or good. Modelling studies were independently assessed by two modelling experts using an adapted version of a published quality assessment tool. Quality was reported as poor, fair or good. Any disagreements were resolved by consensus or with another author.
**Fig. 1. Interventions classified on the upstream–downstream continuum in the systematic review of dietary trans-fatty acid reduction policies**
Upstream interventions
- Legislation or mandatory reformulation of food products
- Taxation and subsidies
- Voluntary reformulation of food products
- Food labelling
- Mass media campaigns
- Community dietary counselling
- Worksite dietary counselling
- Individual dietary counselling
Downstream interventions
Note: Based on the McLaren continuum of interventions.8
**Fig. 2. Flowchart used for the systematic review of dietary trans-fatty acid reduction policies**
1785 records identified through database search
22 additional records identified through other sources
1084 records screened after duplicates removed
1014 records excluded
47 full-text articles excluded
- 23 had no evaluation of TFA interventions (descriptive)
- 11 had no focus on TFA or TFA interventions
- 7 had focus on other outcomes
- 2 studied the effect of TFA reduction rather than effect of a policy
- 3 were opinion articles or conference abstracts
- 1 was a systematic review; studies were included individually if they met the criteria
70 full-text articles assessed for eligibility
23 studies included in qualitative synthesis
Note: Based on the preferred reporting items for systematic reviews and meta-analyses guidelines.14
**Data synthesis and analysis**
We conducted a narrative synthesis to interpret the data, grouping the studies according to intervention type. In accordance with the McLaren continuum of structural–agentic interventions8 and the Nuffield ladder taxonomy,19 we defined downstream (agentic) interventions as those where the principal mechanism of action is dependent on individuals altering their consumption behaviour. Conversely, we defined upstream (structural) interventions as those creating changes that target an entire population – not a subset, however large – thus effectively eliminating individual agency. We then categorized interventions according to their position in the McLaren continuum (Fig. 1).8 We separately analysed multicomponent interventions. An unweighted median regression model was fitted to the TFA intake data from eight empirical studies and four modelling studies.
**Results**
The literature search identified 1785 potentially relevant papers, and 22 additional papers were identified through screening reference lists. After removing 723 duplicates, we screened 1084 publications by title and abstract, after which 70 full-text papers were assessed for eligibility. A total of 23 papers met the inclusion criteria (12 empirical studies20–31 and 11 modelling studies;32–42 Fig. 2). The interventions and their effect sizes are presented in Fig. 3.
**Table 1** summarizes the empirical studies and **Table 2** summarizes modelling studies included in this review (both tables available at: http://www.who.int/bulletin/volumes/95/12/16-189795).
**Individual dietary counselling**
One study in 2007, rated as fair quality, targeted Aboriginal families in Canada. It investigated the effect of dietary counselling on dietary intake, including TFA, using a 24-hour recall. The 29 intervention households significantly reduced their consumption of TFA ($P = 0.02$) from 0.6 to 0.5 g/day over 6 months, while the 28 control households increased consumption from 0.7 to 1.3 g/day.20
Food reformulation
Empirical studies
Two empirical studies examined the effectiveness of reformulating industrially produced foods. A good-quality study reported the results of a TFA monitoring programme after voluntary reformulation limits for TFA content were put in place in Canada in 2005 on vegetable oils and soft margarines and other pre-packaged foods (< 2% and < 5% of total fat, respectively). TFA intake, measured using 24-hour food recalls in the general population (33 030 people), fell from 4.9 g/day in 2004 to 3.4 g/day in 2008.\(^22\)
The other study, rated as fair quality, conducted two meta-analyses of North American and European data. Both analyses investigated the effect of TFA consumption on coronary heart disease risk factors (one included 13 randomized controlled trials and the other covered four prospective studies). The researchers calculated the effect of reformulating the TFA content of partially hydrogenated vegetable oils with other fats. They found higher risk reductions when reformulating oils with higher TFA content. The randomized controlled trials reported an approximately 19% risk reduction, whereas the prospective studies found a 39% reduction in coronary heart disease risk.\(^23\)
Modelling studies
One study of poor quality in 2011 modelled the effect of reformulating foods to reduce TFA in the Netherlands. TFA intake in the population of 750 participants aged 19–30 years was projected to fall from 2.3 to 1.9 g/day after reformulating specific food groups (e.g. bread, pastry, cakes and biscuits; meat snacks and salads; fat and margarines).\(^24\) Another study of fair quality modelled the effect of food reformulation on health outcomes in Denmark in 2016. The researchers estimated reductions in cardiovascular disease deaths of 14.2 (from 441.5 to 427.3) per 100 000 population over the years 2004–2006 and coronary heart disease deaths 26.5 per 100 000 population.\(^25\) Finally, one good-quality modelling study of EU-level policy options investigated the cost-effectiveness of voluntary reformulation compared with no intervention. The study estimated 2.19 of 1075 million DALYs could be averted from coronary artery disease over the course of a person’s lifetime (85 years). A projected 23 billion euros (€) estimated that labelling is at best only half as effective as a total ban on TFAs in terms of health and socioeconomic benefits. Improved labelling might save approximately 3500 deaths from coronary heart disease over the period 2015–2020 (1.3%, 3500 of the total 273 000 deaths) and reduce socioeconomic inequalities by some 1500 deaths (from 20 400 to 18 900).\(^22\) Another modelling study in 2016 investigated the cost-effectiveness of mandatory labelling in products on sale in the European Union (EU) and projected that it may prevent 0.98 million of the 1076 million disability-adjusted life years (DALYs) attributed to coronary artery disease. However, compared with taking no action, this option incurred greater costs than it saved, in terms of health-care costs, lost productivity and implementation costs.\(^26\) In a similar approach to the Dutch study, researchers in Brazil investigated replacing products with ones that complied with a healthier choices logo.\(^27\) Estimated TFA intakes were 0.8 g/day (standard deviation, SD: 1.0) for typical menus, 0.1 g/day (SD: 0.2) for the choices menus and 0.2 g/day (SD: 0.3) for energy-adjusted choices menus, i.e. the same as choices menu, but adjusted for energy of a typical menu.
Worksite dietary counselling
One poor-quality study in 2010 in the USA implemented a dietary counselling programme at a worksite by educating participants on the use of a low-fat vegan diet. After 22 weeks, 3-day dietary records showed that the 45 control participants had increased their TFA intake from 2.4 to 2.5 g/day over 22 weeks, whereas the 68 participants in the intervention group reduced their intake from 2.1 to 1.1 g/day ($P \leq 0.001$).\(^21\)
Food labelling
Modelling studies
We found no empirical studies investigating the sole effect of labels showing the TFA content of food products, but we included five modelling studies (two of good quality,\(^22,23\) two fair,\(^24,25\) and one of poor quality\(^26\)). In the Netherlands a healthier choices logo for food packages was implemented in 2006. Replacing all packaged products with those that complied with a healthier choices logo was projected to lead to a 0.8 g/day reduction in TFA intake from 2.1 to 1.3 g/day.\(^24\) Another Dutch study in 2013, using a similar approach, projected a higher reduction of 1.2 g/day (from 2.2 to 1.0 g/day).\(^24\) A British study in 2015 estimated that labelling is at best only half as effective as a total ban on TFAs in terms of health and socioeconomic benefits. Improved labelling might save approximately 3500 deaths from coronary heart disease over the period 2015–2020 (1.3%, 3500 of the total 273 000 deaths) and reduce socioeconomic inequalities by some 1500 deaths (from 20 400 to 18 900).\(^22\) Another modelling study in 2016 investigated the cost-effectiveness of mandatory labelling in products on sale in the European Union (EU) and projected that it may prevent 0.98 million of the 1076 million disability-adjusted life years (DALYs) attributed to coronary artery disease. However, compared with taking no action, this option incurred greater costs than it saved, in terms of health-care costs, lost productivity and implementation costs.\(^26\) In a similar approach to the Dutch study, researchers in Brazil investigated replacing products with ones that complied with a healthier choices logo.\(^27\) Estimated TFA intakes were 0.8 g/day (standard deviation, SD: 1.0) for typical menus, 0.1 g/day (SD: 0.2) for the choices menus and 0.2 g/day (SD: 0.3) for energy-adjusted choices menus, i.e. the same as choices menu, but adjusted for energy of a typical menu.
could be saved due to reductions in direct health-care costs and in indirect costs linked to informal care, and these outweighed the costs of implementing this policy.31
**Legislation**
**Empirical studies**
Analysing TFA g/purchase not TFA g/day, one good-quality study investigated the effect of regulations on TFA content of food sold in New York city. Instituted in 2007, the policy allowed takeaway food restaurants to sell only products with a ≤0.5 g TFA content per serving. The estimated mean TFA intake per purchase, based on purchase receipts matched to products, decreased from 2.9 g in 2007 (6969 purchases) to 0.5 g in 2009 (7885 purchases).24
Another study of good quality analysed the impact of a upper limit of 0.5 g serving (commonly referred to as a TFA ban) on TFAs in all food-service establishments in the USA in 2016 and found a 4.5% reduction in cardiovascular disease mortality, from 13 per 100000 persons between 2010 and 2013.25
Finally, a study of good quality investigated the effect of TFA restrictions in restaurants in certain New York state counties implemented in 2007, comparing hospital admissions for cardiovascular events in counties with and without restrictions. Three or more years after the restrictions were implemented there were significantly fewer cardiovascular events in the intervention population of 8.4 million adults compared with the reference population of 3.3 million, after adjusting for temporal trends. These changes applied when analysing myocardial infarction and stroke events combined (change of −6.2%; 95% confidence interval, CI: −9.2% to −3.2%) and myocardial infarction alone (−7.8%; 95% CI: −12.7% to −2.8%).26
**Modelling studies**
Five modelling studies, all of good quality, modelled the effect on health outcomes of legislation affecting TFA consumption.29,30,31,33,39–41 One British study projected that a ban on TFAs in sit-down food restaurants to sell only products with a ≤0.5 g TFA content per serving. The estimated mean TFA intake per purchase, based on purchase receipts matched to products, decreased from 2.9 g in 2007 (6969 purchases) to 0.5 g in 2009 (7885 purchases).24
Another study of good quality analysed the impact of a upper limit of 0.5 g serving (commonly referred to as a TFA ban) on TFAs in all food-service establishments in the USA in 2016 and found a 4.5% reduction in cardiovascular disease mortality, from 13 per 100000 persons between 2010 and 2013.25
Finally, a study of good quality investigated the effect of TFA restrictions in restaurants in certain New York state counties implemented in 2007, comparing hospital admissions for cardiovascular events in counties with and without restrictions. Three or more years after the restrictions were implemented there were significantly fewer cardiovascular events in the intervention population of 8.4 million adults compared with the reference population of 3.3 million, after adjusting for temporal trends. These changes applied when analysing myocardial infarction and stroke events combined (change of −6.2%; 95% confidence interval, CI: −9.2% to −3.2%) and myocardial infarction alone (−7.8%; 95% CI: −12.7% to −2.8%).26
**Multicomponent interventions**
Four fair-quality empirical studies investigated the combined effect of more than one policy on TFA consumption.27–30 One cross-sectional study evaluated the effect of both a public health education programme and voluntary reformulation of industrially produced TFAs in soybean oil in Costa Rica. Using 3-day food records the study found a 1.7 g/day reduction in TFA intake among adolescents from 4.5 g/day (276 people) in 1996 to 2.8 g/day (133 people) in 2006.27
Two other studies included both labelling and voluntary reformulation limits on TFA content of industrial food and estimated this could prevent around 1700 coronary heart disease deaths and gain some 15 000 life years annually over a decade. Eliminating both natural- and industrial-derived TFAs may prevent some 3300 coronary heart disease deaths annually and gain approximately 30 000 life years.41 Finally, a study in 2016 modelled the cost-effectiveness of different EU policy options to reduce the TFA intake of the population. Placing legal limits on the TFA content in industrial foods was projected to avoid 3.73 DALYs and save € 51 billion from health-care costs and lost productivity in EU Member States.33
**Discussion**
Our systematic review suggests that multicomponent interventions achieve the biggest reductions in TFA consumption across an entire population, as demonstrated in Denmark, Canada and Costa Rica. Systematic reformulation of products containing TFAs can also help, as observed in Canada and the USA. Interventions targeting individuals typically achieved smaller reductions in TFA consumption.
Several studies, in separate countries, investigated multicomponent interventions; all of them included food reformulation, labelling and voluntary limits on TFA content of industrial food. In Denmark, a progressive series of interventions finally led to a legislative ban on TFA. This package achieved the largest observed reduction in TFA intake over the population over the period from 1976 to 2005 (4.5 g/day).12,31 The USA is now emulating this successful strategy.42 Substantial, but smaller benefits were achieved by multi-intervention strategies lacking a legislative component in Costa Rica32 and Canada.28,32 Multicomponent strategies including upstream policies, such as price regulations or legislation, consistently achieved greater reductions in TFA intake than single interventions, particularly when these were downstream approaches focused on individuals.
Legislation to regulate TFA content in food achieved a 2.4 g/day reduction in intake of TFA in the city of New York.24 This success has now been extended...
nationwide by the United States Food and Drug Administration ruling in June 2015 stating that partially hydrogenated oils are no longer generally recognized as safe for use in food.44 Following the Danish exemplar, several other European countries have subsequently introduced legislation, setting an upper limit for TFAs of 2 g per 100 g in fat or oil.11 However, other countries still rely on voluntary reformulation, which is less effective.12,13 Legislation is routinely opposed by the food industry, fearful of decreased profits or the additional costs of reformulating products.12,13,15 However, the evidence is that such legislation has generally had minimal financial impact on the food industry.12,24,45
Several evaluation studies have reported reductions in TFA content of margarines and industrially produced foods, particularly using mandatory reformulation.24,46 An early concern that manufacturers would substitute saturated fats such as palm oil for TFA has been dismissed. Both Canadian and American studies of reformulation have reported reductions in both TFA and saturated fats, and increases in unsaturated fat,13 the preferred replacements for TFA.13 Food labelling has the potential to help consumers make more informed decisions48 and can also put pressure on the food industry to reformulate products.49 We found no empirical studies of the effects of food labelling alone. However, several modelling studies estimated reductions in TFA consumption ranging from 0.3 g/day to 1.2 g/day.12,14,15 In contrast, most studies involving food labelling have examined the public’s understanding of labels, use of labels, food purchase and purchase-related behaviour rather than quantifying TFA intake.49 The diverse labelling systems and claims currently generated by the industry may confuse many consumers, highlighting a need for package labels that are easier to understand.48
Dietary counselling interventions in different settings, such as communities, worksites, schools and homes, were sparse. One study in the home10 and another in worksites17 both suggested that dietary counselling at an individual level could achieve TFA reductions of 0.8 g/day20 and 1.2 g/day.21 In practice, individuals may struggle to adhere and comply with dietary advice due to competing priorities or financial constraints.48
We found no studies of mass media campaigns focusing on TFA. Media campaigns can achieve modest beneficial behaviour changes in nutrition, physical activity and tobacco and alcohol use – in motivated individuals. However, the overall population benefits tend to be more modest.52
Taxation has been shown to be a potentially powerful tool for reducing consumption of tobacco, alcohol or sugar sweetened beverages. However, we found no studies of taxation focusing specifically on TFAs.13
Our systematic review has several strengths. We did the screening, extraction and analysis in duplicate, and all the included studies were subjected to a rigorous quality appraisal. Furthermore, modelling studies were included, but considered separately, in recognition of their additional uncertainties when compared with empirical evidence. Our review also had limitations. We were unable to conduct a meta-analysis due to the profound heterogeneity of the studies and because several studies included multiple interventions. We only included studies in English. The primary outcome of this study was dietary intake and we excluded studies considering other components of dietary behaviour or changes in product content after reformulation. Generalization of the results should be cautious because countries will vary in their baseline TFA intake. Furthermore, we did not contact authors for missing data. Publication bias also remains possible, potentially overestimating the true effect of some interventions. Finally, some intervention benefits may have been overestimated due to underlying secular trends among the public towards lower TFA consumption.
In conclusion, our results suggest an effectiveness hierarchy similar to those previously demonstrated in salt, tobacco and alcohol control interventions.9,10,53 Multicomponent interventions including a legislative ban on products appear the most effective strategy to reduce TFA intake. By contrast, more downstream interventions targeting individuals in domestic or work settings appear consistently less effective. Future prevention strategies might consider this effectiveness hierarchy to achieve the largest reductions in the consumption of TFAs or other harmful nutrients.
Competing interests: None declared.
The new approach to preparing foods and interventions with trans fatty acids was more effective. The legislative measure that prohibited the content of manufactured foods with trans fatty acids. And it succeeded the legislative measures that prohibited the content of manufactured foods with trans fatty acids. In what other countries, regulatory measures that imposed a new way of preparing foods in reducing the incidence of the problem was higher. And the interventions that involved the workplace were more effective in reducing the incidence of the problem. And the interventions in the workplace were more effective in reducing the incidence of the problem.
In the perspective of the future, future prevention strategies should take into account this hierarchy of effectiveness when considering the interventions that targeted individuals to achieve lower reduction rates. And taking into account this hierarchy of effectiveness when considering the interventions that targeted individuals to achieve lower reduction rates.
They surveyed 2312 empirical and 11 modeling studies. After examining 1084 studies, we excluded studies that only reported trans fatty acids (TFA) consumption. We included trials, observational studies, meta-analyses and modeling studies. We conducted a descriptive synthesis to interpret the data grouped into a continuum ranging from the least effective interventions targeting individuals to achieve lower reduction rates to the most effective interventions targeting individuals to achieve higher reduction rates. And taking into account this hierarchy of effectiveness when considering the interventions that targeted individuals to achieve lower reduction rates.
Revue systématique des interventions visant à réduire les acides gras trans alimentaires
Objectif Effectuer une revue systématique des études publiées portant sur des interventions qui visent à réduire la consommation par les individus d’acides gras trans alimentaires.
Méthodes Nous avons recherché dans des bases de données en ligne (CINAHL, CRD Wider Public Health database, The Cochrane Database of Systematic Reviews, Ovid®, MEDLINE®, Science Citation Index et Scopus) des études évaluant les interventions relatives aux acides gras trans entre 1986 et 2017. Le principal critère d’évaluation était la baisse absolue de la consommation d’acides gras trans/jour. Nous avons exclu les études mentionnant uniquement la teneur en acides gras trans dans les produits alimentaires sans établir de lien avec leur consommation. Nous avons inclus des essais, des études observationnelles, des méta-analyses et des études de modélisation. Nous avons réalisé une synthèse descriptive afin d’interpréter les données en regroupant les études selon un continuum allant des interventions ciblant des individus à des changements structurels au niveau de la population.
Résultats Après avoir examiné 1084 études, nous en avons sélectionné 23 : 12 études empiriques et 11 études de modélisation. Au Danemark, de multiples interventions ont permis de faire passer la consommation d’acides gras trans de 4,5 g/jour en 1976 à 1,5 g/jour en 1995, jusqu’à une élimination virtuelle suite à l’adoption d’une loi en 2004 interdisant les acides gras trans dans les produits alimentaires industriels. Dans d’autres pays, des réglementations imposant une refonte des aliments ont réduit la teneur en acides gras trans d’environ 2,4 g/jour. Des interventions sur les lieux de travail ont induit des baisses de 1,2 g/jour en moyenne. L’étiquetage des produits alimentaires et la fourniture de conseils diététiques individualisés ont quant à eux entraîné des diminutions d’environ 0,8 g/jour.
Conclusion La stratégie la plus efficace consistait en des interventions à plusieurs composantes incluant une loi destinée à éliminer les acides gras trans des produits alimentaires. La refonte des produits alimentaires et d’autres interventions à plusieurs composantes ont également permis de réduire efficacement la consommation d’acides gras trans. En revanche, les baisses induites par les interventions ciblant des individus étaient systématiquement moins importantes. Cette hiérarchie de l’efficacité doit être prise en compte dans les futures stratégies de prévention afin de réduire au maximum la consommation d’acides gras trans.
Цель
Проведение систематического обзора опубликованных исследований мероприятий, направленных на сокращение потребления с пищей транс-изомеров жирных кислот (ТИЖК).
Методы
Был проведен поиск баз данных в Интернете (CINAH, база данных CRD Wider Public Health, Cochrane Database of Systematic Reviews (Кохрэновская база данных систематических обзоров), Ovid®, MEDLINE®, Science Citation Index (Индекс научного цитирования) и Scopus) на предмет исследований, оценивающих мероприятия по сокращению потребления ТИЖК за период с 1986 по 2017 год. Абсолютное снижение потребления ТИЖК (г/сут) было основным критерием результата. Исследования, сообщающие только о содержании ТИЖК в продуктах питания без указания количества потребления, были исключены. В обзор также включались испытания, обсервационные исследования, метаанализы и исследования с применением моделирования. Исследователи провели нарративный синтез, чтобы интерпретировать данные, группируя исследования по континууму, начиная с мероприятий, направленных на отдельных лиц, и заканчивая структурными изменениями, затрагивающими все слои населения.
Результаты
После отбора из 1084 статей-кандидатов в обзор были включены 23 статьи: 12 эмпирических и 11 модельных исследований. Многочисленные мероприятия в Дании привели к сокращению потребления ТИЖК с 4,5 г/сут в 1976 году до 1,5 г/сут в 1995 году, а затем к фактическому устранению ТИЖК из рациона в 2004 году после принятия законодательства, запрещающего присутствие ТИЖК в готовых продуктах питания. В других местах правила, предусматривающие изменение состава продуктов питания, способствуют снижению потребления ТИЖК примерно на 2,4 г/сут. Мероприятия на рабочих местах привели к сокращению потребления ТИЖК в среднем на 1,2 г/сут. Маркировка продуктов питания и индивидуальное консультирование диетологов привела к снижению потребления примерно на 0,8 г/сут.
Вывод
Наиболее эффективная стратегия — многокомпонентные мероприятия, в том числе введение в силу законов, запрещающих присутствие ТИЖК в продуктах питания. Изменение рецептур продуктов питания и другие многочисленные мероприятия также привели к заметному сокращению потребления ТИЖК. Напротив, мероприятия, последовательно направленные на отдельных лиц, приводили к меньшему сокращению потребления. Стратегии будущей профилактики должны учитывать эту иерархию эффективности для достижения максимального сокращения потребления ТИЖК.
References
1. Forouzanfar MH, Alexander L, Anderson HR, Bachman VF, Biryukov S, Brauer M, et al.; GBD 2013 Risk Factors Collaborators. Global, regional, and national comparative risk assessment of 79 behavioural, environmental and occupational, and metabolic risk or clusters of risks in 188 countries, 1990-2013: A systematic analysis for the Global Burden of Disease Study 2013. Lancet. 2015 Dec 5;386(10010):2287–323. doi: http://dx.doi.org/10.1016/S0140-6736(15)00128-2 PMID: 26364544
2. Mozaffarian D, Katan MB, Ascherio A, Stampfer MJ, Willett WC. Trans fatty acids and cardiovascular disease. N Engl J Med. 2006 Apr 13;354(15):1601–13. doi: http://dx.doi.org/10.1056/NEJMr054035 PMID: 16611951
3. Kaur G, Cameron-Smith D, Sinclair AJ. Are trans fats a problem in Australia? Med J Aust. 2012 Jun 18;196(11):666–7. doi: http://dx.doi.org/10.5694/mja1.10372 PMID: 22708749
4. Ascherio A, Willett WC. Health effects of trans fatty acids. Am J Clin Nutr. 1997 Oct;66(4 Suppl):1006S–105S. PMID: 9323581
5. Mozaffarian D, Aro A, Willett WC. Health effects of trans-fatty acids: experimental and observational evidence. Eur J Clin Nutr. 2009 May;63 Suppl 2:S1–21. doi: http://dx.doi.org/10.1038/sj.ejn.1602973 PMID: 19424218
6. Diet, nutrition and the prevention of chronic diseases. Report of a joint WHO/FAO expert consultation. Geneva: World Health Organization; 2003.
7. Backholer K, Peeters A. The impact of trans fat regulation on social inequalities in coronary heart disease in Australia. Med J Aust. 2013 Aug 5;199(3):166. doi: http://dx.doi.org/10.5694/mja1.111942 PMID: 2390932
43. Doell D, Folmer D, Lee H, Hongfort M, Carberry S. Updated estimate of trans fat intake by the US population. Food Addit Contam Part A Chem Anal Control Expo Risk Assess. 2012;29(6):861–74. doi: http://dx.doi.org/10.1080/19440049.2012.664570 PMID: 22439632
44. United States Food and Drug Administration. Final determination regarding partially hydrogenated oils. Docket no. FDA-2013-N-1317. Federal Register. 2015; 80(116). Washington DC: U.S. Government Publishing Office: 2015. Available from: https://www.gpo.gov/fdsys/pkg/FR-2015-06-17/pdf/2015-14883.pdf [cited 2016 Sep 17].
45. Skeaff CM. Feasibility of recommending certain replacement or alternative fats. Eur J Clin Nutr. 2009 May;63 Suppl 2:S34–49. doi: http://dx.doi.org/10.1038/sj.ejcn.1602974 PMID: 19424217
46. Ricciuto L, Lin K, Tarasuk V. A comparison of the fat composition and prices of margarines between 2002 and 2006, when new Canadian labelling regulations came into effect. Public Health Nutr. 2009 Aug;12(8):1270–5. doi: http://dx.doi.org/10.1017/S1368980008003868 PMID: 18986592
47. Ratnayake WMN, L’Abbé MR, Mozaffarian D. Nationwide product reformulations to reduce trans fatty acids in Canada: when trans fat goes out, what goes in? Eur J Clin Nutr. 2009 Jun;63(6):808–11. doi: http://dx.doi.org/10.1038/ejcn.2008.39 PMID: 18594558
48. Campos S, Doxey J, Hammond D. Nutrition labels on pre-packaged foods: a systematic review. Public Health Nutr. 2011 Aug;14(8):1496–506. doi: http://dx.doi.org/10.1017/S1368980010003290 PMID: 21241532
49. Schleifer D. Categories count: trans fat labeling as a technique of corporate governance. Soc Stud Sci. 2013 Feb;43(1):54–77. doi: http://dx.doi.org/10.1177/0306312712461573
50. Hess S, Yanes M, Jourdan P, Edelstein S. Trans fat knowledge is related to education level and nutrition facts label use in health-conscious adults. Topics Clin Nutr. 2005 Apr;20(2):109–17. doi: http://dx.doi.org/10.1097/00008486-200504000-00004
51. Ashenden R, Silagy C, Weller D. A systematic review of the effectiveness of promoting lifestyle change in general practice. Fam Pract. 1997 Apr;14(2):160–76. doi: http://dx.doi.org/10.1093/fampra/14.2.160 PMID: 9137956
52. Wakefield MA, Loken B, Hornik RC. Use of mass media campaigns to change health behaviour. Lancet. 2010 Oct 9;376(9748):1261–71. doi: http://dx.doi.org/10.1016/S0140-6736(10)60809-4 PMID: 20933263
53. Hyseni L, Elliot-Green A, Lloyd-Williams F, Kypridemos C, O’Flaherty M, McGill R, et al. Systematic review of dietary salt reduction policies: Evidence for an effectiveness hierarchy? PLoS One. 2017 05 18;12(5):e0177535. doi: http://dx.doi.org/10.1371/journal.pone.0177535 PMID: 28542317
### Empirical studies categorized according to type of intervention included in the systematic review of dietary trans-fatty acid reduction policies
| Study | Study design | Study aim | Intervention analysed | Geographical scope | Participants and sample size | Methods | Outcomes | Comments | Quality |
|-------|--------------|-----------|-----------------------|--------------------|-----------------------------|---------|----------|----------|---------|
| **Individual dietary counseling** |
| Anand et al., 2007<sup>20</sup> | Randomized open trial | To determine if a household- based lifestyle intervention is effective at reducing energy intake and increasing physical activity among Aboriginal families | Education about healthy weights | Canada | 57 Aboriginal households (174 individuals) | Families were recruited between May 2004 and April 2005 and completed health assessment at baseline and 6 months after randomization to intervention or control groups. Aboriginal health counsellors made regular home visits to assist families in setting their dietary and physical activity goals. Energy and nutrient intake was measured using a 24-hour recall. The Food Processor program (ESP, Research, Salem, USA) was used for nutrient analysis. | Intervention households decreased their consumption of TFAs from 0.6 to 0.5 g/day (–0.1 g/day) versus +0.6 g/day (P = 0.02) after 6 months compared with non-intervention households who increased their consumption from 0.7 to 1.5 g/day. | There may have been bias due to self-reporting of lifestyle changes and the 24-hour dietary recall. The focus of the intervention was not TFA but many other dietary components as well as knowledge and attitudes about healthy dietary practices. | Fair |
| **Worksite dietary counselling** |
| Lof et al., 2010<sup>21</sup> | Workplace- based dietary intervention | To examine whether a workplace nutrition programme using a low-fat vegan diet could significantly improve nutritional intake | Health promotion and nutrition education | USA | 100 participants (65 intervention and 44 control) | At weeks 0 and 22, participants completed 3-day dietary recalls to assess energy and nutrient intake. At the intervention site, participants were asked to follow a low-fat vegan diet and participate in weekly group meetings that included instruction and group support (intervention group). At the control site, participants received no instruction (control group). | In the intervention group, reported intake of TFA decreased significantly (0.110 g) compared with the control group. TFA intakes in the intervention group were 2.1 g (SD 0.13) at baseline and 1.1 g (SD 0.1) after 22 weeks. Mean difference = 1.0 g/day (SD 0.2). TFA intake in the control group was 2.4 g (SD 0.1) at baseline and 2.5 g (SD 0.2) after 22 weeks. Mean difference = 0.2 g/day (SD 0.2). | There may be a significant amount of self-reporting of lifestyle changes and 22-week dietary recall. The focus was on nutrition education about a low-fat, vegan diet not TFA intake. | Poor |
| **Food reformulation** |
| Rathbone et al., 2009<sup>22</sup> | Cross-sectional | To report the results of a TFA monitoring programme | Voluntary food reformulation and limits on TFA in food sold by retailers or food service establishments | Canada | Over 33,000 respondents did a dietary recall repeated by a subset of 10,000. Respondents with null or invalid recalls, children aged <1 year and breastfeeding children were excluded | 519 participants | A previous study showed that the estimated average intake of TFA in Canada was 8.4 g/day in the mid-1990s. This study showed that TFA intake in 2004 dropped significantly to 4.9 g/day and declined further to 3.4 g/day in 2009. On average, there was a 30% decrease in TFA intake between 2004 and 2008. | Good |
| **Mozaffarian and Clarke, 2009<sup>23</sup>** | Meta- analysis of randomized controlled trials | To evaluate the effect on coronary heart disease risk after reformulation of edible oils to reduce TFA consumption | Reformulation of products containing fatty acids | Unlimited | Three different partially hydrogenated vegetable oil formulations (containing 20%, 35% or 45% TFA) were replaced with other fats or oils. Effects on coronary heart disease risk were estimated based on iso-caloric replacement of 7.5% of energy from partially hydrogenated vegetable oil in an individual’s diet. | For partially hydrogenated vegetable oils with 20% TFA, replacement with butter would result in a small net decrease in coronary heart disease risk (2.7%), while replacement with palm oil or lard would modestly increase risk by 7.6% and 6.0%, respectively. Replacement with soybean, canola or high-oleic sunflower oils would produce the largest coronary heart disease risk reductions (8.4–9.9%). For partially hydrogenated vegetable oil with 35% TFA, risk reductions for replacement fats and oils ranged from 11.9% to 16.0%. Largest predicted declines in coronary heart disease risk were for replacement with vegetable oils. For partially hydrogenated vegetable oil with 45% TFA, predicted risk reductions were the highest, including risk reductions of 19.8% and 10.9% for replacement with soybean and canola oil, respectively. | For partially hydrogenated vegetable oil with 20% TFA, replacement with butter would result in a small net decrease in coronary heart disease risk (2.7%), while replacement with palm oil or lard would modestly increase risk by 7.6% and 6.0%, respectively. Replacement with soybean, canola or high-oleic sunflower oils would produce the largest coronary heart disease risk reductions (8.4–9.9%). For partially hydrogenated vegetable oil with 35% TFA, risk reductions for replacement fats and oils ranged from 11.9% to 16.0%. Largest predicted declines in coronary heart disease risk were for replacement with vegetable oils. For partially hydrogenated vegetable oil with 45% TFA, predicted risk reductions were the highest, including risk reductions of 19.8% and 10.9% for replacement with soybean and canola oil, respectively. | Fair |
| **Mozaffarian and Clarke, 2009<sup>23</sup>** | Meta- analysis of prospective cohort studies | To evaluate the effect on coronary heart disease risk after reformulation of vegetable oils to reduce TFA consumption | Reformulation of products containing fatty acids | North America and Europe | 490,000 coronary heart disease cases prospectively ascertained among 1,986,886 participants | These different partially hydrogenated vegetable oil formulations (containing 20%, 35% or 45% TFA) were replaced with other fats or oils. Effects on coronary heart disease risk were estimated based on iso-caloric replacement of 7.5% of energy from partially hydrogenated vegetable oil in an individual’s diet. | For partially hydrogenated vegetable oils with 20% TFA, replacement with butter would result in a small net decrease in coronary heart disease risk (2.7%), while replacement with palm oil or lard would modestly increase risk by 7.6% and 6.0%, respectively. Replacement with soybean, canola or high-oleic sunflower oils would produce the largest coronary heart disease risk reductions (8.4–9.9%). For partially hydrogenated vegetable oil with 35% TFA, risk reductions for replacement fats and oils ranged from 11.9% to 16.0%. Largest predicted declines in coronary heart disease risk were for replacement with vegetable oils. For partially hydrogenated vegetable oil with 45% TFA, predicted risk reductions were the highest, including risk reductions of 19.8% and 10.9% for replacement with soybean and canola oil, respectively. | For partially hydrogenated vegetable oil with 20% TFA, replacement with butter would result in a small net decrease in coronary heart disease risk (2.7%), while replacement with palm oil or lard would modestly increase risk by 7.6% and 6.0%, respectively. Replacement with soybean, canola or high-oleic sunflower oils would produce the largest coronary heart disease risk reductions (8.4–9.9%). For partially hydrogenated vegetable oil with 35% TFA, risk reductions for replacement fats and oils ranged from 11.9% to 16.0%. Largest predicted declines in coronary heart disease risk were for replacement with vegetable oils. For partially hydrogenated vegetable oil with 45% TFA, predicted risk reductions were the highest, including risk reductions of 19.8% and 10.9% for replacement with soybean and canola oil, respectively. | Good |
(continues…)
Multi-component interventions
Loth et al., 2006
Pre-post test
To assess the effectiveness of Denmark’s ban on TFA in industrial food products
TFA ban
New York state, USA
11 countries with TFA restrictions and 25 countries without TFA restrictions were included. In 2006, the years before the first restrictions were implemented, there were 8.4 million adults aged 45+ in 52 high-income countries with TFA restrictions and 3.3 million adults in 51 high-income countries without restrictions.
TFA restrictions were implemented in 2004 in Denmark. Annual hospital admissions for myocardial infarction and stroke in all countries were tracked from January 2002 to December 2011, using central government national monitoring data. Admission rates were calculated by age, sex, and county and residence. A difference-in-differences regression was used to compare admission rates in populations with and without TFA restrictions. Restrictions on TFA content in food products in restaurants were only implemented in highly urban counties. Non-restriction counties of similar urbanity were chosen as a comparison population. Temporal trends and county characteristics were accounted for using fixed effects by county and year, as well as linear time trends by county. Results were adjusted for age, sex, and commuting between restriction and non-restriction counties.
Three or more years after the intervention, myocardial infarction and stroke events combined (per 100 000) and stroke events (per 100 000) were significantly lower in the population with TFA restrictions, compared to the population without TFA restrictions. This was equivalent to 42.45% fewer myocardial infarction events (per 100 000) and 32.82% fewer stroke events (per 100 000) compared with the non-restriction population, after adjusting for temporal trends.
Good
Friesen and Mires, 2006
Pre/Post test
To determine whether introduction of labelling of TFA content in retail foods and removal or reduction of TFA content from vegetable oils in many foods was accompanied by a decline in the TFA consumption of human milk
Food labelling and voluntary limits
Canada
103 breast milk samples in 1998; 87 samples in 2004
TFA labelling was introduced in 2003. Samples of breast milk (0-100 mcg) were collected at 1-month postpartum during the course of breastfeeding. Samples in 1998 were compared with samples from 2004 to 2006.
Mean concentration of total TFA in milk in 1998 was 7.1 mg/100 kcal of fat (range 2.2–18.7). Mean values for milk collected and analysed in three consecutive 5-month periods from November 2004 to January 2006 were 6.3 mg/100 kcal of fat, respectively (range 3.4–13.7 g TFA/100 kcal of fat) (range 1.0–14.3 mg/100 kcal TFA/100 kcal). There was no significant decline in TFA intake in women over 40 years of age, and no trend in TFA intake was found in all three consecutive 5-month periods of November 2004 to January 2006, respectively.
Fair
(continued)
| Study | Study design | Study aim | Intervention analysed | Geographical scope | Participants and sample size | Methods | Outcomes | Comments | Quality |
|------------------|----------------|----------------------------------------------------------------------------|----------------------------------------------------------------------------------------|--------------------------|------------------------------------------------------------------------------------------------|------------------------------------------------------------------------|--------------------------------------------------------------------------|-------------------------------------------|----------|
| Ratnayake et al., 2014 | Pre–post test | To assess the impact of labelling and voluntary limits on the concentration of TFAs in human breastmilk samples | Food labelling and voluntary limits | Canada | 153 breastmilk samples in 2009; 309 samples in 2010; 177 samples in 2011 | Mandatory labelling came into force in 2005 with voluntary limits agreed with the food industry in 2007. Samples were collected from breastfeeding mothers in 10 major cities across Canada. TFA content of human milk was estimated using a previously established linear correlation between the percentage of TFAs in the diet and human milk fat, and assuming that 30% of the energy of a lactating mother’s diet is derived from fat. | Mean TFA content of breastmilk were 2.7% (SD: 0.9; range: 1.4–7.2%), 2.2% (SD: 0.7%; range: 1.0–6.8%) and 1.9% (SD: 0.5; range: 0.3–3.4%) of total milk fat for samples collected in 2009 (n = 152), 2010 (n = 309) and 2011 (n = 177), respectively. These values were considerably lower than the value of 7.2% (SD: 3.0; range: 0.1–17.2%) (n = 198) found previously for Canadian human milk in 1992. Estimated TFA intake of Canadian breastfeeding mothers was 0.9% (2 g/day), 0.5% (1.1 g/day), and 0.3% (0.7 g/day) of total energy in 2009, 2010 and 2011, respectively. Levels of TFAs were detectable in all samples. Levels of vaccenic acid decreased by 56% from 43.7 µmol/L in 2000 to 19.4 µmol/L in 2009 (difference of 24.3 µmol/L; 95% CI: 19.6 to 29.0). Similar changes were seen in elaidic acid, palmitelaidic acid and linolelaidic acid. Weighted geometric mean of the difference for the sum of all four TFAs was 54.1 µmol/L (95% CI: 43.4 to 64.7 µmol/L), or 58% lower in samples from 2009 (93.1 µmol/L) than from 2000 (157.1 µmol/L). Breastmilk values were estimated as % of total energy and converted to g/day. | The study was only reported as a research letter. The study measured percentage decreases of TFAs in blood samples rather than g/day. | Fair | |
| Vesper et al., 2012 | Cross-sectional| To determine plasma concentrations of TFAs in a subset of non-Hispanic white adults after labelling TFA content of foods and voluntary limits on TFAs in restaurants were introduced | Food labelling and limits | USA | 229 participants in 2000 and 292 in 2009 | TFA content of foods was to be declared on the nutrition label after 2003. Some community and state departments required restaurants to limit TFA content in food products. Data were from the National Health and Nutrition Examination Surveys in 2000 and 2001. Participants were selected if they had morning fasting blood samples. Four TFAs (elaidic acid, vaccenic acid, linolelaidic acid and palmitelaidic acid) were measured in plasma. | Levels of TFAs were detectable in all samples. Levels of vaccenic acid decreased by 56% from 43.7 µmol/L in 2000 to 19.4 µmol/L in 2009 (difference of 24.3 µmol/L; 95% CI: 19.6 to 29.0). Similar changes were seen in elaidic acid, palmitelaidic acid and linolelaidic acid. Weighted geometric mean of the difference for the sum of all four TFAs was 54.1 µmol/L (95% CI: 43.4 to 64.7 µmol/L), or 58% lower in samples from 2009 (93.1 µmol/L) than from 2000 (157.1 µmol/L). | The study was only reported as a research letter. The study measured percentage decreases of TFAs in blood samples rather than g/day. | Fair | |
CI: confidence interval; DALYs: disability-adjusted life years; EU: European Union; N/A: not applicable; SD: standard deviation; TFA: trans-fatty acid; US$: United States dollars; WHO: World Health Organization.
* We used the National Heart, Lung and Blood Institute quality assessment tools to assess the quality of empirical studies. 17
### Table 2: Modelling studies included in the systematic review of dietary trans-fatty acid reduction policies
| Study | Study aim | Policy analysed | Geographical scope | Participants and sample size | Methods | Outcomes | Comments | Quality* |
|-------|-----------|----------------|--------------------|-----------------------------|---------|----------|----------|---------|
| Allen et al., 2015 | To determine the health and equity benefits and cost-effectiveness of policies to reduce or eliminate TFAs from processed foods, compared with consumption remaining at most recent levels in England | (i) Total ban on TFAs in processed foods; (ii) improved labelling of TFAs; (iii) ban on TFAs in sit-down and takeaway restaurants | England | Adults aged ≥ 25 years (numbers not stated) | For policies aimed at reducing consumption, health benefits and cost outcomes were calculated for 2015–2020 in England only. Government national data and health economic data from other published studies were used for the model. Adults were stratified by fifths of socioeconomic circumstances. | A total ban on TFAs in processed foods: might prevent or postpone about 7200 (2.6%) of the 275 000 total deaths from coronary heart disease by about 5000 of 20400 deaths (24.7%). Policies to improve labelling could save 5500 (1.7%) of 375 000 total deaths from coronary heart disease and reduce inequalities by 1500 (5.1%) of 25 000 deaths, thus making them at best half as effective as a ban. Policies to simply remove TFAs from restaurants or fast foods: could save between 1800 (0.7%) and 2600 (1.0%) of the 273 000 total deaths from coronary heart disease and reduce inequalities by 600 (0.2%) to 1000 (0.5%) of the 20400 deaths. A total ban would have the greatest net cost savings of about $2.25 million, excluding reformulation costs, at $4.6 million of substantial reformulation costs are incurred. | The health outcomes analysis assumed continuing declines in incidence of and mortality from coronary heart disease. The study used an area-based measure of socioeconomic status. Within an area there will be individuals of higher and lower socioeconomic status. Therefore, the study could not make firm conclusions about individuals. As the effect of TFAs operated on a percentage basis (food energy from TFAs divided by total food energy), differences between surveys could only affect the results if consumption were substantially misrepresented in the surveys used. | Good |
| Martin-Saborido et al., 2015 | To assess the added value of EU-level action by estimating the cost-effectiveness of three possible EU-level policy measures to reduce population dietary TFA intake | (i) Status quo; (ii) mandatory TFA labelling of prepackaged food; (iii) seek voluntary agreements with food industry and retailers towards further reducing industrially produced TFA content in food; (iv) impose a legislative limit for industrially produced TFA in foods | EU | EU population (numbers not stated) | A computer-simulated model was developed, using effect sizes from different studies, complemented with results from a survey of EU Member States. The model considered three types of cost: (i) health care costs, (ii) non-health care costs and (iii) costs of policy-associated measures. | The model estimated that imposing an EU-level legal mandatory limit would avoid the loss of 3.73 of 1073 million DALYs due to coronary artery disease over the course of a person’s lifetime (85 years), and making voluntary agreements would avoid 2.19 of 1076 million DALYs. Imposing EU-level legal limits would save an estimated $5.1 billion of $18 733 635 million in total costs when compared with the reference situation and voluntary agreements would save $2.3 billion (1/10 752 032 million). Implementing mandatory TFA labelling would also avoid the loss of 0.9 of 1076 million DALYs, but this option incurred greater costs ($4.95 billion) than it saved compared with the reference option. | Major sources of potential inaccuracy: the estimated current TFA intake; the wide variability observed for many variables between EU countries; and the lack of data in some instances, e.g. lack of data on number of coronary artery disease events per year (coronary artery disease-related hospital discharges were used instead). The results should be interpreted as a comparison between different policy options rather than considering absolute costs, DALYs or deaths per option. | Fair |
| Vyth et al., 2012 | To investigate the potential impact on cholesterol levels of consuming a diet consisting of products that comply with the criteria for a healthier choice logo | Food labelling | Netherlands | Dutch adult population (aged 18–70 years) (n = 4336) | The healthier choices logo for food packages was implemented in 2006. National food consumption and food composition data were used to estimate the nutrient intake of the Dutch adult population before and after replacing foods that did not comply with the choices criteria. | The study was based on the theoretical food replacements not people’s actual practices. | The study assumed that people would not eat the same amounts of replacement foods as their traditional choice, whereas people may eat high amounts of products they perceive to be healthier. The minimum scenario was based on a single study that may not be representative of the general population. The available national representative food consumption data used were based on self-reports, and were outdated. | Poor |
(continues . . )
| Study | Study aim | Policy analysed | Geographical scope | Participants and sample size | Methods | Outcomes | Comments | Quality |
|-------|-----------|----------------|-------------------|----------------------------|---------|----------|----------|---------|
| Roosendaal et al., 2011[^a] | To describe a nutrient intake modelling method to evaluate the choices programme – a nutrition profiling system with nutrition criteria for TFAs, sodium, added sugar and product groups by investigating the potential effect on nutrient intakes | Food labelling | Netherlands | 758 Young Dutch adults (aged 19–30 years) | Data from the 2003 Dutch food consumption survey in young adults and the Dutch food composition tables were combined into a Monte-Carlo risk assessment model. Three scenarios were calculated: (i) actual intakes; (ii) intakes when all foods that did not comply with the healthy choices criteria were replaced by similar foods that did comply; (iii) intakes when food replacements were adjusted for the difference in energy density between the original and replacement food. Another two scenarios were calculated where snacks were not replaced or partially replaced. For the choices labelling programme compared with the actual scenario. TFA intakes in the different scenarios were 2.2 g/day for the actual scenario, 0.8 g/day for the choices labelling programme and 1.0 g/day for the choices labelling programme, adjusted for energy. TFA intakes were 1.3 g/day and 1.4 g/day, when snacks were partially replaced or not replaced, respectively. | An estimated reduction of 62% for TFA intake was found when foods complied with the choices labelling programme compared with the actual scenario. | Replacements chosen may be susceptible to some subjectivity and bias. Product acceptability was not taken into consideration. The same replacement food was used for a large number of snacks. Snacks are usually eaten for indulgence; therefore it is unrealistic to assume that consumers will replace all snacks with the same healthier alternative | Fair |
| De Menezes et al., 2013[^a] | To evaluate the impact of introducing products in agreement with the choices labelling criteria for TFAs, saturated fatty acids, sodium and added sugar in the typical Brazilian diet | Food labelling | Brazil | 1720 food products in the Brazilian diet | Data on industrialized and packaged products available in the market in São Paulo state were collected in 2011. The sources of nutritional information were product labels or websites. To evaluate the impact of the consumption of products aligned with the choices criteria, ingestion of key compounds was estimated based on the choices menu. Typical menus consumed by the Brazilian population were compared with the choices menu (and with the choices menu with energy adjustment). The estimated menus were based on data from a Brazilian household budget survey carried out between 2008 and 2009. | Replacement of typical products by those meeting the choices criteria was estimated to cause a decrease in the intake of TFAs of 92%. Estimated TFA intakes were 0.8 g/day (SD: 0.1) for typical menus, 0.1 g/day (SD: 0.2) for choices menus, and 0.2 g/day (SD: 0.3) for energy-adjusted choices menus, i.e. the same as choices menu, but adjusted for energy of typical menu. | The study compared the typical menu with the choices criteria to see how the intake of dietary components might change. There was no specific focus on TFA | Good |
| Temme et al., 2011[^a] | To estimate the impact of recent reformulations of food groups in the Netherlands on median intake of TFA and saturated fatty acids | Food reformulation | Netherlands | 758 young adults (aged 19–30 years): 352 men, 398 women | Intakes of TFA were estimated before reformulation (started in 2003), using national food composition data of 2001 as a reference and including most recent TFA composition of foods. Food composition of other foods and food consumption was assumed to be unchanged. | Average TFA intake decreased significantly from 2.3 g/day (95% CI: 2.0 to 2.6) to 1.9 g/day (95% CI: 1.8 to 2.0) in the reformulation scenario. Pasta, cakes and biscuits, and snacks contributed most to the decrease of TFA than potatoes, bread, fats and margarines. Median TFA intakes were 2.3 g/day (95% CI: 2.2 to 2.5) in the reference scenario and 1.9 g/day (95% CI: 1.8 to 2.0) in the reformulation scenario. Estimated reduction in TFA intake was 0.4 g/day (95% CI: 0.2 to 0.6) of total energy. | Composition data provided by members of the Dutch task force for the improvement of fatty acid composition was purchasing data, not actual intake data. Therefore it was not always possible to link this information with food consumption data. | Poor |
| Restepo and Rieger, 2016[^a] | To assess whether Denmark's TFA policy reduced deaths caused by cardiovascular disease | Mandatory food reformulation | Denmark | Danish population (number not stated) | A policy restricting the content of artificial TFAs in certain food ingredients was implemented in 2004. Average mortality rates in OECD and development countries from 1990 to 2012 were used to estimate the effect of Denmark's food policy on cardiovascular disease mortality rates. A synthetic control group was composed of a weighted average of other OECD countries that did not implement the policy. Analyses were conducted in 2015. | In the period before the policy (1990–2003), the mean annual number of deaths per 100,000 people in Denmark was 441.5 and in the synthetic control group were 442.7. In the 3 years after the policy was implemented (2004–2006), mortality attributable to cardiovascular disease decreased on average by 14.2 deaths per 100,000 people per year in Denmark relative to the synthetic control group. The policy reduced male and female cardiovascular disease deaths by 24.4 per 100,000 and 14.3 per 100,000 per year over the period 2004–2006, respectively. For coronary heart disease, the estimated reduction over the period 2004–2006 period was 26.5 deaths per 100,000 people per year. | The study investigated what would have happened if mandatory reformulation had not been applied in Denmark. The paper focuses on 2004–2006 before the anti-smoking law was implemented. | Good |
| Study | Study aim | Policy analysed | Geographical scope | Participants and sample size | Methods | Outcomes | Comments | Quality |
|-------|-----------|-----------------|-------------------|----------------------------|---------|----------|----------|---------|
| Barton et al., 2011<sup>41</sup> | To estimate the potential cost-effectiveness of a population-wide risk factor reduction programme aimed at preventing cardiovascular disease | Legislation to ban industrially produced TFA | England and Wales | Entire population aged 40–79 years (about 50 million) | A spreadsheet model was used, with a range of possible interventions to quantify the reduction in cardiovascular disease over a decade, assuming the benefits applied consistently for men and women across age and risk groups | Legislation to reduce intake of industrial TFA by approximately 0.5% from 0.8% to 0.3% of total energy content could prevent approximately 3700 deaths annually and thus gain 570 000 life years and generate savings to the national health service worth at least £2.3 billion a year | The study made no attempt to consider recurrent events or subsequent deaths. The estimates of deaths avoided, life years gained and cost savings were thus likely to be underestimates, making the analysis conservative. The study only modelled a 10-year timeframe; reduction in cardiovascular disease would clearly be greater over a lifetime. The analysis was pragmatically limited to people aged between 40 and 79 years at the time of the intervention. This initial modelling lacked a full probabilistic sensitivity analysis | Good |
| O’Raherty et al., 2012<sup>40</sup> | To estimate how much more cardiovascular disease mortality could be reduced in the United Kingdom through more progressive nutritional targets | (i) Target of 0.5% decrease in the fraction of total energy derived from TFA by 2011; (ii) legislative ban | United Kingdom | Adults aged 25–84 years (number not stated) | Potential reductions in cardiovascular disease mortality in the United Kingdom between 2006 (baseline) and 2015 were estimated by synthesizing data on population, diet and mortality. The effect of specific dietary changes on cardiovascular disease mortality was obtained from recent meta-analyses. The potential reduction in cardiovascular disease deaths was then estimated for two dietary policy scenarios: (i) conservative scenario, with modest improvements (assuming recent trends would continue until 2015); (ii) aggressive scenario, with more substantial, but feasible reductions (already seen in several countries) in saturated fats, industrial TFAs and salt consumption, plus increased fruit and vegetable intake. A probabilistic sensitivity analysis was conducted | In the conservative scenario: reducing the TFA intake by 0.5% in total energy, approximately 3500 of the 12 500 total cardiovascular disease deaths would be prevented. In the aggressive scenario: effectively eliminating the consumption of TFA (to reach 0% of total energy) could result in approximately 4700 of the 29 900 fewer cardiovascular disease deaths (range: 2500–4800) per year | The study did not explicitly model lag times. The study assumed that the effects of food policies on dietary intake in the United Kingdom would be quantitatively similar to those in other countries, without explicitly considering political, commercial, cultural and socioeconomic differences or whether countries’ baseline dietary values were high or low. The study assumed commercial vested interests could be minimized | Good |
| Pearson-Stuttard et al., 2016<sup>42</sup> | To quantify the potential health effects and costs and benefits of the United Kingdom-wide policies to eliminate dietary intake of TFA | (i) Elimination of industrial TFA; (ii) elimination of both industrial and natural TFA | England and Wales | Population stratified by age, sex and socioeconomic status (number not stated) | The study extended a previously validated model to estimate the potential effects on health and economic outcomes of mandatory reformulation or a complete ban on dietary TFA in manufactured products in England and Wales from 2011 to 2038. Two policy scenarios were modelled: (i) elimination of industrial TFA consumption from 0.8% to 0.4% daily energy; (ii) elimination of all TFA consumption from 0.8% to 0% | Elimination of all TFA resulted in the largest gains in mortality and life years, with slightly larger gains when modelling unequal baseline TFA by socioeconomic status. Scenario 1 (elimination of industrial TFA only) Annual deaths prevented: 1700; Annual life-years gained: 15 900; Annual hospital admissions averted: 4400; Hospital admissions averted over 10 years: 38 000. Scenario 2 (elimination of all TFA) Annual deaths prevented: 3300; Annual life-years gained: 29 000; Annual hospital admissions averted: 8400; Hospital admissions averted over 10 years: 72 000 | The model assumed immediate health benefits. However, rapid improvements might reasonably be expected. The study assumed equal mortality gains from elimination of natural and industrial TFAs | Good |
| Study | Study aim | Policy analysed | Geographical scope | Participants and sample size | Methods | Outcomes | Comments | Quality |
|-------------------------------|---------------------------------------------------------------------------|----------------------------------------|--------------------|------------------------------|--------------------------------------------------------------------------|--------------------------------------------------------------------------|--------------------------------------------------------------------------|---------|
| Rubinstein et al., 2015 | To estimate the impact of policies to reduce TFA on coronary heart disease, DALYs and associated health-care costs in Argentina | Reformulation (voluntary and mandatory) and mandatory food labelling | Argentina | Adults aged 34+ years (number not stated) | Baseline intake of TFA before 2004 was estimated to be 1.5% of total energy intake. A policy model was built including baseline intake of TFA, the oils and fats used to replace artificial TFAs, the clinical effect of reducing artificial TFAs and the costs and DALYs saved due to the coronary heart disease events averted. To calculate the percentage reduction of risk, coronary heart disease risks were calculated on a population-based sample before and after implementation of the intervention. The effect of the policies was modelled in three ways, based on (i) projected changes in plasma lipid profiles; (ii) projected changes in lipid and inflammatory biomarkers; and (iii) the results of prospective cohort studies. The current economic value of DALYs and associated health-care costs of coronary heart disease averted were also estimated | Baseline number of deaths: 24 875 for coronary heart disease and 17 942 for acute myocardial infarction. Baseline costs: US$ 6416 per acute coronary syndrome, US$ 5765 per acute myocardial infarction, US$ 1199 per follow-up and treatment, and US$: 1 290 001 for programmatic costs. The proportion of CHD events averted by the modelled TFA reduction policy in 2014 ranged from 1.3% (scenario 1) to 6.4% (scenario 3) of the total. The estimated reductions in coronary heart disease were sensitive to the assumed baseline TFA intake in 2004. Based on projected changes in plasma lipid profiles: an estimated 101 coronary heart disease deaths, 572 acute myocardial infarctions, 1066 acute coronary heart disease events and 5237 DALYs would be annually averted after 2014. This is calculated compared with the expected events if the policy had not been implemented. In addition, more than US$ 17 million would be saved annually due to acute coronary heart disease events averted and lower costs of chronic treatment and follow-up. Based on projected changes in lipid and inflammatory biomarkers: using the baseline estimate of 1.5% energy intake from TFA, a total of 3 109 acute coronary heart disease events, 15 271 DALYs and more than US$ 50 million in costs would be annually averted after 2014. Based on the results of prospective cohort studies (baseline): on estimated 1517 coronary heart disease deaths, 2884 acute myocardial infarctions, 5373 acute coronary heart disease events and 26 394 DALYs would be averted, resulting in estimated savings of US$ 87 million | The cardiovascular risk calculator used was based on equations developed a couple of decades before when the coronary heart disease incidence was higher in Argentina. The study used global percentage estimates to adjust for under-reporting of mortality from coronary heart disease. The study only looked at cost from a health system perspective and not at the cost for the industry to reformulate. The study did not have precise data on baseline TFA and the level of this would influence the results | Fair |
CI: confidence interval; DALYs: disability-adjusted life-years; EU: European Union; OECD: Organisation for Economic Co-operation; £: Pounds sterling; SD: standard deviation; TFA: trans-fatty acid; US$: United States dollars; WHO: World Health Organization.
* We used adapted version of a published quality assessment tool by Fattore et al. [18]. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
6296,
1
],
[
6296,
12318,
2
],
[
12318,
16750,
3
],
[
16750,
22670,
4
],
[
22670,
28512,
5
],
[
28512,
33108,
6
],
[
33108,
37420,
7
],
[
37420,
41417,
8
],
[
41417,
41417,
9
],
[
41417,
44088,
10
],
[
44088,
52531,
11
],
[
52531,
55491,
12
],
[
55491,
60539,
13
],
[
60539,
66067,
14
],
[
66067,
72331,
15
],
[
72331,
77445,
16
],
[
77445,
81893,
17
]
]
} |
95d4f8acd79af20da3240db879df8353e5a229f9 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 4,
"total-fallback-pages": 0,
"total-input-tokens": 10863,
"total-output-tokens": 3380,
"fieldofstudy": [
"Medicine"
]
} | Pituitary Carcinoma with Mandibular Metastasis: A Case Report
Pituitary carcinomas are rare primary adenohypophyseal tumors with cerebrospinal or extracranial metastasis. The present case, the first report of the disease in Korea, involved a 36-yr-old woman who presented with a 3-week history of headache. Brain magnetic resonance imaging revealed a 2.5-cm sellar and suprasellar mass showing heterogeneous enhancement with suspicious invasion of both cavernous sinuses. The patient underwent gross-total resection. The tumor cells were composed of polygonal cells singly or in variable-sized nests. The nuclei were large and round with prominent nucleoli. The cytoplasm was acidophilic and granular. Marked pleomorphism and frequent mitoses (3 per 10 HPFs) were found. By immunohistochemistry, tumor cells were strongly positive for prolactin, but negative for ACTH and GH. Additional immunostainings for cytokeratin, vimentin, and glial fibrillary acidic protein (GFAP) were negative. After the surgery, the patient received radiotherapy because of the atypical histologic features. The prolactin level fell from 123.17 ng/mL to 5.17 ng/mL after surgery. Nine months after the initial diagnosis, the patient died from mandibular metastasis associated with the pituitary carcinoma.
Key Words: Pituitary Neoplasms; Metastasis; Pituitary Carcinoma
INTRODUCTION
Primary pituitary carcinomas are very rare neoplasms defined as adenohypophyseal tumors with cerebrospinal and/or systemic metastasis. They represent approximately 0.2% of all operated adenohypophyseal neoplasms (1). The disease was first reported in 1931 by Cairns et al. (2) in a 25-yr-old woman with pituitary adenoma and intradural spinal cord metastasis. There have been more than 100 cases of pituitary carcinoma reported, including functional and nonfunctional tumors (3), but none in the Korean literature to the best of our knowledge. The histologic features of pituitary carcinomas are vary from typical adenoma to tumor with marked pleomorphism and frequent mitoses (4, 5). Here we report a case of pituitary carcinoma with left mandibular metastasis in a 36-yr-old woman.
CASE REPORT
A 36-yr-old woman with a 3-week history of headache was presented at the outpatient otorhinolaryngology clinic for the evaluation of suspicious sinusitis in November 2005. Symptoms including nausea, vomiting, myalgia, and general weakness worsened despite ongoing antibiotic therapy. No lesion was found in the initial brain computed tomography (CT) scan. Twenty days later she reported right-side ptosis and amenorrhea. Magnetic resonance imaging revealed a 2.5-cm sellar and suprasellar mass with heterogeneous enhancement, and suspicious invasion to both cavernous sinuses and the left maxillary sinusitis (Fig. 1). The mass compressed the optic chiasm. Her serum prolactin level was 123.17 ng/mL before surgery. She underwent gross-total resection based on the initial diagnosis of pituitary macroadenoma with 3rd and 6th nerve palsies. Histologically, the resected tumor was composed of variable-sized nests and single cells mixed with adenohypophyseal cells, which appeared normal (Fig. 2). The tumor cells had polygonal vesicular nuclei with marked pleomorphism and prominent nucleoli. The cytoplasm was acidophilic and granular (Fig. 3). When presented, mitoses counted 3 per 10 high power fields (HPFs). Immunohistochemistry staining was strongly positive for prolactin (PRL) (Fig. 4), but negative for adenocorticotropic hormone (ACTH) and growth hormone (GH). Immunostaining was also positive for synaptophysin and p53 (Fig. 5A), but not for cytokeratin, vimentin or glial fibrillary acidic protein (GFAP). The Ki-67 labeling index (Fig. 5B) was high. Postoperative CT scanning revealed a residual nodular enhancement at the superior portion of the resected mass. Radiotherapy (total dose 5,040 Gy) was undertaken after surgery, based on the atypical morphology of the pituitary tumor and the presence of the residual enhanced mass. The prolactin level fell to 5.17 ng/mL after surgery and adjuvant radiotherapy.
The patient was readmitted 6 months later with a one-
month history of left cheek and periauricular pain. The patient had problems opening her mouth, and painful swelling and local heating. Radiologic assessment revealed a soft tissue mass with cortical destruction and a permeative osteolytic lesion at the left mandibular ramus (Fig. 6). A partial mandibulectomy was performed. Grossly, the mandible and per-
Fig. 1. Sagittal contrast enhanced T1-weighted image showing a lobulating contoured mass (arrow) with heterogeneous enhancement in the sella. Invasion of the posterior wall of the sphenoid sinus by the sella mass is evident.
Fig. 2. Resected tumor cells showing variable-sized nests and single cells (left lower) mixed with adenohypophyseal cells of normal appearance (right upper) (H&E stain, × 100).
Fig. 3. Tumor cells showing atypical morphology with large round nuclei and prominent nucleoli. The mitoses are 3/10 high power fields (H&E, × 400).
Fig. 4. Tumor cells showing strong and diffuse positive staining for prolactin (Prolactin, × 400).
Fig. 1. Sagittal contrast enhanced T1-weighted image showing a lobulating contoured mass (arrow) with heterogeneous enhancement in the sella. Invasion of the posterior wall of the sphenoid sinus by the sella mass is evident.
Fig. 2. Resected tumor cells showing variable-sized nests and single cells (left lower) mixed with adenohypophyseal cells of normal appearance (right upper) (H&E stain, × 100).
Fig. 3. Tumor cells showing atypical morphology with large round nuclei and prominent nucleoli. The mitoses are 3/10 high power fields (H&E, × 400).
Fig. 4. Tumor cells showing strong and diffuse positive staining for prolactin (Prolactin, × 400).
imandibular soft tissue were involved by the tumor. The histologic features (Fig. 7) and the immunohistochemical results were identical to those observed in the previous pituitary tumor. Primary pituitary carcinoma was confirmed from the histopathologic findings and a metastasis. She died in August 2006, approximately 9 months after the initial diagnosis.
Pituitary Carcinoma with Mandibular Metastasis
DISCUSSION
Pituitary carcinomas are rare adenohypophyal neoplasms and their definition, diagnosis, therapy, and prognosis are controversial. Unlike other malignant neoplasms, the diagnosis of pituitary carcinoma is based on metastasis and not on histologic features such as invasion, cellular pleomorphism, nuclear abnormalities, mitotic activity, and necrosis. More than 100 cases of pituitary carcinoma have been reported worldwide (3), but this is the first report in Korea.
In a review of the literature on pituitary carcinoma, Periccone et al. (1) reported the endocrinological subtypes include PRL-producing tumors (30%), ACTH-producing tumors (28%), GH-producing tumors (10%), thyroid-stimulating hormone-producing tumors (2%), gonadotropin-producing tumors (2%), and nonfunctioning tumors (26%). Malignant prolactinomas, such as the case described here, were the most common (26 cases) subtype of pituitary carcinomas reported (6). They predominantly occur in males (M:F=17:9), in middle age (range: 14-70 yr, mean: 43.6 yr), and have a poor prognosis. Most of the locations of metastasis are extra/intracranial and spinal sites (50%), but can involve other sites including bones and soft tissues (occiput, ribs, vertebra, sacroiliac joint, femur, and mandible), and parenchymal organs (lung, lymph node, liver, ovary, endometrium, and adrenal gland). In addition, nearly all pituitary carcinomas are identified as functioning, mitotically active, and invasive macroadenomas with a tendency for multiple metastasis.
Although the primary pituitary carcinomas have various histologic features, from monotonous proliferation of cells with uniform nuclei to cellular pleomorphism and prominent nucleoli, the cytologic atypia does not predict the tumor behavior. However, there is a slight trend towards a higher degree of cytologic atypia and increased mitotic activity in metastasis lesions compared with primary tumors (1). Our case had atypical morphology with frequent mitoses in both
primary and metastasis tumors. The Ki-67 index was high in the primary and metastasis tumors (10% and 30%, respectively). In the study by Thapar et al. (7), the relationship between proliferative activity in pituitary adenomas and their invasiveness was investigated using the MIB-1 monoclonal antibody, which detects the Ki67 cell cyclonuclear antigen. The growth fractions of noninvasive adenoma, invasive adenoma, and primary pituitary carcinomas were 1.37 ± 0.15%, 4.66 ± 0.47%, and 11.91 ± 3.41%, respectively. In comparison with the above study, the Ki-67 labeling index in the primary and metastasis lesions of the present case exceeded the level reported for pituitary carcinomas. In addition, the p53 immunostaining reported for the tumor suppressor gene was 0%, 15.2%, and 100% for the noninvasive adenoma, invasive adenoma, and carcinoma, respectively (7). Our case was p53-positive in both the primary and metastasis pituitary carcinoma.
Histological changes in pituitary carcinoma after radiation therapy were reported by Yamashita et al. (8). After irradiation approximately three-fifths of tumor cells were unaltered and two-fifths showed degeneration. It was suggested the tumor necrosis after irradiation was not an apoptotic process but an irradiation effect because TUNEL staining of the nuclei of postirradiated cells was negative. However, nonirradiated pituitary carcinoma can show extensive necrosis (up to 50%), as in this case, whether or not apoptosis is involved.
The prognosis of pituitary carcinoma is known to be poor because of the extent of tumor dissemination at the time of diagnosis, with the mean survival time reported to be 2 yr (range 0.25-8 yr) (1). For malignant prolactinomas the latent period between initial diagnosis and detection of metastases ranged from 2 to 228 months, and most patients (80%) died within 18 months of metastasis development. In our case, the latent period was 6 months, and the patient died 3 months after metastasis developed.
In conclusion, pituitary carcinoma is a rare tumor involving cranio-spinal or systemic metastasis. Irrespective of the histological morphology, the behavior of primary pituitary neoplasms cannot be predicted. In view of the poor prognosis, atypical morphology including local invasion, increased mitosis, and a high proliferation index (as in this case), close follow-up is required for proper management.
REFERENCES
1. Pernicone PJ, Scheithauer BW, Sebo TJ, Kovacs KT, Horvath E, Young WF Jr, Lloyd RV, Davis DH, Guthrie BL, Schoene WC. Pituitary carcinoma: a clinicopathologic study of 15 cases. Cancer 1997; 79: 804-12.
2. Cairns H, Russel DS. Intracranial and spinal metastasis in gliomas of the brain. Brain 1931; 54: 377-420.
3. Scheithauer BW, Kovacs KT, Horvath E, Roncaroli F, Ezzat S, Asa SL, Lloyd RV, Nose V, Watwon RE Jr, Lindell EP. Pituitary carcinoma. In: DeLellis RA, Lloyd RV, Hritz PU, Eng C, eds. Pathology & Genetics Tumours of Endocrine Organs. Lyon, France; International Agency for Research on Cancer (IARC) Press, 2004; 36-9.
4. Long MA, Colquhoun IR. Case report: multiple intra-cranial metas-
tases from a prolactin-secreting pituitary tumour. Clin Radiol 1994; 49: 356-8.
5. Bayindir C, Balak N, Gazioglu N. Prolactin-secreting carcinoma of the pituitary: clinicopathological and immunohistochemical study of a case with intracranial and intraspinal dissemination. Br J Neurosurg 1997; 11: 350-5.
6. Lamas C, Nunez R, Garcia-Uria J, Salas C, Saucedo G, Estrada J, Parajon A, Lucas T. Malignant prolactinoma with multiple bone and pulmonary metastases. Case report. J Neurosurg 2004; 101 (Suppl 1): 116-21.
7. Thapar K, Kovacs K, Scheithauer BW, Stefaneanu L, Horvath E, Pernicone PJ, Murray D, Laws ER Jr. Proliferative activity and invasiveness among pituitary adenomas and carcinomas: an analysis using the MIB-1 antibody. Neurosurgery 1996; 38: 99-106.
8. Yamashita H, Nakagawa K, Tago M, Nakamura N, Shiraishi K, Yamauchi N, Ohtomo K. Pathological changes after radiotherapy for primary pituitary carcinoma: a case report. J Neurooncol 2005; 75: 209-14. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4142,
1
],
[
4142,
6166,
2
],
[
6166,
8210,
3
],
[
8210,
12295,
4
]
]
} |
10e78c2dc147092b5da5fd1cabeb05c96b35f21a | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 30,
"total-fallback-pages": 0,
"total-input-tokens": 56470,
"total-output-tokens": 14618,
"fieldofstudy": [
"Biology",
"Psychology"
]
} | Predictive entrainment of natural speech through two fronto-motor top-down channels
Hyojin Park1,3, Gregor Thut3, Joachim Gross2,3
1School of Psychology & Centre for Human Brain Health (CHBH), University of Birmingham, Birmingham, United Kingdom
2Institute for Biomagnetism and Biosignalanalysis, University of Münster, Münster, Germany
3Institute of Neuroscience and Psychology, University of Glasgow, Glasgow, United Kingdom
Author ORCIDs
Hyojin Park: http://orcid.org/0000-0002-7527-8280
Gregor Thut: https://orcid.org/0000-0003-1313-4262
Joachim Gross: http://orcid.org/0000-0002-3994-1006
Twitter
Hyojin Park: @HyojinParkNeuro, Joachim Gross: @Joachim__Gross
Corresponding authors
Hyojin Park and Joachim Gross
-School of Psychology, University of Birmingham, Edgbaston, Birmingham, B15 2TT, United Kingdom
-Institute for Biomagnetism and Biosignalanalysis, University of Münster, Malmedyweg 15, D-48149, Münster, Germany
Email: [email protected] or [email protected]
Running Head
Predictions in speech perception
Abstract
Natural communication between interlocutors is enabled by the ability to predict upcoming speech in a given context. Previously we showed that these predictions rely on a fronto-motor top-down control of low-frequency oscillations in auditory-temporal brain areas that track intelligible speech. However, a comprehensive spatio-temporal characterisation of this effect is still missing. Here, we applied transfer entropy to source-localised MEG data during continuous speech perception. First, at low frequencies (1-4 Hz, brain delta phase to speech delta phase), predictive effects start in left fronto-motor regions and progress to right temporal regions. Second, at higher frequencies (14-18 Hz, brain beta power to speech delta phase), predictive patterns show a transition from left inferior frontal gyrus via left precentral gyrus to left primary auditory areas. Our results suggest a progression of prediction processes from higher-order to early sensory areas in at least two different frequency channels.
Keywords: Speech entrainment, Top-down prediction, Beta, Delta
Introduction
Natural communication between interlocutors may seem effortless, however it relies on a series of complex computational tasks that have to be performed in the human brain in real-time and often in the presence of noise and other interferences. This high performance can only be achieved with the help of very effective prediction mechanisms (Levinson, 2016; Norris, McQueen, & Cutler, 2016). As auditory speech signals enter the sensory auditory system and are complemented by visual signals and cues, the human brain generates and constantly updates predictions about the timing and content of upcoming speech (Friston and Frith, 2015; Pickering and Garrod, 2007, 2013). In this context of natural conversation, we can model human brains as dynamic systems that are coupled through sensory information and operate according to active inference principles (Friston and Frith, 2015). In this framework, the brain relies on internal models to generate predictions about itself and others and updates the internal model to minimize prediction errors.
The temporal structure of this predictive coding mechanism can be mediated by cortical oscillations and previous studies have shown the computational role of cortical oscillations in speech processing as critical elements for parsing and segmentation of connected speech not only for auditory speech (Arnal and Giraud, 2012; Ding, Melloni, Zhang, Tian, & Poeppel, 2016; Giraud and Poeppel, 2012) but also for visual speech (Giordano et al., 2017; Park, Kayser, Thut, & Gross, 2016; Zion Golumbic, Cogan, Schroeder, & Poeppel, 2013). In addition, cortical oscillations track hierarchical components of speech rhythms and cortical oscillations themselves are hierarchically nested for speech tracking (Gross, Hoogenboom, et al., 2013). Importantly, these findings are evident only for intelligible speech processing where top-down modulation by prediction is possible. In our previous study, we found that low-frequency rhythms in the left frontal and motor cortices carry top-down signals to sensory areas, particularly to left auditory cortex, and this top-down signal was correlated with entrainment to speech (Park, Ince, Schyns, Thut, & Gross, 2015).
While previous studies have provided convincing evidence that low-frequency brain rhythms are involved in mediating top-down predictions, several important questions are still unresolved. First, what is the spatio-temporal structure of these prediction processes, or put differently, when and where in the brain is neural activity predictive of upcoming speech in an intelligibility-dependant manner? Second, what is the relationship between these low-frequency rhythms and higher frequency rhythms
that have been implicated in prediction (Morillon and Baillet, 2017)? Third, how do predictive processes (preceding speech) interact with reactive processes (following speech)?
In order to address these questions, we used causal connectivity analysis - transfer entropy (TE) - to identify directed coupling between brain rhythms and speech rhythms for a range of positive delays (brain activity following speech) and negative delays (brain activity preceding speech). For speech signal, we analysed low-frequency phase information which is a dominant spectral component in speech (Chandrasekaran, Trubanova, Stillittano, Caplier, & Ghazanfar, 2009). For brain signal, we analysed both low-frequency phase information as well as high frequency beta power. We hypothesized that beta rhythm in the brain, particularly in higher order areas, are involved in the prediction of forthcoming speech as suggested by the role of beta oscillations in top-down predictive mechanism (Arnal and Giraud, 2012; Bastos et al., 2012; Fontolan, Morillon, Liegeois-Chauvel, & Giraud, 2014). A recent EEG study that used time-compressed speech reported beta oscillations reflecting an endogenous top-down channel which gradually builds up contextual information across time (Pefkou, Arnal, Fontolan, & Giraud, 2017). Particularly, beta oscillations in motor system were shown to be associated with precise temporal anticipation of forthcoming auditory inputs (Morillon and Baillet, 2017). We also hypothesized that this top-down ‘predictive speech coding’ mechanism by beta oscillations (which should be represented at negative delays between brain activity and speech) recurrently interacts with low-frequency ‘speech entrainment’ (represented at positive delays) where better prediction leads to stronger entrainment.
Materials and Methods
Participants and Experiment
Twenty-two volunteers participated in the study (11 females; age range 19–44 years, mean age ± SD: 27.2 ± 8.0 years). None of the participants had a history of psychological, neurological, or developmental disorders. They all had normal or corrected-to-normal vision and were right-handed. Written informed consent was obtained from all participants prior to the experiment and all participants received monetary compensation for their participation. The study was approved by the local ethics committee (FIMS00733; University of Glasgow, Faculty of Information and Mathematical Sciences) and conducted in accordance with the Declaration of Helsinki.
Participants were instructed to listen to a recording of a 7-min-long story (“Pie-man,” told by Jim O’Grady at “The Moth” storytelling event, New York). The stimulus was presented binaurally via a sound pressure transducer through two 5-meter-long plastic tubes terminating in plastic insert earpieces. Stimulus presentation was controlled via Psychtoolbox (Brainard, 1997) in MATLAB (MathWorks, Natick, MA). The experiment consisted of two conditions: standard presentation of story (intelligible speech) and backward played presentation of story (unintelligible speech). Experimental conditions were presented in randomised order across participants.
Data acquisition, Preprocessing and Source localisation
Data recordings were acquired with a 248-magnetometers whole-head MEG system (MAGNES 3600 WH, 4-D Neuroimaging) in a magnetically shielded room. Data were sampled at 1017 Hz and resampled at 250 Hz, denoised with information from the reference sensors, and detrended. The analysis was performed using the FieldTrip toolbox (Oostenveld, Fries, Maris, & Schoffelen, 2011) (http://fieldtrip.fcdonders.nl) and in-house MATLAB scripts according to the guidelines (Gross, Baillet, et al., 2013).
Structural T1-weighted magnetic resonance images (MRI) of each participant were obtained at 3T Siemens Trio Tim Scanner (Siemens, Erlangen, Germany) and co-registered to the MEG coordinate.
system using a semi-automatic procedure. Anatomical landmarks (nasion, left and right pre-auricular points) were manually identified in the individual's MRI. Both coordinate systems were initially aligned based on these three points. Numerical optimisation by using the iterative closest point (ICP) algorithm (Besl and McKay, 1992) was applied.
Individual MRIs were segmented to grey matter, white matter, and cerebrospinal fluid to create individual head models. Leadfield computation was based on a single shell volume conductor model (Nolte, 2003) using a 10-mm grid defined on the standard template brain (Montreal Neurological Institute; MNI). The template grid was transformed into individual head space by linear spatial transformation. Cross-spectral density matrices were computed using Fast Fourier Transform on 1-s segments of data after applying Hanning window.
Frequency-specific spatial filters were computed for delta (1-4 Hz) and beta (14-18 Hz) bands at each voxel. We computed the covariance matrix over the full broad-band 7-minute data to compute LCMV filters for each voxel using 7% regularization. These time series were then subjected to band-pass filtering (4th order Butterworth filter, forward and reverse). Dominant dipole orientation was estimated using SVD at each voxel. Bandpass filtered data were projected through the filter to obtain band-limited time-series for each voxel. This computation was performed for each frequency band, and each experimental condition (intelligible and unintelligible). Hilbert transformation was applied to obtain instantaneous phase and power. We also used regions of interest (ROI) maps from the AAL (Automated Anatomical Labeling) atlas (Tzourio-Mazoyer et al., 2002) in order to delineate the temporal characteristics of directed causal relationship over the delays (see below) within the anatomically parcellated regions. We used ROIs labelled Heschl gyrus, inferior frontal gyrus – opercular part, and precentral gyrus.
**Directed causal connectivity analysis by Transfer Entropy (TE)**
In this paper, we aimed to investigate two relationships of speech entrainment and predictive speech coding between speech and brain signal. In order to assess the relationships, we used transfer entropy (TE) that quantifies directed statistical dependencies between two signals, i.e., time-lagged predictability. TE is also known as Directed Information (Ince, Schultz, & Panzeri, 2014; Massey,
For speech entrainment, we computed TE from speech signal to signal at each brain voxel to quantify to what extent knowledge of speech signal reduces uncertainty in predicting the future of brain signal over and above what could be predicted from knowledge of the past of brain signal alone. For predictive speech coding, we computed TE from brain signal at each voxel to signal to quantify to what extent knowledge of brain signal reduces uncertainty in predicting the future of speech signal over and above what could be predicted from knowledge of the past of speech signal alone.
Specifically, we quantized the phase values from the two signals across all time points during stimulus presentation, and then used 4 bins in which each bin was equally occupied. For a specific delay \( d \), we computed TE from speech \((X)\) to brain \((Y)\) for speech entrainment, and from brain \((X)\) to speech \((Y)\) for predictive speech coding as follows:
\[
TE_d(X \rightarrow Y) = CMI(X_d; Y | Y_d) \\
= H(X_d, Y_d) + H(Y, Y_d) - H(X_d, Y, Y_d) - H(Y_d)
\]
Where CMI is conditional mutual information, \( H \) represents entropy. The suffix \( d \) represents that signal is delayed with respect to the target signal \( Y \) by \( d \) milliseconds (i.e. considers that signal \( d \) milliseconds prior to \( Y \)). We computed entropy terms from the standard formula:
\[
H(Y, Y_d) = \sum_{a,b=1}^{4} P_{Y,Y_d}(a,b) \log_2 P_{Y,Y_d}(a,b)
\]
Where the joint distribution \( P_{Y,Y_d}(a,b) \) is obtained from the multinomial maximum likelihood estimate obtained over time points:
\[
P_{Y,Y_d}(a,b) = \frac{\delta_a(Y(t)) \delta_b(Y(t-d))}{N_t}
\]
With \( \delta_a(Y(t)) \) a Kronecker delta function taking the value 1 if the binned phase value at \( Y(t) \) is quantile \( a \) and 0 otherwise.
In the TE computation bias correction was not applied. Bias of mutual information depends on the number of bins and time points that are used in the analysis (Panzeri, Senatore, Montemurro, &
Petersen, 2007). In our analysis, we performed statistical contrast between conditions in which the same number of bins and time points was used for each calculation (Ince, Mazzoni, Bartels, Logothetis, & Panzeri, 2012). Bias correction reduces bias but increases the variance of the estimator, so comparisons between calculations with the same bias are better with uncorrected estimates.
The TE calculation was repeated for 25 different delays, from 20 ms to 500 ms with a 20-ms step. These computations were performed for each participant, and both conditions (intelligible, unintelligible). For TE computation to study speech entrainment (TE from speech to brain), we analysed the same frequency (delta; 1-4 Hz) phase information for both speech and brain signals. For TE computation to study predictive speech coding (TE from brain to speech), we analysed 1) the same delta (1-4 Hz) phase information for both brain and speech signals as well as 2) beta (14-18 Hz) power for brain signal and delta (1-4 Hz) phase for speech signal. These computations resulted in TE values for each voxel, each delay, and each condition within each participant and then yielded to statistical comparison between conditions for each delay. For ROI analysis using AAL atlas map, TE values were first averaged across the voxels within the ROI, and then yielded to statistics.
Group statistics was performed using non-parametric randomisation in FieldTrip (Monte Carlo randomisation). Specifically, individual volumetric maps were smoothed with a 10-mm Gaussian kernel and subjected to dependent-samples t-test between conditions (intelligible versus unintelligible). The null distribution was estimated using 500 randomisations and multiple comparison correction was performed using FDR.
Results
To investigate the bidirectional nature of speech-brain coupling during listening to continuous speech, we computed transfer entropy (TE) – an information-theoretic measure of directed causal connectivity between speech and brain signals at positive and negative delays. This allows a disambiguation of two different effects that contribute to speech-brain coupling – namely those that follow speech (entrainment) and those that precede speech (prediction).
Since speech-brain coupling is strongest for the low frequency delta rhythm (1-4 Hz) we performed our analysis for this frequency band. In the following, we first present results for positive delays where delta-band brain activity follows speech. Second, we show how delta-band brain activity at different negative delays (i.e. preceding speech) in different brain areas predicts (in a statistical sense) the speech signal. Finally, since beta oscillations in motor cortex have been implicated in temporal predictions (e.g. Arnal and Giraud, 2012; Morillon and Baillet, 2017) we investigate how these higher frequency oscillations in the brain are related to the low-frequency delta rhythm in speech.
Entrained brain signals following speech (positive delays)
We first examine how low-frequency speech signal entrains the same frequency brain rhythms with positive delays (brain signals following speech signals, Figure 2A). We computed TE from speech envelope to brain signals from different ROIs with delays ranging from 20 ms to 500 ms in steps of 20 ms. Based on our recent study (Park, et al., 2015), we focused on three ROIs from the AAL atlas; primary auditory cortex, inferior frontal gyrus (IFG) and precentral gyrus. We computed TE between intelligible (forward played) and unintelligible (backward played) speech conditions (Figure 2B-D). Whole brain results of the same analysis are shown in Supplementary Figure 1.
Figure 2B shows the statistical difference between intelligible and unintelligible speech across different delays for left (pink) and right (green) auditory cortex. Highest t-values are observed at delays of about 80 ms in right auditory cortex and are stronger than corresponding effects in left auditory cortex. The effect of intelligibility remains significant for delays up to about 300 ms for right auditory cortex and for delays up to 500 ms for left auditory cortex. In IFG, the effect of intelligibility is also most pronounced in the right hemisphere (up to about 450 ms) compared to only early transient
effect for the left hemisphere (Figure 2C). Precentral areas in left and right hemisphere show similar sensitivity to intelligibility but are longer lasting in the right hemisphere (Figure 2D).
**Entrained brain signals predicting upcoming speech (negative delays)**
Next, we aimed to characterise the spatio-temporal pattern of predictive processes in the brain in the same low-frequency band (1-4 Hz, Figure 3A). Specifically, we computed TE between speech and brain signals over a range of negative delays (-500 ms to -20 ms in steps of 20 ms) to assess where and when brain signals predict significantly stronger upcoming intelligible compared to unintelligible speech. We performed the computation in the whole brain and show maps of statistical difference between intelligible and unintelligible speech condition. Overall, we found the strongest directed effect in left fronto-motor regions ~-220 ms prior to the forthcoming speech when speech is intelligible compared to unintelligible (Figure 3B; paired t-test; upper red line: \( t_{21} = 3.53, p < 0.05 \), corrected; bottom red line: \( t_{21} = 2.08, p < 0.05 \), uncorrected).
To further characterise the spatio-temporal evolution of brain signals that predict upcoming speech, we averaged TE-maps in 100 ms windows and computed again the statistical contrast of intelligible compared to unintelligible speech (Figure 3C). This revealed a sequence of events that starts around -300 ms in fronto-motor areas (see also Figure 3B) and then moves to right auditory-temporal areas at around -200 ms prior to speech (Figure 3C; \( p < 0.05 \), FDR-corrected). These results indicate that brain activity in fronto-motor areas preceding speech by 300 ms contain information that predicts the forthcoming speech significantly better for intelligible compared to unintelligible speech. The same holds true for right auditory-temporal areas at shorter delays of 200 ms preceding speech. This suggests that predictive speech mechanisms occur first in left inferior fronto-motor areas and later in right auditory-temporal areas.
**Beta rhythms and the prediction of upcoming speech (negative delays)**
As we hypothesized based on the literature that beta rhythms are involved in predictive process during speech processing, we examined the causal relationship between beta rhythm in the brain and low-frequency rhythm in the speech signal. We computed TE from beta (14-18 Hz) power at each
voxel to the low-frequency delta (1-4 Hz) phase of the speech signal for a range of delays (-500 ms to -20 ms in steps of 20 ms, Figure 4A). The same computation was performed for the intelligible and unintelligible conditions and then statistically compared. We first focus on the temporal dynamics of the three ROIs (primary auditory cortex (Heschl gyrus), inferior frontal gyrus, precentral gyrus).
Figure 4B shows over a range of delays for each ROI to what extent the prediction of forthcoming speech depends on speech intelligibility.
Beta power in left inferior frontal gyrus shows strong intelligibility-dependant prediction about 340-500 ms prior to speech (Figure 4B orange line). Left primary auditory cortex and left precentral gyrus show a similar significant effect but at a shorter delay peaking just before -200 ms (Figure 4B, blue and yellow lines). For the precentral gyrus, the ROI map from the AAL atlas is rather big, so we extracted the time series for the voxel with the local maximum in this area (yellow line, but also see the similar pattern for whole precentral gyrus ROI in the AAL atlas; light green line). Interestingly, this intelligibility-dependant prediction in these ROIs was left-lateralized (see Figure 4C and Supplementary Figure 2). Whole brain results averaged across 100 ms-long windows corroborated this pattern that first engages left inferior frontal gyrus about 300-500 ms prior to the speech followed by left precentral gyrus and left primary auditory area 200-300 ms prior to the speech (Figure 4C; p < 0.05, FDR-corrected).
**Relationship between temporal dynamics of speech entrainment and predictive speech coding**
We next assessed how intelligibility-dependant prediction and the temporal dynamics of speech entrainment are related. We hypothesized that predictive control mechanisms interact with the rhythms in the brain driven by speech. We performed correlation analysis between TE values of intelligible speech at negative delays (preceding speech) and positive delays (following speech) using robust Spearman rank correlations across participants (Pernet, Wilcox, & Rousselet, 2012).
Motivated by the role of top-down beta activity particularly in the perception of sustained temporal aspect of speech (Pefkou, et al., 2017), we correlated predictive coding by beta activity (Figure 4) and speech entrainment (Figure 2) at various delays. We studied this mechanism first within early sensory area (auditory cortex) as well as higher order areas where we found strong top-down prediction.
Figure 5A shows this relationship in the left primary auditory cortex where predictive speech coding mechanism of beta power in the left auditory cortex ~250 ms prior to the forthcoming speech with low frequency delta phase information (corresponding to the plot for the left auditory cortex (blue line) at ~250 ms in the Figure 4B) is closely associated with low-frequency delta rhythm in the left auditory cortex driven by speech (corresponding to the plot for the left auditory cortex (pink line) at ~250 ms in the Figure 2B) \( r = 0.46, p = 0.02 \). This suggests that individuals with stronger predictive coding by beta power in the left auditory cortex ~250 ms prior to the upcoming low frequency delta phase information in the speech signal are capable of stronger speech entrainment in the left auditory cortex by speech rhythm with low frequency delta phase at ~250 ms.
Another interesting aspect of this relationship emerged between the left inferior frontal cortex and left precentral gyrus. Figure 5B shows that predictive speech coding of beta power in the left inferior frontal gyrus ~400 ms prior to the forthcoming speech with low frequency delta phase information (corresponding to the plot for left inferior frontal gyrus (orange line) at ~400 ms in the Figure 4B) is associated with low frequency delta phase in the left precentral gyrus modulated by the same frequency phase in the speech at ~200 ms (corresponding to the plot for the left precentral gyrus (pink line) at ~200 ms in the Figure 2D) \( r = 0.53, p = 0.01 \). This suggests that individuals with stronger modulation of predictive coding by beta power in the left inferior frontal gyrus ~400 ms prior to the low frequency delta phase information in the speech signal are capable of stronger speech entrainment in the left precentral gyrus by speech rhythm with low frequency delta phase at ~200 ms.
Discussion
Here, we aimed to study the spatio-temporal characteristics of predictions during continuous speech recognition. We analysed transfer entropy (TE) at various (positive and negative) delays between the speech envelope and brain activity. At low frequencies (1-4 Hz) our results reveal a progression of predictive effects from left fronto-motor regions to right temporal regions. A different pattern emerged when we investigated TE between beta power in the brain and the phase of 1-4 Hz components in the speech envelope. We first see an engagement of left inferior frontal gyrus about 300-500 ms prior to speech followed by left precentral gyrus and left primary auditory areas. Our results suggest a progression of prediction processes from higher-order to early sensory areas.
First, it is important to carefully consider what aspects of predictions are captured using our approach. Transfer entropy (TE) is an information theoretic measure that quantifies directed statistical dependencies between time series. Specifically, TE from signal X to signal Y quantifies to what extent knowledge of X reduces uncertainty in predicting the future of Y over and above what could be predicted from knowledge of the past of Y alone. TE is conceptually similar to Granger causality as it infers causal relationships from time-lagged predictability. Here, when analysing prediction effects, we quantify to what extent the past of brain activity in a certain brain area improves prediction of the future of the speech envelope (over and above what could be predicted from the past of the speech envelope alone). Our main conclusions are then based on the statistical contrast between intelligible (forward played) speech and unintelligible (backward played) speech. This is important for two reasons. First, the power spectrum of the speech envelope is the same for both conditions. Therefore, this statistical contrast controls (to some extent) for the low-level rhythmicity in speech and amplifies sensitivity of the analysis to intelligibility. However, we acknowledge that backward played speech is not a perfect control condition due to differences in the finer temporal structure and attention to the stimulus. Second, similar to Granger causality, the computation of TE on bandpass-filtered data is not without problems (Florin, Gross, Pfeifer, Fink, & Timmermann, 2010; Weber, Florin, von Papen, & Timmermann, 2017). Statistically contrasting two conditions will counteract these problems. In addition, we would like to note that our results are very similar when using delayed mutual information (data not shown), a measure that is less sensitive to the effects of filtering. In summary, our approach is expected to be mostly sensitive to intelligibility-related components in speech.
Still, from our study it is difficult to exactly specify the structure in speech that is the target of the prediction processes presented here. This is in contrast to the many studies that demonstrate that a semantic violation at a certain point in a sentence gives rise to the well-known N400 evoked response (e.g. Kutas and Federmeier (2011)). While being less controlled, our approach benefits from ecological validity and more directly taps into prediction processes that operate during natural speech processing.
Brain oscillations in this frequency range follow the same frequency in the speech envelope robustly across all delays between 20-500 ms. The pattern is stronger in the right hemisphere (higher t-values in the Supplementary Figure 1). In our previous study we showed more right-lateralised speech-brain coupling at a fixed delay using mutual information (Figure 2A in Gross, Hoogenboom, et al. (2013)). This mechanism seems to extend to the directed TE measure used here up to ~400 ms delays. This supports the asymmetric sampling in time (AST) model (Poeppel, 2003) that posits a right hemisphere preference for longer temporal integration window (~150-250 ms). The sustained pattern suggests brain responses modulated by speech envelope are critical to continuous intelligible speech perception.
Directed coupling between speech envelope and brain oscillations across negative delays suggests a predictive coding mechanism where sensory processing is modulated in top-down manner. We investigated this mechanism from low-frequency delta (1-4 Hz) phase in the brain to low-frequency delta phase in the speech envelope and from beta power (14-18 Hz) in the brain to low-frequency delta (1-4 Hz) phase in the speech envelope. Our study reveals robust prediction processes at low-frequencies (1-4 Hz). This is the frequency that represents intonation and prosody (Ghitza, 2011; Giraud and Poeppel, 2012) but also overlaps at the upper end with the mean syllable rate (Ding et al., 2017). It shows an interesting temporal progression from left inferior fronto-motor areas to right auditory-temporal areas. This progression of prediction from higher order areas in the left hemisphere (200-300 ms prior to speech) to the early sensory areas in the right hemisphere suggests that the brain first generates prediction of upcoming sensory input by top-down contextual knowledge that is later used for optimised stimulus encoding in early sensory areas. Indeed, top-down modulations of delta phase (such as temporal expectation or selection of an attended stimulus stream) has been shown to increase sensitivity to external inputs in the auditory (Lakatos, Karmos, Mehta, Ulbert, &
Similarly, we find that beta power in the left frontal cortex and sensorimotor areas reflects prediction of upcoming speech relatively early (200-500 ms prior to speech) and is left-lateralised (Supplementary Figure 2). Active inference by motor systems regarding predictive coding has been studied recently and beta oscillation has been suggested to be working together with low-frequency activity in top-down modulation of ongoing activity during predictive coding (Arnal and Giraud, 2012). Recently an elegant study employing an auditory attention task has shown that interdependent delta and beta activity from left sensorimotor cortex encodes temporal prediction and this is directed towards auditory areas (Morillon and Baillet, 2017). This is consistent with our finding that both delta phase and beta power in the left frontal and sensorimotor engages in the prediction of forthcoming speech from relatively early stage. The temporal progression from inferior frontal to motor areas seems to suggest a hierarchical organisation of prediction processes that warrant further investigation.
During continuous speech perception, brain oscillations entrained by speech (positive delays) and predicting speech (negative delays) are expected to interact in time. In other words, there is a continuous recurrent interaction between stimulus-driven bottom-up processing and top-down prediction processing during continuous speech perception. We focused our analysis on the interaction between top-down predictive coding, i.e., TE from beta power in the brain to delta phase in the speech (Figure 4, negative delays) in the auditory cortex and higher order areas, i.e., fronto-motor areas and speech entrainment, i.e., TE from delta phase in the speech to delta phase in the brain (Figure 2, positive delays). In the left auditory cortex (Figure 5A), subjects with better top-down prediction by beta power at ~250 ms (~peak of LAC in Figure 4B; Supplementary Figure 2B) show better entrainment by speech delta phase at ~250 ms (~crossing point between the LAC and RAC in Figure 2B). This result indicates that although low-frequency brain oscillations following speech envelope seems stronger in the right hemisphere than the left hemisphere (higher t-value for RAC at early delays in Figure 2B; Supplementary Figure 1), the interaction between both (top-down prediction and bottom-up speech entrainment) is modulated by the left auditory cortex (Park, et al., 2015). In the higher order areas (Figure 5B), speech-driven bottom-up information flow to the left motor cortex in the delta phase at delays ~200 ms is strongly associated with top-down predictive information flow from beta power in the left IFG to speech delta phase at ~400 ms prior to the speech. This indicates that subjects with better ability to control top-down predictive speech coding ~400 ms prior to the
upcoming speech by beta power in the left IFG (orange line in Figure 4B; Supplementary Figure 2C) are also better bottom-up entrained by speech delta phase at ~200 ms in the left motor cortex (pink line in Figure 2D).
In summary, our results indicate that predictive processes during continuous speech processing involve fronto-motor areas, operate in at least two frequency channels (delta and beta), follow an organised temporal progression from higher-order areas to early sensory areas and recurrently interact with reactive processes. Further research is needed to decode the exact nature of these predictions, identify the contributions of individual areas and elucidate the mutual dependencies between processes that precede and follow speech.
Acknowledgments
JG is supported by the Wellcome Trust (098433) and GT is supported by the Wellcome Trust (098434). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Disclosure of interest
The authors report no conflict of interest.
References
Arnal, L. H., & Giraud, A. L. (2012). Cortical oscillations and sensory predictions. [Review]. Trends Cogn Sci, 16(7), pp. 390-398. doi:10.1016/j.tics.2012.05.003 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/22682813
Bastos, A. M., Usrey, W. M., Adams, R. A., Mangun, G. R., Fries, P., & Friston, K. J. (2012). Canonical microcircuits for predictive coding. Neuron, 76(4), pp. 695-711. doi:10.1016/j.neuron.2012.10.038 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/23177956
Besl, P. J., & McKay, N. D. (1992). A method for registration of 3-D shapes. IEEE T Pattern Anal, pp. 239–256.
Brainard, D. H. (1997). The Psychophysics Toolbox. Spat Vis, 10(4), pp. 433-436. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/9176952
Chandrasekaran, C., Trubanova, A., Stillittano, S., Caplier, A., & Ghazanfar, A. A. (2009). The natural statistics of audiovisual speech. PLoS Comput Biol, 5(7), e1000436. doi:10.1371/journal.pcbi.1000436 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/19609344
Cravo, A. M., Rohenkohl, G., Wyart, V., & Nobre, A. C. (2013). Temporal expectation enhances contrast sensitivity by phase entrainment of low-frequency oscillations in visual cortex. J Neurosci, 33(9), pp. 4002-4010. doi:10.1523/JNEUROSCI.4675-12.2013 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/23447609
Ding, N., Melloni, L., Zhang, H., Tian, X., & Poeppel, D. (2016). Cortical tracking of hierarchical linguistic structures in connected speech. Nat Neurosci, 19(1), pp. 158-164. doi:10.1038/nn.4186 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/26642090
Ding, N., Patel, A. D., Chen, L., Butler, H., Luo, C., & Poeppel, D. (2017). Temporal modulations in speech and music. Neurosci Biobehav Rev, 81(Pt B), pp. 181-187. doi:10.1016/j.neubiorev.2017.02.011 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/28212857
Florin, E., Gross, J., Pfeifer, J., Fink, G. R., & Timmermann, L. (2010). The effect of filtering on Granger causality based multivariate causality measures. Neuroimage, 50(2), pp. 577-588. doi:10.1016/j.neuroimage.2009.12.050 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/20026279
Fontolan, L., Morillon, B., Liegeois-Chauvel, C., & Giraud, A. L. (2014). The contribution of frequency-specific activity to hierarchical information processing in the human auditory cortex. Nat Commun, 5, p 4694. doi:10.1038/ncomms5694 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/25178489
Friston, K. J., & Frith, C. D. (2015). Active inference, communication and hermeneutics. Cortex, 68, pp. 129-143. doi:10.1016/j.cortex.2015.03.025 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/25957007
Ghitza, O. (2011). Linking speech perception and neurophysiology: speech decoding guided by cascaded oscillators locked to the input rhythm. Front Psychol, 2, p 130. doi:10.3389/fpsyg.2011.00130 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/21743809
Giordano, B. L., Ince, R. A. A., Gross, J., Schyns, P. G., Panzeri, S., & Kayser, C. (2017). Contributions of local speech encoding and functional connectivity to audio-visual speech perception. Elife, 6 doi:10.7554/eLife.24763 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/28590903
Giraud, A. L., & Poeppel, D. (2012). Cortical oscillations and speech processing: emerging computational principles and operations. [Research Support, N.I.H., Extramural Research Support, Non-U.S. Gov't]. Nat Neurosci, 15(4), pp. 511-517. doi:10.1038/nn.3063 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/22426255
Gross, J., Baillet, S., Barnes, G. R., Henson, R. N., Hillebrand, A., Jensen, O., ... Schoffelen, J. M. (2013). Good practice for conducting and reporting MEG research. [Research Support, Non-U.S. Gov't]. Neuroimage, 65, pp. 349-363. doi:10.1016/j.neuroimage.2012.10.001 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/23046981
Gross, J., Hoogenboom, N., Thut, G., Schyns, P., Panzeri, S., Belin, P., & Garrod, S. (2013). Speech rhythms and multiplexed oscillatory sensory coding in the human brain. PLoS Biol, 11(12), p e1001752. doi:10.1371/journal.pbio.1001752 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/24391472
Ince, R. A., Mazzoni, A., Bartels, A., Logothetis, N. K., & Panzeri, S. (2012). A novel test to determine the significance of neural selectivity to single and multiple potentially correlated stimulus features. J Neurosci Methods, 210(1), pp. 49-65. doi:10.1016/j.jneumeth.2011.11.013 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/22142889
Ince, R. A., Schultz, S. R., & Panzeri, S. (2014). Estimating Information-Theoretic Quantities. In D. Jaeger & R. Jung (Eds.), Encyclopedia of Computational Neuroscience.
Kutas, M., & Federmeier, K. D. (2011). Thirty years and counting: finding meaning in the N400 component of the event-related brain potential (ERP). Annu Rev Psychol, 62, pp. 621-647. doi:10.1146/annurev.psych.093008.131123 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/20809790
Lakatos, P., Karmos, G., Mehta, A. D., Ulbert, I., & Schroeder, C. E. (2008). Entrainment of neuronal oscillations as a mechanism of attentional selection. Science, 320(5872), pp. 110-113. doi:10.1126/science.1154735 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/18388295
Levinson, S. C. (2016). Turn-taking in Human Communication--Origins and Implications for Language Processing. Trends Cogn Sci, 20(1), pp. 6-14. doi:10.1016/j.tics.2015.10.010 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/26651245
Massey, J. (1990). Causality, feedback and directed information. In: Proc. Int. Symp. Information Theory Application (ISITA 1990), pp. 303-305.
Morillon, B., & Baillet, S. (2017). Motor origin of temporal predictions in auditory attention. Proc Natl Acad Sci U S A, 114(42), pp. E8913-E8921. doi:10.1073/pnas.1705373114 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/28973923
Nolte, G. (2003). The magnetic lead field theorem in the quasi-static approximation and its use for magnetoencephalography forward calculation in realistic volume conductors. Phys Med Biol, 48(22), pp. 3637-3652. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/14680264
Norris, D., McQueen, J. M., & Cutler, A. (2016). Prediction, Bayesian inference and feedback in speech recognition. Lang Cogn Neurosci, 31(1), pp. 4-18. doi:10.1080/23273798.2015.1081703 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/26740960
Oostenveld, R., Fries, P., Maris, E., & Schoffelen, J. M. (2011). FieldTrip: Open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. [Research Support, Non-U.S. Gov't]. Comput Intell Neurosci, 2011, p 156869. doi:10.1155/2011/156869 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/21253357
Panzeri, S., Senatore, R., Montemurro, M. A., & Petersen, R. S. (2007). Correcting for the sampling bias problem in spike train information measures. J Neurophysiol, 98(3), pp. 1064-1072. doi:10.1152/jn.00559.2007 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/17615128
Park, H., Ince, R. A., Schyns, P. G., Thut, G., & Gross, J. (2015). Frontal Top-Down Signals Increase Coupling of Auditory Low-Frequency Oscillations to Continuous Speech in Human Listeners. Curr Biol, 25(12), pp. 1649-1653. doi:10.1016/j.cub.2015.04.049 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/26028433
Park, H., Kayser, C., Thut, G., & Gross, J. (2016). Lip movements entrain the observers' low-frequency brain oscillations to facilitate speech intelligibility. Elife, 5 doi:10.7554/eLife.14521 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/27146891
Pefkou, M., Arnal, L. H., Fontolan, L., & Giraud, A. L. (2017). theta-Band and beta-Band Neural Activity Reflects Independent Syllable Tracking and Comprehension of Time-Compressed Speech. *J Neurosci*, 37(33), pp. 7930-7938. doi:10.1523/JNEUROSCI.2882-16.2017 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/28729443
Pernet, C. R., Wilcox, R., & Rousselet, G. A. (2012). Robust correlation analyses: false positive and power validation using a new open source matlab toolbox. *Front Psychol*, 3, p 606. doi:10.3389/fpsyg.2012.00606 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/23335907
Pickering, M. J., & Garrod, S. (2007). Do people use language production to make predictions during comprehension? *Trends Cogn Sci*, 11(3), pp. 105-110. doi:10.1016/j.tics.2006.12.002 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/17254833
Pickering, M. J., & Garrod, S. (2013). An integrated theory of language production and comprehension. *Behav Brain Sci*, 36(4), pp. 329-347. doi:10.1017/S0140525X12001495 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/23789620
Poeppel, D. (2003). The analysis of speech in different temporal integration windows: cerebral lateralization as ‘asymmetric sampling in time’. *Speech Communication*, 41, pp. 245-255.
Schreiber, T. (2000). Measuring information transfer. *Phys Rev Lett*, 85(2), pp. 461-464. doi:10.1103/PhysRevLett.85.461 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/10991308
Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O., Delcroix, N., . . . Joliot, M. (2002). Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. *Neuroimage*, 15(1), pp. 273-289. doi:10.1006/nimg.2001.0978 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/11771995
Weber, I., Florin, E., von Papen, M., & Timmermann, L. (2017). The influence of filtering and downsampling on the estimation of transfer entropy. *PLoS One*, 12(11), p e0188210. doi:10.1371/journal.pone.0188210 Retrieved from https://www.ncbi.nlm.nih.gov/pubmed/29149201
Zion Golumbic, E., Cogan, G. B., Schroeder, C. E., & Poeppel, D. (2013). Visual input enhances selective speech envelope tracking in auditory cortex at a "cocktail party". *J Neurosci*, 33(4), pp. 1417-1426. doi:10.1523/JNEUROSCI.3675-12.2013 Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/23345218
Figure Captions
Figure 1. Temporal dynamics of information flow during natural speech perception.
In the present study, we analysed transfer entropy (TE) to investigate two mechanisms of information flow during natural speech perception: 1) Predictive speech coding (top-down prediction mechanism investigated by negative delays between speech-brain; purple color) and 2) Speech entrainment (stimulus-driven bottom-up processing investigated by positive delays between speech-brain; cyan color). For speech signal, we used low-frequency delta phase information, and for brain signal, we used both low-frequency delta phase and high-frequency beta power information.
Figure 2. Entrained brain signals following speech.
(a) A schematic figure for directed causal analysis: TE from speech delta phase to brain delta phase (positive delays between speech-brain). TE computation was performed for each condition (intelligible-forward played and unintelligible-backward played) at each voxel from 20 ms to 500 ms with a 20-ms step. TE values are averaged within each ROI from the AAL atlas and compared statistically between conditions. T-values are shown in each ROI bilaterally (pink: left hemisphere, green: right hemisphere): (b) primary auditory cortex (Heschl gyrus) (c) Inferior frontal gyrus – opercular part (BA44) (d) precentral gyrus. Statistical significance was shown with two red lines depicting t-values by paired t-test (upper red line: t_{21} = 3.53, p < 0.05, corrected; bottom red line: t_{21} = 2.08, p < 0.05, uncorrected).
Figure 3. Entrained brain signals predicting upcoming speech.
(a) A schematic figure for directed causal analysis: TE from brain delta phase to speech delta phase (negative delays between speech-brain). TE computation was performed for each condition (intelligible-forward played and unintelligible-backward played) at each voxel from 20 ms to 500 ms in 20-ms steps and compared statistically between conditions. (b) The strongest prediction was found at ~220 ms in left IFG (upper red line: t_{21} = 3.53, p < 0.05, corrected; bottom red line: t_{21} = 2.08, p < 0.05,
uncorrected). (c) Statistical contrast maps of averaged across 100 ms windows show a sequence of events that start around -300 ms in fronto-motor areas and then move to right auditory-temporal areas at around -200 ms prior to speech (p < 0.05, FDR-corrected).
**Figure 4. Beta rhythms and the prediction of upcoming speech.**
(a) A schematic figure for directed causal analysis: TE from brain beta power to speech delta phase (negative delays between speech-brain). TE computation was performed for each condition (intelligible-forward played and unintelligible-backward played) at each voxel from 20 ms to 500 ms in 20 ms steps and compared statistically between conditions. (b) TE values are averaged within each ROI from the AAL atlas and compared statistically between conditions. T-values are shown in each ROI: left primary auditory cortex (Heschl gyrus) (blue line), left IFG (orange line), and left precentral gyrus (both at local maximum coordinate (yellow line) and whole precentral gyrus ROI (light green line)). Statistical significance was shown with two red lines depicting t-values by paired t-test (upper red line: t_{21} = 3.53, p < 0.05, corrected; bottom red line: t_{21} = 2.08, p < 0.05, uncorrected). Left-lateralised predictions by beta power were observed (see Supplementary Figure 2). (c) Statistical contrast maps of averaged across 100 ms windows show that intelligibility-dependant prediction first engages left inferior frontal gyrus about 300-500 ms prior to the speech followed by left precentral gyrus and left primary auditory area 200-300 ms prior to the speech (p < 0.05, FDR-corrected).
**Figure 5. Relationship between temporal dynamics of speech entrainment and predictive speech coding.**
To assess how the temporal dynamics of speech entrainment (positive delays) and intelligibility-dependant top-down prediction (negative delays) are interacting, we performed correlation analysis using robust Spearman rank correlations across participants between the two mechanisms (but for top-down prediction, we used TE from brain beta to speech delta; Figure 4). We tested the relationship within in early sensory area, i.e. primary auditory cortex as well as higher order areas. (a) Delta phase in the left auditory cortex driven by speech (corresponding to the plot for the left auditory cortex (pink line) at ~250 ms in the Figure 2B) is associated with predictive speech coding mechanism.
of beta power in the left auditory cortex ~250 ms prior to the forthcoming speech with low frequency delta phase information (corresponding to the plot for the left auditory cortex (blue line) at ~250 ms in the Figure 4B) \( (r = 0.46, p = 0.02) \). (b) Delta phase in the left precentral gyrus modulated by the same frequency phase in the speech at ~200 ms (corresponding to the plot for the left precentral gyrus (pink line) at ~200 ms in the Figure 2D) is associated with predictive speech coding of beta power in the left inferior frontal gyrus ~400 ms prior to the forthcoming speech with low frequency delta phase information (corresponding to the plot for left inferior frontal gyrus (orange line) at ~400 ms in the Figure 4B) \( (r = 0.53, p = 0.01) \).
Figure 1
Listener’s Brain Rhythms at both Low and High Frequencies
Predictive Speech Coding (negative delays)
Speech Entrainment (positive delays)
Speaker’s Speech Rhythms at Low Frequency
Figure 3
(a) From Brain Delta to Speech Delta
(b) Strongest Prediction at ~220 ms in L-IFG
(c) From Brain Delta To Speech Delta at each 100 ms delay window
-400~300 ms
-300~200 ms
-200~100 ms
-100~0 ms
Intelligible vs. Unintelligible (t-value)
p < 0.05 (corrected)
Delay (ms)
Colorbar: t-value (FDR-corr)
Figure 4
(a) From Brain Beta to Speech Delta
(b) Predictive Speech Coding
(c) From Brain Beta To Speech Delta at each 100 ms delay window
Legend:
- LAC
- L-JFG
- L-PreC (local max)
- L-PreC
p < 0.05 (corrected)
Delay (ms)
Intelligible vs. Unintelligible (t-value)
-500 to -400 ms
-400 to -300 ms
-300 to -200 ms
-200 to -100 ms
-100 to 0 ms
Supplemental Materials
Supplemental Figure 1 (related to Figure 2). Entrained brain signals following speech at whole brain at each 100 ms window.
TE computation was performed for each condition (intelligible-forward played and unintelligible-backward played) at each voxel from 20 ms to 500 ms with a 20-ms step. To characterise the spatio-temporal pattern across delays at whole brain level, we averaged TE-maps in 100 ms windows and computed again the statistical contrast of intelligible compared to unintelligible speech (only t-values > 4.7, p < 0.0001, FDR-corrected are plotted). Delta phase information in the brain following the same frequency information in the speech envelope is robust across all delays and the pattern is stronger in the right hemisphere (higher t-values).
Supplemental Figure 2 (related to Figure 4). Beta rhythms involving the prediction of upcoming speech are left-lateralised.
Here we show the same plots as in the Figure 4B, but separately for each ROI with the homologous ROI in the right hemisphere; (a) Primary auditory cortex (Heschl gyrus), (b) Inferior frontal gyrus – opercular part, (c) Precentral gyrus. Top-down prediction by beta power in the brain to speech delta phase is left-lateralised (statistics by paired t-test; upper red line: t\(_{21} = 3.53\), p < 0.05, corrected; bottom red line: t\(_{21} = 2.08\), p < 0.05, uncorrected). Our results indicate three different mechanisms in terms of hemispheric asymmetry. 1) Speech-driven entrainment by delta phase is shown bilaterally (Figure 2; Supplementary Figure 1). 2) However, top-down prediction by the same delta phase shows progression from left inferior fronto-motor areas (200-300 ms prior to speech) to right auditory-temporal areas (0-200 ms prior to speech). 3) Top-down prediction by beta power in the fronto-motor areas is left-lateralised from early stage (200-500 ms prior to speech). | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
1039,
1
],
[
1039,
2127,
2
],
[
2127,
4844,
3
],
[
4844,
6644,
4
],
[
6644,
8740,
5
],
[
8740,
11197,
6
],
[
11197,
13190,
7
],
[
13190,
14964,
8
],
[
14964,
17477,
9
],
[
17477,
19922,
10
],
[
19922,
22473,
11
],
[
22473,
24357,
12
],
[
24357,
27154,
13
],
[
27154,
29848,
14
],
[
29848,
32723,
15
],
[
32723,
33475,
16
],
[
33475,
33786,
17
],
[
33786,
37287,
18
],
[
37287,
41307,
19
],
[
41307,
43697,
20
],
[
43697,
45812,
21
],
[
45812,
48241,
22
],
[
48241,
49003,
23
],
[
49003,
49196,
24
],
[
49196,
49196,
25
],
[
49196,
49509,
26
],
[
49509,
49858,
27
],
[
49858,
49858,
28
],
[
49858,
50648,
29
],
[
50648,
51760,
30
]
]
} |
dcbc1d5305994fa47dc6583f7c9aa8121bda65fd | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 8,
"total-fallback-pages": 0,
"total-input-tokens": 22500,
"total-output-tokens": 7928,
"fieldofstudy": [
"Chemistry",
"Biology"
]
} | Covalent Modification by Glyoxals Converts Cytochrome c Into its Apoptotically Competent State
Gurumayum Suraj Sharma1,2, Marina Warepam3, Reshmee Bhattacharya4 & Laishram Rajendrakumar Singh1
The effects of glycation by glyoxal (Gly) and methylglyoxal (MGly) on the early and late conformational alterations in Cytochrome c (Cyt c) were studied. Spectroscopic measurements revealed that Cyt c undergo certain conformational alterations and exposure of heme upon overnight incubation with Gly and MGly. These were followed by the reduction of heme centre and activation of its peroxidase-like, which is crucial for initiation of the intrinsic apoptotic pathway. An extended incubation resulted in appearance of AGE-like fluorescence, with significant alterations in secondary structural compositions. However, no amyloidogenic conversions were observed as suggested by TEM analyses. The study provides an insight to the role of glycating agents, elevated under oxidative stress in inducing Cyt c release and apoptosis.
Hyperglycemia represents a common hallmark of diabetic complications and is characterized by increased levels of sugars and their metabolites1. Gly and MGly are formed via oxidation of reducing sugars under chronic hyperglycemic conditions and endogenous MGly can also be formed from triose phosphate intermediates of glycolysis2. These sugars metabolites have high tendency of covalently modifying proteins, via a process termed as "protein glycation". Such modifications are known to induce protein structural alterations resulting in functional loss and even lead to aggregate/amyloids formation, and have been associated with several age related disorders and neurodegenerative diseases (NDs). Accumulation of advanced glycation end products (AGEs) in plasma as a consequence of glycation is also a primary factor for coronary heart diseases3–7.
Most studies on protein glycation have focused largely on identifying toxic oligomers and formation of AGEs8–12. However, early structural alterations accompanying such modifications have not yet been thoroughly studied. Cyt c is a multi-functional protein present in inner mitochondrial membrane (IMM). In respiration, it is involved in electron transport (ET), whereas in apoptosis, Cyt c release from IMM is known to activate caspase 9 enzyme, initiating a cascade of reactions. Structural alterations in the protein and disruption of its redox state are known to trigger its pro-apoptotic activity13–17 and these changes represent early events in apoptosis16. Hence, maintaining its native structure and redox status is crucial for proper cellular function. Considering the fact that alteration in conformation of Cyt c is known to be an early event in driving cells towards apoptosis, it is worthwhile to study the effects of such agents on the early structural transitions of Cyt c. We have investigated the early and late conformational alterations in Cyt c upon glycation by Gly and MGly. Our results suggest that overnight incubation with Gly and MGly resulted in certain tertiary structural changes with exposed heme, disruption of the redox state and heme-Met80 coordination, ultimately resulting in activation of peroxidase function. Upon prolonged incubation, there was significant loss in secondary structure with appearance of AGEs specific fluorescence. Our study provides an insight to the molecular mechanism for the role of glycated Cyt c in eliciting its toxicity.
Results
Determination of Cyt c Glycation. To determine the extent of glycation, we have carried out carbonyl estimation assay for the untreated and treated Cyt c samples. It is seen in Table 1 that the proteins have been incorporated with Gly and MGly as suggested by increase in carbonyl contents. After overnight incubation, there were 2–4 and 3–5 fold increase for Gly- and MGly-modified Cyt c, respectively. These were further confirmed using MALDI-TOF mass spectrometric analysis (See Fig. 1). For each molecule of Gly or MGly reacting with side
chain of Cyt c, there will be an increase of 58 and 74 Da in the protein mass. It is seen in Fig. 1, that incubation of Cyt c with 10 mM Gly/MGly resulted in appearance of two more peaks, in addition to the major peak which corresponds to the non-glycated form of Cyt c. Our result suggests that the protein was glycated at multiple sites, and presents a mixture of non-glycated Cyt c and Cyt c with Gly/MGly linked to 3–7 sites.
Glycation disrupts Cyt c structure. We further assessed the effects of Gly and MGly on early and late conformational alterations of Cyt c. Early conformational changes were followed after overnight incubation (12–14 Hrs) of Cyt c with Gly and MGly. As seen in Fig. 2A,B, glycation results in disruption of native tertiary interactions. In both cases, there is a significant loss in tertiary contacts as suggested by near-UV CD spectra. There were subtle alterations in secondary structural components (see Fig. 2C, D) indicating that the resulting conformation is not a denatured state, but a non-native state with disordered tertiary interactions. However, we did not observe any ANS binding to these non-native structures (Fig. 2E). Visible absorption spectroscopic measurement further revealed that there is a slight increase in the absorption of heme (409 nm) indicating partial heme exposure to solvent (Fig. 2F).
Redox status of Cyt c upon glycation. Exposure of heme upon glycation could in turn affect the redox state of Cyt c. To verify this, we assessed the redox status of glycated Cyt c. We found that the heme centre has been reduced upon modification. Visible absorption spectra of reduced form of Cyt c is defined by two distinct bands at 520 and 550 nm, while oxidized Cyt c exhibits a single broad band with maximum at 530 nm, as also observed in the present study (Fig. 3A). Furthermore, electron paramagnetic resonance (EPR) analyses confirmed reduction of the heme moiety (Fig. 3B). The EPR spectrum of native Cyt c in its oxidized state consists of two signals, $g = 2.3$ and $g = 3.5^{18,19}$, which were intact for unmodified protein. Upon glycation, these bands disappeared, suggesting the conversion of oxidized Cyt c to reduced form. Furthermore, it was also found that glycation led to activation of peroxidase-like function of Cyt c (Fig. 3C and Fig. S1), which is a crucial step for Cyt c release during intrinsic apoptotic pathway.
Cyt c Heme-Met80 ligation upon glycation. Alterations in heme iron sixth ligand (Met80) are known to bring about certain affects, both in structural and functional properties of Cyt c. To evaluate the heme-Met80 interaction, we carried out visible absorption and soret CD measurements. Figure 3D, E shows the visible absorption and soret CD spectra of unmodified and modified Cyt c. It is seen that there is disruption of the 695 nm (visible absorption) and 416 nm (soret CD) bands in case of modified Cyt c suggesting loss of heme-Met80
interactions in modified proteins. Under normal conditions, heme in Cyt c exists in hexa-coordinated state. The 695 nm absorbance band is assigned diagnostic characteristic for Met80 coordination to heme moiety. Alteration (or disruption) of this axial ligation lead to loss of this band. Indeed heme-Met80 interaction is disrupted upon glycation and Cyt c undergoes conformational transition as evidenced by decrease in absorption band at 695 nm. This alteration was further probed using another spectroscopic signature, soret CD measurement. The negative dichroic band at 416 nm has also been designated to heme-Met80 ligation which is lost in presence of glycating agents. Our results indicate that the non-native structure induced by glycation is a penta-coordinate structure with disrupted tertiary interactions.
Modified Cyt c exhibits AGEs-specific fluorescence characteristics upon prolonged incubation with appearance of non-amyloidogenic structures. Fluorescence measurements after incubation for two weeks revealed that modified Cyt c exhibited AGE-specific fluorescence. Formation of AGEs was monitored using excitation and emission at 350 nm and 450 nm respectively. Increase in fluorescence intensity at 450 nm points towards the formation of AGEs upon glycation (Fig. 4A, B). The conformational alterations upon prolonged incubation were further assessed using far-UV CD measurements (Fig. 4C, D). It is seen that the negative peaks at 208 nm and 222 nm, which are the characteristic features for \( \alpha \)-helix, is gradually lost with increasing concentration of glyoxals. Secondary structure estimation upon extended incubation of Cyt c with Gly and MGly, suggests a prominent transition of the \( \alpha \)-helical components to more unordered structures with increased \( \beta \)-sheet structures (in case of MGly) (Fig. S2). The morphological features of modified proteins were further assessed using TEM. TEM images (Fig. 5) revealed that upon 2-week incubation, Gly-modified Cyt c exists as clusters of spherical structure, whereas MGly-modified Cyt c exists as amorphous structure. However, no amyloidogenic transitions were observed.
Discussion
Glycation preferably targets the side chains of proteins, particularly Lys and therefore Lys-rich proteins could be the major targets for glycation. Thus, Cyt c (with 19 Lys residues) could be a prime target for such modifications. Our results on conformational assessments suggest that upon glycation, Cyt c attains a non-native state characterized by perturbed tertiary interactions and partially exposed heme. Furthermore, heme in the resultant non-native Cyt c was found to be in a penta-coordinated state with disrupted heme-Met80 ligation. Such structural transition may have important physiological consequences. Since Cyt c is a protein essential for intrinsic apoptotic pathway, and the conformational alterations in this protein has been shown to be responsible for intrinsic apoptotic activity, it is important to examine if the non-native intermediate is an apoptotically-competent species. Cardiolipin (CL) oxidation due to the activation of Cyt c peroxidase function is believed to be primary event towards early intrinsic apoptotic pathway leading to release of Cyt c from IMM. We measured the peroxidase activity of modified Cyt c using two different substrates. It was observed that modified Cyt c exhibits peroxidase activity, but not the native protein. Similarly, ribose-5-phosphate (R5P) has also been shown to induce certain alterations in Cyt c. However, peroxidase activation required prolonged incubation period (one week or more).
In addition, modification of Cyt c by homocysteine thiolactone, accumulated under hyperhomocysteinemia has...
also been shown to induce gross conformational alteration leading to reduction of heme moiety and peroxidase activation\(^2^4\). We conclude that these early structural intermediate represents an apoptotically-competent conformation and therefore hint towards possible involvement of such post-translationally modified Cyt c in apoptosis. In support, many studies using different cell lines showed that Gly and MGly induce apoptosis when provided exogenously\(^2^5\)–\(^2^9\). Furthermore, concentrations of both compounds in mitochondria are elevated under hyperglycemic conditions and contribute to mitochondrial dysfunction associated with diabetes and aging\(^3^0\). In addition to this, it has been demonstrated that disruption of Cyt c heme coordination is responsible for mitochondrial injury during ischemia\(^3^1\). Taken together, these results therefore indicate that modification of Cyt c by glycation might be associated with mitochondrial injury leading to apoptosis.
Since glycation is known to result in generation of free radicals\(^3^2\), redox state of Cyt c heme could also be altered upon such modifications. To examine the redox status, we performed visible absorption and EPR spectral measurements of the native and modified proteins. Our absorption and EPR spectral measurements revealed that the heme centre of modified Cyt c was rendered reduced upon glycation. The electrostatic interaction between Cyt c and CL is believed to be responsible for anchoring the protein to IMM and optimizes ET between complex III and complex IV\(^3^1\). In fact, glycation by RSP has also been shown to weaken the ability of Cyt c to transfer electrons in respiratory pathway and to bind membranes\(^3^3\). Hence, it is possible that modifications that lead to
---
**Figure 3.** Heme status of Cyt c upon glycation. Visible absorption (A) and EPR (B) spectra of native and glycated Cyt c. Panel C shows the peroxidase activity of native and glycated Cyt c. Panel D, E show the absorption and soret CD spectra of native and modified Cyt c. Only the spectra of unmodified and Cyt c modified with 10 mM glyoxals are shown.
reduction of heme in Cyt c would thereby not only disrupt normal functionality of the protein in ET process and cellular energy demands, but also help the protein shed off from IMM.
A number of studies have shown the involvement of different proteins in generation of AGEs and their relation with protein aggregation \(^9,22,33\). In addition, the presence of high levels of AGEs in brains of patients with neurologica conditions and their association with amyloid deposition have also been reported \(^34,35\). In fact, glycation has been shown to enhance the severity linked with neurotoxicity of A\(_{β}^{1-42}\) peptide \(^36\). Extended incubation (two weeks) of Cyt c with Gly and MGly resulted in rapid increase in AGEs fluorescence (see Fig. 4A, B). Since AGEs formation has been associated with various clinical complications and human pathological conditions, including NDs \(^37-41\), it is very likely that AGEs formation induced by these agents could also contribute to these pathologies. We have further analysed the morphological nature of glycated Cyt c. Our TEM studies of Gly-modified Cyt c suggest generation of spherical structures with varying sizes, with diameters ranging from 50–100 nm (Fig. 5A). These structures do not have any specificity towards amyloids specific dye (ThT) (Fig. S3), suggesting that these species are non-amyloidogenic. However, MGly-modified Cyt c showed disordered amorphous structures (Fig. 5B). The variation in nature of aggregates could be due to the difference in structural alterations induced by these two molecules. Far-UV CD measurements (Fig. 4D) revealed that MGly induced significant alterations in secondary structural element with appearance of prominent disordered-like character. The 208 nm and 222 nm bands, characteristic signature of \(α\)-helical structure were completely diminished in presence of MGly. At higher concentrations of MGly used (5 and 10 mM), there was the appearance of new signal around 204 nm suggesting that the modified protein now exists as an unordered state. In case of Gly, there were disruptions of \(α\)-helical structures, but no appearance of disordered state. Thus, differences on the effect of the modifying agents on secondary structural levels could be one reason for the differences in final state of aggregates. Similarly, glycation (by D-ribose) has been shown to generate globular aggregates in case of \(α\)-synuclein \(^42\). No fibrillar or amyloidogenic structures were observed. In case of insulin, glycation by MGly was shown to reduce insulin fibrillation, leading to formation of “native-like” aggregates \(^12\). It is suggested that modification led to a less compact and less stable structure which may be associated to increased dynamics, preventing the formation of rigid cross-\(β\) core structure found in amyloid fibrils. The study pointed that MGly could trigger a drifting from an amyloidogenesis towards a “native-like” aggregation pathway, a mechanism that might be important in the context of the amyloidogenicity of AGE-modified proteins involved in conformational diseases \(^12\).
Our study demonstrates the role of glycation-induced early structural alterations and gain of peroxidase function in Cyt c can lead to CL oxidation, hence provides a hint for Cyt c release. Furthermore, prolonged incubation of Cyt c with glyoxals generated AGEs which are also associated with several neurological disorders. Analyses of these AGEs via TEM confirmed that these species exist as clustered spherical or amorphous structures. Studies on how conformational alterations in the protein induce mitochondrial injury could yield useful insights in understanding the mechanism of early apoptotic pathway.
Experimental Procedure
Materials. Cyt c (from bovine heart), and other chemicals were purchased from Sigma-Aldrich. Protein solution was dialyzed extensively against 0.1 M KCl at pH 7.0 at 4°C. Cyt c was oxidized using 0.01% potassium ferrocyanide before dialysis. All solutions for optical measurements were prepared in degassed buffer (0.05 M phosphate buffer, pH 7.4).
Protein modification. For protein modification, Cyt c was incubated in presence of varying concentrations of Gly and M Gly (0–10 mM) in 0.05 M potassium phosphate buffer, pH 7.4 at 37°C. These treated/un treated protein samples were further used for subsequent analyses.
Carbonyl Estimation. Carbonyl content in control and modified proteins were assayed by method described by Levine et al. Briefly, an aliquot of treated Cyt c were incubated for 1 hr at room temperature with DNPH (0.1% w/v in 2 N HCL). The reaction was stop by addition of equal volumes of 20% trichloroacetic acid (TCA) and centrifuged to obtain a pellet. DNPH was removed by extracting the pellets two times using 1 ml of ethyl acetate:ethanol (1:1 v/v) solution. Pellets were dried and dissolved in 6.0 M GdmCl (pH 7.0). Solubilized hydrazones were measured at 370 nm and concentration of DNPH derivatized proteins was determined using molar extinction coefficient of 22,000 M⁻¹ cm⁻¹.
Intact Mass analysis. The treated samples were analyzed using an Applied Biosystems/MDS Scienx 4800 Plus MALDI TOF/TOF Analyzer to obtain the MALDI spectra. Modified protein solutions were mixed with sinapic acid matrix in 1:1 ratio and kept to dry. The protein concentration used was 80 μM.
CD Measurements. CD spectra were procured in Jasco J-810 spectropolarimeter equipped with Peltier-type temperature controller with three accumulations. Protein concentration used for CD measurements were 15–20 μM. Cells of 1 and 10 mm path lengths were used for measurements of far- and near-UV CD spectra, respectively. For Soret CD measurements, cell of 1.0 cm path lengths was used. Necessary blanks were subtracted for each measurement.
UV-Visible Spectrophotometry. Absorption spectra of were recorded in a Jasco V-660 spectrophotometer equipped with Peltier-type temperature controller. The protein concentration used for spectral measurements was 15 μM. For the 695 nm absorption band, protein concentration used was 50 μM. For all measurements, cell of 1.0 cm path length was used.
Electron Paramagnetic Resonance (EPR). EPR measurements were carried out in a Bruker EMX MicroX spectrometer. Following conditions were maintained for the measurements: gain, 1 × 10²; modulation amplitude, 4.0 G; microwave power, 16 mW; temperature, 298 K; conversion time, 20 ms; and time constant, 655 ms. Protein samples were loaded in sealed quartz capillary tubes and transferred to EPR cavity to procure spectra. The protein concentration used was kept at 150 μM.
Fluorescence Measurements. Fluorescence spectra were measured in a Perkin Elmer LS 55 Spectrofluorimeter in a 3 mm quartz cell, with both excitation and emission slits set at 10 nm. Protein concentration for all experiments was 5 μM. For AGEs fluorescence measurements, excitation and emission wavelengths were 300 and 450 nm, respectively. ANS was excited at 360 nm and emission collected at 475–600 nm range and ThT was excited at 450 nm and emission collected at 475–600 nm range. ThT concentration was 25 μM. Necessary blanks were subtracted for each sample.
Peroxidase activity assay. Peroxidase activity was assayed with guaiacol by measuring absorption at 470 nm for tetraguaiacol formed as product. Protein concentration was kept 1 μM, 1 mM guaiacol, and 2 mM H₂O₂. The reactions were also repeated with ABTS and H₂O₂ (100 μM and 1 mM respectively) by measuring absorption at 415 nm.
TEM imaging. Modified Cyt c solutions were placed on copper grid and left to dry at room temperature for 5 min. Negative staining was performed using 1.0% uranyl acetate and allowed to air dry. The samples were examined using FEI Tecnai G2-200kV HRTA TEM (Netherlands) operating at 200kV.
References
1. Basta, G., Schmidt, A. M. & De Caterina, R. Advanced glycation end products and vascular inflammation: implications for accelerated atherosclerosis in diabetes. Cardiovascular research 63, 582–592 (2004).
2. O'Brien, P. J. et al. Fructose-Derived Endogenous Toxins. Endogenous Toxins: Targets for Disease Treatment and Prevention, 173–212 (2010).
3. Castellani, R., Smith, M. A., Richey, G. L. & Perry, G. Glycoxidation and oxidative stress in Parkinson disease and diffuse Lewy body disease. Brain research 737, 195–200 (1996).
4. Gomes, R. et al. Argyryrimidine, a methylglyoxal-derived advanced glycation end-product in familial amyloidotic polyneuropathy. Biochem. J 385, 339–345 (2005).
5. Munch, G. et al. Crosslinking of alpha-synuclein by advanced glycation endproducts—an early pathophysiological step in Lewy body formation? Journal of chemical neuroanatomy 20, 253–257 (2000).
6. Smith, M. A. et al. Advanced Maillard reaction end products are associated with Alzheimer disease pathology. Proceedings of the National Academy of Sciences 91, 195–200 (1994).
7. Yan, S. D. et al. Glycated tau protein in Alzheimer disease: a mechanism for induction of oxidant stress. Proceedings of the National Academy of Sciences 91, 7787–7791 (1994).
8. Booth, A. A., Khalifah, R. G., Todd, P. & Hudson, B. G. In vitro kinetic studies of formation of antigenic advanced glycation end products (AGEs) novel inhibition of post-Amadori glycation pathways. Journal of Biological Chemistry 272, 5430–5437 (1997).
9. Ghosh, S. et al. Prolonged glycation of hen egg white lysozyme generates non amyloidic structures. PloS one 8, e74336 (2013).
10. Iram, A. et al. Molten globule of hemoglobin proceeds into aggregates and advanced glycated end products. Plos One (2013).
11. Oliveira, L. M. A. et al. Insights into the molecular mechanism of protein native-like aggregation upon glycation. Biochimica et Biophysica Acta (BBA)-Proteins and Proteomics 1834, 1010–1022 (2013).
12. Oliveira, L. M. A. et al. Insulin glycation by methylglyoxal results in native-like aggregation and inhibition of fibril formation. BMC biochemistry 12, 41 (2011).
13. Brown, G. C. & Borutaite, V. Regulation of apoptosis by the redox state of cytochrome c. Biochimica et Biophysica Acta (BBA)-Bioenergetics 1777, 877–881 (2008).
14. Garcia-Heredia, J. M. et al. Nitration of tyrosine 74 prevents human cytochrome c to play a key role in apoptosis signaling by blocking caspase-9 activation. Biochimica et Biophysica Acta (BBA)-Bioenergetics 1797, 981–993 (2010).
15. Fluttermann, M. et al. The multiple functions of cytochrome c and their regulation in life and death decisions of the mammalian cell: From respiration to apoptosis. Mitochondrion 11, 369–381 (2011).
16. Jemmerson, R. et al. A conformational change in cytochrome c of apoptotic and necrotic cells is detected by monoclonal antibody binding and mimicked by association of the native antigen with synthetic phospholipid vesicles. Biochemistry 38, 3599–3609 (1999).
17. Ow, Y.-L. P., Green, D. R., Hao, Z. & Mak, T. W. Cytochrome c: functions beyond respiration. Nature Reviews Molecular Cell Biology 9, 332–342 (2008).
18. Gadfly, P. M., Peterson, J., Foote, N., Greenwood, C. & Thomson, A. J. Identification of the ligand-exchange process in the alkaline transition of horse heart cytochrome c. Biochemical Journal 246, 43–54 (1987).
19. Estevez, M. L. et al. Changes in the spin state and reactivity of cytochrome C induced by photochemically generated singlet oxygen and free radicals. Journal of Biological Chemistry 279, 39214–39222 (2004).
20. Stellwagen, E. & Cass, R. Alkaline isomerization of ferricytochrome C from Euglena gracilis. Biochem Biophys Res Commun 60, 371–375, 0026-291X(74)90214-9 (1974).
21. Santucci, R. & Ascoli, F. The Soret circular dichroism spectrum as a probe for the heme Fe (III)-Met (80) axial bond in horse cytochrome c. Journal of inorganic biochemistry 68, 211–214 (1997).
22. Bouma, B. et al. Glycation induces formation of amyloid cross-β structure in albumin. Journal of biological chemistry 278, 41810–41819 (2003).
23. Hildick-Smith, G. J., Downey, M. C., Grewbeck, L. M., Gersten, R. A. & Sandwick, R. K. Ribose 5-phosphate glycation reduces cytochrome c respiratory activity and membrane affinity. Biochemistry 50, 11084–11097 (2011).
24. Sharma, G. S. & Singh, L. R. Conformational status of cytochrome c upon N-homocysteinylation: Implications to cytochrome c release. Archives of Biochemistry and Biophysics 614, 23–27 (2017).
25. Akhand, A. A. et al. Glyoxal and methylglyoxal trigger distinct signals for map family kinases and caspase activation in human endothelial cells. Free Radical Biology and Medicine 31, 20–30 (2001).
26. Chan, W. H., Wu, H. J. & Shiao, N. H. Apoptotic signaling in methylglyoxal-treated human osteoblasts involves oxidative stress, c-Jun N-terminal kinase, caspase-3, and p21-activated kinase 2. Journal of cellular biochemistry 100, 1056–1069 (2007).
27. Kasper, M. et al. Induction of apoptosis by glyoxal in human embryonic lung epithelial cell line L132. American journal of respiratory cell and molecular biology 23, 485–491 (2000).
28. Kim, J., Son, J.-W., Lee, J.-A., Oh, Y.-S. & Shin, S.-H. Methylglyoxal induces apoptosis mediated by reactive oxygen species in bovine retinal pericytes. Journal of Korean medical science 29, 95–100 (2004).
29. Liu, B. F. et al. Methylglyoxal induces apoptosis through activation of p38 mitogen-activated protein kinase in rat mesangial cells. Kidney international 63, 947–957 (2003).
30. Pun, P. B. L. et al. A mitochondria-targeted mass spectrometry probe to detect glyoxals: implications for diabetes. Free Radical Biology and Medicine 47, 437–450 (2014).
31. Birk, A. V., Chao, W. M., Liu, S., Soong, Y. & Szeto, H. H. Disruption of cytochrome c heme coordination is responsible for mitochondrial injury during ischemia. Biochimica et Biophysica Acta (BBA)-Bioenergetics 1847, 1075–1084 (2015).
32. Lee, C., Yim, M. B., Chock, P. B., Yim, H.-S. & Kang, S.-O. Oxidation-reduction properties of methylglyoxal-modified protein in relation to free radical generation. Journal of biological chemistry 273, 25272–25278 (1998).
33. Wei, Y., Chen, L., Chen, J., Ge, L. & He, K. Q. Rapid glycation with D-ribose induces globular amyloid-like aggregations of BSA with high cytotoxicity to SH-SYSY cells. BMC cell biology 10, 10 (2009).
34. Kikuchi, S. et al. Glycation-induced formation of amyloid cross-β structure in albumin. Biochemistry 371–375, 0006-291X(74)90214-9 (1974).
35. Josephs, T. M., Morison, I. M., Day, C. L., Wilbanks, S. M. & Ledgerwood, E. C. Enhancing the peroxidase activity of cytochrome c and free radicals. Biochemical Journal 458, 259–265 (2014).
Acknowledgements
This work is supported partly by R&D Grant and DU-DST Purse Grant, University of Delhi and Department of Biotechnology (BT/PR17096/NER/95/407/2015). We thank Advanced Instrumentation Research Facility, Jawaharlal Nehru University, New Delhi for providing EPR facility and Dr. Smitha Sundaram for her help in EPR spectral measurements, All India Institute of Medical Science (AIIMS), New Delhi for TEM facility and Central Instrumentation Facility (CIF), University of Delhi, South Campus for providing MS facility. We also thank Indian Council of Medical Research for providing fellowship (BMS/FW/Bioch/2014–25490) to GSS and Department of Science and Technology (SERB, DST, PDF/2015/001090) to MW.
Author Contributions
L.R.S. designed the experiments. G.S.S., M.W. and R.B. performed the experiments and analyzed the data. L.R.S. and G.S.S. wrote the manuscript.
Additional Information
Supplementary information accompanies this paper at https://doi.org/10.1038/s41598-019-41282-2.
Competing Interests: The authors declare no competing interests.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
© The Author(s) 2019 | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4011,
1
],
[
4011,
6944,
2
],
[
6944,
10688,
3
],
[
10688,
12820,
4
],
[
12820,
16542,
5
],
[
16542,
22263,
6
],
[
22263,
28064,
7
],
[
28064,
29426,
8
]
]
} |
45ce0213138082882d7f1d65767b1e08b78fabe2 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 15,
"total-fallback-pages": 0,
"total-input-tokens": 46985,
"total-output-tokens": 13736,
"fieldofstudy": [
"Agricultural And Food Sciences"
]
} | Effect of the Rearing Substrate on Total Protein and Amino Acid Composition in Black Soldier Fly
Andrea Fuso 1, Silvia Barbi 2, Laura Ioana Macavei 3, Anna Valentina Luparelli 1, Lara Maistrello 3, Monia Montorsi 3, Stefano Sforza 1 and Augusta Caligiani 1,*
1 Food and Drug Department, University of Parma, Via Parco Area delle Scienze 17/A, 43124 Parma, Italy; [email protected] (A.F.); [email protected] (A.V.L.); [email protected] (S.S.)
2 Interdepartmental Research Center for Industrial Research and Technology Transfer in the Field of Integrated Technologies for Sustainable Research, Efficient Energy Conversion, Energy Efficiency of Buildings, Lighting and Home Automation—En&Tech, University of Modena and Reggio Emilia, Via Amendola 2, 42122 Reggio Emilia, Italy; [email protected]
3 Department of Life Sciences, University of Modena and Reggio Emilia, Via Giovanni Amendola 2, 42122 Reggio Emilia, Italy; [email protected] (L.I.M.); [email protected] (L.M.)
4 Department of Sciences and Methods for Engineering, University of Modena and Reggio Emilia, Via Giovanni Amendola 2, 42122 Reggio Emilia, Italy; [email protected]
* Correspondence: [email protected]
Abstract: Insects are becoming increasingly relevant as protein sources in food and feed. The Black Soldier Fly (BSF) is one of the most utilized, thanks to its ability to live on many leftovers. Vegetable processing industries produce huge amounts of by-products, and it is important to efficiently rear BSF on different substrates to assure an economical advantage in bioconversion and to overcome the seasonality of some leftovers. This work evaluated how different substrates affect the protein and amino acid content of BSF. BSF prepupae reared on different substrates showed total protein content varying between 35% and 49% on dry matter. Significant lower protein contents were detected in BSF grown on fruit by-products, while higher contents were observed when autumnal leftovers were employed. BSF protein content was mainly correlated to fibre and protein content in the diet. Among amino acids, lysine, valine and leucine were most affected by the diet. Essential amino acids satisfied the Food and Agricultural Organization (FAO) requirements for human nutrition, except for lysine in few cases. BSF could be a flexible tool to bio-convert a wide range of vegetable by-products of different seasonality in a high-quality protein-rich biomass, even if significant differences in the protein fraction were observed according to the rearing substrate.
Keywords: Hermetia illucens; insect rearing; vegetable leftovers; protein fraction; amino acids composition; growth substrate
1. Introduction
By 2050, the world will host more than 9 billion people, and the availability of proteins is the main concern in feeding the increasing population, as already pointed out as long ago as 1975 by Meyer-Rochow [1]. Presently, in Western diets the proteins are predominantly introduced through products of animal origin, although the zootechnical production constitutes an important issue from an ecological point of view due to its impact on the environment [2]. A valuable and more sustainable source of protein is represented by insects. They have a high nutritional value [3–5] and, compared with traditionally farmed animals, insects have a much higher conversion efficiency and require much less water [2], and their rearing involves much less greenhouse gas emissions [6].
Nowadays, more than one-third of the edible parts of food produced gets lost or wasted; insects could bear an important role in managing and valorizing food waste, since they can be reared on a large variety of bio-waste substrates [7,8], thus contributing to their mass reduction and preventing unnecessary waste of resources and further emissions of
greenhouse gas [9]. Therefore, insects fit perfectly in the perspective to valorise bio-waste to create a sustainable food and feed production system that embraces the concept of circular economy and increased sustainability [10,11]. The ability of insects to convert waste materials into high-quality nutrients has long been known [12], and the Black Soldier Fly (BSF, Hermetia illucens L., Diptera, Stratiomyidae) is known as one of the most efficient bioconverters among insects, being able to reduce the weight of organic waste up to 75% and converting it into a biomass rich in proteins and lipids [13]. This makes the BSF suitable to be used as feed for farmed animals [14], for biodiesel production [15] and also for cosmetics or pharmaceuticals industries, thanks to its high chitin content [16]. All these applications could be simultaneously tackled through an appropriate fractionation method [17]. Last but not least, the residual larval frass can be employed as a quality soil improver [18,19].
Many studies have shown that BSF larvae can live on many substrates with different characteristics, ranging from mushrooms [20] and winery by-products [21], restaurant waste [8], municipal waste [22], animal manure [10,23] and human faeces [24]. However, according to European legislation (Regulation (EC) No. 1069/2009), when invertebrates are industrially reared, they are considered as “farmed animals”; therefore, the use of animal manure, catering waste or former foodstuffs containing meat and fish as feeding substrates is totally forbidden [25,26]. As a consequence, despite the lower content of nutrients, vegetable and fruit leftovers are increasingly used as rearing substrates for BSF, also due to their high availability in industrialized regions that are fully compliant with the legislation on feed for farmed animals [9]. European Food Safety Authority (EFSA) highlighted that no additional microbiological risks are expected for insect rearing on authorized substrates with respect to other animal farming [26]. In addition, a few studies demonstrated that the eventual pesticides and mycotoxins do not accumulate in BSF biomass after the breeding process [27–30].
An important point in the use of vegetal substrates as feed for insects is their seasonal availability. As a matter of fact, most of the vegetal leftovers are not constantly available throughout the year, thus it is important, both from the economic and technical point of view, to understand the feasibility of rearing BSF on a variety of substrates and a combination thereof, in order to have a constant BSF production and composition throughout the year.
Many authors have studied the influence of rearing substrates on BSF protein fractions in terms of total amount and of amino acid profile. The protein amount showed marked differences, varying from 32 to 58% on dry matter [13], while for the amino acid composition, studies in the literature are often inconsistent, and this is true for other insect species as well [31]. According to Newton et al., BSF larvae reared on animal manure were lacking some essential amino acids, such as cysteine, methionine and threonine [12], whereas according to Liland et al., BSF larvae had sufficient threonine amounts when reared on a conventional diet supplemented with increasing quantities of seaweed [32]. When fed on a vegetable mix, BSF showed high amounts of aspartic acid, glutamate and arginine [7], but also of leucine [8].
In this paper, a wide range of data was collected on the content and quality of BSF prepupae protein in relation to the rearing substrates used. A total of 49 different combinations of vegetable rearing substrates were tested in order to obtain the most complete picture of the effect of diet on BSF proteins and on the possibility of using diverse substrates for their rearing. All these vegetable by-products have been chosen focusing on their availability in a specific region (Regione Emilia-Romagna, Italy) and mostly on their seasonal availability. The diets were formulated according to a Design of Experiment approach (Mixture Design) based on the nutrient composition of each substrate and their seasonal availability, as reported in Barbi et al., in the optic of a possible scale-up application of BSF rearing on vegetable by-products across the whole year [33]. To put this approach into practice and fulfil the requirements of a circular economy, it is of primary importance to verify the composition of BSF reared on the different substrates, and in particular their protein fraction that nowadays is considered as the most valuable component of insects.
2. Materials and Methods
2.1. Materials
The vegetable by-products were collected from different suppliers in the Emilia Romagna Region (northern Italy): Agribologna (Bologna, Italy), Conserve Italia (Bologna, Italy) and Cooperativa Agricola Brisighellesi (Ravenna, Italy). The by-products were stored at −20 °C until use, and further grinded with IKA A10 laboratory grinder.
Chemicals: an AccQ-Fluor reagent kit was obtained from Waters (Milford, MA, USA). DL-norleucine, amino acid standard mixture, L-tryptophan, 5-Methyl-DL-tryptophan, L-Cysteic acid, DL-Dithiothreitol, Ethylenediaminetetraacetic acid (EDTA) and Tris-HCl were purchased from Sigma-Aldrich (St. Louis, MO, USA). Sodium dodecyl sulfate was purchased from Biorad (Hercules, CA, USA). All the other solvents, salts, acids and bases were of analytical grade and were purchased from Sigma-Aldrich or Carlo Erba (Milan, Italy).
2.2. Insect Rearing Conditions and Substrate Selection
BSF prepupae for the analyses were obtained from a series of experiments carried out in the Applied Entomology laboratory of the University of Modena and Reggio Emilia (Italy). Leftovers selection, rearing substrates design and larvae rearing experiments have been managed by the University of Modena and Reggio Emilia (Italy), as discussed in another study [33]. Essentially, BSF larvae were reared under controlled conditions (27 ± 0.5 °C, 60–70% Relative Humidity, RH) in various mixtures of fruit and vegetable by-products with a different seasonal availability (All-year, Summer and Autumn, Table S1), and the mixtures were designed through a Mixture Design approach. Details on the chemical composition of the substrates are reported in Table S1. The “Gainesville House Fly” diet (50% wheat bran, 30% alfalfa meal and 20% corn meal) [34] was used as the control diet (CTR). Forty-nine experiments were conducted, inoculating exactly 300 BSF larvae (of second and third stage) for each substrate. The rearing experiments was conducted for a maximum period of 65 days, checking regularly the prepupae developed. At each control, the prepupae observed were collected, killed by freezing and stored at −20 °C until further analysis.
2.3. Proximate Composition of Rearing Substrates and BSF Prepupae
Proximate composition of agri-food leftovers was determined using standard procedures [35]. Moisture was determined in an oven at 105 °C for 24 h. Crude fat content was determined using an automatized Soxhlet extractor (SER 148/3 VELP SCIENTIFICA, Usmate Velate, Italy) using diethyl ether. Total ash was determined after mineralization at 550 °C for 5 h. Total nitrogen was determined with a Kjeldahl system (DKL heating digestor and UDK 139 semiautomatic distillation unit, VELP SCIENTIFICA) using 6.25 as the mean nitrogen coefficient conversion for vegetable rearing substrates. Dietary fibres of vegetable by-products were determined through the official method AOAC 991.43, while total polyphenol content was determined by using the Folin–Ciocalteu method. Digestible carbohydrates were determined by difference.
Regarding BSF prepupae, the total protein content was determined by the Kjeldahl method, utilizing 4.67 as the nitrogen to protein conversion factor, in order to exclude chitin contributing to total nitrogen, as previously reported [36].
2.4. Amino Acid Profile of BSF Prepupae
2.4.1. Sample Preparation
The total amino acid profile was evaluated according to the protocol proposed by Caligiani et al. with some modifications [17]. An amount of 0.5 g of BSF prepupae was hydrolysed with 6 mL of HCl 6 N at 110 °C for 23 h, then the internal standard (7.5 mL of 5 mM Norleucine in HCl 0.1 M) was added. Cysteine was determined as cysteic acid after performic acid oxidation followed by acid hydrolysis. In this case, an amount of 0.5 g of BSF was added to performic acid freshly prepared (by mixing 9 volumes of formic acid with 1 volume of hydrogen peroxide) and samples were kept in an ice bath for 16 h at 0 °C.
Then, 0.3 mL of hydrobromic acid was added and the bromine formed was removed under nitrogen flow. Then, acid hydrolysis was performed as described above.
2.4.2. UPLC/ESI-MS Analysis
The hydrolysed samples were analysed by ultra-performance liquid chromatography with electrospray ionization and mass spectrometry detector (UPLC/ESI-MS, WATERS ACQUITY) after derivatization with 6-aminoquinolyl-N-hydroxysuccinimidyl carbamate (AQC). In particular, UPLC/ESI-MS analysis was performed by using an ACQUITYU-PLC separation system with an Acquity BEH C18 column (1.7 µm, 2.1 × 150 mm). The mobile phase was composed of H₂O + 0.2% CH₃CN +0.1% HCOOH (eluent A) and CH₃CN + 0.1% HCOOH (eluent B). Gradient elution was performed: isocratic 100% A for 7 min, from 100% A to 75.6% A and 24.4% B by linear gradient from 8 to 28 min, isocratic 100% B from 29 to 32 min, isocratic 100% A from 33 to 45 min. The flow rate was set at 0.25 mL/min, injection volume 2 µL, column temperature 35 °C and sample temperature 18 °C. Detection was performed by using Waters SQ mass spectrometer: the ESI source was in positive ionization mode, capillary voltage 3.2 kV, cone voltage 30 V, source temperature 150 °C, desolvation temperature 300 °C, cone gasflow (N2) 100 L/h, desolvation gas flow (N2) 650 L/h, full scan acquisition (270–518 m/z) and scan duration 1 s. Calibration was performed with standard solutions prepared mixing norleucine, amino acids hydrolysate standard mixture, cysteic acid and deionized water.
2.4.3. Tryptophan Determination
Total tryptophan was determined following the protocol proposed by Delgado-Andrade et al. with some modifications [37]. An amount of 0.2 g of sample was weighed and dissolved in 3 mL of 4N NaOH. An amount of 150 µL of 5-methyl-tryptophan (16 mg/100 mL), used as internal standard, was added and mixed. Hydrolysis was then carried out for 18 h at 110 °C. The hydrolysates were aerated and cooled, then carefully acidified to a pH 6.5 with HCl 37%, diluted to 25 mL with sodium borate buffer (0.1 M, pH 9.0) and allowed to stand for 15 min. Samples were finally centrifuged at 4000 rpm for 5 min and supernatant filtered through a 0.45 µm nylon filter membrane into UPLC vials. UPLC/ESI-MS analysis was performed as for the other amino acids.
2.5. Statistical Analysis
All analyses on prepupae were carried out in duplicate. Data are expressed as the mean ± standard deviation. Protein and amino acid data were subjected to one-way analysis of variance (ANOVA) followed by a Tukey post hoc test using IBM SPSS software version 21.0 (SPSS Inc., Chicago, IL, USA) to determine differences between samples. Significant differences were compared at a level of p < 0.05.
The Pearson correlation coefficient was calculated to evaluate the linear correlations between the nutrients in rearing substrates and protein amounts in BSF prepupae biomass. Both values were taken in absolute terms, calculated according to the following equations:
$$\text{BSF total protein content} = \frac{\text{Protein\%} \times \text{total grams of prepupae at the end of the breeding}}{100}$$
$$\text{Leftovers total nutrient content} = \frac{\text{nutrient\% in diet} \times \text{total grams of diet}}{100}$$
The second equation was applied separately for each nutrient present in the diet: proteins, lipids, fibres, carbohydrates, ashes and polyphenols.
Principal Component Analysis (PCA) was carried out on amino acid data using IBM SPSS software version 21.0 (SPSS Inc., Chicago, IL, USA). PCA was performed through a 52-point matrix (49 samples plus 3 replicates of the control substrate, one for each seasonal period) and 19 variables (18 amino acids and total protein content), and the principal components were derived with the correlation method. As an unsupervised learning approach, PCA allows us to describe the variation in the dataset. Data were visualised by
plotting the score plot and the loading plot, the latter allowing the identification of the amino acids having influences on specific grouping of samples according to the rearing substrate.
The Design Expert 12.0 (Stat-Ease) code was used both to set up the experimental plan and to analyse the results. A mixture design was selected to obtain predictive reliability on the effect of the leftovers’ composition on the amino acid content of BSF’s prepupae. Six factors were considered: proteins, fibres, carbohydrates, polyphenols, lipids, and ashes, from which the experimental plan of Table S1 was derived. ANOVA was employed to estimate the influence of each factor over the responses observing the \( p \)-values (\( \alpha \)-level of 0.05) and F-tests. The quality of the fit in terms of regression analysis and the predictive power of the model were evaluated by using the R2 and Pred-R2, respectively. R2 is the proportion of the dependent variable’s variance predictable from the independent variables. In a similar way, Pred-R2 shows how well the model can predict the responses for new observations. Response contour plots were used as functional tools in explaining graphically the role of the main components on the final considered properties [38].
3. Results and Discussion
3.1. Rearing Substrate Composition and Related BSF Biomass
The composition of rearing substrates, both in terms of leftover combinations and specific nutrient composition is reported in Table S1, while in Table 1, the mean composition of the diets according to the seasonality (substrates available all year, substrates available in summer and substrates available in autumn) is reported and compared with the BSF standard diet (Gainsville diet, CTR). The three groups of tested diets differed between each other and in respect to the control diet. The differences in diet compositions are related to the vegetable by-products employed: (i) fruit peels and pulp (exotic fruits, apple, kiwi, pineapple, melon) for the All-Year diets, (ii) tomato peels/seed and peach pulp/peels for the Summer group and (iii) legume, corn and olive pomace for the Autumn group, although the exotic fruits and melon, as substrates that are available all year round, were introduced in some diets of the AUTUMN and SUMMER groups. Corn, wheat brans and alfalfa were instead the ingredients of the control diet (CTR diet), used as the BSF rearing substrate in the lab colony.
Table 1. Mean percentage composition of the different diets, expressed as g/100 g dry matter and corresponding amount of total prepupal biomasses obtained (g) starting from 300 Black Soldier Fly (BSF) larvae. Nd = not detected.
| | All-Year (n = 21) | Summer (n = 13) | Autumn (n = 15) | CTR |
|-------------------|------------------|-----------------|-----------------|--------------|
| Lipid | 0.78 ± 0.35 | 3.28 ± 2.12 | 3.18 ± 1.46 | 3.37 ± 0.1 |
| Ashes | 6.66 ± 3.20 | 4.87 ± 0.71 | 4.02 ± 1.59 | 5.9 ± 0.5 |
| Fibres | 23.1 ± 17.51 | 56.86 ± 14.79 | 56.69 ± 6.53 | 41 ± 2.5 |
| Polyphenols | 0.03 ± 0.02 | 0.09 ± 0.02 | 0.04 ± 0.01 | Nd |
| Protein | 5.07 ± 1.20 | 9.97 ± 2.71 | 16.03 ± 6.57 | 17.28 ± 1 |
| Available carbohydrates | 64.36 ± 20.07 | 24.93 ± 15.61 | 17.05 ± 10.72 | 32.61 ± 1 |
| Total prepupae biomass (g) | 21.2 ± 2 | 39.8 ± 3 | 57.1 ± 5 | 52.0 ± 4 |
The different ingredients employed in the three groups are also related to a different proximate composition of the diets, as reported in Table 1. The All-Year diets are rich in available carbohydrates and poor in protein and lipids, while the Summer and Autumn groups contain larger amounts of fibres and protein, the latter especially represented in the Autumn group.
Despite differences in composition, all the diets tested allowed BSF to grow (more than 90% of the initial 300 BSF larvae reached the prepupal stage in all experiments), even if with consistent differences in total weight of final prepupae biomass. In Table 1, the mean amount of prepupae (g) obtained for each different rearing mixture is also reported, clearly outlining the prominent effect of seasonality on the BSF prepupae total biomass amount. The specific amount of prepupal biomasses in each rearing experiment is reported in Table S1 of the Supplementary Materials.
As a consequence, the BSF prepupae obtained can be roughly divided into three groups according to their biomass weight at the end of the growing process, which basically correspond to the three groups of vegetable diets administered, classified according to their seasonality. Indeed, the lower quantity of prepupal biomass were obtained by using the All-Year substrates, the intermediate weight with Summer substrates (both of them lower than when using the control diet), while the higher quantities were obtained by employing the Autumn group substrates, which seems to be the most suitable feeding substrate group to maximize the BSF biomass.
3.2. Effect of the Rearing Substrate Composition on BSF Prepupae Total Protein Content
Given that the composition of the diets influences the growth of the prepupae, the target of this work, as previously stated, was to verify how the different substrates, and in turn the different growth performance of prepupae (low, medium and high biomass-weights), influence the final BSF protein quality and its total amount. BSF samples were clustered on the basis of the prepupal biomass, corresponding to the three groups of BSF diets (All-Year, Summer and Autumn).
The total protein content of each BSF prepupa reared on the different substrates is reported in Figure 1 (see Table S1 of the supplementary material for the details about substrate typologies, composition and sample codes). Protein amount, calculated as g/100 g of BSF on dry matter (% DM), varied in the range of 35% to 49.5%. These values agree with the studies in the literature, reporting a range of 32–58% [13].
Figure 1. Total protein content (Nx4.67, g/100 g DM) for each BSF sample grown on the 49 diets considered and compared with the control samples (dotted line). Each datum is the mean of two replicates. Global mean of each group was compared with one-way ANOVA (p-level 0.05). Different letters (a,b) on the bars indicate significantly different values.
In general, most of the samples from the All-Year and Summer groups contain lower amounts of protein compared to the control group, although with a few exceptions (Sample N—All-Year, sample A, M, G—Summer). On the contrary, many of the Autumn group samples contain higher amounts of protein with respect to the control, indicating that a better growth performance, as expected, also corresponds to a higher protein content.
Taken as a whole (mean), the Autumn group did not show a significant difference with respect to the control, outlining that the agrifood leftovers used in this group could fully replace the control diet, while in the All-Year and the Summer groups, a significantly lower amount of protein was recorded (one-way ANOVA, Tukey post hoc test, p-level = 0.05). This indicates that, while growing on these substrates is certainly possible, it would be less efficient and result in a smaller overall amount of protein. Thus, growing on these substrates should be performed by carefully balancing the advantage of re-using leftovers with the disadvantage of having a slightly lower amount of protein.
To better understand which relationship exists between the various diets administered and the production of protein in the prepupae biomass, we correlated the latter and the single nutrients of the diets through linear correlation analysis.
A positive, although moderate, correlation was observed between the lipid content of the diet and the protein content of prepupae (R = 0.54; p < 0.001), while non-structural
carbohydrates (digestible carbohydrates) showed a moderate negative one (R = −0.56, p < 0.001). On the contrary, polyphenols (R = 0.39, p < 0.001) and ashes (R = 0.17, p = 0.37) in the diet did not affect the protein content of BSF prepupae.
A significant positive correlation was detected between the BSF prepupae protein content and the fibres content of the diet (R = 0.87, p < 0.001, Figure 2a). Fibres are typically difficult degrade for most insects, but this positive correlation could be the result of the increased availability of nutrients due to the action of microorganisms able to hydrolyse cellulose, both endogenous and exogenous (i.e., already present in the rearing substrates) [21,39–41]. Indeed, previous studies have identified cellulase genes in the gut microflora of BSF larvae [40,42]. Thus, a likely hypothesis for the correlation found might be that the ability of using more and better fibre biomass leads to better larval growth, which in turn also means a higher amount of protein produced.
A positive correlation was also identified between the BSF prepupae protein content and the diet’s protein content in absolute terms (R = 0.84, p < 0.001, Figure 2b), confirming previous findings [21,23].
Figure 2. Correlation between total protein content (absolute amount) in the BSF prepupae biomass recorded in each experiment (starting from 300 BSF larvae) and (a) total fibre and (b) total protein contained in the rearing substrate (absolute amount).
Observing qualitatively the data in Figure 2a,b, a lack of a linear correlation (a plateau is reached) seems evident, especially when a high amount of fibre or protein was present in the diet, suggesting that there is a specific level of these nutrients in the substrates allowing them to reach the maximum amount of protein in insects.
More specifically, our data show that by increasing the amount of vegetable protein in the diet, BSF larvae convert progressively a smaller part of it into their own animal protein. In fact, whilst for low-protein diets they were able to convert more than 90%, for the most protein-rich diets, this percentage dropped to 10% (Figure 3). This suggests that for BSF larvae growth and protein content, the amount of protein in the rearing substrate is very important until a minimum threshold is reached, and then it becomes less relevant. These experimental data are in agreement with a recent work about digestive enzyme expression and production in BSF [43]. In this paper, the authors clearly demonstrate that the midgut of H. illucens larvae is able to adapt to diets with different nutrient content; an increase in proteolytic activity together with a decrease in α-amylase and lipase activity was
observed as a consequence of nutritionally poor diet. Moreover, Barragán Fonseca (2018) demonstrated that larvae feeding on substrates rich in protein have a higher lipid content, and thereby a reduced protein content (in % dry mass) [44]. Consequently, in order to obtain prepupae with a high protein content, according to Figure 3, a good compromise to maintain an advantageous conversion rate would be to rear them on a substrate containing 7% by weight of protein. This value is also in accordance with a previous study [31].
Figure 3. Conversion rate (%) of vegetable protein present in the substrate into BSF protein (absolute amounts, starting from 300 BSF larvae).
3.3. BSF Amino Acid Content
To better verify the influence of the rearing substrate composition on the BSF protein nutritional value, further insights into the complete amino acid profile were provided. As a matter of fact, information on the nutritional quality of proteins, and therefore on their amino acid composition, is of utmost importance to understand the possible uses of the BSF protein fraction. Results on the complete total amino acid profile of the BSF prepupae are reported in Table S2. An explorative Principal Component Analysis (PCA) was performed to assess if and how different rearing substrates would affect amino acid composition. Principal Components are new variables obtained by linear combinations of the original variables (amino acids and total protein), allowing us to describe the system variability using only a few components, thus reducing the complexity, enabling the visualization of the samples in a two-dimensional graph. The analysis showed that about 39% of the total variation is explained by the first component (PC1), 57% by the first two components and 69% by the first three components. The most important variables for each principal component are reported in Table S3 of the Supplementary Materials. Figure 4a shows the scatter plot of the scores of PC1 versus PC2. The loadings for the first two components are schematized in the component plot (Figure 4b). Glutamic acid/glutamine, lysine, phenylalanine, aspartic acid/asparagine, tyrosine, glycine, valine, serine and histidine turned out to be the most influencing variables for PC1, PC2 was predominantly characterized by cysteine, tryptophan, arginine and threonine, while leucine, isoleucine and methionine had the greatest effect for PC3. Figure 4a shows a partial separation of the BSF prepupae samples, based on the seasonality of the substrates. In particular, BSF prepupae that had been reared on substrates belonging to the Autumn group are found in correspondence with PC1 positive values, well separated from the others. In this group, the most represented essential amino acids are phenylalanine/tyrosine, valine and leucine.
On the other hand, the All-Year and Summer groups showed a less clear separation between each other and both were found at negative values of PC1. They differed from the Autumn group mainly in their higher amounts of lysine, aspartate and glutamate content. Finally, the control group was found in an intermediate position, closer to the All-Year and Summer groups and with greater differences compared to the Autumn one.
In order to verify to what extent the nutritional properties of BSF prepupae proteins were affected by the composition of the rearing substrates, a one-way ANOVA was carried out on the essential amino acids (EAAs), dividing the samples according to the four groups of substrates (Table 2).
As a general consideration, ANOVA results confirm what is evidenced by PCA; in fact, the Autumn group is the one presenting the larger number of significant differences with respect to the other experimental diets (All-Year and Summer), and also with respect to the control diet. The BSF prepupae of the Autumn group contain the highest amounts of essential amino acids, except for lysine, which was detected in significant lower amounts compared to the BSF reared on the other substrates.
Isoleucine, methionine, cysteine and tryptophan did not differ significantly in any of the four diets.
On the other hand, specific differences in essential amino acid content of prepupae reared on the three diets were observed. Leucine turned out to be the EAA most affected from the diet, highlighting a sharp decline when samples of the All-Year group were considered. A similar trend was also observed for valine and histidine. Phenylalanine
and tyrosine were lower in the All-Year and Summer groups when compared with the Autumn group, while threonine in the Summer group was present in a lower amount with respect to the Autumn group. Actually, several peculiar differences were also observed in the single samples (Table S2).
Table 2. Mean values (expressed as mg/g protein) for essential and semi-essential amino acids of BSF prepupae that had been reared on the different groups of substrates.
| | All-Year (n = 21) | Summer (n = 13) | Autumn (n = 15) | CTR (n = 3) |
|----------|------------------|-----------------|-----------------|------------|
| His | 33.50 a | 37.86 ab | 39.00 b | 37.00 ab |
| Ile | 42.89 a | 43.11 a | 44.10 a | 45.39 a |
| Leu | 71.24 a | 77.66 bc | 79.88 c | 74.00 ab |
| Val | 60.49 a | 60.25 a | 66.09 b | 62.43 a |
| Lys | 62.22 a | 62.93 a | 51.73 b | 62.44 a |
| Cys | 18.17 a | 20.42 a | 20.14 a | 20.70 a |
| Met | 18.37 a | 17.54 a | 19.16 a | 18.40 a |
| Phe | 41.69 a | 40.03 a | 46.47 b | 42.53 ab |
| Tyr | 61.45 a | 63.16 ab | 69.67 b | 64.13 ab |
| Thr | 39.50 ab | 37.36 a | 40.58 b | 39.30 ab |
| Trp | 14.80 a | 15.76 a | 15.84 a | 17.60 a |
Values followed by different letters within one row are significantly different (one-way ANOVA, Tukey post hoc test, p < 0.05).
Our results suggested a partial influence of the BSF diet on the total protein production and on their amino acid composition, and similar findings were also obtained by Spranghers et al. [8] and Soetemans et al. [45] on the tenebrionid Alphitobius diaperinus. However, it is not yet clear how BSF larvae convert the amino acids from the diet into amino acids useful for their metabolism and how they store and use differently the different essential amino acids. According to Liland et al. [31], BSF larvae were able to produce certain amino acids, such as tyrosine, which were almost absent in the seaweed-containing media.
Due to the complexity of the system and the possible interaction of diet components on specific essential amino acid content, a multivariate statistical analysis was used as further approach to better understand the influence of the diet composition (Table 1) on the essential amino acid profile. This approach was possible because the experimental diets were formulated according to a Design of Experiments (DoE) [32], allowing us to construct statistically reliable models describing the correlation among food leftovers’ composition and amino acid content, and utilizing ANOVA to verify if the effects of the main factors and their interaction terms are statistically reliable. Results of ANOVA (Table S4) indicate the influence of many significant factors for each response, thus confirming the complex nature of these correlations. These relationships are in general better explained by the interaction between the factors rather than by an independent single factor, confirming the need for the multivariate approach. Furthermore, the fitting parameters show a fairly good fitting (R^2 > 0.50) only for some responses, in particular for leucine, valine and lysine (Table S4). The graphical representation of the results with acceptable fitting quality is reported in Figures 5 and 6. According to these results, for the leucine content in the BSF prepupae (Figure 5), a strong interaction emerges with the content of lipids and proteins in the rearing substrate. In particular, when the rearing diets were lacking carbohydrates, the highest content of leucine was detected when the content of proteins and lipids in the substrate was equal to or above 50 wt% (Figure 5a). In the presence of a higher content of carbohydrates (50 wt%), the highest content of leucine in BSF prepupae is obtained with slightly lower contents of both proteins and lipids in the rearing diet (Figure 5b). Overall, the increase in
carbohydrate content leads to a reduction in the red area, thus indicating that a smaller combination of rearing substrates is suitable for an increase in the leucine content of the BSF prepupae. In conclusion, the highest leucine content in the BSF prepupae can be achieved by increasing the amounts of protein and lipids while reducing the carbohydrates content in the rearing diet. Similarly (data not shown), the valine content in BSF prepupae is enhanced by maximising the content of lipids and proteins and reducing that of carbohydrates in the rearing substrate. The correlation between the content of leucine and valine in the BSF prepupae with the protein content of the rearing substrates could indicate that these amino acids are essential for BSF development.
Finally, the contour plots related to the content of lysine in BSF prepupae (Figure 6) show that carbohydrate variation plays a crucial role in enhancing this amino acid content. In fact, the highest content of lysine in the insects can only be obtained when the rearing substrate has at least 90% DM of carbohydrates (Figure 6b) and a slightly higher amount of lipids compared to protein and fibre. A rearing substrate based on 75% DM of carbohydrates (Figure 6a) will result in prepupae with lower amounts of lysine. In any case, to obtain average values of lysine in the BSF prepupae, in both situations a well-balanced amount of protein and fibre in the rearing substrate is needed.
As a general consideration, it is important to highlight that none of the BSF prepupae reared on experimental diets contained significantly lower amounts of essential amino acids with respect to the control group, except in the above-mentioned case of lysine. Despite the specific differences, the majority of the proteins obtained from the BSF prepupae reared on the experimental substrates contain on average a sufficient quantity of each EAA required for human consumption, as shown in Table 3. Recent studies have shown that BSF prepupae contain optimal amounts of all the essential amino acids (EAA) to satisfy the human adult requirements, as established in the reference of FAO/WHO [7,17,46].
Figure 5. Graphical model variation of the amount of leucine in the proteins of BSF prepupae in relation to the composition of the rearing substrate in terms of lipids, proteins and fibres, considering two scenarios: (a) carbohydrates = 0 wt%; (b) carbohydrates = 50 wt%. The region representing the highest value of the response is shown in red colour whereas the lowest values are in blue. For each response, the most significant factor has been considered for the graphical model, expecting a higher variation in the response behaviour.
Figure 6. Graphical model variation of the amount of lysine in the proteins of BSF prepupae in relation to the composition of the rearing substrate in terms of lipids, protein and fibre, considering two scenarios: (a) carbohydrates = 75 wt%; (b) carbohydrates = 90 wt%. The region representing the highest value of the response is shown in red colour, whereas the lowest values are in blue. For each response, the most significant factor has been considered for the graphical model, expecting higher variation in the response behaviour.
Table 3. Highest, lowest and average essential amino acid content of BSF prepupae, compared with the FAO protein reference for human adults (2011).
| Reference Protein FAO 2011 (mg/g Protein) | Average Values Found in BSF Prepupae Protein (mg/g Protein) | Lowest Values Found in BSF Prepupae Protein (mg/g Protein) | Highest Values Found in BSF Prepupae Protein (mg/g Protein) |
|------------------------------------------|----------------------------------------------------------|----------------------------------------------------------|----------------------------------------------------------|
| His | 16 | 36 | 46 |
| Ile | 30 | 44 | 50 |
| Leu | 61 | 76 | 89 |
| Lys | 48 | 58 | 72 |
| Cys + Met | 23 | 38 | 45 |
| Phe + Tyr | 41 | 107 | 127 |
| Thr | 25 | 39 | 44 |
| Trp | 6.6 | 15 | 19 |
| Val | 40 | 62 | 72 |
According to these figures, all the analysed BSF prepupae can meet the FAO requirements, and, as in the case of histidine and tryptophan, the EAA content is even double the required amount. Moreover, the lowest values found in the samples turned out to be higher than the recommended amounts, except for lysine, which resulted in being at the lower limit in few cases (two of the BSF samples reared on the Autumn substrates). This finding is particularly relevant, considering that the analysed prepupae samples did not undergo any thermal treatment, which, on the other hand, is very likely to occur when BSF proteins are to be used in food and feed formulations, thus further lowering the lysine amount through the Maillard reaction. Moreover, one factor that could affect the lysine content is the killing method for BSF prepupae [47]. Killing by freezing, which was the method used in this experimental plan, leads to some alteration of the total amino acid fraction, with the notable decrease of lysine and cysteine, likely involved in the process of melanisation, reacting with quinones with their side chain.
4. Conclusions
Due to its physiological characteristics and excellent nutritional properties, the BSF is considered one of the most promising candidates in insect farming for feed and food purposes. Aiming at evaluating the effect of the nutrients of the rearing substrate on the protein content and composition of BSF prepupae, we examined 49 different substrates that consisted of variable proportions of different vegetable by-products and were divided into three groups according to their seasonal availability. The results showed that the total protein content of BSF prepupae ranged between 35% and 49% DM, with the highest values in the Autumn group substrates. It has also been observed that a higher protein content in the diet has resulted in a higher prepupae protein content up to a certain value; this vegetable-to-animal protein conversion lowered gradually as dietary proteins were increased, indicating the existence of a minimal critical amount of protein that has to be present in the diet, and that once reached makes any further protein addition to the rearing substrate much less relevant. Dietary fibres also seem to play a positive role in the achievement of BSF prepupae protein biomass. An important outcome of this work focuses on the essential amino acids. Higher amounts of essential amino acids in the BSF prepupae that had been reared with the substrates of the Autumn group were observed. Lysine, leucine and valine were found to be the most correlated with the presence of nutrients of the feeding diet. Leucine and valine were strongly dependent on the content of protein and lipid in the diet, while lysine is correlated to the amount of carbohydrates. Despite these important differences, the essential amino acid composition almost always fully satisfied the FAO nutritional requirements for humans. The only exception was lysine, and only in a very limited number of cases.
In conclusion, this study shows that by providing BSF larvae with substrates based on a very wide range of combinations of vegetable by-products, it is possible to obtain in the BSF prepupae a protein quality very similar to the one obtained with the control diet. However, when the employed leftovers have a very low-quality nutritional content, the development of BSF biomass is less efficient, and as a consequence a lower amount of protein is obtained. Thus, growing on these substrates should be performed by carefully balancing the advantage of re-using leftovers with the disadvantage of having a slightly lower amount of protein. These findings are very important in view of promoting BSF as a flexible tool able to bio-convert a wide range of vegetable by-products, which can vary according to seasonality or areas of production.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10.3390/foods10081773/s1, Table S1: Formulation and proximate composition of the rearing substrates, and total weight of corresponding prepupae biomass, Table S2: Complete amino acid composition (mg AA/g protein) of BSF prepupae grown on different diets, Table S3: PCA loadings, Table S4: Summary of ANOVA results.
Author Contributions: Conceptualization, A.C., S.S., L.M. and M.M.; methodology, A.F., S.B., L.I.M., L.M., M.M. and A.C.; software, S.B.; formal Analysis, A.F. and S.B.; investigation, A.F., L.I.M. and A.V.L.; writing—original draft preparation, A.F. and A.C.; writing—review and editing, L.M., M.M. and S.S.; supervision, A.C.; project administration, L.M., M.M. and A.C. All authors have read and agreed to the published version of the manuscript.
Funding: This work was supported by the Emilia-Romagna Region within the Rural Development Plan 2014–2020 Op. 16.1.01—GO EIP-Agri—FA 5C, Pr. “BIOECO-FLIES”.
Data Availability Statement: The data presented in this study are available here and in the Supplementary Material.
Acknowledgments: The authors are grateful to Andrea Polacco for helping in laboratory analyses.
Conflicts of Interest: The authors declare that they have no known competing financial interest or personal relationship that could have appeared to influence the work reported in this paper.
References
1. Meyer-Rochow, V.B. Can insects help to ease the problem of world food shortage? Search 1975, 6, 261–262.
2. van Huis, A.; van Itterbeeck, J.; Klunder, H.; Mertens, E.; Halloran, A.; Muir, G.; Vantomme, P. Edible Insects: Future Prospects for Food and Feed Security; Food and agriculture Organization of the United Nation: Rome, Italy, 2013; ISBN 978-92-5-107596-8.
3. Rumpold, B.A.; Schlüter, O.K. Nutritional composition and safety aspects of edible insects. Mol. Nutr. Food Res. 2013, 57, 802–823. [CrossRef] [PubMed]
4. Chen, X.; Feng, Y.; Zhang, H. Review of the nutrition value of edible insects. In Proceedings of the Forest Insects as Food: Humans Bite Back, Chiang Mai, Thailand, 19–21 February 2008.
5. Bessa, L.W.; Pieterse, E.; Marais, J.; Hoffman, L.C. Why for feed and not for human consumption? The black soldier fly larvae. CRSFS 2020, 19, 2747–2763. [CrossRef]
6. Oonincx, D.G.A.B.; van Itterbeeck, J.; Heetkamp, M.J.W.; van den Brand, H.; van Huis, A. An exploration on greenhouse gas and ammonia production by insect species suitable for animal or human consumption. PLoS ONE 2010, 5, e14445. [CrossRef] [PubMed]
7. Cappellozza, S.; Leonardi, M.G.; Savoldelli, S.; Carminati, D.; Rizzolo, A.; Cortellino, G.; Terova, G.; Moretto, E.; Badaile, A.; Concheri, G.; et al. A first attempt to produce proteins from insects by means of a circular economy. Animals 2019, 9, 278. [CrossRef] [PubMed]
8. Spranghers, T.; Ottoboni, M.; Klootwijk, C.; Ovyn, A.; Deboosere, S.; De Meulenaer, B.; Michiels, J.; Eeckhout, M.; De Clercq, P.; et al. A first attempt to produce proteins from insects by means of a circular economy. Animals 2019, 9, 278. [CrossRef] [PubMed]
9. Gustavsson, J.; Cederberg, C.; Sonesson, U.; van Otterdijk, R.; Meybeck, A. Global Food Losses and Food Waste: Extent, Causes and Prevention; Food and Agriculture Organization of the United Nations: Rome, Italy, 2011.
10. Bortolini, S.; Macavei, L.I.; Hadj Saadoun, J.; Foca, G.; Ulrici, A.; Bernini, F.; Malferri, D.; Setti, L.; Ronga, D.; Maistrello, L. Hermetia illucens (L.) larvae as chicken manure management tool for circular economy. J. Clean. Prod. 2020, 262, 121289. [CrossRef] [PubMed]
11. Madau, F.A.; Arru, B.; Furesi, R.; Pulina, P. Insect farming for feed and food production from a circular business model perspective. Sustainability 2020, 12, 5418. [CrossRef]
12. Newton, G.L.; Booram, C.V.; Parker, R.W.; Hale, O.M. Dried Hermetia illucens Larvae Meal as a Supplement for Swine. J. Anim. Sci. 1977, 44, 395–400. [CrossRef]
13. Gold, M.; Tomberlin, J.K.; Diener, S.; Zurbrügg, C.; Mathys, A. Decomposition of biowaste macronutrients, microbes, and chemicals in black soldier fly larval treatment: A review. Waste Manag. 2018, 82, 302–318. [CrossRef]
14. Barragan-Fonseca, K.B.; Dicke, M.; van Loon, J.J.A. Nutritional value of the black soldier fly (Hermetia illucens L.) and its suitability as animal feed—A review. J. Insects Food Feed 2017, 3, 105–120. [CrossRef]
15. Li, W.; Li, Q.; Zheng, L.; Wang, Y.; Zhang, J.; Yu, Z.; Zhang, Y. Potential biodiesel and biogas production from corncob by anaerobic fermentation and black soldier fly. Bioreour. Technol. 2015, 194, 276–282. [CrossRef] [PubMed]
16. Gortari, M.C.; Hours, R.A. Biotechnological processes for chitin recovery out of crustacean waste: A mini-review. Electron. J. Biotechnol. 2013, 16, 14. [CrossRef]
17. Caligiani, A.; Marsiglia, A.; Leni, G.; Baldassarre, S.; Maistrello, L.; Dossena, A.; Sforza, S. Composition of black soldier fly prepupae and systematic approaches for extraction and fractionation of proteins, lipids and chitin. Food Res. Int. 2018, 105, 812–820. [CrossRef]
18. Choi, Y.-C.; Choi, J.-Y.; Kim, J.-G.; Kim, M.-S.; Kim, W.-T.; Park, K.-H.; Bae, S.-W.; Jeong, G.-S. Potential Usage of Food Waste as a Feedstock in the Circular Economy. Waste Biomass Valorization 2019, 10, 265–273. [CrossRef]
19. Caligiani, A.; Marsiglia, A.; Leni, G.; Baldassarre, S.; Maistrello, L.; Dossena, A.; Sforza, S. Composition of black soldier fly prepupae and systematic approaches for extraction and fractionation of proteins, lipids and chitin. Food Res. Int. 2018, 105, 812–820. [CrossRef]
20. Choi, Y.-C.; Choi, J.-Y.; Kim, J.-G.; Kim, M.-S.; Kim, W.-T.; Park, K.-H.; Bae, S.-W.; Jeong, G.-S. Potential Usage of Food Waste as a Feedstock in the Circular Economy. Waste Biomass Valorization 2019, 10, 265–273. [CrossRef]
21. Liu, W.; Li, Q.; Zheng, L.; Wang, Y.; Zhang, J.; Yu, Z.; Zhang, Y. Potential biodiesel and biogas production from corncob by anaerobic fermentation and black soldier fly. Bioreour. Technol. 2015, 194, 276–282. [CrossRef] [PubMed]
22. Gortari, M.C.; Hours, R.A. Biotechnological processes for chitin recovery out of crustacean waste: A mini-review. Electron. J. Biotechnol. 2013, 16, 14. [CrossRef]
23. Li, W.; Li, Q.; Zheng, L.; Wang, Y.; Zhang, J.; Yu, Z.; Zhang, Y. Potential biodiesel and biogas production from corncob by anaerobic fermentation and black soldier fly. Bioreour. Technol. 2015, 194, 276–282. [CrossRef] [PubMed]
24. Newton, G.L.; Booram, C.V.; Parker, R.W.; Hale, O.M. Dried Hermetia illucens Larvae Meal as a Supplement for Swine. J. Anim. Sci. 1977, 44, 395–400. [CrossRef]
25. Newton, G.L.; Booram, C.V.; Parker, R.W.; Hale, O.M. Dried Hermetia illucens Larvae Meal as a Supplement for Swine. J. Anim. Sci. 1977, 44, 395–400. [CrossRef]
26. Newton, G.L.; Booram, C.V.; Parker, R.W.; Hale, O.M. Dried Hermetia illucens Larvae Meal as a Supplement for Swine. J. Anim. Sci. 1977, 44, 395–400. [CrossRef] [PubMed]
| 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3874,
1
],
[
3874,
8515,
2
],
[
8515,
12507,
3
],
[
12507,
16395,
4
],
[
16395,
20860,
5
],
[
20860,
24399,
6
],
[
24399,
27119,
7
],
[
27119,
30357,
8
],
[
30357,
31585,
9
],
[
31585,
35771,
10
],
[
35771,
38473,
11
],
[
38473,
42728,
12
],
[
42728,
46904,
13
],
[
46904,
52463,
14
],
[
52463,
52463,
15
]
]
} |
ab5594874a0c2a235d3d9fa09f301a0c85ddf2d6 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 6,
"total-fallback-pages": 0,
"total-input-tokens": 14963,
"total-output-tokens": 3321,
"fieldofstudy": [
"Education"
]
} | The development of thematic modules based on Strengthening Character Education (PPK) and Quick Response (QR) code
S Rahayu*, E D P Andayani, D D Chrisyarani and I Ladamay
Elementary School Teacher Education Study Program, Universitas PGRI Kanjuruhan Malang, Indonesia
*[email protected]
Abstract. Less attractive and effective learning modules in thematic learning and less reinforcement in character education make students bored and do not really implement strengthening character education in the school environment and society. Therefore, alternative teaching materials are needed in thematic learning and strengthening student character education. The purpose of this research is to develop thematic modules based on PPK and QR code. This study applied the Bog & Gall development research model. The data analysis technique used was quantitative and qualitative data analysis. The research instruments used were observation, questionnaires and documentation. The results of research on the development of thematic modules based on PPK and QR Code, it is known that the results of the analysis of media experts obtained a score of 92.86 percent with a very valid category, material experts obtained a score of 88.64 percent in the very valid category, linguists obtained a score of 75 percent with a fairly valid category, the response of educators obtained a score of 89.92 percent in the very practical category, the response of students obtained a score of 87.78 percent in the practical category. Based on the results of the explanation above, the thematic module based on PPK and QR Code can be used as a supplementary teaching material for the thematic book to make it easier for students to acquire the lesson content and apply characters education in schools and communities.
1. Introduction
A number of studies have reported on the importance of character education for students to support their future success. Numerous countries in the world have strengthened the implementation of character education in their education systems, including Indonesia. The balance between cognitive abilities and positive character of students is one of the key factors that determine student success.
By having good character, students will have the tendency to increase their academic achievement [1]. In the last decades, character education has always been considered in every educational design [2]. In addition, it also found a correlation between character education programs on students' academic and social achievement [3].
In Indonesia, the implementation of character education has been carried out intensively since the implementation of the 2004 Competency-Based Curriculum (KBK). Technically, the implementation of character education is through formal education and is included in the curriculum. In Indonesia these days, the 2013 Curriculum is the latest curriculum that has integrated character education into the learning process as stated in the Content Standards and has been simplified into five main characters.
packaged in Strengthening Character Education (PPK). Where the thematic learning system is a strategy in implementing the 2013 Curriculum.
In its implementation, teachers need to provide innovative media and teaching materials as a source of learning for students. The teaching material that is often encountered is modules. Thus, it is expected that educators can create a more innovative module, thematic modules based on printed and electronic texts to help students. The use of modules to support learning activities is important to improve mastery of the material for both teachers and students.
Learning models are currently starting to develop along with the rapid development of technology and information [4,5]. Rapid technological developments allow changes in improving the teaching process, particularly in the use of electronic teaching materials. The electronic module enables students to acquire the complex skills needed in today's era of globalization [6,7]. The advantage of the electronic module is that it can be accessed via computers, tablets and so on, which are equipped with audio-video tutorials, worksheets and evaluations [8].
The difference with the results of previous studies is that researchers focus on designing thematic modules based on PPK and Quick Response (QR) code. Quick response code is a quick response that is useful for conveying information quickly and can respond quickly too. The use of a quick response code must be accompanied by a barcode capable of storing information horizontally as well as vertically.
2. Methods
This research model applied the Borg & Gall development research model which used five out of ten stages, namely (Figure 1):
Figure 1. Research model.
However, this researcher arrived at the main product revision stage. The trial subjects in this study were ten students of grade V Elementary School. The instruments used were observation, questionnaires, and documentation. The questionnaire consisted of a student needs analysis questionnaire, a media expert questionnaire, material and language experts, educator response practitioners and students. The data analysis technique used was quantitative data and qualitative data. The results of input or suggestions from the validator were described by the researchers using qualitative or descriptive analysis. While the results of validation and student responses were analysed by the researchers by using quantitative analysis.
3. Results and discussion
Research that has been conducted by the researchers has resulted a product in the form of a thematic module based on PPK and Quick Response (QR) code. It is known that the results of the analysis of media experts obtained a score of 92.86 percent with a very valid category, material experts obtained a
score of 88.64 percent with a very valid category, linguists obtained a score of 75 percent in a moderately valid category, the response of educators obtained a score of 89.92 percent with very practical category, the response of students obtained a score of 87.78 percent with the practical category. Overall, the thematic modules based on PPK and the Quick Response (QR) code are described as follows (Table 1).
Table 1. List of thematic module visualization descriptions.
| No. | Information | Visual |
|-----|----------------------------------------------|--------|
| 1. | The cover of the Thematic Module |  |
| 2. | Table of contents |  |
| 3. | Instructions for use |  |
| 4. | Mapping of KI & KD |  |
| 5. | The limitation of each study |  |
| 6. | Theory |  |
Table 1. Cont.
| No. | Information | Visual | No. | Information | Visual |
|-----|-------------------|--------------|-----|-------------------|--------------|
| 7 | Activities |  | 11 | Answer key |  |
| 8 | Quick Response |  | 12 | Glossary |  |
| 9 | Summary |  | 13 | References |  |
| 10 | Evaluation |  | 14 | Developer Profile |  |
Based on the research results above, this study tries to develop a module based on Strengthening Character Education (PPK) and Quick Response (QR) code in the material for Theme 4 Subtheme 3 for Fifth Graders of Elementary Schools. This module is used in learning The Importance of Health which can be implemented to students at home and school. This module aims at increasing students'
understanding of The Importance of Health and to strengthen character education for students. The researchers tested the feasibility and practicality of the products developed.
The development of this thematic module is motivated by several problems including the lack of innovative teaching materials because the teacher remains utilizing the theme book from the government, the lack of teaching materials that contain subject matter as well as instilling the strengthening of character education, the minimum use of technology in the teaching and learning process.
Modules as a support for the teaching and learning process as well as an alternative interactive learning solution are feasible to be developed to improve student learning outcomes [9]. The development of learning modules is very feasible and effective to improve student learning outcomes [10]. The development of an electronic module by integrating local wisdom is very effective in the learning process [11]. Other research results state that the electronic module with a multimedia approach helps students to better understand the learning content [12].
In the era of the industrial revolution 4.0 which has shaken the progress of education, the stakeholders related to education require to be aware and able to equip their students to be mentally prepared based on knowledge and character. They must acquire new skills, adapt, manage and take advantage of the Industrial Revolution 4.0 by becoming critical thinkers, problem solvers, innovators, communicators and have a leadership spirit where all of these criteria cannot be separated from character education [13].
The concept of character education during the industrial revolution 4.0 these days, based on the obtained findings, still does not involve students to participate in. The conventional teaching method used by the teacher does not allow students to involve themselves actively in the learning process. Students, then, only play a role as a passive listener. This condition will not be effective to instill character education among students. In addition, this condition is inversely proportional to the learning theory and expectations of students, especially in the digital era like today. The involvement of students using digital and computer technology significantly helps improve students’ understanding of learning in today’s digital era [14].
Education in the era of the industrial revolution 4.0 emphasizes student skills known as 4C, specifically creativity, critical thinking, communication, and collaboration [15]. In the context of schools, character education requires the cooperation of various parties and even the development of an appropriate system. Character education is a national movement that creates and fosters young generations who are ethical, responsible and caring [16].
4. Conclusion
The results of research on the development of thematic modules based on PPK and QR Code, it is known that the results of the analysis of media experts obtained a score of 92.86 percent with a very valid category, material experts obtained a score of 88.64% in the very valid category, linguists obtained a score of 75 percent with a fairly valid category, the response of educators got a score of 89.92 percent in the very practical category, the response of students got a score of 87.78 percent with the practical category. Based on the results of the explanation above, the thematic module based on PPK and QR Code can be used as a companion teaching material for the theme book to make it easier for students to master the material and apply characters in schools and communities.
References
[1] Nurhasanah N and Nida Q 2016 Character Building Of Students By Guidance And Counseling Teachers Through Guidance And Counseling Services J. Ilm. Peuradeun 4 65–76
[2] Abu L, Mokhtar M, Hassan Z and Darmanita Suhan S Z 2015 How to Develop Character Education of Madrassa Students in Indonesia J. Educ. Learn. 9 79
[3] Chang F and Muñoz M A 2007 School personnel educating the whole child: Impact of character education on teachers’ self-assessment and student development J. Pers. Eval. Educ. 19 35–49
[4] Ratheeswari K 2018 Information Communication Technology in Education J. Appl. Adv. Res. 3
[5] Maslin N M, Consultant P and Ltd S S 2010 Impact of Modern Technology *HF Commun.* 3 33–5
[6] Getuno D M, Kiboss J K, Changeiywo J and Ogola L B 2015 Effects of an E-Learning Module on Students’ Attitudes in an Electronics Class *J. Educ. Pract.* 6 80–6
[7] Mills S C 2006 *Using the Internet for Active Teaching and Learning* (Ohio: Pearson Meerill Prentice Hall)
[8] Sugiani K A, Degeng I N S, Setyosari P and Sulton 2019 The Effects of Electronic Modules in Constructivist Blended Learning Approaches to Improve Learning Independence *Int. J. Innov. Creat. Chang.* 9 82–93
[9] Perinpasingam P T S, Arumugam N, Sathyaperbha and Mylvaganam G 2014 Development of a science module through interactive whiteboard *Rev. Eur. Stud.* 6 31–8
[10] Mangesa R T and Dirawan G D 2016 Development of Learning Module Work Competence Integrated Character Value of Electricity in Vocational High School *Int. J. Appl. Eng. Res.* 11 6943–8
[11] Sumarti S S, Supartono and Diniy H H 2014 Material Module Development of Colloid Orienting on Local-Advantage-Based Chemo- Entrepreneurship to Improve Students’ Soft Skill *Int. Humanit. Manag. Sci.* 2 42–6
[12] Lim J S C, Jailani Md Yunos and Ghazally Spahat 2005 The Development and Evaluation of an E-Module for Pneumatics Technology *Malaysian Online J. Instr. Technol.* 2 25–33
[13] Watson S and Sutton J M 2012 An Examination of the Effectiveness of Case Method Teaching Online: Does the Technology Matter? *J. Manag. Educ.* 36 802–21
[14] Heriyanto, Sator D, Komariah A and Suryana A 2019 Character Education in the Era of Industrial Revolution 4.0 and its Relevance to the High School Learning Transformation Process *Utop. y Prax. Latinoam.* 24 327–40
[15] Facione P A 2011 Critical thinking: What it is and why it counts *Insight assessment* 2007(1) 1-23
[16] Pala A 2011 The Need For Character Education *Int. J. Soc. Sci. Humanit. Stud.* 3 23–32 | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3039,
1
],
[
3039,
5850,
2
],
[
5850,
6930,
3
],
[
6930,
7878,
4
],
[
7878,
12132,
5
],
[
12132,
14023,
6
]
]
} |
96a84d54d6cf780d1cd39a3f19e567895d45559f | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 9,
"total-fallback-pages": 0,
"total-input-tokens": 35722,
"total-output-tokens": 11727,
"fieldofstudy": [
"Environmental Science"
]
} | MLOD: Awareness of Extrinsic Perturbation in Multi-LiDAR 3D Object Detection for Autonomous Driving
Jianhao Jiao*, Peng Yun*, Lei Tai, Ming Liu
Abstract—Extrinsic perturbation always exists in multiple sensors. In this paper, we focus on the extrinsic uncertainty in multi-LiDAR systems for 3D object detection. We first analyze the influence of extrinsic perturbation on geometric tasks with two basic examples. To minimize the detrimental effect of extrinsic perturbation, we propagate an uncertainty prior on possible fusion schemes in real tests for an excellent detection. As described in [4], current data fusion methods all have pros and cons. It requires us to perform extensive efforts to boost an approach for 3D geometric tasks. Then we extend our findings to propose a multi-LiDAR 3D object detector called MLOD. MLOD is a two-stage network where the multi-LiDAR information is fused through various schemes in stage one, and the extrinsic perturbation is handled in stage two. We conduct extensive experiments on a real-world dataset, and demonstrate both the accuracy and robustness improvement of MLOD. The code, data and supplementary materials are available at: https://ram-lab.com/file/site/mlod.
I. INTRODUCTION
3D object detection is a fundamental module in robotic systems. As the front-end of a system, it enables vehicles to recognize key objects such as cars and pedestrians, which is indispensable for high-level decision making. Compared with camera-based detectors, the LiDAR-based detectors perform better in several challenging scenarios because of their activeness as sensors and distance measurement ability for surroundings. However, LiDARs commonly suffer from data sparsity and a limited vertical field of view [1]. For instance, LiDARs’ points distribute loosely, which induces a mass of empty regions between two nearby scans. In this paper, we consider a multi-LiDAR system, a setup with excellent potential to solve 3D object detection. Compared with single-LiDAR setups, multi-LiDAR systems enable a vehicle to maximize its perceptual awareness of environments and obtain sufficient measurements. The decreasing price of LiDARs also makes such systems accessible to many modern self-driving cars [1–3].
However, two challenges affect the development of multi-LiDAR object detection. One difficulty is multi-LiDAR fusion. As described in [4], current data fusion methods all have pros and cons. It requires us to perform extensive efforts on possible fusion schemes in real tests for an excellent detection algorithm. Another issue is that extrinsic perturbation is inevitable for sensor fusion during long-term operation because of factors such as vibration, temperature drift, and calibration error [5]. Especially, wide baseline stereo cameras or vehicle-mounted multi-LiDAR systems suffer even more extrinsic deviations than the normal one. Even though online calibration has been proposed to handle this issue [6], lifelong calibration is always challenging [7]. These methods typically require environmental or motion constraints with full observability. Otherwise, the resulting extrinsics may become suboptimal or unreliable. As identified in both Fig. 1 and Section IV-B, extrinsic perturbation is detrimental to the measurement accuracy even with a small change. But this fact is often neglected by the research community.
To tackle the challenges, we propose a Multi-LiDAR Object Detector (MLOD) to predict states of objects on point clouds. We design a two-stage procedure to estimate 3D bounding boxes, and demonstrate two innovations in MLOD. First, we explore three fusion schemes to exploit multi-view point clouds in a stage-1 network for generating object proposals. These schemes perform LiDAR fusion at the input, feature, and output phases, respectively. Second,
we develop a stage-2 network to handle extrinsic perturbation and refine the proposals. The experimental results demonstrate that MLOD outperforms single-LiDAR detectors up to 9.7AP in perturbation-free cases. When considering extrinsic perturbation, MLOD consistently improves the performance of its stage-1 counterpart. To the best of our knowledge, this is the first work to systematically study the multi-LiDAR object detection with consideration of extrinsic perturbation. Overall, our work has the following contributions:
1) We analyze the influence of extrinsic perturbation on multi-LiDAR-based geometric tasks, and also demonstrate that the usage of input uncertainty prior improves the robustness of an approach against such effect.
2) We propose a unified and effective two-stage approach for multi-LiDAR 3D object detection, where the multi-LiDAR information is fused in the first stage, and extrinsic perturbation is handled in the second stage.
3) We exhaustively evaluate the proposed approach in terms of accuracy and robustness under extrinsic perturbation on a real-world self-driving dataset.
II. RELATED WORK
We briefly review methods on LiDAR-based object detection and uncertainty estimation of deep neural networks.
A. 3D Object Detection on Point Clouds
The LiDAR-based object detectors are generally categorized into grid-based [8]–[10] and point-based methods [11]–[13]. 3D-FCN [8] implemented 3D volumetric CNN on voxelized point clouds. But it commonly suffers a high computation cost due to dense convolution on sparse point clouds. VoxelNet [9] was proposed as an end-to-end network to learn features. However, operation on grids is inefficient since LiDAR’s points are sparse. To cope with this drawback, SECOND [10] utilized the spatially sparse convolution [14] to replace the 3D dense convolution layers. In this paper, we adopt SECOND as our basic proposal generator and propose three schemes for multi LiDAR fusion in the first stage.
B. Uncertainty Estimation in Object Detection
The problem of uncertainty estimation is essential to the reliability of an algorithm, and has attracted much attention in recent years. As the two main types of uncertainties in deep neural networks: aleatoric and epistemic uncertainty, they are explained in [17], [18]. In [18], the authors also demonstrated the benefits of modeling uncertainties for vision tasks. As an extension, Feng et al. [19] proposed an approach to capture uncertainties for 3D object detection. Generally, the data noise is modeled as a unique Gaussian variable in these works. However, the sources of data uncertainties in multi-LiDAR systems are much more complicated. In this paper, we take both the measurement noise and extrinsic perturbation into account, and analyze their effect on multi-LiDAR-based object detection. Furthermore, we propagate the Gaussian uncertainty prior of both the extrinsics and measurement to model the input data uncertainty. This additional cue is utilized to improve the robustness of MLOD against extrinsic perturbation.
III. NOTATIONS AND PROBLEM STATEMENT
The nomenclature is shown in Tab. I. The transformation from \{l\}_i to \{l\}_i is denoted by \( T_{i} \). Our perception system consists of one primary LiDAR which is denoted by \( l^1 \) and multiple auxiliary LiDARs. The primary LiDAR defines the base frame and the auxiliary LiDAR provides an additional field of view (FOV) and measurements to alleviate the occlusion problem and the sparsity drawback of the primary LiDAR. Extrinsics describe the relative transformation from the base frame to frames of auxiliary LiDARs. With the extrinsics, all measurements or features from different LiDARs are transformed into the base frame for data fusion. LiDARs are assumed to be synchronized that multiple point clouds are perceived at the same time.
In this paper, we focus on 3D object detection with a multi-LiDAR system. The extrinsic perturbation is also considered, which indicates the small but unexpected change on transformation from the base frame to other frames over time. Our goal is to estimate a series of 3D bounding boxes covering objects of predefined classes. Each bounding box \( b \in B \) is parameterized as \([c, x, y, z, w, l, h, \gamma]\), \( c \) is the class
| Notation | Explanation |
|----------|-------------|
| \{l\}_i | Frame of the base and the \( i^{th} \) LiDAR. |
| \( T_{i} \) | Transformation matrix in the Lie group SE(3). |
| \( \Theta \) | Measurement and extrinsic uncertainty prior. |
| \( \Xi \) | Associated covariance of each point. |
| \( \alpha \) | Scaling parameter of \( \Theta \). |
TABLE I
NOMENCLATURE
of a bounding box, \([x, y, z]\) denotes a box’s bottom center, \([w, h, l]\) represent the sizes along the \(x\), \(y\), and \(z\) axes respectively, as well as \(\gamma\) for the rotation of the 3D bounding box along the \(z\)-axis in the range of \([0, \pi]\).
IV. Propagating Extrinsic Uncertainty on Points
We begin by providing preliminaries about the uncertainty representation and propagation. We then use an example to demonstrate the negative effect on points caused by extrinsic perturbation. Finally, we also conduct a plane fitting experiment to show that the extrinsic covariance prior can be utilized to make a fitting approach robust.
A. Preliminaries
We employ the method in [20] to represent the uncertainty. For convenience, rotation and translation are used to indicate a transformation. We first define a random variable for \(\mathbb{R}^3\) with small perturbation according to
\[
t = \rho + \bar{t}, \quad \rho \sim \mathcal{N}(0, P),
\]
(1)
where \(\bar{t}\) is a noise-free translation and \(\rho \in \mathbb{R}^3\) is a zero-mean Gaussian variable with covariance \(P\). We can also define a rotation for \(SO(3)\) as
\[
R = \exp(\phi \& \bar{R}), \quad \phi \sim \mathcal{N}(0, \Phi),
\]
(2)
where \(\bar{R}\) is a noise-free rotation and \(\phi \in \mathbb{R}^3\) is a small zero-mean Gaussian variable with covariance \(\Phi\). With (2), we can represent a noisy transformation by storing the mean as \(\bar{t}, \bar{R}\) and using \([\rho, \phi]\) for perturbation on the vector space. Similarly, a point with the perturbation in \(\mathbb{R}^3\) is written as
\[
p = \zeta + p, \quad \zeta \sim \mathcal{N}(0, Z),
\]
(3)
where \(\zeta\) is zero-mean Gaussian with the covariance \(Z\). With the above representations, we can pass the Gaussian representation of a point through a noisy rotation and translation to produce a mean and covariance for its new measurement. Here, we can use \([\bar{t}_i, \bar{R}_i]\) to indicate the ground-truth extrinsics of a multi-LiDAR system, and \([\rho, \phi, \zeta]\) to indicate both the extrinsic and measurement perturbation. By transforming \(p \in P^i\) into \(\{b\}\), we have
\[
y \triangleq \bar{R}_i^b p + \bar{t}_i^b = \exp(\phi \& \bar{R}_i)p + (\rho + \bar{t}_i) \\
\approx \left(1 + \phi \& \bar{R}_i\right)(\zeta + p) + (\rho + \bar{t}_i),
\]
(4)
where we have kept the first-order approximation of the exponential map. If we multiply out the equation and retain only those terms that are first-order in \(\phi\) or \(\zeta\), we have
\[
y \approx h + H\theta,
\]
(5)
where
\[
\theta = [\rho, \phi, \zeta],
\]
\[
h = \bar{R}_i^b p + \bar{t},
\]
\[
H = [I - (\bar{R}_i^b)^\wedge \bar{R}_i^b],
\]
(6)
\footnote{The \& operator turns a \(3 \times 1\) vector into the corresponding skew symmetric matrix in the Lie algebra \(so(3)\). The exponential map \(\exp\) associates an element of \(so(3)\) to a rotation in \(SO(3)\). The closed-form expression for the matrix exponential can be found in [20].}
We embody the uncertainties of extrinsics and measurements into \(\theta\) that is subjected to a zero-mean Gaussian with \(9 \times 9\) covariance \(\Theta = \text{diag}(\Phi, \Phi, Z)\). Linearly transformed by \(\theta\), \(y\) is a Gaussian variable with mean and covariance as
\[
\mu \triangleq E[y] = h
\]
\[
\Xi \triangleq E[(y - \mu)(y - \mu)^\top] = \Theta \Theta^\top.
\]
(7)
where we follow [21] to use the trace, i.e., \(\text{tr}(\Xi)\), as the criterion to quantify the magnitude of a covariance.
B. Uncertainty Propagation Example
In this section, a simple example of uncertainty propagation on a point is presented, where we quantify the data noise caused by both extrinsic and measurement perturbation. The associated covariance is propagated by passing the Gaussian uncertainty from extrinsics and measurements through a transformation. We use the ground-truth extrinsics having rotation with \([10, 10, 10]^T\) in roll, pitch, yaw and translation with \([1, 1, 1]^T\) along the \(x\), \(y\), and \(z\)-axes respectively. Let \(p = [10, 10, 10]^T m\) be a landmark. \(\Theta\) is treated as our prior knowledge and considered as a constant matrix\footnote{The measurement noise is found in LiDAR datasheets. The extrinsic perturbation typically has the variances around \(5\times m\) in translation and \(5^\circ\) in rotation. This is based on our studies on multi-LiDAR systems [1], [22].}
\[
\Theta = \text{diag}\left\{\frac{1}{20^2}, \frac{1}{20^2}, \frac{1}{10}, \frac{1}{10}, \frac{1}{10^2}, \frac{1}{50^2}, \frac{1}{50}, \frac{1}{50}\right\}.
\]
(8)
where the first six diagonal entries can be multiplied by a scaling parameter \(\alpha\), allowing us to parametrically increase the extrinsic covariance in experiments. According to (7), we have an approximate expression
\[
\Xi \approx \alpha \times \begin{bmatrix}
2.36 & -1.24 & -1.20 \\
-1.24 & 2.41 & -1.18 \\
-1.20 & -1.18 & 2.49
\end{bmatrix},
\]
(9)
where the new position has a high variance around \(0.22 m\) on each axis even for a small input covariance (i.e., \(\alpha = 0.02\)). This accuracy is totally unacceptable for autonomous driving.
C. Plane Fitting Experiment
We denote \(P_m\) a merged point cloud which is obtained by transforming \(P_{i1}\) and \(P_{i2}\) into \(\{b\}\) with ground-truth extrinsics. We assume that there is a dominant plane in \(P_m\), and our task is to estimate the plane coefficients. They can
be obtained by solving a linear system \( Ax = b \) from \( \mathcal{P}_m \), where row elements of \( A \) are the point coordinates, and \( b \) is set as an identity vector with least-squares methods.
Here, we show how the input uncertainty prior \( \Theta \) can be utilized to acquire better fitting results. We define \( W \) as a diagonal matrix to weight the linear system. For a point \( p_j \in \mathcal{P}_m \), we propagate its associated covariance \( \Xi \) using (7). The corresponding entry (i.e., \( W_{jj} \)) is set as the inverse of \( \text{tr}(\Xi) \). After that, the fitting problem is turned into a weighted least-squares regression and the optimal results can be obtained as \( x^* = (A^\top W A)^{-1} A^\top W b \).
An experiment is conducted on a toy example to compare the performance of two fitting methods (i.e., with or without weights). We use the ground-truth extrinsics and the prior covariance in Section V-B to sample perturbation with 100 trials. We also generate 10000 points according to a planar surface which are subjected to zero-mean Gaussian noise with a standard deviation of 0.02m to produce \( \mathcal{P}_m \). \( \mathcal{P}_m \) is randomly split into two equal parts to form \( \mathcal{P}_l \) and \( \mathcal{P}_r \). At each trial, we evaluate the plane fitting results of each method by comparing them to the ground truth as \( e_x = \| x_{gt} - x_{est} \| \).
The mean fitting error of each method on two different cases over \( \alpha \in [0, 0.1] \) is shown in Fig. 2. We see that the weighted least-squares method does better in estimating the plane coefficients, and the mean fitting error does not increase along with \( \alpha \). It shows that the point-wise uncertainty information can be utilized to improve the robustness of algorithms in geometric tasks. But in practice, the value of \( \alpha \) should be carefully set by manual or adaptively obtained from an online method. Otherwise, a few correct measurements are discarded during operation.
V. METHODOLOGY
In this section, we extend our findings from the basic geometric tasks to multi-LiDAR-based 3D object detection. The overall structure of the proposed two-stage MLOD is illustrated in Fig. 3. We adopt SECOND [10], which is a sparse convolution improvement of VoxelNet [9], to generate 3D proposals in the first stage. It consists of three components: a feature learning network (FLN) for feature extraction; middle layers (ML) for feature embedding with sparse convolution; and a region proposal network (RPN) for box prediction and regression. Readers are referred to [9], [10] for more details about the network structures. We first present three species of 3D proposal generation with different multi-LiDAR fusion schemes: Input Fusion, Feature Fusion and Result Fusion. Then, we introduce the architecture of our stage-2 network to tackle the extrinsic uncertainty and refine the proposals.
A. Proposal Generation With Three Fusion Schemes
According to stages at which the information from multiple LiDARs is fused, we propose three general fusion schemes, which we call Input Fusion, Feature Fusion, and Result Fusion. These approaches are developed from SECOND with several modifications.
1) Input Fusion: The fusion of point clouds is performed at the input stage. We transform raw point clouds perceived
3We use SECOND V1.5 in our experiments: https://github.com/traveller59/second.pytorch Different from the original SECOND [10], the FLN of SECOND V1.5 is simplified by computing the mean value of points within each voxel, which consumes less memory.
by all LiDARs into the base frame to obtain the fused point cloud, and then feed it to the network as the input.
2) Feature Fusion: To enhance the feature interaction, the LiDAR data is also fused in the feature level. The extracted features from the FLN and ML are transformed into the base frame, and then fused by adopting the maximum value as
$$F_{\text{fused}} = F_i^{\text{fl}} \oplus (T_{b}^{\text{fl}} F_i^{\text{fl}}) \oplus \cdots \oplus (T_{b}^{\text{ml}} F_i^{\text{ml}}),$$
where $F_i^{\text{fl}}$ are the extracted features, $\oplus$ denotes the max operator, and $I$ is the number of LiDARs.
3) Result Fusion: The result fusion takes the box proposals and the associated points as inputs, and produces a set of boxes with high scores. We transform all box proposals into the base frame, and then filter them as
$$B_{\text{fused}} = B_i^{\text{fl}} \oplus (T_{b}^{\text{fl}} B_i^{\text{fl}}) \oplus \cdots \oplus (T_{b}^{\text{ml}} B_i^{\text{ml}}),$$
where $\oplus$ denotes the non-maximum suppression (NMS) on 3D intersection-over-Union (IoU). After transformation, each object is associated with several candidate boxes, and these boxes with low confidences are filtered by the NMS.
B. Box Refinement From Uncertain Points
Although the stage-1 fusion-based network generates promising proposals, its capability in handling uncertain data uncertainty is fragile. Therefore, we propose a stage-2 module to improve the robustness of MLOD. A straightforward idea is to eliminate the highly uncertain points according to their associated covariances. But this method is sensitive to a pre-set threshold. Inspired by [13], it is more promising that training a neural network to embed features and refine the proposals with an awareness of uncertainty.
We thus use a deep neural network to deal with this problem. The network takes a series of points of each proposal (a 2m margin along $x$, $y$, and $z$-axes is expanded) generated by the stage-1 network, and refines the stage-1 proposals. In addition to three-dimensional coordinates, each point is also embedded with its uncertain quantity. As defined in [7], we use the trace, i.e., $\text{tr}(\Sigma)$, to quantify the uncertainties. We employ PointNet [11] as the backbone to extract global features. We also encode extra features, including the scores and parameters of each proposal, to concatenate with the global features. At the end of the network, fully connected layers are used to provide classification scores and refinement results with two heads. To reduce the variance of input, we normalize the coordinates of each point of a proposal as
$$p_n = Tp \odot s,$$
where
$$T = \begin{bmatrix} \cos(\gamma) & \sin(\gamma) & 0 & -x \\ -\sin(\gamma) & \cos(\gamma) & 0 & -y \\ 0 & 0 & 1 & -z \\ 0 & 0 & 0 & 1 \end{bmatrix},$$
$$s = \begin{bmatrix} 1/l \\ 1/w \\ 1/h \\ 1 \end{bmatrix}^T,$$
where $T$ and $s$ are defined in terms of the parameters of a proposal $b = [c, x, y, z, l, w, h, \gamma]$, $\odot$ is the Hadamard product, and both $p_n$ and $p$ are represented in homogeneous coordinates. The output of the classification head is a binary variable indicating the probability of objectiveness. We regress residuals of the bounding box based on the proposal instead of directly regressing the final 3D bounding box. The regression target of the stage-2 network is
$$u_i \triangleq \begin{bmatrix} x-x_p \\ y-y_p \\ z-z_p \\ \frac{l}{l_p} \\ \frac{w}{w_p} \\ \frac{h}{h_p} \sin(\gamma-\gamma_p), \cos(\gamma-\gamma_p) \end{bmatrix},$$
where $u_i \in \mathbb{R}^8$ are regression outputs for the $i^{th}$ positive proposal respectively. We adopt the classification and regression loss which are defined in [23] as
$$L_{\text{cls}} = \sum_i f_{\text{cls}}(p_i), \quad L_{\text{reg}} = \sum_i f_{\text{reg}}(u_i, u^*_i),$$
where $p_i$ represents the posterior probability of objectiveness, $L_{\text{cls}}$ is defined as the classification loss, $f_{\text{cls}}(\cdot)$ denotes the focal loss [24], $L_{\text{reg}}$ is defined as the normalized regression loss, and $f_{\text{reg}}(\cdot)$ denotes the smooth-L1 loss. Since we have the observation that the regression outputs $u_i$ should not be large when the uncertainty is small, we add a regularization term to penalize large $u_i$ in the low uncertainty cases:
$$L_{\text{uct}} = \sum_i \exp \left( 1 - f \left[ \max_j \text{tr}(\Sigma_j) \right] \right) \cdot ||u^*_i||_2,$$
where $\text{tr}(\Sigma_j)$ is the point-wise uncertainty of the $j^{th}$ point within the $i^{th}$ proposal, $MAX$ is the max operator, $u^*_i$ is the regressed residual without the last cosine term for the the $i^{th}$ proposal, and $f(\cdot)$ clamps a value into the range $[10^{-5}, 0.5]$ to stabilize training. Finally, we define the loss function for training the stage-2 network as
$$L = L_{\text{reg}} + \eta L_{\text{cls}} + \lambda L_{\text{uct}},$$
where $\eta$ and $\lambda$ are hyper-parameters to balance the weight of the classification loss and the uncertainty regularizer. We use $\eta = 2$ and $\lambda = 0.005$ in our experiments.
Fig. 4 visualizes the refinement results of our stage-2 network, which shows that our method produces similar results to the ground truths even if many points are uncertain. More quantitative examples are provided in Section VI-C.
VI. EXPERIMENT
In this section, we evaluate our proposed MLOD on LYFT multi-LiDAR dataset [3] in terms of accuracy and robustness under different levels of extrinsic perturbation. In particular, we aim to answer the following questions:
1) Can a multi-LiDAR object detector perform better accuracy than single-LiDAR methods?
2) Can MLOD improve the robustness of a multi-LiDAR object detector under extrinsic perturbation?
A. Implementation Details
1) Dataset: The LYFT dataset [3] provides a large amount of data collected in a variety of environments for the task of 3D object detection. Three 40-beam and calibrated Hesai LiDARs are mounted on the top, front-left and front-right position of the vehicle platform. The top LiDAR is set as the primary LiDAR, which is denoted by $l^1$. The left and right LiDARs are auxiliary LiDARs, which are denoted by $l^2$ and $l^3$ respectively. This setting is according to LiDAR’s field of view (FOV). We select all multi-LiDAR data samples for our experiment. The data contain 4031 samples, 2500 of which are for training as well as validation and 1531 of which are for testing. The testing scenes are different from those in the training and validation sets.
2) Metric: Following the KITTI evaluation metrics [25], we compute the average precision on 3D bounding boxes (AP$_{3D}$) to measure the detection accuracy. We compute the AP$_{3D}$ for 360$^\circ$ around the vehicle instead of evaluating only the 90$^\circ$ front view. According to the distance of objects, we set three different evaluation difficulties: easy ($< 20 m$), moderate ($< 30 m$), and hard ($< 50 m$).
3) Extrinsic Perturbation Injection: We inject extrinsic perturbation on the original LYFT dataset to generate another set of data for robustness evaluation. We take $\Theta$ defined in (8) and adjust $\alpha \in [0, 0.1]$ with a 0.02 interval to simulate different levels of perturbation. To remove the effects from outliers, we bound the sampled perturbation within the $\sigma$ position. Although the maximum value of $\alpha$ is small, but the effect of input perturbation on points is obvious. The noisy extrinsics are obtained by adding the sampled perturbation to the ground-truth extrinsics according to (1) and (2).
4) Case Declaration: We declare all cases of our methods which are tested in the following experiments as
- **LiDAR-Top, LiDAR-Left, LiDAR-Right**: single-LiDAR object detectors based on SECOND with the input data which are captured by the top LiDAR, front-left LiDAR and front-right LiDAR, respectively.
- **Input Fusion, Feature Fusion, Result Fusion**: multi-LiDAR object detectors with different fusion schemes, which are introduced in Section V-A
- **MLOD-I, MLOD-F, MLOD-R**: MLOD with Input Fusion, Feature Fusion, Result Fusion as its stage-1 network respectively, which are described in Section V-B
- **MLOD-I (OC), MLOD-F (OC), MLOD-R (OC)**: Variants of MLOD with an online calibration method to reduce extrinsic perturbation. This method is implemented with a point-to-plane ICP approach [26].
5) Training of Stage-One Network: We follow [10] to train the stage-1 network. During training, we conduct dataset sampling as in [10] and an augmentation of random flips on the $y$-axis as well as translation sampling within $[-0.5, 0.5] m$, $[-0.3, 0.3] m$ on the $x$-, $y$-, and $z$-axes respectively, as well as rotation around the $z$-axis between $[-5,5]^\circ$. In each epoch, the augmented data accounts for 30% of the whole training data. In Result Fusion and Feature Fusion, we directly take the calibrated and merged point clouds from different LiDARs as input. In Result Fusion, we calibrate all the proposal generators with temperature scaling [27].
6) Training of Stage-Two Network: To train our stage-2 network, we use all cases of the well-trained stage-1 networks to generate proposals. We use the training set to create proposals and inject uncertainties by selecting different $\alpha$ like the above section. In addition, we force the ratio between uncertainty-free samples and uncertain samples to be about 1:1.5 to stabilize the training process. A proposal is considered to be positive if its maximum IoU with ground-truth boxes is above 0.6, and is treated as negative if its maximum 3D IoU is below 0.45. During training, we conduct data augmentation of random flipping, scaling with a scale factor sampled from $[0.95, 1.05]$, translation along each axis between $[-0.1, 0.1] m$ and rotation around the $z$-axis between $[-3,3]^\circ$. We also randomly sample 1024 points within the proposals to increase their diversity.
B. Results on the Multi-LiDAR LYFT Dataset
We evaluate both single-LiDAR and multi-LiDAR detectors on the LYFT test set, as reported in Tab. II. In this experiment, we do not consider any extrinsic perturbation between LiDARs. **LiDAR-Top** performs the best among all single-LiDAR detectors. This is because the top LiDAR has a complete 360$^\circ$ horizontal field of view (FOV). The left and right LiDARs, which are partially blocked by the vehicle, only have a 225$^\circ$ horizontal FOV. With sufficient measurements and decreasing occlusion areas, all multi-LiDAR detectors outperform **LiDAR-Top**, and the accuracy improvement gains up to 9.8AP. Tab. II also shows that all MLOD variants perform better or comparable to their one-stage counterparts. This proves that our stage-2 network refines the stage-1 proposals in perturbation-free cases.
| Case | AP$_{3D}$ @IoU $\geq$ 0.7 | AP$_{3D}$ @IoU $\geq$ 0.5 |
|---------------------|----------------------------|-----------------------------|
| | easy | mod. | hard | easy | mod. | hard |
| LiDAR-Top | 62.7 | 52.9 | 37.8 | 89.8 | 80.1 | 61.8 |
| LiDAR-Left | 41.1 | 37.0 | 27.5 | 62.6 | 53.9 | 43.1 |
| LiDAR-Right | 40.9 | 31.5 | 23.4 | 53.9 | 52.4 | 35.5 |
| Input Fusion | 71.9 | 61.8 | 45.0 | 89.4 | 87.9 | 61.9 |
| MLOD-I | 72.4 | 62.7 | 45.6 | 89.6 | 88.3 | 61.8 |
| Feature Fusion | 71.1 | 60.2 | 44.1 | 89.6 | 87.4 | 61.6 |
| MLOD-F | 72.0 | 61.8 | 44.9 | 89.7 | 88.2 | 61.6 |
| Result Fusion | 71.4 | 61.6 | 44.5 | 89.5 | 87.7 | 62.0 |
| MLOD-R | 72.1 | 62.3 | 45.2 | 89.6 | 88.0 | 61.7 |
C. Robustness Under Extrinsic Perturbation
In this section, we evaluate the robustness of our proposed methods under extrinsic perturbation. Three levels of perturbation are performed: no perturbation ($\alpha = 0$), moderate perturbation ($\alpha = 0.02$, $\alpha = 0.04$) and high perturbation ($\alpha = 0.1$). An example is displayed in Fig. 6(a) ($\alpha = 0.04$), where massive noisy points (50cm misplacement) appear at 35m away. This is a typical phenomenon of extrinsic perturbation in multi-LiDAR systems. We randomly select 300 data samples from the test set with 10 trials to conduct evaluations at each level.
In Tab. III the detection results in terms of the means and variances as $AP_{3D}(IoU \geq 0.7)$ are detailed. We see that all variants of MLOD (MLOD-I, MLOD-F, MLOD-R) perform better or comparable to their one-stage counterparts. Online calibration baselines (Input Fusion (OC), Feature Fusion (OC), Result Fusion (OC)) demonstrate better results than those without online calibration. With the assistance of MLOD, their performance is further enhanced. Feature Fusion and its variants perform better than the others. We explain that Feature Fusion fuses data in a high dimensional embedding space, and each feature is related to an area of the cognitive region. Regarding LiDAR-Top, MLOD’s variants also perform better or comparable results. A qualitative result is illustrated in Fig. 6 and more results are shown in the supplementary materials.
We also conduct a sensitivity analysis of extrinsic uncertainty with a fixed $\alpha$ as input in the supplementary material. The same conclusion still holds. To further explore the reason behind the performance of MLOD, we investigate the characteristics of active points. These points are activated by PointNet before the max-pooling layer (Fig. 5(a)). We compute the difference between the input points and active points on the proportion of uncertain point for each sample. Fig. 5(b) plots a histogram of the proportional difference
**Fig. 5.** (a) Active points (Red dotted) form the global features after applying the max-pooling symmetric function. (b) The proportional differences between input uncertain points and active uncertain points.
Table III: Mean and Variance of Accuracy ($AP_{3D}(IoU \geq 0.7)$) Under Different Level of the Extrinsic Perturbation.
| Cases | $\alpha = 0$ mod. | $\alpha = 0.02$ mod. | $\alpha = 0.04$ mod. | $\alpha = 0.1$ mod. |
|-------|------------------|----------------------|----------------------|------------------|
| Top LiDAR | 63.3 | 53.8 | 38.2 | 63.3 | 53.8 | 38.2 | 63.3 | 53.8 | 38.2 |
| MLOD | 71.2 | 62.3 | 45.4 | 64.7 ± 2.5 | 60.8 ± 0.4 | 44.4 ± 0.2 | 63.0 ± 0.5 | 53.9 ± 2.1 | 38.0 ± 0.2 | 60.9 ± 1.0 | 51.0 ± 0.6 | 36.1 ± 0.4 |
| Feature Fusion | 78.1 | 73.2 | 46.4 | 66.3 ± 3.6 | 61.4 ± 0.4 | 44.8 ± 0.2 | 62.8 ± 0.4 | 54.3 ± 3.7 | 35.7 ± 0.2 | 61.6 ± 0.9 | 51.5 ± 0.6 | 35.8 ± 0.5 |
| Result Fusion | 72.1 | 69.2 | 42.9 | 63.9 ± 2.5 | 59.2 ± 0.9 | 42.9 ± 0.9 | 62.9 ± 0.4 | 58.6 ± 2.5 | 41.1 ± 3.3 | 61.9 ± 0.6 | 51.1 ± 1.0 | 35.7 ± 0.7 |
Fig. 6. Percentage of Data Samples as input in the supplementary material.
In this paper, we analyze the extrinsic perturbation effect on multi-LiDAR-based 3D object detection. We propose a two-stage network to both fuse the data from multiple LiDARs and handle extrinsic perturbation after data fusion. We conduct extensive experiments on a real-world dataset and discuss the results in different levels of extrinsic perturbation. In the perturbation-free situation, we show the multi-LiDAR fusion approaches obtain better accuracy than single-LiDAR detectors. Under extrinsic perturbation, MLOD performs great robustness with the assistance of the uncertainty prior. A future direction concerns the combination with sensors in various modalities, e.g., LiDAR-camera-radar setups, for
---
4The proportion of the uncertain points is defined as $|P_u|/|P|$, where $P_u \subset P$ is the set of uncertain points with the uncertain quantity $> 0.05$.
---
VII. Conclusion
In this paper, we analyze the extrinsic perturbation effect on multi-LiDAR-based 3D object detection. We propose a two-stage network to both fuse the data from multiple LiDARs and handle extrinsic perturbation after data fusion. We conduct extensive experiments on a real-world dataset and discuss the results in different levels of extrinsic perturbation. In the perturbation-free situation, we show the multi-LiDAR fusion approaches obtain better accuracy than single-LiDAR detectors. Under extrinsic perturbation, MLOD performs great robustness with the assistance of the uncertainty prior. A future direction concerns the combination with sensors in various modalities, e.g., LiDAR-camera-radar setups.
developing a more accurate object detector.
REFERENCES
[1] J. Jiao, Y. Yu, Q. Liao, H. Ye, R. Fan, and M. Liu, “Automatic calibration of multiple 3d lidars in urban environments,” in 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2019, pp. 15–20.
[2] P. Sun, H. Kretzschmar et al., “Scalability in perception for autonomous driving: Waymo open dataset,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 2446–2454.
[3] R. Kesten, M. Usman et al., “Lyft level 5 av dataset 2019,” https://levels.lyft.com/dataset/, 2019.
[4] X. Chen, H. Ma, J. Wan, B. Li, and T. Xia, “Multi-view 3d object detection network for autonomous driving,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 1907–1915.
[5] Y. Ling and S. Shen, “High-precision online markerless stereo extrinsic calibration,” in 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2016, pp. 1771–1778.
[6] J. Lin, X. Liu, and F. Zhang, “A decentralized framework for simultaneous calibration, localization and mapping with multiple lidars,” arXiv preprint arXiv:2007.01483, 2020.
[7] W. Maddern, G. Pascoe, C. Linegar, and P. Newman, “1 year, 1000 km: The oxford robocar dataset,” The International Journal of Robotics Research, vol. 36, no. 1, pp. 3–15, 2017.
[8] B. Li, “3d fully convolutional network for vehicle detection in point cloud,” in 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2017, pp. 1513–1518.
[9] Y. Zhou and O. Tuzel, “Voxelnet: End-to-end learning for point cloud based 3d object detection,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 4490–4499.
[10] Y. Yan, Y. Mao, and B. Li, “Second: Sparsely embedded convolutional detection,” Sensors, vol. 18, no. 10, p. 3337, 2018.
[11] C. R. Qi, H. Su, K. Mo, and L. J. Guibas, “Pointnet: Deep learning on point sets for 3d classification and segmentation,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 652–660.
[12] C. R. Qi, L. Yi, H. Su, and L. J. Guibas, “Pointnet++: Deep hierarchical feature learning on point sets in a metric space,” in Advances in neural information processing systems, 2017, pp. 5099–5108.
[13] S. Shi, X. Wang, and H. Li, “Pointrcnn: 3d object proposal generation and detection from point cloud,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2019, pp. 770–779.
[14] B. Graham and L. van der Maaten, “Submanifold sparse convolutional networks,” arXiv preprint arXiv:1706.01307, 2017.
[15] C. R. Qi, W. Liu, C. Wu et al., “Frustum pointnets for 3d object detection from rgb-d data,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 918–927.
[16] D. Xu, D. Anguelov, and A. Jain, “Pointfusion: Deep sensor fusion for 3d bounding box estimation,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 244–253.
[17] A. Kendall and Y. Gal, “What uncertainties do we need in bayesian deep learning for computer vision?” in Advances in neural information processing systems, 2017, pp. 5574–5584.
[18] A. Kendall, V. Badrinarayanan, and R. Cipolla, “Bayesian segnet: Model uncertainty in deep convolutional encoder-decoder architectures for scene understanding,” in British Machine Vision Conference 2017, BMVC 2017, 2017.
[19] D. Feng, L. Rosenbaum, F. Timm, and K. Dietmayer, “Leveraging heteroscedastic aleatoric uncertainties for robust real-time lidar 3d object detection,” in 2019 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2019, pp. 1280–1287.
[20] T. D. Barfoot and P. T. Furgale, “Associating uncertainty with three-dimensional poses for use in estimation problems,” IEEE Transactions on Robotics, vol. 30, no. 3, pp. 679–693, 2014.
[21] Y. Kim and A. Kim, “On the uncertainty propagation: Why uncertainty on lie groups preserves monotonicity?” in 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2017, pp. 3425–3432.
[22] T. Liu, Q. Liao, L. Gan et al., “Hercules: An autonomous logistic vehicle for contact-less goods transportation during the covid-19 outbreak,” arXiv preprint arXiv:2004.07480, 2020.
[23] P. Yun, L. Tai, Y. Wang, C. Liu, and M. Liu, “Focal loss in 3d object detection,” IEEE Transactions on Robotics and Automation Letters, vol. 4, no. 2, pp. 1263–1270, 2019.
[24] T.-Y. Lin, P. Goyal, R. Girshick, K. He, and P. Dollár, “Focal loss for dense object detection,” in Proceedings of the IEEE international conference on computer vision, 2017, pp. 2980–2988.
[25] A. Geiger, P. Lenz, C. Stiller, and R. Urtasun, “Vision meets robotics: The kitti dataset,” The International Journal of Robotics Research, vol. 32, no. 11, pp. 1231–1237, 2013.
[26] F. Pomareleau et al., “Comparing icp variants on real-world data sets,” Autonomous Robotic, vol. 34, no. 3, pp. 133–148, 2013.
[27] C. Guo, G. Pleiss, Y. Sun, and K. Q. Weinberger, “On calibration of modern neural networks,” in *International Conference on Machine Learning*, 2017, pp. 1321–1330. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3828,
1
],
[
3828,
8486,
2
],
[
8486,
13922,
3
],
[
13922,
17524,
4
],
[
17524,
23301,
5
],
[
23301,
29588,
6
],
[
29588,
34390,
7
],
[
34390,
39416,
8
],
[
39416,
39585,
9
]
]
} |
1f11a079315857250546f70b9c98577ce89483d8 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 14,
"total-fallback-pages": 0,
"total-input-tokens": 49956,
"total-output-tokens": 15483,
"fieldofstudy": [
"Agricultural And Food Sciences"
]
} | Mannan Oligosaccharides Application: Multipath Restriction From Aeromonas hydrophila Infection in the Skin Barrier of Grass Carp (Ctenopharyngodon idella)
Zhiyuan Lu†, Lin Feng1,2,3†, Wei-Dan Jiang1,2,3, Pei Wu1,2,3, Yang Liu1,2,3, Jun Jiang1,2,3, Sheng-Yao Kuang4,5, Ling Tang4,5, Shu-Wei Li4,5, Xiang-An Liu4, Cheng-Bo Zhong5 and Xiao-Qiu Zhou1,2,3*
1 Animal Nutrition Institute, Sichuan Agricultural University, Chengdu, China, 2 Fish Nutrition and Safety Production University Key Laboratory of Sichuan Province, Sichuan Agricultural University, Chengdu, China, 3 Key Laboratory for Animal Disease-Resistance Nutrition of China Ministry of Education, Sichuan Agricultural University, Chengdu, China, 4 Sichuan Animal Science Academy, Sichuan Animtech Feed Co. Ltd, Chengdu, China, 5 Animal Breeding and Genetics Key Laboratory of Sichuan Province, Animal Nutrition Institute, Sichuan Academy of Animal Science, Chengdu, China
The objective of this study was to evaluate the efficacy of dietary Mannan oligosaccharides (MOS) supplementation on skin barrier function and the mechanism of on-growing grass carp (Ctenopharyngodon idella). Five hundred forty grass carp were fed for 60 days from the growing stage with six different levels of MOS diets (0, 200, 400, 600, 800, and 1,000 mg kg⁻¹). At the end of the growth trial, the 14-day Aeromonas hydrophila challenge experiment has proceeded. The obtained data indicate that MOS could (1) decline skin lesion morbidity after being challenged by the pathogenic bacteria; (2) maintain physical barrier function via improving antioxidant ability, inhibiting excessive apoptosis, and strengthening the tight junction between the epithelial cell and the related signaling pathway (Nrf2/Keap1, p38MAPK, and MLCK); and (3) regulate immune barrier function by modulating the production of antimicrobial compound and expression of involved cytokines and the related signaling pathway (TOR and NFκB). Finally, we concluded that MOS supplementation reinforced the disease resistance and protected the fish skin barrier function from Aeromonas hydrophila infection.
Keywords: mannan oligosaccharides, antioxidant, apoptosis, tight junction, skin immune, grass carp (Ctenopharyngodon idella)
INTRODUCTION
Due to large-scale intensive production facilities, fish are exposed to potential various pathogens that often result in massive economic losses (1). The disease resistance of fish mainly depends on the main defense organs’ immune function (2). Skin, an important mucosal defense organ in fish, has developed a better barrier system (including physical barriers and immune barriers) to protect the whole body from natural pathogen invasion (3). A previous report has confirmed that mechanical
skin barrier injury could further lead to high morbidity and mortality of fish (4). Therefore, a protected fish skin barrier function must be necessary for fish health. An effective strategy is to supplement the dietary with prebiotics. The current definition of prebiotics is, “Prebiotics are food constituents that well thought-out to be non-digestible selectively fermented, confers benefits of growth and activity of beneficial microbes present in gastrointestinal tract and improve the health of host” (5). In general, the prebiotics used in animal production are some functional oligosaccharides (6). Studies on fish reported that skin physical and immune barrier function could be improved by functional oligosaccharides such as xylooligosaccharides (XOS), galactooligosaccharides (GOS), fructooligosaccharides (FOS), and so on (7–9). Mannan oligosaccharide (MOS), a kind of functional oligosaccharide, is widely used in the feed formulation of aquatic (10). However, no systematic research has been conducted, and no in-depth exploration has been performed about the relationship between MOS and skin barrier function. Limited research on skin has shown that MOS supplementation promoted mucus production in rainbow trout (Oncorhynchus mykiss) and European sea bass (Dicentrarchus labrax), and upregulated IFNγ and IL-10 expression in greater amberjack (Seriola dumerili Risso 1810) (11–13). Thus, a comprehensive understanding of the effect of MOS on fish skin barrier function and the in-depth possible mechanisms is necessary.
Previous studies have reported that cellular structure and intercellular junctions comprise skin physical barrier, which is mainly related to the antioxidant capacity, apoptosis levels, and tight junctions (1, 14). As far as we know, the NF-E2-related factor 2 (Nrf2) could eliminate excess free radicals by regulating the levels of antioxidant enzymes, while p38 mitogen-activated protein kinase (MAPK) could dynamically regulate apoptosis by regulating apoptosis promoter and effector, thus corporately protecting the cellular structure integrity (15, 16). Myosin light chain kinase (MLCK) is an important signaling molecule that could maintain intercellular junctions by regulating the expression of downstream tight junction protein molecules (17, 18). However, the research to date about the effects of MOS on fish skin cellular structure and intercellular junctions and their possible mechanism has not been investigated. It is worth noting that available evidence suggests a probable correlation between MOS and skin physical barrier. A study on chicken macrophages demonstrated that MOS could increase the production of nitric oxide (NO) (19), which could activate Nrf2 in PC12 cells (20). Furthermore, MOS supplementation could improve calcium (Ca⁺++) absorption and retention in layer hens (21). Other reports revealed that Ca⁺⁺ induced apoptosis via activating p38MAPK signaling pathways in murine macrophage cells (22). Besides, IL-1β gene expression was upregulated by MOS in European sea bass (23). And occludin expression could be decreased by IL-1β in Caco-2 cells (24). These intriguing observations implicate a probably delicate link between MOS and fish skin physical barrier, and the underlying mechanism warrants further exploration.
Fish physical barrier function of the skin is also associated with the immune barrier function, which is closely related to antimicrobial compounds [such as lysozyme (LZ), complement 3 (C3), and immunoglobulins (Ig)] and inflammatory cytokines (25–27). However, available literature describing the skin barrier’s function affected by MOS supplementation after pathogen infections is particularly scarce. A study in human macrophages showed that cytokines were mediated by nuclear factor kappa B (NFκB) (28) and the target of rapamycin (TOR) signaling pathways (29). It has been reported that MOS increased the digestibility of protein in the ileum of piglets (30). Our lab’s previous work in grass carp confirmed that protein increased the activity of LZ and the concentration of C3 (31). A study on weaned piglets demonstrated that fed MOS diet could enhance the digestibility of phosphorus in ileum (30). Another study from our lab in grass carp described that phosphorus could upregulate interleukin 15 (IL-15) expression, which is regulated by the TOR signaling pathway (32). Furthermore, Pinheiro et al. (33) demonstrated that MOS increased butyrate concentration in growing rabbit cecum. It was of note that butyrate could inhibit the activation of the NFκB signaling pathway in grass carp (34). All of these studies imply that MOS might regulate skin immune barrier function via acting on multiple pathways, the mechanism of which is worth in-depth exploration.
Based on the lab’s previous MOS study of growth and intestinal health (35), the objectives of the present study were to elaborate on the protective effects of dietary MOS supplementation on the skin barrier function of on-growing grass carp under the condition of pathogen infection. For this purpose, this work explores the influence of MOS on antioxidant parameters, apoptosis parameters, tight junction (TJ) proteins, antibacterial compounds, and cytokines, as well as the possible signal molecule Nrf2, p38MAPK, MLCK, NFκB, and TOR in the skin of grass carp after being challenged with Aeromonas hydrophila for the first time. Furthermore, as we all know, the grass carp is a broadly distributed species over the world (36). These results will shed new light on the understanding of freshwater fish defense mechanisms to bacterial pathogens, and also provide a more effective alternative reference for antibiotics.
### MATERIALS AND METHODS
#### Study Design
The method of MOS (Sciphar Hi-Tech Industry, Xi’an, purity: 99.12%) diet preparation and storage was based on our published work (35, 37). The experimental diet formulation and proximate composition analyses are displayed in Supplementary Table 1. The different levels of MOS (0, 200, 400, 600, 800, and 1,000 mg kg⁻¹) were added to the control diet in place of cornstarch. All completed diets were stored at 4°C until feeding.
#### Determination of Antioxidant Properties
MOS antioxidant properties were determined mainly by the kit list in Supplementary Table 2. In short, DPPH, 2,2′-azino-bis(3-ethylbenzthiazoline-6-sulphonic acid) (ABTS), and 2,2′-azino-bis(3-ethylbenzthiazoline-6-sulphonic acid) (ABTS) radical scavenging activities of MOS at different levels were determined to reflect the antioxidant properties of MOS in vitro.
The method used is spectrophotometry as previously described (38, 39).
**Animals and Experimental Management**
The guidelines for the Laboratory Animals Care and Use of Animal Nutrition Institute (LACUANI), Sichuan Agricultural University were strictly followed (permit no. LZY-2018114005) during the whole feeding trial. All healthy on-growing grass carp were obtained from Tong Wei fisheries (Sichuan, China) and acclimated to the fishpond culture condition for a month before the experiment. A total of 540 individuals (215.85 ± 0.30 g) were randomized to 18 nylon cages (n=30), and the feeding frequency and experimental period had the same description as our previous study (35). Routine test control parameters were as follows: dissolved oxygen > 6.0 mg L⁻¹, water temperature at 28.5 ± 2.0°C, pH value 7.5 ± 0.3, and experiment condition with a natural light cycle during the whole experimental period.
**Challenge Test**
After the growth trial, a 14-day challenge test (CT) was conducted to study the effect of dietary MOS on the fish skin barrier function according to our published work (35). Briefly, randomly selected five fish per replicates from each MOS group were intraperitoneally injected with 1.0 ml A. hydrophila (FDL20120711), and the concentration of bacteria is 2.5 × 10⁸ colony-forming units (CFU) ml⁻¹. Concurrently, the saline group was injected with the same amount of normal saline. The situation and management were in line with the feeding trial. In a previous study, we successfully establish the A. hydrophila challenge model.
**Sampling and Biochemical Parameter Analysis**
At the end of the CT, all grass carp were anesthetized in a benzocaine bath according to LACUANI requirements. Then fish skin was rapidly collected and temporarily stored in liquid nitrogen. Finally, the sample was stored at -80°C for later analysis. The methods of the skin lesion morbidity scoring system were from a previous study (40). For the determination of physical and immune barrier-related parameters, 10% (w/v) of skin tissue homogenates were prepared with saline (4°C) and centrifuged (6,000 g, 20 min). Then the supernatant was collected. The biomarkers and related enzyme activity analysis methods are shown in Supplementary Table 2.
**DNA Fragment Analysis**
The fragmented DNA of the skin tissue was isolated as previously described (41). And then, DNA was extracted following the instructions and analyzed on a 2% agarose gel to verify DNA fragmentation. Electrophoresis duration and related parameters were 90 min and 80 V, respectively. Finally, Gene Genius (Syngene, Frederick, MD, USA) is used to analyze the results of visualizations.
**Real-Time PCR**
qRT-PCR was conducted to refer to the method from our previous work (35). In short, the total RNA of skin samples was isolated by using an RNAiso Plus Kit (Takara, Dalian, China). RNA quality was assessed by 1% agarose gel electrophoresis and quantified by spectrophotometry at 260/280 nm using Nanodrop 2000 (Thermo Scientific, USA). Afterward, RNA was reverse-transcribed into cDNA by using a PrimeScript™ RT reagent kit (Takara, Dalian, China). For qRT-PCR, specific primers were designed according to the sequences we cloned (Supplementary Table 3). Our preliminary experiment screened four internal reference genes and finally selected β-Actin and GAPDH as previously described (35, 42). Preparation of melting curves and calculation of amplification efficiency of target genes were according to the manufacturer's instruction. The gene transcription level was calculated as described by the method (2^−ΔΔCT) from Livak and Schmittgen (43).
**Western Blot Analysis**
Preparation method-related parameters of skin homogenates, primary and second antibodies, and blotting analysis were performed as our lab previously described (35, 44, 45). Extraction and determination of tissue protein were performed by using the RIPA and BCA assay kit (Beyotime). The prepared sample (40 μg lane⁻¹) was separated by SDS-PAGE (10%) and transferred to a PVDF membrane. Membranes were incubated overnight with primary antibody (14 h, 4°C). Afterward, membranes were washed and secondary antibody was incubated (90 min, room temperature). Then, protein signals were visualized and quantified (NIH Image J, 1.42q) as previously described (35, 37). All antibodies’ detailed information in the current study is listed in Supplementary Table 4.
**Statistical Analysis Method**
Before statistical analysis, the Shapiro–Wilk test of normality, as well as Levene’s test of variance homogeneity, was conducted. All data underwent one-way analysis of variance (ANOVA) followed by Duncan’s multiple comparisons at P < 0.05 with SPSS 25.0 (SPSS Inc., Chicago, IL, USA). Data visualization was done using the GraphPad 8.0 software (GraphPad Software, Inc.), R (v4.0.2), and Hiplot platform (https://hiplot.com.com).
**RESULTS**
**Skin Morbidity and Phenotype**
To investigate the effect of MOS on fish skin morbidity and phenotype with A. hydrophila challenge, we performed intraperitoneal injection of bacteria solution. We obtained the results of skin morbidity and phenotype as showed in Figure 1; compared to the control (14.40%), the morbidity of skin hemorrhages and lesions after being challenged was significantly decreased with MOS at 400 mg kg⁻¹ diet. At this optimal MOS supplementation, the skin morbidity was reduced to a minimum 8.27% (P < 0.05). Then, it showed an upward trend (from 9.87% to 13.33%, P < 0.05) with the increase in MOS (600-1000 mg kg⁻¹). These results suggest that the optimal level of MOS could effectively reduce skin morbidity.
Antioxidant Properties of MOS
To investigate whether MOS have antioxidant properties in vitro, we designed MOS with different concentrations to test their antioxidant properties (Supplementary Figure 1). Our results showed that the free radical scavenging rate (DPPH, ASA, and AHR) increased gradually (from 0 to 60%) with the increase in the dosage of MOS (from 0 to 5 mg ml⁻¹) in a dose-dependent manner. These data suggest that MOS has excellent antioxidant properties.
Biochemical Analysis Parameters
To uncover the MOS effect on fish skin antioxidant capacity with A. hydrophila challenge, we determined the content of oxidative damage biomarkers and the activity of key antioxidant enzymes (Figures 2A–L). Oxidative damage biomarkers are indicators that reflect the state of oxidative damage. In Figures 2A–E, compared with the control diet (ROS: 100% DCF florescence; MDA: 9.88 nmol g⁻¹ tissue; ASA: 64.71 U g⁻¹ protein), the ROS and MDA contents were decreased, the ASA were increased with the MOS supplementation, and ROS and MDA reached their minimum value (ROS: 53.20% DCF florescence; MDA: 6.88 nmol g⁻¹ tissue, P < 0.05), whereas ASA reached its maximum value (ASA: 71.46 U g⁻¹ protein, P < 0.05) with 400 mg kg⁻¹ MOS supplementation. Then ROS and MDA showed an upward trend (ROS: from 56.08 to 75.50% DCF florescence, P < 0.05; MDA: from 7.07 to 9.60 nmol g⁻¹ tissue), and ASA showed a downward trend (ASA: from 70.28 to 64.94 U g⁻¹ protein) with the increase in MOS (600–1,000 mg kg⁻¹). The PC contents were significantly reduced, and the AHR were increased with MOS supplementation with MOS at 600 mg kg⁻¹ diet; at this optimal MOS supplementation, the PC content was obviously reduced to a minimum 2.47 nmol mg⁻¹ protein, and AHR was increased to a maximum 117.4 U mg⁻¹ protein (P < 0.05) compared with the control group (PC: 4.10 nmol mg⁻¹ protein; AHR: 97.12 U mg⁻¹ protein). Then PC showed an upward trend (from 3.76 to 3.98 nmol mg⁻¹ protein), and AHR showed a downward trend (from 66.52 to 64.94 U mg⁻¹ protein) with the increase in MOS (800–1,000 mg kg⁻¹). These data suggest that the MOS could effectively alleviate oxidative damage caused by A. hydrophila.
Antioxidant enzymes are key proteins that scavenge free radicals, and their activities reflect antioxidant capacity. Figures 2F–L presents the results obtained from the biochemical analysis of the antioxidant enzymes and non-enzymatic antioxidants (GSH). Compared with the control diet (CuZnSOD: 5.85 U mg⁻¹ protein; MnSOD: 5.70 U mg⁻¹ protein; GPx: 78.00 U g⁻¹ protein; GST: 120.54 U mg⁻¹ protein, GSH: 5.28 mg g⁻¹ protein), the activity of CuZnSOD, MnSOD, GPx, and GST and the content of GSH were increased with the MOS supplementation, and all of them reached their maximum value (CuZnSOD: 7.18 U mg⁻¹ protein; MnSOD: 6.28 U mg⁻¹ protein; GPx: 87.11 U g⁻¹ protein; GST: 142.34 U mg⁻¹ protein, GSH: 8.49 mg g⁻¹ protein, P < 0.05) with 400 mg kg⁻¹ MOS supplementation. Then all of them showed a downward trend (CuZnSOD: from 6.84 to 5.94 U mg⁻¹ protein; MnSOD: from 5.71 to 5.26 U mg⁻¹ protein; GPx: from 84.77 to 76.73 U mg⁻¹ protein; GST: from 136.80 to 101.19 U mg⁻¹ protein, GSH: from 7.55 to 5.88 mg g⁻¹ protein) with the increase in MOS (600–1,000 mg kg⁻¹). The activity of CAT and GR increased with the MOS supplementation, and both of them reached their maximum value (CAT: 4.85 U mg⁻¹ protein; GR: 33.23 U mg⁻¹ protein, P < 0.05) with 600 mg kg⁻¹ MOS supplementation compared with the control group (CAT: 4.01 U mg⁻¹ protein; GR: 17.79 U mg⁻¹ protein). Then both of them showed a downward trend (CAT: from 4.09 to 3.98 U mg⁻¹ protein; GR: from 25.90 to 26.12 U mg⁻¹ protein) with the increase in MOS (800–1,000 mg kg⁻¹).
The antimicrobial compound-related parameters are displayed in Figures 2M–Q. Compared with the control (LZ: 248.64 U mg⁻¹ protein; ACP: 209.33 U mg⁻¹ protein; C3: 35.01 mg g⁻¹ protein; C4: 5.97 mg g⁻¹ protein), the activity of LZ and ACP and the contents of C3 and C4 were increased with the MOS supplementation. And all of them reached their maximum value (LZ: 277.72 U mg⁻¹ protein; ACP: 412.41 U mg⁻¹ protein; C3: 49.19 mg g⁻¹ protein; C4: 7.17 mg g⁻¹ protein, P < 0.05) with 600 mg kg⁻¹ MOS supplementation. Then all of them showed a downward trend (LZ: from 269.13 to 265.80 U mg⁻¹ protein; ACP: from 373.30 to 355.22 U mg⁻¹ protein; C3: from 36.29 to 30.57 mg g⁻¹ protein; C4: from 6.65 to 6.28 mg g⁻¹ protein) with the increase in MOS (800–1,000 mg kg⁻¹). The IgM content was significantly increased with MOS supplementation with MOS at
400 mg kg⁻¹ diet; at this optimal MOS supplementation, the IgM content was obviously increased to a maximum 114.49 mg g⁻¹ protein (P < 0.05) compared with the control group (IgM: 94.90 mg g⁻¹ protein). Then it showed a downward trend (IgM: from 103.00 to 91.59 mg g⁻¹ protein) with the increase in MOS (600 – 1,000 mg kg⁻¹).
Skin Physic Barrier Function Gene Expression
To further determine the MOS effect on fish skin physic barrier function with A. hydrophila challenge, the mRNA expression of the antioxidant, apoptosis, and tight junction-related gene was examined by real-time RT-PCR (Figures 3A-C). The enzymatic antioxidant pathway is an important part of the antioxidant system in fish (46). Figure 3A provides the heat map of the antioxidant-related gene expression. In comparison with the control group, almost all antioxidant enzyme-related isoforms, CuZnSOD (1.84-fold change), MnSOD (1.70-fold change), CAT (1.98-fold change), GR (1.79-fold change), GPx1a (1.69-fold change), GPx1b (1.63-fold change), GPx4a (1.69-fold change), GSTp1 (1.49-fold change), GSTp2 (1.61-fold change), and GSTo1 (1.65-fold change), were significantly upregulated with optimal MOS supplementation up to 400 mg kg⁻¹ (P < 0.05), and GSTo4b (1.87-fold change), GSTo2 (1.81-fold change), and GSTR (1.53-fold change) were significantly upregulated with optimal MOS supplementation up to 600 mg kg⁻¹ (P < 0.05); then all of them followed a gradual downward trend with the increase in MOS (600–1,000 mg kg⁻¹). Conversely, the keap1a (0.56-fold change) mRNA level had a significant downward trend with 400 mg kg⁻¹ MOS supplementation (P > 0.05).
and then plateaued. However, one of the interesting results we found was that the MOS supplementation did not affect keap1b mRNA levels. As expected, our antioxidant gene expression data were consistent with enzyme activities results, suggesting that the optimal level of MOS could enhance the antioxidant capacity of fish skin under A. hydrophila challenge.
We investigated the effect of MOS on the apoptosis level by DNA fragmentation and determination of apoptotic pathway gene expression (Supplementary Figure 2 and Figure 3B). Supplementary Figure 2 provides the visualization results that revealed that skin DNA showed an obvious fragmentation after being challenged (control group). Interestingly, MOS supplementation (600 and 800 mg kg\(^{-1}\)) performed the obvious reduction of DNA fragmentation. In Figure 3B, our results showed that compared with the control, the pro-apoptotic factors, Caspase-3 (0.67-fold change), Caspase-7 (0.31-fold change), FasL (0.44-fold change), BAX (0.51-fold change), and p38MAPK (0.61-fold change), were significantly downregulated with optimal MOS supplementation up to 400 mg kg\(^{-1}\) (\(P < 0.05\)), and then all of them followed a gradual downward trend with the increase in MOS (600–1,000 mg kg\(^{-1}\)) compared with the control group. However, we found that the MOS supplementation did not affect JNK mRNA levels. These data suggest that the MOS could effectively inhibit fish skin excessive apoptosis caused by A. hydrophila.
The tight junction proteins contribute to the skin barrier function (47). In Figure 3C, our results showed that compared with the control, most TJ protein genes, ZO-1 (1.95-fold change), ZO-2 (1.80-fold change), Occludin (1.87-fold change), Claudin-3c (1.56-fold change), Claudin-7b (1.73-fold change), Claudin-11 (1.74-fold change), Claudin-12 (1.61-fold change), Claudin-15a (1.77-fold change), and Claudin-15b (1.39-fold change), were significantly upregulated with optimal MOS supplementation up to 600 mg kg\(^{-1}\) (\(P < 0.05\)), and Claudin-c (1.83-fold change), Claudin-f (1.93-fold change), and Claudin-7a (1.46-fold change) were significantly upregulated with optimal MOS supplementation up to 600 mg kg\(^{-1}\) (\(P < 0.05\)); then all of them followed a gradual downward trend with the increase in MOS (600–1,000 mg kg\(^{-1}\) or 800–1,000 mg kg\(^{-1}\)). Furthermore, the key regulation molecules MLCK mRNA levels (0.58-fold change) were significantly downregulated with optimal MOS supplementation up to 400 mg kg\(^{-1}\) (\(P > 0.05\)) and plateaued with the increase in MOS (600–1,000 mg kg\(^{-1}\)) compared with the control group. We also found that the MOS supplementation did not affect Claudin-b mRNA levels. These data suggest that MOS
could enhance tight junctions of fish skin under *A. hydrophila* challenge.
**Skin Immune Barrier Function**
**Gene Expression**
To investigate the effect of MOS on fish skin immune barrier function with *A. hydrophila* challenge, the mRNA expression of the pro-inflammatory cytokines and anti-inflammatory cytokines and key signaling molecule gene was examined by real-time RT-PCR (Figure 3D). As is well known, inflammatory cytokines are crucial for fighting off infections and are involved in immune responses (48). In Figure 3D, compared with the control, the expression of pro-inflammatory cytokines, IL-1β (0.48-fold change), TNF-α (0.50-fold change), IL-6 (0.57-fold change), IL-12p40 (0.48-fold change), IL-15 (0.43-fold change), and IL-17D (0.47-fold change), was significantly downregulated with MOS supplementation up to 400 mg kg⁻¹ (*P* < 0.05), that of IFNγ2 (0.65-fold change) and IL-12p35 (0.47-fold change) was significantly downregulated with MOS supplementation up to 600 mg kg⁻¹ (*P* < 0.05), and IL-8 (0.58-fold change) was significantly downregulated with MOS supplementation up to 800 mg kg⁻¹ (*P* < 0.05), followed by a gradual upward trend or plateau with the increase in MOS (600–1,000 mg kg⁻¹). Besides, the anti-inflammatory cytokine factors IL-4/13A (1.64-fold change) and IL-11 (1.66-fold change) were significantly upregulated with MOS supplementation up to 400 mg kg⁻¹ (*P* > 0.05), and TGFβ1 (1.76-fold change), TGFβ2 (1.42-fold change), and IL-10 (1.79-fold change) were significantly upregulated with MOS supplementation up to 600 mg kg⁻¹ (*P* > 0.05), followed by a gradual downward trend with the increase in MOS (600–1,000 mg kg⁻¹ or 800–1,000 mg kg⁻¹), compared with the control group. Our results showed that the MOS supplementation did not affect IL-4/13B mRNA levels.
Many inflammatory cytokines could be mediated by NFKB and the TOR signaling pathway (29, 49). The present study displayed that compared with the control, the expression of several inflammatory signal molecular-related genes, NFXBp65 (0.57-fold change) and 4E-BP2 (0.59-fold change), was significantly downregulated with MOS supplementation up to 400 mg kg⁻¹, that of NFXBp52 (0.55-fold change), c-Rel (0.61-fold change), and 4E-BP1 (0.58-fold change) was significantly downregulated with MOS supplementation up to 600 mg kg⁻¹, that of IKKβ (0.75-fold change) and IKKγ (0.64-fold change) was significantly downregulated with MOS supplementation up to 800 mg kg⁻¹, followed by a gradual upward trend with the increase in MOS (600–1,000 mg kg⁻¹). Besides, IKBα (1.98-fold change) and TOR (2.11-fold change) were significantly upregulated with MOS supplementation up to 400 mg kg⁻¹, and S6K1 (1.92-fold change) was significantly upregulated with MOS supplementation up to 600 mg kg⁻¹, followed by a gradual downward trend with the increase in MOS (600–1,000 mg kg⁻¹), compared with the control group. Our results showed that the MOS supplementation did not affect IKKα mRNA levels. These results suggest that MOS is involved in the regulation of inflammatory cytokines under *A. hydrophila* challenge.
**Correlation Analysis**
To investigate the correlation between the expression of genes related to the skin barrier function and the signal molecules involved in regulation, correlation analysis was performed. Figures 4A–D provides the diagram of the correlation analysis. These data showed the gene expression correlation analyses of physic barrier-related parameters and immune barrier-related parameters. Gene expression of studied antioxidant enzymes revealed a positive correlation with Nrf2 mRNA levels, whereas Keap1a and Keap1b revealed a negative correlation. Gene expression of the studied pro-apoptotic factor showed a positive correlation with p38MAPK, whereas the anti-apoptotic factor showed a negative correlation. Gene expression of studied TJ proteins (except Claudin-b) showed a negative correlation with MLCK. Besides, gene expression of studied pro-inflammatory cytokine factors presented a positive correlation with NFκB, and the anti-inflammatory cytokine presented a positive correlation with TOR.
**Key Role Protein Levels of Skin Barrier Function**
To verify the results of skin barrier function expression, we further performed Western blot analysis to test several key regulatory signaling molecules. The protein expression of Nrf2, TOR, and NFκB p65 in the skin of fish is exhibited in Figures 5A–C, respectively. Compared with the control group, the nuclear Nrf2 (1.42-fold change) in the skin of fish was elevated with MOS supplementation up to 400 mg kg⁻¹ (*P* > 0.05) and then plateaued with MOS supplementation up to 1,000 mg kg⁻¹. Besides, fish fed with 600 and 400 mg kg⁻¹ MOS presented the maximum p-TOR Ser2448 (1.80-fold change) and total TOR (T-TOR) (1.33-fold change) expression (*P* < 0.05), respectively, and then gradually decreased with MOS supplementation up to 1,000 mg kg⁻¹ compared with the control group. With dietary MOS supplementation up to 600 mg kg⁻¹, NFκB p65 expression (0.70-fold change) weakened obviously (*P* < 0.05) and then gradually increased with MOS supplementation up to 1,000 mg kg⁻¹. As expected, these results of protein expression were consistent with those of gene expression.
**DISCUSSION**
This research used the same growth trial from our previous work in grass carp (35), which is a part of a larger study conducted to investigate the protective effect of fish skin barrier function by MOS supplementation. Our previous works have demonstrated that optimal MOS supplementation could promote fish growth and improve multiple functional organs (such as intestine, head-kidney, and spleen) health (35, 37). As is well known, fish growth and development are closely related to skin health (1). Therefore, to investigate the effects of prebiotics on fish skin health, we conducted relevant experiments based on previous studies.
MOS Supplementation Enhanced Skin Disease Resistance
As is well known, skin health is mainly reflected by disease resistance (50). *Aeromonas hydrophila* is one of the most common pathogenic microorganisms associated with the aquatic environment, which could cause skin lesions in fish (51). In this study, our results displayed that optimal MOS (400 mg kg⁻¹) could decrease skin lesion morbidity (8.27%) after being challenged while the control group caused skin lesion morbidity (14.40%), indicating that MOS supplementation enhanced fish resistance against skin lesions. Our data also showed that MOS supplementation attenuated skin hemorrhages and lesions, which suggested that MOS supplementation enhanced the ability to resist *A. hydrophila* invasion. Based on the quadratic regression analysis, the recommend suitable MOS supplementation against skin lesions morbidity was estimated to be 508.2 mg kg⁻¹. Generally, skin health is closely related to physical barriers and immune barriers in fish (1). Therefore, at first, we investigated the effects of MOS supplementation on physical barrier function in the skin of on-growing grass carp.
MOS Supplementation Enhanced Skin Physical Barrier Function
As mention above, the physical barrier function of the skin is related to cellular integrity and intercellular integrity, which were related to antioxidant capacity, apoptosis, and tight junction. Generally, MDA and PC were usually recognized to reflect the level of cell damage resulting from reactive oxygen (ROS) metabolites, which could be reduced by the antioxidant system (52, 53). We found that optimal MOS dosage decreased the biomarker content of oxidative damage of lipid and protein, whereas it enhanced the antioxidant enzyme activities. These data implied that MOS supplementation enhanced the antioxidant capacity to inhibit oxidative damage in fish skin. In general, antioxidant enzyme activities were strongly associated with their corresponding mRNA gene expression (54). We found that antioxidant enzymes and related isoform gene expression were upregulated by optimal MOS supplementation in the skin, indicating that MOS-enhanced activity of the antioxidant enzyme might be partly related to the upregulation of their mRNA levels. To our knowledge, Nrf2 is a major factor accounting for promoting the expression of various antioxidant enzymes.
enzyme genes to defend against oxidative stress, which is degenerated by Keap1 in the nucleus (55, 56). A study on mice liver showed that the Nrf2 protein level in the nucleus could evaluate the nuclear translocation of Nrf2 (57). Our result showed that MOS supplementation upregulated Nrf2 and downregulated Keap1a (rather than Keap1b) and increased the protein levels of nucleus Nrf2, suggesting that MOS supplementation activated the Nrf2 signaling pathway by the activation of Nrf2 nuclear translocation in the skin. Notably, we found that MOS only downregulated the Keap1a expression in the skin, which might be partly relevant to threonine. A study on piglets revealed that threonine absorbed from the intestine could be enhanced by MOS supplementation (58). Our lab previously has confirmed that threonine has no influence on Keap1b gene expression in the grass carp gill (59). Thus, these data might partially support our hypothesis. However, the specific mechanism needs further investigation. In addition, a study reported that excessive oxidative damage could induce cell apoptosis in MN9D cells (60). Therefore, we further examined the effects of MOS supplementation on fish skin apoptosis.
Apoptosis, a tightly controlled physiological process, and internal environment homeostasis, plays important roles not only in the normal development and homeostasis of organisms but also in the pathogenesis of bacterial infections (61). However, excessive apoptosis could destroy the physical barrier of the skin in fish (62). In mammals, there are two major apoptosis pathways, the death receptor pathway (FasL/caspase-8) and the mitochondria pathway [(Bcl-2, Mcl-1, and Bax)/Apaf-1/caspase-9], which were modulated by signal molecule p38MAPK and JNK (63–65). These two apoptosis pathways converge on caspase-3 activation, which is the key apoptotic protein. As is well known, the apoptosis-related protein includes the apoptotic promoter (caspase-8 and caspase-9) and effector (caspase-3 and caspase-7). In addition, DNA fragmentation is a hallmark of apoptosis (66). The visualization of apparent index results clearly showed that the level of apoptosis was significantly reduced with MOS supplementation. Our gene expression results also displayed that the optimal MOS supplementation could suppress the excess apoptosis process under-challenged, which was partly associated with p38MAPK (not JNK), leading to the inhibition of both apoptosis pathways in fish skin. As mentioned above, intercellular structure integrity also played a crucial role in the physical barrier, which is associated with TJ proteins (67). Thus, we next examined the influences of MOS supplementation on TJs as well as the related signaling pathway in fish skin.
The intercellular junction complex function has maintained the integrity of the skin barrier, which mainly consists of TJ proteins (68, 69). It has been reported that inhibition of MLCK expression could improve epithelial TJ barrier function in Caco-2 cells (70). Our result displayed that optimal MOS upregulated the expression of most of the tight junction proteins (except claudin-b) and downregulated MLCK, suggesting that MOS improved tight junction partly by inhibiting the MLCK signaling pathway. We surprisingly found that MOS did not affect claudin-b gene expression, which could involve both IL-6 and cortisol. Our result exhibited that MOS supplementation could downregulate IL-6 gene expression. Steensberg et al. (71) confirmed that IL-6 could increase the content of cortisol in humans. Studies showed that cortisol did not affect claudin-b mRNA levels in the gill epithelial cell of pufferfish and goldfish (72, 73), which supports our hypothesis. However, determining the underlying mechanism warrants further investigation.
**MOS Supplementation Enhanced Skin Immune Barrier Function**
To our knowledge, the existence and function of the secretory cell in teleost skin (such as mucous goblet cells, squamous cells, pigment cells, and so on) have been confirmed and provided the first line of defense against pathogen invasion (1, 74). The mucus secreted by these cells contains a large number of antimicrobial substances (75, 76). Previous studies have demonstrated that MOS can increase the LZ activity and bactericidal activity in the skin of greater amberjack (13). The present study focuses on antibacterial compounds, and the results revealed that MOS could promote LZ production in the skin of grass carp, agreeing with previous findings in the skin of greater amberjack. Coincidentally, we also found a study that showed that other prebiotics also have antimicrobial properties, which reported the antimicrobial ability to be enhanced in the skin of Caspian white fish (Rutilus frisii kutum) with xylooligosaccharide (77). These interesting results partly reflect the commonality of prebiotic to improve skin antimicrobial capacity. In addition, the skin immune function is closely related to the inflammatory response mediated by cytokines (14). Thus, we next examined the effects of MOS supplementation on fish skin immune barrier function.
In the immune system, there is a dynamic balance between pro-inflammatory cytokines and anti-inflammatory cytokines. The imbalance of inflammatory cytokines caused by external stimuli (pathogenic bacteria) is one of the causes of the excessive inflammatory response (78). A study on channel catfish, Ictalurus punctatus, revealed that Actigen® (a commercial MOS product from Alltech) could improve inflammatory cytokine balance in multiple mucosal immune organs by using RNA-seq, indicating that MOS additives may provide protection extending beyond the intestine to surface mucosa (79). As we expected, our result displayed that optimal MOS dosage downregulated pro-inflammatory cytokine expression; in contrast to the former, the anti-inflammatory cytokine (except IL-4/13B) expression was upregulated, indicating that MOS supplementation attenuated the inflammation in fish skin. Notably, part of these data differed with other similar studies (parasite challenged) in the skin of greater amberjack, which found that TNFα, IL-1β, IFNγ, and IL-8 were upregulated by MOS supplementation (2 g kg⁻¹) (13). Differences in species, MOS purity, and challenged type might account for this disparity. Notably, another interesting result showed that dietary MOS only upregulated IL-4/13A expression in the skin. This phenomenon might be associated with the content of phosphorus. A study on weaned piglets confirmed that MOS increased the digestibility of phosphorus (30). Past work in our lab has confirmed that phosphorus has no effect on the IL-4/13B expression, and our results also showed that dietary MOS did the same effect on IL-4/13B expression (32). Thus, we speculated that MOS supplementation upregulates the IL-4/13A (rather than IL-4/13B), which might relate to improving the digestibility of phosphorus, thus leading to a disposition of only upregulated IL-4/13A in fish.
As we all know, the pro-inflammatory cytokines could be activated by the NFκB family of transcription factors (such as NFκB, p52, and c-Rel), which required a sequestering protein named IκBα that could be catalyzed by the IKK complex (IKKa, IKKβ, and IKKγ) (80, 81). We found that optimal MOS supplementation downregulated NFκB-related signal molecule (rather than IKKα) gene expression and decreased the protein levels of NFκB p65, suggesting that MOS supplementation activated the NFκB signaling pathway by decreasing the nuclear NFκB p65 protein expression in the skin. Interestingly, what is noteworthy of this study is that MOS supplementation did not have influence on IKKα in the skin; the possible reasons for this difference might be due to TNF-α and PKCζ. Our result revealed that MOS could downregulate TNF-α expression. A study on rat showed that downregulated TNF-α expression could decrease the activity of PKCζ (82), which could downregulate IKKβ and IKKγ (rather than IKKα) expression in Kupffer cells, and did not have an effect on IKKα expression (83), supporting our hypothesis. However, the underlying molecular mechanism is still unknown and warrants further investigation. In addition, it has been reported that anti-inflammatory cytokines could be modulated by the mTOR/S6K1, 4EBP-1 signaling cascades in humans (84). One study on rainbow trout reported that the phosphorylation of TOR on residue Ser2448 can be used to monitor the activation of TOR signaling (85). We found that MOS supplementation downregulated 4EBP-1 and 4EBP-2 gene expression and upregulated TOR and S6K1 expression, and increased the protein levels of TOR and p-TOR Ser2448, suggesting that MOS supplementation upregulated the anti-inflammatory cytokine mRNA levels partly due to the activation of the TOR signaling pathway cascades in fish skin.
In summary, the current work presented a clear outline of dietary MOS enhanced fish skin immune barrier and physical barrier function after infection with A. hydrophila. Our study confirmed that dietary MOS supplementation could improve the status of skin health, as demonstrated by the following findings (1): MOS supplementation enhanced the immune barrier function via increasing the skin disease resistance, producing antibacterial compounds and immunoglobulins, upregulating anti-
inflammatory cytokines (except IL-4/13B), and downregulating pro-inflammatory cytokines gene expression (2). MOS supplementation protected the physical barrier function via increasing the antioxidant capacity, inhibited excessive apoptosis, and enhanced the tight junction barriers (except claudin-b). Moreover, MOS supplementation improved fish physical and immune barrier function by modulating multiple signaling pathways (such as Nrf2, TOR, NFlxB, and so on).
DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.
ETHICS STATEMENT
The animal study was reviewed and approved by Laboratory Animals Care and Use of Animal Nutrition Institute of Sichuan Agricultural University.
AUTHOR CONTRIBUTIONS
ZyL performed formal analysis, investigation and writing original draft. LF performed conceptualization, funding acquisition, and writing review & editing. PW performed conceptualization, methodology, validation, data curation and project administration. YL and JJ performed project administration. YL and JJ performed project supervision and funding acquisition. All authors contributed to the article and approved the submitted version.
FUNDING
This research was financially supported by the National Key R&D Program of China (2019YFD0900200 and 2018YFD0900400), the National Natural Science Foundation of China for Outstanding Youth Science Foundation (31922086), and the Young Top-Notch Talent Support Program, Supported by China Agriculture Research System of MOF and MARA (CARS-45), and supported by Sichuan Science and Technology Program (2019YFN0036). The authors would like to thank the personnel of these teams for their kind assistance.
SUPPLEMENTARY MATERIAL
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fimmu.2021.742107/full#supplementary-material
REFERENCES
1. Ángeles Esteban M. An Overview of the Immunological Defenses in Fish Skin. ISRN Immunol (2012) 2012:1–29. doi: 10.5402/2012/853470
2. Alvarez-Pellitero P. Fish Immunity and Parasite Infections: From Innate Immunity to Immunomodulatory Prospects. Vet Immunol Immunopathol (2008) 126:171–98. doi: 10.1016/j.vetimm.2008.07.013
3. Fast MD, Sims DE, Burka JF, Mustafa A, Ross NW. Skin Morphology and Humoral non-Specific Defence Parameters of Mucous and Plasma in Rainbow Trout, Coho and Atlantic Salmon. Comp Biochem Physiol Part A Mol Integr Physiol (2002) 132:645–57. doi: 10.1016/s1095-6433(02)00109-5
4. Noga EJ, Skin Ulcers in Fish: Pfiesteria and Other Etiologies. Toxicol Pathol (2000) 28:807–23. doi: 10.1177/019262330002800667
5. Bindels LB, Delzenne NM, Cani PD, Walter J. Towards a More Comprehensive Concept for Prebiotics. Nat Rev Gastroenterol Hepatol (2015) 12:303–10. doi: 10.1038/nrgastro.2015.47
6. Nawaz A, Bakhsh Javid A, Ishad S, Hoseinifar SH, Xiong H. The Functionality of Prebiotics as Immunostimulant: Evidences From Trials on Terrestrial and Aquatic Animals. Fish Shellfish Immunal (2018) 76:272–8. doi: 10.1016/j.fsi.2018.03.004
7. Hoseinifar SH, Ahmad M, Ateis M, Hoseini SM, Khallil M, Behnampour N. Comparative Study on Immunomodulatory and Growth Enhancing Effects of Three Prebiotics (Galactooligosaccharide, Fructooligosaccharide and Inulin) in Common Carp (Cyprinus Carpio). Aquacult Res (2017) 48:3298–307. doi: 10.1111/are.13156
8. Modanloo M, Soltanian S, Akhlaghi M, Hoseinifar SH. The Effects of Single or Combined Administration of Galactooligosaccharide and Pediococcus Acidilactici on Cutaneous Mucus Immune Parameters, Humoral Immune Responses and Immune Related Genes Expression in Common Carp (Cyprinus Carpio) Fingerlings. Fish Shellfish Immun (2017) 70:391–7. doi: 10.1016/j.fsi.2017.09.032
9. Van Doan H, Hoseinifar SH, Faggio C, Chitmanat C, Mai NT, Jaturasitha S, et al. Effects of Corncoth Derived Xyloooligosaccharide on Immune Immune Response, Disease Resistance, and Growth Performance in Nile Tilapia (Oreochromis Niloticus) Fingerlings. Aquaculture (2018) 495:786–93. doi: 10.1016/j.aquaculture.2018.06.068
10. Gelibolu S, Yanar Y, Genc MA, Genc E. The Effect of Mannan- Oligosaccharide (MOS) as a Feed Supplement on Growth and Some Blood Parameters of Gilthead Sea Bream (Sparus Atrata). Turkish J Fisheries Aquat Sci (2018) 18:817–23. doi: 10.4194/1303-2712-v18_6_08
11. Uriel R. Effects of Single and Combined Supplementation of Enterococcus Faecalis, Mannan Oligosaccharide and Polyhydroxybutyrate Acid on Growth Performance and Immune Response of Rainbow Trout Oncorhynchus Mykiss. Aquaculture (2009) 57:609–14. doi: 10.11233/aquaculturesci.57.609
12. Torrecillas S, Montero D, Caballero MJ, Pittman KA, Custodio M, Campo A, et al. Dietary Mannan Oligosaccharides: Counteracting the Side Effects of Soybean Meal Oil Inclusion on European Sea Bass (Dicentrarchus Labrax) Gut Health and Skin Mucosa Mucus Production? Front Immunol (2015) 6:397. doi: 10.3389/fimmu.2015.00397
13. Fernandez-Montero A, Torrecillas S, Izquierdo M, Caballero MJ, Milne DJ, Secombes CJ, et al. Increased Parasite Resistance of Greater Amberjack (Seriola Dumerili Risso 1810) Juveniles Fed a cMOS Supplemented Diet is Associated With Upregulation of a Discrete Set of Immune Genes in Mucosal Tissues. Fish Shellfish Immun (2019) 86:35–45. doi: 10.1016/j.fsi.2018.10.034
14. Zheng L, Feng L, Jiang W-D, Wu P, Tang L, Kuang S-Y, et al. Selenium Deficiency Impaired Immune Function of the Immune Organs in Young
Grass Carp (Ctenopharyngodon Idella). Fish Shellfish Immunol (2018) 77:53–70. doi: 10.1016/j.fsi.2018.03.024
15. Homemop, L, Isihy Y, Morishima Y, Yamadori T, Matsuno Y, Haraguchi N, et al. Nrf2 Enhances Cell Proliferation and Resistance to Anticancer Drugs in Human Lung Cancer. Clin Cancer Res (2009) 15:3423–32. doi: 10.1158/1078-0432.CCR-08-2822
16. Ki YW, Park JH, Lee JE, Shin IC, Koc HC. JNK and P38 MAPK Regulate Oxidative Stress and the Inflammatory Response in Chlorpyrifos-Induced Apoptosis. Toxicol Lett (2013) 218:235–45. doi: 10.1016/j.toxlet.2013.02.003
17. Ma TY, Boivin MA, Ye D, Pedram A, Said HM. Mechanism of TNF-Alpha Modulation of Caco-2 Intestinal Epithelial Tight Junction Barrier: Role of Myosin Light-Chain Kinase Protein Expression. Am J Physiol Gastrointest Liver Physiol (2005) 288:G422. doi: 10.1152/ajpgi.00412.2004
18. Baker OJ, Camden JM, Redman RS, Jones JE, Seye CI, Erb L, et al. Proinflammatory Cytokines Tumor Necrosis Factor-Alpha and Interferon-Gamma Alter Tight Junction Structure and Function in the Rat Parotid Gland Par-C10 Cell Line. Am J Physiol Cell Physiol (2008) 295:C1191. doi: 10.1152/ajpcell.00144.2008
19. Ibuki M, Kovacs-Nolan J, Fukui K, Kanatani H, Mine Y. Beta-1-4 Mannobiose Enhances Salmonella-Killing Activity and Activates In innate Immune Responses in Chicken Macrophages. Vet Immunol Immunopathol (2011) 139:289–95. doi: 10.1016/j.vetimm.2010.10.011
20. Um HC, Jang JH, Kim DH, Lee C, Surh YJ. Nitric Oxide Activates Nrf2 Through S-Nitrosylation of Keap1 in PC12 Cells. Nitric Oxide: Biol Chem (2011) 25:161–8. doi: 10.1016/j.niox.2011.06.001
21. Hassanein SM, Soliman NK. Effect of Probiotic (Saccharomyces Cerevisiae) Adding to Diets on Intestinal Microflora and Performance of Hy-Line Layers Hens. J Am Sci (2010) 6:159–69. doi: am0611/21_3171am0611_159_169
22. Kim SH, Sharma RP. Mercury-Induced Apoptosis and Necrosis in Murine Human Lung Cancer. Toxicol Appl Pharmacol (2004) 196:47–57. doi: 10.1016/j.taap.2003.11.020
23. Torrecillas S, Montero D, Caballero MJ, Robaina L, Zamorano MJ, Sweetman L, et al. Copper-Induced Apoptosis. Toxicol Lett (2013) 226:161–72. doi: 10.1016/j.toxlet.2013.02.003
24. Chen K, Jiang WD, Wu P, Liu Y, Kuang SY, Tang L, et al. Effect of Dietary Phosphorus Deficiency on the Growth, Immune Function and Structural Integrity of Head Kidney, Spleen and Skin in Young Grass Carp (Ctenopharyngodon Idella). Fish Shellfish Immunol (2017) 63:103–26. doi: 10.1016/j.fsi.2017.02.027
25. Pinheiro V, Alves A, Mourao JL, Guedes CM, Pinto L, Spring P, et al. Effect of Mannan Oligosaccharides on the Ileal Morphometry and Cecal Fermentation of Growing Rabbits, in: Proceedings of the 8th World Rabbit Congress, Pueblo, Mexico, September 7–10, 2004. pp. 936–41
26. Tian L, Zhou XQ, Jiang WD, Liu Y, Wu P, Jiang J, et al. Sodium Butyrate Improved Intestinal Immune Function Associated With NF-kappaB and P38 MAPK Signalling Pathways in Young Grass Carp (Ctenopharyngodon Idella). Fish Shellfish Immunol (2017) 66:548–63. doi: 10.1016/j.fsi.2017.05.049
27. Lu Z-Y, Feng L, Jiang W-D, Wu P, Liu Y, Kuang S-Y, et al. Mannan Oligosaccharides Improved Growth Performance and Antioxidant Capacity in the Intestine of on-Growing Grass Carp (Ctenopharyngodon Idella). Aquaculture (2020) 17:100313. doi: 10.1016/j.aqrep.2020.100313
28. Bacon AGJ. Trends in Global Aquaculture and Aquafeed Production: 2000–2017. Rev Fisheries Sci Aquatul (2019) 28:43–56. doi: 10.1080/23308249.2019.1649634
29. Lu ZY, Jiang WD, Wu P, Liu Y, Kuang SY, Tang L, et al. Mannan Oligosaccharides Supplementation Enhanced Head-Kidney and Spleen Immune Function in on-Growing Grass Carp (Ctenopharyngodon Idella). Fish Shellfish Immunol (2020) 106:596–608. doi: 10.1016/j.fsi.2020.08.035
30. Xu S, Zhang Y, Jiang K. Antioxidant Activity In Vitro and In Vivo of the Polysaccharides From Different Varieties of Auricularia Auricula. Food Funct (2016) 7:3868–79. doi: 10.1039/c6fo00686b
31. Zhang YH, Song XN, Lin Y, Xiao Q, Du XP, Chen YH, et al. Antioxidant Capacity and Prebiotic Effects of Gracilaria neoagaro Oligosaccharides Prepared by Agarase Hydrolysis. Int J Biol Macromol (2019) 137:177–86. doi: 10.1016/j.ijbiomac.2019.06.207
32. Seguin MJ, Barney D, Wodowski TM. Assessment of a Group-Housing System for Gestating Sows: Effects of Space Allowance and Pen Size on the Incidence of Superficial Skin Lesions, Changes in Body Condition, and Farrowing Performance. J Swine Health Product (2006) 14:89–96. doi: 10.1111/j.1748-8277.2006.00023.x
33. Wang B, Feng L, Jiang WD, Wu P, Kuang SY, Jiang J, et al. Copper-Induced Tight Junction mRNA Expression Changes, Apoptosis and Antioxidant Responses via NF-kappaB, TOR and Nrf2 Signaling Molecules in the Gills of Fish: Preventive Role of Arginine. Aquat Toxicol (2015) 158:125–37. doi: 10.1016/j.aquatox.2014.10.025
34. Su J, Zhang R, Dong J, Yang C. Evaluation of Internal Control Genes for qRT-PCR Normalization in Tissues and Cell Culture for Antiviral Studies of Grass Carp (Ctenopharyngodon Idella). Fish Shellfish Immunol (2011) 30:830–5. doi: 10.1016/j.fsi.2011.01.006
35. Tian B, Han R. Schmittgen T. Analysis of Relative Gene Expression Data Using Real-Time Quantitative PCR and the 2(-Delta Delta C(T)) Method. Methods (2001) 25:402–8. doi: 10.1006/mlm.2001.1262
36. Hu K, Zhang JX, Feng L, Jiang WD, Wu P, Kuang SY, Jiang J, et al. Cytokine Mediators of Immunity and Inflammation. Fish Shellfish Immunol (2015) 162:2939–9. doi: 10.1016/j.fsi.2015.03.007
37. Lowry SF. Cytokine Mediators of Immunity and Inflammation. Annu Rev Immunol (1998) 16:255–87. doi: 10.1146/annurev.immunol.16.1.225
Rainbow Trout (*Oncorhynchus Mykiss*). *Am J Physiol* (2008) 295:329–35. doi: 10.1152/ajpregu.00146.2008
**Conflict of Interest:** S-YK, LT, S-WL, X-Al and C-BZ were employed by Sichuan Animtech Feed Co. Ltd.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
**Publisher’s Note:** All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
*Copyright © 2021 Lu, Feng, Wu, Jiang, Kuang, Tang, Li, Liu, Zhong and Zhou. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.* | 2025-03-06T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2744,
1
],
[
2744,
9315,
2
],
[
9315,
14954,
3
],
[
14954,
19514,
4
],
[
19514,
21143,
5
],
[
21143,
23936,
6
],
[
23936,
29852,
7
],
[
29852,
32225,
8
],
[
32225,
34972,
9
],
[
34972,
41553,
10
],
[
41553,
47089,
11
],
[
47089,
52717,
12
],
[
52717,
52717,
13
],
[
52717,
54008,
14
]
]
} |
7c02ebd0885cef15c4d4177665b5c128d7ba498b | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 3,
"total-fallback-pages": 0,
"total-input-tokens": 10917,
"total-output-tokens": 3154,
"fieldofstudy": [
"Medicine"
]
} | The use of electronic alerts in primary care computer systems to identify the over-prescription of short-acting beta₂-agonists in people with asthma: a protocol for a systematic review
Shauna McKibben¹, Andy Bush², Mike Thomas³ and Chris Griffiths³
npj Primary Care Respiratory Medicine (2017) 27:30; doi:10.1038/s41533-017-0033-y
BACKGROUND
Asthma is a heterogeneous disease, usually characterised by (a) chronic airway inflammation with variable symptoms of wheeze, shortness of breath, chest tightness and/or cough, and (b) variable expiratory airflow limitation.¹ Despite increasing evidence-based guidelines for asthma gaps between recommended care and current practice remain.², ³ Frequent and increasing use of short-acting beta₂-agonists (SABA) or reliever therapy is a marker for poor asthma control and increased risk of asthma attacks;⁴ with control defined as the degree to which the manifestations of asthma are minimised by treatment.⁵ Asthma control can be assessed by reviewing both current symptoms and risk factors (modifiable or non-modifiable) of future asthma attacks.⁵, ⁶ Poor asthma control and risk of asthma attacks can be determined in part by SABA use,⁷–¹² with high or increasing SABA use a potentially modifiable risk factor for asthma attacks.⁸, ¹¹, ¹³, ¹⁴ and asthma related death.⁷, ⁸, ¹⁵ Poor asthma control is commonly due to suboptimal asthma management and can result not only in loss of school and workdays at a high cost for countries¹⁶–¹⁸ but also in unnecessary morbidity and even mortality.¹⁹ The National Review of Asthma Deaths identified that 39% of people who died from asthma had been prescribed more than 12 SABA inhalers in the year before death and 4% had been prescribed more than 50 SABA inhalers in the year before death.¹⁵ Recent figures show that asthma deaths are at the highest level for a decade with a 17% increase in the number of asthma related deaths from 2014 to 2015 in England and Wales.²⁰
Computer decision support systems (CDSSs) are increasingly being used to improve the prevention and management of chronic conditions such as asthma.²¹, ²² CDSSs include electronic alerts and reminders that use patient-specific information and clinical data to help healthcare providers make decisions that enhance patient care.²¹, ²² Whilst CDSSs have the potential to improve prescribing efficiency for healthcare professionals²³–²⁶ overall effectiveness in clinical practice is unclear.²⁷ Recommendations have called for the electronic surveillance of prescription refill frequency in primary care to alert clinicians to patients being prescribed excessive quantities of SABA¹⁵; however it is unclear to what extent alerts have been used in the management of SABA prescribing and what impact, if any, they have on patient outcomes.
AIMS
We aim to identify and critically appraise studies that have used electronic alerts to identify people with asthma being prescribed excessive SABA in primary care.
Specific objectives are as follows:
1. Evaluate the effectiveness of electronic alerts within CDSSs to identify people with asthma being prescribed excessive SABA in primary care.
2. Determine the features of electronic surveillance systems that have the potential to improve process outcomes for healthcare providers and clinical outcomes for people with asthma.
DISCUSSION AND CONCLUSION
CDSS interventions can potentially increase adherence to evidence-based medical knowledge, reduce unnecessary variation in clinical practice and improve clinical decision-making processes²⁹, ³⁰ particularly in the prevention and management of chronic conditions.²¹ Studies addressing the use of CDSSs in the care of people with asthma have shown varying results. One study reported that CDSSs had little effect on clinical and process outcomes for asthma due to low clinician use²² whilst another reported that CDSSs can improve chronic disease processes and outcomes particularly in the support of asthma self-management.²¹ Alerts represent an important category of decision support to clinicians, often having a substantial effect on prescribing behaviour.³¹ However few studies have assessed the impact of computerised alerts on clinical or health service management outcomes.³¹ Current recommendations include the use of alerts within general practice computer software to identify patients with asthma being prescribed excessive quantities of SABA.¹⁵, ³² A thorough synthesis of the evidence is required to: (i) determine the extent to which electronic alerts to identify people with asthma being prescribed excessive SABA in primary care can improve asthma management and patient outcomes; (ii) clarify the design and implementation of CDSSs alerts to improve asthma prescribing decisions for clinicians.
METHODS
Study eligibility criteria
Types of studies. We will include all types of randomised controlled trials in which patients have been treated by clinical teams informed by an
electronic SABA prescribing alert compared with usual care. As a surrogate measure of prescribing, studies using dispensing data will be included. We will exclude non-randomised trial designs (quasi-experimental, observational studies); study protocols; paper-based tools (e.g., flow charts and non-electronic clinical pathway tools); CDSS alerts used for conditions that are not asthma, e.g., COPD or other respiratory conditions; CDSS alerts used in secondary or tertiary care.
Types of participants. We will include studies involving healthcare professionals and non-clinical staff in primary care who provide care to adults and/or children with a physician coded asthma diagnosis.
Types of intervention. We will include studies which used CDSS based alerts initiated by the excessive prescribing of SABA for people with asthma. Definitions of excessive prescribing will be analysed on an individual study basis.
Types of outcome measures. The primary outcome will be study-defined SABA over-prescription. Secondary outcomes will be SABA prescribing, inhaled corticosteroid (ICS) prescribing alone or with a long-acting beta$_2$-agonist, the ratio of ICS-SABA prescribed, asthma reviews, study-defined asthma exacerbations, study-defined asthma exacerbations requiring oral steroids, unscheduled consultations for asthma (including general practice visits, emergency department visits and hospitalisations for asthma) and study-defined asthma control assessment.
Search strategy
We will search the international electronic databases: MEDLINE (Ovid), EMBASE (Ovid), CINAHL (Cumulative Index to Nursing and Allied Health Literature), SCOPUS (Elsevier) and Cochrane Library (Wiley). Additional studies will be retrieved by searching the references of eligible papers. Unpublished and in-progress studies will be identified by searching online trial registries; ISRCTN registry and ClinicalTrials.gov. All databases will be searched from 1990 to July 2016. No language restrictions will be imposed; translations will be sought where possible. Supplementary Appendix 1 presents details of our search strategy, which was developed for MEDLINE and will be adapted in searching other databases.
Screening of retrieved literature. The titles and abstracts of all papers retrieved from the databases will be checked independently by two reviewers against the criteria of the study. The full texts of papers that are potentially eligible will be retrieved and further assessed for inclusion independently by two reviewers. Discrepancies in the screening processes between the two reviewers will be resolved by consensus, and disagreements will be arbitrated by a third reviewer.
Data extraction
A customised data collection form will be used by two reviewers, independently, to extract relevant study data from full-text papers selected for inclusion. The form will be piloted and refined before being applied to full-text reports. Included papers will be discussed by the two reviewers after data extraction, and disagreements will be arbitrated by a third reviewer. Where necessary, clarification and additional data will be sought from study authors. Key findings from each included study will be summarised and tabulated.
Quality assessment
We will assess the risk of bias in each trial using the seven-criteria approach described in section eight of the Cochrane Handbook for Systematic Reviews of Interventions. Overall, each study will be rated as follows: A: low risk of bias—no bias found; B: moderate risk of bias—one criterion for risk of bias; C: high risk of bias—more than one criterion for risk of bias.
Data synthesis
Narrative synthesis of heterogeneous process outcomes (prescribing and asthma reviews) and clinical outcomes (exacerbations, unscheduled consultations and asthma control) will be conducted. Data will be presented in tabular form. Where possible, meta-analysis will be performed on process and clinical variables of interest, specifically: study-defined SABA over-prescription, study-defined asthma exacerbations and study-defined asthma control.
Heterogeneity will be assessed using the I-squared statistic. Where possible, subgroup analyses will be performed on age categories as defined by BTS/ SIGN Guidelines; less than 5 years, aged 5–12 years and greater than 12 years of age.
Registration and reporting
This study is registered with PROSPERO, the University of York Centre for Reviews and Dissemination International prospective register of systematic reviews (CRD42016035633). We will report according to the PRISMA guidelines for reporting systematic reviews.
ACKNOWLEDGEMENTS
The authors wish to thank the Asthma UK Centre for Applied Research for funding this work. AB is an National Institute for Health Research (NIHR) Senior Investigator and additionally was supported by the NIHR Respiratory Disease Biomedical Research Unit at the Royal Brompton and Harefield NHS Foundation Trust and Imperial College London. MT was supported by the NIHR Collaboration for Leadership in Applied Health Research and Care (CLAHRC) Wessex, NIHR School of Primary Care Research and NIHR Southampton Biomedical Research Centre. CG was supported by the NIHR CLAHRC North Thames at Bart’s Health NHS Trust. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health.
AUTHOR CONTRIBUTIONS
This work forms part of an Asthma UK Centre for Applied Centre PhD Studentship being undertaken by S.M. It was drafted by S.M. and revised after several rounds of critical comments from A.B., M.T. and C.G.
COMPETING INTERESTS
CG is an Assistant Editor of npj Primary Care Respiratory Medicine and MT is an Associate Editor of npj Primary Care Respiratory Medicine but were involved in neither the editorial review of, nor any decision to publish or not publish this article.
REFERENCES
1. Global Initiative for Asthma. Global strategy for asthma management and prevention. Available at http://www.ginasthma.org. Accessed 6 June 2016 (2016).
2. Klomp, H. et al. Examining asthma quality of care using a population-based approach. Can. Med. Assoc. J. 178, 1013–1021, doi:10.1503/cmaj.070426 (2008).
3. Vermeire, P. A., Rabe, K. F., Soriano, J. B. & Maier, W. C. Asthma control and differences in management practices across seven European countries. Respir. Med. 96, 142–149 (2002).
4. British Thoracic Society and Scottish Intercollegiate Guidelines Network. British guideline on the management of asthma. Available at https://www.brit-thoracic.org.uk/standards-of-care/guidelines/btssh Bethesda-British-guideline-on-the-management-of-asthma/. Accessed 30 September 2016 (2016).
5. Bouquet, J. et al. Uniform definition of asthma severity, control, and exacerbations: document presented for the World Health Organization Consultation on Severe Asthma. J. Allergy. Clin. Immunol. 126, 926–938 (2010).
6. Reddel, H. K. et al. A summary of the new GINA strategy: a roadmap to asthma control. Eur. Respir. J. 10.1183/13993003.00835-2015 (2015).
7. Suissa, S., Blais, L. & Ernst, P. Patterns of increasing beta-agonist use and the risk of fatal or near-fatal asthma. Eur. Respir. J. 7, 1602–1609 (1994).
8. Spitzer, W. O. et al. The use of beta-agonists and the risk of death and near death from asthma. N. Engl. J. Med. 326, 501–506, doi:10.1056/nejm199202203260801 (1992).
9. Schatz, M. et al. Asthma quality-of-care markers using administrative data. Chest. 128, 1968–1973, doi:10.1378/chest.128.4.1968 (2005).
10. Donahue, J. G. et al. Inhaled steroids and the risk of hospitalization for asthma. JAMA. 277, 887–891 (1997).
11. Schatz, M. et al. Validation of a beta-agonist long-term asthma control scale derived from computerized pharmacy data. J. Allergy. Clin. Immunol. 117, 995–1000, doi:10.1016/j.jaci.2006.01.053 (2006).
12. Diette, G. B. et al. Treatment patterns among adult patients with asthma: factors associated with overuse of inhaled beta-agonists and underuse of inhaled corticosteroids. Arch. Intern. Med. 159, 2697–2704 (1999).
| 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4946,
1
],
[
4946,
12974,
2
],
[
12974,
12974,
3
]
]
} |
a406e60d68da1c65bd3ae18059eb56c73e6c7dc9 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 29,
"total-fallback-pages": 0,
"total-input-tokens": 67869,
"total-output-tokens": 15369,
"fieldofstudy": [
"Engineering",
"Mathematics"
]
} | Abstract
This paper describes the design process for developing a nonlinear model predictive controller for fault tolerant flight control. After examining and implementing a number of numerical techniques, this paper identifies pseudospectral discretisation as the most suitable for this design. Applying the controller to a 2D robot model shows that the nonlinear controller performs much better than the linear controller, especially in the closed loop scenario. Assuming fault detection information, applying the technique to the longitudinal motion of a generic aircraft model shows the design to be eminently suitable for flight control.
Keywords: nonlinear model predictive control, pseudospectral, optimal control, 2D robot, flight control
1 Introduction
Most research on fault tolerant control sits within the context of large manned aircraft. With regards to unmanned aerial vehicles (UAVs), the majority of the literature describes the application of fault tolerant control (FTC) to rotorcraft rather than fixed wing aircraft. In this paper we develop a nonlinear model predictive control (NMPC) based controller for a fixed wing aircraft model suitable for the purposes of fault tolerant flight control. The controller is first tested and analysed on a lower order 2D robot model. Then assuming that fault information is available we successfully demonstrate for the first time the use of an NMPC based fault tolerant
flight control system for a fixed wing aircraft.
The main characteristic of a fault tolerant control (FTC) system is its ability to automatically cope with system faults before they turn into a serious system failure. The integration of an FTC scheme significantly increases the ability of the system to maintain overall stability in the presence of a fault \[10\]. Model predictive control, a highly promising approach to re-configurable and fault tolerant control (\[2\], \[3\], \[4\] and \[5\]), focuses on what to control, instead of how to control. This is a subtle, but illuminating, difference allowing for the exploitation of inherent system characteristics such as nonlinearities and cross-coupling effects by the controller, rather than their influence being minimised. This is possible due to the underlying daisy chaining capability of MPC \([11]\). For example \([1]\), the primary function of the rudder of an aircraft is to provide yaw or sideways control. However, the rudder can also have some effect on the roll of the aircraft. Therefore, in the event of the failure of an aileron actuator, the primary control surface for roll, it is still possible to execute a limited roll manoeuvre with the rudder. This degree of fault-tolerance in the flight control system requires a suitable re-configurable architecture to be purposefully designed and implemented \([6]\), which is able to re-establish control, albeit with limited capacity, and execute the required manoeuvres. The mission can then either be continued with the failed component or aborted, the primary objective being to avert a catastrophic failure or the loss of the aircraft, and to ensure that it is brought back to ground safely. The aim of this paper is to design a controller for exactly this purpose.
Much of the research in model predictive control (MPC) is based on linear MPC where a linear model of the plant is utilised. This paper develops an NMPC controller with the ultimate goal of application to fault tolerant flight control for UAVs. The next section (Section 2) will detail the workings of MPC, followed by a discussion of NMPC. The implementation of NMPC requires the solution of an optimal control problem at each time step, and in Section 2 several methods of achieving a solution are investigated. In Section 3 a small selection of these optimal control methods, namely direct single shooting, direct multiple shooting and two collocation methods the first based on Euler integration and the second on pseudospectral discretisation are applied to the Brachistochrone problem, with the results aiding in the selection of the best technique to integrate into the NMPC controller design. The results of this section showed the pseudospectral collocation method to be the best choice. Section 4 applies a pseudospectral based linear and nonlinear MPC controller to a 2D robot model in both open loop and closed loop scenarios. The results show that the nonlinear controller outperforms the linear controller particularly as perturbations increase and linear assumptions are violated as is the case when faults occur. Section
applies the final NMPC controller design to the longitudinal motion of a generic aircraft model assuming fault detection information is available. The results successfully demonstrate the capability of our controller as a fault tolerant flight control system. Many fault tolerant control system designs require two controllers one for the nominal case and one for the fault case. Our system requires the design of only one controller which can handle both nominal and fault cases. Finally Section 6 gives the analysis followed by the conclusion.
2 Nonlinear Model Predictive Control
Model predictive control (MPC), also referred to as receding horizon control, is an advanced control technology developed by practitioners within the industrial process industry that has had considerable impact on industrial process control. MPC, being capable of handling equipment and safety constraints [7], allows systems to operate at or near constraints, yielding a more efficient and profitable operation. Unlike many other control system designs, where the model of the plant is used only for design and analysis purposes, in MPC the model is an integral part of the control algorithm and is used to predict future behaviour of the plant in order to calculate the optimal control trajectory.
Linear MPC, where the internal model is linear, is a thoroughly researched area and is commonly used in practice. The internal prediction model predicts the behaviour of the plant over a future prediction Horizon, $H_p$. The idea is to select the best input that will produce the best predicted behaviour. A number of coincidence points are placed over the horizon with the aim of bringing the predicted output as close as possible to the reference trajectory. This is achieved by optimising a cost function, commonly a quadratic cost via quadratic programming in the case of linear MPC. Only the first input of the calculated trajectory is applied to the plant and the prediction window slides along by the sampling time. Once the window slides to the next time step and the calculated input is applied to the plant, the new plant states are fed back to the controller and the whole cycle begins again. The length of the prediction window remains fixed but slides forward by one sampling interval at each step; a process referred to as the receding horizon strategy. To reduce the computational burden a control horizon, $H_u$, can be defined which is smaller than $H_p$. The control inputs are calculated only along the control horizon, beyond which point the value of $u$ remains constant. Performance stability increases as the length of $H_u$ approaches the length of $H_p$.
Whilst NMPC uses the same structure as linear MPC the advantage of the former is the incorporation of a nonlinear process model for highly non-
linear systems. These nonlinear models are based on “first principles” and are obtained from an understanding of the physical nature of the system [7]. Increased ability to handle the computing demands of solving nonlinear optimization problems has lead to a rise in interest in NMPC within the control community.
The interest in NMPC, which began in the 90s, has been driven by the fact that today’s processes need to be operated under tighter performance specifications, with more environmental and safety constraints that can only be met when process nonlinearities and constraints are explicitly considered in the controller design [8]. The major limitation of linear MPC is that plant behaviour is described by a linear dynamic model, making it unsuitable for both moderately as well as highly nonlinear processes with large operating regimes [12]. NMPC is more frequently used in the process industry because the time scales encountered are in the order of minutes, making real-time requirements less severe than in, for example, aerospace applications.
While NMPC has the potential to improve process operation, it poses theoretical and practical problems that are more challenging than those associated with its linear counterpart, mainly due to the nonlinear program that must be solved online at each sampling period [9]. However, the inherent robustness of NMPC that allows it to deal with input model uncertainties without taking them directly into account, is a definite advantage, which is what makes it the focus of this paper. The next sub-section discusses optimal control techniques that can be used to solve NMPC problems.
2.1 Problem Formulation - Optimal Control Problem
For the NMPC methodology to be practically feasible the optimisation must be performed within the time constraints governed by the sampling period of the application [13]. Hence, when designing and implementing NMPC strategies, consideration must be given to computational efficiency [13]. Globally optimal NMPC methods can provide benefits over local techniques and can be successfully used for online control [14]. NMPC methods are generally based around tailoring nonlinear programming algorithms [13] to fit the structure of the online optimization or parametrising the predictions in terms of degrees of freedom. This directly affects the size of the online optimisation problem and in turn the computational burden of the NMPC strategy.
Methods used to solve optimal control problems commonly fall under two categories: direct and indirect [17]. Direct methods have better convergence properties than indirect methods and can be used quickly to solve a number of practical trajectory optimisation problems, hence only direct
methods are considered in this paper.
The path-constrained trajectory optimisation problem as detailed in [17], is an area that has been heavily researched and forms an integral part of the design of an NMPC controller. The main aim of optimal control is to determine the state and control pair that minimises a cost functional. That is, if a state and control pair is represented by \( \{x(t), u(t)\} \) then the aim is to minimise:
\[
J = M[x(t)] + \int_{t_0}^{t_f} [L(x(t), u(t), t)] \, dt,
\]
subject to the nonlinear state equations:
\[
\dot{x}(t) = f[x(t), u(t), t],
\]
the initial and terminal constraints
\[
\psi_0[x(t_0)] = 0,
\]
\[
\psi_0[x(t_f)] = 0,
\]
the mixed state-control path constraints
\[
g_L \leq g[x(t), u(t), t] \leq g_U,
\]
and the box constraints
\[
x_L \leq x(t) \leq x_U, \quad u_L \leq u(t) \leq u_U.
\]
Here: \( x \in \mathbb{R}^{n_x} \) are the state variables, \( u \in \mathbb{R}^{n_u} \) the control inputs, \( t \in \mathbb{R} \) the time, \( M : \mathbb{R}^{n_x} \times \mathbb{R} \rightarrow \mathbb{R} \) the terminal non-integral cost (also known as Mayer component), \( L : \mathbb{R}^{n_x} \times \mathbb{R}^{n_u} \times \mathbb{R} \rightarrow \mathbb{R} \) is integral cost (known as the Bolza component), \( \psi_0 \in \mathbb{R}^{n_x} \times \mathbb{R} \rightarrow \mathbb{R}^{n_0} \) represents the initial point conditions, \( \psi_f \in \mathbb{R}^{n_x} \times \mathbb{R} \rightarrow \mathbb{R}^{n_f} \) the final point conditions, \( g_L \in \mathbb{R}^{n_x} \times \mathbb{R}^{n_u} \times \mathbb{R} \rightarrow \mathbb{R}^{n_g} \) the lower bounds on the path constraints, and \( g_U \in \mathbb{R}^{n_x} \times \mathbb{R}^{n_u} \times \mathbb{R} \rightarrow \mathbb{R}^{n_g} \) the upper bounds on the path constraints.
Solving the problem defined by equations (1) to (6) is difficult. The direct methods detailed here solve this problem by applying a discretisation process and using standard algorithms to solve the resulting discrete optimisation problem.
A multitude of discretisation techniques exist for converting the continuous time problem to discrete time. In direct methods the mathematical programming problem, equations (1) to (6), is solved by considering either discretised inputs, or a combination of discretised inputs and states, as decision variables. The most common direct methods use the controls and
When deciding on a discretisation process many important factors need to be considered [17], such as the accuracy of the solution for a particular discretisation method given a number of optimisation variables, the computational expense of a particular discretisation method and the robustness of the discretisation method to the initial guess. The methods chosen for implementation and analysis are direct single shooting [16], direct multiple shooting [16], direct collocation using Euler integration [15] and direct collocation using pseudospectral discretisation [25].
2.1.1 Shooting methods
- are used to solve the given problem with initial and terminal conditions. They convert two point boundary value problems (BVPs) into initial value problems (IVPs) by guessing the value of the derivative at the initial boundary. Every time a guess is made a “shot” (using DE solvers to find a solution), is fired in an attempt to hit the end boundary. It is an iterative process where “shots” are made until the end boundary is reached within a desired tolerance.
The main difference between direct single and direct multiple shooting methods is that in the latter a BVP is converted into multiple IVPs. The interval of computation, \([t_0, t_f]\) is divided into \(M\) subintervals and an IVP is solved over each subinterval. All of the solutions over the subintervals are pieced together to form a continuous trajectory/solution, whenever the solutions to the IVPs match at the beginning and end of each subinterval, the matching conditions. These matching conditions introduce algebraic equations which must be satisfied along with the boundary conditions.
2.1.2 Direct Transcription
- involves fully discretising the problem (all controls and all states) and then solving the discrete problem numerically. The discretisation method, either integration or differentiation based, used to approximate the state equations must be combined with a method for approximating the integral in the generalised Bolza problem. Pseudospectral methods, for example, are differentiation based methods relying on differentiating Lagrange Polynomial expansions of the approximating polynomials for the states, while Hermite-Simpson based techniques are often thought of as integration methods. A comparison of various methods can be found in [17]. We investigated both an integration based method as well as a differentiation based method.
Derivative Based (or Pseudospectral) methods provide a better convergence rate known as spectral accuracy \([18], [19]\). The underlying idea is to represent the solution \(f\) via a truncated series expansion and to use analytic differentiation of the series to obtain spatial derivatives of \(f\). The spectral differentiation matrix, \(D_N\), is a linear mapping of a vector of \(N\) function values \(\{f(x_i)\}\) to a vector of \(N\) derivative values \(\{f'(x_i)\}\), and its calculation depends on the choice of the approximating series and the location of the points \(\{x_i\}\).
One advantage pseudospectral methods have over finite element or finite difference methods is that the underlying polynomial space is spanned by orthogonal polynomials that are infinitely differentiable global functions \([19], [20], [21]\). The choice of collocation points is crucial in pseudospectral methods \([22]\) and the Legendre-Gauss-Lobatto (LGL) points are implemented here because they provide maximum accuracy for quadrature approximations while at the same time avoiding the Runge phenomenon during interpolation \([23]\). In the case of LGL, the nodal points, which are the zeros of the derivatives of the Legendre polynomials, lie in the interval \([-1, 1]\) with the end points of the interval being included in the discretisation.
Finally an NLP solver is required as it forms an integral part of NMPC. Different discretisation methods are affected by the choice of NLP solvers in terms of the speed and robustness of the solution obtained \([17]\). For this work SNOPT \([24]\) is the solver of choice due to its popularity and because it is readily available. SNOPT solves the quadratic programming subproblem with a quasi-Newton approximation to the Hessian, via a large-scale sparse sequential quadratic programming algorithm.
### 3 Brachistochrone
The Brachistochrone problem is a nonlinear, nontrivial problem with an analytic solution very similar to the 2D robot problem to be addressed later. The analytical solution allows determination of the accuracy of the implementation of each method and provides a benchmark in choosing a numerical method to continue development of an NMPC controller. The Brachistochrone problem, simply stated, is to find the shape of a wire such that a bead sliding on the wire without friction, in uniform gravity, will reach a given horizontal displacement in minimum time.
The analytical solution is given by:
\[
\begin{align*}
x_b &= \frac{g}{\omega^2} \left( \omega t - \sin \omega t \right), \\
y_b &= \frac{g}{\omega^2} \left( 1 - \cos \omega t \right),
\end{align*}
\]
where \(\omega = \sqrt{\frac{2g}{x_f}}\), \(x_b\) and \(y_b\) are the horizontal and vertical displacements of...
the bead in the $xy$-plane, $g$ is the gravitational force and $x_f$ is the final x-displacement. The optimal control problem is to minimise the cost function:
$$
\min t_f,
$$
subject to the equations of motion of the bead:
$$
\dot{x}_b = V \sin \theta, \quad (9)
$$
$$
\dot{y}_b = V \cos \theta, \quad (10)
$$
$$
\dot{V} = g \cos \theta, \quad (11)
$$
and the initial and terminal constraints:
$$
x_b(0) = 0, \quad (12)
$$
$$
y_b(0) = 0, \quad (13)
$$
$$
V(0) = 0, \quad (14)
$$
$$
x_b(t_f) = x_f. \quad (15)
$$
Here $t_f$ is the time taken to reach $x_f$ and $V$ is the speed of the bead.
The number of discretisation points was varied for each method to investigate their effect and to determine the method most suitable for developing the fault tolerant controller. The value of $x_f$ is set to 0.5m and the value of $g$ for this work is 1m/s$^2$. For the direct single shooting method the control points, $N_u$, were varied from 5 to 500 and for each value of $N_u$ the state points were varied from 10 to 1000. Similarly the number of sections for the direct multiple shooting method was varied from $M = 2$ to $M = 30$ and the control points chosen for each section went from $N_u = 2$ to $N_u = 50$. The coincidence points for both the collocation methods (Euler integration and Pseudospectral) varied from $N = 5$ to $N = 800$.
The accuracy of the different numerical methods was assessed by comparing the solutions produced by them and the analytical solution given in [7]. The comparison was performed by producing plots of the magnitude of the errors in $x_b$ and $y_b$. The results showed that the Pseudospectral method produced the most accurate solution and increasing the number of coincidence points beyond $N = 50$ did not increase the accuracy of the solution. Hence for this reason the Pseudospectral method with $N = 50$ was used as the nominal solution in the analysis of the CPU time.
The time taken to reach $x_f = 0.5$ in the solution of the analytical problem given by equation (7) is $t_f = 1.2533$ secs. Plots of the CPU time taken to reach an optimal solution as a percentage of the nominal were produced.
The error plots show the percentage error between the optimal solution produced by each method and the nominal solution.
An example plot for the direct single shooting method is given in figure 1 for \( N_u = 5 \). The plots showed that for \( N_u = 5 \) the CPU time is less than the nominal for all \( N_x \) however the correct final time is unattainable. The CPU times for the pseudospectral method are given in figure 2. As expected the CPU time increases for increasing \( N \) and in general the CPU time taken by the pseudospectral method is higher when compared with the other methods.
Overall the results showed that the pseudospectral method can produce more accurate results with fewer discretisation points consequently requiring
less time. While, for large values of \( N \), the pseudospectral method results in a greater CPU time, larger values of \( N \) are deemed unnecessary to obtain a high level of accuracy. For this reason only the Pseudospectral method with \( N = 50 \) points is used in design of the NMPC controller.
4 Linear and Nonlinear MPC
In this part of the design phase of the NMPC controller the pseudospectral method is integrated into an MPC framework. An NMPC controller is designed, implemented and tested for a 2D robot model for trajectory following. Both the open and closed loop problems are addressed and comparisons are made to linear MPC. The next section details the 2D robot model.
4.1 Equations of motion
The 2D robot model given in figure 3 is used for both the linear and nonlinear implementations of MPC.
\[\begin{align*}
\dot{x} &= V \cos \psi, \\
\dot{y} &= V \sin \psi, \\
\dot{\psi} &= \frac{R(\omega_R - \omega_L)}{2b},
\end{align*}\]
where \( x \) is the \( x \)-coordinate of the point \( C \), \( y \) the \( y \)-coordinate of the point \( C \), \( \psi \) the heading angle, \( \omega_R \) the right wheel angular velocity, \( \omega_L \) the left
wheel angular velocity and $V$ the speed given by $V = \frac{R(\omega_R + \omega_L)}{2}$.
The next few sub-sections detail the development of the linear and non-linear MPC controllers. For a fair comparison the pseudospectral method with 50 collocation points is chosen as the method for discretisation for both controllers.
### 4.2 The Open Loop Problem
In MPC an open loop problem is solved at each time step. Hence it was important to implement and test the controller on the open loop problem.
Many tuning parameters can be used to determine the performance of the controller; the weighting factors on the cost function, the design of the cost function, the length of the prediction horizon, the initial condition, the integration time step, and the number of discretisation points required for an acceptable solution. From the previous analysis the numerical technique chosen for the application of NMPC is the Pseudospectral method with 50 discretisation/coincidence points. As a part of this work the effect of the choice of the cost function, the prediction window length, the integration time step and the effect of the initial condition on the solution were all considered.
The 2D robot is required to follow the path:
\[
\forall x \geq 0 : y = 5,
\]
(19)
travelling with a velocity of $1 \text{m/s}$ and constraints of $\pm 1000 \text{deg/sec}$ on the wheel speeds $\omega_R$ and $\omega_L$. The objective is to drive the robot back to the reference path, from $y = 6$ to $y = 5$.
### 4.2.1 Effect of Different Cost Functions
Five different cost functions were developed:
**Cost Type 1:** Errors between the reference/nominal path and the robot path are minimised:
\[
J_{N1} = \frac{(t_f - t_0)}{2} \sum_{j=0}^{N} \left( \| x - x_{\text{ref}} \|_{Q_x}^2 \right) w_j.
\]
(20)
**Cost Type 2:** Errors between the robot path and the nominal path, plus the error between the actual wheel speeds, $\omega_R$ and $\omega_L$ and the nominal wheel speeds are minimised:
\[
J_{N2} = \frac{(t_f - t_0)}{2} \sum_{j=0}^{N} \left( \| x - x_{\text{ref}} \|_{Q_x}^2 + \| u - u_{\text{ref}} \|_{Q_u}^2 \right) w_j.
\]
(21)
Cost Type 3: Errors between the robot path and the nominal path, plus the difference between the wheel speeds are minimised:
\[ J_{N3} = \frac{(t_f - t_0)}{2} \sum_{j=0}^{N} \left( \| \mathbf{x} - \mathbf{x}_{ref} \|_{Q_x}^2 + \| \omega_R - \omega_L \|_{Q_\omega}^2 \right) w_j. \] (22)
Cost Type 4: Errors between the nominal speed and robot speed as well as the errors between the nominal angular acceleration and the robot’s angular acceleration are minimised:
\[ J_{N4} = \frac{(t_f - t_0)}{2} \sum_{j=0}^{N} \left( \| V - V_{ref} \|_{Q_V}^2 + \| \dot{\psi} - \dot{\psi}_{ref} \|_{Q_\psi}^2 \right) w_j. \] (23)
Cost Type 5: Errors between the nominal speed and robot speed as well as the errors between the nominal angular acceleration and the robot’s angular acceleration along with the errors between the nominal path and the robot path are minimised:
\[ J_{N5} = \frac{(t_f - t_0)}{2} \sum_{j=0}^{N} \left( \| \mathbf{x} - \mathbf{x}_{ref} \|_{Q_x}^2 + \| V - V_{ref} \|_{Q_V}^2 + \| \dot{\psi} - \dot{\psi}_{ref} \|_{Q_\psi}^2 \right) w_j. \] (24)
Each cost type was tested using both linear and nonlinear MPC. The robot initial \( x, y \) and \( \psi \) was set to \( x_0 = [0 \ 6 \ 0]^T \) for both controllers. The prediction window length was varied between \( H_p = 1 \) sec, \( H_p = 5 \) secs and \( H_p = 10 \) secs. Through trial and error the weights were set to:
\[ Q_x = 10, \ Q_u = 1, \ Q_\omega = 1, \ Q_V = 1, \ Q_\psi = 1 \]
The optimal trajectories produced by all the different cost functions for the varying window lengths were plotted for both the linear and nonlinear MPC cases. Results showed that for a window length of 1 sec cost types 1, 2, 3 and 5 were able to drive the robot back onto the desired path by the end of the window for both the linear and nonlinear cases, however cost type 4 was unsuccessful in doing so.
The error plots for all the prediction window lengths were generated showing the magnitude of the error in the y-direction between the nominal path \( (y = 5) \) and the actual robot path. Overall the results showed that for a path following scenario, it is best to not only minimise the path errors, but to also follow a velocity profile to obtain a smoother non oscillating solution. For this reason cost type 5 (equation (24)) was chosen for the final controller design and is used in the remainder of this analysis. In addition, the error plots show that the difference in errors produced by the linear and nonlinear controllers are the least for this cost function, making it the best...
candidate for comparison purposes.
The effect of the integration time step was also investigated. The analysis given above considers only the optimal solution produced by the controller, however in an MPC framework only the first output is applied to the plant, with the plant then providing sensor information to the navigation subsystem, for example, (on an aircraft) to calculate location and orientation information. Hence it is important to understand the effect of the integration time step in conjunction with the optimal control input. The integration time step was varied as follows; $dt = [0.1, 0.01, 0.001]$ for varying $H_p$ lengths, namely 1 sec, 5 secs and 10 secs, and the optimal trajectories were plotted. All results showed that the integration time step has very little effect on the results. Upon further investigation the results showed that the smallest integration time step of 0.001secs was able to give a solution closest to the optimal. The length of the prediction horizon was seen to have the greatest effect on the integrated solution with the integrated solution getting closer to the optimal solution as the look ahead increased. Another point to note is that the integrated output produced with the nonlinear controller more closely matched the optimal solution compared to the linear integrated output.
Varying the length of the prediction window showed that it is always best to have a longer window as this produced the lowest errors, particularly in the linear controller case. In addition, the longer window allowed the robot to reach the nominal path more quickly. The accuracy of the integrated solution increased as the window length increased, but, unfortunately a longer window resulted in an increase in computation time.
From the results obtained a window length of 5 secs was chosen for the final NMPC controller design as it was a good compromise between efficiency and accuracy. A window length of 1 second proved to be too short to produce an accurate solution particularly in the case of the integrated solution. While a window length of 10 secs produced an integrated solution closely matching the optimal solution in the nonlinear controller case, the solution produced by a window length of 5 secs, while not as precise, still managed to develop a solution closely resembling the optimal solution. Hence for the rest of this research a window length of 5 seconds is used along with cost type 5 and an integration time step of 0.01secs.
The initial condition is another factor that must be considered in the design and selection of the controller. The sensitivity of the starting point on the overall solution is critical particularly in the case of linear techniques. In this analysis the robot is required to follow the same path as above, $y = 5$, and the initial $y$ is varied from 0m to 10m in steps of 0.1m. The errors
between the nominal path and the actual robot position are calculated at various points along the prediction horizon namely at 1 sec, 2 secs, 3 secs, 4 secs and 5 secs for all initial \(y\) values. Plots of errors versus initial \(y\) for the different times were obtained and figures 4 and 5 show the results at \(t = 1\) sec and \(t = 5\) secs. The results show the errors between the optimal solution and the nominal path as well as the errors between the integrated output and the nominal path for both the linear and nonlinear controllers. The errors arising from the integrated output of the linear controller are shown on a separate plot underneath the main plots as these errors were much higher compared to the others and by plotting all errors on the one graph the errors produced by the other solutions were not as clearly visible. The results show that as the time increases from 1 second to 5 secs the errors decrease as the robot approaches the nominal path. The results clearly show that the further away the robot is from the nominal path (i.e. the greater the perturbation) the higher the error in the case of the linear controller.
At the 1 second mark along the prediction window (figure 4) the errors between the solution produced by the nonlinear controller and the nominal path \((y = 5)\) were seen to be linear as a function of initial \(y\). Moving further along, the prediction window shows that these errors decrease and are very close to zero for any \(y_0\). There is only a small region around the nominal path, \(y = 5\), during which the errors produced by the linear controller are zero and match those produced by the nonlinear solution at any time along the prediction window.
Figure 4: Open Loop: Initial Conditions vs y-Displacement Error, time = 1sec
The next subsection investigates the closed loop problem and compares the output produced by both linear and nonlinear MPC.
4.3 The Closed Loop Problem
The aim of these simulations is to implement and investigate the behaviour of both linear and nonlinear MPC regarding the closed loop problem. The fault tolerant problem is essentially closed loop, hence to apply NMPC to fault tolerant control the pseudosepectral NMPC controller design is tested on the 2D robot model where the robot is required to travel on a circular trajectory.
Based on the analysis from the previous subsections a prediction window length of 5 seconds is used with 50 collocation points, cost type 5 and an integration time step of 0.01secs. Constraints of $\pm 1000\text{deg/sec}$ are placed on the control inputs which are the angular velocities produced by the right and left wheels. Three different scenarios were set up: Scenario 1 the robot begins on the path with initial conditions $y_0 = [5, 0, 0]^\top$, Scenario 2 the robot begins slightly off the path with initial conditions $y_0 = [-2, 4, 0]^\top$ and Scenario 3 the robot begins well off the path with initial conditions $y_0 = [0, 20, 0]^\top$.
For stability $H_u = H_p$ [8]. In many MPC\NMPC formulations $H_u$ is less than $H_p$. While this greatly reduces the computational expense it does however produce a suboptimal solution [9], hence for the purposes of this research the control horizon is equal to the prediction horizon.
For scenario 1 where the robot begins on the path the trajectories produced by both linear and nonlinear controllers were the same. The calculated
optimal inputs produced by both controllers were also identical. In the case of scenario 2 where the robot begins slightly off the path, both the linear and nonlinear controllers managed to bring the robot back onto the path. The plots of the optimal inputs showed that initially both controllers work at the maximum constraint to drive the robot back onto the path. Once the path is reached (i.e. perturbations are small) both controllers exhibit the same performance.
Figure 6: Closed Loop: Scenario 3 - Trajectory
Figure 7: Closed Loop: Scenario 3 - Angular Rates
The trajectory plots for scenario 3 show that only the nonlinear controller is able to bring the robot back onto the path with the linear controller unable to drive the robot back to the path. The control inputs produced by
both controllers clearly showed that the linear controller worked very hard to take the robot back onto the path by consistently working at the constraint limits however it was still unable to return the robot back to the path.
The results clearly showed that the pseudospectral NMPC solution to the nonlinear model predictive controller is a viable choice outperforming its linear counterpart when the perturbations are large. In the next section we apply the solution developed thus far to a generic aircraft model. Note that FDI is assumed for the simulation exercise carried out in section 5.
5 Application to Flight Control
The NMPC controller developed in the previous sections was applied to the longitudinal motion of an aircraft to demonstrate fault tolerant control. The generic aircraft model developed here for control law design and validation is based on the McDonnell Douglas F-4 aircraft [26]. It is a fixed wing aircraft equipped with throttle, elevators, ailerons and a rudder for control. Longitudinal motion is predominantly controlled via the throttle and elevators which is used to pitch the aircraft nose up and down and hence the remaining controls will not be considered here. Figure 8 presents a sketch of a generic aircraft which identifies the location of the elevators and defines the coordinate frames in which the equations of motion are defined. These equations are all carried out in the body axis which has its origin at the centre of gravity (c.g.) on the body of the aircraft however position and velocity information are commonly presented in the NED frame which is an earth fixed coordinate system.
Figure 8: Aircraft controls and co-ordinate systems
The generic aircraft model has the aerodynamic characteristics of the McDonnell Douglas F-4 aircraft however the dimensional and mass properties are those given in tables 1 and 2 respectively.
**Table 1: Aircraft Dimensional Properties**
| Parameter | Value |
|---------------------------|-------|
| Wing Area $S$ | 20m |
| Mean Aerodynamic Chord $\bar{c}$ | 3m |
| C.G location $x_{c.g}$ | 0 |
| C.G reference location $x_{c.g.ref}$ | 0 |
**Table 2: Mass Properties of model used for simulation.**
| Parameter | Weight (kg) | $I_X$ (kg.m²) | $I_Y$ (kg.m²) | $I_Z$ (kg.m²) | $I_{XZ}$ (kg.m²) |
|-----------|-------------|---------------|---------------|---------------|------------------|
| Value | 1,177 | 2,257 | 11,044 | 12,636 | 106 |
The process model used by the NMPC controllers and the plant model is given by following equations of motion:
\[
V_t = \sqrt{V_N^2 + V_D^2} \tag{25}
\]
\[
u = V_N \cos(\theta) - V_D \sin(\theta) \tag{26}\]
\[
w = V_N \sin(\theta) + V_D \cos(\theta) \tag{27}\]
\[
\alpha = \arctan\left(\frac{w}{u}\right) \tag{28}\]
\[
\bar{q} = \frac{1}{2} \rho V_t^2 \tag{29}\]
\[
a_x = \frac{\bar{q} S \bar{c} C_X + T}{m} \tag{30}\]
\[
a_z = \bar{q} S \bar{c} C_Z m \tag{31}\]
\[
a_N = a_x \cos(\theta) + a_z \sin(\theta) \tag{32}\]
\[
a_D = g - a_x \sin(\theta) + a_z \cos(\theta) \tag{33}\]
\[
\dot{q} = \bar{q} S \bar{c} C_m \left(\frac{1}{I_Y}\right) \tag{34}\]
\[ CX = -0.0434 + 2.93 \times 10^{-3} \alpha + 2.53 \times 10^{-5} \beta^2 - 1.07 \times 10^{-6} \alpha \beta^2 + 9.5 \times 10^{-4} \delta_e
- 8.5 \times 10^{-7} \delta_e \beta^2 + \left( \frac{180 q \bar{e}}{\pi 2V_t} \right) \left( 8.73 \times 10^{-3} + 0.001 \alpha - 1.75 \times 10^{-4} \alpha^2 \right), \]
\[ C_m = -6.61 \times 10^{-3} - 2.67 \times 10^{-3} \alpha - 6.48 \times 10^{-5} \beta^2
- 2.65 \times 10^{-6} \alpha \beta^2 - 6.54 \times 10^{-3} \delta_e - 8.49 \times 10^{-5} \delta_e \alpha
+ 3.74 \times 10^{-6} \delta_e \beta^2 - 3.5 \times 10^{-5} \delta_e^2 \]
+ \left( \frac{180 q \bar{e}}{\pi 2V_t} \right) (-0.0473 - 1.57 \times 10^{-3} \alpha) + (x_{c,g,ref} - x_{c,g}) C_Z. \]
Where \( V_T \) is the true airspeed, \( V_N, V_D \) are the velocities in the north and down directions in the NED frame and \( u \) and \( w \) are the velocities in the \( x \) and \( z \) directions in the body axis frame. The accelerations \( a_N \) and \( a_D \) are given in the NED frame in the North and Down directions respectively and \( a_x \) and \( a_z \) are accelerations in the body axis. \( \dot{q} \) is the pitch rate derivative, \( \bar{q} \) is known as dynamic pressure and \( \alpha \) is an aerodynamic angle called the angle of attack. \( CX \) is a non-dimensional force coefficient in the body \( X \)-direction and the \( C_M \) is the non-dimensional pitching moment coefficient. The force and moment coefficients used for this model are valid for angle of attack \( \alpha \leq 15 \text{ deg} \). The thrust force, \( T \) is modelled by \[ T_{max} = ((30.21 - 0.668 h_T - 6.877 h_T^2 + 1.951 h_T^3 - 0.1512 h_T^4)
+ \left( \frac{V_t}{v_s} \right) (-33.8 + 3.347 h_T + 18.13 h_T^2 - 5.865 h_T^3 + 0.4757 h_T^4)
+ \left( \frac{V_t}{v_s} \right)^2 (100.8 - 77.56 h_T + 5.441 h_T^2 + 2.864 h_T^3 - 0.3355 h_T^4)
+ \left( \frac{V_t}{v_s} \right)^3 (-78.99 + 101.4 h_T - 30.28 h_T^2 + 3.236 h_T^3 - 0.1089 h_T^4)
+ \left( \frac{V_t}{v_s} \right)^4 (18.74 - 31.6 h_T
+ 12.04 h_T^2 - 1.785 h_T^3 + 0.09417 h_T^4) \right) \frac{4448.22}{20}, \]
\[ T = T_{max} \delta_{th}, \]
The equations of motion are integrated forward in the plant model using a Runge-Kutta integration method with the Matlab subroutine ode45. The controller runs at 10Hz and the equations of motion are used as constraints to the optimal control problem. As developed in the previous sections, a pseudospectral discretisation method is used with 50 collocation points and a prediction window $H_p$ of 5 secs. The optimal control inputs are calculated via SNOPT. The aircraft is required to follow the trajectory given in figure 9.
Adopting the pseudospectral discretisation method where both the states and controls are discretised, the NMPC optimisation vector is:
$$x_{\text{nmcp}} = [x_D, V_N, V_D, \theta, q, \delta_e, \delta_{th}, \Delta\delta_e]^T,$$
where $\Delta\delta_e$ is the rate of change of the elevator deflection $\delta_e$.
The following optimal control problem is then solved:
$$\min_{x, u} \frac{H_p}{2} \sum_{j=1}^{N+1} \left( \|x_D(j) - x_{D\text{vel}}(j)\|^2_{Q_x} + \|V_t(j) - V_{t\text{vel}}(j)\|^2_{Q_v} + \|\Delta\delta_e\|^2_{Q_u} \right) w(j),$$
subject to
$$\left(\frac{t_f - t_0}{2}\right) D_{j,k} x_j - \dot{x}_j = 0, \quad (42)$$
$$x(j_0) - x_{\text{dem}}(j_0) = 0, \quad (43)$$
$$x_{lb} \leq x \leq x_{ub}, \quad (44)$$
$$u_{lb} \leq u \leq u_{ub}, \quad (45)$$
$$\Delta\delta_{e\text{lb}} \leq \Delta\delta_e \leq \Delta\delta_{e\text{ub}}, \quad (46)$$
20
where $x_D$ and $x_{D_{ref}}$ are the actual and reference heights respectively, and $V_t$ and $V_{t_{ref}}$ are the actual and reference true airspeeds respectively. The constraints applied are given in table 3.
| Variable | Upper Constraint | Lower Constraint |
|----------|------------------|------------------|
| $x_D$ | 300 m | 1 m |
| $V_N$ | 100 m/s | 30 m/s |
| $V_D$ | 3 m/s | −3 m/s |
| $\theta$ | None | None |
| $q$ | None | None |
| $\delta_e$ | 20 deg | −20 deg |
| $\delta_{th}$ | 100% | 0% |
| $\Delta \delta_e$ | 200 deg/s | −200 deg/s |
The weighting matrices are diagonal matrices with the following values along the diagonal for each state which were set through trial and error: $Q_x = 10$, $Q_V = 20$, $Q_u = 1$. Note: the Control surface rates given in table 3 are realistic for a high performance unstable airframe or for a lower weight aircraft with a stable airframe; in either case a feasible fictional aircraft model has been produced for simulation purposes to demonstrate proof of concept.
5.1 Numerical Results
To illustrate the concept of FTC, seven different scenarios were set up, with the first one being the no fault case. The next five scenarios had the throttle stuck at 70, 50, 35, 30 and 20% for the entire duration of the flight. Scenario seven simulated the throttle getting stuck at 20% 80 secs into the flight. It is also assumed that fault detection information is available. This includes the time at which the throttle becomes stuck and the position at which it is stuck. This information is used to update the constraint values of the NMPC controller. Providing the controller with the most accurate and up to date information enables it to make better use of the healthy actuators. As previously mentioned the force and moment coefficients are valid for $\alpha \leq 15$ deg. For all scenarios $\alpha$ was checked to ensure that 15 deg was never exceeded. The plots of $\alpha$ vs time (they have not been provided here due to space constraints) showed $\alpha$ to remain below 15 deg hence the equations of motion were never violated. Another means of avoiding this would be to place a constraint on $\alpha$ in the NMPC controller.
5.1.1 True Airspeed
- The demanded airspeed was 50 m/s true airspeed. The plot given in figure 10 shows the aircraft true airspeed for each of the scenarios. The results show that in a fault free case the aircraft is able to fly at the demanded true airspeed. However, when the throttle is stuck at 70% or even 50% there is too much power continually being provided to the aircraft resulting in a large airspeed response. When the throttle is stuck at 35% the aircraft is able to maintain the demanded $V_t$ for only a short period of time, at the beginning of the flight mission. However at 30% throttle the maximum deviation from the demanded airspeed is approximately 5 m/s at any given time. When the throttle drops below 30% the aircraft is unable to maintain the true airspeed which drops to approximately between 35 m/s and 30 m/s. The results for scenario 7 show that once the fault occurs at 80 secs the true airspeed immediately begins to drop, as expected. One of the main points to note is that the stall speed was never reached; the controller was able to avoid the aircraft stalling regardless of the severity of the fault.
5.1.2 Vertical Speed
- The vertical speed response of the aircraft was also analysed and plots for scenarios 1-4 are given in figure 11. The plots show the aircraft response along with the constraints (in red) placed on the vertical speed. For high values of throttle (70% and 50%) the vertical speed is continuously bouncing between the constraints in an attempt to maintain the true airspeed demand. For the case when the throttle is stuck at 35% the vertical speed profile is seen to be similar to the no fault case, except in the descent phase. During this phase, when the aircraft is descending and gaining speed, the vertical speed response can be seen to continuously move between the constraints to regulate the speed. Results showed that for throttle values less than 30% there is insufficient power to maintain a climb hence the vertical speed is
seen to operate at the lower constraint or at zero. In the case of scenario 7 it was found that once the fault occurred at 80 secs the vertical speed moved between the constraints, working hard to maintain the true airspeed.
Figure 11: Stuck Throttle - Vertical Speed Response, constraints (red lines), aircraft response (blue)
5.1.3 Elevator Activity
- In regards to fault tolerance, the elevator activity is of the most interest. If the throttle is stuck the elevator provides a level of redundancy to maintain the aircraft speed. Figure 12 shows plots of elevator activity for the different scenarios. The plots clearly show that any change in throttle increases the elevator activity when compared to the no fault case. The elevator activity increases in an attempt to regulate the airspeed of the aircraft. In the case of the high throttle values (70% and 50%) the elevator activity is the highest because a higher level of power is continually being provided to the aircraft exceeding the amount required to fly at the demanded speed. Hence the elevator constantly jumps between the constraints in an attempt to compensate for the excess power. For the 30% stuck throttle case the elevator activity does increase compared to the no fault case; however 35% throttle was found to be closer to the amount required to maintain the given height profile, hence the elevator does not need to work as hard compared to the 70% and 50% cases. For the lower throttle values activity increases during the climb and descent phases. In the climb phase of the mission there is not enough power available to the aircraft, so it compensates by erratically deflecting the elevator. During the descent phase however there is too much power; to regulate this and to stay within the velocity constraints the activity increases. The last scenario shows that at the fault occurrence
time of 80 secs the elevator increases activity to compensate for the faulty throttle.
Figure 12: Stuck Throttle - Elevator Activity, constraints (red lines), aircraft response (blue)
The plots indicate large oscillations in the elevator activity however the data presented shows 200 seconds of flight. Actuator dynamics have been modelled in both the controller and the plant model. Zooming in on the 70% stuck throttle case (figure 13) it can be seen that the rate dynamics are respected.
5.1.4 Height Profile
- The trajectory flown by the aircraft during the different scenarios is given in figure 14. The no fault case, as expected, follows the reference height profile perfectly. The 35% case is also able to closely maintain the profile. In the 70% and 50% cases the aircraft continually tries to regulate the airspeed to compensate for the excess power. The solutions produced by both scenarios show the aircraft overshooting followed by an undershoot, so the solution oscillates around the reference. When the throttle becomes stuck at 30% the aircraft begins the climb phase of the mission but is only able to continue climbing for 20 secs before it begins gliding towards the ground. In the 20% stuck throttle case the aircraft completes the straight and level phase of the mission but does not have enough power to begin climbing, and descends to the ground. The final scenario shows that the elevator is able to compensate for the stuck throttle in mid-flight and successfully finish the mission.
6 Conclusion
This paper was dedicated to the theoretical and practical implementation aspects of NMPC. A viable design for a nonlinear MPC controller applicable to FTC was sought. A number of discretisation methods were implemented using the well known Brachistochrone problem and the Pseudospectral numerical method was found to have the best performance. This method was applied in an NMPC framework to a 2D robot model in both open and closed loop settings. Comparisons were made between linear MPC and the nonlinear MPC solutions. For small perturbations the two controllers produced the same results. However, for large perturbations where nonlinearity effects are more significant, the pseudospectral nonlinear MPC controller was found to produce more accurate results than the linear MPC controller. The final design was also applied to a generic aircraft model where it was assumed that fault detection information was provided. The motivation behind this was to demonstrate the concept of fault tolerant flight control using NMPC. It was found that the nonlinear NMPC controller designed in this work is a viable controller for fault tolerant flight control. This was evident from the results which show that in the event of a stuck throttle the controller is able to manipulate the movement of the elevator to compensate for the fault.
References
[1] Pachter M, Huang Y. Fault tolerant flight control. Journal of Guidance, Control, and Dynamics 2003; 26(1):151-160.
[2] Kale MM, Chipperfield AJ. Robust and stabilized MPC formulations for fault tolerant and reconfigurable flight control. Proceedings of the 2004 IEEE International Symposium on Intelligent Control 2004; 222-227.
[3] Boskovic JD, Sai-Ming L, Mehra RK. On-line failure detection and identification (FDI) and adaptive reconfigurable control (ARC) in aerospace applications. Proceedings of the 2001 American Control Conference 2001; 4:2625-2626. DOI:10.1109/ACC.2001.946269
[4] Gopinathan M, Boskovic JD, Mehra RK, Rago C. A multiple model predictive scheme for fault-tolerant flight control design. Proceedings of the 37th IEEE Conference on Decision and Control 1998; 2:1376-1381. DOI:10.1109/CDC.1998.758477.
[5] Maciejowski JM. Fault-tolerant aspects of MPC. IEE Two-Day Workshop on Model Predictive Control: Techniques and Applications-Day 2 (Ref. No. 1999/096) 1999; 1/1–1/4.
[6] Maciejowski JM. Modelling and predictive control: Enabling technologies for reconfiguration. Annual Reviews in Control 1999; 23:13-23.
[7] Maciejowski JM. Predictive Control: With Constraints. Pearson Education 2002. Prentice Hall.
[8] Allgöwer F, Zheng A. Nonlinear Model Predictive Control. Progress in Systems and Control Theory 2000; Volume 26. Birkhäuser Basel.
[9] Rau M, Schroder D. Model predictive control with nonlinear state space models. 7th International Workshop on Advanced Motion Control 2002: 136–141. IEEE.
[10] Zhang Y, Jiang J. Bibliographical review on reconfigurable fault-tolerant control systems. Annual Reviews in Control, Elsevier 2008; 32(2):229–252
[11] Maciejowski JM. The implicit daisy-chaining property of constrained predictive control. Applied Mathematics and Computer Science 1998; 8:695–712.
[12] Henson MA. Nonlinear model predictive control: current status and future directions. Computers and Chemical Engineering 1998; 23(2):187–202
[13] Cannon M. Efficient nonlinear model predictive control algorithms. Annual Reviews in Control 2004; 28(2):229 - 237. DOI:http://dx.doi.org/10.1016/j.arcontrol.2004.05.001
[14] Long C.E, Polisetty PK, Gatzke EP. *Nonlinear model predictive control using deterministic global optimization*. Journal of Process Control 2006; 16(6):635 - 643. DOI: [http://dx.doi.org/10.1016/j.jprocont.2005.11.001](http://dx.doi.org/10.1016/j.jprocont.2005.11.001)
[15] Diehl M, Bock HG, Diedam H, Wieber PB. *Fast Direct Multiple Shooting Algorithms for Optimal Robot Control*. Fast Motions in Biomechanics and Robotics - Lecture Notes in Control and Information Sciences 2006; 340:65–93. DOI: [10.1007/978-3-540-36119-0-4](10.1007/978-3-540-36119-0-4).
[16] Diehl M, Bock HG, Schilder JP, Findeisen R, Nagy Z, Allgwer F. *Real-time optimization and nonlinear model predictive control of processes governed by differential-algebraic equations*. Journal of Process Control 2002; 12(4):577–585. DOI: [http://dx.doi.org/10.1016/S0959-1524(01)00023-3](http://dx.doi.org/10.1016/S0959-1524(01)00023-3).
[17] Williams P. *A Comparison of Differentiation and Integration Based Direct Transcription Methods (AAS 05-128)*. Advances in the Astronautical Sciences American Astronautical Society by Univelt 2005; 120(1):389–409.
[18] Greengard L. *Spectral integration and two-point boundary value problems*. SIAM Journal on Numerical Analysis 1991; 28(4):1071–1080.
[19] Ross IM, Fahroo F. *Legendre pseudospectral approximations of optimal control problems*. New Trends in Nonlinear Dynamics and Control and their Applications, Springer 2003; 327–342.
[20] Elnagar G, Kazemi MA, R M. *The pseudospectral Legendre method for discretizing optimal control problems*. IEEE Transactions on Automatic Control 1995; 40(10):1793–1796.
[21] Fahroo F, Ross MI. *Direct trajectory optimization by a Chebyshev pseudospectral method*. Journal of Guidance, Control, and Dynamics AIAA 2002; 25(1):160–166.
[22] Ross MI, Karpenko M. *A review of pseudospectral optimal control: From theory to flight*. Annual Reviews in Control, Elsevier 2012; 36(2):182–197.
[23] Williams P. *Hermite-Legendre-Gauss-Lobatto direct transcription in trajectory optimization*. Journal of Guidance, Control, and Dynamics, AIAA 2009; 32(4):1392–1395.
[24] Gill PE, Murray W, Saunders MA. *Users guide for SNOPT version 7: Software for large-scale nonlinear programming*. Citeseer 2006; [http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.217.5929](http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.217.5929) [Last Accessed: 20-01-2016].
[25] Williams Paul, *Quadrature discretization method in tethered satellite control*. Applied Mathematics and Computation, Elsevier 2011; 217(21):8223–8235.
[26] Garza F, Morelli EA. *A collection of nonlinear aircraft simulations in Matlab*. NASA Langley Research Center, Technical Report NASA/TM 2003; 212145.
[27] Bryson AE. *Dynamic Optimization*. Addison Wesley Longman, Menlo Park, CA 1999. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
1433,
1
],
[
1433,
4560,
2
],
[
4560,
7371,
3
],
[
7371,
10099,
4
],
[
10099,
12487,
5
],
[
12487,
14915,
6
],
[
14915,
17665,
7
],
[
17665,
19809,
8
],
[
19809,
20745,
9
],
[
20745,
21919,
10
],
[
21919,
24053,
11
],
[
24053,
26610,
12
],
[
26610,
29490,
13
],
[
29490,
31369,
14
],
[
31369,
32994,
15
],
[
32994,
33788,
16
],
[
33788,
35481,
17
],
[
35481,
36975,
18
],
[
36975,
39088,
19
],
[
39088,
40554,
20
],
[
40554,
42921,
21
],
[
42921,
44981,
22
],
[
44981,
46850,
23
],
[
46850,
47389,
24
],
[
47389,
48408,
25
],
[
48408,
49887,
26
],
[
49887,
51930,
27
],
[
51930,
54351,
28
],
[
54351,
54749,
29
]
]
} |
a69ca62f551d4d4d76232de8f14bd58fc4a7e9ba | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 12,
"total-fallback-pages": 0,
"total-input-tokens": 39771,
"total-output-tokens": 15262,
"fieldofstudy": [
"Medicine"
]
} | Evaluating the impact of a new educational tool on understanding of polygenic risk scores for alcohol use disorder
Morgan N. Driver†, Sally I-Chun Kuo2, Lia Petronio3, Deanna Brockman4, Jacqueline S. Dron5, Jehannine Austin6,7 and Danielle M. Dick2,8
1Department of Human and Molecular Genetics, Virginia Commonwealth University, Richmond, VA, United States, 2Department of Psychiatry, Robert Wood Johnson Medical School, Rutgers University, Piscataway, NJ, United States, 3Broad Institute of MIT and Harvard, Cambridge, MA, United States, 4Color Health Inc., Burlingame, CA, United States, 5Department of Medicine, Center for Genomic Medicine, Massachusetts General Hospital, Boston, MA, United States, 6Department of Psychiatry, The University of British Columbia, Vancouver, BC, Canada, 7Department of Medical Genetics, The University of British Columbia, Vancouver, BC, Canada, 8Rutgers Addiction Research Center, Brain Health Institute, Rutgers University, Piscataway, NJ, United States
Introduction: As gene identification efforts have advanced in psychiatry, so have aspirations to use genome-wide polygenic information for prevention and intervention. Although polygenic risk scores (PRS) for substance use and psychiatric outcomes are not yet available in clinical settings, individuals can access their PRS through online direct-to-consumer resources. One of these widely used websites reports that alcohol use disorder is the third most requested PRS out of >1,000 conditions. However, data indicate that there are misunderstandings about complex genetic concepts, with a lower understanding of PRS being associated with a more negative impact of receiving polygenic risk information. There is a need to develop and evaluate educational tools to increase understanding of PRS.
Methods: We conducted a randomized controlled trial to evaluate the impact of web-based educational information on understanding of PRS for alcohol use disorder. A total of 325 college students (70.4% female; 43.6% White; mean age = 18.9 years) from an urban, diverse university completed the study.
Results: Overall, participants were highly satisfied with the educational information. Results from a one-way ANOVA indicated that there was a significant increase in overall understanding of PRS for alcohol use disorder (p-value < 0.001), among individuals who received educational information about PRS and alcohol use disorder, as compared to receiving no accompanying information (adj. p-value < 0.001), or educational
information about alcohol use disorder only (adj. $p$-value < 0.001).
**Discussion:** These findings suggest that the web-based educational tool could be provided alongside polygenic risk information in order to enhance understanding and interpretation of the information.
**Clinical trial registration:** [ClinicalTrials.gov], identifier [NCT05143073].
**KEYWORDS**
polygenic risk scores, alcohol use disorder, personalized medicine, genetic risk, prevention
### Introduction
The basis of precision medicine is to use an individual's personal genetic information, along with lifestyle information and personal medical history, to make more effective clinical decisions regarding health outcomes (1). For many common complex health outcomes, including psychiatric conditions, an individual's genetic liability is calculated using estimates from genome-wide association studies (GWAS), with risk estimates provided in the form of polygenic risk scores (PRS) which sum information about risk-enhancing variants detected across the genome (2).
Providing genetic risk information in the form of a PRS is quite different from genetic feedback that has typically been provided in medicine. Historically, genetic testing has focused on single gene disorders, with results indicating either the presence or absence of a disease-causing variant. These genetic testing results have traditionally been presented in clinical settings by health care professionals such as genetic counselors, who are trained to educate people about the inheritance of genetic conditions and communicate genetic test results (3). However, there are limited numbers of genetic counselors (4) and PRS are most commonly accessed through online direct-to-consumer (DTC) websites, not in a clinical setting (5, 6).
Although PRS are not yet commonly used in clinical settings, they are already available through free, online resources. There has been an exponential increase in the provision of DTC genetic information, with more than 26 million individuals having participated in DTC genetic testing by 2019 (7). Public websites allow individuals to upload raw genetic data obtained from DTC genetic testing companies to compute PRS for a variety of health conditions, including cancer, coronary artery disease, and psychiatric conditions (6). User data from one of these websites illustrates a parallel exponential increase in individuals accessing PRS over the last several years, with the third most frequently accessed PRS being for alcohol dependence (6). Additionally, a recent study found that 80–90% of young adults were interested in receiving their genetic feedback for alcohol use disorder, depression, and anxiety (8).
However, there are concerns about how individuals will understand and interpret PRS for complex health outcomes. PRS represent complex genetic risk information and their interpretation is further complicated by logistical constraints surrounding the calculation of genetic liability in the form of PRS (9, 10). PRS currently only capture a small amount of the variance in a trait which limits their predictive ability (9). Additionally, GWAS are primarily conducted in samples of European ancestry which limits the utility of PRS in diverse populations because PRS based on samples of European ancestries predict poorly in samples of non-European ancestries (9). Many people may not be aware of these issues and limitations when receiving and interpreting PRS information. These complexities may require individuals to have greater genetic and statistical knowledge to understand PRS information as compared to what is needed to understand genetic test results for a single gene disorder.
Indeed, there is evidence of substantial misunderstandings about the contribution of genetic and environmental factors to complex conditions (8, 11, 12), which may impact one's ability to accurately interpret complex genetic feedback. For example, one study found that 25% of young adults did not know whether substance use and psychiatric conditions were influenced by one gene or many different genes (8). Results of a recent study showed that 74.4% of the individuals who accessed PRS through a DTC website incorrectly answered at least one of the questions that assessed understanding and interpretation of PRS (13). Concerningly, the individuals who had a lower understanding of PRS had a more negative psychological reaction to receiving their PRS (13).
Given the misunderstandings related to complex genetic concepts and the increasing access to PRS via public resources, there is an urgent need to design and evaluate educational materials that can be used to help individuals understand and interpret their PRS information. One of the first efforts in this area has been led by a team of geneticists, clinicians, data visualization specialists, and software developers at the Broad Institute (14). They created a mock PRS report for cardiovascular disease with educational information about PRS, cardiovascular disease, and ways to reduce disease risk. User
feedback \((n = 10)\) suggested that the use of color, simple graphics, and percentiles helped with PRS interpretation, but the use of static reports were not ideal for disclosure and education of complex genetic results. The team adapted the mock report to create a user-centered, dynamic narrative visualization tool composed of animated graphics aided by simple, repetitive text to help communicate information about PRS and personalized medicine \((14)\). In the present study, we adapted the visualization tool to include additional text and graphics that communicate information about alcohol use disorder since alcohol use disorder is one of the most accessed PRS through DTC resources \((6)\). More importantly, alcohol use disorder is influenced by both genetic factors and environmental factors and there are actionable ways to reduce one's risk for developing alcohol use disorder through behavior, such as reducing or eliminating alcohol consumption. This uniquely positions alcohol use disorder as a starting phenotype as researchers begin to explore the feasibility and appropriateness of returning PRS for substance use and psychiatric outcomes.
The primary goal of the present study was to adapt the web-based educational tool from Brockman et al. \((14)\) for alcohol use disorder and use a randomized controlled trial to evaluate the impact of the tool on understanding of PRS for alcohol use disorder in a sample of emerging adults. We recruited a sample of college students to participate in the randomized controlled trial because emerging adulthood is a critical period for the onset of problematic alcohol use behavior \((15)\), with college students engaging in high rates of risky drinking behavior and typically using more alcohol than their non-college attending peers \((16)\). The randomized controlled trial consisted of three conditions: (1) A control condition in which participants only received a survey assessment, (2) an intervention condition in which participants received general information about alcohol use disorder, and (3) an intervention condition in which participants received information about PRS and alcohol use disorder. We evaluated the efficacy of the educational information by assessing participants' understanding of PRS for alcohol use disorder across the three conditions. We hypothesized that receiving information about PRS and alcohol use disorder would result in greater understanding of PRS for alcohol use disorder compared to both the control condition and intervention condition in which participants received information only about alcohol use disorder. Additionally, we assessed satisfaction with the educational information and whether the effect of the intervention varied across participants' demographic characteristics.
Materials and methods
Participants and recruitment
Participants for the study were recruited through Psychology's SONA (PsychSONA) system at an urban, public university during the fall 2021 semester. PsychSONA is a cloud-based participation and experiment management software that allows researchers to schedule both online and lab studies, recruit participants, monitor participants' study related activities, and grant study completion credit. Undergraduate students aged 18 years or older had the option of signing up for the study through PsychSONA. The study was described as an hour-long study in which participants could complete an online survey and learn more about alcohol use disorder and genetic risk. Participants were randomly assigned to one of three conditions: (1) A control condition in which participants received only the survey assessment, (2) an intervention condition in which participants received general information about alcohol use disorder, and (3) an intervention condition in which participants received general information about alcohol use disorder and information about PRS. Participants were emailed a unique study link to a REDCap survey which provided additional information about the study and a consent form. Following consent, participants were directed to initial survey items that assessed demographic information, personality, and current alcohol use. Depending on which condition the participants were assigned to, participants were either directed to (1) the remaining survey items, (2) a website that provided educational information about alcohol use disorder and ways to reduce risk, or (3) a website that provided educational information about PRS, alcohol use disorder, and ways to reduce risk. At the bottom of the website, participants were instructed to click a link in order to be redirected back to the survey. The remaining survey items assessed understanding and interpretation of PRS for alcohol use disorder through the use of hypothetical PRS scenarios. Participants within each condition of the randomized controlled trial were randomly assigned to receive the PRS scenarios in one of two different orders: (1) Below-average PRS, average PRS, above-average PRS or (2) above-average PRS, average PRS, below-average PRS. Approximately 50% of the participants in each condition received the scenarios in the first order and 50% received the scenarios in the second order. The additional randomization ensured that there were no order effects associated with the presentation of the various levels of genetic risk in the scenarios. A flow chart of the study design can be visualized in Supplementary Figure 1. After completion of the study, the research coordinator granted one credit to each participant through the PsychSONA system. All data was collected through REDCap \((17)\). All procedures were approved by the University's Institutional Review Board.
Educational information
The educational information evaluated in this study was adapted from the coronary artery disease PRS dynamic explainer that was designed and developed by a team of geneticists,
clinicians, software developers, and data visualization experts as a collaboration between the Broad Institute’s Cardiovascular Disease Initiative, IBM Research, and Massachusetts General Hospital (14). The original content was developed to help educate users about PRS and ways to reduce risk for cardiovascular disease using both text and visual aids. The website URL for the original PRS dynamic explainer can be found in the footnote1. The development of this PRS dynamic explainer was informed by interview feedback on mock PRS reports. Scroll-based techniques were used to introduce concepts in manageable chunks, step-by-step, with corresponding graphics that animate as the user scrolls, allowing users to control the pace of information they receive and create a clear connection between the textual explanation and graphical representation of each concept. Short sentences that use simple wording and repetitive phrases were used throughout the site to enhance comprehension. Color coding was used as the key element to communicate different levels of genetic risk with teal indicating lower genetic risk, gray indicating average genetic risk, and red indicating higher genetic risk (14). These colors were used to highlight key terms throughout the written narrative and to encode the risk information in the corresponding graphics and labels, in order to establish a continuum between the written explanation and visual representation of each concept. The educational information in the present study was adapted from the original site to focus on alcohol use disorder rather than coronary artery disease. See Supplementary Figures 2–5 for more details about the educational information.
**Intervention condition 1 (AUD Edu)**
The information provided to participants in the AUD Edu condition was related to alcohol use disorder, including a definition, prevalence, consequences, risk factors, and risk-reducing strategies. The content was developed based on educational information available through public websites, including the National Institute of Alcohol Abuse and Alcoholism, Mayo Clinic, and the National Survey on Drug Use and Health. Simple graphics were designed to easily communicate risk-reducing strategies, such as measuring drinks, finding alternative activities, and avoiding places that trigger drinking.
**Intervention condition 2 (PRS + AUD Edu)**
The information provided to participants in the PRS + AUD Edu condition explained PRS by discussing genetic variation, risk variants, how PRS are calculated, and how they can be interpreted. The participants also received the same information as the AUD Edu condition that related to alcohol use disorder, including a definition, prevalence, consequences, risk factors, and risk-reducing strategies.
---
1 http://polygenicscores.org/explained/
**Control condition**
Participants in the control condition did not receive an educational intervention prior to completing the study.
**Measures**
**Time spent accessing educational information**
For participants in the AUD Edu condition and PRS + AUD Edu condition, time spent accessing the website which contained the educational information was calculated using timestamps recorded through REDCap. The first timestamp was recorded when participants were presented with the link to the educational information and the second timestamp was recorded when participants clicked the link at the end of the website containing the educational information and returned to the survey. The time between these two time points was calculated in order to best estimate how long participants spent accessing the educational information presented to them.
**Satisfaction with the educational materials**
Satisfaction was assessed using a series of items directly related to the presentation of information and content provided. Example items included “The animation helped explain concepts,” “I learned new information about alcohol use disorder as part of this program,” and “The pacing of the information was manageable.” These items were rated on a five-point scale from strongly disagree to strongly agree. Participants were also asked to rate how useful and satisfied they were with the different sections of the educational material on a five-point scale.
**Understanding of polygenic risk scores**
The items used to assess understanding of PRS for alcohol use disorder were adapted from items used in Peck et al. (13). Two items were used to assess general understanding of PRS with response options of “true,” “false,” and “don’t know.” An example item was “A genetic risk score includes only some of the genetic factors that can contribute to risk for the condition.” Additionally, 12 items were used to assess understanding of different levels of PRS for alcohol use disorder. Participants were asked to imagine that they received each PRS for alcohol use disorder (below-average, average, and above-average) and respond to a series of questions. Below-average risk was indicated using a graph in which the PRS was in the 30th percentile, average risk was indicated using a graph in which the PRS was in the 50th percentile, and above-average risk was indicated using a graph in which the PRS was in the 75th percentile. See Supplementary Figure 6 for additional details regarding the display of hypothetical PRS. Example items for the below-average PRS scenario included “You will definitely develop alcohol use disorder” and “You have a lower chance than the average person to develop alcohol use disorder.”
primary demographic variables included in the analyses.
Demographic variables
Dimension (extraversion, neuroticism, and sensation seeking). Items were coded from the short UPPS-P Impulsive Behavior Scale (SUPPS-P) included to assess extraversion and neuroticism. Three items from the Big Five Inventory (BFI) (20) were included to assess sensation seeking. Items were coded according to the scoring guidelines of the BFI and SUPPS-P and averaged to create an overall score for each personality dimension (extraversion, neuroticism, and sensation seeking).
Sex assigned at birth, race/ethnicity, and age were the primary demographic variables included in the analyses.
A detailed breakdown of race/ethnicity is reported in Table 1. Because approximately half of the sample self-identified as White (43.6%), race/ethnicity was coded as a binary variable. Individuals who self-identified as White were coded as 0, and individuals who self-identified as Asian, Black/African American, Hispanic/Latino, American Indian/Alaska Native, Native Hawaiian/Pacific Islander, or more than one race were coded as 1.
Drinking status
Drinking status was measured using the frequency item from the Alcohol Use Disorder Identification Test (AUDIT) (18). A total of 34.8% of the full sample “never” used alcohol, 27.4% used alcohol “monthly or less,” 23.4% used alcohol “2 to 4 times a month,” 10.5% used alcohol “2 to 3 times a week,” 1.5% used alcohol “4 or more times a week,” and 2.5% chose not to answer. In view of the distribution, participants who responded “never” to the alcohol frequency item were coded as 0 to indicate that they had not previously used alcohol and participants who responded at least “monthly or less” were coded as 1 to indicate that they had previously used alcohol.
Personal history of alcohol problems
Personal history of having an alcohol use disorder was assessed using the question “Have you ever been diagnosed with an alcohol use disorder by a healthcare professional?” Response options were “yes,” “no,” and “don’t know.”
Family history of alcohol problems
Participants answered separate questions about alcohol problems for four types of biological relatives: Mother, father, aunts/uncles/grandparents, and siblings (19). An example question was: “Do you think your biological mother has ever had a drinking problem? (By drinking problem we mean that her drinking caused problems at home, at work, with her health, or with the police, or that she received alcohol treatment.).” The questions were repeated for each type of relative. Response options for each question were “yes,” “no,” and “I don’t know.”
Family history items related to alcohol problems were combined into an overall binary variable to indicate whether or not the participant had any first- or second-degree relatives with a history of alcohol problems.
Personality
Six items from the Big Five Inventory (BFI) (20) were included to assess extraversion and neuroticism. Three items from the short UPPS-P Impulsive Behavior Scale (SUPPS-P) (21) were included to assess sensation seeking. Items were coded according to the scoring guidelines of the BFI and SUPPS-P and averaged to create an overall score for each personality dimension (extraversion, neuroticism, and sensation seeking).
Analyses
Descriptive analyses were used to describe demographic information for each condition and for the full sample. ANOVA methods and chi-squared tests were used to ensure that randomization was successful across the three conditions of the randomized controlled trial. Counts and frequencies were used to examine satisfaction items presented to participants in the AUD Edu condition and PRS + AUD Edu condition. Counts and frequencies were also used to investigate individual items that assessed understanding and interpretation of PRS. To examine the effectiveness of the educational information, a one-way ANOVA was used to compare mean understanding of PRS for alcohol use disorder between the control condition, AUD Edu condition, and PRS + AUD Edu condition. Post-hoc tests were conducted to examine where the differences occurred. To examine moderators influencing the effects of the educational information, two-way ANOVAs were conducted to examine interactions between the interventions and demographic characteristics (i.e., sex, race/ethnicity, drinking status, and family history of alcohol problems) on understanding of PRS for alcohol use disorder. Additionally, a linear regression model was conducted to assess the robustness of the effect of the intervention on overall understanding of PRS for alcohol use disorder between the control condition, AUD Edu condition, and PRS + AUD Edu condition. Post-hoc tests were conducted to examine where the differences occurred. To examine moderators influencing the effects of the educational information, two-way ANOVAs were conducted to examine interactions between the interventions and demographic characteristics (i.e., sex, race/ethnicity, drinking status, and family history of alcohol problems) on understanding of PRS for alcohol use disorder. Additionally, a linear regression model was conducted to assess the robustness of the effect of the intervention on overall understanding of PRS for alcohol use disorder between the control condition, AUD Edu condition, and PRS + AUD Edu condition. Post-hoc tests were conducted to examine where the differences occurred. To examine moderators influencing the effects of the educational information, two-way ANOVAs were conducted to examine interactions between the interventions and demographic characteristics (i.e., sex, race/ethnicity, drinking status, and family history of alcohol problems) on understanding of PRS for alcohol use disorder.
Results
Sample characteristics
Figure 1 displays the recruitment details for each condition of the randomized controlled trial. A total of 477 participants signed up to participate in the study and were randomly assigned to a study condition, 371 participants (77.8%) consented to participate in the study, and 338 participants (70.9%) completed the study. In total, 12 participants were removed from analyses due to high response rates of “I choose not to answer” (>25% of all survey items). One participant completed the study twice, and the participant’s second survey completion was removed from the analyses.
The final analytic sample included 325 participants: 109 participants in the control condition, 105 participants in the AUD Edu condition, and 111 participants in the PRS + AUD Edu condition. Participant demographic characteristics for each condition of the randomized controlled trial, as well as characteristics of the entire study sample, are displayed in Table 1. A total of 70.4% of the sample self-reported sex assigned at birth as female. A total of 43.6% of the sample self-identified as White, 24.3% of the sample self-identified as Black/African American, and 18.7% of the sample self-identified as Asian. The demographic characteristics of the full sample are generally reflective of overall university demographics, with a greater percentage of females (62%) than males (38%) and 45.8% of the undergraduate student population self-identifying as White, 17.5% as Black/African American, and 13.3% as Asian. The mean age of the sample was 18.93 years. Using a series of comparison tests with a Bonferroni corrected p-value threshold of 0.005, there were no significant differences in demographic characteristics across the three conditions, indicating that randomization of participants was effective.
Satisfaction with educational websites
On average, participants in the AUD Edu condition spent 0.87 min (SD = 0.8) accessing the educational information regarding alcohol use disorder and participants in the PRS + AUD Edu condition spent 1.77 min (SD = 1.4) accessing the educational information regarding PRS and alcohol use disorder. Overall, participants in both the AUD Edu condition and the PRS + AUD Edu condition appeared to be satisfied with different aspects of the educational websites, including length, order of the content, pacing, and online delivery method. A total of 73 participants (69.5%) in the AUD Edu condition and 87 participants (78.4%) in the PRS + AUD Edu condition agreed or strongly agreed that they enjoyed the website. Additionally, almost all of the participants in
TABLE 1 Demographic characteristics for participants in each condition of the randomized controlled trial and for the total sample.
| Demographic characteristic | Control (n = 109) | AUD Edu (n = 105) | PRS + AUD Edu (n = 111) | Total (n = 325) | F/X² (P-value) |
|----------------------------|------------------|------------------|-------------------------|-----------------|--------------|
| Age | | | | | 0.38 (0.69) |
| Mean (SD) | 18.85 (1.25) | 19.06 (2.66) | 18.89 (1.25) | 18.93 (1.82) | |
| Sex assigned at birth, n (%)| | | | | 5.43 (0.07) |
| Male | 41 (57.6) | 29 (27.6) | 26 (23.6) | 96 (29.6) | |
| Female | 68 (82.4) | 76 (72.4) | 84 (76.4) | 228 (70.4) | |
| Race/Ethnicity, n (%) | | | | | 7.50 (0.82) |
| American Indian/Alaska native | 0 (0.0) | 1 (1.0) | 0 (0.0) | 1 (0.3) | |
| Asian | 20 (18.5) | 21 (20.2) | 19 (17.4) | 60 (18.7) | |
| Black/African American | 24 (22.2) | 24 (23.1) | 30 (27.5) | 78 (24.3) | |
| Hispanic/Latino | 4 (3.7) | 7 (6.7) | 8 (7.3) | 19 (5.9) | |
| More than one race | 8 (7.4) | 7 (6.7) | 7 (6.4) | 22 (6.9) | |
| Native Hawaiian/Other Pacific Islander | 1 (0.9) | 0 (0.0) | 0 (0.0) | 1 (0.3) | |
| White | 51 (47.2) | 44 (42.3) | 45 (41.3) | 140 (43.6) | |
| Unknown | 0 (0.0) | 0 (0.0) | 0 (0.0) | 0 (0.0) | |
| Race/Ethnicity (Binary), n (%)| | | | | 0.88 (0.64) |
| White | 51 (47.2) | 44 (42.3) | 45 (41.3) | 140 (43.6) | |
| Racial/Ethnic minority | 57 (52.8) | 60 (57.7) | 64 (58.7) | 181 (56.4) | |
| Previous AUD Diagnosis, n (%)| | | | | 4.07 (0.40) |
| Yes | 0 (0.0) | 2 (1.9) | 1 (0.9) | 3 (0.9) | |
| No | 109 (100.0) | 102 (98.1) | 109 (98.2) | 320 (98.8) | |
| Don’t know | 0 (0.0) | 0 (0.0) | 1 (0.9) | 1 (0.3) | |
| Alcohol use, n (%) | | | | | 6.04 (0.05) |
| Yes | 59 (55.7) | 67 (65.7) | 78 (71.6) | 204 (64.4) | |
| No | 47 (44.3) | 35 (34.3) | 31 (28.4) | 113 (35.7) | |
| Family history of alcohol problems, n (%) | | | | | 0.40 (0.82) |
| Yes | 53 (48.6) | 55 (52.9) | 55 (50.0) | 163 (50.5) | |
| No | 56 (51.4) | 49 (47.1) | 55 (50.0) | 160 (49.5) | |
| Extraversion | | | | | 1.34 (0.26) |
| Mean (SD) | 2.05 (1.02) | 2.08 (1.13) | 2.27 (1.12) | 2.13 (1.09) | |
| Neuroticism | | | | | 0.02 (0.98) |
| Mean (SD) | 2.22 (1.01) | 2.24 (0.86) | 2.25 (0.98) | 2.24 (0.95) | |
| Sensation seeking | | | | | 0.69 (0.51) |
| Mean (SD) | 1.73 (0.77) | 1.83 (0.65) | 1.83 (0.68) | 1.8 (0.70) | |
Bonferroni corrected significance threshold p < 0.05/10 = 0.005. Statistical information presented in the last column of the table assess differences across the three study conditions.
the PRS + AUD Edu condition (96.4%) reported that they learned new information about complex genetic risk through the website. Over 90% of participants in the PRS + AUD Edu condition thought that the way in which the genetic risk information was displayed, including the animation and color choice, was helpful.
Responses to items that assess understanding of polygenic risk scores
Descriptive analyses were used to investigate responses to the individual items which assessed understanding and interpretation of PRS for alcohol use disorder. Counts and frequencies of response options for each item are displayed in Table 2. Strikingly, participants across the three conditions had the highest incorrect response rate to the item “you have a chance of over X% to develop alcohol use disorder” in each PRS scenario. On average approximately 30% of participants in the PRS + AUD Edu condition correctly responded to these statements, while approximately 15% of participants in both the control condition and AUD Edu condition correctly responded to these statements. Interestingly, there was a discrepancy in understanding one's own risk compared to understanding one's risk as it relates to others in the population. For example, in the control condition, 93 participants (85.3%) understood that they...
### TABLE 2 Counts and frequencies of response options for each item that assessed understanding of polygenic risk scores for alcohol use disorder.
| Item | Correct response | Control | AUD Edu | PRS + AUD Edu |
|---------------------------------------------------------------------|------------------|---------|---------|---------------|
| | $n$ | $%$ | $n$ | $%$ | $n$ | $%$ |
| **A genetic risk score:** | | | | | | |
| Includes only some of the genetic factors that can contribute to | True | 48 | 44.0 | 66 | 62.9 | 69 | 62.2 |
| risk for the condition | False | 14 | 15.6 | 9 | 8.6 | 17 | 15.3 |
| | Don’t know | 44 | 40.4 | 30 | 28.6 | 25 | 22.5 |
| Shows that your lifestyle and environment play no role in | False | 60 | 55.1 | 65 | 61.9 | 86 | 77.5 |
| whether you develop the condition | True | 13 | 11.9 | 13 | 12.4 | 12 | 10.8 |
| | Don’t know | 36 | 33.0 | 27 | 25.7 | 13 | 11.7 |
| **Below-average genetic risk score:** | | | | | | |
| You have a lower chance than the average person to develop | Agree | 94 | 87.0 | 89 | 84.8 | 100 | 90.1 |
| alcohol use disorder | Disagree | 5 | 4.6 | 8 | 7.6 | 6 | 5.4 |
| | Don’t know | 9 | 8.3 | 8 | 7.6 | 5 | 4.5 |
| You will definitely develop alcohol use disorder | Agree | 79 | 73.2 | 84 | 80.0 | 70 | 63.1 |
| | Disagree | 15 | 13.9 | 12 | 11.4 | 33 | 29.7 |
| | Don’t know | 14 | 13.0 | 9 | 8.6 | 8 | 7.2 |
| You have a chance of just over 50% to develop alcohol use | Agree | 75 | 68.8 | 80 | 76.2 | 69 | 62.2 |
| disorder | Disagree | 19 | 17.4 | 17 | 16.2 | 33 | 29.7 |
| | Don’t know | 15 | 13.8 | 8 | 7.6 | 9 | 8.1 |
| You will definitely NOT develop alcohol use disorder | Agree | 73 | 67.0 | 72 | 68.6 | 94 | 84.7 |
| | Disagree | 30 | 27.5 | 23 | 21.9 | 15 | 13.5 |
| **Average genetic risk score:** | | | | | | |
| You have a similar chance as the average person to develop | Agree | 77 | 70.6 | 77 | 74.0 | 89 | 80.2 |
| alcohol use disorder | Disagree | 16 | 14.7 | 17 | 16.4 | 14 | 12.6 |
| | Don’t know | 16 | 14.7 | 10 | 9.6 | 8 | 7.2 |
| You will definitely develop alcohol use disorder | Agree | 67 | 62.0 | 72 | 68.6 | 93 | 83.8 |
| | Disagree | 23 | 21.3 | 20 | 12.4 | 8 | 7.2 |
| You have a chance of just over 50% to develop alcohol use | Agree | 75 | 68.8 | 80 | 76.2 | 69 | 62.2 |
| disorder | Disagree | 19 | 17.4 | 17 | 16.2 | 33 | 29.7 |
| | Don’t know | 15 | 13.8 | 8 | 7.6 | 9 | 8.1 |
| You will definitely NOT develop alcohol use disorder | Agree | 73 | 67.0 | 72 | 68.6 | 94 | 84.7 |
| | Disagree | 30 | 27.5 | 23 | 21.9 | 15 | 13.5 |
| **Above-average genetic risk score:** | | | | | | |
| You have a higher chance than the average person to develop | Agree | 93 | 85.3 | 87 | 83.7 | 97 | 87.4 |
| alcohol use disorder | Disagree | 8 | 7.3 | 13 | 12.5 | 8 | 7.2 |
| | Don’t know | 8 | 7.3 | 4 | 3.9 | 6 | 5.4 |
| You will definitely develop alcohol use disorder | Agree | 54 | 50.0 | 70 | 67.3 | 75 | 67.6 |
| | Disagree | 20 | 18.5 | 5 | 4.8 | 7 | 6.3 |
| You have a chance of just over 75% to develop alcohol use | Agree | 84 | 77.1 | 83 | 79.8 | 73 | 66.4 |
| disorder | Disagree | 15 | 13.8 | 16 | 15.4 | 31 | 28.2 |
| | Don’t know | 10 | 9.2 | 5 | 4.8 | 6 | 5.5 |
| You will definitely NOT develop alcohol use disorder | Agree | 81 | 74.3 | 78 | 75.0 | 93 | 83.8 |
| | Disagree | 20 | 18.4 | 12 | 11.5 | 9 | 8.1 |
Boldface text indicates the correct response to the item, as well as the $n$ ($\%$) of that correct response in each of the three conditions.
had a higher chance than the average person to develop alcohol use disorder when provided with the above-average genetic risk score, but 84 participants (77.1%) incorrectly agreed that their chance of developing alcohol use disorder was over 75% when provided with a PRS that was greater than the 75th percentile. This pattern occurred across the three conditions, as well as for each level of PRS provided.
Overall understanding of polygenic risk scores for alcohol use disorder
There was a significant difference in overall understanding of PRS between the three conditions (p-value < 0.001). The mean score for the control condition was 7.86 (SD = 3.06), the mean score for the AUD Edu condition was 8.38 (SD = 2.62), and the mean score for the PRS + AUD Edu condition was 9.67 (SD = 2.79), where higher scores indicated a greater understanding of polygenetic risk scores for alcohol use disorder. Significant mean differences occurred between the control condition and the PRS + AUD Edu condition (diff = 1.81; adj. p-value < 0.001) and between the AUD Edu condition and the PRS + AUD Edu condition (diff = 1.29; adj. p-value < 0.005). There was not a significant difference in understanding of PRS between the control condition and the AUD Edu condition (diff = 0.52; adj. p-value = 0.37). These results are summarized in Table 3.
There were no significant interactions between the intervention condition and demographic characteristics (i.e., sex, race/ethnicity, drinking status, and family history of alcohol problems) on understanding of PRS for alcohol use disorder (Table 4). The effect of the intervention was consistent across females and males, individuals who self-identified as White and individuals who self-identified as other races/ethnicities, individuals who use and do not use alcohol, and individuals with and without a family history of alcohol problems. Additional exploratory analyses demonstrated no significant interactions between the intervention condition and race/ethnicity when categorized as White, Black/African American, and Asian.
Additionally, the effect of receiving information about PRS on understanding appears to be robust while controlling for demographic characteristics and characteristics generally associated with substance use behaviors (e.g., family history of alcohol problems, drinking status, and personality). Receiving educational information about PRS was significantly associated with understanding of PRS for alcohol use disorder (p < 0.01) while controlling for individual characteristics (Supplementary Table 1).
### Understanding of different levels of polygenic risk scores
Lastly, we assessed whether understanding of PRS varies across different levels of genetic risk. Table 5 displays the mean understanding scores for each level of PRS (below-average, average, and above-average) for alcohol use disorder in each condition. Participants in the AUD Edu condition and PRS + AUD Edu condition had a similar understanding of each PRS regardless of whether the PRS was below-average, average, or above-average. There was a significant difference in understanding of the different PRS in the control condition (p-value < 0.01), with post-hoc analyses demonstrating a significant
---
**Table 3** Mean (SD) understanding of polygenic risk scores for alcohol use disorder across the three conditions of the randomized controlled trial.
| Condition | Mean (SD) | F-test |
|------------------|-----------|--------|
| Control | 7.86 (3.06)a | F (2, 322) = 11.96; p < 0.0001 |
| AUD Edu | 8.38 (2.62)b | |
| PRS + AUD Edu | 9.67 (2.79) | |
Values that share a superscript are significantly different (adj. p < 0.01). aRefers to significant differences between the control condition and the PRS + AUD Edu condition. bRefers to significant differences between the AUD Edu condition and PRS + AUD Edu condition.
| Moderator | Binary category | Control | AUD Edu | PRS + AUD Edu | Interaction |
|----------------------------|-----------------|---------|---------|---------------|-------------|
| | M (SD) | M (SD) | M (SD) | F-value | df | P-value |
| Sex | Male | 7.97 (3.26) | 8.14 (2.92) | 10.60 (2.73) | 1.51 | 2 | 0.22 |
| | Female | 7.79 (2.96) | 8.47 (2.51) | 9.40 (2.78) | | |
| Race/Ethnicity | White | 8.85 (2.62) | 8.82 (2.70) | 10.10 (3.14) | 1.34 | 2 | 0.26 |
| | Non-white | 7.05 (3.16) | 8.06 (2.56) | 9.39 (2.55) | | |
| Drinking status | Drinker | 8.14 (2.94) | 8.51 (2.33) | 9.56 (3.01) | 1.12 | 2 | 0.33 |
| | Non-drinker | 7.40 (3.20) | 8.40 (3.02) | 10.00 (2.33) | | |
| Family history of alcohol problems | No family history | 7.59 (3.26) | 8.63 (2.92) | 9.79 (2.81) | 0.94 | 2 | 0.40 |
| | Family history | 8.14 (2.83) | 8.22 (2.32) | 9.49 (2.78) | | |
Results from four separate two-way ANOVAs are displayed in this table.
TABLE 5 Mean (SD) understanding of each level of hypothetical polygenic risk score for alcohol use disorder.
| Condition | Below-average PRS | Average PRS | Above-average PRS | F-test |
|--------------------|-------------------|-------------|-------------------|----------------|
| | M (SD) | M (SD) | M (SD) | |
| Control | 2.48 (1.04) | 2.16 (1.20) | 2.23 (1.02) | F (1,9,201.4) |
| | | | | p = 0.01 |
| AUD Edu | 2.45 (0.93) | 2.27 (0.99) | 2.41 (0.80) | F (2,206) = 2.55; |
| | | | | p = 0.08 |
| PRS + AUD Edu | 2.82 (0.96) | 2.78 (0.99) | 2.67 (0.95) | F (2,220) = 2.22; |
| | | | | p = 0.11 |
Four items were included to assess understanding of each PRS. Items were scored as correct or incorrect and summed to create a sum score for each scenario. Range = 0–4. PRS, polygenic risk score.
mean difference between understanding of below-average PRS for alcohol use disorder and understanding of average PRS for alcohol use disorder (diff = −0.32; adj. p-value < 0.05). The mean difference between understanding of a below-average PRS and an above-average PRS was −0.25 (adj. p-value = 0.08). This suggests that participants in the control condition had a slightly better understanding of below-average PRS compared to average or above-average PRS.
Discussion
This is the first randomized controlled trial designed to evaluate the efficacy of educational information delivered through a web-based educational tool intended to increase understanding and interpretation of PRS for alcohol use disorder. The randomized controlled trial consisted of a control condition in which participants only received a survey assessment, an intervention condition in which participants received educational information about alcohol use disorder, and an intervention condition in which participants received information about PRS and alcohol use disorder. The educational information which explained PRS focused on genetic variation, risk variants, how PRS are created, and how they can be interpreted. Results showed that the educational information about PRS and alcohol use disorder significantly increased participants’ ability to understand and interpret hypothetical PRS for alcohol use disorder. The effect of receiving educational information about PRS was consistent across several demographic characteristics, including sex and race/ethnicity, demonstrating that the intervention has the potential to be equally effective across individuals of diverse backgrounds.
Additionally, participants in the PRS + AUD Edu condition had a similar understanding of each PRS regardless of whether the PRS was below-average, average, or above-average, suggesting that the intervention increased general understanding of PRS. However, participants in the control condition had a slightly better understanding of below-average PRS compared to average or above-average PRS. This could further exacerbate a negative impact of receiving an above-average PRS, as a lower understanding of PRS has been previously shown to be associated with more negative psychosocial reactions (13). These findings further demonstrate the importance of providing individuals with educational information about PRS prior to receiving genetic risk information.
It is important to note that although understanding of PRS did increase significantly, participants on average answered approximately 10 out of 14 items correct (69.1%). This suggests that there are still ways that the educational information and provision of PRS could be improved. Response rates for the individual items that assessed understanding and interpretation of PRS revealed that a majority of participants understood how the PRS impacted their chance of developing alcohol use disorder as compared to others but less than 30% of participants understood how the PRS related to their overall chance of developing alcohol use disorder. Presenting PRS as percentiles may have confused participants, as a majority of participants seemed to believe that the percentile reflected their chance of developing alcohol use disorder. Our data support existing evidence (23, 24) that providing absolute risk information to participants may aid in their understanding and interpretation of PRS. At the time of this study, translating PRS into absolute risk was complex, and not yet being routinely performed so we focused on ways to improve the comprehensibility of available PRS information. Currently, new tools have been developed that can convert PRS into absolute risk (25), and future studies could assess the outcomes of using these strategies in risk communication interventions. Further, it is possible that the limited amount of variance currently accounted for by PRS (9), and limitations in portability of PRS across ancestral groups (9) may be contributing to confusion about how PRS relate to one’s risk of developing problems.
Promisingly, participants assigned to the intervention conditions were highly satisfied with the educational information. Participants liked many aspects of the website, including the animation, colors, and presentation of the information. In the future, the content can be adapted to discuss genetic risk for many different complex disorders and diseases, which can be easily implemented as the PRS dynamic explainer is currently an online, publicly accessible website. Additionally, providing educational information about PRS through an effective online website can be a cost-effective strategy for education and can broaden access to education. As demand is high for genetic counseling services, an effective educational resource that can accompany the provision of PRS could be well-utilized and allow genetic counselors to operate to the top of their scope of practice focusing on counseling rather than providing information.
These findings should be interpreted in light of several limitations. First, the randomized controlled trial was conducted in a sample of college students. Although the educational information regarding PRS effectively increased understanding and interpretation of PRS for alcohol use disorder in this sample, the results may not be generalizable to other populations. Replicating these findings in diverse samples is important. Efficacy of the PRS dynamic explainer should be assessed in unaffected populations with diverse educational backgrounds and ages, a sample of clinicians, and affected patient populations. Second, the sample disproportionately consisted of female participants (comprising 70%); accordingly, we did not have power to test for potential sex differences. Our results may be more representative for females. Future studies should aim for equal representation of males and females. Third, the PRS were presented as hypothetical scenarios and may not reflect how an individual would understand and interpret their true PRS for alcohol use disorder. Additional research should be conducted to assess understanding of one’s true PRS information for a variety of different disorders. Additionally, the time participants spent accessing the website was estimated using two timestamps recorded with the survey software; however, we cannot assume that participants used that time to engage with the website. Due to the online nature of this study, there is no way to guarantee that participants read through all of the educational information provided to them or to assess level of engagement with the website.
Conclusion
In conclusion, the educational information utilized in this randomized controlled trial effectively increased understanding of PRS for alcohol use disorder in a sample of emerging adults. As the possibility of providing PRS information to inform prevention programing, screening, and treatment increases, the need to educate individuals about complex genetic concepts increases as well. Findings of the present study suggest that the PRS dynamic explainer could be used alongside the return of polygenic risk information in order to enhance understanding and interpretation of the genetic risk information. Future research should focus on assessing the effectiveness of the educational information in diverse samples across different age groups and educational background and assess how the information may impact one’s understanding of their true genetic risk information.
Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
Ethics statement
The studies involving human participants were reviewed and approved by the Virginia Commonwealth University Institutional Review Board. The patients/participants provided their informed consent to participate in this study through REDCap.
Author contributions
MD, SK, JA, and DD contributed to conception and design of the study. MD collected data for the study, performed the statistical analysis, and wrote the first draft of the manuscript. LP, DB, and JD created the educational tool used in the study. LP created graphics and adapted versions of the educational tool for use in the study. All authors contributed to manuscript revision, read, and approved the submitted version.
Acknowledgments
We thank the Pattern Team at the Broad Institute for their collaboration. We also thank the research participants who completed this study.
Conflict of interest
Author DB was employed by Color Health Inc. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg.2022.1025483/full#supplementary-material
References
1. Collins FS, Varmus H. A new initiative on precision medicine. N Engl J Med. (2015) 372:793-5. doi: 10.1056/NEJMp1500523
2. Visscher PM, Wray NR, Zhang Q, Sklar P, McCarthy MJ, Brown MA, et al. 10 years of GWAS discovery: biology, function, and translation. Am J Hum Genet. (2017) 101:5–22. doi:10.1016/j.ajhg.2017.06.005
3. National Society of Genetic Counselors’ Definition Task Force, Resta R, Biesecker BB, Bennett RL, Blum S, Hahn SE, et al. A new definition of genetic counseling. National Society of Genetic Counselors’ Task Force report. J Genet Couns. (2008) 17:15–77. doi:10.1007/s10897-007-9014-3
4. Hoskovec JM, Bennett RL, Carey ME, DaVanzo JE, Dougherty M, Hahn SE, et al. Projecting the supply and demand for certified genetic counselors: a workforce study. J Genet Couns. (2018) 27:16–20. doi:10.1007/s10897-017-0158-8
5. Polygenic Risk Score Task Force of the International Common Disease Alliance. Responsible use of polygenic risk scores in the clinic: potential benefits, risks and gaps. Nat Med. (2021) 27:1786–84. doi:10.1038/s41591-021-01549-6
6. Folkesen L, Pain O, Ingaason A, Werpe T, Lewis CM, Austin J. Impute-me: an open-source, non-profit tool for using data from direct-to-consumer genetic testing to calculate and interpret polygenic risk scores. Front Genet. (2020) 11:578. doi:10.3389/fgene.2020.00578
7. Regalado, A. More than 26 million people have taken an at-home ancestry test. MIT Technol Rev. (2019). Available online at: https://www.technologyreview.com/2019/02/11/103446/more-than-26-million-people-have-taken-an-at-home-ancestry-test/ (accessed Nov 6, 2022).
8. Driver MN, Kuo SIC, Dick DM, On Behalf Of The Spit For Science Working Group. Interest in genetic feedback for alcohol use disorder and related substance use and psychiatric outcomes among young adults. Brain Sci. (2020) 10:1007. doi:10.3390/brainsci101201007
9. Lewis ACF, Green RC. Polygenic risk scores in the clinic: new perspectives needed on familiar ethical issues. Genome Med. (2021) 13:14. doi:10.1186/s13073-021-00829-7
10. Palk AC, Dalvie S, de Vries J, Martin AR, Stein DJ. Potential use of clinical polygenic risk scores in psychiatry – Ethical implications and communicating high polygenic risk. Philos Ethics Humanit Med. (2019) 14:4. doi:10.1186/s13010-019-0073-8
11. Molster C, Charles T, Samanez A, O’Leary P. Australian study on public knowledge of human genetics and health. PHG. (2009) 12:84–91. doi:10.1159/000164864
12. Chapman R, Likhanov M, Selina F, Zakharov I, Smith-Woolley E, Kovas Y. New literacy challenge for the twenty-first century: genetic knowledge is poor even among well educated. J Community Genet. (2019) 10:73–84. doi:10.1007/s12687-018-0163-7
13. Peck L, Borle K, Folkesen L, Austin J. Why do people seek out polygenic risk scores for complex disorders, and how do they understand and react to results? Eur J Hum Genet. (2022) 30:81–7. doi:10.1038/s41431-021-00929-3
14. Brockman DG, Petronio L, Dron JS, Kwon BC, Vosburg T, Nip L, et al. Design and user experience testing of a polygenic score report: a qualitative assessment of prospective users. BMC Med Genomics. (2021) 14:238. doi:10.1186/s12920-021-01056-0
15. Kessler RC, Berglund P, Demler O, Jin R, Merikangas KR, Walters EE. Lifetime prevalence and age-of-onset distributions of DSM-IV disorders in the National Comorbidity Survey Replication. Arch Gen Psychiatry. (2005) 62:593–602. doi:10.1001/archpsyc.62.6.593
16. Schulenberg JE, Johnston LD, O’Malley PM, Bachman JG, Miech RA, Patrick ME. Monitoring the Future National Survey Results on Drug Use, 1975-2017. Volume II, College Students & Adults Ages 19-55. Institute for Social Research. (2018). Available online at: https://eric.ed.gov/?id=ED589764 (accessed Nov 13, 2020). doi:10.3998/2027.421.146531
17. Harris PA, Taylor R, Thieke R, Payne J, Gonzalez N, Conde JG. Research electronic data capture (REDCap): a metadata-driven methodology and workflow process for providing translational research informatics support. J Biomed Inform. (2009) 42:377–81. doi:10.1016/j.jbi.2008.08.010
18. Bohn MJ, Babor TF, Kranzler HR. The Alcohol Use Disorders Identification Test (AUDIT): validation of a screening instrument for use in medical settings. J Stud Alcohol. (1995) 56:423–32. doi:10.15288/jsa.1995.56.423
19. Dick DM, Nasim A, Edwards AC, Salvatore JE, Cho SB, Adkins A, et al. Spit for science: launching a longitudinal study of genetic and environmental influences on substance use and emotional health at a large US university. Front Genet. (2014) 5:47. doi:10.3389/fgene.2014.00047
20. John OP, Srivastava S. The big five trait taxonomy: history, measurement, and theoretical perspectives. 2nd ed. In: Pervin LA, John OP, editors. Handbook of Personality: Theory and Research. New York, NY: Guilford Press (1999). P.102–38.
21. Lynam D, Smith G, Whiteside S, Cyders M. The UPPS-P: Assessing Five Personality Pathways to Impulsive Behavior (Technical Report). West Lafayette, IN: Purdue University (2006).
22. R Core Team. R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing (2018).
23. Zipkin DA, Umscheid CA, Keating NL, Allen E, Aung K, Beyth R, et al. Evidence-based risk communication: a systematic review. Ann Intern Med. (2014) 161:270–80. doi:10.7326/M14-0295
24. Gigerenzer G, Gaissmaier W, Kurz-Müleke E, Schwartz LM, Woloshin S. Helping doctors and patients make sense of health statistics. Psychol Sci Public Interest. (2007) 8:53–96.
25. Pain O, Gillett AC, Austin JC, Folkesen L, Lewis CM. A tool for translating polygenic scores onto the absolute scale using summary statistics. Eur J Hum Genet. (2022) 30:339–48. doi:10.1038/s41431-021-01028-z | 2025-03-06T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2516,
1
],
[
2516,
7577,
2
],
[
7577,
13494,
3
],
[
13494,
19042,
4
],
[
19042,
24794,
5
],
[
24794,
27455,
6
],
[
27455,
33578,
7
],
[
33578,
40363,
8
],
[
40363,
45506,
9
],
[
45506,
51672,
10
],
[
51672,
56002,
11
],
[
56002,
61720,
12
]
]
} |
e77654602b76627cbcc874b7adfa5cc890c60c77 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 12,
"total-fallback-pages": 0,
"total-input-tokens": 44392,
"total-output-tokens": 15471,
"fieldofstudy": [
"Biology"
]
} | Analysis of Arabidopsis glucose insensitive growth Mutants Reveals the Involvement of the Plastidial Copper Transporter PAA1 in Glucose-Induced Intracellular Signaling
Shin Ae Lee, Eun Kyung Yoon, Jung-Ok Heo, Mi-Hyun Lee, Indeok Hwang, Hyeonsook Cheong, Woo Sung Lee, Yong-sic Hwang, and Jun Lim*
Department of Bioscience and Biotechnology, Konkuk University, Seoul 143–701, Korea (S.A.L., E.K.Y., J.-O.H., M.-H.L., Y.-s.H., J.L.); Department of Biotechnology, Chosun University, Gwangju 501–759, Korea (I.H., H.C.); and Department of Biological Science, Sungkyunkwan University, Suwon 440–746, Korea (W.S.L.)
Sugars play important roles in many aspects of plant growth and development, acting as both energy sources and signaling molecules. With the successful use of genetic approaches, the molecular components involved in sugar signaling have been identified and their regulatory roles in the pathways have been elucidated. Here, we describe novel mutants of Arabidopsis (Arabidopsis thaliana), named glucose insensitive growth (gig), identified by their insensitivity to high-glucose (Glc)-induced growth inhibition. The gig mutant displayed retarded growth under normal growth conditions and also showed alterations in the expression of Glc-responsive genes under high-Glc conditions. Our molecular identification reveals that GIG encodes the plastidial copper (Cu) transporter PAA1 (for P_{gi}g-type ATPase 1). Interestingly, double mutant analysis indicated that in high Glc, gig is epistatic to both hexokinase1 (hxk1) and aba insensitive4 (abi4), major regulators in sugar and retrograde signaling. Under high-Glc conditions, the addition of Cu had no effect on the recovery of gig/paa1 to the wild type, whereas exogenous Cu feeding could suppress its phenotype under normal growth conditions. The expression of GIG/PAA1 was also altered by mutations in the nuclear factors HXK1, ABI3, and ABI4 in high Glc. Furthermore, a transient expression assay revealed the interaction between ABI4 and the GIG/PAA1 promoter, suggesting that ABI4 actively regulates the transcription of GIG/PAA1, likely binding to the CCAC/ACGT core element of the GIG/PAA1 promoter. Our findings indicate that the plastidial Cu transporter PAA1, which is essential for plastid function and/or activity, plays an important role in bidirectional communication between the plastid and the nucleus in high Glc.
1 Sugars have multifaceted roles in plant growth and development, including their roles as energy sources and signaling molecules (Smeekens, 2000; Ramon et al., 2008). During the life cycle, plants can sense sugar levels, control the expression of many genes involved in physiological and developmental processes accordingly, and thus modulate growth and development to adapt to changes in sugar levels (Smeekens, 2000; Gibson, 2005; Rolland et al., 2006; Ramon et al., 2008). Genetic approaches have been fruitful in elucidating the molecular components in sugar signaling; several Arabidopsis (Arabidopsis thaliana) mutants insensitive to high-sugar conditions have been isolated and well characterized at the molecular level (Smeekens, 2000; Rook and Bevan, 2003; Gibson, 2005; Rolland et al., 2006; Ramon et al., 2008). In particular, glucose insensitive (gin) mutant seedlings are able to grow in the presence of 6% (w/v) Glc, which causes developmental arrest in wild-type seedlings (León and Sheen, 2003; Gibson, 2005). Some of these mutants are allelic to abscisic acid (ABA) biosynthesis (e.g. gin1/aba2 and gin5/aba3) or signaling mutants (e.g. gin6/aba insensitive4 [abi4]), demonstrating extensive interactions between sugar and phytohormone ABA (Arenas-Huertero et al., 2000; Cheng et al., 2002; Finkelstein and Gibson, 2002; León and Sheen, 2003; Gibson, 2005). In addition, the GIN2 locus encodes a hexokinase (HXK1), which phosphorylates Glc to glucose-6-phosphate, and the gin2 mutants overcome developmental arrest in the presence of 6% Glc (Moore et al., 2003). It has also been well established that HXK1 acts as an evolutionarily conserved Glc sensor in sugar signaling (Moore et al., 2003; Rolland et al., 2006). Coordination of growth and developmental responses to sugar levels is a complex process due to the existence of cross talk between sugar and other signaling pathways. Plastids are known as sites of photosynthesis and of the synthesis and storage of biomolecules such as...
carbohydrates and hormones (Buchanan et al., 2000; Jung and Chory, 2010). Thus, the developmental, functional, and metabolic states of plastids can act as signals that modify the expression of nuclear genes (Nott et al., 2006; Pogson et al., 2008; Kleine et al., 2009; Pfannschmidt, 2010). Extensive genetic screens have been undertaken to identify mutants impaired in the plastid-to-nucleus retrograde signaling. Such mutants showed loss of intercompartmental communication, including aberrant control of the expression of nucleus-encoded plastid genes at the transcriptional level (Nott et al., 2006; Koussevitzky et al., 2007; Pogson et al., 2008). Interestingly, the ABA signaling pathway is also implicated in retrograde signaling (Penfield et al., 2006; Shen et al., 2006; Koussevitzky et al., 2007; Kim et al., 2009; Priest et al., 2009; Jung and Chory, 2010; Leister et al., 2011). Indeed, Koussevitzky et al. (2007) demonstrated that ABI4, an AP2-type transcription factor, serves as a point of convergence and regulates nuclear gene expression in retrograde signaling. However, the impact of sugars on interorganellar communication between the plastid and the nucleus is relatively unexplored.
Copper (Cu) is a microelement essential for living organisms as a cofactor (Palmer and Guerinot, 2009). However, excess Cu causes visible toxicity in Arabidopsis, indicating that adequate amounts need to be delivered to the various subcellular compartments (Shikanai et al., 2003; Abdel-Ghany et al., 2005; Palmer and Guerinot, 2009). In particular, the Arabidopsis plastidial P1B-type ATPase Cu transporters, PAA1 (also known as AtHMA6) and its closest homolog PAA2 (also known as AtHMA8), play an important role in Cu delivery to plastids and, as a result, in the maintenance of Cu homeostasis (Shikanai et al., 2003; Abdel-Ghany et al., 2005). The Cu transporter PAA1, localized to the inner chloroplast envelope, transports Cu across the envelope into the stroma, and PAA2, localized to the thylakoid membrane, further transports Cu into the thylakoid lumen (Shikanai et al., 2003; Abdel-Ghany et al., 2005). Not surprisingly, paa1 paa2 double mutants are seedling lethal, indicating their important roles in Cu delivery during postembryonic growth and development (Shikanai et al., 2003; Abdel-Ghany et al., 2005). However, the potential role of PAA1 in sugar-induced intercompartmental signaling is largely unknown.
In a screen for altered response in sugar signaling, we identified novel Arabidopsis mutants with insensitivity to growth inhibition in the presence of 6% Glc. We report here the genetic and physiological analyses of the recessive Glc-insensitive gig mutants (for glucose insensitive growth). In the presence of 1% Glc, however, the gig mutants showed a reduction of division potential in the root meristem, resulting in the retardation of root growth. Interestingly, under high-Glc conditions, gig is epistatic to both hlxk1 and ab14, and the expression levels of the nuclear genes, in particular AB4 and HXK1, were significantly decreased in gig. We found that the GIG locus encodes the plastidial P1B-type ATPase Cu transporter PAA1 (Shikanai et al., 2003; Williams and Mills, 2005). Under high-Glc conditions, the addition of Cu had no effect on the recovery of gig/paa1 to the wild type, whereas exogenous Cu feeding, as reported previously (Shikanai et al., 2003), could suppress its phenotype under normal growth conditions. In the presence of 6% Glc, the expression of GIG/PAA1 was also altered by mutations in the nuclear factors HXK1, AB13, and AB14. A transient expression assay further revealed the interaction between AB14 and the GIG/PAA1 promoter, suggesting that AB14 actively regulates the transcription of GIG/PAA1, likely binding to the CCAC/ACGT core element of the GIG/PAA1 promoter. Our findings provide evidence for a novel function of the plastidial...
RESULTS
Identification of Novel Mutants Insensitive to High-Glc-Induced Growth Inhibition
To date, genetic approaches have been successful in identifying constituents and elucidating their roles in sugar signaling (Smeekens, 2000; Rook and Bevan, 2003; Gibson, 2005; Rolland et al., 2006; Ramon et al., 2008). In an attempt to identify additional components in the sugar signaling pathway, we screened a population of approximately 1,500 M2 activation-tagged lines for insensitivity to growth arrest under high-Glc conditions. In the mutant screening, we identified one line that exhibited insensitivity to the inhibition of seedling growth in the presence of 6% Glc (Fig. 1A). In a root growth assay, root length of the line was longer, by approximately 3-fold, than that of wild-type Columbia (Col-0) seedlings (Fig. 1B). In addition, we analyzed anthocyanin accumulation, which is enhanced by high-Glc-induced growth inhibition (Tsukaya et al., 1991; Mita et al., 1997; Xiao et al., 2000; Baier et al., 2004; Teng et al., 2005; Jeong et al., 2010), in this mutant line. In the presence of 6% Glc, anthocyanin accumulation in the mutant was reduced by approximately 2.5-fold compared with that in Col-0 (Fig. 1C). Our findings indicate that the mutant line exhibits an insensitive growth phenotype in high (6%) Glc.
To determine whether the insensitive growth phenotype is Glc specific, we analyzed this mutant line under high-Suc conditions. In the presence of 6% Suc, root growth of the mutant was slightly more insensitive than in the wild type (Fig. 1, D and E). When grown in the presence of 12% Suc, which is nearly the same as 6% Glc in molarity, the growth phenotype of the mutant line was almost indistinguishable from Col-0 seedlings (Fig. 1, F and G). To test whether the phenotype was attributable to osmotic stress, we cultured both Col-0 and the mutant in the presence of 300 mM mannitol, which is also the same as 6% Glc in molarity. We found that both the mutant and Col-0 seedlings were indistinguishably similar in growth (Fig. 1, H and I). Our observations indicate that the insensitive growth phenotype of the mutant was attributed primarily to high levels of exogenous Glc, and hence we named the mutant gig. For further genetic analysis, gig was segregated in the typical 3:1 ratio, indicating that the gig mutation was inherited as a single recessive Mendelian locus (Table I).
The gig Mutant Exhibits Growth Retardation under Normal Growth Conditions
To address whether its insensitive growth phenotype in the presence of 6% Glc is due to relatively
---
Table I. Genetic analysis of gig mutants
| Mutant | Phenotype | Total | Segregation Ratio | \( \chi^2 \) |
|----------|-----------|-------|------------------|-------------|
| Col-0 x gig F2 | Sensitive | 99 | Insensitive | 37 |
| | | 136 | 3:1 | 0.35 (P ≈ 0.05) |
---
PAA1 Cu transporter in bidirectional (“plastid-to-nucleus” and “nucleus-to-plastid”) communication in response to high Glc levels.
lower inhibition of vigor in gig seedlings, we investigated the gig mutant in the presence of 1% Glc (hereafter referred to as normal growth conditions in this study). Interestingly, gig showed a short-root phenotype compared with Col-0 seedlings under normal growth conditions (Fig. 2A). In a root growth assay, we found that differences in the growth rate between Col-0 and gig increased gradually as plants became mature, indicating that gig root growth was retarded (Fig. 2B). We further investigated the root meristem size of gig by measuring the number and length of ground cells from the quiescent center (QC) in the presence of 1% Glc, as described previously (Dello Ioio et al., 2007; Achard et al., 2009; Ubeda-Tomás et al., 2009; Heo et al., 2011). The root meristem size of gig was smaller than that of Col-0, suggesting a reduction of cell division in the meristem zone (MZ; Fig. 2, C–E). To further characterize this defect, a CYCB1;1::GUS mitotic marker (Donnelly et al., 1999) was monitored both in gig and Col-0, and indeed cell division potential in gig roots was significantly reduced (Fig. 2, F and G). Our findings indicate that the retarded root growth of gig is attributable to a decrease in cell division potential. In Arabidopsis root meristem, the stem cell niche, including the QC and initials, replenishes all the cell files (Scheres, 2007). We thus investigated developmental defects in and around the stem cell niche of gig using cell-specific markers, including SCARECROW (SCR), SHORT-ROOT (SHR), QC25, and WUSCHEL RELATED HOMEBOX5 (WOX5; Di Laurenzio et al., 1996; Helariutta et al., 2000; Nakajima et al., 2001; Sabatini et al., 2003; Gallagher et al., 2004; Sarkar et al., 2007). In the presence of 1% Glc, the stem cell niche of gig was indistinguishable from that of Col-0, indicating that the reduction of the root meristem size of gig is not due to defects in the stem cell niche (Supplemental Fig. S1). In addition to its short-root phenotype, gig adult plants were also dwarf compared with Col-0 under our growth conditions (16-h-light/8-h-dark cycles; Supplemental Fig. S2). Taken together, gig exhibited growth retardation under normal (1% Glc) growth conditions, unlike its insensitive growth under high-Glc (6% Glc) conditions.
The gig Mutant Is Epistatic to abf4 and hxk1 under High-Glc Conditions
The insensitive phenotype under high-Glc conditions raised the question of whether GIG plays a role in sugar signaling. To address this question, we adopted a genetic approach by using hxk1/gin2 and abf4/gin6 mutants, which are well-characterized mutants with defects in sugar signaling (Arenas-Huertero et al., 2000; Huijser et al., 2000; Finkelstein and Gibson, 2002; Arroyo et al., 2003; Moore et al., 2003; Acevedo-Hernández et al., 2005; Dekkers et al., 2008). In the presence of 6% Glc, both hxk1 and abf4, as expected, exhibited insensitivity to growth arrest compared with Col-0 seedlings (Fig. 3, A, C, and D). Intriguingly, gig was more insensitive in root growth compared with both hxk1 and abf4 under high-Glc conditions (Fig. 3, B–D). To investigate genetic interactions of gig and these mutants, we generated the double mutant combinations of gig hxk1 and gig abf4, respectively, and examined the growth phenotype in the presence of 6% Glc. To our surprise, the seedling growth of both gig hxk1 and gig abf4 was indistinguishable from that of the gig single mutant (Fig. 3, B, E, and F). Our findings
Figure 3. Genetic analysis of gig hxk1 and gig abf4 double mutants. Twelve-day-old seedlings of Col-0 (A), gig (B), hxk1 (C), abf4 (D), gig hxk1 (E), and gig abf4 (F) were grown on MS agar plates with 6% Glc. gig is epistatic to both hxk1 and abf4. Bar = 1 cm.
Figure 4. Molecular identification of the GIG locus. A, Location of T-DNA insertions. The boxes depict the coding regions, whereas the lines represent the noncoding regions. The red triangle denotes the position of a T-DNA insertion in gig-1 that is isolated from an activation-tagged population, whereas the blue triangle indicates the T-DNA insertion site in gig-2 that was identified from the SALK T-DNA database. The GIG locus encodes the plastidial P1B-type ATPase Cu transporter PAA1. B, Expression of PAA1 in Col-0, gig-1, and gig-2 seedlings. No expression was detected in either gig-1 or gig-2 seedlings. C, Allelism test of gig-1 and gig-2. From left to right, Col-0, gig-1, gig-2, and F1 progeny of a cross between gig-1 and gig-2 (gig-1 gig-2). The F1 progeny show insensitivity to growth inhibition in 6% Glc. Bar = 1 cm.
indicate that gig is epistatic to both hxxk1 and abi4 in high Glc.
**GIG Encodes the Plastidial P1b-Type ATPase Cu Transporter PAA1**
To determine the molecular basis of the gig phenotypes, we identified the GIG locus using thermal asymmetric interlaced PCR (Liu et al., 1995), since the mutant was initially isolated in an activation-tagged population. We found a T-DNA insertion in the sixth intron of the GIG locus (At4g33520; Fig. 4A), which encodes the plastidial P1b-type ATPase Cu transporter PAA1 (also known as AtHMA6; Shikanai et al., 2003; Williams and Mills, 2005). Previous studies have demonstrated that PAA1, localized in the chloroplast inner membrane, mediates Cu delivery into the stroma (Shikanai et al., 2003; Abdel-Ghany et al., 2005). With specific primers for PAA1, we detected no expression, indicating that, in accordance with our genetic analysis, gig is a loss-of-function mutant (Fig. 4B; Table I). Additionally, we identified another T-DNA insertion allele in the SALK database (http://signal.salk.edu) and found that this new allele, named gig-2 (SALK_043208; Fig. 4, A and B), was also similarly insensitive to growth inhibition under high-Glc conditions (Fig. 4C). Subsequently, we reciprocally crossed these mutants for a complementation test. In the presence of 6% Glc, the F1 progeny of gig-1 and gig-2 showed indistinguishably insensitive growth compared with each parental line (Fig. 4C), corroborating their allelic relation. To further verify whether mutations in the GIG/PAA1 locus cause the growth insensitivity to high levels of exogenous Glc, we transformed gig mutants with a translational fusion (hereafter referred to as pGIG::GIG-GFP), including 443 bp of the putative promoter region and the open reading frame (ORF) fused to GFP. In the presence of 1% Glc, the translational fusion restored the retarded gig growth phenotype to Col-0 (Fig. 5A). As expected, transgenic seedlings with pGIG::GIG-GFP exhibited similar sensitivity to high-Glc-induced growth inhibition compared with Col-0 (Fig. 5B). Taken together, our findings indicate that the insensitivity of gig to high-Glc-induced growth inhibition is, indeed, due to the loss of the plastidial Cu transporter PAA1 function.
**The gig/paa1 Mutant Is Not Rescued by Cu Addition In High Glc**
Previously, it was shown that paa1 mutants had a lower Cu content in the chloroplast and that the addition of exogenous 10 μM CuSO4 could suppress growth defects, indicating that PAA1, which mediates Cu delivery across the plastid envelope, is an essential component of the plastidial Cu transport system (Shikanai et al., 2003). Hence, we cultured gig/paa1 on Murashige and Skoog (MS) agar plates with increasing Cu concentrations (5, 10, and 50 μM). When grown on MS agar plates supplemented with 10 μM CuSO4, the growth of gig in the presence of 1% Glc was nearly indistinguishable from Col-0 seedlings (Fig. 5, C–E; Supplemental Fig. S3). As Shikanai et al. (2003) demonstrated previously, in the presence of 50 μM CuSO4, the growth of both Col-0 and gig/paa1 roots was inhibited (data not shown), indicating that the addition of high (50 μM) CuSO4 adversely affected seedling development of both Col-0 and gig/paa1. To further
assess the relationship between Cu and Glc, we analyzed the root growth of Col-0 and gig/paa1 in the absence or presence of 10 μM CuSO₄ with increasing Glc concentrations. In the absence of 10 μM CuSO₄, the growth of gig/paa1 was retarded, as expected, compared with Col-0, up to the point where exogenous Glc was added to 3% (w/v; Supplemental Fig. S4A). Whereas both Col-0 and gig/paa1 were nearly identical in the presence of 4% Glc, the difference in growth between Col-0 and gig/paa1 became reversed under high-Glc conditions, in that gig/paa1 was more insensitive. When supplemented with 10 μM CuSO₄, however, gig/paa1 was nearly indistinguishable from Col-0 seedlings up to 5% Glc (Supplemental Fig. S4B). Interestingly, gig/paa1 was still insensitive to high (6% and 7%) Glc, with no recovery of the growth defect (Fig. 5, F and G; Supplemental Fig. S4B). These findings indicate that the addition of Cu had no effect on reverting gig/paa1 to the wild type under high-Glc conditions, whereas exogenous Cu feeding could suppress its phenotype under normal growth conditions.
Expression Analysis of GIG/PAA1
To obtain further insight into gig/paa1 phenotypes, we investigated the in planta expression patterns of GIG/PAA1. To this end, we generated transgenic lines harboring a transcriptional fusion of the GUS marker gene under the control of the GIG/PAA1 cis-regulatory sequence located upstream of the ORF (hereafter referred to as pGIG:GUS). The putative promoter sequence selected was the longest that was used for molecular complementation of the gig/paa1 mutant. Therefore, we expected that this cis-regulatory sequence would be informative in monitoring the expression patterns of GIG/PAA1 in planta. In the shoot of 12-d-old seedlings, we observed GUS staining in marginal regions of cotyledons and leaves and primarily in the vasculature (Fig. 6, A and B). In the root, the GIG/PAA1 expression was rather cell type specific, being detected only in the vascular tissues (Fig. 6C). In parallel, we also analyzed the seedling root with pGIG::GIG-GFP (translational fusion), and similarly, localization of GIG-GFP was observed only in the vasculature (Fig. 6D). For a more detailed analysis, we generated transverse sections of the primary root and detected GUS expression in the vascular bundle (Fig. 6E). Interestingly, however, no GUS expression was detected in the root tip (Fig. 6F), implying that cells in the MZ are more sensitive to fluctuations in Glc levels. Our findings suggest the involvement of the plastidial Cu transporter PAA1 in nongreen tissues, in which we primarily observed the growth retardation in gig. We also analyzed the levels of GIG/PAA1 expression by an independent, complementary method: reverse transcription-based quantitative (qRT)-PCR. In our analysis, the expression of GIG/PAA1 was detected in various organs, with the highest expression in the flower (Fig. 6G). Additionally, the levels of GIG/PAA1 mRNA accumulation were significantly increased, by approximately 3-fold, in the presence of 6% Glc, indicating that its expression is also subject to regulation by high Glc (Fig. 6H).
Expression of Glc-Responsive Genes in gig/paa1
In addition to growth arrest, high levels of exogenous Glc also regulate a wide variety of genes at the transcriptional level, including APL3 (a large subunit of ADP-Glc pyrophosphorylase involved in starch biosynthesis) and CHS (for chalcone synthase; Koch, 1996; Li et al., 2006). Hence, we analyzed the expression levels of these Glc-responsive genes in both Col-0 and gig/paa1 in the presence of 1% and 6% Glc. Under high-Glc conditions, the APL3 mRNA level was, as expected, markedly elevated in Col-0, whereas the level of induction was significantly reduced in gig/paa1 (Fig. 7A). Likewise, the CHS mRNA level was increased in the presence of 6% Glc, but in gig/paa1, the level of induction was significantly reduced (Fig. 7B). On the basis of our findings that gig is epistatic to both
Figure 6. Expression pattern of GIG/PAA1. A to C, Tissue-specific expression of the transcriptional GUS fusion (pGIG::GUS) in the whole seedling (A), shoot (B), and root (C). D, Confocal image of the translational GFP fusion (pGIG::GIG-GFP) in the root. E, Transverse root section of 12-d-old seedling. Blue GUS staining is detected only in the vasculature. F, No expression is detectable in the root tip. G, GIG transcript levels in different organs as determined by qRT-PCR. The GIG mRNA level in the shoot is arbitrarily set to 1. H, Expression of GIG in the presence of 1% and 6% Glc by qRT-PCR. The statistical significance of differences was determined by Student’s t test (* P < 0.05). Error bars indicate SE from three biological replicates.
abi4 and hxxk1 in response to high-Glc concentrations, we subsequently analyzed the transcript levels of ABI4 and HXXK1 in Col-0 and gig/paa1 in the presence of 1% and 6% Glc. Under high-Glc concentrations, the HXXK1 mRNA level was, as expected, also induced by approximately 3-fold in Col-0, whereas it was significantly reduced in gig/paa1 (Fig. 7C). Interestingly, high-Glc activation of ABI4 was completely abolished in gig/paa1 (Fig. 7D). To investigate the extent of Glc-induced gene regulation, we also examined the ABI3 and ABI5 expression levels, which are known to be induced by high Glc (Cheng et al., 2002; Arroyo et al., 2003; Yuan and Wysocka-Diller, 2006; Dekkers et al., 2008). In the presence of 6% Glc, such Glc activation of ABI3 was almost completely eliminated in gig/paa1 (Fig. 7E). Likewise, induction of ABI5 expression was markedly reduced in gig/paa1 (Fig. 7F). Furthermore, we also analyzed the expression of APL3, HXXK1, and ABI4 in the presence of 10 μM CuSO4. Interestingly, the expression levels of these genes in gig/paa1 were not
---
**Figure 7.** Expression analysis of Glc-responsive genes. qRT-PCR in Col-0 and gig is shown in the presence of 1% (blue) and 6% (red) Glc. A, APL3. B, CHS. C, HXXK1. D, ABI4. E, ABI3. F, ABI5. The statistical significance of differences was determined by Student’s t test (* P < 0.05). Error bars indicate ± from biological triplicates.
**Figure 8.** Regulation of GIG expression levels by ABA. A, GIG transcript levels in abi3, abi4, abi5, and hxxk1 in the presence of 1% and 6% Glc. B, Expression of GIG in the absence or presence of 10 μM ABA. C, Transient expression assay for the interaction between ABI4 and the GIG promoter. The effector and reporter plasmids are schematically shown. Relative LUC activity was determined by the addition of Glc alone, ABI4 alone, and both Glc and ABI4. Error bars indicate ± from biological triplicates.
recovered to the Col-0 levels (Supplemental Fig. S5). This observation is consistent with our previous finding (Fig. 5, F and G; Supplemental Fig. S4) that the addition of Cu in high Glc could not suppress the gig/paa1 phenotype.
Taken together, our expression studies indicate that the loss of the plastidial Cu transporter PAA1 function significantly alters the expression levels of Glc-responsive genes. In particular, a reduction of the expression of nuclear factors ABI3, ABI4, ABI5, and HXK1 in gig/paa1 under high-Glc conditions suggests a novel retrograde plastid-to-nucleus signaling, in which the plastidial Cu transporter PAA1 is involved.
**ABI4 Is Essential for the Activation of GIG/PAA1 Expression**
Our findings that the loss of GIG/PAA1 function results in a dramatic reduction of the nuclear gene expression of ABI3, ABI4, ABI5, and HXK1 in the presence of high Glc and that the GIG/PAA1 promoter itself contains a CCAC/ACGT sequence for ABI4 binding (Supplemental Fig. S6; Strand et al., 2003; Koussevitzky et al., 2007) raised the question of whether the transcription of GIG/PAA1 itself is subject to regulation by the nuclear transcription factor ABI4. To test this, we first examined the expression of GIG/PAA1 in abi3, abi4, and abi5 in the presence of 1% and 6% Glc. As shown in Figure 8A, the GIG/PAA1 mRNA levels were significantly reduced in both abi3 and abi4, but not in abi5, under high-Glc conditions. In addition, we also monitored the GIG/PAA1 expression in hxx1 in the presence of 1% and 6% Glc, since the Glc sensor HXK1 can be localized in the nucleus and regulate gene expression (Cho et al., 2006). In high Glc, the induction of GIG/PAA1 mRNA was almost completely abolished in hxx1 (Fig. 8A). Next, we addressed whether the expression of GIG/PAA1 is regulated by ABA. Indeed, the level of GIG/PAA1 mRNA accumulation was substantially increased in the presence of 10 μM ABA (Fig. 8B).
To further investigate the interaction between ABI4 and the GIG/PAA1 promoter, we performed a transient expression assay using Arabidopsis protoplasts as described previously (Yoo et al., 2007). The reporter plasmid containing the 443-bp fragment that was used for both transcriptional fusion and molecular complementation, and the effector plasmid 35S::ABI4, were introduced into Arabidopsis protoplasts in the absence or presence of high Glc (300 μM). When relative luciferase (LUC) activity was monitored, the expression of GIG/PAA1 was, as expected, induced by high Glc alone and ABI4 alone (Fig. 8C). Interestingly, GIG/PAA1 expression was synergistically induced by both high Glc and ABI4, suggesting that the nuclear transcription factor ABI4 actively regulates the transcription of GIG/PAA1 that is essential for plastid function and/or activity. Our findings suggest a molecular mechanism of bidirectional (plastid-to-nucleus and nucleus-to-plastid) communication in response to high Glc levels.
**DISCUSSION**
In this study, we identified novel mutants, designated gig, that showed insensitivity to growth inhibition in high (6%) Glc. The expression levels of Glc-responsive ABI3, ABI4, ABI5, APL3, CHS, and HXK1 genes were significantly altered in gig/paa1 in the presence of 6% Glc. Interestingly, gig abi4 and gig hxx1 double mutants were indistinguishable from the gig single mutant under high-Glc conditions, indicating that gig is epistatic to both abi4 and hxx1. Subsequent molecular cloning led us to the conclusion that the insensitivity to high-Glc-induced growth inhibition was caused by the loss of plastidial Cu transporter PAA1 function (Shikanai et al., 2003). When complemented with the GIG/PAA1 genomic fragment or supplemented with exogenous Cu, growth defects in the gig/paa1 mutants were completely restored to levels indistinguishable from Col-0 seedlings. These results, together with those from the phenotypic analyses of gig/paa1, indicate the important role of the GIG/PAA1 Cu transporter in Glc-induced retrograde plastid-to-nucleus signaling.
In retrograde signaling, particularly plastid signaling revealed in this study, integrated information on developmental, functional, and metabolic states is conveyed to the nucleus, in which the expression of nuclear genes is modified accordingly (Nott et al., 2006; Pogson et al., 2008; Kleine et al., 2009; Pfannschmidt, 2010; Leister et al., 2011). In particular, ABI4 plays an important role in the integration of retrograde signaling to regulate the transcription of nuclear genes (Koussevitzky et al., 2007). Furthermore, intracellular communication between the plastid and the nucleus, as
suggested previously (Jung and Chory, 2010), is essentially bidirectional. Thus, we investigated whether the *GIG/PAA1* gene itself is subject to transcriptional regulation by the nuclear transcription factor ABI4. First, we found a significant reduction of the *GIG/PAA1* transcript levels in *abi3*, *abi4*, and *hxk1* mutants. Next, we analyzed the interaction between ABI4 and the *GIG/PAA1* promoter, which contains a combination of the CCAC/ACGT core element for ABI4 binding and retrograde signaling (Strand et al., 2003; Koussevitzky et al., 2007). Indeed, our transient expression assay reveals that the transcription of *GIG/PAA1* is synergistically regulated by high Glc and ABI4. These results lend support to our hypothesis that the plastidial Cu transporter PAA1 plays a role in the coordination of the bidirectional intercompartmental communication between the plastid and the nucleus in the presence of high Glc.
Previously, it was shown that miRNA398 levels, which function to repress two Cu/zinc superoxide dismutases (CSD1 and CSD2), were decreased by the addition of Cu (Sunkar et al., 2006; Yamasaki et al., 2007; Dugas and Bartel, 2008). Moreover, Dugas and Bartel (2008) demonstrated that in the presence of Suc, miRNA398 levels were increased, whereas the protein levels of CSD1 and CSD2, the miRNA398 targets, were substantially decreased. The inverse relationship between Cu and sugars in the regulation of miRNA398 levels, and in turn the CSD1 and CSD2 levels, suggests that sugars affect the results of Cu feeding. Interestingly, we also found that the addition of Cu could not revert gig/paa1 to the wild type in the presence of high Glc. Since it was demonstrated that both transcript and protein levels of CSD1 and CSD2 were higher in gig/paa1 than in the wild type (Shikanai et al., 2003; Abdel-Ghany et al., 2005), it will be interesting to determine the levels of miRNA398 accumulation in gig/paa1 in the absence or presence of high Glc with the addition of Cu.
Since *GIG/PAA1* is expressed in roots as well as in green tissues (Shikanai et al., 2003; Abdel-Ghany et al., 2005) and reactive oxygen species (ROS) can modulate the balance between cell proliferation and differentiation in roots (Tsukagoshi et al., 2010), it is tempting to speculate that ROS and/or sugar levels in plastids of both roots and green tissues may be influenced by proper Cu delivery to CSD1 and CSD2, which is controlled by PAA1 (Shikanai et al., 2003; Abdel-Ghany et al., 2005). Thus, changes in ROS, sugar levels, functional states, and/or activities of plastids caused by high Glc can act as retrograde signals (Oswald et al., 2001; Fey et al., 2005) and in turn regulate the expression of nuclear genes such as *ABI3, ABI4, ABI5, and HXK1* (Nott et al., 2006; Koussevitzky et al., 2007; Pogson et al., 2008; Kleine et al., 2009; Pfannschmidt, 2010). Subsequently, alterations in the expression levels of the nuclear factors, in particular ABI4, which can bind directly to the CCAC/ACGT core element of the *GIG/PAA1* promoter, regulate the expression of the plastidial Cu transporter PAA1 that is essential for plastid function and/or activity (Fig. 9).
In summary, we report a previously unrecognized role for the plastidial Cu transporter PAA1 in the coordination of Glc-induced intracellular signaling. Further molecular and biochemical characterization with respect to the nature of retrograde signals (e.g. ROS, sugar levels, or both), for which GIG/PAA1 may well be responsible, will be necessary to elucidate the precise molecular mechanism of Glc-induced inter-compartmental signaling between the plastid and the nucleus.
**MATERIALS AND METHODS**
**Plant Materials and Growth Conditions**
The Arabidopsis (*Arabidopsis thaliana*) ecotype Col-0 was used as the wild-type control in this study, except for *abi3-1* (*Landsberg erecta*) and *abi5-1* (Wassilevskija). The Col-0 T-DNA insertion mutants *hxk1* (SALK_070739) and *gig-2* (SALK_043208) were identified in the SGN-AN Web site (http://signal.salk.edu) and obtained from the Arabidopsis Biological Resource Center. Other mutant and transgenic lines used in this study have been described previously: *abi1-1* (Finkelstein et al., 1998), CYCB1::GUS (Donnelly et al., 1999), QC25 (Sabatini et al., 2003), WOX5 (Sarker et al., 2007), SCR (Di Laurenzio et al., 1996; Gallagher et al., 2004), and SHR (Helariutta et al., 2000; Nakajima et al., 2001). Seeds were sterilized with 2% sodium hypochlorite and 0.15% Tween 20 for 2 min and rinsed three to five times in sterile water. After stratification for 4 d in the dark at 4°C, seeds were cultured vertically on one-half-strength MS agar plates containing 1% (w/v) Glc, 6% (w/v; approximately 300 mM) Suc, 6% (w/v; approximately 350 mM) mannitol, or 10 mM CuSO₄ as indicated (in culture rooms with a 16-h-light/8-h-dark cycle at 22°C). For adult plants, MS agar plate-grown seedlings were transferred to soil and grown to full maturity in culture rooms with a 16-h-light/8-h-dark cycle at 22°C as described previously (Heo et al., 2011).
**Screening for gig and Isolation of the GIG Locus**
For the identification of *gig*, an activation tagging library was generated as described (Weigel et al., 2000; Hwang et al., 2010). A population of approximately 1,500 M₂ lines was screened for insensitivity to growth arrest on one-half-strength MS agar plates containing 6% (w/v) Glc. The *gig* mutant with enhanced high-Glc tolerance relative to wild-type seedlings was visually identified and was subsequently backcrossed to Col-0 at least four times for further analysis. To identify the *GIG* locus, thermal asymmetric interlaced PCR was performed as described previously (Liu et al., 1995). The primer sequences are listed in Supplemental Table S1.
**Root Growth Assays and Statistical Analysis**
Digital images of seedlings grown vertically on one-half-strength MS agar plates were taken using an SP-560UZ digital camera (Olympus) at each time point as indicated. Root length was measured from the digital images of the plates using ImageJ software (http://rsbweb.nih.gov/ij). The experiments were independently repeated three times, and the data were analyzed using the Excel statistical package (Microsoft). Student’s *t* test was performed to compare the mean values of triplicates, and *SE* values are indicated.
**Anthocyanin Quantitation**
Anthocyanin accumulation in Arabidopsis seedlings was quantitatively determined as described previously (Mita et al., 1997; Teng et al., 2005; Sollanelli et al., 2006) with minor modifications. For anthocyanin extraction, frozen, homogenized seedlings (100 mg) at 12 d after germination were incubated with 800 μL of 1% (v/v) hydrochloric acid in methanol overnight at 4°C. Subsequently, 400 μL of distilled water and 200 μL of chloroform were added to the mixture and mixed vigorously by vortexing. After centrifugation at 13,000 rpm for 15 min, the supernatant was separated and the absorbance was measured at 530 and 657 nm using a spectrophotometer, as described previously (Mita et al., 1997; Teng et al., 2005; Sollanelli et al., 2006).
Plasmid Construction and Transformation
To generate a transcriptional fusion of the GIG/PAA1 locus to the GUS reporter gene, a 443-bp fragment containing the cis-sequence upstream of the start codon, which is the longest intergenic region of the GIG/PAA1 locus, was PCR amplified and subsequently cloned into the pENTR/D-TOPO vector (Invitrogen). The error-free promoter fragment of GIG/Pah was subcloned into the binary pMDC162 vector (Curtis and Grossniklaus, 2003) using Gateway recombination cloning technology (Invitrogen).
For molecular complementation, we generated a translational fusion of the wild-type GIG/PAA1 ORF to GFP under the control of the 443-bp GIG/PAA1 promoter into the binary pMDC107 vector (Curtis and Grossniklaus, 2003) using Gateway technology (Invitrogen). The resulting plasmids were introduced into Agrobacterium tumefaciens (GV3101) and then introduced into Col-0 and ggi plants by the floral dipping method (Clough and Bent, 1998).
For the transient expression assay, the reporter and effector plasmids were constructed using the pBl221 vector. In the reporter plasmid, the GUS gene was replaced with the firefly LUC gene, and the cauliflower mosaic virus 35S promoter was replaced with the 443-bp GIG/PAA1 promoter. To generate the effector plasmid, the full-length ABI4 coding region was amplified by PCR and then inserted into pBI221, replacing the GUS gene. As a control plasmid, the GUS gene in pBI221 was replaced with the Renilla LUC gene.
Histochemical GUS Assays and Microscopy
The protocols used for the histochemical localization of GUS activity were essentially the same as those described previously (Yu et al., 2010; Heo et al., 2011) with minor modifications. MS agar plate-grown seedlings at 12 d after germination were incubated overnight at 37°C in GUS staining solution [0.1 mM 5-bromo-4-chloro-3-indoxyl-beta-D-glucuronic acid, 2 mM K2Fe(CN)6, 2 mM K3Fe(CN)6, 0.1 mM sodium phosphate, 10 mM EDTA, and 0.1% Triton X-100]. After staining overnight, the samples were washed and cleared as described (Yu et al., 2010; Heo et al., 2011). Plastic sections of GUS-stained plants were used for qRT-PCR using SYBR Premix ExTaq reagents (Takara) with the ABI5 reporter plasmids for 5 min. The plasmids were prepared with the AxyPrep Maxi-Plasmid kit (Axygen), and a total of 10 μg of plasmid DNA was used at a ratio of 9:2 (effector:reporter:control). After transformation, protoplasts were incubated in the washing and incubation buffer (500 μM mannitol, 4 mM MES, and 20 mM KCl) in the absence or presence of 300 μM Glc at 22°C for 16 h in the dark. Harvested protoplasts were lysed and measured for LUC activity using the Dual-Luciferase Reporter Assay System (Promega). The reporter gene activity was normalized by Renilla LUC activity. Experiments were independently repeated three times for biological replicates.
Sequence data from this article can be found in The Arabidopsis Information Resource data libraries under accession numbers AT4G33520 (GIG/PAA1), AT3G18780 (ACTIN2), AT4G39210 (APL3), AT5G13930 (CHS), AT1G29130 (HMK3), AT1G16540 (ABI3), AT2G40220 (ABI4), and AT2G36270 (AB15).
Supplemental Data
The following materials are available in the online version of this article.
Supplemental Figure S1. Analysis of the stem cell niche in Col-0 and ggi roots.
Supplemental Figure S2. Comparative analysis of Col-0 and ggi adult plants.
Supplemental Figure S3. Complementation of ggi with Cu supplementation.
Supplemental Figure S4. Root growth assay in the absence or presence of Cu with increasing Glc concentrations.
Supplemental Figure S5. Expression analysis of Glc-responsive genes with the addition of Cu.
Supplemental Figure S6. Prediction of an ABA-responsive element sequence in the GIG promoter.
Supplemental Table S1. Sequence information of PCR primers used in this study.
ACKNOWLEDGMENTS
We thank Eun Shion Hong, Dong Sung Jang, Gyuree Kim, and Gyu Min Lee for technical assistance. We also thank the Arabidopsis Biological Resource Center and the Nottingham Arabidopsis Stock Centre for plant materials.
Received November 30, 2011; accepted May 10, 2012; published May 11, 2012.
dehydrogenase/reductase in Arabidopsis thaliana glucose signaling and asbics acid biosynthesis and functions. Plant Cell 14: 2723–2743
Co YH, Yoo S-D, Sheen J (2006) Regulatory functions of nuclear hexas- kinasel complex in glucose signaling. Cell 127: 579–589
Clough SJ, Bent AF (1998) Floral dip: a simplified method for Agrobacterium-mediated transformation of Arabidopsis thaliana. Plant J 16: 735–743
Curts MD, Grossniklaus U (2003) A Gateway cloning vector set for high-throughput functional analysis of genes in plants. Plant Physiol 133: 462–469
Dekkers BJ, Schuurmans JA, Smeekens SC (2008) Interaction between sugar and asbicsic acid signaling during early seedling development in Arabidopsis. Plant Mol Biol 67: 151–167
Dello Ioio Ri, Linhares FS, Scacchi E, Casamitjana-Martinez E, Heidstra R, Costantino P, Sabatini S (2007) Cytokinins determine Arabidopsis root- meristem size by controlling cell differentiation. Curr Biol 17: 678–682
Di Laurenzio I, Wysocka-Diller J, Malamy J, Pysy I, Helariutta Y, Frenshour G, Hahn MG, Feldmann KA, Beney PN (1996) The SCARECROW gene reg- ularizes an asymmetric cell division that is essential for generating the radial organization of the Arabidopsis root. Cell 86: 423–433
Donnelly PM, Bonetta D, Tsukaya H, Dengler RE, Dengler NG (1999) Cell cycling and cell enlargement in developing leaves of Arabidopsis. Dev Biol 215: 407–419
Dugas DV, Bartel B (2008) Sucrose induction of Arabidopsis mii398 re- presses two Cu/Zn superoxide dismutases. Plant Mol Biol 40: 403–417
Fey V, Wagner R, Brautigam K, Wirtz M, Hel R, Dietzmann A, Leister D, Oelmüller R, Pfannschmidt T (2005) Retrograde plastid redox signals in the expression of nuclear genes for chloroplast proteins of Arabidopsis thaliana. J Biol Chem 280: 5318–5328
Finkelstein RR, Gibson SI (2002) ABA and sugar interactions regulating development: cross-talk or voices in a crowd?Curr Opin Plant Biol 5: 26–32
Finkelstein RR, Wang ML, Lynch TJ, Rao S, Goodman HM (1998) The Arabidopsis asbicsic acid response locus ABA1 encodes an APETALA2 domain protein. Plant Cell 10: 1043–1054
Gallaher KL, Paquette AJ, Nakajima K, Beney PN (2004) Mechanisms regulating SHORT-ROOT intercellular movement. Curr Biol 14: 1847–1851
Gibson SI (2005) Control of plant development and gene expression by sugar signaling. Curr Opin Plant Biol 8: 93–102
Helariutta Y, Fukaki H, Wysocka-Diller J, Nakajima K, Jung J, Sena G, Hauser MT, Beney PN (2000) The SHORT-ROOT gene controls radial patterning of the Arabidopsis root through radial signaling. Cell 101: 555–567
Heo JO, Chang KS, Kim IA, Lee MH, Lee SA, Song SK, Lee MM, Lim J, Hwang I, Kim SY, Kim CS, Park Y, Tripathi GR, Kim SK, Cheong H (2010) Over-expression of the IG1 leading to altered shoot-brancing development related to MAX pathway in Arabidopsis. Plant Mol Biol 73: 629–641
Jeong SW, Das PK, Jeoung SC, Song JY, Lee HK, Kim YK, Kim WJ, Park YI, Yoo SD, Choi SB, et al (2010) Ethylene suppression of sugar-induced anthocyanin pigmentation in Arabidopsis. Plant Physiol 154: 1514–1531
Jung HS, Chory J (2010) Signaling between chloroplasts and the nucleus: can a systems biology approach bring clarity to a complex and highly regulated pathway? Plant Physiol 152: 453–459
Kim C, Lee KP, Barah A, Nater M, Göbel C, Feussner I, Apel K (2009) O2-mediated retrograde signaling during late embryoogenesis pre-de- termines plastid differentiation in seedlings by recruiting asbicsic acid. Proc Natl Acad Sci USA 106: 9920–9924
Kleine T, Voigt C, Leister D (2009) Plastid signalling to the nucleus: cross-talk or voices in a crowd? Curr Opin Plant Biol 12: 185–192
Koch KE (1996) Carbohydrate-modulated gene expression in plants. Ann Rev Plant Physiol Plant Mol Biol 47: 509–530
Kousovitzky S, Nott A, Mockler TC, Hong F, Sachetto-Martins G, Sunjariyam J, Li M, Mittler R, Chory J (2007) Signals from chloroplasts converge to regulate nuclear gene expression. Science 316: 715–719
Lee MH, Kim B, Song SK, Heo JO, Yu NI, Lee SA, Kim M, Kim DG, Sohn SO, Lim CE, et al (2008) Large-scale analysis of the GRAS gene family in Arabidopsis thaliana. Plant Mol Biol 67: 659–670
Leister D, Wang X, Haberer G, Mayer KFX, Klein T (2011) Intracom- partmental and intercompartamental transcriptional networks coordi- nate the expression of genes for organelar functions. Plant Physiol 157: 386–404
León P, Sheen J (2003) Sugar and hormone connections. Trends Plant Sci 8: 110–116
Li Y, Lee KK, Walsh S, Smith C, Hadingham S, Sorefan K, Cawley G, Bevan MW (2006) Establishing glucose- and ABA-regulated transcriptional networks in Arabidopsis by microarray analysis and promoter classification using a Relevance Vector Machine. Genome Res 16: 414–427
Liu YG, Mitsukawa N, Oosumi T, Whittier RF (1995) Efficient isolation and mapping of Arabidopsis thaliana T-DNA insert junctions by thermal asymmetric interlaced PCR. Plant J 8: 457–463
Mita S, Murano N, Azaize M, Nakamura K (1997) Mutants of Arabidopsis thaliana with pleiotropic effects on the expression of the gene for β-amylose and on the accumulation of anthocyanin that are inducible by sugars. Plant J 11: 841–851
Moore B, Zhou L, Rolland F, Hall Q, Cheng WH, Liu YX, Hwang I, Jones T, Sheen J (2003) Role of the Arabidopsis glucose sensor HXK1 in nutri- ent, light, and hormonal signaling. Science 300: 332–336
Nakajima K, Sena G, Nawy T, Beney PN (2001) Intracellular movement of the putative transcription factor SHR in root patterning. Nature 415: 307–311
Nott A, Jung HS, Kousovitzky S, Chory J (2006) Plastid-to-nucleus retro- grade signaling. Annu Rev Plant Biol 57: 739–759
Osvald O, Martin T, Dominy PJ, Graham IA (2001) Plastid redox state and sugars: interactive regulators of nuclear-encoded photosynthetic gene expression. Proc Natl Acad Sci USA 98: 2047–2052
Palmer CM, Guerinot ML (2009) Facing the challenges of Cu, Fe and Zn homeostasis in plants. Nat Chem Biol 5: 333–340
Penfield S, Li Y, Gilday AD, Graham S, Graham IA (2006) Arabidopsis ABA INSENSITIVE4 regulates lipid mobilization in the embryo and reveals re- pression of seed germination by the endosperm. Plant Cell 18: 1887–1899
Pfannschmidt T (2010) Plastidial retrograde signalling: a true “plastid factor” or just metabolic signatures? Trends Plant Sci 15: 427–435
Pogson BJ, Woo NS, Füster B, Small ID (2008) Plastid signalling to the nucleus and beyond. Trends Plant Sci 13: 602–609
Priest HD, Filichkin SA, Mockler TC (2009) cis-regulatory elements in plant cell signaling. Curr Opin Plant Biol 12: 643–649
Ramon M, Rolland F, Sheen J (2008) Sugar sensing and signaling. The Arabidopsis Book. 6: e0117, doi/10.1199/tab.0117
Rolland F, Baena-Gonzalez E, Sheen J (2006) Sugar sensing and signaling in plants: conserved and novel mechanisms. Annu Rev Plant Biol 57: 675–709
Rook F, Bevan MW (2003) Genetic approaches to understanding sugar- response pathways. J Exp Bot 54: 495–501
Sabatini S, Heidstra R, Wildwater M, Scheres B (2003) SCARECROW is involved in positioning the stem cell niche in the Arabidopsis root meristem. Genes Dev 17: 472–488
Sarkar AK, Luijten M, Miyashima S, Lenhard M, Hashimoto T, Nakajima K, Scheres B, Heidstra R, Laux T (2007) Conserved factors regulate signalling in Arabidopsis thaliana shoot and root stem cell organizers. Nature 446: 811–814
Scheres B (2007) Stem-cell niches: nursery rhymes across kingdoms. Nat Rev Mol Cell Biol 8: 345–354
Shen YY, Wang XF, Wu FQ, Du SY, Cao Z, Shang Y, Wang XL, Peng CC, Yu XC, Zhu SY, et al (2006) The Mg-chelatase H subunit is an asbicsic acid receptor. Nature 443: 823–826
Shikanai T, Müller-Moulé P, Munekage Y, Niyogi KK, Pilon M (2003) PAA1, a P-type ATPase of Arabidopsis, functions in copper transport in chloroplasts. Plant Cell 15: 1333–1346
Smeekens S (2000) Sugar-induced signal transduction in plants. Annu Rev Plant Physiol Plant Mol Biol 51: 49–81
Solfanelli C, Poggi A, Loreti E, Alpi A, Perata P (2006) Sucrose-specific induction of the anthocyanin biosynthetic pathway in Arabidopsis. Plant Physiol 140: 637–646
Strand A, Asami T, Alonso J, Ecker JR, Chory J (2003) Chloroplast to nu- cleus communication triggered by accumulation of Mg-protoporphrinIX. Nature 421: 79–83
Sunkar R, Kapoor A, Zhu JK (2006) Posttranscriptional induction of two Cu/Zn superoxide dismutase genes in Arabidopsis is mediated by downregulation of miR398 and important for oxidative stress tolerance. Plant Cell 18: 2051–2065
Teng S, Keurentjes J, Bentsink L, Koornneef M, Smeekens S (2005) Sucrose-specific induction of anthocyanin biosynthesis in Arabidopsis requires the MYB75/PAP1 gene. Plant Physiol 139: 1840–1852
Tsukagoshi H, Busch W, Benfey PN (2010) Transcriptional regulation of ROS controls transition from proliferation to differentiation in the root. Cell 143: 606–616
Tsukaya H, Obshima T, Naito S, Chino M, Komeda Y (1991) Sugar-dependent expression of the CHS-A gene for chalcone synthase from petunia in transgenic Arabidopsis. Plant Physiol 97: 1414–1421
Ubeda-Tomás S, Federici F, Casimiro I, Beemster GT, Bhalerao R, Swarup R, Doerner P, Haseloff J, Bennett MJ (2009) Gibberellin signaling in the endodermis controls Arabidopsis root meristem size. Curr Biol 19: 1194–1199
Weigel D, Ahn JH, Blázquez MA, Borevitz JO, Christensen SK, Fankhauser C, Ferrándiz C, Kardailsky I, Malancharuvil EJ, Neff MM, et al (2000) Activation tagging in Arabidopsis. Plant Physiol 122: 1003–1013
Williams LE, Mills RF (2005) P(1B)-ATPases: an ancient family of transition metal pumps with diverse functions in plants. Trends Plant Sci 10: 491–502
Xiao W, Sheen J, Jang JC (2000) The role of hexokinase in plant sugar signal transduction and growth and development. Plant Mol Biol 44: 451–461
Yamasaki H, Abdel-Ghany SE, Cohu CM, Kobayashi Y, Shikanai T, Pilón M (2007) Regulation of copper homeostasis by micro-RNA in Arabidopsis. J Biol Chem 282: 16369–16378
Yoo SD, Cho YH, Sheen J (2007) Arabidopsis mesophyll protoplasts: a versatile cell system for transient gene expression analysis. Nat Protoc 2: 1565–1572
Yu NI, Lee SA, Lee MH, Heo JO, Chang KS, Lim J (2010) Characterization of SHORT-ROOT function in the Arabidopsis root vascular system. Mol Cells 3: 113–119
Yuan K, Wysocka-Diller J (2006) Phytohormone signalling pathways interact with sugars during seed germination and seedling development. J Exp Bot 5: 3359–3367 | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4405,
1
],
[
4405,
8310,
2
],
[
8310,
11353,
3
],
[
11353,
15911,
4
],
[
15911,
19141,
5
],
[
19141,
23871,
6
],
[
23871,
25787,
7
],
[
25787,
30394,
8
],
[
30394,
37521,
9
],
[
37521,
41686,
10
],
[
41686,
50048,
11
],
[
50048,
51963,
12
]
]
} |
32dc12e25009e6651094a947fffd445a5f154ce1 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 23,
"total-fallback-pages": 0,
"total-input-tokens": 70519,
"total-output-tokens": 27120,
"fieldofstudy": [
"Engineering"
]
} | Energy Solutions for Wearable Sensors: A Review
Guoguang Rong 1,2, Yuqiao Zheng 1,2 and Mohamad Sawan 1,2,*
1 CenBRAIN Lab., School of Engineering, Westlake University, Hangzhou 310024, China; [email protected] (G.R.); [email protected] (Y.Z.)
2 CenBRAIN Lab., Institute for Advanced Study, Westlake Institute for Advanced Study, Hangzhou 310024, China
* Correspondence: [email protected]; Tel.: +86-571-8738-1206
Abstract: Wearable sensors have gained popularity over the years since they offer constant and real-time physiological information about the human body. Wearable sensors have been applied in a variety of ways in clinical settings to monitor health conditions. These technologies require energy sources to carry out their projected functionalities. In this paper, we review the main energy sources used to power wearable sensors. These energy sources include batteries, solar cells, biofuel cells, supercapacitors, thermoelectric generators, piezoelectric and triboelectric generators, and radio frequency (RF) energy harvesters. Additionally, we discuss wireless power transfer and some hybrids of the above technologies. The advantages and drawbacks of each technology are considered along with the system components and attributes that make these devices function effectively. The objective of this review is to inform researchers about the latest developments in this field and present future research opportunities.
Keywords: energy sources; power solutions; wearable devices; biosensors; health monitors; scavengers; harvesters
1. Introduction
Over the years, wearable sensors have attracted considerable attention due to their ability to offer constant and real-time physiological information. Wearable sensors dynamically and noninvasively measure biochemical markers found in biological fluids, including sweat, tears and interstitial fluids. Recently, the non-invasive monitoring of biomarkers through the use of biosensors for both healthcare and sports analytics has become a research hotspot [1]. Researchers have miniaturized composite biosensors, transmission systems and microfluidic sampling platforms so that these techniques can be integrated with flexible materials to enhance wearability and facilitate corresponding operations [2].
Wearable sensors have various clinical uses and are capable of sensing changes in human physiology, biochemistry and motion, which is required for both diagnostics and treatments in sports applications (Figure 1). Additionally, these technologies can play a vital role in treating chronic illnesses by realizing continuous drug monitoring, which is a promising technique intended to replace current therapeutic drug monitoring strategies [3]. Physicians have used wearable sensors in several formats that are built to be attached to the human body to measure bio-signals. Biosensors can be embedded in glasses [4], contact lenses [5], mouth guards [6], clothing [7], wrist bands [8] and many other products. Wearable sensors can measure cardiovascular activity [9], body signals such as heart rate [10], blood pressure and respiration rate [11], and sweat content and interstitial fluid [12], which includes glucose, sodium and potassium levels. There have also been studies of the development of wearable electroencephalography (EEG) devices and using deep neural networks for EEG data analysis [13]. Wearable technologies based on multimodal EEG-functional near-infrared spectroscopy (EEG-fNIRS) have been used to monitor physiological parameters related to strokes, which can help prevent stroke-related disability or death [14].
Figure 1. Wearable medical and healthcare devices for various regions of the body.
Wearable sensors have strong potential for improving current medical systems, as they are able to provide extensive monitoring of various physiological functions and highly efficient economical services in a variety of fields. However, wearable sensors need appropriate energy sources, given that their maintenance depends on a continuous, stable and sufficient supply of power. Different forms and applications of wearable sensors require different amounts of energy to work. Therefore, it is of great importance to design and select reliable energy supply systems with suitable structures, sizes and energy densities. Wearable sensors generally use batteries as a power source. Battery technology is very mature, and various power management strategies for prolonging battery life have been proposed [15]. However, the use of batteries to power wearable sensors is associated with risks such as leakage of toxic substances and limited device lifetimes. An alternative solution involves wearable sensors harvesting energy from the human body or from the ambient environment. There are a number of technologies that can be used to harvest energy, including photovoltaic (PV) or solar cells, thermoelectric generators (TEGs), piezoelectric nanogenerators (PENGs), triboelectric nanogenerators (TENGs), biofuel cells (BFCs), electromagnetic generators (EMGs) and radio frequency (RF) harvesters. These technologies can be used in either single or hybrid formats [16]. When applied to wearable sensors, all of these approaches need to be well designed in terms of their structure, size, power density and energy output according to the application of interest. At the same time, these power generators need to conform well to the human body and be flexible, durable, stable, and non-toxic.
The remainder of this paper is composed of four sections. Section 2 discusses the methods most widely used to power biosensors, including batteries, solar cells, TEGs, BFCs and nanogenerators. Section 3 introduces three hybrid-generator systems and discusses their properties and performance. Section 4 compares the different types of generator systems and their performance parameters, and Section 5 draws conclusions with regard to the main research fields for which power generators are of interest and offers future directions of study. Table 1 summarizes the parameters used in this article.
2. Power Supply Solutions for Wearable Sensors
Available biosensors and wearable sensors under development require various power levels to be delivered from different sources of energy. Figure 2 shows some of the sources of energy that can be harvested, converted, and used to supply power to wearable sensors. Energy collected from light, heat, radio waves, or vibrations can be converted into electrical energy to power wearable sensors. The main energy solutions for wearable sensors are reviewed in the following sections.
Table 1. Parameters in this article.
| Symbol | Meaning | Symbol | Meaning |
|--------|------------------|--------|----------------------|
| $V_{oc}$ | Open circuit voltage | $P_o$ | Output power density |
| $V_o$ | Output voltage | $R_L$ | Load resistance |
| $J_{SC}$ | Short circuit current | $\Delta T$ | Temperature difference |
| $P_o$ | Output power | $T$ | Temperature |
2.1. Batteries
Wearable technologies, similar to any embedded system or portable device, require batteries to operate, and consumers would like smaller and thinner gadgets with batteries that can last for an extended period of time. Over the years, a number of different kinds of batteries have been introduced with the intention of meeting the power requirements of various wearable sensors.
Alkaline batteries, available since the 1960s, have been tested and widely used. These batteries are safe and easily replaceable. The main types of alkaline batteries are referred to as AA and AAA. These batteries are also available in coin cells and have a 1.5 V standard voltage and an 11.6 mm diameter and 5.4 mm height [17]. However, alkaline battery voltages drop with continued use; a steady and stable voltage is not achievable, and...
the voltage drops sharply near the end of the battery’s life. Nickel-metal hybrids are an alternative class of batteries that can be used to power wearable sensors. These types of batteries are rechargeable and have a capacity that is three times that of nickel-sodium batteries, with a value approaching that of lithium-ion batteries [15]. Lithium-ion and lithium-ion polymer batteries are the most commonly used batteries for wearable sensors. These batteries use Li-polymer cells or Li coin cells to make the batteries rechargeable and sustain power for an extended period of time to secure the normal operation of the wearable sensor [18].
Table 2 summaries typical batteries that can be used for wearable sensors. From this table we understand that lithium-ion batteries have a high working voltage, high specific energy, high specific power, long life cycle and very low self-discharge rate (2% per month) compared to other kinds of batteries. Table 3 compares some low-modulus and flexible substrate wearable sensors powered by different batteries. These sensors can be used to detect molecular or physical parameters such as pressure. In general, batteries can simplify device structure and size, power biosensors directly without the need for complex circuits [19].
Table 2. Comparison of different types of batteries.
| Battery | Working Voltage [V] | Specific Energy | Specific Power | Life Cycle |
|---------------|---------------------|----------------|---------------|------------|
| | | Gravimetric [W \cdot h/kg] | Volumetric [W \cdot h/L] | |
| Li-ion | 3.7 | ~200 | ~550 | High |
| Ni-MH | 1.25–1.10 | ~100 | <500 | Low |
| Ni-CD | 1.25–1.00 | <40 | <100 | High |
| Ni zinc | 1.60–1.40 | ~100 | <400 | High |
| Alkaline | 1.3–1.0 | 100 | <300 | Medium |
Li-ion: Lithium-ion; Ni-MH: Nickel metal hydrogen; Ni-CD: Nickel chromium; Ni zinc: Nickel zinc.
Table 3. Typical low-modulus and flexible substrate wearable sensors powered by batteries.
| Battery Chemistry | Voltage (V) | Capacity (Mah) | Sensor’s Target | Ref. |
|-------------------|-------------|----------------|-----------------|-----|
| Li-ion polymer battery | 3.7 | 150 | Glucose, lactate, pH, temperature | [19]|
| Li-ion polymer battery | 3.7 | 105 | Glucose, lactate | [20]|
| Thin film Li-ion battery | N/A | N/A | Lactate | [21]|
| Li-ion battery | 3.0 | 100 | Lactate, potassium | [4] |
| Li-Po battery | N/A | N/A | OP nerve-agent compounds | [22]|
| Li coin battery | 3.6 | 180 | Lactate, Li ions | [23]|
| Button cell battery (Type: N/A) | 1.5, 3.0 | N/A | H2O2 | [24]|
| Seat-type cell battery (Type: N/A) | N/A | 25 | Pressure | [25]|
Ref: Reference; OP: organophosphate; Li: lithium; Po: polonium.
Regardless of the type of battery used to supply energy to wearable sensors, whether it is a lithium-ion or alkaline battery, various designs have been implemented to ensure batteries fit user needs with regard to keeping wearable sensors safely powered [26]. However, the high toxicity of battery electrolytes is a disadvantage that cannot be ignored. On the other hand, wearable sensors have large limitations in size. Thus, in order to minimize device size, large and rigid battery packs are not expected. For some applications, batteries eventually may need to be replaced if they are of low capacity, and will run out of energy sooner or later in real applications. Devices powered by batteries require frequent charging, which can bring inconvenience. Therefore, except for enhancing the battery capacity, energy solutions that can provide an alternative power supply are being developed, such as energy harvesting technologies. On the other hand, although many studies are searching for alternatives to batteries, batteries and supercapacitors can also
be combined in an energy harvesting system as the energy storage section. Therefore, optimizing battery performance inside wearable sensors is still a topic worth studying.
2.2. Solar Cells
Substantial attempts have been made by researchers to replace batteries as an energy source for numerous applications. Solar cells have been introduced to replace batteries to make devices lighter and more efficient [27]. Solar energy has various benefits in maintaining wearable device functionality. Solar cells can charge wearable devices through a USB connection integrated into a piece of cloth, which is very convenient in constructing different modules of devices via a standard interface, making devices more user-friendly and reducing their power demand. Since the introduction of miniaturized solar cells, it has become possible to generate power by solar energy in a variety of wearable objects. Thus, the use of wearable sensors has become more convenient than ever before [28].
Over the years, solar cells with reduced lengths and diameters have been synthesized; for example, the most recently developed solar cells are 3 mm in length and 1.5 mm in diameter, making them lighter and more suitable as a power source for wearable devices [29]. Solar-powered wearable devices are similar to any other textile but contain fibers with miniaturized cells that create electricity. This kind of flexible solar cells have been proved to be of high flexibility and durability. Table 4 provides some typical examples of these flexible textile solar cells. It has been found that 200 miniature solar cells embedded in a 20 cm$^2$ section of fabric can yield a power of up to 43.5 milliwatts, which is sufficient to power a mobile phone [30].
Table 4. Typical examples of flexible solar cells.
| Device Structure | Fill Factor | $V_{oc}$ [V] | $J_{SC}$ [mA·cm$^{-2}$] | PCE | Ref. |
|------------------|-------------|--------------|--------------------------|-----|------|
| FTO | co-doped TiO$_2$ | perovskite | spiro-OMeTAD | Au | 0.701 | 1.18 | 23.34 | 19.44% | [31] |
| MgF2 | Willow Glass | TCO | SnO$_2$ | FAMACs | Spiro-MeOTAD | MoO$_3$ | Al | 0.752 | 1.084 V | 22.16 | 18.1% | [32] |
| PET | IZO | PEDOT: PSS | perovskite | C$_{60}$ | BCP | Ag | 0.6011 ± 0.04 | 0.96 ± 0.04 | 18.05 ± 0.68 | 10.39 ± 0.41% | [33] |
| ITO | F-TiO$_2$ | KCsFAMA perovskite | spiro-OMeTAD | Au | 0.7703 ± 0.019 | 1.17 ± 0.02 | 22.90 ± 0.61 | 20.66 ± 0.97% | [34] |
| ITO | PEDOT: PSS | perovskite | PCBM | Ag | 0.745 | 0.75 | 27.8 | 15.59% | [35] |
| Glass | ITO | SnO$_2$ | MaPbI$_3$ | Carbon | PEDOT: PSS (PH1000) | | 0.61 | 0.95 | 22.94 | 13.08% | [36] |
| PEDOT: PSS (F VP AH4083) | PM6: Y6 | PBN-Br | Al | 0.7729 ± 0.006 | 0.840 ± 0.010 | 23.92 ± 0.15 | 14.66 ± 0.24% | [37] |
PCE: power conversion efficiency; Ref: Reference; FTO: Fluorine-doped tin oxide; TiO$_2$: titania; spiro-OMeTAD: 2,2',7,7'-Tetrakis[N,N-di(4-methoxyphenyl)amino]-9,9'-spirobi-fluorene; TCO: transparent conductive oxide; Al: aluminum; PRT: poly(ethylene terephthalate); IZO: indium zinc oxide; PEDOT:PSS: poly(3,4- ethylenedioxythiophene) polystyrene sulfonate; BCP: bathocuproine; Ag: silver; ITO: indium tin oxide; F-TiO$_2$: TiO$_2$ nanocrystals; Gr: graphene; GO: graphene oxide; KCsFAMA: K$_{0.025}$Cs$_{0.05}$FA$_{0.95}$MA$_{0.12}$Pb$_{0.55}$Br$_{0.45}$; PCBM: [6,6]-Phenyl-C$_{61}$butyric acid methyl ester; PEN: polyethylene naphthalate; Ag-NW: silver nanowire; Ti: titanium; Pt: platinum; Cu: copper.
Solar energy can also help optimize the performance of energy storage modules and thus can be more stable and adaptable for wearable electronics. Researchers used graphene oxide as solar-thermal conversion section to improve the low-temperature resilience of a supercapacitor [38]. In addition to high flexibility and scalability, this new design of a fiber-based supercapacitor can absorb and make use of sunlight energy to guarantee operation under extreme low temperatures (0 °C), and thus is a promising choice of a combined energy harvesting and storage device for wearable and portable biosensors.
Perovskite solar cells are a class of solar cells that offer an efficient, flexible and lightweight energy solution for wearable electronic sensors. These solar cells are thin, flexible and very popular in, for example, portable electric chargers and wearable electronic
Flexible perovskite solar cells with a novel light absorber have been developed. An efficiency of 10.2% has been achieved by introducing a 3 µm-thick layer of this absorber. In other studies, research groups have succeeded in fabricating perovskite solar cells at a low temperature (130 °C) using another kind of light absorber [40]. These solar cells exhibit a higher power conversion efficiency (PCE) (15%) than those of optimized single-junction flexible organic solar cells. The most recent applications of perovskite solar cells based on organic light-absorbing halides have been shown to be economically and practically viable [41]. These solar cells are fabricated at low temperatures and can achieve PCEs above 16%. Researchers also studied bendable solar cells that can be applied to highly diffuse light environments. Jaehyun Park and his group build an analytical model in quantifying the relationship between harvested energy and the radius of cell curvature. The energy collected is increased by up to 25.0% under their proposed model [42]. Another work aimed at applying bending silicon photovoltaic cells for lighting poles. The authors studied the influence of illumination type (direct or diffuse irradiance), geometry condition (flat, half-bent or full-bent) and dye concentration on cell energy-harvesting performance [43]. The results show that the bent photovoltaic cells are less likely to be influenced by incoming irradiance angle compared with flat ones, and they also show stable performance under bad weather conditions. However, surface loss for the bending group was larger than that of the flat group. Although the authors only mentioned the application of their bent solar cell in architecture, it is still very promising for wearable sensors.
Table 4 summarizes the parameters of some high-PCE solar cells with high durability and stability. High flexibility can be significant in reducing wearing inconvenience, thus can benefit the commercialization of wearable sensors and their power supplying systems. To realize required flexibility, cell substrate and electrodes should be carefully designed. These solar cells can realize competitive PCE and output voltage levels for many portable electronic devices. They are suitable for wearable sensors without complicating the whole system. Note that all the experiments listed in Table 4 are done under illumination conditions of AM 1.5 G (100 mW⋅cm⁻²).
Besides flexibility and stretchability, another important factor that can greatly affect the performance of solar cells is light intensity. Solar cells need further optimization to enhance their indoor performance, as the weather is not controllable and users spend most of their time inside buildings. It is worth mentioning that presently, many studies have been devoted to improving the indoor performance of solar cells [44,45], known as indoor photovoltaics. One work studying a photovoltaic-thermoelectric hybrid energy generator aims at achieving satisfactory self-powering ability in indoor environments when used in wearable devices [46]. Lee et al. tested the indoor performance of a perovskite photovoltaic cell under 200–1000 lux of illumination conditions. This cell realizes 18 µW⋅cm⁻² power density under 200 lux cool white light, and the fill factor kept stable among the light intensity range [47]. Biswas et al. enhanced the PCE of indoor organic solar cell by studying the doping concentration of the cell’s hole extraction layer. By testing under a 500 lx LED light, this cell realizes 8.1% PCE level [48]. Park and his group also proposed an indoor organic solar cell with very high PCE of up to 42% under indoor light conditions. After 1500 cycles of bending tests, this organic solar cell can maintain 84% of initial PCE [49]. All these works are aimed at improving the electrical performance of solar cells under low light environments and have made great progress.
Although solar cells exhibit higher PCEs than other types of energy generators [16], their efficiencies can be greatly reduced with increasing temperature [50]. The temperature coefficient of PV cells depends greatly on the type of material used [51]. Additionally, because some textile-based solar cells may suffer severe reductions in PCE following fatigue [52], the mechanical properties of solar cells should also be seriously taken into account. For example, when applied to wearable devices, these flexible devices face more complex body motions compared with experimental conditions. How to avoid material failure and performance degradation has become a problem that cannot be ignored. On
the other hand, the temperature of solar cells can exceed the temperature of the human body significantly. For example, one experiment conducted by a research group tested the performance of organic solar cells under temperatures up to 85 °C [53]; thus, cooling systems may be needed, and methods that utilize wasted heat should be considered.
2.3. Thermoelectric Generators
TEGs are another type of energy source that can be used to power wearable devices. TEGs last for a long time, produce no noise, and generate the energy needed to directly power wearable devices by converting heat into electrical energy. TEGs have been shown to be effective in powering wearable sensors by using waste heat from the human body. The human body is a great portable source of energy, generating up to 58.2 W/m² in waste heat at rest [54]. Even a small percentage of this waste heat is sufficient to power most low-energy wearable devices without the need for batteries as a backup energy source. Advancements in TEGs have allowed for more efficient power conversion. Thinner devices can reduce the profile and complexity of devices and remove the need for regular changes or swaps of energy cells. Additionally, TEGs can be directly integrated with wearable textiles [7]. Wearable sensors that operate with body heat can be used for long-term monitoring of human vitals and chronic illnesses. Human body heat-powered sensors have been used to monitor glucose levels and have been applied in hearing aids and accelerometer-based rehabilitation devices [55].
Thermoelectric energy has some advantages over other wearable sensor power sources. For example, the conversion of mechanical energy into electrical energy requires the user to be active, which may not be possible for elderly or bed-ridden individuals. Additionally, when users are exposed to low-lighting conditions, solar cells may not function properly. On the other hand, TEGs yield constant power as long as a difference between the temperature of the skin and the ambient temperature exists, which is typical for most practical conditions [56].
TEGs utilize the Seebeck effect to convert heat into electrical energy. When dissimilar materials, for instance, metals or semiconductors, which are n-type and p-type components, have junctions at different temperatures, the carriers of electrons and holes will move to the cold ends [57]. The Seebeck effect results in the creation of electric fields in both materials that are proportional to the temperature gradient. If there is a circuit connection, the current is able to flow [58]. TEGs are constructed by introducing a heat sink between the n- and p-type semiconductors and the heat source.
People generate heat as a result of metabolic functions, which serve to maintain core body temperature. The body temperature required for humans to maintain standard functionalities is approximately 37 °C. Heat exiting the skin is transferred to the ambient environment through convection and radiation at 1~10 mW cm⁻². The rate at which heat is transferred depends on which body part is being considered. For example, muscles act as insulators, while arteries exhibit the highest heat-transfer efficiency of any body part. Clothing can obstruct heat transfer, leading to an average body heat transfer rate of approximately 5 mW cm⁻² [59]. For accurate heat measurements, a body part such as the radial artery in the wrist may be targeted since it has a heat flow amounting to approximately 25 mW cm⁻² at room temperature [59].
Electric generators that use body heat as an energy source may experience some shortcomings as a result of an insufficient conversion of body heat to electricity. For efficient conversion, there is a need for thin TEGs that do not consume much energy. Additionally, in some applications, there is a need for flexible wristbands with TEG modules that capture sufficient accelerometer data from a user [59].
Furthermore, industry has introduced flexible and organic TEGs with roll-by-roll methods. This is an efficient manufacturing process utilizing the rolling of different layers, which has led to the increased availability and reduced cost of TEGs. Better design in structure can improve the flexibility and durability of TEGs. Researchers have proposed highly flexible thermoelectric (TE) devices that can be integrated on arm bands. The
output power density can reach 5.60 µW·cm⁻² by harvesting human body heat energy [60]. Table 5 lists some examples of flexible TEGs that function as a result of temperature differences between the human body and the environment. Usually, the temperature difference between the hot and cold side for wearable sensor should not be too large for general applications. Sun et al. prepared a fiber-based TEG that can be directly woven into textiles [61]. Additionally, a TEG-powered bracelet that can simultaneously monitor the temperature, humidity and motion of the human body has been reported [62]. Furthermore, the voltage increases linearly with increasing temperature difference $\Delta T$, and the output power is positively correlated with $\Delta T$ [63].
Table 5. Comparison of different flexible TEGs.
| TE Material | Temperature | $P_o$ (or $\rho_o$) | Ref. |
|--------------------------------------|-------------|---------------------|------|
| n-type Ag₂Se | $T_{cold} = RT$ | 6.6 µW cm⁻² | [64] |
| Ag-modified Bi₀.₅Sb₁.₅Te₃ | $T_{cold} = RT$ | 12.4 µW·cm⁻¹K⁻² at 300 K | [65] |
| Bi₀.₅Sb₁.₅Te₃ and Bi₂Sb₀.₃Te₂.₇ | $\Delta T = 5 \sim 35$ K | 0.1-10 nW | [63] |
| Bi₂Te₃ grains | $T_{hot} = 307$ K, $T_{air} = 293$ K | 153 µW·cm⁻² | [62] |
| (D-A)-type polymer/few-walled CNTs | $\Delta T = 20$ K | 210 nW | [66] |
Ref: Reference; TE: thermoelectric; CNT: carbon nanotube; PI: polyimide; PF: power factor; PDMS: polydimethylsiloxane; PVDF: polyvinylidene fluoride; D-A: Donner-acceptor.
2.4. Radio Frequency Energy Harvesters and Wireless Power Transfer (WPT)
RF harvesters, which use wireless power, offer an energy solution for wireless sensor networks [67]. RF technology harvests energy from the surrounding or dedicated energy sources. RF harvesters are attractive for many applications. Unlike other energy harvesters, such as solar cells or chemical generators, RF harvesters offer a continuous and controllable source of power, which makes them desirable for applications that require higher levels of energy [68]. Additionally, RF signals have been used to carry wireless information in wireless communications and can be utilized for sensor data transmission.
Depending on the WPT range, RF power transfer can be categorized as near-field inductive and capacitive coupling power transfer, ultrasonic power transfer, and mid- or far-field electromagnetic power transfer. Near-field inductive coupling is relatively mature and has been used to power cochlear implants [69]; however, careful alignments are needed to generate high power. Near-field capacitive coupling is used to power flexible patches, but the associated power transfer efficiency (PTE) drops dramatically when the transmitting coil and receiving coil are separated. Ultrasonic energy transfer is limited by large PTE fluctuations. Mid-field and far-field RF harvesters are less efficient when the frequency exceeds tens of gigahertz (GHz). Table 6 summarizes RF harvesters used to generate power for wearable sensors. Their PTE ranges from 5% to 90%, dependent on input power.
Over the past few years, WPT and data transfer have been investigated. Since these transfers are both enabled by RF signaling, many scholars have investigated simultaneous wireless data transfer and power transfer, which combines the two techniques [70].
There are also some problems in applying RF harvesters to power wearable sensors. For example, RF energy harvesting cannot generate enough power without perfect alignment in nearfield systems. Far-field RF harvester systems have low PTE inherently and may not satisfy the power requirements of wearable devices. One of the main concerns of RF harvesters is their capability to provide enough energy. On the other hand, RF harvesters need wake-up power, given that Complementary Metal-Oxide-Semiconductor (CMOS) transistor has a threshold. Although RF energy seems to act as a desirable and reliable source of energy for wearable devices, there is still room for constant advancement.
Table 6. Comparison of typical RF harvesters used for wearable devices.
| Flexibility | Frequency | P_o (or p_o) | PTE | Ref. |
|-------------|-----------------|-------------------------------|----------------------------|--------|
| Flexible | 2.45 GHz | NA | NA | [71] |
| Non-flexible| 2.45 GHz | 1–10 µW · cm^{-2} | 5.9–27.7% | [72] |
| Flexible | 2.45 GHz | 600 µW at 10 cm from source | 91% | [73] |
| Flexible | 2.45 GHz | 80 µW at 60 cm from source | | |
| Flexible | 915 MHz & 1.85 GHz| NA | 43.2% at -18 dBm input power | [74] |
| Flexible | 5.2 GHz | NA | 67% at +20 dBm input power | [75] |
| Flexible | 0.868 GHz | NA | 65.8% at +6 dBm input power | [76] |
PTE: power transfer efficiency.
2.5. Biofuel Cells
Biofuel cells (BFCs) are another energy recovery method used for wearable sensors. A fuel cell refers to an electrochemical cell wherein current is generated by reactions occurring between the chemical species flowing into the cell at the anodic site and the oxidant at the cathodic site [77]. Fuel cells are different from standard batteries, as they can produce continuous energy as long as the reactants are present. While there are various fuel cell forms, the most commonly used fuel cell involves a proton-exchange membrane [78]. In this type of fuel cell, a membrane separates the fuel and oxidant, allowing only protons produced at the anodic site to cross the membrane and minimizing the amount of oxidant present at the cathodic site. Electrons generated at the anode cannot pass through the membrane to reach the cathode; as such, they have to follow an alternative path, which results in the generation of current. The use of fuel cells to power wearable sensors has a number of advantages. The most important advantage is that the presence of reactants inside the fuel cell makes it unnecessary to replace the batteries [79]. In addition, elderly or bed-ridden individuals may utilize wearable sensors powered by BFCs using reactants available in human body, such as glucose or lactic acid.
When BFCs are used to power wearable sensors, the power supply can be combined with biosensing to simplify the design. Epidermal BFCs have been used to oxidize lactic acid in sweat to generate energy [80]. The energy generated from sweat helps the biofuels create ten times more energy per unit area than any other biofuel used in wearable sensors [81]. Jia et al. [82] prepared a BFC with bridge and island structures that was flexible, stretchable, and compatible with wearable sensors. This BFC contained two dotted rows linked by spring-shaped structures. Half of the dots comprised the anode, and the other half comprised the cathode. In addition, the spring-like structures between the dots stretched and bent without breaking or changing the initial structure of either the cathode or anode. The island and bridge structures were prepared from gold using lithography [83]. Then, researchers used a screen-printing strategy to place biofuel layers on top of the anode and cathode dots. Even though the approach was proven effective in producing energy, it was difficult to identify the amount of energy that these fuel cells could create per unit area. Thus, there is a need to quantify the amount of material to use and the combination ratio of different materials since these factors determine the amount of power generated [84]. Researchers are also trying to enhance the flexibility of biofuel cells. Shitanda et al. proposed a lactate paper-based biofuel cell [85] that can be very promising in wearable applications. It can provide ~3.4 V open circuit voltage (OCV) with six cells in series and 4.3 mW output power with a 6 × 6 cell array. Yin and his group demonstrated a flexible bracelet BFC that can collect sweat and utilize lactate to power wearable devices. It can generate 74 µW maximum output power and 0.39 V OCV at 20 mM lactate solution [86].
The biofuel cell array is proved to be of high performance in powering devices with low power cost. Table 7 lists some examples of BFCs that are or can be used to power wearable sensors. In general, BFCs are highly biocompatible, with lower fabrication costs. They make use of biochemical substances to supply power for wearable sensors. Recently, more BFCs with a high flexibility and small size have been investigated. However, BFCs
require further studies to enhance their PCE. On the other hand, their output power relies heavily on analyte concentration.
Table 7. Comparison of different BFCs that can be used to power wearable sensors.
| Target | Sensitivity/LOD | $P_o$ (or $p_o$) | $V_{oc}$ or $V_o$ | Ref. |
|-------------------------------|-----------------|------------------|------------------|-------|
| D-fructose | 3.82 ± 0.01 mW·cm$^{-2}$·mM$^{-1}$ | Pulse mode: 17.6 mW·cm$^{-2}$ | NA | [87] |
| Exosomes in cancer cells | 300 particles·mM$^{-1}$ | 619 µW·cm$^{-2}$ | ($V_{oc}$) 0.46 V | [88] |
| Glucose | 64.97 µA·cm$^{-2}$·mM$^{-1}$ | 1011.21 µW·cm$^{-2}$ | NA | [89] |
| Glucose | NA | 2.24 mW·cm$^{-2}$ | ($V_o$) ~0.3 V | [90] |
| Glucose | 0.1 mM | 31.3 µW·cm$^{-2}$ | ($V_{oc}$) ~0.65 V | [91] |
LOD: limit of detection; Ref: Reference; RNA: ribonucleic acid.
2.6. Kinetic Energy Harvesters
Energy harvesters (EHs) are another type of energy source used to power wearable sensors. EHs function by trapping and accumulating vibrational energy produced by either human body movements or natural phenomena. EHs are considered green because they are biocompatible and environmentally friendly [92]. EHs provide low voltages and are suitable for applications requiring low power.
EH technology uses various techniques. The most important technique involves kinetic energy harvesting, where human motion is converted into energy. Kinetic energy harvesters can make use of vibration or motion to generate electrical power. They are classified by their transduction mechanisms: electromagnetic, electrostatic, piezoelectric and triboelectric [93].
2.6.1. Electromagnetic Kinetic Energy Harvesters
An electromagnetic kinetic energy harvester contains electromagnetic transducers that can generate an electromotive force in response to changes in the external magnetic flux of a closed-loop circuit, by which electrical power is generated. Influx fluctuations can also be induced by, for example, making a circuit rotate around an axis, which changes the surface aligned with the magnetic flux [94]. Seiko has used this approach in its quartz wristwatch, which can self-charge through wrist motion due to energy transfer from an oscillating weight to a magnetic rotor attached to the watch’s coil. Kinetic harvesters can also have a charge pump circuit with various multiplicative factors to increase the battery voltage [95]. Wang et al. proposed an electromagnetic kinetic energy harvester that can convert walking, running and jumping mechanical energy to electrical power [96]. It is sensitive to vibrations with a frequency less than 100 Hz and is very small in size (10.58 × 2.06 × 2.55 mm$^3$). This generator can provide 98.3 µW·cm$^{-3}$·g$^{-1}$ normalized power density and 43.7 µ average output power.
2.6.2. Electrostatic Kinetic Energy Harvesters
Electrostatic energy harvesters generate electrical power based on electrostatic induction. They contain a variable capacitor composed of two electrodes. External vibration disturbance can change the capacitor’s overlapping area, which leads to capacitance change. The devices resonate in response to this vibration disturbance and generate electricity [97]. The output power level is proportional to the harvesters’ operating frequency. Therefore, to gain maximum electrical power, the harvesters are expected to work at resonance.
An electrostatic kinetic energy harvester is able to greatly reduce the size of kinetic energy harvesters, which is very competitive in wearable devices. Microelectromechanical System (MEMS) electrostatic energy harvesters produce capacitance variation by mechanical vibrations and are typically designed in comb format [98]. MEMS electrostatic energy harvesters consist of a central mass and attached parallel electrostatic transducers. Hassana et al. designed a MEMS kinetic energy harvester generating an average output power of up to 195 nW at only 6 × 7 mm$^2$ in size [99]. Lu’s group demonstrated a model that can
predict the frequency performance of a MEMS kinetic energy harvester and concluded that bias voltage and stopper position can be designed to control the harvester power performance [100]. In addition to theoretical models, the author proposed a paper-based electrostatic kinetic energy harvester with a thickness less than 1 mm. It can provide 45.6 $\mu$W maximum output power with 16 M$\Omega$ load resistance [101]. In general, with further study and experiments, electrostatic kinetic energy harvesters with soft substrate are very promising in wearable sensors due to their small device size.
### 2.6.3. Piezoelectric Nanogenerators
PENGs harvest energy by converting kinetic energy into electrical energy through the actions of nanostructured piezoelectric materials. A substantial amount of mechanical energy can be harvested from a variety of biological functions, including fluid flow, walking, heartbeats, breathing and muscle movements [102]. As such, PENGs, which convert mechanical stress into electrical charges, are the most promising energy harvesters for microsystems.
PENGs can be attached under shoes to generate power from leg motion [103]. In addition, a highly sensitive PENG can generate an output voltage near 11 mV from the vibration of a human throat when speaking [103]. PENGs have also been used to generate power for artificial skin, which can monitor the health condition of individuals [104].
The electric energy supplied by PENGs can be used to energize systems with power consumption requirements ranging from microwatts to milliwatts, which is a suitable range for wearable sensors. Experiments have shown that a 3D-composite PENG can generate an output voltage near 65 V and an output current near 75 nA when 15% stretching stress is applied. This PENG is highly flexible and can generate electricity when pressed, stretched or bent [105]. Although PENGs have not been widely used for wearable sensors, this technology has ample potential for future breakthroughs and has been applied to further miniaturize conventional energy harvesters [106]. Additionally, this technology can potentially be integrated with other types of energy harvesting mechanisms [107].
Furthermore, piezoelectricity can be harvested from the human body irrespective of time or location [108]. The design of garments plays a vital role in energy collection, since clothing covers the human body as it moves [109], and an optimal design is needed to achieve mechanical deformation for efficient energy harvesting. The efficiency of piezoelectric energy harvesting depends on the properties and design of the textile being considered [110]. However, in addition to the garment texture, human factors such as the frequency of movement or the type of body part moving also play significant roles in the efficiency of piezoelectric energy harvesting [111].
PENGs have significant advantages over other power generation devices; for example, they are generally flexible and thus can be used for a broad range of applications. Table 8 lists some examples of PENGs that have been used to power wearable sensors. In real applications, the system supplying power should not be too large or complex in structure. In tradition, PENGs produce an AC voltage; as such, additional rectifiers are needed to convert the AC voltage to DC voltage. At the same time, human motion is typically of low frequency and not predictable; therefore, frequency up-conversion devices are also needed. These extra modules can lead to bulky and less-flexible generators. Recently, non-resonant piezoelectric energy harvesting techniques have been proposed to offset these defects. Bassani et al. proposed non-resonant macro-fiber composites in harvesting mechanical energy from human joint movements [112]. Benefiting from the netlike electrodes that can introduce a transverse mode inside fibers, P2 type macro-fiber composites used in this work are more suitable for energy harvesters compared to P1 types. These P2 type devices can also gain higher capacitance and generate more charge under the same strain conditions. As a result, both periodic and non-periodic movement can be made use of to generate electricity because the harvester’s power output depends on the bending velocity. On the other hand, walking with specific frequency is not necessary, and any motion that involves knee-bending can be used to gain energy. Another work demonstrated by Tunce.
et al. built a theoretical model, and performed experiments to assess the amount of energy their harvester scavenged from human body joint motions [113]. The authors applied two kinds of macro-fiber composite patches and performed 16 groups of experiment to validate their theoretical model. The device can tolerate a large degree of bending (60° and 90° were tested) and can generate 13 μW average energy level for walking motions when attached to both knees. Cha and his group developed piezoelectric energy harvester to scavenge mechanical energy from finger’s clicking mouse [114]. The device is sealed onto gloves and tested by a robot finger under both one-click and double-click of a mouse. The maximum energy gained is in 1~10 nJ for 30–70 MΩ load resistance. In summary, it is promising to study further how to minimize the influence of unpredictable human motion and how to dispense the use of AC-DC converters. Further study is expected of the relationship between internal impedance, motion frequency, motion intensity and electric power.
Table 8. Comparison of different PENGs.
| Form | Material | Periodical Pressure: | $R_t$ | $P_o$ (or $p_o$) | Ref. |
|-----------------------|---------------------------------|----------------------|-------|-----------------|------|
| Flexible film | MASnBr$_3$-PDMS | 0.5 MPa/5 Hz | 6 MΩ | ~75.52 μW · cm$^{-2}$ | [115]|
| Fibers | PTrEE-PVDF | NA | NA | 1.35 μW · cm$^{-3}$ | [116]|
| 3D composite foam | (Sm-PMN-PT)-PDMS | NA | NA | 11.5 μW · cm$^{-2}$ | [117]|
| Flexible film | MASnl$_3$-PDMS | 0.5 MPa/1 Hz | NA | 21.6 μW · cm$^{-2}$ | [118]|
| Flexible composite | FAPnBr$_3$ NPs/PVDF | 0.5 MPa/5 Hz | 200 kΩ| 27.4 μW · cm$^{-2}$ | [119]|
| Flexible thin film | AIN buffer layer/Al$_x$ interlayer/top GaN layer | NA | 5 MΩ | 167 μW | [120]|
Ref: Reference; MASnBr$_3$: methylammonium tin bromide; SM-PMN-PT: samarium-doped Pb (Mg$_{1/3}$Nb$_{2/3}$) O$_3$-PbTiO$_3$; PDMS: polydimethylsiloxane; Pb: lead; PTrEE: polyrifluoroethylene; PVDF: polyvinylidene fluoride; MASnl$_3$: methylammonium tin iodide; NPs: nanoparticles; AIN: aluminum nitride; GaN: gallium nitride.
2.6.4. Triboelectric Nanogenerators
Different from electrostatic energy harvesters, TENGs generate surface static charge by contact electrification. When two materials with different electron affinities contact each other, there will be opposite electrostatic charges on the joint surface. Mechanical disturbance can build electric potential difference, which will then establish electric current through the contact surface. Based on this, TENGs can have broad application fields, given that we generate large amounts of wasted mechanical energy in daily activities. TENGs are highly flexible and stretchable. They can be attached to the elbow (armbands), legs (kneepads), wrists, fingers or feet (shoes) to harvest waste mechanical energy from human motion or the ambient environment. The combination of electrical textiles, TENGs and clothing can be promising in commercial applications. Table 9 provides some examples of TENG-powered biosensors.
TENGs have been used to power biosensor systems, wearable devices and electronic skin. Single fiber TENGs can provide an open circuit voltage of 140 V and a short circuit current density of 0.18 μA/cm. With an optimum load resistance of 320 MΩ, these TENGs can provide a power of 5.5 μW [121]. Non-texturized triboelectric devices can be worn on the skin to monitor electrophysiological signals, body temperature and hydration levels. Nanotexturized triboelectric devices can also measure electrophysiological signals and simultaneously convert imperceptible time-variant motions of the body into electrical signals. This process enables the self-powered monitoring of respiration, swallowing and arterial pulses [122]. Recently, fire-resistant and self-extinguishing TENGs with a stable electrical output under 200 °C have been proposed for special users such as firefighters [123]. These TENGs can be integrated into gloves or shoes and generate power by users walking, running or falling down.
**Table 9. Comparison of various TENGs.**
| Form | Sensor Application | $R_L$ | $P_o$ (or $p_o$) | Ref. |
|-----------------------------|--------------------------|---------|-------------------|--------|
| All-nanofiber-based TENG | Human movement monitoring| 4610 MΩ | 48.6 µW · m$^{-2}$| [124] |
| Flexible chip | Detecting NH$_3$ ammonia at RT | 46.2 MΩ | 10.84 W · m$^{-2}$| [125] |
| Electronic skin | Tactile sensing | 140 MΩ | 2.9 µW · cm$^{-2}$| [126] |
Ref: Reference; NH$_3$ ammonia; RT: room temperature.
AC voltage generated by TENGs needs to be converted into DC voltage. The extra circuit will increase the complexity of wearable and portable devices. To be applied to clinical medicine, they are expected to be sensitive enough to utilize tiny mechanical motions such as arterial pulsation.
### 3. Hybrid Energy Solutions
The combination of different energy sources allows for the generation of a larger amount of power and therefore increases the PCE of the energy system. A typical hybrid energy harvesting system consists of the energy harvester module, energy storage system and wearable sensor module. The remainder of this section introduces the currently most-popular hybrid systems.
#### 3.1. Combination of Solar and Thermoelectric Energy Sources
Solar energy sources have a higher spectrum coverage for higher energy conversion. However, solar cells do not make full use of the photons outside of their band gap energy. Photons outside of the solar cell’s band gap energy are converted into waste heat.
Figure 3 shows a diagram of a conceptual photovoltaic and thermoelectric (PV-TE) system. As mentioned above, an increase in temperature will lead to a decrease in the PCE. Therefore, the combination of TE and PV energy systems could help broaden the use of these technologies and increase the total output power [127]. The two technologies can be combined effectively if there is a significant temperature difference across the thermoelectric module with contrasting flow of heat [128]. A PV-TE system that will be exposed to concentrated thermal radiation can be fabricated with optimized thermal management characteristics. Research collaborations have been conducted that allowed the theoretical and numerical calculation of heat flow and temperature distribution to determine the amount of energy generated from such a system [129]. Additionally, a copper plate introduced in the system as a thermal concentrator guaranteed that a difference in temperature on both sides of the thermoelectric module was obtained [130]. Due to the additional electrical energy generated from the TEG, the developed PV-TE hybrid system can achieve a theoretical efficiency of 23%, which is higher than that attainable by either single system. In wearable applications, device flexibility and indoor performance is also of great significance. One work demonstrated a flexible PV-TE bracelet to realize self-sustainability by harvesting solar energy and wasted human body heat [46]. This PV-TE cell can support a camera, a microphone, an accelerometer and temperature sensors under indoor illumination conditions.
The optimal power output of a PV-TE hybrid device is nearly equal to the combined maximum power outputs of the individual PV and TE devices. One work suggested that the power supplied by an individual TEG or solar cell is not sufficient to light a commercial light-emitting diode (LED) with a 1.8 V turn-on voltage [132], while the PV-TE hybrid system can support this light. At the same time, lossless coupling between PV and TE devices is feasible, and the total PCE can be raised from 12.5% to 16.3% with only a 15 °C temperature gradient by simply incorporating a TE device with a PV system [133]. However, this hybrid generator is rigid and thus less compatible with wearable sensors than single PV or TE systems.
Additionally, TEG can produce an electrical energy amounting to 648 joules in 90 min even when there is no sunlight [134]. Since the system is cost effective and power efficient, it is desirable and economical as an energy source for wearable sensors.
3.2. Combination of TENGs and Solar Cells
A hybrid self-charging textile capable of simultaneously collecting solar and body motion energy has been recently introduced, with the energy collected being stored in a supercapacitor [134]. Figure 4 shows a schematic of this self-charging textile, which consists of a double layer structure. It contains F-DSSC as the top layer to harvest sunlight energy and bottom F-SC layer to store energy. The connection of this F-DSSC and F-SC forms the F-TENG module that can make use of body motion energy. In this system, both solar and mechanical energy are converted into electricity. Solar radiation is converted into electrical energy by fiber-shaped dye-sensitized solar cells, and body motion is converted into electrical energy by fiber-shaped TENGs. This electrical energy is further converted to chemical energy in fiber-shaped supercapacitors.
Due to its all-fiber-shaped structure, the proposed self-charging textile can be easily integrated with electronic textiles to manufacture smart clothing to sustain the operations of wearable electronics and sensors.
3.3. Combination of Electromagnetic and Thermal Effects
A common strategy used by ESs to harvest energy is the utilization of electromagnetic effects. Scavengers can also be based on the Seebeck effect. Because the temperature can vary within metals and semiconductors, a voltage drop is observed across these materials. The most suitable method for measuring electromagnetic effects involves thermocouple technology. The two materials are integrated while maintaining their junctions at different temperatures. ESs that are based on this approach have many parallel thermocouples linked through an electrical connection to yield a thermopile [135]. Additional elements, such as radiators, may be used to transfer heat to the thermopile legs so as to substantially increase efficiency.
3.4. Combination of TENGs and PENGs
The TENG-PENG hybrid system can harvest mechanical energy from human motion or the ambient environment. One TENG-PENG system was able to successfully power a commercial digital watch and a temperature-humidity meter using wind as the energy source [136]. This type of hybrid generator can also be made portable and has been used to generate power for cell phones from hand vibrations [137].
In general, hybrid generators have fewer applications than single source generators because they are generally less flexible; however, they are able to offset the limitations caused by harvesting energy from a single source. To realize high compatibility with
various applications, a good hybrid energy harvester should be of high flexibility. Table 10 lists some examples of flexible hybrid power generators that have been applied or can be used to power different kinds of wearable sensors.
Table 10. Performance of latest flexible hybrid energy harvesting systems.
| EH Type | Device Type | Energy Storage System | P_\text{o} (or p_\text{o}) | Comment | Ref. |
|---------|-------------|-----------------------|---------------------------|---------|------|
| PV-TE | Wristband | Medical sensor for Temperature Heartbeat Blood oxygen saturation Body acceleration | Super capacitor | PV: 207 mW TEG: 50 mW at \( \Delta T = 20 \) K | Can support the integration of multiple medical systems. | [138] |
| PV-TE | Bracelet | NA | NA | In helical structure. High stretchability and stability. | \( \Delta T = 70 \) K is very high in wearable applications | [139] |
| PV-TE | Wristband | Data acquisition from the on-board camera and multiple sensors Visualization and wireless connectivity | Battery | 550 \( \mu \)W for PV in door 250 \( \mu \)W for TEG at \( \Delta T = 5 \) K | Tested the indoor performance of PV section | [46] |
| PV-PENG | NA | NA | Super capacitor | 0.97 W/cm$^3$ | Power conversion efficiency: 0.13% Stable output performance | [141] |
| TENG-PENG | Flexible substrate | LED lights | Light LED without storage section | 151.42 · W/cm$^2$ | High Sensitivity against tiny motion | [142] |
| TENG-PENG | Flexible slice | NA | NA | V_{\text{oc}}: 5.2 V J_{\text{sc}}: 500 nA | High Sensitivity against tiny motion | [143] |
EH: energy harvester; Ref: Reference; TENG: triboelectric nanogenerator; PV: photovoltaic; TE: thermoelectric; PENG: piezoelectric nanogenerator; V_{\text{oc}}: open circuit voltage; J_{\text{sc}}: short circuit current.
4. Comparison of Various Energy Sources
Table 11 compares the various energy sources that can be used to power wearable sensors. These energy sources can act in single format or in combination; their operation and performance make them suitable for a variety of applications, and they can be used to power many different kinds of sensing devices. The suitability of a given energy source for a given application depends on the ambient environment, continuous sensing and sensing frequency requirements, the target analyte being detected, the cost, and human behavior, among many other factors. For any given application, chances are there exists a solution to power a dedicated wearable sensor to satisfy the performance requirements.
Table 11. Characteristics of various energy sources for wearable sensors.
| Power Source | Advantages | Drawbacks | Comments |
|------------------------|-----------------------------------------------------------------------------|---------------------------------------------------------------------------|--------------------------------------------------------------------------|
| Batteries | Provide the greatest amount of power | Need replacement or frequent charge. not flexible for wearable applications. toxic | Suitable for applications that require larger components and more power |
| Perovskite solar cells | Flexible; lightweight; compatible with wearable applications [39] Sensing and power modules are integrated for enhanced miniaturization; no need for light, RF waves, or body motion [77] | Require light | Exhibit higher efficiency than that of organic flexible solar cells [34] |
| Biofuel cells | | Power density can be affected by the analyte concentration | Suitable for wearable sensors that monitor sweat [80] |
| Power Source | Advantages | Drawbacks | Comments |
|------------------------------|------------------------------------------------|------------------------------------------------|--------------------------------------------------------------------------|
| Radio frequency harvesters | Reliable source of energy | Low power; need sufficient RF signal levels | Offer a more continuous and controllable source of power than chemical sources or solar cells [71] |
| Thermoelectric generators | No need for light or body motion; cost effective. | Low-power source [64] | Suitable for applications requiring continuous or uninterrupted monitoring |
| Triboelectric generators | No need for light or RF waves | Low-power source; require body motion [124] | Usually combined with other techniques to provide useful power [134] |
| Piezoelectric generators | No need for light or RF waves | Low-power source; require body motion [102] | Usually combined with other techniques to provide useful power [134] |
| Hybrid techniques | Higher levels of power production | More complex in terms of system design and materials; power management modules are generally needed [137] | More cost and power effective than single source technologies [132] |
5. Summary and Conclusions
Sensing technologies have gained much popularity in health monitoring over the past few years. Wearable sensors dynamically and noninvasively measure biochemical markers found in biological fluids, including sweat, tears and interstitial fluids. Recently, the noninvasive monitoring of biomarkers such as metabolites, bacteria and hormones via electrochemical and optical biosensors has become a research hotspot. Researchers and scientists have miniaturized composite biosensors, transmission systems and microfluidic sampling platforms and integrated these modules into single wearable devices. However, these wearable sensing components cannot operate without energy sources. Various technologies have been introduced to offer a wide range of energy solutions that allow these wearable microsystems to function as designed. In this review, we report the various energy solutions that have been used to provide either a constant or temporal power supply to wearable devices. We present the latest developments in solar cell, BFC, thermo-electric, triboelectric, piezoelectric, RF energy harvester and WPT technologies, as well as hybrid systems that combine these single energy sources. Each technology has advantages and disadvantages. In general, there are many single-format energy generators that perform well in powering wearable sensors. These generators are more mature and compatible with the human body and are less expensive than hybrid energy-harvesting systems. However, to enhance the PCE of the system, hybrid power generators are promising solutions.
Wearable sensors need to be flexible, twistable and durable, and must conform well to the human body. Most single-source energy generators and some hybrid generators meet these requirements; however, only a small proportion can function without the need for batteries as energy storage units. Therefore, additional research is needed to determine the capabilities of generators, especially hybrid generators, to supply stable and continuous power for wearable biosensors. Currently, the performance of these generators is simply measured by evaluating their electrical properties. Furthermore, a high PCE and low toxicity are of great importance. According to the research reviewed in this paper, optimizing the geometric structure and material fabrication process and developing new materials are the main approaches to improve the performance of energy generators. With continued technological advancements, further improvements in the available energy solutions for wearable sensors are achievable, and enhanced systems can offer permanent and reliable power supplies.
Author Contributions: Conceptualization, G.R. and M.S.; writing—review and editing, G.R., Y.Z. and M.S.; supervision, M.S. All authors have read and agreed to the published version of the manuscript.
Funding: The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The work is supported by Leading Innovative and Entrepreneur Team Introduction Program of Zhejiang (No. 2020R01005), Westlake University (Grant No. 10318A992001), Tencent Foundation (Grant No. XHTX202003001), Zhejiang Key R&D Program (No. 2021C03002), and the Bright Dream Joint Institute for Intelligent Robotics (10318H991901).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.
References
1. Ye, S.; Feng, S.; Huang, L.; Bian, S. Recent Progress in Wearable Biosensors: From Healthcare Monitoring to Sports Analytics. *Biosens. Bioelectron.* 2020, 10, 205. [CrossRef]
2. Kim, J.; Campbell, A.S.; De Ávila, B.E.-F.; Wang, J. Wearable biosensors for healthcare monitoring. *Nat. Biotechnol.* 2019, 37, 389–406. [CrossRef]
3. Bian, S.; Zhu, B.; Rong, G.; Sawan, M. Towards wearable and implantable continuous drug monitoring: A review. *J. Pharm. Anal.* 2020. [CrossRef]
4. Sempionatto, J.R.; Nakagawa, T.; Pavinatto, A.; Mensah, S.T.; Imani, S.; Mercier, P.; Wang, J. Eyeglasses based wireless electrolyte and metabolite sensor platform. *Lab Chip* 2017, 17, 1834–1842. [CrossRef]
5. Kim, S.; Jeon, H.-J.; Park, S.; Lee, D.Y.; Chung, E. Tear Glucose Measurement by Reflectance Spectrum of a Nanoparticle Embedded Contact Lens. *Sci. Rep.* 2020, 10. [CrossRef]
6. Arakawa, T.; Kuroki, Y.; Nitta, H.; Toma, K.; Mitsubayashi, K.; Takeuchi, S.; Sekita, T.; Minakuchi, S. Mouth Guard Type Biosensor “Cavitous Sensor” for Monitoring of Saliva Glucose with Telemetry System. In Proceedings of the 2015 9th International Conference on Sensing Technology (ICST), Auckland, New Zealand, 8–10 December 2015; pp. 46–49.
7. Hyland, M.; Hunter, H.; Liu, J.; Veety, E.; Vashee, D. Wearable thermoelectric generators for human body heat harvesting. *Appl. Energy* 2016, 182, 518–524. [CrossRef]
8. Rachim, V.P.; Chung, W.-Y. Wearable-band type visible-near infrared optical biosensor for non-invasive blood glucose monitoring. *Sens. Actuators B Chem.* 2019, 286, 173–180. [CrossRef]
9. Tandon, A.; de Ferranti, S.D. Wearable Biosensors in Pediatric Cardiovascular Disease Promises and Pitfalls toward Generating Actionable Insights. *Circulation* 2019, 140, 350–352. [CrossRef]
10. Couts, L.V.; Plans, D.; Brown, A.W.; Collomosse, J. Deep learning with wearable based heart rate variability for prediction of mental and general health. *J. Biomed. Inform.* 2020, 112. [CrossRef]
11. Wang, D.L.T.; Yu, J.; Li, Y.; Deepu, C.J.; Ngo, D.H.; Zhou, C.; Singh, S.R.; Kob, A.S.L.; Hong, R.T.Y.; Veeavalli, B.; et al. An Integrated Wearable Wireless Vital Signs Biosensor for Continuous Inpatient Monitoring. *IEEE Sens. J.* 2020, 20, 448–462. [CrossRef]
12. Kim, J.; Sempionatto, J.R.; Imani, S.; Hartel, M.C.; Barfidokht, A.; Tang, G.; Campbell, A.S.; Mercier, P.P.; Wang, J. Simultaneous Monitoring of Sweat and Interstitial Fluid Using a Single Wearable Biosensor Platform. *Adv. Sci.* 2018, 5. [CrossRef]
13. Zhao, S.; Cui, X.; Fan, Y.; Zou, C.; Yu, D. Binary Convolutional Neural Network for Brain Computer Interfaces. In Proceedings of the 2019 China Semiconductor Technology International Conference (CSTIC), Shanghai, China, 17–18 March 2019.
14. Chen, Y.-H.; Sawan, M. Trends and Challenges of Wearable Multimodal Technologies for Stroke Risk Prediction. *Sensors 2021*, 21, 460. [CrossRef]
15. Arora, H.; Sherratt, R.S.; Janko, B.; Harwin, W. Experimental validation of the recovery effect in batteries for wearable sensors and healthcare devices discovering the existence of hidden time constants. *J. Eng.* 2017, 10. 9. [CrossRef]
16. Zhao, J.; Ghannam, R.; Htet, K.O.; Liu, Y.; Law, M.; Roy, V.A.L.; Michel, B.; Imran, M.A.; Heidari, H. Self-Powered Implantable Medical Devices: Photovoltaic Energy Harvesting Review. *Adv. Healthc. Mater.* 2020, 9. [CrossRef]
17. Corbishley, P.; Rodriguez-Villegas, E. Breathing Detection: Towards a Miniaturized, Wearable, Battery-Operated Monitoring System. *IEEE Trans. Biomed. Eng.* 2008, 55, 196–204. [CrossRef] [PubMed]
18. Fafoutis, X.; Elsts, A.; Vafeas, A.; Oikonomou, G.; Pechocki, R. On Predicting the Battery Lifetime of IoT Devices: Experiences from the SPHERE Deployments. In Proceedings of the 7th International Workshop on Real-World Embedded Wireless Systems and Networks, Shenzhen, China, 4 November 2018; pp. 7–12.
19. Yokus, M.A.; Songkakul, T.; Fozdiz, V.A.; Bozkurt, A.; Daniele, M.A. Wearable multiplexed biosensor system toward continuous monitoring of metabolites. *Biosens. Bioelectron.* 2020, 153. [CrossRef]
20. Gao, W.; Emaminejad, S.; Nyein, H.Y.Y.; Challa, S.; Chen, K.; Peek, A.; Fahad, H.M.; Ota, H.; Shiraki, H.; Kiriya, D.; et al. Fully integrated wearable sensor arrays for multiplexed in situ perspiration analysis. *Nature* 2016, 529, 509–514. [CrossRef] [PubMed]
21. Curran, L.J.; Sage, F.C.; Hagedon, M.; Hamilton, L.; Patrone, J.; Gerasopoulos, K. Wearable Sensor System for Detection of Lactate in Sweat. *Sci. Rep.* 2018, 8. [CrossRef] [PubMed]
22. Mishra, R.K.; Hubble, L.J.; Martin, A.; Kumar, R.; Barfidothk, A.; Kim, J.; Musameh, M.M.; Kyratizis, I.L.; Wang, J. Wearable Flexible and Stretchable Glove Biosensor for On-Site Detection of Organophosphorus Chemical Threats. *ACS Sens.* 2017, 2, 553–561. [CrossRef]
23. Ny Hanitra, I.; LoBello, L.; Stradolini, F.; Tuoheti, A.; Criscuolo, F.; Kilic, T.; Demarchi, D.; Carrara, S.; De Micheli, G. A Flexible Front-End for Wearable Electrochemical Sensing. In Proceedings of the 2018 IEEE International Symposium on Medical Measurements and Applications (MeMeA), Rome, Italy, 11–13 June 2018; pp. 1–6. [CrossRef]
24. Luo, X.; Yao, X.; Zhang, Y.; Zheng, X.; Xie, G.; Cui, Y. Amperometric biosensing system directly powered by button cell battery for lactate. *PLoS ONE* 2019, 14. [CrossRef]
25. Sekine, T.; Sugano, R.; Tashiro, T.; Sato, J.; Takeda, Y.; Matsui, H.; Kumaki, D.; Dos Santos, F.D.; Miyabo, A.; Tokito, S. Fully Printed Wearable Vital Sensor for Human Pulse Rate Monitoring using Ferroelectric Polymer. *Sci. Rep.* 2018, 8, 4442. [CrossRef] [PubMed]
26. Lee, Y.-H.; Kim, J.-S.; Noh, J.; Lee, I.; Kim, H.J.; Choi, S.; Seo, J.; Jeon, S.; Kim, T.-S.; Lee, J.-Y.; et al. Wearable Textile Battery Rechargeable by Solar Energy. *Nano Lett.* 2013, 13, 5753–5761. [CrossRef]
27. Gurung, A.; Qiao, Q. Solar Charging Batteries: Advances, Challenges, and Opportunities. *Joule* 2018, 2, 1217–1230. [CrossRef]
28. Arbab, A.A.; Ali, M.; Memon, A.A.; Sun, K.C.; Choi, B.J.; Jeong, S.H. An all carbon dye sensitized solar cell: A sustainable and low-cost design for metal free wearable solar cell devices. *J. Colloid Interface Sci.* 2020, 569, 386–401. [CrossRef] [PubMed]
29. Umetsu, Y.; Nakamura, Y.; Arakawa, Y.; Fujimoto, M.; Suwa, H. EHAAS: Energy Harvesters as a Sensor for Place Recognition on Wearables. In Proceedings of the 2019 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops), Kyoto, Japan, 11–15 March 2019; pp. 1–10.
30. Satharasinghe, A.; Hughes-Riley, T.; Dias, T. An investigation of a wash-durable solar energy harvesting textile. *Prog. Photovolt.: Res. Appl.* 2020, 28, 578–592. [CrossRef]
31. Duan, Y.; Zhao, G.; Liu, X.; Ma, J.; Chen, S.; Song, Y.; Pi, X.; Yu, X.; Yang, D.; Zhang, Y.; et al. Low-temperature processed tantalum/nioibium co-doped TiO2 electron transport layer for high-performance planar perovskite solar cells. *Nanotechnology* 2021, 32, 11. [CrossRef]
32. Xie, Z.; Xu, W.; Ran, G.; Li, Y.; Qin, G. Enhancing performance of flexible perovskite solar cell by secondary methamiozole treating. *Mater. Lett.* 2021, 290, 4. [CrossRef]
33. Gong, C.; Zhang, L.; Meng, X.; Xing, Z.; Rao, L.; Wang, H.; Huang, Z.; Tan, L.; Hu, T.; Hu, X.; et al. A non-wetting and conductive polyethylene dioxythiophene hole transport layer for scalable and flexible perovskite solar cells. *Sci. China Ser. B Chem.* 2021, 64, 10. [CrossRef]
34. Hu, W.; Wen, Z.; Yu, X.; Qian, P.; Lian, W.; Li, X.; Shang, Y.; Wu, X.; Chen, T.; Lu, Y.; et al. In Situ Surface Fluorination of TiO2 Nanocrystals Reinforces Interface Binding of Perovskite Layer for Highly Efficient Solar Cells with Dramatically Enhanced Ultraviolet-Light Stability. *Adv. Sci.* 2021. [CrossRef]
35. Sun, Y.; Yang, S.; Pang, Z.; Quan, Y.; Song, R.; Chen, Y.; Qi, W.; Gao, Y.; Wang, F.; Zhang, X.; et al. Preferred Film Orientation to Achieve Stable and Efficient Sn–Pb Binary Perovskite Solar Cells. *ACS Appl. Mater. Interfaces* 2021, 13, 10822–10836. [CrossRef]
36. Xie, H.; Yin, X.; Guo, Y.; Liu, D.; Liang, T.; Wang, G.-F.; Que, W. Hole transport free flexible perovskite solar cells with cost-effective carbon electrodes. *Nano Letters* 2021, 32, 9. [CrossRef]
37. Wan, J.; Wen, R.; Xia, Y.; Dai, M.; Huang, H.; Xue, L.; Zhang, Z.-G.; Fang, J.; Hui, K.F.; Fan, X. All annealing-free solution-processed highly flexible organic solar cells. *J. Mater. Chem. A* 2021, 9, 5425–5433. [CrossRef]
38. Zhao, T.; Yang, D.; Xu, T.; Zhang, M.; Zhang, S.; Qin, L.; Yu, Z. Cold-Resistant Nitrogen/Sulfur Dual-Doped Graphene Fiber Supercapacitors with Solar–Thermal Energy Conversion Effect. *Chem. A Eur. J.* 2021. [CrossRef]
39. Kim, B.J.; Kim, D.H.; Lee, Y.-Y.; Shin, H.-W.; Han, G.S.; Hong, J.S.; Mahmood, K.; Ahn, T.K.; Joo, Y.-C.; Hong, K.S.; et al. Highly efficient and bending durable perovskite solar cells: Toward a wearable power source. *Energy Environ. Sci.* 2015, 8, 916–921. [CrossRef]
40. Docampo, P.; Ball, J.M.; Darwish, M.; Eperon, G.E.; Snaith, H.J. Efficient organometal trihalide perovskite planar-heterojunction solar cells on flexible polymer substrates. *Nat. Commun.* 2013, 4. [CrossRef]
41. Kang, S.; Jeong, J.; Cho, S.; Yoon, Y.J.; Park, S.; Lim, S.; Kim, J.Y.; Ko, H. Ultrathin, lightweight and flexible perovskite solar cells with an excellent power-per-weight performance. *J. Mater. Chem. A* 2019, 7, 1107–1114. [CrossRef]
42. Park, J.; Joshi, H.; Lee, H.G.; Kiae, S.; Ogras, U.Y. Flexible PV-cell Modeling for Energy Harvesting in Wearable IoT Applications. *ACM Trans. Embed. Comput. Syst.* 2017, 16. [CrossRef]
43. Vishwanathan, B.; Reinders, A.; de Boer, D.; Desmet, L.; Ras, A.; Zahn, F.; Debije, M. A comparison of performance of flat and bent photovoltaic luminescent solar concentrators. *Sol. Energy* 2015, 112, 120–127. [CrossRef]
44. Li, B.; Hou, B.; Amaratunga, G.A.J. Indoor photovoltaics, The Next Big Trend in solution-processed solar cells. *InfoMat* 2021, 3, 15. [CrossRef]
45. Park, J.; Lee, M.; Karade, V.; Shin, S.J.; Yoo, H.; Shim, H.; Gour, K.S.; Kim, D.; Hwang, J.; Shin, D.; et al. Suppression of Defects Through Cation Substitution: A Strategic Approach to Improve the Performance of Kesterite Cu2ZnSn(S,Se)4 Solar Cells Under Indoor Light Conditions. *Solar RRL* 2021, 5, 10. [CrossRef]
46. Magno, M.; Brunelli, D.; Sigrist, L.; Andri, R.; Cavigelli, L.; Gomez, A.; Benini, L. InfiniTime: Multi-sensor wearable bracelet with human body harvesting. *Sustain. Comput. Informatics Syst.* 2016, 11, 38–49. [CrossRef]
47. Lee, M.; Choi, E.; Soufiani, A.M.; Lim, J.; Kim, M.; Chen, D.; Green, M.A.; Seidel, J.; Lim, S.; Kim, J.; et al. Enhanced Hole-Carrier Selectivity in Wide Bandgap Halide Perovskite Photovoltaic Devices for Indoor Internet of Things Applications. Adv. Funct. Mater. 2021, 31, 10. [CrossRef]
48. Biswas, S.; You, Y.-J.; Lee, Y.; Shim, J.W.; Kim, H. Efficiency improvement of indoor organic solar cell by optimization of the doping level of the hole extraction layer. Dyes Pigments 2020, 183, 8. [CrossRef]
49. Park, J.W.; Takalo, A.V.; Kim, S.H.; Son, K.R.; Kang, D.Y.; Kang, S.K.; Lee, C.B.; Choi, H.; Shim, J.W.; Kim, T.G. Surface-modified ultra-thin indium zinc oxide films with tunable work function for efficient hole transport in flexible indoor organic photovoltaics. J. Power Sources 2021, 489, 11. [CrossRef]
50. Radziemska, E. The effect of temperature on the power drop in crystalline silicon solar cells. Renew. Energy 2003, 28, 1–12. [CrossRef]
51. Virtuani, A.; Pavanello, D.; Friesen, G. Overview of Temperature Coefficients of Different Thin Film Photovoltaic Technologies. In Proceedings of the 25th European Photovoltaic Solar Energy Conference and Exhibition, Valencia, Spain, 6–10 September 2010.
52. Yun, M.J.; Sim, Y.H.; Cha, S.I.; Seo, S.H.; Lee, D.Y. Three-Dimensional Textile Platform for Electrochemical Devices and its Application to Dye-Sensitized Solar Cells. Sci. Rep. 2019, 9. [CrossRef]
53. Jörgensen, M.; Norrman, K.; Georgyvan, S.A.; Tromholt, T.; Andreasen, B.; Krebs, F.C. Stability of Polymer Solar Cells. Adv. Mater. 2012, 24, 580–612. [CrossRef]
54. Thielen, M.; Sigrist, L.; Magni, M.; Hierold, C.; Benini, L. Human body heat for powering wearable devices: From thermal energy to application. Energy Convers. Manag. 2017, 131, 44–54. [CrossRef]
55. Proto, A.; Bibbo, D.; Cerny, M.; Vala, D.; Kaš, V.; Peter, L.; Conforto, S.; Schmid, M.; Penbaker, M. Thermal Energy Harvesting on the Bodily Surfaces of Arms and Legs through a Wearable Thermo-Electric Generator. Sensors 2018, 18, 1927. [CrossRef]
56. Wang, Y.; Shi, Y.; Mei, D.; Chen, Z. Wearable thermoelectric generator for harvesting heat on the curved human wrist. Appl. Energy 2017, 205, 710–719. [CrossRef]
57. Kim, S.J.; We, J.H.; Cho, B.J. A wearable thermolectric generator fabricated on a glass fabric. Energy Environ. Sci. 2014, 7, 1959–1965. [CrossRef]
58. Leonov, V. ThermoElectric Energy Harvesting of Human Body Heat for Wearable Sensors. IEEE Sens. J. 2013, 13, 2284–2291. [CrossRef]
59. Siddique, A.R.M.; Mahmud, S.; Van Heyst, B. A review of the state of the science on wearable thermoelectric power generators (TEGs) and their existing challenges. Renew. Sustain. Energy Rev. 2017, 73, 730–744. [CrossRef]
60. Park, H.; Lee, D.; Kim, D.; Cho, H.; Eom, Y.; Hwang, J.; Kim, H.; Kim, J.; Han, S.; Kim, W. High power output from body heat harvesting based on flexible thermolectric system with low thermal contact resistance. J. Phys. D Appl. Phys. 2018, 51. [CrossRef]
61. Sun, T.; Zhou, B.; Zheng, Q.; Wang, L.; Jiang, W.; Snyder, G.J. Stretchable fabric generates electric power from woven thermoelectric fibers. Nat. Commun. 2020, 11. [CrossRef]
62. Yuan, J.; Zhu, R. Self-Powered Wearable Multi-Sensing Bracelet with Flexible Thermoelectric Power Generator. In Proceedings of the 2019 20th International Conference on Solid-State Sensors, Actuators and Microsystems & Eurosensors XXXIII, Berlin, Germany, 23–27 June 2019; pp. 1431–1434.
63. Zhao, X.; Han, W.; Zhao, C.; Wang, S.; Kong, F.; Ji, X.; Li, Z.; Shen, X. Fabrication of Transparent Paper-Based Flexible Thermoelectric Generator for Wearable Energy Harvester Using Modified Printer Printing Technology. ACS Appl. Mater. Interfaces 2019, 11, 10301–10309. [CrossRef]
64. Mallick, M.; Rösch, A.G.; Franke, L.; Ahmad, S.; Gall, A.; Geßwein, H.; Aghassi-Hagmann, J.; Lemmer, U.; Roesch, A. High-Performance Ag-Se-Based n-Type Printed Thermoelectric Materials for High Power Density Folded Generators. ACS Appl. Mater. Interfaces 2020, 12, 19655–19663. [CrossRef]
65. Shang, H.; Li, T.; Luo, D.; Yu, L.; Zou, Q.; Huang, D.; Xiao, L.; Gu, H.; Ren, Z.; Ding, F. High-Performance Ag-Modified Bi0.5Sb1.5Te3 Films for the Flexible Thermoelectric Generator. ACS Appl. Mater. Interfaces 2020, 12, 7358–7365. [CrossRef]
66. Jung, J.; Suh, E.H.; Jeong, Y.J.; Yang, H.S.; Lee, T.; Jang, J. Efficient Debundling of Few-Walled Carbon Nanotubes by Wrapping with Donor–Acceptor Polymers for Improving Thermoelectric Properties. ACS Appl. Mater. Interfaces 2019, 11, 47330–47339. [CrossRef]
67. Selvan, S.; Zaman, M.; Gobbi, R.; Wong, H.Y. Recent advances in the design and development of radio frequency-based energy harvester for powering wireless sensors: A review. J. Electromagn. Waves Appl. 2018, 32, 2110–2134. [CrossRef]
68. Visser, H.J.; Vullers, R.J.M. RF Energy Harvesting and Transport for Wireless Sensor Network Applications: Principles and Requirements. Proc. IEEE 2013, 101, 1410–1423. [CrossRef]
69. Hochmair, I.; Nopp, P.; Jolly, C.; Schmidt, M.; Schöffler, H.; Garnham, C.; Anderson, I. MED-EL Cochlear Implants: State of the Art and a Glimpse into the Future. Trends Amplif. 2006, 10, 201–219. [CrossRef]
70. Yao, Y.; Cheng, H.; Wang, Y.; Mai, J.; Lu, K.; Xu, D. An FDM-Based Simultaneous Wireless Power and Data Transfer System Functioning with High-Rate Full-Duplex Communication. IEEE Trans. Ind. Inform. 2020, 16, 6370–6381. [CrossRef]
71. Boopathi, C.S.; Sivaram, M.; Sundararajan, T.V.P.; Maheswar, R.; Yupapin, P.; Amir, I.S. Bandenna for RF energy harvesting and flexible electronics. Microsyst. Technol. 2021. [CrossRef]
72. Xu, G.; Yang, C.-Y.; Wu, J.-W.; Chang, C.-C. Harvesting Electromagnetic Energy in Air: A Wireless Energy Harvester at 2.45 GHz Using Inexpensive Materials. IEEE Microw. Mag. 2020, 21, 88–95. [CrossRef]
73. Vital, D.; Bhardwaj, S.; Volakis, J.L. Textile-Based Large Area RF-Power Harvesting System for Wearable Applications. *IEEE Trans. Antennas Propag.* 2020, 68, 2323–2331. [CrossRef]
74. Noshghabaei, S.M.; Radin, R.L.; Sawan, M. Efficient Dual-band Ultra-Low-Power RF Energy Harvesting Front-End for Wearable Devices. In Proceedings of the 2018 IEEE 61st International Midwest Symposium on Circuits and Systems (MWSCAS), Windsor, ON, Canada, 5–8 August 2018; IEEE: New York, NY, USA, 2018; pp. 444–447.
75. Abbasizadeh, H.; Hejazi, A.; Rikan, B.S.; Kim, S.Y.; Bae, J.; Lee, J.M.; Moon, J.H.; Park, J.J.; Pu, Y.G.; Hwang, K.C.; et al. A High-Efficiency and Wide-Input Range RF Energy Harvester Using Multiple Rectenna and Adaptive Matching. *Energies* 2020, 13, 1023. [CrossRef]
76. Wagih, M.; Weddell, A.S.; Beeby, S. Sub-1 GHz Flexible Concealed Rectenna Yarn for High-Efficiency Wireless-Powered Electronic Textiles. In Proceedings of the 2020 14th European Conference on Antennas and Propagation (EuCAP), Copenhagen, Denmark, 15–20 March 2020; IEEE: New York, NY, USA, 2020.
77. Sharma, H.; Haque, A.; Jafferzy, Z.A. Solar energy harvesting wireless sensor network nodes: A survey. *J. Renew. Sustain. Energy* 2018, 10. [CrossRef]
78. Kamalinejad, F.; Mahapatra, C.; Sheng, Z.; Mirabbasi, S.; Leung, V.C.M.; Guan, Y.L. Wireless energy harvesting for the Internet of Things. *IEEE Commun. Mag.* 2015, 53, 102–108. [CrossRef]
79. Shaikh, F.K.; Zeadally, S. Energy harvesting in wireless sensor networks: A comprehensive review. *Renew. Sustain. Energy Rev.* 2016, 55, 1041–1054. [CrossRef]
80. Sempionatto, J.R.; Raymundo-Pereira, P.A.; Azeredo, N.F.B.; Silva, A.N.D.E.; Angnes, L.; Wang, J. Enzymatic biofuel cells based on protective hydrophobic carbon paste electrodes: Towards epidermal bioenergy harvesting in the acidic sweat environment. *Chem. Commun.* 2020, 56, 2004–2007. [CrossRef]
81. Jeerapan, I.; Sempionatto, J.R.; Pavinatto, A.; You, J.-M.; Wang, J. Stretchable biofuel cells as wearable textile-based self-powered sensors. *J. Mater. Chem. A* 2016, 4, 18342–18353. [CrossRef]
82. Jia, W.; Valdés-Ramírez, G.; Bandodkar, A.J.; Windmiller, J.R.; Wang, J. Epidermal Biofuel Cells: Energy Harvesting from Human Perspiration. *Angev. Chem. Int. Ed.* 2013, 52, 7233–7236. [CrossRef]
83. Halámek, L.; Halámková, L.; Bocharová, V.; Szczupak, A.; Alfonta, L.; Katz, E. Implanted Biofuel Cell Operating in a Living Snail. *J. Am. Chem. Soc.* 2012, 134, 5040–5043. [CrossRef] [PubMed]
84. Zhou, M. Recent Progress on the Development of Biofuel Cells for Self-Powered Electrochemical Biosensing and Logic Biosensing: A Review. *Electroanalysis* 2015, 27, 1786–1810. [CrossRef]
85. Shitanda, I.; Morigayama, Y.; Iwashita, R.; Goto, H.; Aikawa, T.; Mikawa, T.; Hoshi, Y.; Itagaki, M.; Matsu, H.; Tokito, S.; et al. Paper-based lactate biofuel cell array with high power output. *J. Power Sources* 2021, 489, 7. [CrossRef]
86. Yin, S.; Liu, X.; Kaji, T.; Nishina, Y.; Miyake, T. Fiber-crafted biofuel cell bracelet for wearable electronics. *Biosens. Bioelectron.* 2021, 179, 6. [CrossRef] [PubMed]
87. Bollella, P.; Boeva, Z.; Latonen, R.-M.; Kano, K.; Gorton, L.; Bobacka, J. Highly sensitive and stable fructose self-powered biosensor based on a self-charging biosupercapacitor. *Biosens. Bioelectron.* 2021, 176, 9. [CrossRef] [PubMed]
88. Gu, C.; Bai, L.; Pu, L.; Gai, P.; Li, F. Highly sensitive and stable self-powered biosensors for exosomes based on dual metal-organic frameworks nanocarriers. *Biosens. Bioelectron.* 2021, 176, 6. [CrossRef] [PubMed]
89. Huang, J.; Zhang, Y.; Ding, F.; Chen, D.; Wang, Y.; Jin, X.; Zhu, X. Rational design of electroactive redox enzyme nanocapsules for high-performance biosensors and enzymatic biofuel cell. *Biosens. Bioelectron.* 2021, 174, 7. [CrossRef] [PubMed]
90. Yang, J.H.; Kim, H.R.; Lee, J.H.; Jin, J.-H.; Lee, H.U.; Kim, S.W. Electrochemical properties of enzyme electrode covalently immobilized on a graphite oxide/cobalt hydroxide/chitosan composite mediator for biofuel cells. *Int. J. Hydrog. Energy* 2021, 46, 3251–3258. [CrossRef]
91. Chansena, H.; Kamkaew, A.; Lisnund, S.; Prachai, P.; Ratwirunkit, P.; Jingpho, T.; Blay, V.; Pinyou, P. Development of a Sensitive Self-Powered Glucose Biosensor Based on an Enzymatic Biofuel Cell. *Biosensors* 2021, 11, 16. [CrossRef] [PubMed]
92. Olivo, J.; Carrara, S.; De Micheli, G. Energy Harvesting and Remote Powering for Implantable Biosensors. *IEEE Sens. J.* 2011, 11, 1573–1586. [CrossRef]
93. Bandodkar, A.J.; Wang, J. Wearable Biofuel Cells: A Review. *Electroanalysis* 2016, 28, 1188–1200. [CrossRef]
94. Zhang, X.-S.; Han, M.; Kim, B.; Bao, J.-F.; Brugger, J.; Zhang, H. All-in-one self-powered flexible microsystems based on triboelectric nanogenerators. *Nano Energy* 2018, 47, 410–426. [CrossRef]
95. Cohen, J.L.; Volpe, D.J.; Westly, D.A.; Pechenik, A.; Aburná, H.D. A Dual Electrolyte H₂/O₂ Planar Membraneless Microchannel Fuel Cell System with Open Circuit Potentials in Excess of 1.4 V. *Langmuir* 2005, 21, 3544–3550. [CrossRef]
96. Xinixin, L.; Wang, N.; Han, R.; Chen, C.; Gu, J. Double-Deck Metal Solenoids 3D Integrated in Silicon Wafer for Kinetic Energy Harvester. *Micromachines* 2021, 12, 74. [CrossRef]
97. Wang, F.; Hansen, O. Electrostatic energy harvesting device with out-of-the-plane gap closing scheme. *Sens. Actuators A Phys.* 2014, 211, 131–137. [CrossRef]
98. Lu, Y.; Martı, F.; Galayko, D.; Lhéurre, J.-M.; Basset, P. A power supply module for autonomous portable electronics: Ultralow-frequency MEMS electrostatic kinetic energy harvester with a comb structure reducing air damping. *Microsyst. Nanoeng.* 2018, 4, 28. [CrossRef]
99. Galayko, D.; Ben Hassana, M.A.; Bessaad, A.; Carulli, P.; Karmakar, S.; Samaiali, H.; Najar, F.; Basset, P. Design of a MEMS Electrostatic Kinetic Energy Harvester and its Bennet Conditioning Circuit in Integrated Technologies. In Proceedings of the IEEE 2019 Symposium on Design, Test, Integration & Packaging of MEMS and MOEMS (DTIP), Paris, France, 12–15 May 2019; pp. 1–4.
100. Basset, P.; Lu, Y.; Juillard, J.; Cottone, F.; Galayko, D. An Impact-Coupled MEMS Electrostatic Kinetic Energy Harvester and Its Predictive Model Taking Nonlinear Air Damping Effect into Account. J. Microelectromech. Syst. 2018, 27, 1–13.
101. Lu, Y.; Amroun, D.; Leprince-Wang, Y.; Basset, P. A paper-based electrostatic kinetic energy harvester with stacked multiple electret films made of electroporus polymer nanofibers. J. Phys. Conf. Ser. 2016, 773. [CrossRef]
102. Wu, Y.; Qu, J.; Daoud, W.A.; Wang, L.; Qi, T. Flexible composite-nanofiber based piezo-triбоelectric nanogenerators for wearable electronics. J. Mater. Chem. A 2019, 7, 13347–13355. [CrossRef]
103. Bairagi, S.; Ghosh, S.; Ali, S.W. A fully sustainable, self-poled, bio-waste based piezoelectric nanogenerator: Electricity generation from pomelo fruit membrane. Sci. Rep. 2020, 10. [CrossRef]
104. Qiu, Y.; Zhang, H.; Hu, L.; Yang, D.; Wang, L.; Wang, B.; Ji, J.; Liu, G.; Liu, X.; Lin, J.; et al. Flexible piezoelectric nanogenerators based on ZnO nanorods grown on common paper substrates. Nanoscale 2012, 4, 6568–6573. [CrossRef] [PubMed]
105. Zhang, G.; Zhao, P.; Zhang, X.; Han, K.; Zhao, T.; Zhang, Y.; Jeong, C.K.; Jiang, S.; Zhang, S.; Wang, Q. Flexible three-dimensional interconnected piezoelectric ceramic foam based composites for highly efficient concurrent mechanical and thermal energy harvesting. Energy Environ. Sci. 2018, 11. [CrossRef]
106. Liu, Z.; Zhang, S.; Jin, Y.M.; Ouyang, H.; Zou, Y.; Wang, X.X.; Xie, L.X.; Li, Z. Flexible piezoelectric nanogenerator in wearable self-powered active sensor for respiration and healthcare monitoring. Semicond. Sci. Technol. 2017, 32. [CrossRef]
107. Ji, S.H.; Cho, Y.-S.; Yun, J.S. Wearable Core-Shell Piezoelectric Nanofiber Yarns for Body Movement Energy Harvesting. Nanomaterials 2019, 9, 555. [CrossRef]
108. Yang, J.-H.; Cho, H.-S.; Park, S.-H.; Song, S.-H.; Yun, K.-S.; Lee, J.H. Effect of garment design on piezoelectricity harvesting from joint movement. Smart Mater. Struct. 2016, 25. [CrossRef]
109. Kim, M.; Yun, K.-S. Helical Piezoelectric Energy Harvester and Its Application to Energy Harvesting Garments. Micromachines 2017, 8, 115. [CrossRef]
110. Dąbrowska, A.; Greszta, A. Analysis of the Possibility of Using Energy Harvesters to Power Wearable Electronics in Clothing. Adv. Mater. Sci. Eng. 2019, 2019. [CrossRef]
111. Lund, A.; Rundqvist, K.; Nilsson, E.; Yu, L.; Hagström, B.; Müller, C. Energy harvesting textiles for a rainy day: Woven piezoelectricals based on melt-spun PVDF microfibres with a conducting core. NPJ Flex. Electron. 2018, 2. [CrossRef]
112. Bassani, G.; Filippeschi, A.; Ruffaldi, E. Nonresonant Kinetic Energy Harvesting Using Macrofiber Composite Patch. IEEE Sens. J. 2018, 18, 2068–2076. [CrossRef]
113. Tuncel, Y.; Bandyopadhyay, S.; Kulshrestha, S.V.; Mendez, A.; Ogras, U.Y. Towards Wearable Piezoelectric Energy Harvesting: Modeling and Experimental Validation. In Proceedings of the ISLPED '20: ACM/IEEE International Symposium on Low Power Electronics and Design, Boston, MA, USA, 10–12 August 2020.
114. Cha, Y.; Hong, J.; Lee, J.; Park, J.-M.; Kim, K. Flexible Piezoelectric Energy Harvesting from Mouse Click Motions. Sensors 2016, 16, 1045. [CrossRef] [PubMed]
115. Ippili, S.; Jella, V.; Kim, J.; Hong, S.; Yoon, S.-G. Unveiling Predominant Air-Stable Organotin Bromide Perovskite toward Mechanical Energy Harvesting. ACS Appl. Mater. Interfaces 2020, 12, 16469–16480. [CrossRef] [PubMed]
116. Kim, S.-R.; Yoo, J.-H.; Cho, Y.-S.; Park, J.-W. Flexible piezoelectric energy generators based on P(VDF-TrFE) nanofibers. Mater. Res. Express 2019, 6. [CrossRef]
117. Zhang, Y.; Jeong, C.K.; Wang, J.; Sun, H.; Li, F.; Zhang, G.; Chen, L.-Q.; Zhang, S.; Chen, W.; Wang, Q. Flexible energy harvesting polymer composites based on biofibbr-templateld 3-dimensional interconnected piezoceramics. Nano Energy 2018, 50, 35–42. [CrossRef]
118. Ippili, S.; Jella, V.; Eom, J.-H.; Kim, J.; Hong, S.; Choi, J.-S.; Tran, V.-D.; Van Hieu, N.; Kim, Y.-J.; Kim, H.-J.; et al. An eco-friendly flexible piezoelectric energy harvester that delivers high output performance is based on lead-free MASnI3 films and MASnI3-PVDF composite films. Nano Energy 2019, 57, 911–923. [CrossRef]
119. Ding, R.; Zhang, X.; Chen, G.; Wang, H.; Kishor, R.; Xiao, J.; Gao, F.; Zeng, K.; Chen, X.; Sun, X.W.; et al. High-performance piezoelectric nanogenerators composed of formamidinium lead halide perovskite nanoparticles and poly(vinylidene fluoride). Nano Energy 2017, 37, 126–135. [CrossRef]
120. Chen, J.; Oh, S.K.; Nabulsi, N.; Johnson, H.; Wang, W.; Ryu, J.-H. Biocompatible and sustainable power supply for self-powered wearable and implantable electronics using III-nitride thin-film-based flexible piezoelectric generator. Nano Energy 2019, 57, 670–679. [CrossRef]
121. He, X.; Zi, Y.; Guo, H.; Zheng, H.; Xi, Y.; Wu, C.; Wang, J.; Zhang, W.; Lu, C.; Wang, Z.L. A Highly Stretchable Fiber-Based Triboelectric Nanogenerator for Self-Powered Wearable Electronics. Adv. Funct. Mater. 2017, 27. [CrossRef]
122. Sadri, B.; Abete, A.M.; Martinez, R.V. Simultaneous electrophysiological recording and self-powered biosignal monitoring using epidermal, nanotexturized, triboelectric devices. Nanotechnology 2019, 30. [CrossRef] [PubMed]
123. Ahmed, A.; El-Kady, M.F.; Hassan, I.; Negm, A.; Pourrahimi, A.M.; Muni, M.; Selvagananpathy, P.R.; Kaner, R.B. Fire-retardant, self-extinguishing triboelectric nanogenerators. Nano Energy 2019, 59, 336–345. [CrossRef]
124. Jiang, Y.; Dong, K.; An, J.; Liang, F.; Yi, J.; Peng, X.; Ning, C.; Ye, C.; Wang, Z.L. UV-Protective, Self-Cleaning, and Antibacterial Nanofiber-Based Triboelectric Nanogenerators for Self-Powered Human Motion Monitoring. *ACS Appl. Mater. Interfaces* 2021. [CrossRef]
125. Wang, D.; Zhang, D.; Yang, Y.; Mi, Q.; Zhang, J.; Yu, L. Multifunctional Latex/Polytetrafluoroethylene-Based Triboelectric Nanogenerator for Self-Powered Organ-like MXene/Metal-Organic Framework-Derived CuO Nanohybrid Ammonia Sensor. *ACS Nano* 2021. [CrossRef]
126. Han, X.; Jiang, D.; Qu, X.; Bai, Y.; Cao, Y.; Luo, R.; Li, Z. A Stretchable, Self-Healable Triboelectric Nanogenerator as Electronic Skin for Energy Harvesting and Tactile Sensing. *Materials* 2021, 14, 1689. [CrossRef]
127. Zhu, W.; Deng, Y.; Wang, Y.; Shen, S.; Gulfam, R. High-performance photovoltaic-thermoelectric hybrid power generation system with optimized thermal management. *Energy* 2016, 100, 91–101. [CrossRef]
128. Li, G.; Shittu, S.; Diallo, T.M.; Yu, M.; Zhao, X.; Ji, J. A review of solar photovoltaic-thermoelectric hybrid system for electricity generation. *Energy* 2018, 158, 41–58. [CrossRef]
129. Kraemer, D.; Hu, L.W.; Muto, A.; Chen, X.H.; Chen, G.T.Y.; Chiesa, M. Photovoltaic-thermoelectric hybrid systems: A general optimization methodology. *Appl. Phys. Lett.* 2008, 92. [CrossRef]
130. Van Sark, W. Feasibility of photovoltaic—Thermoelectric hybrid modules. *Appl. Energy* 2011, 88, 2785–2790. [CrossRef]
131. Beeri, O.; Rotem, O.; Hazan, E.; Katz, E.A.; Braun, A.; Gelbstein, Y. Hybrid photovoltaic-thermoelectric system for concentrated solar energy conversion: Experimental realization and modeling. *J. Appl. Phys.* 2015, 118. [CrossRef]
132. Zhang, Y.; Fang, J.; He, C.; Yan, H.; Wei, Z.; Li, Y. Integrated Energy-Harvesting System by Combining the Advantages of Polymer Solar Cells and Thermoelectric Devices. *J. Phys. Chem. C* 2013, 117, 24685–24691. [CrossRef]
133. Park, K.-T.; Shin, S.-M.; Tazebay, A.S.; Um, H.-D.; Jung, J.-Y.; Jee, S.-W.; Oh, M.-W.; Park, S.-D.; Yoo, B.; Yu, C.; et al. Lossless hybridization between photovoltaic and thermoelectric devices. *Sci. Rep.* 2013, 3. [CrossRef] [PubMed]
134. Wen, Z.; Yeh, M.-H.; Guo, H.; Wang, J.; Ji, Y.; Xu, W.; Deng, J.; Zhu, L.; Wang, X.; Hu, C.; et al. Self-powered textile for wearable electronics by hybridizing fiber-shaped nanogenerators, solar cells, and supercapacitors. *Sci. Adv.* 2016, 2. [CrossRef]
135. Mitcheson, P.D. Energy Harvesting for Human Wearable and Implantable Biosensors. In Proceedings of the 2010 Annual International Conference of the IEEE Engineering in Medicine and Biology, Buenos Aires, Argentina, 31 August–4 September 2010; pp. 3432–3436.
136. Zheng, H.; Zi, Y.; He, X.; Guo, H.; Lai, Y.-C.; Wang, J.; Zhang, S.; Wu, C.; Cheng, G.; Wang, Z.L. Concurrent Harvesting of Ambient Energy by Hybrid Nanogenerators for Wearable Self-Powered Systems and Active Remote Sensing. *ACS Appl. Mater. Interfaces* 2018, 10, 14708–14715. [CrossRef]
137. Chung, J.; Yong, H.; Moon, H.; Duong, Q.V.; Choi, S.; Kim, D.; Lee, S. Hand-Driven Gyroscopic Hybrid Nanogenerator for Recharging Portable Devices. *Adv. Sci.* 2018, 5. [CrossRef] [PubMed]
138. Mohsen, S.; Zekry, A.; Youssef, K.; Abouelatta, M. A Self-powered Wearable Wireless Sensor System Powered by a Hybrid Energy Harvester for Healthcare Applications. *Wirel. Pers. Commun.* 2021, 116, 3143–3164. [CrossRef]
139. Sa, C.; Xu, X.; Wu, X.; Chen, J.; Zuo, C.; Fang, X. A wearable helical organic–inorganic photodetector with thermoelectric generators as the power source. *J. Mater. Chem. C* 2019, 7, 13097–13103. [CrossRef]
140. Wu, B.; Guo, Y.; Hou, C.; Zhang, Q.; Li, Y.; Wang, H. High-Performance Flexible Thermoelectric Devices Based on All-Inorganic Hybrid Films for Harvesting Low-Grade Heat. *Adv. Funct. Mater.* 2019, 29, 11. [CrossRef]
141. Liu, X.; Li, J.; Fang, Z.; Wang, C.; Shu, L.; Han, J. Ultraviolet-protecting, flexible and stable photovoltaic-assisted piezoelectric hybrid unit nanogenerator for simultaneously harvesting ultraviolet light and mechanical energies. *J. Mater. Sci.* 2020, 55, 15222–15237. [CrossRef]
142. Liu, J.; Yu, D.; Zheng, Z.; Huangfu, G.; Guo, Y. Lead-free BiFeO3 film on glass fiber fabric: Wearable hybrid piezoelectric-triboelectric nanogenerator. *Ceram. Int.* 2021, 47, 3573–3579. [CrossRef]
143. Zou, Y.; Liao, J.; Ouyang, H.; Jiang, D.; Zhao, C.; Li, Z.; Qu, X.; Liu, Z.; Fan, Y.; Shi, B.; et al. A flexible self-arched biosensor based on combination of piezoelectric and triboelectric effects. *Appl. Mater. Today* 2020, 20, 9. [CrossRef] | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3632,
1
],
[
3632,
6101,
2
],
[
6101,
7897,
3
],
[
7897,
12112,
4
],
[
12112,
16464,
5
],
[
16464,
21088,
6
],
[
21088,
25460,
7
],
[
25460,
29594,
8
],
[
29594,
34272,
9
],
[
34272,
38413,
10
],
[
38413,
42859,
11
],
[
42859,
47187,
12
],
[
47187,
50397,
13
],
[
50397,
52262,
14
],
[
52262,
53957,
15
],
[
53957,
58028,
16
],
[
58028,
62233,
17
],
[
62233,
67628,
18
],
[
67628,
73646,
19
],
[
73646,
79478,
20
],
[
79478,
85180,
21
],
[
85180,
91099,
22
],
[
91099,
95719,
23
]
]
} |
a6368efd63e0d596f09d159674906cd379634b7a | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 9,
"total-fallback-pages": 0,
"total-input-tokens": 24461,
"total-output-tokens": 7533,
"fieldofstudy": [
"Environmental Science"
]
} | Biodiversity and seasonal variations of zooneuston in the northwestern Mediterranean Sea
France Collard1,2,*, Amandine Collignon1,3, Jean-Henri Hecq1,3, Loïc Michel1 & Anne Goffart1,3
1 Laboratory of Oceanology, MARE Centre, University of Liège, B6C, 4000 Liège, Sart Tilman, Belgium.
2 Laboratory of Functional and Evolutionary Morphology, AFFISH-RC, University of Liège, B6C, 4000 Liège, Belgium.
3 Station de Recherches Sous-Marines et Océanographiques (STARESO), BP 33, 20260 Calvi, France.
* Corresponding author: [email protected]
ABSTRACT. Neuston includes animals and plants inhabiting the surface layer of the water column. The neustonic area is an accumulation zone for bacteria, organic molecules but also terrestrial debris. The surface layer is also the air/water exchange region. Therefore, neustonic organisms are directly exposed to several constraints such as wind stress and turbulence. The present study aims to characterize the zooneuston in terms of abundance and biodiversity and to evaluate the impacts of wind stress on neustonic abundance. Zooneustonic and zooplanktonic (depth of 5 meters) samples were collected twice a month between 30th August 2011 and 10th July 2012 in Calvi Bay, Corsica. Zooneustonic biodiversity was high and, notably, twenty-eight copepod genera were identified. Among these copepods, several organisms, belonging to the Pontellidae family, were much more frequent in neuston than in underlying plankton and their abundance depended on wind direction. Taxon-specific trends in seasonal abundance variation were present. For example, individuals of the Acantharia Lithoptera spp. were found in summer whereas the Pontellidae Anomalocera patersoni appeared in winter. Overall, our data provide a first step towards a better knowledge of neuston community structure in the Mediterranean Sea.
KEY WORDS: Neuston, Pontellidae, Mediterranean Sea, Plankton, Wind forcing.
INTRODUCTION
The neuston comprises organisms inhabiting the surface layer of the water column, whereas planktonic organisms inhabit the subsurface layer (NAUMANN, 1917). Neustonic organisms live under a particular, not well-structured surface microlayer (SML), composed of colloids and macromolecules coming from dissolved organic matter and bacteria (SIEBURTH, 1983). The surface layer is a critical zone for marine species, and is notably a feeding area for fish (CARDINALE et al., 2003; PUSINERI et al., 2005). The surface layer is also an accumulation zone of pollutants (GARCIA-FLOR et al., 2008) and debris such as terrestrial debris or (micro)plastics (RYAN et al., 2009) that can threaten marine organisms (LAIST, 1997; MATO et al., 2001; GREGORY, 2009; COLLIGNON et al., 2012 & 2014). Due to its intermediate position between the atmosphere and the water column, the neuston is exposed to many constraints related to atmospheric conditions: high light intensity, wind stress, turbulence and temperature variations. Both physical and chemical parameters of seawater can affect zooneuston, and sea surface temperature and salinity notably influence the distribution of Pontellidae copepods (ZAITSEV, 1971). Copepods are known to migrate vertically according to light (HANEY, 1988), water density (HARADA et al., 1985), oxygen concentration (HERMAN, 1984), or phytoplankton abundance (and therefore nutrient concentration, TISELIUS, 1992). In the Black Sea, neustonic communities resemble the underlying plankton (ZAITSEV, 1971). However, some neustonic species display particular morphological or ecophysiological adaptations.
For example, neustonic bacteria communities are different from planktonic communities (Franklin et al., 2005) and phytoneuston differs from phytoplankton (Hardy & Apt, 1984; Lyalyuk & Lipnitskaya, 2003). Zaitsev (1971) also reported that Pontellidae copepods are characteristic of the neuston. Very few studies deal with neuston in the Mediterranean sea (Olivar & Sabates, 1997). The diversity and abundance of mediterranean zooneustonic organisms, as well as seasonal and daily variations of these organisms remain unknown to date. In this context, aims of this study were (1) to assess zooneuston diversity in a well-preserved Mediterranean area, (2) to quantify the abundance of dominant organisms at each season, and (3) to examine relations between neuston abundance and environmental parameters. To achieve this goal, neuston community structure in Calvi Bay (NW Corsica, France) was analyzed over a 10-month-long sampling period.
MATERIAL AND METHODS
Sample collection
Zooneuston and zooplankton samples were collected twice a month between 30th August 2011 and 10th July 2012. The sampling site, located near the STARESO oceanographic station (Calvi Bay, NW Corsica; Fig. 1), is characterized by very low anthropogenic influence (Gobert et al., 2009). Neustonic and planktonic samples were collected with WP2 nets (200 µm mesh). The frame of the neuston net was rectangular (0.60 m x 0.25 m) and the trawl sampled the top 20 cm of the water column. The frame of the plankton net was circular (diameter of 0.60 m) and the trawl sampled at a depth of 5 m. Nets were towed following a trajectory fixed by two points (42°35'7.80''N 8°43'46.37''E and 42°35'5.09''N 8°43'44.39''E; Fig. 1) for 20 minutes between 7.00 AM and 8.00 AM, at a speed of 2 km/h. After collection, samples were concentrated to a volume of 0.2L and fixed in 2.5% formalin.
A short period of ten days (from 30th August to 8th September 2011) was chosen to evaluate the impact of wind stress on zooneustonic abundances. During this period, twelve neuston samples were collected at the same location (Fig. 1), using the methodology described above.
Biovolume measurements and community structure assessment
In order to rapidly and easily estimate the quantity of organisms, their biovolume was measured using a non-destructive method. Organisms were placed in a graduated cylinder, and after a 24 hour sedimentation period, it was possible to visually estimate the biovolume, i.e. the volume of the graduated cylinder occupied by all organisms. In two samples, we had to remove exceptional proliferations of Siphonophora (family of Hippopodiidae) and Velella that made precise biovolume estimation impossible. Neustonic organisms were examined and counted under a binocular microscope. For each neuston sample, occurring groups were listed. Important taxa (i.e. taxa supposedly characteristic of the neustonic environment and/or proportionally abundant in the samples)
were counted. Particular attention was given to Pontellidae copepods. The main species identification criteria for this family (ROSE, 1933; TREGOUBOFF & ROSE, 1957) included the number of ocular lenses, the number of cephalic lateral hooks, the pincer type on males’ last right appendages, and the presence and the shape of spikes on the last thoracic segment.
Environmental parameters
Wind speed and wind direction were measured near STARESO (42°34'43.32"N 8°43'8.36"E; altitude 169 m) using an AWS2700 weather station (Aanderaa Data Instruments, Bergen, Norway) equipped with a wind speed sensor 2740 and a wind direction sensor 3590. Measurements were done every 20 minutes, and data were averaged over the twelve hours before sampling. Wind speed is one of the most important parameters in the estimation of turbulence. Wind speed to the power of 3 can indeed be used as a turbulence proxy (STACEY & POND, 1997). Precipitation data were obtained from Météo France. They were taken at Calvi airport (42°31'23.88"N 8°47'30.01"E; altitude 57 m).
Statistical analyses
The relation between zooneuston abundance and environmental parameters was investigated using correlation analysis. Statistical analyses were conducted using Prism 5.03 (Graphpad Software, La Jolla, CA, U.S.A.). Significance threshold was fixed at $\alpha = 0.05$.
RESULTS
Neustonic biodiversity
Twenty-one higher taxa belonging to 9 phyla were identified and counted (Table 1). Arthropoda was the most abundant group in the neuston. It was mainly represented by copepods, particularly the genera Clausocalanus and Paracalanus (regrouped under the “Others” entry, Table 1). Pontellidae copepods were present in 16 of the 22 neuston samples. Contrastingly, they were rarely found in plankton (2 of the 22 samples), and only encountered in autumn. Most of the neustonic Pontellidae were immature individuals that could not be identified to the genus or species level. Besides those, 7 Pontellidae species belonging to 4 genera were observed during the sampling period. Pontella mediterranea (Table 1) was the most common adult Pontellidae in neuston samples, and the only one to be encountered in underlying plankton. Small copepods (Clausocalanus spp., Paracalanus spp., Oithona sp., Acartia sp.) were more abundant than large ones (Centropages sp., Temora sp., Pontellidae, Candacia sp.). Twenty-eight genera of copepods were found, including four from the Pontellidae family.
Other major zooneustonic groups included Cladocera, Mollusca, Appendicularia, Chaetognatha, ichthyoplankton and Siphonophora, as well as Collozoum inerme (Table 1). Members of Foraminifera, Polychaeta, Decapoda (adults), Amphipoda or Ostracoda were more rare.
Seasonal variations
The mean neustonic biovolume was 5.1 ml.(100 m²)$^{-1}$, varying between 0.8 ml.(100 m²)$^{-1}$ and 14.0 ml.(100 m²)$^{-1}$ throughout the studied year (Fig. 2). Neustonic biovolume did not seem to follow a clear seasonal pattern, and high values did not occur consistently: the highest value was found on 16th January 2012 (sample 10), the second highest on 24th April 2012 (sample 16) and the third on 8th May 2012 (sample 17).
No general seasonal pattern was found for total number of species, but some group-specific trends were present (Table 1). The Acantharia Lithoptera spp. and the Pontellidae Pontellopsis regalis were the only organisms found in a single season. The acantharians were found in summer (beginning of September) and the copepods in autumn. By contrast,
### TABLE 1
Abundances and seasonal variations of dominant neustonic taxa. X: present, -: absent.
| Taxa | Global mean abundance (ind.100$^{-2}$ m$^{-2}$) | Summer | Autumn | Winter | Spring |
|-------------------------------|------------------------------------------------|--------|--------|--------|--------|
| Foraminifera | - | X | X | - | X |
| Globothalamea | | | | | |
| Radiozoa | | | | | |
| Acantharia | 1.40 | X | - | - | - |
| Lithoptera spp. | | | | | |
| Amoebozoa | 30.4 | X | X | X | X |
| Lobosa incertae sedis | | | | | |
| Collozoum inerme | | | | | |
| Cnidaria | | | | | |
| Hydrozoa | 24.6 | X | X | X | X |
| Siphonophora | | | | | |
| Annelida | | | | | |
| Polychaeta | - | - | X | X | X |
| Mollusca | | | | | |
| Gastropoda | 342 | X | X | X | - |
| Creseis sp. | | | | | |
| Chaetognatha | 29.2 | X | X | X | X |
| Sagittoidea | | | | | |
| Arthropoda | | | | | |
| Maxillopoda | | | | | |
| Copepoda | | | | | |
| Calanoida | | | | | |
| Pontellidae | 96 | X | X | - | - |
| Immature Pontellidae | | | | | |
| Anomalocera patersoni | 0.58 | - | - | X | X |
| Labidocera brunescens | 30.8 | X | X | X | - |
| Labidocera wollastoni | 0.08 | X | X | - | - |
| Pontella lobianco | 0.07 | - | - | - | - |
| Pontella mediterranea | 23.6 | X | X | - | - |
| Pontellopis regalis | 0.02 | - | X | - | - |
| Pontellopis villosa | 0.10 | X | X | - | - |
| Other families | | | | | |
| Acartia sp. | 154 | X | X | X | X |
| Candacia sp. | 1.70 | - | X | X | - |
| Centropages spp. | 163 | X | X | X | - |
| Oithona sp. | 314 | X | X | X | - |
| Temora sp. | 162 | X | X | - | - |
| Others | 9375 | X | X | X | X |
| Cladocera | 1024 | X | X | - | X |
| Malacostraca | | | | | |
| Isopoda | 1.35 | - | X | X | X |
| Chordata | | | | | |
| Appendicularia | | | | | |
| Salpa spp. | 180 | X | X | X | X |
| Actinopterygii | | | | | |
| Eggs | 25.6 | X | X | X | X |
| Larvae | 2.31 | X | X | X | X |
several groups were present throughout the whole sampling period: *Collozoum inerme*, *Paracalanus* spp., *Clausocalanus* spp., *Oithona* sp., *Acartia* sp., *Centropages* spp., Appendicularia, Chaetognatha, Siphonophora, and ichthyoplankton. Most Pontellidae species were found in summer or in autumn, with the exception of *Anomalocera patersoni* which was found in spring and winter. The maximal Pontellidae abundance (1602 ind.100 m$^{-2}$) occurred on 4th October 2011 (Fig. 3).
Molluscan *Creseis* sp. and copepods *Temora* sp. were present in autumn and absent in spring. Cladocerans were abundant in September 2011, June and July 2012. Their numbers strongly decreased in October 2011 and they progressively disappeared from November 2011 to April 2012.
**Relationship with wind stress and turbulence**
During the ten days of measurement, wind directions associated with neuston sampling events fluctuated between 71° (ENE) and 260° (W) and wind speed ranged between 0.6 m/s and 14.8 m/s.
Although a non-significant negative trend seemed to be present for Pontellidae (data not shown), no correlation was found between turbulence and abundance of any of the neustonic taxa.
Mean wind direction, however, influenced the abundance of Pontellidae (Fig. 4, $r^2 = 0.7848$, $p = 0.0001$) and Chaetognatha (Fig. 4, $r^2 = 0.4969$, $p = 0.0105$). Pontellidae were more abundant when wind blew from the north-east than when it blew from the south-west. Inversely, Chaetognatha were less abundant when wind blew from the east (Fig. 4). No correlation between abundance and wind direction was found for any of the other counted taxa.
**DISCUSSION**
The neustonic environment contains a diverse assemblage of organisms. Crustacea, and mostly copepods, were the most important group in terms of biodiversity and abundance. Other major groups included Mollusca, Appendicularia, early life stages of fish, Siphonophora and Amoebozoa. Other species of Cnidaria, Acantharia and
Foraminifera were less abundant, but regularly observed. Our results are comparable with another study from the NW Mediterranean (Licandro & Icardi, 2009). Neuston biodiversity highlights the ecological importance of this zone.
Pontellidae copepods are known to be specific to the neuston. In Calvi Bay, they were accordingly much more frequently observed in neuston than in underlying plankton. In the Red Sea, they seem to be absent from samples taken “just beneath the surface” (i.e. a bit deeper than our neuston samples) of the sea (Khalil et al., 1997), supporting the fact that Pontellidae copepods mostly live in the first centimeters of the water column. Species of Pontellidae have surface attachment structures (SAS), formed by several patches of setae on the dorso-anterior surface of the cephalosome and the second thoracic segment (Ianora et al., 1992).
These SAS are thought to be an energy-saving way to adhere to the surface film, and may therefore represent a morphological adaptation to neustonic lifestyle. SAS are not present in all species of Pontoellidae (Pennell, 1973). It appears that only species of Pontoellidae with bright colors possess an SAS. This pigmentation is thought to play a protective role against the ultraviolet rays to which Pontoellidae with SAS are more exposed.
A negative linear correlation between abundance of Pontoellidae and wind direction was highlighted. When wind blows from the northeast, abundance of Pontoellidae is high, while abundance is low when wind blows from the south-west. The offshore opening of Calvi Bay is towards the northeast (Fig. 1). A northeasterly wind therefore causes offshore surface water to enter the bay. High Pontoellidae abundance in this advected water could therefore explain the observed correlation.
As explained by Holdway & Maddock (1983), abundance and diversity in the neuston vary with the day/night cycle. In their study, some taxa were more abundant during the night (e.g. Amphipoda and Ostracoda) or during the day (e.g. Appendicularia). Several taxa were more numerous at dusk or dawn (e.g. Lucifer spp. and Cnidaria except Siphonophora). We are aware of this influence on diversity and abundance. Nevertheless, for practical reasons (boat and staff availability), samples have been collected at the same hour each day. Further studies based on samples taken during multiple times of the day are needed to improve our knowledge of dynamics of Mediterranean communities.
Chaetognatha abundances, on the other hand, were inversely impacted by wind direction. These two correlations should be confirmed by concomitant sampling in the bay and offshore.
Although our study is a first step towards a better knowledge of neuston community structure in the Mediterranean Sea, information about this particular layer of the sea remains scarce. It would be interesting to know what drives dominance of certain taxa in the neuston, and what are their ecological impacts on other compartments of the pelagic ecosystem. The surface layer is indeed a feeding location for fish, as well as an accumulation zone of pollutants. In the context of increasing human population, and consequently of increasing pollution and fishing activities, neuston assemblages could undergo drastic changes. Whether or not these changes could have adverse effects on the underlying parts of the water column is a question that will only be answered through further understanding of neuston ecology.
ACKNOWLEDGEMENTS
This work was supported by the F.R.S.-FNRS (Fonds de la Recherche Scientifique), the FRIA (Fonds pour la Recherche dans l’Industrie et l’Agriculture) and the IFREMER (Institut Français de Recherche pour l’Exploitation de la Mer). We thank the crew of STARESO for sampling and logistical support, and J. Schnitzler (Laboratory of Oceanology, ULg) for his language advice and corrections. This paper is MARE publication nr. 281.
REFERENCES
Cardinale M, Casini M, Arrhenius F, Håkansson N (2003). Diel spatial distribution and feeding activity of herring (Clupea harengus) and sprat (Sprattus sprattus) in the Baltic Sea. Aquatic Living Resources 16:283-92.
Collignon A, Hecq JH, Glagani F, Voisin P, Collard F & Goffart A (2012). Neustonic microplastic and zooplankton in the North Western Mediterranean Sea. Marine Pollution Bulletin, 64:861-864.
Collignon A, Hecq JH, Galgani F, Collard F, Goffart A (2014). Annual variation in neustonic micro- and meso-plastic particles and zooplankton in the Bay of Calvi (Mediterranean-Corsica). Marine Pollution Bulletin 79:293-98.
Franklin MP, McDonald IR, Bourne DG, Owens NJP, Upstill-Goddard RC & Murrell JC
Neuston biodiversity in the northwestern Mediterranean Sea
(2005). Bacterial diversity in the bacterioneuston (sea surface microlayer): The bacterioneuston through the looking glass. Environmental Microbiology, 7:723-736.
GARCIA-FLOR N, ALZAGA R, ORTIZ L, BAYONA JM & ALBAIGES J (2008). Determination of organochlorine compounds in neuston from the Mediterranean. Environmental Technology, 29:1275-1283.
GOBERT S, SARTORETTO S, RICO-RAIMONDINO V, ANDRAL B, CHERY A, LEJEUNE P, BOISSERY P (2009). Assessment of the ecological status of Mediterranean French coastal waters as required by the Water Framework Directive using the Posidonia oceanica Rapid Easy Index: PREI. Marine Pollution Bulletin, 58:1727-1733.
GREGORY MR (2009). Environmental implications of plastic debris in marine settings-entanglement, ingestion, smothering, hangers-on, hitch-hiking and alien invasions. Philosophical Transactions of the Royal Society B-Biological Sciences, 364:2013-2025.
HANEY JF (1988). Diel patterns of zooplankton behavior. Bulletin of Marine Science, 43:583-603.
HARADA E, NISHINO M & NARITA T (1985). A partite plankton sampler and summer vertical structure of zooplankton revealed using it in shallow brackish waters, Lake Shinji-ko and Lake Naka-umi. Physiology & Ecology Japan, 22:37-57.
HARDY JT & APTS CW (1984). The sea-surface microlayer: phytoneuston productivity and effects of atmospheric particulate matter. Marine Biology, 82:293-300.
HERMAN AW (1984). Vertical copepod aggregations and interactions with chlorophyll and production on the Peru shelf. Continental Shelf Research, 3:131-146.
HOLDWAY P, MADDOCK L (1983). A comparative survey of neuston: geographical and temporal distribution patterns. Marine Biology, 76:263-70.
IANORA A, MIRALTO A, VANUCCI S (1992). The surface attachment structure: a unique type of integumental formation in neustonic copepods. Marine Biology, 113:401-407.
KHALIL MT & ABD EL-RAHMAN NS (1997). Abundance and diversity of surface zooplankton in the Gulf of Aqaba, Red Sea, Egypt. Journal of Plankton Research, 19:927-936.
LAIST DW (1997). Impacts of marine debris: entanglement of marine life in marine debris including a comprehensive list of species with entanglement and ingestion records. In: COE JM & ROGERS DB (eds), Marine debris, sources, impacts, and solutions, New York: Springer-Verlag: 99-139.
LICANDRO P & ICARDI P (2009). Basin scale distribution of zooplankton in the Ligurian Sea (north-western Mediterranean) in late autumn. Hydrobiologia 617:17-40.
LYALYUK NM & LIPNITSKAYA GP (2003). Distribution of unicellular algae at the surface of the pelagic zone of the sea of Azov. Hydrobiological Journal, 39:59-64.
MATO Y, ISOBE T, TAKADA H, KANEHIRO H, OHTAKE C & KAMINUMA T (2001). Plastic resin pellets as a transport medium for toxic chemicals in the marine environment. Environmental Science & Technology, 35:318-324.
NAUMANN E (1917). Beiträge zur kenntnis des teichnanoplanktons. II. Über das neuston des süßwasser. Biologische Zentralblatt, 37:98-106.
OLIVAR MP & SABATES A (1997). Vertical distribution of fish larvae in the north-west Mediterranean Sea in spring. Marine Biology, 129:289-300.
PENNELL WM 1973. Studies on a member of the pleuston, Anomalocera opalus n.s. (Crustacea, Copepoda) in the Gulf of St. Lawrence. Ph.D. thesis, Marine Sciences Centre, McGill University, Montreal.
PUSINERI C, VASSEUR Y, HASSANI S, MEYNIER L, SPITZ J, RIDOUX V (2005). Food and feeding ecology of juvenile albacore, Thunnus alalunga, off the Bay of Biscay: A case study. ICES Journal of Marine Science 62:116-22.
ROSE M (1933). Faune de France : Copépodes pélagiques. Kraus, Paris, 374 pp.
RYAN PG, MOORE CJ, VAN FRANEKER JA & MOLONEY CL (2009). Monitoring the abundance of plastic debris in the marine environment. Philosophical Transactions of the Royal Society B-Biological Sciences, 364:1999-2012.
SIEBURTH JM (1983). Microbiological and organic-chemical processes in the surface and mixed layers. In: LISS PS & SLINN WGN (eds), Air-Sea Exchange of Gases and Particles, Reidel Publishers Co: Hingham, MA:121-172.
STACEY MW & POND S (1997). On the Mellor–Yamada Turbulence Closure Scheme: The Surface Boundary Condition for q2. Journal of Physical Oceanography, 27:2081-2086.
Tiselius P (1992). Behavior of *Acartia tonsa* in patchy food environments. Limnology & Oceanography, 37:1640-1651.
Tregouboff G & Rose M (1957). Manuel de planctonologie méditerranéenne. Centre national de la recherche scientifique, Paris, 587 pp.
Vermeulen S, Sturaro N, Gobert S, Bouquegneau JM, Lepoint G. (2011). Potential early indicators of anthropogenically derived nutrients: A multiscale stable isotope analysis. Marine Ecology Progress Series 422:9-22.
Zaitsev YP (1971). Marine neustonology. Jerusalem: Israel Program for Scientific Translations (I.P.S.T.).
Received: June 6th, 2014
Accepted: December 1st, 2014
Branch editor: Marleen de Troch | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3564,
1
],
[
3564,
6512,
2
],
[
6512,
10026,
3
],
[
10026,
15977,
4
],
[
15977,
17954,
5
],
[
17954,
18823,
6
],
[
18823,
22592,
7
],
[
22592,
26850,
8
],
[
26850,
27511,
9
]
]
} |
246c9fbe36efdbeeb159eaa671f09a8df5b3cb4e | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 13,
"total-fallback-pages": 0,
"total-input-tokens": 44377,
"total-output-tokens": 15849,
"fieldofstudy": [
"Biology"
]
} | Altered Circulating MicroRNA Profiles After Endurance Training: A Cohort Study of Ultramarathon Runners
Ceren Eyileten, Zofia Wicik, Alex Fitas, Mikolaj Marszalek, Jenny E. Simon, Salvatore De Rosa, Szczepan Wiecha, Jeffrey Palatini, Marek Postula* and Lukasz A. Malek
1 Department of Experimental and Clinical Pharmacology, Centre for Preclinical Research and Technology, Medical University of Warsaw, Warsaw, Poland, 2 Genomics Core Facility, Centre of New Technologies, University of Warsaw, Warsaw, Poland, 3 Division of Cardiology, Department of Medical and Surgical Sciences, Magna Graecia University, Catanzaro, Italy, 4 Department of Physical Education and Health in Biała Podlaska, Józef Piłsudski University of Physical Education in Warsaw, Biała Podlaska, Poland, 5 Department of Epidemiology, Cardiovascular Disease Prevention and Health Promotion, National Institute of Cardiology, Warsaw, Poland
Background: Despite the positive effects of endurance training on the cardiovascular (CV) system, excessive exercise induces not only physiological adaptations but also adverse changes in CV system, including the heart. We aimed to evaluate the selected miRNAs expression based on bioinformatic analysis and their changes before and after an ultramarathon run.
Materials and Methods: Cardiac tissue-specific targets were identified with the Tissue 2.0 database. Gene-gene interaction data were retrieved from the STRING app for Cytoscape. Twenty-three endurance athletes were recruited to the study. Athletes ran to completion (100 km) or exhaustion (52–91 km, median 74 km). All participants completed pre- and post-run testing. miRNAs expressions were measured both before and after the race.
Results: Enrichment analysis of the signaling pathways associated with the genes targeted by miRNAs selected for qRT-PCR validation (miR-1-3p, miR-125a-5p, miR-126, miR-223, miR-125a-5p, miR-106a-5p, and miR-15a/b). All selected miRNAs showed overlap in regulation in pathways associated with cancer, IL-2 signaling, TGF-β signaling as well as BDNF signaling pathway. Analysis of metabolites revealed significant regulation of magnesium and guanosine triphosphate across analyzed miRNA targets. MiR-1-3p, miR-125a-5p, miR-126, and miR-223 expressions were measured in 23 experienced endurance athletes, before and after an ultramarathon wherein athletes ran to completion (100 km) or exhaustion (52–91 km, median 74 km). The expressions of miR-125a-5p, miR-126, and miR-223 were significantly increased after the race (p = 0.007, p = 0.001, p = 0.014, respectively). MiR-1-3p expression post-run showed a negative correlation with the post-run levels of high-sensitivity C-reactive protein (hs-CRP) (r = −0.632, p = 0.003). Higher miR-1-3p expression was found in runners, who
INTRODUCTION
Cardiovascular diseases (CVDs) are the leading cause of death globally, accounting for nearly 18 million deaths annually (World Health Organization, 2019). It is estimated that the majority of CVDs could be prevented by modifications of lifestyle including change of dietary habits and regular physical activity. Regular exercise of moderate (at least 150 min per week) to high intensity (75 min/week) is recommended by the European Society of Cardiology to reduce atherosclerotic CV risk (Pellliccia et al., 2020). Despite the positive effects of endurance training on the CV system, excessive exercise may induce not only physiological adaptations but also adverse changes in CV system, including the heart. Cardiac alterations comprise modifications of its structure, electrical activity, or function, toward a phenotype resembling pathological states (George et al., 2012).
MicroRNAs (miRNAs, miRs) are small, endogenous RNAs that form complex signaling networks responsible for regulating cell differentiation, development and homeostasis. MiRNAs are able to regulate gene expression on the post-transcriptional level suppressing or enhancing the degradation of messenger RNA (mRNA) (Ha and Kim, 2014; Eyleten et al., 2018, 2020; Pordzik et al., 2018, 2019; Sabatino et al., 2019; Gasecka et al., 2020; Jarosz-Popek et al., 2020; Soplinska et al., 2020; Wicik et al., 2020; Wolska et al., 2020; Zareba et al., 2020; Jakubik et al., 2021). MiRNAs take part in regulation of cell growth, cell differentiation, apoptosis, proliferation and they are thus involved in pathophysiology of cardiovascular pathology such as hypertrophy, inflammation, fibrosis and cardiomyocyte damage (Feinberg and Moore, 2016; Tahamtan et al., 2018; Ji et al., 2019; Soplinska et al., 2020; Jakubik et al., 2021). Circulating miRNAs are changed due to acute and endurance exercise and can be related in the adaptations to exercise. As it was reviewed before, previous studies assessed miRNA plasma levels in marathon runners and showed altered expression levels of some miRNAs after marathon race and their relation with standard fitness parameters (Mooren et al., 2014; Soplinska et al., 2020). Moreover, other studies documented correlations between miRNAs expression and cardiac injury markers such as troponin plasma levels, n-terminal b-type natriuretic peptide (NT-pro-BNP), or creatine kinase-MB (CK-MB) (Baggish et al., 2014). Therefore, these findings can suggest their potential use as biomarkers of adaptive changes in response to endurance exercise.
Several miRNAs were found to correlate with individual anaerobic lactate threshold (LT) (Mooren et al., 2014), which is defined as the exercise intensity above which blood lactate concentrations increase rapidly. LT is an indicator of endurance performance corresponding to low/moderate exercise in high-level endurance athletes (Garnacho-Castaño et al., 2015). Various evidence on the role of miRNAs in lactate dehydrogenase’s (LDH) activity exist, and it is postulated that some miRNAs may aggravate cell injury via LDH action enhancement (Ge et al., 2019). It is established that participation in ultramarathon runs leads to significant elevation in high-sensitivity troponin T (hs-TnT) concentration, although the mechanism of this phenomenon is poorly understood (Malek et al., 2020). Moreover, some miRNAs were found to be independently associated with the increase in hs-TnT levels (Widera et al., 2011).
Computational approaches are very useful in explaining the complex regulatory networks of miRNAs, and the identification of their functions and target genes in a cost-effective manner. Since each miRNA has multiple targets, it would be unrealistic to rely solely on laboratory experiments. Integrating in silico target prediction into the workflow is a powerful approach to orient the selection of promising molecules and enrich laboratory results with biological knowledge (Zhang et al., 2016).
Endurance intensive and long-term exercise is characterized by cardiovascular adaptation. Recently, the scarce clinical imaging studies on ultramarathon runners indicated that cardiovascular changes attributed to intensive training can resemble pathological states (Malek et al., 2021). Moreover, several molecular based studies showed that endurance training may increase inflammation and cardiac fibrosis (Parry-Williams and Sharma, 2020). We hypothesize that excessive endurance intensive exercise induces on a molecular basis pathological cardiac fibrosis, muscle hypertrophy and inflammation, which can come evident through the alterations of miRNA expression.
Therefore, we have aimed to perform bioinformatic analysis to determine the miRNAs related to angiogenesis, cardiac muscle function, muscle hypertrophy, coagulation, inflammation, and fibrosis processes based on detailed literature search which studied acute and chronic exercises. Results of bioinformatic analysis informed the selection of a panel of miRNAs that
finished the race under 10 h compared to runners who finished over 10 h (p = 0.001). Post-run miR-125a-5p expression showed a negative correlation with the peak lactate during the run (r = −0.576, p = 0.019).
Conclusion: Extreme physical activity, as exemplified by an ultramarathon, is associated with changes in circulating miRNAs’ expression related to inflammation, fibrosis, and cardiac muscle function. In particular, the negative correlations between miR-125a-5p and lactate concentrations, and miR-1-3p and hs-CRP, support their role in specific exercise-induced adaptation. Further studies are essential to validate the long-term effect of these observations.
Keywords: microRNAs, miRNA, bioinformatics analysis, endurance sport, in silico prediction
were then validated in a cohort of elite ultramarathon runners. Consequently, we aimed to evaluate the selected miRNAs expression changes before and after an ultramarathon run, as well as the association of miRNAs with hs-CRP and lactate concentrations in elite ultramarathon runners.
**MATERIALS AND METHODS**
**Bioinformatics Analysis MicroRNA Targets Prediction, Data Filtering, and Visualization as Interaction Networks**
**Article Search Process**
Electronic databases PubMed and Scopus were searched up to January 2021. Original studies were reviewed based on: the clinical usefulness of miRNAs as novel biomarkers of adaptive alteration in response to endurance exercise, namely running and cycling based on human subjects. We also investigated review articles and meta-analyses, and their secondary references were examined for possible inclusion. Papers describing strength exercises were excluded from our analysis (Supplementary Table 1).
The following search syntax was used: “Search (“microRNAs” [MeSH Terms] OR “miR” [MeSH Terms] OR “miRNA” [MeSH Terms] OR “circulating miRNA” [MeSH Terms] OR “circulating microRNA” [MeSH Terms]) AND (“endurance training” [MeSH Terms] OR “adaptation” [MeSH Terms] OR “change” [All Fields]) Filters: Humans”. Our search was limited to human studies and did not exclude studies on the basis of ethnicity.
**Target Prediction**
To identify targets of analyzed miRNAs we used multiMiR 1.4 R package (Ru et al., 2014). We searched the top 10% hits among all conserved and non-conserved target sites in 14 target prediction databases. For input miRNAs without mature versions of id’s we performed target predictions using all three combinations: stem-loop miRNA and -3p and -5p versions.
**Selection of Cardiovascular Disease-Related Lists of Genes**
To identify the genes associated with analyzed processes (angiogenesis, cardiac muscle functions, coagulation, fibrosis, hemopoiesis, inflammation, muscle hypertrophy, and platelet activity) among identified miRNAs targets we performed a screening of the Gene Ontology (GO) terms for the presence of key words using the biomaRt package in R (Durinck et al., 2009). GO terms associated with gene lists are available as Supplementary Table 2. In order to identify genes associated with CVD, we screened the DisGeNET database for this term and gene-disease associations (Piñero et al., 2020). Next, we selected genes which had at least five CVD related publications or three gene-disease associations. The following tissues were used for further analysis: Atrium, Capillary pericyte, Cardiac muscle, Cardiac Purkinje_cell, Cardiac Purkinje fiber, Cardiofibroblast, Cardiomyoblast, Cardiovascular system, Heart endothelial cell, Heart, Heart ventricle, Left atrium, Left ventricle, and Pericyte.
**Data Aggregation, Summarization, and Visualization**
In order to aggregate and summarize miRNA-target interactions we used our wizbionet R package (Wicik et al., 2020, 2021; Walkiewicz et al., 2021). After aggregation we ranked the obtained results using clusterizer_oneR function. Clusterizer_oneR utilizes Jenks natural breaks optimization algorithm from the original OneR package as a non-arbitrary classification dividing numbers of regulated genes into four categories (clusters). Obtained results were sorted and visualized as heatmap using Morpheus software https://software.broadinstitute.org/morpheus/.
**Enrichment Analysis**
Enrichment analysis of the ontological terms associated with the targets of miRNAs selected for quantitative polymerase chain reaction (qPCR) validation was performed using EnrichR database API plugin and databases BioPlanet_2019, HMDB_Metabolites and Jensen_DISEASES datasets. In all analyses FDR corrected p-value cutoff was set as lower than 0.05. Visualization of the overlap between the enriched terms associated with the analyzed miRNAs was performed using https://software.broadinstitute.org/morpheus/.
**Study Group**
The study was conducted on November 10, 2018 at the University of Physical Education in Warsaw. Ultramarathon runners did 65.10 laps of 1535.89 m distance on flat terrain (asphalt, bitumen track and short parts of cobblestone) which in total gave 100 km ultra-marathon run. The race was accredited by the Polish Athletics Association as the National Championships of 100 km. Total number of 23 healthy amateur runners (20 males) volunteered to participate in the study and to follow the whole protocol of the study. Initial screening was performed in case of each participant in the form of a medical questionnaire to exclude any known medical conditions as described previously (Malek et al., 2020). Blood was drawn from an antecubital vein to perform baseline analysis of hs-TnT and high sensitivity C-reactive protein (hs-CRP) levels. Simultaneously the analysis of baseline capillary lactate and glucose concentration was performed. Moreover, every six laps (approximately every 9.2 km) runners had fingertip capillary lactate and glucose assessment. The final assessment of capillary lactate and glucose concentration followed by venous blood draw for hs-TnT and hs-CRP was performed immediately after participants completed the run. In collaboration with a certified company (dataSport.pl) we have collected data on the time of the race, together with the mean pace and total distance covered by each runner participating in the study.
---
1. https://github.com/wizbionet/wizbionet/blob/master/doc/vignette_wizbionet.md
2. https://www.maratonypolskie.pl/mp_index.php?dzial=3&action=5&code=48684
Blood Collection and Biomarker Measurement
Approximately 9 ml of blood was collected from an antecubital vein into a plasma separator tube just before and within 30 min after the race. The sample was kept at room temperature for 30 min prior to centrifugation at 1,500 × g for 15 min at 18–25°C. Plasma was aliquoted into 500 µl volumes and stored in −80°C freezer. In order to determine hs-CRP and hs-TnT, an electrochemiluminescence immunoassay method (ECLIA) Roche Cobas e411 analyser (Roche Diagnostics, Mannheim, Germany) was used. Reference values were set to <5 mg/dL and <14 ng/L, for hs-CRP and hs-TnT, respectively. Glucose and lactate assessment was carried out in intervals during the run through a fingertip capillary test with Biosen C-Line analyser (EKF Diagnostics, Cardiff, United Kingdom), as described previously (Małek et al., 2020).
RNA Preparation, Detection, and Quantification of MicroRNAs by Quantitative Polymerase Chain Reaction
In order to purify samples from cell debris, plasma samples after thawing at room temperature were subjected to centrifugation at 16,000 × g for 10 min at 4°C. Total RNA was extracted using mirVana PARIS Kit (invitrogen, Applied technologies) from 500 µl of plasma. Subsequently, the obtained RNA template was subjected to a reverse transcription reaction using the TaqMan miRNA Reverse Transcription kit (ABI, CA, United States) according to guidelines provided by the manufacturer. Afterward, miRNA expressions were detected by qPCR using TaqMan miRNA Assay kits (ABI, CA, United States) for the corresponding miRNAs on a the CFX384 Touch Real-Time PCR Detection System (BioRad Inc., Hercules, CA, United States). Cel-miR-39 was added as an exogenous spike-in normalizer. Mean values of all reactions—performed in triplicate—were used in statistical analysis (De Rosa et al., 2017, 2018). MiRNA expressions are expressed as 2−ΔCT (miRNA–cel-miR-39) (De Rosa et al., 2017, 2018), then log-transformed for statistical analysis.
Statistical Analysis
All results for categorical variables were presented as a number and percentage. Continuous variables were expressed as mean ± standard deviation (SD) or median and interquartile range (IQR), depending on the normality of distribution assessed by means of the A Shapiro–Wilks test. The Student t-test or the Mann-Whitney test for unpaired samples, and Wilcoxon test for paired samples, were applied depending on the normality of the distribution. To assess the correlation between continuous variables, a Spearman test was applied. All tests were two-sided with the significance level of ρ < 0.05. Calculations were performed using SPSS version 22.0 (IBM Corporation, Chicago, IL, United States).
Ethical Considerations
At the time of design, the study was congruent with the of the Declaration of Helsinki. Written informed consent form was obtained from all participants. Both the study protocol and the informed consent form were approved by the Ethics Committee of the Regional Medical Chamber in Warsaw (no. 52/17).
RESULTS
Bioinformatic Analysis Results
To investigate miRNAs contribution to the type and intensity of endurance exertion, we conducted a bioinformatics analysis based on detailed literature search. We performed two simultaneous bioinformatics analyses: (i) tissue-specific and (ii) CV process-specific. We used as an input 55 miRNAs gathered from the literature, related to acute, and chronic exercises of runners and cyclers. We have ranked the miRNAs based on the number of process-related and tissue-related targets as presented in Figure 1. Our analysis didn’t identify any miRNA which would be uniquely associated with both chronic or both acute types of training. Thus rather we focused on identification of miRNAs the most affected by the physical activity.
Among the highly ranked miRNAs, miR-1 expression changes were related to most of the training programs and regulated the highest number of targets associated with the CV system. MiR-125a appeared in relation to chronic training programs, and targeted genes related to inflammation and CV system. Both miR-15b and miR-223 were found to be associated with acute phases of physical activity and were found in our previously published bioinformatic analysis as an important regulator of CV disease pathophysiology (Sabatino et al., 2019). Also, miR-126 expression responded to acute exposure of physical activity and regulated expression of targets related to capillary pericytes. Lastly, miR-106a was found to be influenced by acute training and had the highest number of targets related to cardiac muscle functions and haemopoiesis. Therefore, miR-1-3p, miR-126, miR-223, miR-125a-5p, miR-106a-5p, and miR-15b were selected for the further qRT-PCR validation in our study. Moreover, although miR-15a did not appear in current bioinformatic analysis, we have included it to our qRT-PCR validation analysis as well, as miR-15a and miR-15b are in the same miRNA precursor family (miR-15) and was previously found by our team as a novel regulator of insulin signaling and glucose metabolism in bioinformatic analysis (Pordzik et al., 2019).
In order to identify the pathways regulated by the miRNAs selected for qRT-PCR validation (miR-1-3p, miR-126, miR-223, miR-125a-5p, miR-106a, and miR-15a/b) we performed enrichment analysis of the genes targeted by those miRNAs. All the selected miRNAs showed overlap in regulation in pathways associated with cancer, IL-2 signaling, TGF-β signaling as well as BDNF signaling pathway (Supplementary Figure 1). Analysis of metabolites revealed significant regulation of magnesium and guanosine triphosphate across analyzed miRNA targets. We also observed the highest overlap among the following diseases: Cancer, intellectual disability and kidney cancer (six miRNAs). Analyzed miRNAs also regulated genes associated with multiple other cancerous diseases including liver and endometrial cancer, as well as neurodegenerative and acquired metabolic disease (Figure 2).
Participants
Twenty-three endurance athletes were recruited to the study. Athletes ran to completion (100 km) or exhaustion (52–91 km, median 74 km). All participants completed pre- and post-run testing. Detailed demographics are tabulated in Table 1 [also previously provided elsewhere (Małek et al., 2020)]. hs-CRP, lactate, hs-TnT and glucose levels for five runners were unavailable. The median running time in the studied group was 10.6 h (IQR 8.6–11.5) and the median pace was 8.7 min/km (IQR 8.0–9.4). Post-run values of blood pressure (both systolic and diastolic) and heart rate were significantly higher in comparison to pre-run values in all participants as expected ($p = 0.0004$, $p < 0.0001$, $p < 0.0001$, respectively). Importantly, hs-CRP values before the race were in normal range in all runners. On the other hand, hs-CRP levels significantly increased in all athletes after the race ($N = 18$; $p < 0.0001$) (Table 2), and they exceeded reference value in six cases (33%). Runners were divided into two different groups based on their post-run hs-CRP (cut off: $\leq 5$ mg/dL) and maximum lactate concentration during the race (cut off: $\leq 4$ mmol/L) concentration. Patients in the high hs-CRP group had significantly higher hs-TnT concentration changes (delta hs-TnT) compared to patients in the low hs-CRP group ($p = 0.018$).
### Table 1: Baseline and running characteristics of the studied group.
| Parameter | Ultra-marathon runners ($N = 23$) |
|----------------------------------|-----------------------------------|
| Male sex (%) | 20 (87) |
| Age, yrs (IQR) | 45 (37–54) |
| BMI, kg/m² (IQR) | 24.7 (22.7–25.7) |
| Years of running (IQR) | 4.5 (3.5–7.0) |
| Years of ultra running (IQR) | 2 (0–3) |
| Weekly running distance, km (IQR)| 66 (40–80) |
| Number of ultra races completed (IQR)| 3 (0–10) |
| Longest completed race, km (IQR) | 55 (42–80) |
Data are presented as number and percentage or median and IQR. BMI, Body mass index; IQR, Interquartile range; yrs, Years.
**TABLE 2** | Pre- and post-run values of the analyzed parameters.
| Parameter | Pre-run | Post-run | P |
|----------------------------|---------|----------|---------|
| HR, bpm (N = 23) | 54.5 (50–60) | 81.5 (76–93) | <0.0001 |
| SBP, mmHg (N = 18) | 137 (130–148) | 123 (109–133) | 0.0004 |
| DBP, mmHg (N = 18) | 84 (92–91) | 73 (70–78) | <0.0001 |
| CRP, mg/dL (N = 18) | 0.7 (0.43–1.1) | 3.2 (1.9–8.1) | <0.0001 |
| hs-TnT, ng/L (N = 18) | 5 (3–7) | 14 (12–26) | <0.0001 |
| Lactate mmol/L (N = 18) | 2 (1.7–2.4) | 2.2 (1.4–3.5) | 0.22 |
| Glucose, mg/L (N = 18) | 89 (86–95) | 93 (80–100) | 0.83 |
bpm, Beats per minute; CRP, C-reactive protein; DBP, Diastolic blood pressure; HR, Heart rate; hs-TnT, High-sensitivity troponin T; SBP, Systolic blood pressure. P values marked with bold indicate statistically significant differences between the groups < 0.05.
**Circulating MicroRNAs**
Expression levels of the selected miRNAs, miR-1-3p, miR-126, miR-223, miR-125a-5p, miR-106a-5p, and miR-15a/b were measured in 23 athletes in both pre- and post-ultramarathon run. The expression levels of miR-125a-5p (p = 0.007), miR-126 (p = 0.001), and miR-223 (p = 0.014) were significantly increased after the ultramarathon, whereas miR-15b was significantly decreased (p = 0.028). No significant difference was observed for miR-1-3p and miR-15a, while miR-106a was not detectable in any blood plasma sample (Figure 3).
**Correlations of MicroRNAs Expressions With Clinical Parameters and Race Duration**
MiR-125a-5p expression post-run was negatively correlated with max lactate levels during run (r = −0.576, p = 0.019) and miR-125a-5p pre-run showed positive correlation with delta hs-TnT levels (r = 0.807, p = 0.005). Moreover, miR-1-3p expression post-run was negatively correlated with hs-CRP levels post-run (r = −0.632, p = 0.003) (Table 3). MiR-1-3p (p = 0.006) and miR-125a-5p (p = 0.048) were significantly lower in post-run samples in the high hs-CRP group (Figure 4). Similarly, miR-125a-5p was found significantly lower (p = 0.012) in the high lactate group (Figure 5). MiR-1-3p was significantly lower in post-run samples in the runners who finished the race over 10 h (p = 0.001) (Figure 6). There were no correlations observed between miRNAs changes and changes of levels of biomarkers in ultramarathon runners (data not shown).
**FIGURE 3** | Differences in miRNA expression pre- and post-ultramarathon.
**FIGURE 4** | Differences in miR-1-3p and 125a-5p expression after completion the run in subgroups divided based on hs-CRP concentration.
**TABLE 3** | Spearman’s correlation analysis.
| miRNA | Parameter | R and P |
|----------------|-----------------------------|------------------|
| miR-125a-5p pre-run | hs-TnT delta | \( r = 0.807, p = 0.005 \) |
| miR-125a-5p post-run | max lactate (during the race) | \( r = -0.576, p = 0.019 \) |
| miR-1-3p post-run | hs-CRP post-run | \( r = -0.632, p = 0.003 \) |
| miR-15a post-run | Glucose post-run | \( r = -0.489, p = 0.019 \) |
**DISCUSSION**
Our study evaluated the effect of excessive physical fitness in a unique population of ultra-marathon runners on the expression changes of miRNAs associated with hemopoiesis, angiogenesis, cardiac muscle functions, and muscle hypertrophy selected on the basis of a bioinformatic analysis. Among the possible strategies to select miRNAs for validation studies, in silico bioinformatic analysis are beneficial, allowing to sum all available evidence and to generate predictions of specific targets and molecular interactions. Applying this framework, the current study is the first to identify the most relevant targets and to provide a validation on a very specific cohort of elite athletes. The results generated might be useful to establish new biomarkers of physiological adaptation in the endurance sportsman. Bioinformatics and computational analysis of data from systematic literature search highlighted the most promising circulating miRNAs modulated by physical exercise and endurance training. Taking into account that in silico analysis did not identify the miRNAs which would be specific only for chronic or acute types of training, we hypothesize that the difference in miRNAs profile is more related to the level of the miRNA, rather than its presence/absence. We confirmed that expressions of miR-125a-5p, miR-126, and miR-223 were significantly higher after the 100-km race in a cohort of 23 elite endurance sportsmen.
Moreover, in enrichment analysis we found several connections between miRNAs and pathways that may play a role in physiological regeneration or differentiation in endurance sport including BDNF and insulin signaling pathway (Walsh et al., 2020). BDNF is a member of the neurotrophin family, which is a well-known mediator in the development of the nervous system while supporting the survival of neurons and induces neurogenesis (Eyileten et al., 2017, 2020; Gasecka et al., 2020). BDNF acts through several different pathways including the MAPK pathway and PI3K-Akt cascade, which stimulates cell survival (Eyileten et al., 2020). Apart from the nervous system, it is well-known that BDNF can play an important role in glucose/energy homeostasis. Interestingly, many studies provided viable evidence of miRNAs-mediated post-transcriptional regulation of BDNF (Eyileten et al., 2017, 2020). Several human studies and meta-analyses showed that physical exercise leads to elevation of blood BDNF levels (Mrówczyński, 2019; Ruiz-González et al., 2021). Physical exercise induces BDNF secretion which leads to an increase of lactate levels in the blood, and elevation of CV response (Schiffer et al., 2011). If the physical activity is excessive training, increased sympathetic activity can cause platelet activation which contributes to the BDNF release to the blood circulation, as platelets contain the major source of BDNF (Fujimura et al., 2002; Eyileten et al., 2016, 2019; Walsh et al., 2017). In our bioinformatic analysis, among other important pathways also magnesium and guanosine triphosphate were found to be regulated by analyzed miRNAs. Of note, animal studies reported that magnesium can enhance exercise performance via ameliorate glucose availability in the brain, muscle and blood circulation by reducing/delaying lactate accumulation in the muscle, and thus reducing fatigue (Zhang et al., 2017). Also free extracellular guanosine 5’-triphosphate (GTP) has been demonstrated to be an improver of myogenic cell differentiation in both mouse and human cell line, which may suggest that guanosine triphosphate may serve as a miRNA-myogenic...
regulatory factor modulation and thus influence adaptation to physical activity (Pietrangelo et al., 2018).
Interestingly, we found that miR-1-3a was negatively correlated with hs-CRP after completion of the run. More importantly, we found higher miR-1-3p expressions in runners, who finished the race in under 10 h compared to runners who finished over 10 h. MiR-1 is a member of the subgroup of striated muscle-specific or muscle-enriched miRNAs called myomiRs. These molecules are involved in the regulation of muscle development, homeostasis and regeneration, as well as hypoxia/reoxygenation-induced cardiomyocytes apoptosis (Zilahi et al., 2019). It has been suggested that myomiRs show a dose-response correlation at different levels of exercise intensity and duration, and miR-1 expression pattern is dose-dependent with exercise intensity, and other miRNAs such as miR-133a or miR-222 depend on duration of exercise (Ramos et al., 2018). Previously, miR-1 has been reported to respond particularly to aerobic exercise as its concentration increases post-marathon in runners and in young males after high-intensity interval training (HIIT) or vigorous distance-matched exercise (Clauss et al., 2016; Denham and Prestes, 2016). However, specific circulating miR-1 was significantly up-regulated 3 h post exercise but not immediately or shortly after (within <1 h), which concords with our findings as blood samples for analysis were obtained within 30 min after finishing the race (Nielsen et al., 2014; Denham et al., 2018). We previously described that several studies found elevated miR-1-3p in response to endurance training, however, only one study correlated this finding with myoglobin at 24 h after the run (Soplinska et al., 2020; Yin et al., 2020). Moreover, in the current study we found a negative association between miR-1-3p and CRP levels after the 100 km race. There is a temporary rise in blood CRP levels after and during endurance exercise, caused by exercise-induced acute phase response regulated by cytokines, especially IL-6 (Kasapis and Thompson, 2005). Both short-term and long-term endurance sport can induce both anti- and pro-inflammatory responses (Barros et al., 2017). It was also previously reported that CRP level may continuously increase throughout ultra-endurance runs and strongly correlates with distance covered (Klapcińska et al., 2013). Thus, miR-1-3p may be considered as an indicator of the reparative processes elicited in response to stressful external stimuli. Moreover, it was found that its expression is decreased in myocardial damage induced by epirubicin and that miR-1 negatively modulates the expression of phosphoinositide 3-kinases catalytic subunit alpha (PIK3CA). As a result, downregulation of PIK3CA yields a significant decrease in phosphorylation of protein kinase B (Akt) and mammalian target of rifampicin kinase (mTOR), a key downstream target gene of the PI3K/Akt pathway that can inhibit apoptosis and increase autophagy ameliorating cardiac injury (Wu et al., 2018). MiR-125a plays an angiogenic role in hypoxia and inflammation in both endothelium and cardiac muscle (Wade et al., 2019). It was shown that the expression of miR-125a-5p was downregulated in ischemic myocardium shortly after myocardial infarction, whereas in another study reported increased circulating expressions of miR-125a-5p among heart failure patients, especially those with reduced ejection fraction (Boštjančič et al., 2012). Further studies reported increased expressions of miR-125a-5p as a response to paroxysmal atrial fibrillation (da Silva et al., 2017). Additionally, other reports indicate a dynamic character in their circulatory presence after myocardial injury (Niculescu et al., 2015; Yuan et al., 2019). As a summary, damaged myocardium may release miR-125a into the circulation. Moreover, in vitro studies have shown that endothelial secretion of miR-125a-5p is induced by shear rates which can be also observed during intensive training (Schmitz et al., 2019). This type of training was found to influence miRNAs expression changes as miR-125a-5p was significantly elevated in response to HIIT and thus may be affected by the intermittent nature of HIIT (Schmitz et al., 2019). Indeed, the potential existence of an intensity-dependent threshold for exercise-induced miRNA release from CV cells was reported for other miRNA species in studies comparing moderate intensity continuous (MOD) exercise to HIIT exercise (Sapp et al., 2020).
In the current study we found a higher expression of miR-125a-5p in individuals who participated in the 100 km run. The HUNT study revealed that increased levels of miR-125a were associated with low VO2 max in male participants, which is an indicator of cardiopulmonary fitness (Bye et al., 2013). In our previous study we also found that the ultramarathon runners with the highest quartile of VO2 max had a lower expression of miR-125a-5p (Eyleten et al., 2021). In the current study we observed a negative correlation between miR-125a-5p and maximum lactate concentration. Ultramarathon runners predominantly rely on energy generated in aerobic pathways. Correspondingly, such athletes possess more active oxidative enzymes (Costill and Fox, 1969; Costill et al., 1987). It was previously reported that mean lactate accumulation increases with the running distance covered and, further, that mean running speed is related to lactate concentration (Costill and Fox, 1969; Pollock, 1977; Sjödin and Jacobs, 1981; Rhodes and McKenzie, 1984; Tanaka and Matsuura, 1984; Saltin et al., 1995). Drawn from these findings was the conclusion that marathon runners adjust their running speed to achieve a level of oxygen uptake that minimizes the increase in lactate concentration in the blood, preventing an exponential lactate rise. Tendency of miR-125a-5p to impede aerobic glycolysis and lactate production has been reported in the context of several types of cancer (Huang et al., 2018). However, data on the active and healthy population is very scarce. Cancer cell's metabolism is unique, in that the majority of glucose is converted to lactate irrespective of the O2 availability. This is achieved through the overexpression of several glucose metabolism-related proteins, such as: (a) glucose transporter (GLUT) 1, (b) hexokinase (HK2), and (c) monocarboxylate transporters. They act in canalizing glucose metabolites from catabolic to anabolic processes, resulting in accelerated cell proliferation, migration, and invasion. Notably, other studies pointed to the negative role of miR–125a in the regulation of HK2, a rate-limiting enzyme for glycolysis (Jin et al., 2017; Luo et al., 2020). While the clinical applicability of these findings is yet to be determined, one study suggested miR-125a as a promising molecular target for laryngeal squamous cell carcinoma (LSCC) on the basis that miR-125a suppresses LSCC progression through targeting of HK2 both in vitro and in vivo analysis (Sun et al., 2017). Additionally, miR-125a-regulated mitochondrial fission is postulated to be a component of cellular energy disorders. Indeed, upregulated mitochondrial fission is believed to impair mitochondrial adenosine triphosphate (ATP) production by reducing electron transport chain activity (Perdiz et al., 2017), resulting in a decrease in glucose consumption and lactate production. These findings reflect those of the present study, advocating for miR-125a being the key regulator of cell energy metabolism. Increase in miR-125a expression reduced autophagy and cell proliferation, while enhancing the apoptotic rate and pro-inflammatory cascade–TNFα, IL-1β, IL-6, and IL-18 were all elevated—through the downregulation of the PI3K/Akt/mTOR signaling pathway. This suggests that PI3K inhibition intensified the ability of miR-125a to harness the inflammatory response in vitro through the regulation of the PI3K/Akt/mTOR signaling pathway (Chen et al., 2019). The PI3K/Akt signaling pathway lies at the nexus of numerous biological processes, which include the cell cycle, apoptosis, angiogenesis, and glucose metabolism. PI3K/Akt can thereby regulate the reduction of glycogen synthesis and enhance glycolysis (Xie et al., 2019). Thus, it may be hypothesized that miR-125a-5p inhibits lactate production in high intensity and long-lasting physical activity via control of pyruvate synthesis and, with that, control of acetyl-coenzyme A which feeds into the tricarboxylic acid cycle for ATP production. However, additional studies are needed to better explain the role of miR-125a-5p in energy metabolism in endurance athletes. Moreover, it can be also hypothesized that the analyzed miRNAs play an adaptive role in ultra-marathon runners as was described for miR-1, miR-125a-5p, and miR-126, also alleviating endothelial cell damage through restoration of autophagic flux by PI3K/Akt/mTOR signaling inhibition (Tang and Yang, 2018; Wu et al., 2018). It is important to note that our previous study evaluated several typical CV-related biomarkers alterations during 100 km race in elite athletes. We showed that participation in a 100 km ultra-marathon leads to a modest, but significant hs-TnT increase in the majority of runners. Additionally, mean lactate concentration during the race and change in hs-CRP correlated with troponin change (Malek et al., 2020).
Another miRNA- miR-15a may play an important role in glucose metabolism during physical activity. Several in vivo and in vitro analysis showed that miR-15a regulates the expression of transcription factors which leads to insulin resistance development (Chakraborty et al., 2014). In limited studies miR-15a was found downregulated in patients with type 2 diabetes mellitus (T2DM) compared to non-diabetic subjects (Zampetaki et al., 2010; Jiménez-Lucena et al., 2018). Importantly, our previous bioinformatic analysis pointed out that miR-15a may have an influence on blood coagulation, platelet activation, insulin signaling and glucose metabolism processes (Pordzik et al., 2019). In the current study, we found a negative correlation between miR-15a expression and glucose levels after the race, which is in line with the previous findings. Interestingly, there was no difference in miR-15a expression before and after the run. It can be hypothesized that complex insulin action on peripheral tissues may be associated with miR-15a expression, which may influence other regulatory mechanisms that prevent high glucose level, however, further studies are essential to explain this issue.
Endurance training leads to various changes and processes. It can cause damage of skeletal muscles (Armstrong, 1986) which is, besides i.e. oxidative stress (Duca et al., 2006), one of the factors enhancing training-related inflammation (Kaufmann et al., 2021). Furthermore, high intensity endurance training can result in transient acute volume overload of the atria and right ventricle (Patil et al., 2012). We hypothesize that these phenomena may be related to elevated levels of miR-223 in ultramarathon runners post-race. MiR-223 synthesis is thought to be stimulated by damaged myofibrils and muscle ischaemia. Additionally, the upregulation of miR-223 is correlated with the presence of infiltrating inflammatory cells (Greco et al., 2009; Rangrez et al., 2012; Taïbi et al., 2014), probably due to its essential role in macrophage differentiation and function. MiR-223 can target PBX/knotted 1 homeobox 1 causing macrophage polarization into anti-inflammatory type (Taïbi et al., 2014;
Zhou et al., 2015). MiR-223 may also regulate inflammation by inhibiting synthesis and preventing the excess accumulation of NLRP3, a part of inflammasome acting in response to the cellular damage (Taïbi et al., 2014). All of the aforementioned regulatory functions that maintain the balance between inflammatory and anti-inflammatory factors may be beneficial for skeletal muscle regeneration (Cheng et al., 2020). Another suspected function of miR-223, potentially of value for endurance athletes, is its ability to influence metabolic signaling. Indeed, miR-223 may influence the insulin sensitivity of adipose tissue and stimulate GLUT4 expression in cardiomyocytes, which enables increased glucose uptake by those cells (Taïbi et al., 2014). Paralleling this phenomenon is the finding that ablation of miR-223 results in enhanced IFNγ/LPS-induced nitric oxide synthase 2 (NOS2) expression, an enzyme which lies at the nexus of insulin resistance as a potent source of oxidative stress (Deiuliiis et al., 2016). Collating these results, it can be suggested that the overexpression of miR-223 in ultramarathon runners is related to metabolic adaptation, skeletal muscle damage, and inflammation.
**STUDY LIMITATIONS**
One of the major limitations of our study is that miRNAs expression was assessed in a relatively small group of ultramarathon runners and with a lack of positive and negative control group comparison. Nonetheless, our cohort of athletes remains one of the largest for which circulating miRNAs have been evaluated, to date. Other limitations are related to the demographic characteristics of our subjects including gender, as the majority of participants were male (87%). As cardiac function and structure measurement after the race and long-term follow-up was unavailable in this study, we are not equipped to draw firm conclusions on long-term effects and we cannot comment whether the changes observed were harmful. Additionally, the present study did not obtain amplification of the hsa-miR-106-5p. This could be explained by low yield of target sequence and influence of additives present in plasma separation tubes which may have affected miRNA quantitation. Indeed, chelator K2EDTA has the ability to interfere with PCR reaction by binding with magnesium chloride, an important source of magnesium for PCR, thereby affecting primer annealing temperature, specificity, and amount of qPCR product (Beekman et al., 2009). The RNA isolation procedure usually removes EDTA contamination, but traces of K2 can persist and affect the amplification process if the amount of miRNAs targets is very low or if primer design is not optimal. This can be improved by choosing NaF/KOx treated tubes, different primers of method of expression analysis (Cacheux et al., 2019). Moreover, we did not have the possibility to assess the miRNAs alterations after a resting period at the recovery phase. Additionally, a 30 min blood collection time frame post the termination of the trial may slightly affect the credibility of our findings. A more comprehensive genome-wide analysis of miRNAs regulated by acute and long-term exercise training is also warranted. Studies acknowledged that differences in health habits and status—diet (Kura et al., 2019), regular physical activity (Barber et al., 2019), and, more broadly, medical history—could produce differences in baseline miRNA expressions between subjects such that results cannot be conclusively attributed to participation in the ultramarathon (Kura et al., 2019). In our analysis we have aimed to perform bioinformatic analysis to determine the miRNAs related to angiogenesis, cardiac muscle function, muscle hypertrophy, coagulation, inflammation, and fibrosis processes. However, extreme physical activity, such as ultramarathon, could affect many organs' and tissues' function, not only heart function. Therefore, a future study recruiting more subjects and allowing for propensity score matching between exposure and control groups could clarify the role of intense exercise bouts such as ultramarathon races on miRNA profiles and miRNAs involved in signaling pathways.
**CONCLUSION**
Changes in circulating miRNAs' expression in response to endurance exercise point to their key role in evaluating specific exercise-induced adaptation. Similarly, this may in turn exert an impact on adaptation to training. The negative correlations between miR-125a-5p and lactate concentrations, and miR-1-3p and CRP further support this hypothesis. However, their protective effects should be further evaluated in this setting. Future studies will be essential in elucidating the impact of extensive and regular physical exercise on selected miRNAs. Overall, our results stand in a context of the need for novel biomarkers to facilitate the diagnostic differentiation of physiological and pathological ends of the cardiac remodeling spectrum in athletes, and the broader promise to catalog the miRNAome to prognostic ends.
**DATA AVAILABILITY STATEMENT**
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author.
**ETHICS STATEMENT**
The studies involving human participants were reviewed and approved by the Ethics Committee of the Regional Medical Chamber in Warsaw (no. 52/17). The patients/participants provided their written informed consent to participate in this study.
**AUTHOR CONTRIBUTIONS**
CE, AF, MP, ZW, MM, JS, and SW: writing—original draft preparation. ZW: bioinformatic analysis. CE, MP, LM, SD, SW, and JP: writing—review and editing. CE and MP: visualization. CE, LM, MP, SD, and JP: supervision. CE and ZW: graphical design. CE, MM, AF, and JS: laboratory analysis. All authors contributed to the article and approved the submitted version.
FUNDING
This work was funded by the Medical University of Warsaw, implemented with grant 1M9/1/M/02/G/20 and the CEPT infrastructure financed by the European Union-the European Regional Development Fund within the Operational Program “Innovative economy” for 2007–2013.
ACKNOWLEDGMENTS
This work was written by the members of the International and Intercontinental Cardiovascular and Cardiometabolic Research Group (I-COMET; www.icomet.science). We thank Pamela Czajka for the laboratory analysis.
REFERENCES
Armstrong, R. B. (1986). Muscle damage and endurance events. Sports Med. 3, 370–381. doi: 10.2165/00007256-19860305-00006
Bagnis, A. L., Park, J., Min, P.-K., Isaacs, S., Parker, B. A., Thompson, P. D., et al. (2014). Rapid upregulation and clearance of distinct circulating microRNAs after prolonged aerobic exercise. J. Appl. Physiol. 116, 522–531. doi: 10.1152/ japplphysiol.01141.2013
Barber, J. L., Zellars, K. N., Barrington, K. G., Bouchard, C., Spinale, P. G., and Sarzynski, M. A. (2019). The effects of regular exercise on circulating cardiovascular-related MicroRNAs. Sci. Rep. 9:7527. doi: 10.1038/s41598-019- 43978-x
Barros, E. S., Nascimento, D. C., Prestes, J., Nóbrega, O. T., Córdova, C., Sousa, F., et al. (2017). Acute and chronic effects of endurance running on inflammatory markers: a systematic review. Front. Physiol. 8:779. doi: 10.3389/fphys.2017. 00779
Beekman, J. M., Reischl, J., Henderson, D., Bauer, D., Ternes, R., Peña, C., et al. (2016). Serum brain-derived neurotrophic factor is related to platelet reactivity but not to genetic polymorphisms within bdnf encoding gene in patients with type 2 diabetes. Acta Diabetol. 53, 22, 69ñ76. doi: 10.12659/msm. 2021.792931/full#supplementary-material
SUPPLEMENTARY MATERIAL
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphys. 2021.792931/full#supplementary-material
Supplementary Figure 1 | Enrichment analysis results of the selected miRNAs based on bioinformatic analysis.
Fujimura, H., Altar, C. A., Chen, R., Nakamura, T., Nakahashi, T., Kambayashi, J.-I., et al. (2002). Brain-derived neurotrophic factor is stored in human platelets and released by agonist stimulation. *Thromb. Haemost.* 87, 728–734.
Garrachó-Castaño, M. V., Domínguez, R., and Maté-Muñoz, J. L. (2015). Understanding the meaning of lactate threshold in resistance exercises. *Int. J. Sports Med.* 36, 371–377. doi: 10.1055/s-0034-1384945
Gasecka, A., Siwik, D., Gajewska, M., Jaguszewski, M. J., Mazurek, T., Filipiak, K. J., et al. (2020). Early biomarkers of neurodegenerative and neurovascular disorders in diabetes. *J. Clin. Med. Res.* 9:2807. doi: 10.3390/jcim9092807
Ge, L., Cai, Y., Ying, F., Liu, H., Zhang, D., He, Y., et al. (2019). miR-181c-5p exacerbates hypoxia/reoxygenation-induced cardiomyocyte apoptosis via targeting PTPN4. *Oxid. Med. Cell. Longev.* 2019:1957920. doi: 10.1155/2019/1957920
George, K., Whyte, G. P., Green, D. J., Oxborough, D., Shave, R. E., Gaze, D., et al. (2020). The effects of exercise on the heart: acute stress and chronic adaptation. *Br. J. Sports Med.* 46 Suppl 1, i29–i36. doi: 10.1136/bjsports-2019-101141
Greco, S., De Simone, M., Colussi, C., Zaccagnini, G., Fasanaro, P., Pescatori, M., et al. (2009). Common micro-RNA signature in skeletal muscle damage and regeneration induced by Duchenne muscular dystrophy and acute ischemia. *FASEB J.* 23, 3333–3346. doi: 10.1096/fj.08-128579
Ha, M., and Kim, V. N. (2014). Regulation of microRNA biogenesis. *Nat. Rev. Mol. Cell Biol.* 15, 509–524. doi: 10.1038/nrn3883
Huang, P., Mao, L.-F., Zhang, Z.-P., Lv, W.-W., Feng, X.-P., Liao, H.-J., et al. (2018). Down-regulated miR-125a-5p promotes the reprogramming of glucose metabolism and cell malignancy by increasing levels of CD147 in thyroid cancer. *Thyroid* 28, 613–623. doi: 10.1089/thy.2017.0401
Jakubik, D., Fitas, A., Eyileten, C., Jarosz-Popek, J., Nowak, A., Czajka, P., et al. (2020). Early markers of neurodegenerative and neurovascular disorders in diabetes. *J. Clin. Med. Res.* 12, 9014ñ9030. doi: 10.18632/aging.103163
Jarosz-Popek, J., Wolska, M., Gasecka, A., Czajka, P., Jakubik, D., Sharif, L., et al. (2015). Brain-derived neurotrophic factor is stored in human platelets and released by agonist stimulation. *Diagnostics (Basel)* 10:167. doi: 10.3390/diagnostics10030167
Ruiz-González, D., Hernández-Martínez, A., Valenzuela, P. L., Morales, J. S., and Soriano-Maldonado, A. (2021). Effects of physical exercise on plasma brain-derived neurotrophic factor in neurodegenerative disorders: a systematic review and meta-analysis of randomized controlled trials. *Neurosci. Biobehav. Rev.* 128, 394–405. doi: 10.1016/j.neubiorev.2021.05.025
Sabatino, J., Wick, Z., De Rosa, S., Eyileten, C., Jakubik, D., Spaccarotella, C., et al. (2019). MicroRNAs fingerprint of bicuspid aortic valve. *J. Mol. Cell. Cardiol.* 134, 98–106. doi: 10.1016/j.yjmcc.2017.09.001
Saltin, B., Larsen, H., Terrados, N., Bangsbo, J., Bak, T., Kim, C. K., et al. (1995). Aerobic exercise performance at sea level and at altitude in Kenyan boys, junior and senior runners compared with Scandinavian runners. *Scand. J. Med. Sci. Sports* 5, 209–221. doi: 10.1111/j.1600-0838.1995.tb00357.x
Sapp, R. M., Chesney, C. A., Eagan, L. E., Evans, W. S., Zietowski, E. M., Prior, S. J., and Evans, W. S. (2011). Effect of acute voluntary lower extremity exercise on plasma BDNF concentration. *Biochem. Biophys. Res. Commun.* 408, 1443–1448. doi: 10.1016/j.bbrc.2011.07.065
Schiffer, T., Schulte, S., Sperlich, B., Achtzehn, S., Fricke, H., and Strüder, H. K. (2013). Diagnostic and prognostic impact of six circulating microRNAs in acute coronary syndrome. *J. Mol. Cell. Cardiol.* 51, 872–875. doi: 10.1016/j.yjmcc.2011.07.011
Wolska, M., Jarosz-Popek, J., Junger, E., Wick, Z., Porshoor, T., Sharif, L., et al. (2020). Long non-coding RNAs as promising therapeutic approach in ischemic stroke: a comprehensive review. *Mol. Neurobiol.* 58, 1664–1682. doi: 10.1007/s12035-020-02206-8
World Health Organization (2019). *World Health Statistics 2019: Monitoring Health for the SDGs, Sustainable Development Goals*. Geneva: WHO.
Wu, J., Sun, C., Wang, R., Li, J., Zhou, M., Yan, M., et al. (2018). Cardioprotective effect of paonol against epirubicin-induced heart injury via regulating miR-1 and PI3K/AKT pathway. *Chem. Biol. Interact.* 286, 17–25. doi: 10.1016/j.cbi.2018.02.035
Xie, Y., Shi, X., Sheng, K., Han, G., Li, W., Zhao, Q., et al. (2019). PI3K/Akt signaling transduction pathway, erythropoiesis and glycolysis in hypoxia (Review). *Mol. Med. Rep.* 19, 783–791. doi: 10.3892/mmr.2018.9713
Yin, X., Cui, S., Li, X., Li, W., Lu, Q. J., Jiang, X. H., et al. (2020). Regulation of circulatory muscle-specific MicroRNA during 8 km Run. *Int. J. Sports Med.* 41, 582–588. doi: 10.1016/j.ijsm.2019.113459
Yuan, S., Scherer, R., Tanriverdi, K., Martin, I., Rahalkar, S., and Huse, P. (2019). MicroRNA biomarkers associated with type 1 myocardial infarction in HIV-positive individuals. *AIDS* 33, 2351–2361. doi: 10.1097/QAD.0000000000002368
Zampetaki, A., Kiechl, S., Drozidov, I., Willeit, P., Mayr, U., Prokop, M., et al. (2010). Plasma microRNA profiling reveals loss of endothelial miR-126 and other microRNAs in type 2 diabetes. *Circ. Res.* 107, 810–817. doi: 10.1161/CIRCRESAHA.110.226357
Zareba, L., Fitas, A., Wolska, M., Junger, E., Eyileten, C., Wick, Z., et al. (2020). MicroRNAs and long noncoding RNAs in coronary artery disease: new and potential therapeutic targets. *Cardiol. Clin.* 38, 601–617. doi: 10.1016/j.ccl.2020.07.005
Zhang, W., Le, T. D., Liu, L., Zhou, Z.-H., and Li, J. (2016). Predicting miRNA targets by integrating regulatory knowledge with expression profiles. *PLoS One* 11:e0152860. doi: 10.1371/journal.pone.0152860
Zhang, Y., Xun, P., Wang, R., Mao, L., and He, K. (2017). Can magnesium enhance exercise performance? *Nutrients* 9:496. doi: 10.3390/nu9090496
Zhou, H., Xiao, J., Wu, N., Liu, C., Xu, J., Liu, F., et al. (2015). MicroRNA-223 regulates the differentiation and function of intestinal dendritic cells and macrophages by targeting C/EBPβ. *Cell Rep.* 13, 1149–1160. doi: 10.1016/j.celrep.2015.09.073
Zilahi, E., Adamaczek, Z., Bodoki, L., Griger, Z., Pölska, S., Nagy-Vincze, M., et al. (2019). Dysregulated expression profile of myomiRs in the skeletal muscle of patients with polymyositis. *EJIFCC* 30, 237–245.
**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
**Publisher's Note:** All claims expressed in this article are solely those of the authors and do not necessarily reflect representation of those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2787,
1
],
[
2787,
8551,
2
],
[
8551,
14118,
3
],
[
14118,
20118,
4
],
[
20118,
22392,
5
],
[
22392,
24893,
6
],
[
24893,
29170,
7
],
[
29170,
33673,
8
],
[
33673,
40725,
9
],
[
40725,
46559,
10
],
[
46559,
48595,
11
],
[
48595,
50942,
12
],
[
50942,
55600,
13
]
]
} |
509741c723fed56a343c43e3e2fccffcb149bafd | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 8,
"total-fallback-pages": 0,
"total-input-tokens": 20187,
"total-output-tokens": 7104,
"fieldofstudy": [
"Economics"
]
} | Assessing The Impact of Infrastructure on Economic Growth and Global Competitiveness
Tatyana Palei a*
a Associate Professor, Department of General Management, Kazan Federal University, Kazan, Russian Federation
Abstract
The aim of this research is to examine the degree of the influence of infrastructure on national competitiveness. Through an effectiveness of infrastructure management can improve industrial policy and gain national competitiveness. According to research of the World Bank there are several factors influencing the economy growth effectiveness and national competitiveness, including institutions, infrastructure, macroeconomic environment, health and primary education, technological readiness, market size, etc and also, there are various frameworks, models, and analytical tools that can be used in studying the causal relationships between some key infrastructure factors and national competitiveness. Based on existing models, this study aims to identify and discuss the key infrastructure factors that determine national competitiveness, which in turn influence positively on the total results of industrial policy. The results of study showed that national competitiveness is influenced basically by the level of institutional development and other seven factors, including infrastructure, in turn infrastructure factor is determined mainly by the quality of roads, railroad infrastructure, air transport and electricity supply. The key institutional traps were singled out that prevent the development of the national economy. These findings contribute to an understanding of the key factors that determine economic growth, help to explain what infrastructure factors allows to be more successful in raising income levels and offer policymakers and business leaders an important tool in the formulation of improved economic policies and institutional reforms.
© 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Selection and/ peer-review under responsibility of Academic World Research and Education Center
Keywords: national competitiveness; infrastructure; the economy growth
* Tatyana Palei. Tel.: +7-987-2961196; fax: +7-843-517-4578.
E-mail address: [email protected]
1. Introduction
The subject of this study is to evaluate the impact of infrastructure on global competitiveness. In general, infrastructure problems research is considering only a narrow part of the infrastructure capital that is in public ownership. Even if there is an opportunity to evaluate private infrastructure capital, it is difficult to separate its impact on industrial growth from the effects of public infrastructure. Therefore, in our study, we will consider only the infrastructure assets in public ownership.
Fourie (2006) argues that the infrastructure consists of two elements - "capitalness" and "publicness". Accordingly, it consists of assets that have a major, but not necessarily social. We classified the degree of capital intensity of infrastructure and social significance (Table 1).
Table 1. Classification of the degree of capital intensity of infrastructure and public works
| Publicity | Capital | Low level |
|------------|------------------------------------------|-------------------------------------|
| High level | roads, highways, railways, airports, ports, electricity, water and sewerage, telecommunications | schools, hospitals, parks, courts, museums, theatres, libraries, universities, hospitals |
| Low level | industrial infrastructure | Fountains and Statues |
Therefore, the infrastructure may include capital-intensive facilities that are not of public interest. But the public actively uses most of the infrastructure. Economists refer to such objects as physical infrastructure or infrastructure capital. In the scientific literature, the role of infrastructure is evaluated by the services provided by the physical infrastructure assets. Infrastructure services, such as energy, transport, telecommunications, provision of water, sanitation and safe disposal of waste are fundamental to all kinds of household activities and economic production. We agree with the Prud'homme (2004), Baldwin and Dixon (2008), that infrastructure is a long-term, spatially bound, capital-intensive asset with a long life cycle and the period of return on investment is often associated with a "market failure" (a situation in which the market system crashes, and economic efficiency is not achieved). For example, monopoly (if there is only one seller on the market, who can abuse their position and put a price on his product much higher than it costs), or a natural monopoly, it is a form of public goods with favourable externalities (including through external networks), which leads to reduce costs in the business, or provides significant social benefit (merit goods). Baldwin and Dixon (2008), in accordance with these features, divided infrastructure into three groups: machinery and equipment, buildings, engineering structures. The field of our study includes only the basic physical infrastructure, except the social, environmental and institutional infrastructure (schools, hospitals, prisons, etc.). In such a manner, under the infrastructure we mean the basic physical infrastructure consisting of: transport infrastructure, water infrastructure, telecommunication infrastructure and energy infrastructure. This infrastructure will be called the public infrastructure because it creates benefits for a large number of users.
2. Related Literature and Research Results
Last years the fact of the positive impact of infrastructure on productivity and economic growth is in increased attention. Fig. 1 depicts the most famous work on the subject in this area over the last 20 years.
Aschauer (1989) found out that almost simultaneously with a reduction of public investment almost everywhere the productivity growth fell sharply. He was the first who proposed that the reduction of productive public services in the United States may be crucial in explaining the overall reduction in the rate of productivity growth in the country. Mamatzakis' (2008) calculations suggest that the infrastructure is an important component of economic activity in Greece. His estimates show that the public infrastructure reduces costs in the most manufacturing industries, as it strengthens the growth of productivity of resources. The efficient infrastructure supports economic growth, improves quality of life, and it is important for national security (Baldwin, Dixon, 2008). The researchers analyze the impact of infrastructure in various aspects: regional competitiveness, economic growth, income inequality, output, labour productivity, the impact on the environment and well-being (in time and cost savings, increased safety, the development of information networks) (Bristow and Nellthorp (2000)). Some authors argue that investment in infrastructure can stimulate organizational and management changes: the construction of the railway system will lead to the standardization of the schedule, which leads to increased revenue in addition to having railway service (Mattoon, 2004). Public infrastructure provides the geographic concentration of economic resources and wider and deeper markets for output and employment (Gu, Macdonald, 2009). It affects the markets and resources of the finished product, helps to determine the spatial patterns of development and provides an extensive network of individual users at low prices. Public infrastructure is generally seen as a foundation on which to build the economy (Macdonald, 2008). Grundey (2008), Burinskiene and Rudzkiene (2009) have conducted an analysis of the implementation of sustainable development policies, they note the development of infrastructure as one of the most important aspects in the field of strategic planning for sustainable spatial and socio-economic development of the country. Aschauer (1998) confirms that the public infrastructure is the basis of the quality of life: good roads reduce the number of accidents and increase public safety, water supply system reduces the level of disease, waste management improves the health and aesthetics of the environment. Agenor and Moreno-Dodson (2006) examined the association between the presence of infrastructure and health and education in the community, and proved that infrastructure services are essential to ensure the quality and availability of health and education, which provide a wealth effect to a large extent. Damaskopoulos, Gatautis, Vitkauskaite (2008) attributed to the sources of infrastructure performance. Demetriades and Mamuneas (2000) suggest that social capital infrastructure has a significant positive impact on earnings, the demand for private means of production and delivery of products in 12 OECD countries. The results of the assessments that were made by Mentolio, Sole-Olle (2009) confirmed the idea that productive public investment in
roads positively influenced by the relative increase in labour productivity in the Spanish regions. Macdonald (2008) analyzed the impact of public infrastructure on the level of private production and found that private infrastructure is vital for the private manufacturing sector. Companies are looking at social capital as an unpaid factor of production while maximizing profits.
Nijkamp (1986) confirms that the infrastructure is one of the tools for the region development. It can affect, directly or indirectly, on the social-economic activities and other regional capacity, as well as factors of production. The author emphasizes that infrastructure policy is a condition of the regional development policy: it does not guarantee regional competitiveness, but creates the necessary conditions for achieving regional development objectives. Snieska and Draksaite (2007) say that the competitiveness of the economy is determined by many different factors, and indicator of infrastructure is one of them. Snieska and Brunecki (2009) identified infrastructure as one of the indicators of the competitiveness of regions within the country. It refers to the physical infrastructure (consisting of road transport infrastructure, telecommunications, newly built property, external accessibility of the region by land, air and water) as an indicator of the factors of production, competitive conditions in the region. Martinkus and Lukasevicius (2008) consolidate that the infrastructure services and physical infrastructure are factors that affect the investment climate at the local level and increase the attractiveness of the region. Further, we examine the extent of the infrastructure influence for global competitiveness and sources.
3. Methodology
There is no agreement among researchers on a set of variables that characterize the infrastructure: some authors explore the infrastructure as a set, the others study one particular type of infrastructure such as transport, and ignore any relationship between different types of infrastructure. In most scientific works researchers use physical indicators of public infrastructure, but not the cost parameters to avoid the difficulty of estimating the infrastructure, but there is no agreed methodology for assessing the infrastructure variables. Agénor and Moreno-Dodson (2006) define infrastructure generally, it includes transportation, water and sanitation, information and communication technology (ICT) and energy. Seeethepalli, Bramati, Veredas (2008), Seeethepalli and others (2007) and Straub (2008) examine the physical attributes of the communication infrastructure (number of telephone lines, mobile subscribers), power (energy consumption), roads (kilometres of paved roads, the percentage of paved roads), sanitation (percent of population with access to improved sanitation conditions), water supply (the percentage of the population with access to improved sources of water). Grubesic (2009), Straub, Vellutini, Warlters (2008), Yeaple, Golub (2007), Canning and Pedroni (2008) also analyze the physical characteristics of infrastructure, they assess the performance of three different sectors - telecommunications, energy and transport: main telephone lines or phone number, electricity production capacity, the length of railway lines or the length of paved roads. The use of physical indicators (on their opinion) better reflects the investment in infrastructure than in monetary terms.
It is difficult to estimate the stock of social capital reliably. Researchers commonly use the sum of past investment flows, adjusted for depreciation. In the application of the so-called perpetual inventory method, it is necessary to make certain assumptions about the duration of the life cycle and wear. You also need to know the initial amount of capital. All these assumptions are far from trivial to infrastructure. There is a huge difference in the duration of the life cycle of various types of infrastructure: for example, a railway bridge and power lines. Europe still uses roads and sewers that were built by the Roman Empire. This characteristic has serious implications in terms of funding and maintenance. Further infrastructure is divided into sub-sector, that are defined by a set of physical quantities.
The literature can be divided into four approaches to measure the effect of social capital on competitiveness. 1. In the so-called behavioural approach, estimated the cost function or profit, which includes social capital. This allows the use of flexible functional forms and some better account of the different characteristics of public and private capital. 2. The introduction of various economic restrictions with VAR-model (Vector auto regression) solved the problem of causality (cause) and endogenous. 3. The final alternative way to model the effect of public spending on social capital includes the slice (cross-section) regression analysis. Each approach has its advantages and its own set of problems. Nevertheless, the general conclusions drawn using different approaches are remarkably similar. Or, at least, the differences in the identified effects do not depend on what kind of approach
was used. We used regression analysis as a statistical process for estimating the relationships among variables. It includes many techniques for modeling and analyzing several variables with the focus on the relationship between a national competitiveness and ten independent variables, the same as the Global Competitiveness Index pillars from the World Economic Forum’s annual Global Competitiveness Report at the first stage, followed by determination of relationship between quality of overall infrastructure and its eight components at the second stage. So regression analysis helps us to understand how the typical value of the dependent variable changes when any one of the independent variables is varied, while the other independent variables are held fixed. Most commonly, regression analysis estimates the conditional expectation of the dependent variable given the independent variables – that is, the average value of the dependent variable when the independent variables are fixed. Less commonly, the focus is on a Quintile, or other location parameter of the conditional distribution of the dependent variable given the independent variables. In all cases, the estimation target is a function of the independent variables called the regression function. Regression analysis is also helps to understand which among the independent variables are related to the dependent variable, and to explore the forms of these relationships.
4. Result
To estimate the conditional expectation of the dependent variable given the independent variables and to explore the forms of these relationships we used the data from the Global competitiveness Report-2012 covering 124 economies. So dependent variable is global competitiveness (y) and global competitiveness (y) independent variables are institutions, infrastructure, macroeconomic environment, health and primary education, higher education and training, goods market efficiency, labor market efficiency, financial market development, technological readiness, market size. When testing the statistical significance of the regression coefficients null hypothesis was rejected by two factors - higher education and training, goods market efficiency, the rest were statistically significant and were included in the regression equation.
As a result regression function of the impact of economic factors on the global competitiveness (y) has the form (1) with certainty R²=98%:
\[ Y = 0.2x_1 + 0.05x_2 + 0.1x_3 + 0.1x_4 + 0.1x_7 + 0.06x_8 + 0.07x_9 + 0.13x_{10} + 0.175 \] \hspace{1cm} (1).
where \( x_1 \) – Institutions, \( x_2 \) – Infrastructure, \( x_3 \) – Macroeconomic environment, \( x_4 \) – Health and primary education, \( x_7 \) – Labor market efficiency, \( x_8 \) – Financial market development, \( x_9 \) – Technological readiness, \( x_{10} \) – Market size
According to the model, the main problems that hinder the development of national competitiveness are questions of an institutional nature. The low level of institutional development makes investment highly risky and, therefore, ineffective. We single out the key institutional traps that prevent the development of the national economy:
1) The trap selection of catch-up development model, involving the combination of leadership in some segments, which are (or can be quickly set up) a competitive advantage, but with the implementation of the catch-up strategy in most sectors of the economy and industry. If the specific guidelines for industrial policy are not delivered, the manufacturer is expected to acquire of the cheapest and, as a consequence, the non-competitive equipment that may be advanced by national standards technologies. As a result, the problem of scientific and technological backwardness of many branches will appear in a few years due to technological change. Deep gap in the scientific and technological development can lead to the abandonment of the production of their own products and their replacement by an assembly of non-competitive imported counterparts.
2) The trap of non-system of strategic thinking. Process contingency of already established production causes the synchronization of complementary and mutually supportive of each other innovations. This kind of feedback with a strong positive effect forms the growth trajectory of the new technological order. Therefore it is necessary to adopt a common strategy of innovative development of the national economy, taking into account the complementarity and interdependence of industries.
3) The state governmental orders placement corruption. Creation of a common information field does not exclude the possibility of corruption. Various indicators of corruption on the stability of the various levels of government say that the corruption is a major factor in the level of authority of customers contracting system.
4) The trap of limiting the mechanism of refinancing. The need of raising funds for the modernization of most enterprises highlights the issue of the cost of credit. If the average of credit rate on loans in the country higher than similar rates in other countries because of the imperfection of the mechanism of refinancing, manufacturers of imported products having sufficient credit, have an advantage in the cost of investment of resources, which leads to further intensification of competition between foreign and domestic businesses and the latest crisis.
5) The trap of short horizon of strategic planning. The development of public-private partnership (PPP) requires changes in the strategic planning of the State: A long-term financial planning, the development of the system of guaranteeing private sector investment, state property investing, development of monitoring of PPP projects, the economic efficiency of projects. In this case, the uncertainty for the investor is reducing.
6) The trap of inefficient owner and inefficient management. The liberalization of the economy through the large-scale privatization of state property can lead to the elimination of the administrative institutions of government rule, the loss of control of economic processes. As a consequence, the new owners of the enterprises are not able to organize an effective enterprise management, and management of enterprises, in turn, choose an effective strategy.
Further, we examine the contribution of various types of infrastructure in the level of overall quality. Of the eight factors four were significant. The model with the reliability $R^2 = 0.93$ becomes:
$$Y = 0.33x_1 + 0.089x_2 + 0.27x_4 + 0.18x_6$$
where $Y$ - Quality of overall infrastructure, $x_1$ - Quality of roads, $x_2$ - Quality of railroad infrastructure, $x_4$ - Quality of air transport infrastructure, $x_6$ - Quality of electricity supply.
Quality of port infrastructure, Available airline seat kms/week, billions, Mobile telephone subscriptions/1000 pop., Fixed telephone lines/1000 pop are not included in the list of important factors.
It is assumed that the services of social capital are clean, non-competitive public goods and are proportional to the capital stock. Thus they can, however, be deficient: number of vehicles can not exceed the performance of the road. More roads will reduce congestion and improve performance. However, the increase above a certain threshold will no longer affect the gross domestic product and competitiveness, as it will not "embroider bottlenecks" (Sanchez-Robles, 1998).
A theoretical analysis of the impact of infrastructure on economic growth and the competitiveness of domestic producers can be concluded that the impact of infrastructure is expressed as follows:
- Infrastructure enables businesses to generate additional production capacity, reduce the cost of inputs in the production and transaction costs. This is called a direct impact performance;
- Infrastructure increases the productivity of workers, and this effect is known as an indirect effect;
- The impact of infrastructure on economic growth achieves in the initial period of construction work: creating jobs in construction and related industries. Investments in infrastructure require maintenance; it further increases the number of created jobs;
- The infrastructure also has a positive impact on education and health: good health and a high level of education of labour causes economic growth;
- Infrastructure contributes to the accession of the poor and undeveloped areas to the core business activities, public communications, which can raise the value of their assets, and increase human capital.
5. Conclusion
The empirical studies on the relationship between social capital and national competitiveness should provide answers to two important questions. First is whether increase in public capital causes the increase of the national competitiveness? Second is the "politically relevant" question of investment in infrastructure is not, "what is the effect of additional infrastructure, assuming everything else is constant?" But "what is the net effect of more infrastructure, despite the fact that the construction of infrastructure diverts resources from other uses? "In other words, is the existing stock of social capital is optimal?
References
Agénor, P-R., Moreno-Dodson, B. (2006), Public infrastructure and growth: new channels and policy implications. Banca d'Italia, Italia.
Aschauer, D.A. (1989), Is public expenditure productive? Journal of Monetary Economics 23 (1989) 177-200.
Baldwin, J.R., Dixon, J. (2008), Infrastructure Capital: What is it? Where is it? How much of it is there? Canadian Productivity Review. No 16.
Ottawa: Statistics Canada.
Ballagi, B.H., Pinnoi, N. (1995), Public Capital Stock and State Productivity Growth: Further Evidence from an Error Components Model. Empirical Economics, Vol. 20, 351-359.
Behrman, J. and Wolfe, B. (1987), How does mother’s schooling affect family health, nutrition, medical care usage and household sanitation? Journal of econometrics, 36, 185–204.
Brenneman, A., Kerf, M. (2002), Infrastructure and Poverty Linkages: A Literature Review. The World Bank, Mimeo.
Bristow, A.L., Neithorp, J. (2000), Transport project appraisal in the European Union, Transport Policy 7, 51-60.
Burinskiene, M., Rudzkiene, V. (2009), Plėtros krypčių vertinimo ir valdymo informacinių modelių. Vilnius: Technika.
Calderón, C. and Servén, L. (2003b), Macroeconomic Dimensions of Infrastructure in Latin America. Presented at the Fourth Annual Stanford Conference on Latin American Economic Development, November 13-15, 2003.
Canning, D. (1999), Infrastructure's Contribution to Aggregate Output, World Bank Policy Research Working Papers No.2246.
Canning, D. and Pedroni, P., December (1999), Infrastructure and Long-Run Economic Growth. CAER II Discussion Paper No. 57. Harvard
Clark, T., Woodley, R., De Halas, D. (1962), Gas-Graphite Systems, in “Nuclear Graphite”. Academic Press, New York, p. 387.
Damaskopoulos T., Gatavis, R., & Vitkauskaitė, E. (2008), Extended and Dynamic Clustering of SMEs. Inzinerine Ekonomika-Engineering Economics(1), 11-21.
Deal, B., Grove, A. (1965), General Relationship for the Thermal Oxidation of Silicon. Journal of Applied Physics 36, p. 3770.
Demetriades, P. O., Mamuneas, T. P. (2000), Intertemporal Output and Employment Effects of Public Infrastructure Capital: Evidence from 12 OECD Economies. The Economic Journal (110), 687-712.
Devarajan, S., et al. (1996), The composition of public expenditure and economic growth. Journal of Monetary Economics 37, 313-344.
Easterly, W & Levine, R. (1996), Africa's Growth Tragedy: Policies and Ethnic Divisions, Papers 536, Harvard.
Easterly, W., Rebelo, S. (1993), Fiscal policy and economic growth: An empirical investigation. Journal of Monetary Economics, Elsevier, vol. 32(3), pages 417-458, December.
Esfahani, H.S., Ramirez M.T. (2003), Journal of Development Economics 70 (2003) 443–477.
Estache, A. (2003), On Latin America’s Infrastructure Privatization and Its Distributional Effects, the Center for Global Development, Washington, D.C., February 24-25, 2003.
Fay, M., Leipziger, D., Wodon, Q., Yepes, T. (2003), Achieving the Millennium Development Goals: The role of infrastructure, Policy Research Working Paper Series 3163, The World Bank.
Fernald, J. (1997), Roads to Prosperity? Assessing the Link Between Public Capital and Productivity. Board of Governors of the Federal Reserve System, International Finance Discussion Papers Number 592.
Ferreira, C.F. (1995), Growth and Fiscal Effects of Infrastructure Investment in Brazil. Graduate School of Economics –Fundação Getulio Vargas. Carlos Hamilton Vasconcelos Arabújo. Fundação Getulio Vargas and Banco Central do Brasil.
Grubesic, T. H. (2009), The Management and Measurement of Infrastructure: Performance, Efficiency and Innovation. Growth and Change (1), 184–187.
Grundey, D. (2008), Managing sustainable tourism in Lithuania: Dream or reality? Technological and Economic Development of Economy, 14(2), 118-129.
Gu, W., & Macdonald, R. (2009), The Impact of Public Infrastructure on Canadian Multifactor Productivity Estimates. The Canadian Productivity Review. Research paper (21).
Holtz-Eakin D. (1994), Public Sector Capital and the Productivity Puzzle, Review of Economics and Statistics, Vol 76, pp 12-21.
Hulten, C. (1996), Infrastructure Effectiveness as a Determinant of Economic Growth: How Well You Use it May Be More Important than How Much You Have. NBER Working Paper 5847, December 1996 (revised December 2005).
Jacoby, Hanan G. (2000), Access to Markets and the Benefits of Rural Roads. The Economic Journal. 110 (July): 713-737.
Jalan, J., Ravallion, M. (2002), Household Income Dynamics in Rural China, Working Papers UNU-WIDER Research Paper, World Institute for Development Economic Research (UNU-WIDER).
Krypštėnienė, A. (2003), Infrastructure and Poverty Linkages: A Literature Review. The World Bank, Mimeo.
Lavy, V., Strauss, J., Thomas, D., and Vreyer, P. (1996), Quality of Health Care, Survival and Health Outcomes in Ghana. Journal of Health Economics, 15 (June 1996), 333-57.
Lee, K., Pesaran M.H. and Smith R. (1997), Growth and Convergence in a Multicountry Empirical Stochastic Solow Model. Journal of Applied Econometrics, Vol. 12 pp 357-392.
López, H. (2004), Macroeconomics and Inequality. The World Bank Research Workshop.
Macdonald, R. (2008), An Examination of Public Capital’s Role in Production. Economic Analysis Research Paper Series. No. 50. Ottawa: Statistics Canada.
Mamatzakis, E.C. (2008), Economic performance and public infrastructure: an application to Greek manufacturing. Bulletin of Economic Research (60), 307-326.
Martinkus, B., Lukasevicius, K. (2008), Investment environment of Lithuanian resorts: Researching national and local factors in the Palanga case. Transformations in Business & Economics, 7(2), 67-83.
Mattoon, R. H. (2004), Infrastructure and State Economic Development: A survey of the issues (I-G). Economic Conference.
Mentolio, D., & Solé-Ollé, A. (2009), Road investment and regional productivity growth: the effects of vehicle intensity and congestion. Papers in Regional Science (88), 99-118.
Nijkamp, P. (1986), Infrastructure and Regional development: A multidimensional policy analysis. Empirical Economics (1), 1-21.
Prud’homme, R., 2005. Infrastructure and Development. Lessons of Experience. Proceedings of the 2004 Annual Bank conference on Development Economics. 153-181.
Roller, L.-H., Waverman L. (2001), Telecommunications Infrastructure and Economic Development: A Simultaneous Approach, American Economic Review, American Economic Association, vol. 91(4), pages 909-923, September.
Sanchez-Robles, B. (1998), Infrastructure Investment And Growth: Some Empirical Evidence, Contemporary Economic Policy. Western Economic Association International, vol. 16(1), pages 98-108, 01.
Seethepalli K., Bramati M.C., Veredas D. (2008), How Relevant Is Infrastructure To Growth In East Asia? The world bank e-library. Research Working Papers No.: 4597.
Snieska, V., Bruneckiene, J. (2009), Measurement of Lithuanian Regions by Regional Competitiveness Index. Inzinerine Ekonomika-Engineering Economics(1), 45-57.
Straub, S., Vellutini, C., Warlters, M. (2008), Infrastructure and economic growth in East Asia, Policy Research Working Paper Series 4589, The World Bank.
The Global Competitiveness Report (2012 – 2013), World Economic Forum - Geneva Switzerland 2012
Yeaple, S.R., Golub S.S. (2007), International Productivity Differences, Infrastructure, and Comparative Advantage. Review of International Economics, Wiley Blackwell, vol. 15(2), pages 223-242, 05. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2320,
1
],
[
2320,
5943,
2
],
[
5943,
9142,
3
],
[
9142,
14330,
4
],
[
14330,
19175,
5
],
[
19175,
23514,
6
],
[
23514,
28671,
7
],
[
28671,
30964,
8
]
]
} |
85d529001feb0d13139122cc71cc5f5527c062a9 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 31,
"total-fallback-pages": 0,
"total-input-tokens": 101377,
"total-output-tokens": 23306,
"fieldofstudy": [
"Physics"
]
} | Application of a Spectral Method to Simulate Quasi-Three-Dimensional Underwater Acoustic Fields
Houwang Tu\textsuperscript{a}, Yongxian Wang\textsuperscript{b,*}, Wei Liu\textsuperscript{b}, Chunmei Yang\textsuperscript{c}, Jixing Qin\textsuperscript{d}, Shuqing Ma\textsuperscript{b}, Xiaodong Wang\textsuperscript{a}
\textsuperscript{a}College of Computer, National University of Defense Technology, Changsha, 410073, China
\textsuperscript{b}College of Meteorology and Oceanography, National University of Defense Technology, Changsha, 410073, China
\textsuperscript{c}Key Laboratory of Marine Science and Numerical Modeling, First Institute of Oceanography, Ministry of Natural Resources, Qingdao, 266061, China
\textsuperscript{d}State Key Laboratory of Acoustics, Institute of Acoustics, Chinese Academy of Sciences, Beijing, 100190, China
Abstract
The calculation of a three-dimensional underwater acoustic field has always been a key problem in computational ocean acoustics. Traditionally, this solution is usually obtained by directly solving the acoustic Helmholtz equation using a finite difference or finite element algorithm. Solving the three-dimensional Helmholtz equation directly is computationally expensive. For quasi-three-dimensional problems, the Helmholtz equation can be processed by the integral transformation approach, which can greatly reduce the computational cost. In this paper, a numerical algorithm for a quasi-three-dimensional sound field that combines an integral transformation technique, stepwise coupled modes and a spectral method is designed. The quasi-three-dimensional problem is transformed into a two-dimensional problem using an integral transformation strategy. A stepwise approximation is then used to discretize the range dependence of the two-dimensional problem; this approximation is essentially a physical discretization that further reduces the range-dependent two-dimensional problem to a one-dimensional problem. Finally, the Chebyshev–Tau spectral method is employed to accurately solve the one-dimensional problem. We provide the corresponding numerical program SPEC3D for the proposed algorithm and describe several representative numerical examples. In the numerical experiments, the consistency between SPEC3D and the analytical solution/high-precision finite difference program COACH verifies the reliability and capability of the proposed algorithm. A comparison of running times illustrates that the algorithm proposed in this paper is significantly faster than the full three-dimensional algorithm in terms of computational speed.
Keywords: Chebyshev–Tau spectral method, coupled modes, range-dependent, computational ocean acoustics
*Corresponding author: [email protected]
Email addresses: [email protected] (Houwang Tu), [email protected] (Wei Liu), [email protected] (Chunmei Yang), [email protected] (Jixing Qin)
1. Introduction
Accurately and efficiently identifying and locating underwater targets have always been the core pursuits of computational ocean acoustics [1]. The most critical step in achieving these objectives is accurately simulating the acoustic field [2]. The acquisition of an accurate three-dimensional numerical sound field is computationally expensive, as it involves a numerical solution of the three-dimensional Helmholtz equation [3–6]. A quasi-three-dimensional sound field (one of the two horizontal directions is range-independent) is also useful, as it is usually associated with continental slopes, continental shelves, or transitional seas.
Simplified solutions exist for quasi-three-dimensional sound fields that do not incur the high computational cost of full three-dimensional problems. Fawcett proposed using the integral transformation technique to transform the Helmholtz equation satisfied by the quasi-three-dimensional marine environment into a governing equation similar to the two-dimensional line source problem, which greatly reduces the amount of computation [7]. The two-dimensional line source problems can be solved accurately using the coupled modes. When using the step approximation proposed by Evans to address the range dependence, the two-dimensional line source problems become one-dimensional modal equations [8–10]. The modal equations need to be solved by discretization. The accuracy of the solution of the one-dimensional modal equations determines the accuracy of the quasi-three-dimensional sound field.
There are many numerical discretized methods in scientific and engineering computing, and the spectral method is a highly precise numerical technique similar to the finite difference method (FDM) and the finite element method (FEM) [11]. Compared with the FDM and FEM, the spectral method is a global approach, meaning its basis functions are defined in the entire solution domain, whereas the basis functions of the FDM and FEM are defined on the grid points and in the elements, respectively [12]. Strictly speaking, the spectral method is a weighted residual method, but the basis functions selected by the spectral method—unlike those of the general weighted residual method—are sets of orthogonal polynomials, and the orthogonality of these basis functions ensures that the spectral method converges quickly [13]. The research conducted by Orszag and Gottlieb illustrated that for smooth problems, the numerical error of the spectral method decreases exponentially with increasing truncation order [14, 15]. Consequently, benefiting from its high precision, the spectral method has been widely used in many numerical simulations of scientific and engineering problems [16–18].
Spectral methods have also been gradually applied to computational ocean acoustics in recent years. In 1993, Dzieciuch wrote a program to simulate the ideal fluid waveguide of the Munk sound speed profile using the Chebyshev–Tau spectral method [19–20]. However, this program could calculate only a single-layer water body in which the seawater density does not vary with depth and the seabed is a pressure-release boundary, and thus, it was not capable of solving higher-complexity acoustic propagation problems; nevertheless, this program represented the first use of the spectral method to solve the underwater acoustic propagation problem, and it demonstrated remarkable accuracy. More recently, in 2016, Evans proposed the Legendre–Galerkin spectral method to solve for the normal modes of underwater acoustic propagation in two-layer media and developed the program rimLG [21]. Because rimLG also requires the seabed to be a pressure-release boundary and applies Gaussian integration in each parameter layer, its computational speed slows when the acoustic profile data are dense, but it exhibits extremely high accuracy. Moreover, rimLG can simulate underwater acoustic waveguides with attenuation in the sediment, and the diversity of sound waveguides that can be synthesized is much greater than the capability of Dzieciuch’s program. Subsequently, in 2019, Sabatini proposed a multidomain spectral collocation method that can accurately calculate the normal modes in the open ocean [22]. This method can accurately compute the half-space boundary of the seafloor, although the size of the matrix eigenvalue problem is doubled. Since 2020, Tu et al. have performed a series of studies to design highly precise numerical algorithms based on the spectral method for solving underwater acoustic propagation problems. In March 2020, Tu et al. proposed an algorithm based on the Chebyshev–Tau spectral method to solve the underwater acoustic propagation problem in two-layer media [23, 24] and developed a corresponding simulation software named NM-CT [25]. Numerical simulations have shown that the accuracy of NM-CT is comparable to that of rimLG and that the numerical errors are all on the order of $10^{-14}$, which is much smaller than the order of $10^{-6}$ of the classic FDM; in addition, the computational speed is much faster than that of rimLG. Then, in June 2021, Tu et al. proposed an algorithm for simulating underwater acoustic propagation in multilayer media based on the Legendre collocation spectral method [26, 27]. The algorithm adopts an absorbing layer to simulate the acoustic half-space; in fact, there is a compromise between computational cost and computational accuracy. The application of an absorption layer does not double the matrix size but reduces the precision of eigenpairs. To date, the spectral method has shown remarkable
1. Introduction
Accurately and efficiently identifying and locating underwater targets have always been the core pursuits of computational ocean acoustics [1]. The most critical step in achieving these objectives is accurately simulating the acoustic field [2]. The acquisition of an accurate three-dimensional numerical sound field is computationally expensive, as it involves a numerical solution of the three-dimensional Helmholtz equation [3–6]. A quasi-three-dimensional sound field (one of the two horizontal directions is range-independent) is also useful, as it is usually associated with continental slopes, continental shelves, or transitional seas.
Simplified solutions exist for quasi-three-dimensional sound fields that do not incur the high computational cost of full three-dimensional problems. Fawcett proposed using the integral transformation technique to transform the Helmholtz equation satisfied by the quasi-three-dimensional marine environment into a governing equation similar to the two-dimensional line source problem, which greatly reduces the amount of computation [7]. The two-dimensional line source problems can be solved accurately using the coupled modes. When using the step approximation proposed by Evans to address the range dependence, the two-dimensional line source problems become one-dimensional modal equations [8–10]. The modal equations need to be solved by discretization. The accuracy of the solution of the one-dimensional modal equations determines the accuracy of the quasi-three-dimensional sound field.
There are many numerical discretized methods in scientific and engineering computing, and the spectral method is a highly precise numerical technique similar to the finite difference method (FDM) and the finite element method (FEM) [11]. Compared with the FDM and FEM, the spectral method is a global approach, meaning its basis functions are defined in the entire solution domain, whereas the basis functions of the FDM and FEM are defined on the grid points and in the elements, respectively [12]. Strictly speaking, the spectral method is a weighted residual method, but the basis functions selected by the spectral method—unlike those of the general weighted residual method—are sets of orthogonal polynomials, and the orthogonality of these basis functions ensures that the spectral method converges quickly [13]. The research conducted by Orszag and Gottlieb illustrated that for smooth problems, the numerical error of the spectral method decreases exponentially with increasing truncation order [14, 15]. Consequently, benefiting from its high precision, the spectral method has been widely used in many numerical simulations of scientific and engineering problems [16–18].
Spectral methods have also been gradually applied to computational ocean acoustics in recent years. In 1993, Dzieciuch wrote a program to simulate the ideal fluid waveguide of the Munk sound speed profile using the Chebyshev–Tau spectral method [19–20]. However, this program could calculate only a single-layer water body in which the seawater density does not vary with depth and the seabed is a pressure-release boundary, and thus, it was not capable of solving higher-complexity acoustic propagation problems; nevertheless, this program represented the first use of the spectral method to solve the underwater acoustic propagation problem, and it demonstrated remarkable accuracy. More recently, in 2016, Evans proposed the Legendre–Galerkin spectral method to solve for the normal modes of underwater acoustic propagation in two-layer media and developed the program rimLG [21]. Because rimLG also requires the seabed to be a pressure-release boundary and applies Gaussian integration in each parameter layer, its computational speed slows when the acoustic profile data are dense, but it exhibits extremely high accuracy. Moreover, rimLG can simulate underwater acoustic waveguides with attenuation in the sediment, and the diversity of sound waveguides that can be synthesized is much greater than the capability of Dzieciuch’s program. Subsequently, in 2019, Sabatini proposed a multidomain spectral collocation method that can accurately calculate the normal modes in the open ocean [22]. This method can accurately compute the half-space boundary of the seafloor, although the size of the matrix eigenvalue problem is doubled. Since 2020, Tu et al. have performed a series of studies to design highly precise numerical algorithms based on the spectral method for solving underwater acoustic propagation problems. In March 2020, Tu et al. proposed an algorithm based on the Chebyshev–Tau spectral method to solve the underwater acoustic propagation problem in two-layer media [23, 24] and developed a corresponding simulation software named NM-CT [25]. Numerical simulations have shown that the accuracy of NM-CT is comparable to that of rimLG and that the numerical errors are all on the order of $10^{-14}$, which is much smaller than the order of $10^{-6}$ of the classic FDM; in addition, the computational speed is much faster than that of rimLG. Then, in June 2021, Tu et al. proposed an algorithm for simulating underwater acoustic propagation in multilayer media based on the Legendre collocation spectral method [26, 27]. The algorithm adopts an absorbing layer to simulate the acoustic half-space; in fact, there is a compromise between computational cost and computational accuracy. The application of an absorption layer does not double the matrix size but reduces the precision of eigenpairs. To date, the spectral method has shown remarkable
success and potential in designing highly precise numerical algorithms for simulating underwater acoustic propagation. However, there has been no research on the application of spectral methods to simulate three-dimensional acoustic propagation. Therefore, the development of a high-precision quasi-three-dimensional underwater acoustic model based on spectral methods is a research topic worthy of gradual exploration.
In this paper, we used the spectral method combined with the integral transformation technique and stepwise coupled modes to develop a high-precision quasi-three-dimensional numerical algorithm and develop the corresponding numerical program. We have also carefully designed several representative numerical experiments to verify the accuracy and efficiency of the algorithm and program by comparing them with analytical solutions or high-precision full-three-dimensional models.
2. Mathematical Modeling
2.1. Quasi-three-dimensional waveguide
Figure 1: Schematic diagram of the quasi-three-dimensional waveguides.
Quasi-three-dimensional waveguides can model common typical scenarios, including continental slopes, continental shelves, and transitional seas, and their structures are shown in Fig. 1. Terrain and acoustic parameters are range-dependent variables in the $xoz$-plane; in other words, parameters such as the sound speed profile, density and attenuation, as well as terrain, vary across the plane. In contrast, in the $yoz$–plane, the acoustic parameters are $y$-independent. Thus, this scenario is not a real three-dimensional problem but rather a quasi-three-dimensional problem. In the quasi-three-dimensional waveguide here, the sound speed $c \equiv c(x, z)$, density $\rho \equiv \rho(x, z)$, and attenuation coefficient $\alpha \equiv \alpha(x, z)$. To accurately simulate this problem physically, we can always place the sound source in the plane of $y = 0$; thus, the governing equation of the problem can be written as [2]:
$$
\rho(x, z) \frac{\partial}{\partial x} \left[ \frac{1}{\rho(x, z)} \frac{\partial p}{\partial x} \right] + \frac{\partial^2 p}{\partial y^2} + \rho(x, z) \frac{\partial}{\partial z} \left[ \frac{1}{\rho(x, z)} \frac{\partial p}{\partial z} \right] + k^2(x, z)p = -\delta(x - x_s)\delta(y)\delta(z - z_s) \tag{1}
$$
where the harmonic point source is located at a horizontal range $x = x_s$ and a depth $z = z_s$, $k(x, z) = \frac{2\pi f}{c(x, z)}$ is the wavenumber, $f$ is the frequency of the sound source ($f$ arises from a temporal Fourier transform of the wave equation [2]), $p \equiv p(x, y, z)$, and $\delta(\cdot)$ is the Dirac function. Since the marine environment is $y$-independent, Eq. (1) can be
transformed using the following Fourier transformations \[7\]:
\[
\tilde{p}(x, k_y, z) = \int_{-\infty}^{\infty} p(x, y, z) e^{-i k_y y} \, dy \quad (2a)
\]
\[
p(x, y, z) = \frac{1}{2\pi} \int_{-\infty}^{\infty} \tilde{p}(x, k_y, z) e^{i k_y y} \, dk_y \quad (2b)
\]
This is accomplished by applying the following operators to both sides of Eq. (1):
\[
\int_{-\infty}^{\infty} (\cdot) e^{-i k_y y} \, dy
\]
From Eq. (2a), Eq. (1) can be transformed into the following form:
\[
\rho(x, z) \frac{\partial}{\partial x} \left[ \frac{1}{\rho(x, z)} \frac{\partial \tilde{p}}{\partial x} \right] + \rho(x, z) \frac{\partial}{\partial z} \left[ \frac{1}{\rho(x, z)} \frac{\partial \tilde{p}}{\partial z} \right] + \left( k_y^2 - k_s^2 \right) \tilde{p} = -\delta(x - x_s) \delta(z - z_s) \quad (3)
\]
For each particular \(k_y\), Eq. (3) has almost the same form as the governing equation for an infinitely long line source sound field in a two-dimensional plane. Eq. (3) is solved by subdividing the environment into segments along the \(x\)-axis; within each segment, the density and speed of sound are assumed to vary only with depth. Details are described in Sec. 2.2. After calculating the sound pressure \(\tilde{p}(x, k_y, z)\) in the \(k_y\)-domain, the sound pressure \(p(x, y, z)\) in the three-dimensional domain can be easily calculated through the inverse Fourier transform of Eq. (2b).
2.2. Stepwise approximation of a two-dimensional marine environment
The previous section demonstrates that Eq. (3) is highly similar in form to the governing equation of the sound field of an infinitely long line source in a two-dimensional plane. As shown in Fig. 2 for an \(x\)-dependent marine environment, the classic solution is to divide the environment into a sufficiently large number \(J\) of narrow segments along the \(x\)-direction [28]. In the \(j\)-th flat segment, the sound field is approximated as [2, 28]:
![Figure 2: Stepwise coupled modes of a line source [28].](image-url)
\[ \tilde{p}^j(x, z) \approx \sum_{m=1}^{M} \left[ a^j_m E^j_m(x) + b^j_m F^j_m(x) \right] \Psi^j_m(z) \]
\[ E^j_m(x) = \exp \left[ i \mu^j_m (x - x_{j-1}) \right] \]
\[ F^j_m(x) = \exp \left[ -i \mu^j_m (x - x_j) \right] \]
where \( j = 1, 2, \cdots, J \); \( M \) is the total number of normal modes needed to synthesize the sound field; \( \{a^j_m\}_{m=1}^{M} \) and \( \{b^j_m\}_{m=1}^{M} \) indicate the amplitudes of the forward and backward propagation modes, respectively, which are also called the coupling coefficients of the \( j \)-th segment [23]; \( E^j_m(x) \) and \( F^j_m(x) \) are the normalized range functions for special cases when \( j = 1, x_{j-1} = x_1 \); and \( \{\mu^j_m, \Psi^j_m(z)\} \) is the \( m \)-th eigensolution of the \( j \)-th segment and satisfies the following modal equation:
\[ \rho(z) \frac{d}{dz} \left[ \frac{1}{\rho(z)} \frac{d\Psi(z)}{dz} \right] + \left[ k^2(z) - k_0^2 - \mu^2 \right] \Psi(z) = 0 \]
\[ k(z) = 2 \pi f (1 + i \eta \alpha / c(z)), \quad \eta = (40 \pi \log_{10} e)^{-1} \]
We redefine \( k(z) \) in Eq. (1) to account for the effect of attenuation on acoustic propagation. Here, \( \mu^2 \), the constant obtained by the separation of variables, physically represents the horizontal wavenumber, and \( \alpha \) is the attenuation coefficient in units of dB/\( \lambda \), where \( \lambda \) is the wavelength.
Ignoring the contribution of the continuous spectrum and adding appropriate boundary conditions, Eq. (5) has a set of eigensolutions \( \{\mu_m, \Psi_m\}_{m=1}^{\infty} \), where \( \Psi_m \) is also called the eigenmode. The eigenmodes of Eq. (5) should be normalized as:
\[ \int_0^H \frac{\Psi_m^2(z)}{\rho(z)} dz = 1, \quad m = 1, 2, \ldots \]
or:
\[ \int_0^\infty \frac{\Psi_m^2(z)}{\rho(z)} dz = \int_0^H \frac{\Psi_m^2(z)}{\rho(z)} dz + \frac{\Psi_m^2(H)}{2 \rho_\infty \rho_\infty} = 1, \quad m = 1, 2, \ldots \]
\[ \rho_\infty = \sqrt{\mu^2 - k_\infty^2}, \quad k_\infty = 2 \pi f (1 + i \eta \alpha_\infty) / c_\infty \]
where \( H \) is the ocean depth. Eq. (6) applies to the case where the upper and lower boundaries are ideal boundaries, while Eq. (7) applies to the case where the upper boundary is an ideal boundary but the lower boundary is an acoustic half-space.
Segmenting the environment means that we solve the sound field in \( J \) segments. Subsequently, the complete sound field can be obtained by simply imposing the continuity of the pressure perturbation and the continuity of the horizontal velocity perturbation. The method of coupling segments explicitly imposes two continuity conditions on the sides of the segments. In other words, the coefficients \( a^j_m \) and \( b^j_m \) are computed by enforcing the continuity of pressure and the continuity of horizontal velocity at the interface between regular segments. The first segment condition requires that the acoustic pressure be continuous at the \( j \)-th side, and the second segment condition requires that the horizontal velocity be continuous at the \( j \)-th side:
\[ \frac{\dot{p}^{j+1}(x_j, \bar{z}) - \dot{p}^j(x_j, \bar{z})}{\rho_{j+1}(x_j, \bar{z})} = \frac{1}{\rho_j(x_j, \bar{z})} \frac{\partial \dot{p}^j(x_j, \bar{z})}{\partial x} \]
For the first segment condition, we have:
\[ \sum_{m=1}^{M} \left[ a^j_m E^j_m(x_j) + b^j_m F^j_m(x_j) \right] \Psi^j_m(\bar{z}) = \sum_{m=1}^{M} \left[ a^j_m E^j_m(x) + b^j_m F^j_m(x) \right] \Psi^j_m(\bar{z}) \]
Next, we apply the following operator to the above equation:
$$
\int \frac{\Psi'^{j+1}(z)}{\rho_{j+1}(z)} \, dz
$$
Furthermore, we apply the orthogonal normalization formula in Eq. (6) or (7) to the eigenmodes of the \((j+1)-th\) segment:
$$a'^{j+1} + b'^{j+1} F'^{j+1}(x_j) = \sum_{m=1}^{M} \left[ a^j_{m} E^j_m(x_j) + b^j_{m} \right] \hat{c}_{lm} \tag{10a}$$
Next, we apply the following operator to the above equation:
$$\hat{c}_{lm} = \int_{0}^{\infty} \frac{\Psi'^{j+1}(z) \Psi^j_m(z)}{\rho_{j+1}(z)} \, dz, \quad \ell = 1, 2, \ldots, M \tag{10b}$$
Yielding the following:
$$\hat{c}_{lm} = \int_{0}^{\infty} \frac{\Psi'^{j+1}(z) \Psi^j_m(z)}{\rho_{j+1}(z)} \, dz = \int_{0}^{\infty} \frac{\Psi'^{j+1}(z) \Psi^j_m(z)}{\rho_{j+1}(z)} \, dz + \frac{\Psi'^{j+1}(H) \Psi^j_m(H)}{\rho_{j+1}(z)} \tag{10c}$$
Next, we find the partial derivatives on both sides of Eq. (10a):
$$\frac{1}{\rho_j(z)} \frac{\partial \hat{p}(x, z)}{\partial x} = \frac{1}{\rho_j(z)} \sum_{m=1}^{M} \mu^j_m \left[ a^j_m E^j_m(x) - b^j_m \right] \Psi^j_m(z) \tag{11}$$
For the second segment condition, we have:
$$\frac{1}{\rho_{j+1}(z)} \sum_{m=1}^{M} \mu^j_m \left[ a^j_m E^j_m(x) - b^j_m \right] \Psi^j_m(z) = \frac{1}{\rho_{j+1}(z)} \sum_{m=1}^{M} \mu^j_m \left[ a^j_m E^j_m(x) - b^j_m \right] \Psi^j_m(z) \tag{12}$$
Similarly, we apply the following operator to the above equation:
$$\int \frac{\Psi'^{j+1}(z)}{\rho_{j+1}(z)} \, dz$$
Next, we apply the orthogonal normalization formula in Eq. (6) or (7) to the eigenmodes of the \((j+1)-th\) segment, yielding the following:
$$a'^{j+1} - b'^{j+1} F'^{j+1}(x_j) = \sum_{m=1}^{M} \left[ a^j_{m} E^j_m(x_j) - b^j_{m} \right] \hat{c}_{lm} \tag{13a}$$
Next, we apply the following operator to the above equation:
$$\hat{c}_{lm} = \frac{\mu^j_m}{\mu'^{j+1}_l} \int_{0}^{\infty} \frac{\Psi'^{j+1}(z) \Psi^j_m(z)}{\rho_{j+1}(z)} \, dz, \quad \ell = 1, 2, \ldots, M \tag{13b}$$
Yielding the following:
$$\hat{c}_{lm} = \frac{\mu^j_m}{\mu'^{j+1}_l} \int_{0}^{\infty} \frac{\Psi'^{j+1}(z) \Psi^j_m(z)}{\rho_{j+1}(z)} \, dz + \frac{\Psi'^{j+1}(H) \Psi^j_m(H)}{\rho_{j+1}(z)} \tag{13c}$$
The coupling integrals in Eqs. (10) and (13) can be obtained using general numerical integration. In the numerical model developed later in this paper, the coupling integrals are evaluated by the trapezoidal rule.
Then, the above Eqs. (10a) and (13a) can be naturally written in the following matrix-vector form:
$$a'^{j+1} - F'^{j+1} b'^{j+1} = \tilde{C}^j \left( \bar{E}' a' - b' \right) \tag{14a}$$
$$a'^{j+1} + F'^{j+1} b'^{j+1} = \tilde{C}^j \left( \bar{E}' a' + b' \right) \tag{14b}$$
and can be combined into the following form:
\[
\begin{bmatrix}
a^{i+1} \\
b^{i+1}
\end{bmatrix} = \begin{bmatrix}
R^i_1 & R^i_2 \\
R^i_3 & R^i_4
\end{bmatrix} \begin{bmatrix}
a^i \\
b^i
\end{bmatrix}
\]
(15a)
\[
R^i_1 = \frac{1}{2} (\hat{C}^i + \hat{C}^i) E^i
\]
(15b)
\[
R^i_2 = \frac{1}{2} (\hat{C}^i - \hat{C}^i)
\]
(15c)
\[
R^i_3 = \frac{1}{2} (F^{i+1})^{-1} (\hat{C}^i - \hat{C}^i) E^i
\]
(15d)
\[
R^i_4 = \frac{1}{2} (F^{i+1})^{-1} (\hat{C}^i + \hat{C}^i)
\]
(15e)
Note that since the sound source is located at \((x_s, z_s)\), a virtual boundary \(j_i\) needs to be introduced at the horizontal distance \(x_s\). The introduction of this virtual boundary at the sound source increases the number of segments to \(J + 1\). All boundaries except \(j_i\) still satisfy the boundary conditions of Eq. (5). However, due to the existence of the sound source, the boundary conditions at the virtual boundary \(j_i\) need to be modified. We can obtain the relationship between the coupling coefficients of the \(j_i\)-th and \((j_i + 1)\)-th segments \(23\):
\[
\begin{bmatrix}
a^{i+1} \\
b^{i+1}
\end{bmatrix} = \begin{bmatrix}
E^i & 0 \\
0 & (F^{i+1})^{-1}
\end{bmatrix} \begin{bmatrix}
a^i \\
b^i
\end{bmatrix} + \begin{bmatrix}
-\frac{s}{2}
\end{bmatrix}
\]
\[
s_m = -\frac{i}{\rho(z_s)} \frac{\Psi_m(z_s)}{\mu_m}, \quad m = 1, 2, \cdots, M
\]
(16)
Since the sound source is located at \(x_s \neq 0\), the segment conditions at \(x = 0\) and \(x = \infty\) are the radiation conditions, i.e., \(a^1 = 0\) and \(b^j = 0\). The global matrix used to determine the coupling coefficients can be assembled as follows:
\[
\begin{bmatrix}
R_1^1 & -I \\
R_2^1 & 0 & -I \\
R_3^1 & R_2^1 & -I & 0 \\
R_4^1 & R_3^1 & 0 & -I
\end{bmatrix}
\begin{bmatrix}
b^1 \\
a^2 \\
b^3 \\
a^4 \\
\vdots
\end{bmatrix}
\begin{bmatrix}
0 \\
a^2 \\
b^3 \\
a^4 \\
\vdots
\end{bmatrix}
\]
(17)
Note that the global matrix is a sparse band matrix of order \((2J - 2) \times M\); the upper bandwidth is \((2M - 1)\), and the lower bandwidth is \((3M - 1)\). The linear algebraic equations are then solved to obtain the coupling coefficients, and Eq. (4) is implemented to synthesize the sound pressure field of the entire waveguide.
Specifically, when \(x_s = 0\), the segment condition at the acoustic source \(x_s = 0\) is:
\[
a^i = s
\]
\[
s_m = \frac{i}{2\rho(z_s)} \Psi_m(z_s) \frac{e^{i_0 x_s}}{\mu_m}, \quad m = 1, 2, \cdots, M
\]
(18)
Therefore, the global matrix becomes:
\[
\begin{bmatrix}
I & 0 & 0 \\
R_1^{-1} & R_1^{-1} & -I & 0 \\
R_2^{-1} & R_2^{-1} & 0 & -I \\
\vdots & \vdots & \vdots & \vdots \\
R_{J-2}^{-1} & R_{J-2}^{-1} & -I & 0 \\
R_{J-1}^{-1} & R_{J-1}^{-1} & 0 & -I \\
\end{bmatrix}
\begin{bmatrix}
a_1 \\
b_1 \\
a_2 \\
\vdots \\
b_{J-2} \\
b_{J-1} \\
\end{bmatrix}
=
\begin{bmatrix}
s \\
a_1 \\
b_1 \\
\vdots \\
a_{J-2} \\
b_{J-1} \\
\end{bmatrix}
\]
(19)
3. Numerical Discretization
3.1. Range-independent marine environment
In the previous section, we fully derived the solution of a quasi-three-dimensional waveguide. Here, we solve for the eigensolutions in each segment, that is, the solutions \(\{\mu_j^m, \Psi_j^m\}_{m=1}^M\) of Eq. (5). For the sediment-covered marine environment considered in Figs. 1 and 2, within range-independent segments, \(\rho(z), c(z)\) and \(\alpha(z)\) are discontinuous at the interface \(z = h\); thus, the ocean is divided into a discontinuous water column and a bottom sediment. The marine environmental parameters are defined as follows:
\[
c(z) = \begin{cases}
c_w(z), & 0 \leq z \leq h \\
c_b(z), & h \leq z \leq H \\
c_\infty, & z \geq H
\end{cases}
\]
\[
\rho(z) = \begin{cases}
\rho_w(z), & 0 \leq z \leq h \\
\rho_b(z), & h \leq z \leq H \\
\rho_\infty, & z \geq H
\end{cases}
\]
\[
\alpha(z) = \begin{cases}
\alpha_w(z), & 0 \leq z \leq h \\
\alpha_b(z), & h \leq z \leq H \\
\alpha_\infty, & z \geq H
\end{cases}
\]
(20)
To solve Eq. (5), it is necessary to impose boundary conditions at the sea surface \((z = 0)\) and the seabed \((z = H)\) and interface conditions at the interface \((z = h)\). The sea surface is usually set as a pressure-release boundary:
\[
\Psi(z = 0) = 0
\]
(21)
In contrast, the seabed can be either a pressure-release boundary or a rigid seabed:
\[
\Psi(z = H) = 0
\]
\[
\Psi'(z = H) = 0
\]
(22a) (22b)
Furthermore, for an acoustic half-space, the lower boundary should generally satisfy the following [2]:
\[
\Psi(H) + \frac{\rho_\infty}{\rho_b(H)\gamma_\infty} \Psi'(H) = 0
\]
(23)
The interface conditions are defined as follows:
\[
\Psi(h^-) = \Psi(h^+)
\]
\[
\frac{1}{\rho(z = h^-)} \frac{d\Psi(h^-)}{dz} = \frac{1}{\rho(z = h^+)} \frac{d\Psi(h^+)}{dz}
\]
(24a) (24b)
where the superscripts \(h^-\) and \(h^+\) denote above and below \(h\), respectively.
3.2. Chebyshev–Tau spectral method
Traditionally, two numerical methods are used to solve for the local normal modes: the FDM, which is implemented by the KRAKEN/KRAKENC programs [29], and the Galerkin method, which is implemented by COUPLE [10]. The spectral method is a new type of numerical method known for its high precision that has been introduced into computational ocean acoustics in recent years. Generally, the spectral method includes both the Galerkin method (with orthogonal polynomials as the basis functions) and the collocation method (with the extreme points of orthogonal polynomials as the nodes). Numerical experiments have shown that the former method is slightly more accurate than the latter [12]. This paper employs the Chebyshev–Tau spectral method. Different from the classic Galerkin-type spectral method, the Chebyshev–Tau spectral method uses the original Chebyshev polynomial as the basis function and does not require the basis function to satisfy the boundary conditions but instead transforms the boundary conditions into the spectral space to enforce the requirement [30].
Here, we employ the Chebyshev–Tau spectral method to solve for the local modes (Eq. (5)) in the J range-independent segments illustrated in Fig. 2. Since the Chebyshev polynomial \(T_i(t)\), that is, the basis function, is defined in \(t \in [-1, 1]\), the equation to be solved, Eq. (5), must first be scaled to \(t \in [-1, 1]\) as:
\[
\frac{4}{\Delta h^2} \rho(t) \frac{d}{dt} \left( \frac{1}{\rho(t)} \frac{d\Psi(t)}{dt} \right) + \left[ k^2(t) - k^2_\text{c} \right] \Psi(t) = \mu^2 \Psi(t)
\]
(25)
where \(\Delta h\) denotes the thickness of the medium. The variable to be determined, \(\Psi(t)\), is transformed into the spectral space spanned by the basis functions \(\{T_i(t)\}_{i=1}^N\):
\[
\Psi(t) \approx \sum_{i=0}^N \hat{\Psi}_i T_i(t)
\]
(26)
where \(\{\hat{\Psi}_i\}_{i=1}^N\) represents the spectral coefficients of \(\Psi(t)\). This function approximation becomes increasingly accurate as \(N\) increases. Due to the advantageous properties of the Chebyshev polynomial, the following is easy to prove [13]:
\[
\hat{\Psi}' \approx \frac{2}{c_l} \sum_{j+i+1, j+i+i=\text{odd}}^N J \hat{\Psi}_j, \quad c_0 = 2, c_{i+1} = 1 \iff \hat{\Psi}' = \mathbf{D}_c \hat{\Psi}
\]
(27a)
\[
(\nabla \hat{\Psi})_i \approx \frac{1}{2} \sum_{m+n=i}^N \hat{\Psi}_m \hat{v}_n + \frac{1}{2} \sum_{m+n=i}^N \hat{\Psi}_m \hat{v}_n \iff (\nabla \hat{\Psi}) \approx \mathbf{C}_v \hat{\Psi}
\]
(27b)
where \(\hat{\Psi}\) is a column vector composed of \(\{\hat{\Psi}_i\}_{i=1}^N\), Eq. (27a) describes the relationship between the spectral coefficients of a function and the spectral coefficients of its derivative. Similarly, Eq. (27b) represents the relationship between the spectral coefficients of the product of two functions and the spectral coefficients of the individual functions, where the right-hand side is the matrix-vector representation of the relationship.
As shown in Eqs. (26) and (27), Eq. (5) can be discretized into the following matrix-vector form:
\[
\left( \frac{4}{\Delta h^2} \mathbf{C}_\rho \mathbf{D}_c \mathbf{C}_1/J \mathbf{D}_N + \mathbf{C}_{l^2} - k^2 c_\text{c} \right) \hat{\Psi} = \mu^2 \hat{\Psi}
\]
(28)
where \(\mathbf{I}\) is the identity matrix. The above equation is a matrix eigenvalue problem, but the boundary conditions are not considered at this time.
For the waveguide in Eq. (20), the modal equations (Eqs. (5) and (28)) must be established in both the water column and the bottom sediment. In the flat segments, a single Chebyshev polynomial approximation (continuous with continuous derivatives) cannot be used, as the pressure perturbation \(p\) is continuous at \(z = h\) but its vertical derivative is not. Thus, we apply the domain decomposition strategy [31] to Eq. (5) and split the domain interval into two subintervals:
\[
\Psi(z) = \begin{cases}
\Psi_\nu(z) = \Psi_\nu(t) \approx \sum_{i=0}^{N_d} \hat{\Psi}_\nu_i T_i(t_\nu), & t_\nu = -\frac{z}{h} + 1, \quad 0 \leq z \leq h \\
\Psi_\delta(z) = \Psi_\delta(t) \approx \sum_{i=0}^{N_b} \hat{\Psi}_\delta_i T_i(t_\delta), & t_\delta = -\frac{z}{h} + \frac{h_\delta}{h-\delta}, \quad h \leq z \leq H
\end{cases}
\]
(29)
where \( N_w \) and \( N_b \) are the spectral truncation orders in the water column and bottom sediment, respectively, and \( \{ \Psi_w \}_{i=0}^{N_w} \) and \( \{ \Psi_b \}_{i=0}^{N_b} \) are the spectral coefficients in these two layers. Similar to Eq. (25), the modal equations in the water column and bottom sediment can be discretized into matrix-vector form:
\[
A \hat{\Psi}_w = k_z^2 \hat{\Psi}_w, \quad A = \frac{4}{H^2} C_{\rho b} D_{N_b} C_{1/\rho_b} D_{N_w} + C_{\kappa^2} - k_z^2 I \tag{30a}
\]
\[
B \hat{\Psi}_b = k_z^2 \hat{\Psi}_b, \quad B = \frac{4}{(H-h)^2} C_{\rho w} D_{N_b} C_{1/\rho_w} D_{N_w} + C_{\kappa^2} - k_z^2 I \tag{30b}
\]
where \( A \) and \( B \) are square matrices of order \((N_w + 1)\) and \((N_b + 1)\), respectively, and \( \hat{\Psi}_w \) and \( \hat{\Psi}_b \) are column vectors composed of \( \{ \Psi_w \}_{i=0}^{N_w} \) and \( \{ \Psi_b \}_{i=0}^{N_b} \), respectively. Since the interface conditions are related to both the water column and the bottom sediment, Eqs. (30a) and (30b) can be solved simultaneously as follows:
\[
\begin{bmatrix}
A & 0 \\
B & \end{bmatrix}
\begin{bmatrix}
\hat{\Psi}_w \\
\hat{\Psi}_b \\
\end{bmatrix}
= \mu^2
\begin{bmatrix}
\hat{\Psi}_w \\
\hat{\Psi}_b \\
\end{bmatrix},
\quad E = \begin{bmatrix}
A & 0 \\
0 & B \\
\end{bmatrix}
\tag{31}
\]
The boundary conditions and interface conditions in Eqs. (21), (22) and (24) must also be expanded to the spectral space and explicitly added to Eq. (31). For details regarding the treatment of the boundary conditions and interface conditions, please see Eq. (36) in Ref. [24]. Rearranging and combining the modified rows by an elementary row transformation, Eq. (31) can be rewritten into the form of the following block matrix:
\[
\begin{bmatrix}
L_{11} & L_{12} \\
L_{21} & L_{22} \\
\end{bmatrix}
\begin{bmatrix}
\hat{\Psi}_1 \\
\hat{\Psi}_2 \\
\end{bmatrix}
= \mu^2
\begin{bmatrix}
\hat{\Psi}_1 \\
\hat{\Psi}_2 \\
\end{bmatrix},
\quad L = \begin{bmatrix}
L_{11} & L_{12} \\
L_{21} & L_{22} \\
\end{bmatrix}
\tag{32}
\]
where \( L_{11} \) is a square matrix of order \((N_w + N_b - 2)\) and \( L_{22} \) is a square matrix of order 4, \( \hat{\Psi}_1 = [\hat{\psi}_{w,0}, \hat{\psi}_{w,1}, \ldots, \hat{\psi}_{w,N_b-2}, \hat{\psi}_{b,0}, \hat{\psi}_{b,1}, \ldots, \hat{\psi}_{b,N_b-2}]^T \) and \( \hat{\Psi}_2 = [\hat{\psi}_{w,-N_b-1}, \hat{\psi}_{w,-N_b}, \hat{\psi}_{b,-N_b-1}, \hat{\psi}_{b,-N_b}]^T \). Solving this linear algebraic system yields the wavenumbers and spectral coefficients of the eigenmodes \((\mu, \Psi_w, \Psi_b)\). This matrix eigenvalue problem can be solved by a mature numerical software library, such as the zgeev() function in LAPACK [32].
For the acoustic half-space boundary condition in Eq. (23), since \( \gamma_{i\infty} \) contains the eigenvalue to be solved, \( \mu \), the elements of \( L \) on the left side of Eq. (32) also contain \( \mu \); thus, Eq. (32) is no longer an algebraic eigenvalue problem and can be solved iteratively only by a root-finding algorithm [22]. Since a prior estimate of \( \mu \) is usually not available, many of the existing numerical programs following similar principles fail to converge to a specific root in some cases. To avoid the same issue, we apply the method proposed by [22]: we use \( k_{z,\infty} = \sqrt{k_{\infty}^2 - \mu^2} \) to transform the modal equation and Eq. (23) as follows:
\[
\rho(z) \frac{d}{dz}\left( \frac{1}{\rho(z)} \frac{d\Psi}{dz} \right) + \left[ k_z^2(z) - k_{z,\infty}^2 \right] \Psi = 0 \tag{33a}
\]
\[
\frac{[\rho_{\infty}(\mu)]}{\rho_H(H)} \frac{d\Psi(H)}{dz} \bigg|_{z=H} + k_{z,\infty} \Psi(H) = 0 \tag{33b}
\]
In the Chebyshev–Tau spectral method, Eq. (33a) can be discretized into:
\[
\begin{bmatrix}
U + k_{z,\infty}^2 I \\
0 \\
\end{bmatrix} \hat{\Psi} = 0, \quad \hat{\Psi} = \begin{bmatrix}
1 \\
0 \\
0 \\
\end{bmatrix}
\tag{34}
\]
After adding the boundary condition in Eq. (33b) related to \( k_{z,\infty} \), Eq. (34) finally expresses the following polynomial eigenvalue problem:
\[
\begin{bmatrix}
U + k_{z,\infty} V + k_{z,\infty}^2 W \\
0 \\
\end{bmatrix} \hat{\Psi} = 0
\tag{35}
\]
\( \hat{U} \) in Eq. (35) is not identical to that in Eq. (34), as it has been modified by boundary conditions and interface conditions; nevertheless, we denote it as \( U \). This polynomial eigenvalue problem can be efficiently solved by transforming into a general matrix eigenvalue problem, although the matrices are twice as large:
\[
\begin{bmatrix}
-\nabla & -U \\
I & 0 \\
\end{bmatrix} \begin{bmatrix}
\hat{W} \\
\Psi \\
\end{bmatrix} = \begin{bmatrix}
0 \\
0 \\
\end{bmatrix}, \quad \hat{W} = \begin{bmatrix}
W \\
0 \\
I \\
\end{bmatrix}, \quad \Psi = \begin{bmatrix}
k_{z,\infty} \hat{\Psi} \\
\hat{\Psi} \\
\end{bmatrix}
\tag{36}
\]
Here, U, V, and W are all very sparse. Similarly, this generalized matrix eigenvalue problem can be solved by the zggev() function in LAPACK [32].
Using Eq. (26), the eigenvectors \( \hat{\Psi}_u \) and \( \hat{\Psi}_v \) can be easily transformed into \( \Psi_u(z), z \in [0, h] \) and \( \Psi_v(z), z \in [h, H] \). The vectors \( \Psi_u \) and \( \Psi_v \) are stacked into a single column vector to form \( \Psi \). Then, either Eq. (6) or (7) is used to normalize \( \Psi \), and finally, a set of eigenpairs \( \{ \mu, \Psi(z) \} \) is obtained.
### 3.3. Numerical sound field synthesis
According to the above analysis, after calculating \( \tilde{p}(x, k_y, z) \), the spatial sound field can be obtained by the inverse Fourier transform in Eq. (2b). However, \( \tilde{p}(x, k_y, z) \) has singularities along the real \( k_y \) axis (please see Sec. IC of Ref. [7]); therefore, we refer to the approach in Ref. [7, 33] and convert the integral of Eq. (2b) into a contour integral on the complex plane. The contour, as shown in Fig. 3, takes the following form:
\[
k_y(q) = q - i\varepsilon \tanh(\delta q), \quad -\infty < q < \infty
\]
(37)
![Figure 3: Complex integration contour to avoid singularities on the real-\(k_y\)-axis [33]; red dots indicate the eigenvalues of the two-dimensional problem.](image)
The integral in Eq. (2b) is transformed into the following form:
\[
p(x, y, z) = \frac{1}{\pi} \int_0^\infty \tilde{p}(x, k_y(q), z) \cos(y k_y(q)) \left| 1 - i\varepsilon \sech^2(\delta q) \right| dq
\]
(38)
Since the interval is infinite, integrating over this interval is impossible in an actual numerical evaluation. Luo et al. demonstrated that sufficiently accurate results can be obtained by setting the integration interval of \( q \) in the above formula as \([0, 1.5k_0]\), \( k_0 = 2\pi f / \min\{c_w(x, z), c_b(x, z)\} \). Here, \( q \) takes equidistant, discrete values in the interval; thus, the above formula can be transformed into a general summation problem [34].
### 4. Algorithm and Complexity
#### 4.1. Algorithm
Based on the above derivation, the proposed algorithm can be summarized as follows:
**Input:** Data from the marine environment and the program parameters.
**Output:** The quasi-three-dimensional sound pressure field.
1. Set the parameters.
- The parameters include the frequency \( f \), the truncation interval \( q \) of the Fourier integral (Eq. (35)), the number of sampling points \( N_q \) in the \( k_y \) wavenumber domain, the location of the source \( (x_s, z_s) \), the total depth of the ocean \( H \), the topography of the seabed, the number of acoustic profiles, and the specific information pertaining
to each group of acoustic profiles. In addition, the parameters should include the spectral truncation orders \((N_x, N_y)\), the horizontal and vertical resolutions \((\Delta x, \Delta z)\), and the number of modes \(M\) to be coupled (the choice of \(M\) is usually in the interval \(\left[ \frac{1}{\min(c_x, c_z)}, \frac{1}{\max(c_x, c_z)} \right] \)). Based on the cutoff frequency estimate for the normal modes. If the lower boundary is the upper interface of an acoustic half-space, the speed \(c_{\infty}\), density \(\rho_{\infty}\) and attenuation \(\alpha_{\infty}\) of the half-space must also be specified.
2. Segment the marine environment based on the seabed topography, frequency of the source and acoustic profiles. We assume that the \(x\)-dependence is divided into \(J\) segments. The larger \(J\) is, the higher the computational accuracy is, but this generates expensive computational overhead. Jensen confirmed that a strict segmentation criterion is \(\lambda x \leq \Delta x/4\), where \(\lambda = \min(c_x(x, z), c_z(x, z))\) [35].
3. Discretize the \(q\) wavenumber domain into \(N_q\) equidistant points and perform the calculation from the fourth to the seventh step for each \(k_y(q)\).
4. Apply the Chebyshév–Tau spectral method to solve for the eigenpairs \(\left(\{\mu_m^j, \Psi_m^j\}\right)_{m=1}^M\) of the \(J\) \(x\)-independent segments (see Sec. 3.2).
5. Calculate the coupling submatrices \(\{R^j_{j-1} \}, \{R^j_{j-1} \}, \{R^j_{j-1} \}, \{R^j_{j-1} \}\) and \(\{R^j_{j-1} \}\). The sound pressure subfields of the segments are calculated according to Eq. (4). The subfields of the \(J\) segments are then spliced to form the sound pressure field of the entire \(xoz\)-plane. The fourth to the seventh steps are naturally parallel because the solution of \(\tilde{p}(x, k_y(q), z)\) is independent for each \(k_y(q)\) (see Sec. 2.1).
8. Calculate the sound field \(\tilde{p}(x, y, z)\).
According to Eq. (18), the \(N_q\) discrete \(\tilde{p}(x, k_y(q), z)\) fields are accumulated in the following manner to obtain the spatial sound field \(\tilde{p}(x, y, z)\):
\[
\tilde{p}(x, y, z) = \frac{3k}{2n(N_q - 1)} \sum_{i=1}^{N_q} \tilde{p}(x, k_y(q), z) \cos\left(yk_x(q)\right) \times \left(1 - \frac{i}{4\pi e \text{sech}^2\left(\frac{(N_q - 1)k_y(q)}{9}\right)}\right)
\]
The above formula is equivalent to taking \(\frac{3k}{2n(N_q - 1)}\) for \(\Delta y\), \(\frac{3k}{2\pi e}\) for \(\varepsilon\) and \(\frac{1}{6N_q}\) for \(\delta\) in Eq. (38) [2, 33, 35].
4.2. Computational complexity
The computational cost of the above algorithm can be approximately divided into two parts: one is to solve Eq. (3), and the other is to inverse integral transform Eq. (3b), corresponding to the fourth to eighth steps of the above algorithm. Next, we analyze the algorithm step by step.
In step 4, when using the Chebyshév–Tau spectral method to solve the local modes, it is necessary to solve the eigenvalues of the \((N_x + N_y + 2)\)-order dense matrix. If the lower boundary is an acoustic half-space, the generalized matrix eigenvalues of the \(2(N_x + N_y + 2)\)-order matrix are solved. Let \(N = N_x + N_y + 2\); then, the computational complexity is \(O(N^3)\). \(N_x\) and \(N_y\) should be adapted to the complexity of the sound speed profiles. The complicated marine environment requires a larger spectral truncation order; generally speaking, \(N \geq 2M\). For each \(k_y\), the \(J\)-segment local modes need to be solved: \(J = \left[ \frac{4/\min(c_x, c_z)}{\max(c_x, c_z)} \right]\). Thus, the total computational complexity of the fourth step is \(O(N_y \times J \times N^3)\). In step 5, the process of forming the global matrix, the matrices \(\mathbf{C}^j\) and \(\mathbf{C}^j\) need to be computed. The integrals are calculated by the trapezoidal rule, and the calculation amount is related to the number of vertical discrete points, i.e., \(O(J \times \frac{N}{\Delta z})\). In addition, \(M\)-order matrix inverse and multiplication is performed, and the computational complexity is \(O(J \times M^3)\). Thus, the total computational complexity of the fifth step is \(O\left(N_y \times J \times \frac{N}{\Delta z}\right) + O\left(N_y \times J \times M^3\right)\). In step 6, the global matrix is a sparse band matrix of order \((2J - 2) \times M\); the upper bandwidth is \((2M - 1)\), and the lower bandwidth is \((3M - 1)\). The direct solution method of the banded linear equation system typically uses LU decomposition first and then solves the problem step by step. The total computational cost is \(O\left(12M^3 - 8M^2 - 2M \times (2J - 2)\right)\). The seventh step is essentially \(J\) matrix multiplication operations, and the calculation amount is \(O(N_y \times J \times \frac{N}{\Delta z} \times \frac{4M}{\Delta z})\). The inverse integral transform in Eq. (39) is essentially a matrix
multiplication with a computational complexity of $O \left( N_q \times \frac{H}{\Delta z} \times \frac{\Delta x}{\Delta z} \right)$. Therefore, the total computational complexity of this algorithm is approximately:
$$O(N_q \times J \times N^3) + O(N_q \times J \times M^3) + O\left(N_q \times J \times \frac{H}{\Delta z} \times \frac{\Delta x}{\Delta z}\right)$$
5. Numerical Simulation
Here, we present a program named SPEC3D that we developed based on the above algorithm, and we verify the accuracy of the algorithm through several numerical experiments. The transmission loss (TL) field is often used in actual displays to compare and analyze sound fields [2]. Therefore, to present the sound field results, the TL of the acoustic pressure is defined as $\text{TL} = -20 \log_{10}(|p|/|p_0|)$ in dB, where $p_0 = \exp(i k_0) / 4\pi$ is the acoustic pressure 1 m from the source and $k_0$ is the wavenumber of the medium at the location of the source.
For convenience, in SPEC3D, the spectral truncation orders of all the examples are set to $N_w = N_b = 20$; in an actual simulation, however, the user can arbitrarily specify the spectral truncation orders in the SPEC3D input file. In the following eight numerical experiments, the speed of sound in seawater is set to 1500.0 m/s, and the density of seawater is set to 1.0 g/cm$^3$; the speed of sound in the sediment is set to 1700.0 m/s, the density in sediment is set to 1.5 g/cm$^3$, and the attenuation coefficient in the sediment is set to 0.5 dB/λ; the sound speed in the acoustic half-space is set to 2000.0 m/s, the density is set to 2.0 g/cm$^3$, and the attenuation coefficient is set to 1.0 dB/λ.
5.1. Analytical example: a three-dimensional ideal fluid waveguide
Here, we use an ideal fluid waveguide, which is a highly simplified case in underwater acoustics, to validate SPEC3D. The structure of the three-dimensional ideal fluid waveguide is shown in Fig. 4. The frequency of the sound source is $f = 20$ Hz, and the source is located at $z_s = 36$ m. The number of coupled modes is $M = 2$. The number of discrete points in the $k_y$-domain is $N_q = 512$. Due to the horizontal independence of the marine environment, there is no coupling in the $x$-direction, so segmentation is not needed, but for the smooth operation of SPEC3D, $J$ is taken as 2. The horizontal independence also makes the sound field cylindrically symmetrical around the axis where the sound source is located. First, consider the case when the lower boundary is also the pressure-release boundary. Fig. 5 presents the sound field calculated by SPEC3D, featuring prominent cylindrical symmetry.
Figs. 6(a) and 6(b) depict slices of the sound field calculated by SPEC3D at two depths. According to the above analysis, the rings formed by the sound field on each slice should be strictly concentric, but the sound fields are not standard rings, so there are some errors. There is an analytical solution for an ideal fluid waveguide with a pressure-release boundary. The analytical solution for each plane of symmetry is:
$$p(x, z) = \frac{2\pi}{H} \sum_{n=1}^{\infty} \sin \left( k_{z,n} z_s \right) \sin \left( k_{x,n} x \right) H_1^{(1)}(k_{x,n} x)$$
$$k_{z,n} = \frac{n\pi}{H}, \quad k_{x,n} = \sqrt{k_0^2 - k_{z,n}^2}, \quad n = 1, 2, 3, \ldots$$
13
where \( H_0^{(1)}(\cdot) \) is the Hankel function. Figs. 6(c) and 6(d) show the analytical solution and the sound field calculated by SPEC3D, respectively, indicating that the two sound fields are reasonably consistent. Fig. 6(e) plots the TL curve along the \( x \)-direction at the depth of the sound source. The consistency between the SPEC3D solution and the analytical solution verifies the accuracy and reliability of SPEC3D.
When the lower boundary is perfectly rigid, the analytical solution of the ideal fluid waveguide is the same as Eq. (40), except that the vertical wavenumber becomes:
\[
\begin{align*}
k_{z,n} &= \left( n - \frac{1}{2} \right) \frac{\pi}{H}, \\
k_{x,n} &= \sqrt{k_0^2 - k_{z,n}^2}, \\
n &= 1, 2, 3, \ldots
\end{align*}
\]
The number of coupled modes is \( M = 3 \). Similarly, the results of Fig. 7 can lead to the same conclusion as Fig. 6.
5.2. Analytical example: an ideal wedge-shaped waveguide
Consider the ideal wedge-shaped waveguide shown in Fig. 8, which is a primary benchmark problem for range-dependent waveguides. Both the sea surface and the seabed are pressure-release boundaries, and the source is located at \((x_s, 0, z_s)\). Buckingham [37–39] developed an analytical solution to this benchmark problem, which was originally proposed and discussed at two consecutive conferences of the Acoustic Society of America (ASA). Here, we briefly introduce this problem. The velocity potential in the water body is:
\[
\begin{align*}
\Phi &= \frac{1}{\theta_0} \sum_{n=1}^{\infty} I_{\nu_n}(r, r_s, z) \sin(\nu_n \theta) \sin(\nu_n \theta_s) \\
I_{\nu_n}(r, r_s, z) &= i \int_0^\infty k_r \eta \exp(i \eta |z|) J_{\nu_n}(k_r r) J_{\nu_n}(k_r r_s) dk_r \\
\eta &= \sqrt{k^2 - k_r^2}, \\
\nu_n &= \frac{n \pi}{\theta_0}, \quad n = 1, 2, \ldots
\end{align*}
\]
where \( r \) and \( r_s \) are the distances from the receiver and source to the apex of the wedge, respectively; \( \theta \) and \( \theta_s \) are the angles measured from the apex to the depths of the receiver and source, respectively; \( \theta_0 \) is the wedge angle; \( k \) is the
Figure 6: Sound fields of the three-dimensional ideal fluid waveguide with a free bottom calculated by SPEC3D at depths of 36 m (a) and 80 m (b); analytical solution (c) and sound field calculated by SPEC3D (d) on the y–plane at y = 0 m; TL along the x-direction at the depth of the sound source (e).
Figure 7: Sound fields of the three-dimensional ideal fluid waveguide with rigid bottom calculated by SPEC3D at depths of 25 m (a) and 80 m (b); analytical solution (c) and sound field calculated by SPEC3D (d) on the y-plane at $y = 0$ m; TL along the x-direction at the depth of the sound source (e).
wavenumber of the seawater; and \( J_\nu(\cdot) \) is the Bessel function of order \( \nu_n \). The TL field is calculated by the following formula:
\[
\text{TL} = -20 \log_{10} \left| \frac{\Phi(r, r_s, \theta, \theta_s)}{\Phi_0(1)} \right| \quad (43a)
\]
\[
\Phi_0(r) = \frac{\exp(ikr)}{4\pi r} \quad (43b)
\]
Note that directly calculating the integral in Eq. (42b) may trigger a numerical overflow, which can be approximated by taking the Debye asymptotic expansion [40]. In this paper, we use the Gauss–Kronrod quadrature.
In this case, the frequency of the sound source is \( f = 25 \) Hz, and the source is located at \( x_s = 2000 \) m and \( z_s = 50 \) m. The wedge is \( x = 4000 \) m long in the horizontal direction and \( z = 200 \) m deep in the vertical direction. The number
of coupled modes is $M = 6, J = 267$. The number of discrete points in the $k_y$–domain is $N_q = 2048$. Fig. 9 shows the sound field of the three-dimensional wedge-shaped waveguide calculated by SPEC3D; the sound field exhibits prominent three-dimensional characteristics.
Next, we compare slices through the sound field with slices from the analytical solution in detail. As observed from the slices through the sound field in Fig. 10, the sound field calculated by SPEC3D effectively matches the analytical sound field. Thus, despite the small (typically less than 1 dB) error at long distances, SPEC3D can successfully reproduce the sound field of the waveguide.
The analytical solution for the three-dimensional wedge-shaped waveguide with a rigid bottom has the same form as Eq. (42), but:
$$\nu_n = \left(n - \frac{1}{2}\right) \frac{\pi}{\theta_0}, \quad n = 1, 2, \cdots$$
(44)
Fig. 11 displays a comparison between the results of SPEC3D and the analytical solution. The number of coupled
modes is $M = 7$, revealing clear agreement, which further confirms the accuracy and reliability of SPEC3D.
Figure 11: Sound fields of the three-dimensional wedge-shaped waveguide with a rigid bottom calculated by the analytical solution (a) and SPEC3D (b) at a depth of $z = 10$ m; sound fields calculated by the analytical solution (c) and SPEC3D (d) on the $y$-plane at $y = 1000$ m; TLs along the $x$-direction (e) and $y$-direction (f) at a depth of $z = 10$ m.
In addition to the ideal bottom, the penetrable bottom is a classic configuration of a wedge-shaped waveguide. Fig. 12 shows the sound field simulated by SPEC3D. The sound source is located at $x_s = 0$, $z_s = 100$ m, and the other configurations are similar to the above example. Since there is a homogeneous half-space under the penetrable bottom, SPEC3D is truncated from $H = 400$ m in the simulation, and a total of $M = 13$ modes are involved in the coupling. For a wedge-shaped waveguide with a penetrable bottom, there is only an exact solution constructed by the image source method. Deane and Buckingham proposed the image source solution to calculate the three-dimensional acoustic field in wedge-shaped sea water [41]. Yang et al. further improved the above three-dimensional image source solution, enabling it to calculate not only the sound field in seawater but also the sound field in the sediment [42]. Fig. 13 shows a comparison between slices of the sound field calculated by SPEC3D and the image source solution. Both Figs. 12 and 13 fully demonstrate the capability of SPEC3D in simulating this example.
5.3. Acoustic half-space
Here, we consider an example involving an acoustic half-space, as shown in Fig. 14 where the seafloor topography is also \( x \)-independent. In this case, the frequency of the sound source is \( f = 50 \) Hz, and the source is located at \( x_s = 0 \) m, \( z_s = 36 \) m, \( h = 50 \) m, and \( H = 100 \) m. The number of coupled modes is \( M = 6 \), and the number of discrete points in the \( k_y \)-domain is \( N_q = 512 \). Due to the horizontal independence of the marine environment, \( J \) is taken as 2. Fig. 15 approximately shows the three-dimensional sound field structure of the waveguide.
Figs. 16(a) and 16(b) present cross-sections of the sound field calculated by SPEC3D at depths of 36 m and 50 m, respectively. The sound field still clearly exhibits the symmetry of horizontally independent terrain, but a certain degree of error remains at different azimuth angles. Figs. 16(c) and 16(d) display the two-dimensional sound fields calculated by KRAKENC [29] and NM-CT [24, 25], respectively, and Fig. 16(e) plots a slice of the sound field computed by SPEC3D at \( y = 0 \) m; the simulation results of all three programs exhibit good agreement. Fig. 16(f) shows the line graph of TL in Figs. 16(c) to 16(e). The results of KRAKENC, NM-CT and SPEC3D are almost identical on the whole, although there are slight differences in the near field.
5.4. Ridge-shaped waveguide
Consider a ridge-shaped waveguide, as shown in Fig. 17. In the \( xo\z \)-plane, the expression for the seabed topography is given by the following equation:
\[
h(x) = \begin{cases}
50 - 25 \cos \left( \frac{\pi(x - 500)}{200} \right), & 300 < x < 700, \\
75, & \text{elsewhere},
\end{cases}
\]
The frequency of the sound source is \( f = 25 \) Hz, and the source is located at \( x_s = 0 \) m and \( z_s = 25 \) m. The ocean depth \( H = 100 \) m. Fig. 18 shows the three-dimensional sound field structure of the waveguide calculated by SPEC3D. Due to the flat terrain in both the near field and the far field, the sound field still exhibits a certain symmetry. The uplift of the ridge at 300–700 m induces prominent changes in the sound field within this region.
For the cases where there is no analytical solution, we use the results of the benchmark program ‘COACH’ developed by Liu et al. as a reference [6]. COACH is a three-dimensional finite difference program with fourth-order accuracy developed for the ocean acoustic Helmholtz equation, which can be used to address arbitrary bathymetry and provide more accurate benchmark solutions for other three-dimensional underwater acoustic approximation models. The derivatives in the acoustic Helmholtz equation are numerically discretized based on regular grids, and a perfectly matched layer is introduced to absorb unphysical reflections from the boundaries where Sommerfeld radiation conditions are deployed. Here, running COACH is equivalent to directly numerically solving Eq. (1).
Figure 13: Sound fields of the three-dimensional wedge-shaped waveguide with a penetrable bottom calculated by the image source method (a) and SPEC3D (b) at a depth of $z = 30$ m; sound fields calculated by the image source (c) and SPEC3D (d) on the $y$–plane at $y = 0$ m; TLs along the $x$-direction (e) and $y$-direction (f) at a depth of $z = 30$ m.
Figure 14: Schematic diagram of the three-dimensional horizontally independent waveguide with an acoustic half-space.
Figure 15: Sound field of the three-dimensional horizontally independent waveguide with an acoustic half-space calculated by SPEC3D.
Figure 16: Sound fields of the three-dimensional horizontally independent waveguide with an acoustic half-space calculated by SPEC3D at depths of $z = 36$ m (a) and $z = 50$ m (b); sound fields calculated by KRAKEN-C (c), NM-CT (d) and SPEC3D (e) on the $y$-plane at $y = 0$ m; TLs on the $y = 0$ m-plane along the $x$-direction at a depth of $z = 20$ m (f).
Figure 17: Schematic diagram of the three-dimensional ridge-shaped waveguide.
Figure 18: Sound field of the three-dimensional ridge-shaped waveguide calculated by SPEC3D.
Figure 19: Sound fields of the three-dimensional ridge-shaped waveguide calculated by COACH (a) and SPEC3D (b) at $z = 50$ m; TLs along the $x$-direction at a depth of $z = 50$ m (c).
Figs. 19(a) and 19(b) show the sound field slices calculated by COACH and SPEC3D at z = 50 m, respectively. The results of SPEC3D and COACH are in good agreement, and even in a finer comparison of the TL curves in Fig. 19(c), the errors of both are satisfactory. In addition, the backscattering and forward scattering generated by the uplifted topography of the ridge show how the ridge perturbs the near-field and far-field sound fields.
5.5. Trench-shaped waveguide
Figure 20: Schematic diagram of the three-dimensional trench-shaped waveguide.
In contrast to the previous example, Fig. 20 shows a marine environment with a trench-shaped waveguide. The seafloor topography is given by:
\[
h(x) = \begin{cases}
65 + 25 \cos \left(\frac{\pi(x-500)}{200}\right), & 300 < x < 700, \\
40, & \text{elsewhere},
\end{cases}
\] (46)
The remainder of the configuration is exactly the same as the above example. Fig. 21 shows the three-dimensional sound field structure of the waveguide calculated by SPEC3D. The horizontal independence of both the near field and the far field makes the sound field retain a certain symmetry. The occurrence of refraction in the trench causes the sound field to exhibit a notable three-dimensional effect, which further affects the near and far fields through forward scattering and backscattering.
Likewise, Figs. 22(a) and 22(b) plot sound field slices calculated by COACH and SPEC3D at z = 50 m, respectively. Fig. 22(c) displays the TL curves calculated by the two programs along the x direction at a depth of z = 50 m. In both slice and line plots, the results of SPEC3D are very similar to those of COACH. Therefore, we conclude that the algorithm proposed in this paper is reliable.
6. Discussion, Remarks and Conclusion
6.1. Discussion
1. The numerical experiments in the previous section verify the accuracy of the algorithm proposed in this paper and demonstrate the ability of SPEC3D to address quasi-three-dimensional underwater acoustic propagation problems. Here, we give the running times of SPEC3D and COACH at roughly the same accuracy. The results of SPEC3D and COACH are shown in Table 1. Since COACH is a parallel program, both programs are run
Figure 21: Sound field of the three-dimensional trench-shaped waveguide calculated by SPEC3D.
Table 1: Running times of the numerical examples (unit: core-hour).
| Example | SPEC3D | COACH |
|------------------|--------|--------|
| free wedge | 1.19 | >200.00|
| rigid wedge | 1.20 | >300.00|
| penetrable wedge | 44.62 | >5000.00|
| penetrable ridge | 0.06 | >3.00 |
| penetrable trench| 0.06 | >3.00 |
Figure 22: Sound fields of the three-dimensional trench-shaped waveguide calculated by COACH (a) and SPEC3D (b) at $z = 50$ m; TLs along the $x$-direction at a depth of $z = 50$ m (c).
on the Tianhe-2 supercomputer [43] during the timing period, both programs use the same compiler and nodes with the same hardware configuration, and the running time is measured in core-hour. In terms of computational speed, SPEC3D is much faster than the full three-dimensional model within approximately similar accuracy—almost two orders of magnitude faster. Therefore, SPEC3D can also be said to trade off efficiency by sacrificing generality.
2. According to the above derivation and analysis, we can obtain a quasi-three-dimensional sound field through an inverse Fourier transform as long as we solve the sound field in Fig. 2 corresponding to different \( k_y \). Note that \( k_y \) is a set of complex constants and that the marine environment parameters in Fig. 2 do not change for different \( k_y \), which means that the stepwise segmentation of the x-dependences in Fig. 2 also does not change. Let us re-examine Eq. (5), which becomes the matrix eigenvalue problem in Eq. (28) after spectral discretization. Eq. (28) is equivalent to the following matrix eigenvalue problem:
\[
\frac{4}{|\Delta h|^2} C_p D_N C_{1/\mu} D_N + C_{1c} \hat{\Psi} = (k_y^2 + \mu^2) \hat{\Psi}
\]
(47)
Notably, the above indicates that the sum of squares of \( k_y \) and \( \mu \) is always equal to the eigenvalue in the above equation and that the eigenvectors do not change with \( k_y \). Therefore, Eq. (47) needs to be solved only once after the derivation of Sec. 3.2 to obtain the eigenvalues of the \( \mu \) values corresponding to different \( k_y \). Since the eigenvectors are invariant, the coupling submatrix in Eq. (15) is also shared for different \( k_y \).
However, in fact, the above conjecture includes a notable misunderstanding: that is, for different \( k_y \), it is not Eq. (28) that needs to be solved but Eq. (32) or (35) (boundary conditions are imposed). The boundary and interface conditions in Eqs. (32) and (35) do not change with \( k_y \). Therefore, in the \( k_y \)-domain, \( N_q \) sound fields still need to be calculated strictly according to the derivation in Secs. 2.2 and 3.2.
6.2. Remarks
From the above analysis, we can directly summarize the following advantages of the algorithm and program developed in this article:
1. The two-dimensional problem formed after implementing the Fourier transform can be solved in parallel because the evaluation of \( \hat{p}(x, k_y(q), z) \) is independent. This computational efficiency makes this algorithm more efficient than similar algorithms.
2. Furthermore, the two-dimensional problem formed after implementing the Fourier transform can be solved efficiently by using the global matrix method. This model is considered to involve two-way coupled normal modes, and its accuracy is higher than that of a model with only one-way coupled modes.
3. The Chebyshev–Tau spectral method used for the single-segment solution can accurately simulate free, rigid and half-space seafloor boundaries, especially for a half-space, and the eigenvalue transformation technique used avoids the problem of missing roots in traditional iterative root-finding programs.
6.3. Conclusion
In this paper, we develop an algorithm that can solve for quasi-three-dimensional sound fields. The results of numerical simulations verify the reliability and efficiency of the model and code. In the algorithm, the Fourier transform is used to transform the equation governing quasi-three-dimensional acoustic propagation from the \( y \)-domain to the \( k_y \)-domain. After the sound pressure at each discrete \( k_y \) is obtained, the sound pressure in the \( y \)-domain can be synthesized by applying an inverse Fourier transformation; this process can also be regarded as \( k_y \) wavenumber integration to some extent. The governing equation in the \( k_y \)-domain adopts a similar form to that of a two-dimensional line source sound field. The stair-step approximation strategy is used to accommodate the \( x \)-dependence of the two-dimensional sound field, and the coupling coefficients are assembled by means of a global matrix, a favorable sparse band matrix that can be solved efficiently. In each \( x \)-independent segment, the Chebyshev–Tau spectral method is used to solve for the eigenpairs, and the proposed method achieves good accuracy, speed and robustness.
In terms of its application scope, this algorithm requires that the marine environment be independent in the \( y \) direction, which limits the practicality of SPEC3D to a certain extent. However, for marine environments such as continental slopes and continental shelves, \( y \)-independence provides an appropriate approximation, and SPEC3D greatly reduces computational costs compared to full three-dimensional models that require the use of more sophisticated techniques [5, 6].
Therefore, Eq. (10c) as an example to deduce the integral change process in Eqs. (10c) and (13c).
\[
\int_{0}^{\infty} \frac{\Psi_{\ell}^{j+1}(z) \Psi_{m}^{j}(z)}{\rho_{j+1}(z)} \, dz = \int_{0}^{H} \frac{\Psi_{\ell}^{j+1}(z) \Psi_{m}^{j}(z)}{\rho_{j+1}(z)} \, dz + \int_{H}^{\infty} \frac{\Psi_{\ell}^{j+1}(z) \Psi_{m}^{j}(z)}{\rho_{j+1}(z)} \, dz
\]
(48)
At the bottom of the semi-infinite space, the modes decay exponentially with increasing depth as follows:
\[
\Psi(z) = \Psi(H) \exp[-\gamma(z - H)], \quad z \geq H
\]
(49)
Therefore,
\[
\int_{H}^{\infty} \frac{\Psi_{\ell}^{j+1}(z) \Psi_{m}^{j}(z)}{\rho_{j+1}(z)} \, dz = \frac{1}{\rho_{\infty}} \int_{H}^{\infty} \Psi_{\ell}^{j+1}(H) \Psi_{m}^{j}(H) \exp \left[ -\left( \gamma_{\ell}^{j+1} + \gamma_{m}^{j} \right) (z - H) \right] \, dz
\]
(50)
Eq. (7) presents a simplified case \((j = j + 1\) and \(\gamma_{\ell} = \gamma_{m} = \gamma_{\infty}\)) in the range-independent case.
Acknowledgments
The authors thank Prof. Wenyu Luo from the Institute of Acoustics, Chinese Academy of Sciences, for his valuable guidance on the analytical solution of the sound field for the wedge-shaped waveguide.
This work was supported by the National Natural Science Foundation of China [grant number 61972406] and the National Key Research and Development Program of China [grant number 2016YFC1401800].
Appendix
Here, we take Eq. (10c) as an example to deduce the integral change process in Eqs. (10c) and (13c).
\[
\int_{0}^{\infty} \frac{\Psi_{\ell}^{j+1}(z) \Psi_{m}^{j}(z)}{\rho_{j+1}(z)} \, dz = \int_{0}^{H} \frac{\Psi_{\ell}^{j+1}(z) \Psi_{m}^{j}(z)}{\rho_{j+1}(z)} \, dz + \int_{H}^{\infty} \frac{\Psi_{\ell}^{j+1}(z) \Psi_{m}^{j}(z)}{\rho_{j+1}(z)} \, dz
\]
(48)
References
[1] P. C. Etter, Underwater Acoustic Modeling and Simulation, CRC Press, Boca Raton, USA, 2018. doi:10.1201/9781315166346
[2] F. B. Jensen, W. A. Kuperman, M. B. Porter, H. Schmidt, Computational Ocean Acoustics, Springer-Verlag, New York, 2011. doi:10.1007/978-1-4419-8678-8
[3] D. Lee, M. H. Schultz, W. L. Siegmann, Numerical Ocean Acoustic Propagation in Three Dimensions, World Scientific, 1995. doi:10.1121/1.5126013
[4] Y. T. Lin, M. B. Porter, F. Sturm, M. J. Isakson, C. S. Chiu, Introduction to the special issue on three-dimensional underwater acoustics, The Journal of the Acoustical Society of America 146 (3) (2019) 1855–1857.
[5] S. M. Ivansson, Coupled-mode field computations for media with locally reacting irregular boundaries, The Journal of the Acoustical Society of America 150 (4) (2021) 2985–2998. doi:10.1121/1.5126013
[6] W. Liu, L. Zhang, W. Wang, Y. Wang, S. Ma, X. Cheng, W. Xiao, A three-dimensional finite difference model for ocean acoustic propagation and benchmarking for topographic effects, The Journal of the Acoustical Society of America 150 (2) (2021) 1140–1156. doi:10.1121/10.0005853
[7] J. A. Fawcett, T. W. Dawson, Fourier synthesis of three-dimensional scattering in a two-dimensional oceanic waveguide using boundary integral equation methods, The Journal of the Acoustical Society of America 88 (4) (1990) 1913–1920. doi:10.1121/1.400214
[8] R. B. Evans, A coupled mode solution for acoustic propagation in a waveguide with stepwise depth variations of a penetrable bottom, The Journal of the Acoustical Society of America 74 (1983) 188–195. doi:10.1121/1.389707
[9] R. B. Evans, The decoupling of stepwise coupled modes, The Journal of the Acoustical Society of America 80 (1986) 1414–1418. doi:10.1121/1.394395
[10] R. B. Evans, COUPLE: A coupled normal-mode code (Fortran)(2007).
URL: https://oalib-acoustics.org/models-and-software/normal-modes/
| 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2893,
1
],
[
2893,
14106,
2
],
[
14106,
16812,
3
],
[
16812,
18809,
4
],
[
18809,
22255,
5
],
[
22255,
24872,
6
],
[
24872,
27362,
7
],
[
27362,
29727,
8
],
[
29727,
33974,
9
],
[
33974,
38766,
10
],
[
38766,
41458,
11
],
[
41458,
46328,
12
],
[
46328,
49635,
13
],
[
49635,
51763,
14
],
[
51763,
52064,
15
],
[
52064,
52366,
16
],
[
52366,
53162,
17
],
[
53162,
54534,
18
],
[
54534,
56129,
19
],
[
56129,
59098,
20
],
[
59098,
59452,
21
],
[
59452,
59704,
22
],
[
59704,
60063,
23
],
[
60063,
60235,
24
],
[
60235,
60419,
25
],
[
60419,
62619,
26
],
[
62619,
63057,
27
],
[
63057,
63242,
28
],
[
63242,
68078,
29
],
[
68078,
71719,
30
],
[
71719,
71719,
31
]
]
} |
f63d2674a72043df2efeafaef2f6509529005bb4 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 6,
"total-fallback-pages": 0,
"total-input-tokens": 20357,
"total-output-tokens": 7425,
"fieldofstudy": [
"Medicine"
]
} | Phenotypic Distribution and Cluster Analysis in Asthma Patients
Sakine Nazik Bahçeçoğlu¹, Haluk Türktaş²
¹ Division of Allergy and Clinical Immunology, Department of Chest Diseases, Erciyes University Faculty of Medicine, Kayseri, Turkey.
² Department of Chest Diseases, Gazi University School of Medicine, Ankara, Turkey.
ABSTRACT
Objective: Several diagnostic and treatment algorithms regarding asthma have been described in previous guidelines. Yet these descriptions fail at reflecting different phenotypes of asthma encountered in clinical practice. The purpose of this study is to retrospectively analyze the data of asthma patients that have presented to the outpatient clinic and to group the patients according to the pre-bronchodilator FEV1 value, post-bronchodilator FEV1 value, age of asthma onset while evaluating the common characteristics of the different clusters.
Methods: 246 patients that had been diagnosed with asthma and had complete data records were recruited for this study. These patients were categorized under five phenotypic clusters according to the three variables (pre-bronchodilator FEV1 value, post-bronchodilator FEV1 value, age of asthma onset) of the SARP (Severe Asthma Research Program) algorithm and were evaluated accordingly.
Results: Cluster 4 had the highest number of patients while Cluster 5 had the least number of patients within our study. Obesity and gastro-esophageal reflux was thought to be the reason behind the fixed obstruction seen in patients of Cluster 5. Multiple drug treatment regimens were also mostly used for patients in Cluster 5. This led us to think that Cluster 5 asthma was the most difficult group to obtain control. Unlike the SARP study, atopy was encountered the most in Cluster 2.
Conclusions: In conclusion, phenotypical distribution and cluster analysis using the pre-bronchodilator FEV1 value, post-bronchodilator FEV1 value and age of asthma onset is an easy and effective classification system that can both be used for the Turkish population and to set guidelines and strategies for treatment of difficult asthma cases according to different clusters.
Key Words: Asthma, cluster analysis, phenotypes
Received: 05.11.2020
Accepted: 11.27.2020
ÖZET
Amaç: Astım ile ilgili, rehberlerde fikir birliği bulunan tanı ve tedavi algoritmaları tanımlanmıştır. Ancak bu tanımlamaların hastalığın kliniğinde görülen farklı fenotiplerin hepsini birden yansıtmama mümkün olmamaktadır. Bu çalışmanın amacı; Gazi Üniversitesi Tıp Fakültesi (GÜTF) Göğüs Hastalıkları Polikliniğinde başvuran hastaların dosyalarının retrospektif incelenmesi ile pre-bronkodilatör FEV₁, post-bronkodilatör FEV₁, astım başlangıç yaşını kullanarak hastaların kümlere ayrılmalarını ve kümlerin ortak özelliklerini incelemektir.
Yöntem: GÜTF Göğüs Hastalıkları polikliniğine başvuran, astım tanıtı ve verileri mevcut olan 246 hasta çalışmaya alındı. Çalışmaya alınan hastalar Ağır Astım Araştırma Programı (Severe Asthma Research Program, SARP) algoritmasına göre pre-bronkodilatör FEV₁, post-bronkodilatör FEV₁ ve astım başlangıç yaşını kullanılarak beş fenotipik küme ayrılarak incelendi.
Bulgular: Çalışmamızda Küme 4 hasta sayısı en fazla olan en geniş küme, Küme 5 ise en az sayılan en az olan kümeler olarak sahip çıktı. Obesite ve gastro-özafageal reflü en sık Küme S’ti sahip çıktı. Eşlik eden obezite ve gastro-özafageal reflünün Küme S’teki sıfır olması, astım kontrolünün güç olduğunu düşündürmektedir. Kullanılan tedavilerin incelenmesinde birden fazla integrasyon kontrol etkisi olmasına rağmen sık Küme S’tının sahip çıktığı saptanmıştır. Bu durum Küme S’tin astım kontrolünün güçlü olduğunu düşündürmektedir. Apto ipe SARP çalışmasında farklı olarak en sık Küme Z’de izlenmiştir.
Sonuç: Pre-bronkodilatör FEV₁, post-bronkodilatör FEV₁ ve astım başlangıç yaşını kullanarak astım olan hastaların en çok uygulanabilecek olan fenotipik sınıflandırma ve kümeler analizi yöntemlerinin Türk populasyonunda da uygulanabileceğini ve astım kontrolünün güçlenmesi için etkilerini artırmayacağını düşündürülmekte ve bu yöntemleri yönlendirebileceği düşünülmektedir.
INTRODUCTION
Asthma is a reversible chronic inflammatory disease of the airways. The symptoms particularly occur at night and in the morning. These symptoms are due to diffuse airway obstruction of varying degree (1).
Various diagnostic and therapeutic algorithms have been defined in various guidelines on asthma. However, it is not possible for these definitions to cover all the different phenotypes. A phenotype is defined as "the characteristics that occur in the external appearance of the organism after the interaction of its genetic properties with the environment". Various asthma phenotypes have been identified and all these different phenotypes have been grouped under the “asthma” definition over time. Phenotypic evaluations have accelerated in asthma patients in recent years as different phenotypes have different treatment strategies and different responses to treatment. Asthma has been divided into three groups by the first few phenotypic classifications as clinical and physiological phenotypes, trigger-based phenotypes and inflammatory phenotypes (2).
Many phenotypic cluster analyses related to asthma have been performed in recent years. The first few cluster analysis studies have evaluated various treatment strategies for the inflammatory asthma phenotypes (3).
The 2009 Severe Asthma Research Program (SARP) study has divided 726 severe asthma patients into five clusters by using 34 variables and shown that the patients could also be divided into phenotypic clusters by using only three variables in the form of pre-bronchodilator FEV1(forced expiratory volume in 1 second), post-bronchodilator FEV1, and asthma onset age (4).
The aim of our study was to divide the patients into five phenotypic clusters with the SARP algorithm by using three variables (pre-bronchodilator FEV1 value, post-bronchodilator FEV1 value, asthma onset age), reveal the characteristics of the clusters, and evaluate the applicability of the SARP algorithm to the Turkish population.
MATERIAL and METHOD
We retrospectively reviewed the charts of the patients who presented to the Gazi University School of Medicine's Chest Diseases outpatient department between 1995 and 2013 in this study, which was evaluated and approved by the Gazi University Rectorship Clinical Research Ethics Committee under the code number of G.Ü.ET- 2013-89.
The age, gender, smoking history, body mass index, age of onset of asthma, pre-bronchodilator FEV1 value, post-bronchodilator FEV1 value, total immunoglobulin E(IgE) value, skin tests, blood eosinophil percentage, comorbid diseases (rhinitis, sinusitis, gastroesophageal reflux, hypertension) and the treatment data of the 246 patients included in the study were recorded. The patients were divided into five phenotypic clusters according to the SARP where three variables (pre-bronchodilator FEV1, post-bronchodilator FEV1, asthma onset age) were used (figure 1).
Cluster 1: Patients with a pre-bronchodilator FEV1 value of 68% or more and a post-bronchodilator FEV1 value of 108% or more were included in this group.
Cluster 2: Patients with a pre-bronchodilator FEV1 value of 68% or more and a post-bronchodilator FEV1 value of less than 108% with an asthma onset age under 40 years were included in this group.
Cluster 3: Patients with a pre-bronchodilator FEV1 value of 68% or more and a post-bronchodilator FEV1 value of less than 108% with an asthma onset age under 40 years were included in this group.
Cluster 4: Patients with a pre-bronchodilator FEV1 value of less than 68% and a post-bronchodilator FEV1 of 65% or above were included in this group.
Cluster 5: Patients with a pre-bronchodilator FEV1 value of less than 68% and a post-bronchodilator FEV1 of 65% or above were included in this group.
The files of 1621 patients who presented to the Gazi University School of Medicine’s Chest Diseases Outpatient Department and were diagnosed with asthma between 1995 and 2013 were retrospectively reviewed and 246 patients with recorded data were included in the study. Data for the body mass index (BMI), pre-bronchodilator FEV1, post-bronchodilator FEV1, total immunoglobulin E (IgE), serum eosinophil percentage, drugs used for asthma treatment and the presence of any allergic rhinitis, sinusitis, gastro-esophageal reflux (GER), hypertension or heart failure accompanying the asthma were recorded.
RESULTS
The files of asthma patients who presented to the Gazi University School of Medicine's Chest Diseases outpatient department between 1995 and 2013 were retrospectively reviewed and 246 patients with complete data were included in the study.
The study group consisted of 179 (72.8%) females and 67 (27.2%) males. Nonsmokers made up 68.3% of the group with 168 subjects. The mean age was 42.5 ± 12.6 years and the mean asthma onset age was 34.1 ± 11.8 years. The mean body mass index was 28.8 ± 5.8 kg/m² (Table 1). The demographic data characteristics according to clusters are summarized in Table 2. Evaluation of the comorbidities revealed the most common comorbidity to be allergic rhinitis (n = 126, 51.2%). In addition, the asthma was associated with sinusitis in 115 (46.7%) patients, GER in 62 (25.2%) patients, hypertension in 40 (16.3%) patients and heart failure in 12 (4.9%) patients (Figure 2).
Table 1. The demographic data of the patients
| Demographic Data | General distribution |
|---------------------------|----------------------|
| Gender | N | % |
| Female | 179 | 72.8% |
| Male | 67 | 27.2% |
| Smoking status | | |
| Smoker | 78 | 31.7% |
| Non-smoker | 168 | 68.3% |
| Age | | |
| Mean | 42.5 years | |
| Minimum | 17 years | |
| Maximum | 83 years | |
| Asthma onset age | | |
| Mean | 34.1 years | |
| Minimum | 10 years | |
| Maximum | 82 years | |
Table 1. The demographic data of the patients
| Demographic Data | General distribution |
|---------------------------|----------------------|
| Body mass index | | |
| Mean | 28.8 kg/m² | |
| Minimum | 16.9 kg/m² | |
| Maximum | 56.4 kg/m² | |
The type of control drugs of the patients was reviewed and an inhaler corticosteroid (ICS) + long-acting β2 agonist (LABA) combination was the most commonly used type of control drug (n = 145, 58.9%). None of the patients used only anti-IgE drugs or leukotriene receptor antagonists (LTRA).
The patients included in the study were divided into five clusters using three variables (pre-bronchodilator FEV1, post-bronchodilator FEV1 and asthma onset age) with the SARP algorithm. Cluster 1 consisted of patients with a pre-bronchodilator FEV1 value of 68% or more and a post-bronchodilator FEV1 value of less than 68%, and an asthma onset age under 40 years; Cluster 3 consisted of patients with a pre-bronchodilator FEV1 value of less than 68% and a post-bronchodilator FEV1 value of less than 108%, and an asthma onset age of 40 years or more; Cluster 4 consisted of patients with a pre-bronchodilator FEV1 value of 68% or more, a post-bronchodilator FEV1 value of less than 108%, and an asthma onset age of 40 years or more; Cluster 2 consisted of patients with a pre-bronchodilator FEV1 value of less than 68% and a post-bronchodilator FEV1 value of less than 108%, and an asthma onset age of 40 years or more; Cluster 5 consisted of patients with a pre-bronchodilator FEV1 value of less than 108%, and an asthma onset age of 40 years or more; Cluster 1 consisted of patients with a pre-bronchodilator FEV1 value of 68% or more, a post-bronchodilator FEV1 value of less than 108%, and an asthma onset age of 40 years or more.
Patients with pulmonary function test (PFT) values within normal limits were included in this cluster. Females made up the majority. The cluster included patients with asthma onset at an advanced age (mean 35.3 ± 10.5 years) and a shorter duration of asthma (mean 6 years). Rhinitis was the most common comorbidity (p<0.05).
Table 2. Demographic Data Characteristics According to Clusters
| | Cluster 1 | Cluster 2 | Cluster 3 | Cluster 4 | Cluster 5 |
|-------------------------|-----------|-----------|-----------|-----------|-----------|
| n,% | 65,26.4% | 50,20.3% | 49,19.9% | 66,26.8% | 16, 6.5% |
| Age (years) | 41,4±11.7 | 31,6±7.7 | 48.7±8.2 | 44.2±13.1 | 54.9±12.8 |
| Asthma onset age | 35.3±10.5 | 26±2±2.9 | 39.5±11.1 | 34.9±12.9 | 34.2±15.3 |
| BMI (kg/m²) | 28.1±5.6 | 28.3±5.7 | 29.2±5.3 | 28.8±5.0 | 32.6±9.5 |
| Hypertension (n,% | 11, 16.9% | 0, 0% | 13, 26.5% | 10, 15.2% | 6, 37.5% |
| GER (n,% | 15,23.1% | 7, 14% | 11,22.4% | 19, 28.8% | 10, 61.5% |
| Heart Failure (n,% | 4, 6.2% | 0, 0% | 4, 8.2% | 2, 3% | 2, 12.5% |
| Allergic rhinitis(n,% | 39, 60% | 29, 58% | 26, 53.1% | 24, 36.4% | 8, 50% |
| ICS (n,% | 26, 40% | 14, 28% | 14, 28.6% | 11, 16.7% | 1, 6.3% |
| ICS+LABA (n,% | 38, 58.5% | 33, 66% | 32, 65.3% | 35, 53% | 7, 43.8% |
| ICS+LABA+LTRA (n,%) | 1, 1.5% | 3, 6% | 2, 4.1% | 16, 24.2% | 8, 50% |
There was no statistically significant difference between the clusters for sinusitis distribution (p>0.05). Sinusitis was most commonly observed in Cluster 2 (n=31, 62%). GER was most commonly found in Cluster 5 (n=10, 61.5%) and statistical analysis revealed that this difference was significant (p=0.003). GER was least common in Cluster 2 (n=7, 14%). Hypertension was most commonly found in Cluster 5 (n=6, 37.5%) and this difference in prevalence was found to be statistically significant (p=0.001). No hypertension was found in the asthma patients in Cluster 2. There was no statistically significant difference between the clusters regarding the distribution of heart failure (p=0.05) There was also no statistically significant difference between the clusters for BMI (p>0.05). The mean peripheral blood eosinophil percentages were within normal limits in all the clusters. Total IgE value was found to be above the normal limits (0-100 IU/mL) in all clusters.
The patients were divided into groups according to the number of positive allergens in the skin test and the distribution of these groups by cluster was investigated. A status of no allergen positivity on the skin test was most commonly found in Cluster 5 (n=12, 76.8%) and statistical analysis revealed that this difference was significant (p=0.003). No allergen positivity on the skin test was most commonly found in Cluster 2 (n=30, 60%).
The distribution of the treatments used by the patients was also evaluated by cluster. There was no patient using an anti-IgE drug or LTRA alone in the study. Multiple control drugs were most commonly used in Cluster 5 (n=15, 93.8%) (p=0.0001). Only one patient in Cluster 5 was using a single control drug type (ICS).
Our patients were divided into clusters according to the three basic criteria (pre-bronchodilator FEV1 value, post-bronchodilator FEV1 value and asthma onset age) recommended by the SARP, and cluster analysis revealed the following information:
Cluster 1: Patients with pulmonary function test (PFT) values within normal limits were included in this cluster. Females made up the majority. The cluster included patients with asthma onset at an advanced age (mean 35.3 ± 10.5 years) and a shorter duration of asthma (mean 6 years). Rhinitis was the most common comorbidity (p<0.05).
Cluster 2: Patients with a post-bronchodilator FEV₁ value lower than Cluster 1 on PFT and an asthma onset age under 40 years were included in this cluster. Cluster 2 had a majority of females again together with a shorter duration of asthma (mean 5 years). It was also found to be the cluster with the youngest patients (mean age 31.6±7.7 years).
Skin test positivity accompanied by increased total IgE was also most commonly found in Cluster 2. Sinusitis was the most common comorbidity (p<0.05).
Cluster 3: Patients whose PFT values were similar to Cluster 2 but where the asthma onset age was 40 years or more were included in this cluster. The cluster consisted of older patients (mean age 48.7 ± 8.2 years) with a majority of females and comorbidities were not common.
Cluster 4: Patients whose PFT values were lower than in the first three clusters but had no fixed obstruction were included in this cluster. Cluster 4 was found to contain the highest number of patients (n=66, 26.8%). This cluster consisted of mostly female patients where comorbidities were not common and the PFT values were low, unlike Cluster 3.
Cluster 5: Patients who were found to have fixed obstruction on PFT were included in this cluster. Cluster 5 consisted of an equal proportion of females and males with asthma onset at an advanced age (mean age 54±12.8 years) and the longest mean duration of asthma (10 years). GER (p=0.003) and hypertension (p=0.001) was most commonly observed in Cluster 5. The highest mean BMI value was also found in this cluster (mean 32.6±9.5 kg/m²) (p<0.05). In addition, the highest number of control drug types were used in Cluster 5 (p<0.05) and a negative skin test was also most commonly found here (p<0.05).
DISCUSSION
The idea of treatment according to phenotype has emerged in recent years and resulted in more detailed phenotyping studies and various phenotypic classifications. The aim of these classifications is to predict the common prognostic evaluation groups with similar clinical findings in addition to the treatment protocol and response to treatment (2).
Several cluster analysis studies have been conducted with such phenotypic evaluations in recent years. Cluster analysis is concerned with "identifying similarities according to a large number of variables predetermined in a population".
Asthma patients were evaluated after dividing them into five different clusters by using the 34 variables in the SARP study conducted with a large patient population consisting of 726 participants (4). Despite the 34 variables used in that study, we found that it was possible to include 80% of the subjects in the correct cluster by using only three variables: pre-bronchodilator FEV₁, post-bronchodilator FEV₁, and asthma onset age. According to the analysis based on the SARP, Cluster 1 consisted of young, atopic, female asthmatics with asthma onset at an early age and 40% of these patients were not using medication. Cluster 2 consisted of patients who were atopic and mostly female, who were somewhat older than in Cluster 1, and with earlier onset of asthma and normal PFT values. No medication was being used by 26% of the patients in Cluster 2. Cluster 3 consisted of obese female patients of an advanced age with asthma onset at a young age, low degree of atopy, and moderate obstruction on PFT. Cluster 4 consisted of an equal proportion of females and males with asthma onset at an early age, most of whom were atopic and required high medication doses. Cluster 5 consisted of obese female patients with asthma onset at an advanced age and who suffered from atopy less commonly. According to the distribution by SARP category, Cluster 2 had the highest and Cluster 3 the lowest number of patients.
The aim of our study was to divide the study subjects into clusters using the three variables and evaluate their characteristics as in the SARP study. It is known that allergic rhinitis is present in 75% of asthma cases and conversely asthma is present in 10-40% of allergic rhinitis cases (5). The physician should be aware that asthma and allergic rhinitis affect each other’s course negatively and the treatment should take this into account (1,5). Sinusitis is also frequently concurrent with asthma and can make its control difficult (5). The patients were evaluated under “sinus diseases” in the SARP study. Accordingly, the incidence of sinus diseases in our entire patient population was found to be 45% and these were most commonly observed in Cluster 3. Allergic rhinitis was seen in 51.2% and sinusitis in 46.7% of all patients in our study. No statistical difference was found between the clusters in terms of allergic rhinitis and sinusitis.
However, allergic rhinitis was most commonly seen in Clusters 1 and 2, and sinusitis in Cluster 2. Based on our patient population, we believe that allergic rhinitis and sinusitis should especially be investigated in patients with the phenotypic characteristics of Clusters 1 and 2.
Asthma is more common and more difficult to control in obese patients (body mass index > 30 kg/m²) (6-8). Asthma control can be provided more easily in obese patients who lose weight thanks to the relevant decrease in the accompanying reflux and elimination of the negative effects on lung mechanics (5). The review of the treatments used by the patients in Cluster 5, which was the cluster where obesity was most common in our study, revealed this cluster to most commonly include patients requiring multiple types of control drugs.
Weight loss could have positive effects on asthma in this cluster where asthma control could have been difficult judging by the variety of drugs used.
GER has been reported to aggravate asthma findings (9). The disorder may play a role in patients whose asthma is difficult to treat and cannot be controlled. The asthma can be controlled and the quality of life improved by adding a proton pump inhibitor to the treatment in these patients (9-12). Similar to the SARP study, gastro-esophageal reflux was most commonly found in Cluster 5 in our study. GER and obesity associated with asthma could therefore be responsible for the fixed obstruction in Cluster 5 and make asthma control difficult. When the treatments used were investigated, it was found that multiple control drug types were most commonly used in Cluster 5, suggesting that asthma was difficult to control in this group. The addition of proton pump inhibitors to the treatment could possibly improve asthma control in this group of patients.
Unlike the SARP study, the skin test negativity rate was found to be high in Cluster 5 in our study. However, similar to the SARP study, total IgE values were found to be high in Cluster 2, and skin test positivity was also most commonly seen in Cluster 2 in our study. Atopy was therefore thought to be common in Cluster 2.
Hypertension was most commonly found in Cluster 5 in this study, similar to the SARP study. We believe that hypertension occurring most commonly in Cluster 5 may be related to the older age of Cluster 5 patients. Evaluation of the comorbidities revealed that obesity and reflux were mostly seen in Cluster 5 with fixed obstruction. These two factors were thought to make it difficult to control asthma and may be the reason multiple types of control drugs were required in the relevant cluster in this study where phenotypic evaluation and cluster analysis were investigated by using pre-bronchodilator FEV₁, post-bronchodilator FEV₁ and asthma onset age in patients with asthma. Allergic rhinitis and sinusitis were found to be particularly common in Clusters 1 and 2.
CONCLUSION
We believe that the phenotypic classification and cluster analysis methods that can be applied easily using the criteria of pre-bronchodilator FEV₁, post-bronchodilator FEV₁ and asthma onset age values can also be useful in the Turkish population and the factors that make it difficult to control asthma can be predicted by cluster to guide the treatment strategy. This issue should be further clarified by future studies and increasing clinical experience.
Conflict of interest
No conflict of interest was declared by the authors.
REFERENCES
1. Global Initiative for Asthma. Global Strategy for Asthma Management and Prevention. 2019. Available from: www.ginasthma.org.
2. Wenzel SE. Asthma: Defining of the persistant adult phenotypes. Lancet 2006; 26: 368: 804-813
3. Haldar P, Pavord ID, Shaw DE, Berry MA, Thomas M, Brightling CE, et al. Cluster analysis and clinical asthma phenotypes. Am J Respir Crit Care Med 2008; 178(3): 218-224.
4. Moore WC, Meyers DA, Wenzel SE et al. Identification of asthma phenotypes using cluster analyses in the asthma research program Am J Respir Crit Care Med 2010; 181: 315-323
5. Bousquet J, Khaltaev N, Cruz AA et al. ZarAllergic Rhinitis and its Impact on Asthma (ARIA) 2008 Update, in collaboration with the World Health Organization, GA(2)LEN and AllerGen) Allergy 2008: 63 (Suppl. 86): 8-160.
6. Lavoie KL, Bacon SL, Labrecque M, Cartier A, Ditto B. High BMI is associated with worse asthma control and quality of life but not asthma severity. Respir Med 2006; 100(4): 648-657.
7. Pakhale S, Doucette S, Vandemheen K, Boulet LP, McIvor R Fitzgerald JM. A comparison of obese and non-obese people with asthma: exploring an asthma-obesity interaction Chest 2010; 137(6): 1316-1323.
8. Weiss ST, Shore S. Obesity and asthma: directions for research. Am J Respir Crit Care Med 2004; 169(8): 963-968.
9. Expert Panel Report 3 (EPR–3): Guidelines for the diagnosis and management of asthma-Full Report 2007, J Allergy Clin Immunol 2007; 120: 94–138.
10. Gibson PG, Henry RL, Coughlan JL. Gastro-oesophageal reflux treatment for asthma in adults and children. Cochrane Database Syst Rev 2003;2:CD0011496.
11. Klijendar TO, Salomoa ER, Hietanen EK, Terho EO. Gastroesophageal reflux in asthmatics. A double blind, placebo controlled crossover study with omeprazole. Chest 1999; 116; 1257-1264.
12. Littne MR, Leung WF, Ballard ED, Huang B, Sarma NK. Effect of 24 weeks of lansoprazole therapy on asthma symptoms, exacerbations, quality of life, and pulmonary function in adult asthmatic patients with acid reflux symptoms. Chest 2005; 128: 1128-1135. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4125,
1
],
[
4125,
8581,
2
],
[
8581,
10938,
3
],
[
10938,
16318,
4
],
[
16318,
25079,
5
],
[
25079,
26549,
6
]
]
} |
fcd876767d55ae9283cb2b1d8eed2c0f3ce6baa0 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 12,
"total-fallback-pages": 0,
"total-input-tokens": 39260,
"total-output-tokens": 12784,
"fieldofstudy": [
"Materials Science"
]
} | Broadband Polarization Conversion Metasurface Based on Metal Cut-Wire Structure for Radar Cross Section Reduction
Jia Ji Yang 1, Yong Zhi Cheng 2,∗, Chen Chen Ge 1 and Rong Zhou Gong 1,∗
1 School of Optical and Electronic Information, Huazhong University of Science and Technology, Wuhan 430074, China; [email protected] (J.J.Y.); [email protected] (C.C.G.)
2 School of Information Science and Engineering, Wuhan University of Science and Technology, Wuhan 430081, China
* Correspondence: [email protected] or [email protected] (Y.Z.C.); [email protected] (R.Z.G.)
Received: 13 March 2018; Accepted: 17 April 2018; Published: 19 April 2018
Abstract: A class of linear polarization conversion coding metasurfaces (MSs) based on a metal cut-wire structure is proposed, which can be applied to the reduction properties of radar cross section (RCS). We firstly present a hypothesis based on the principle of planar array theory, and then verify the RCS reduction characteristics using linear polarization conversion coding MSs by simulations and experiments. The simulated results show that in the frequency range of 6–14 GHz, the linear polarization conversion ratio reaches a maximum value of 90%, which is in good agreement with the theoretical predictions. For normal incident x- and y-polarized waves, RCS reduction of designed coding MSs 01/01 and 01/10 is essentially more than 10 dB in the above-mentioned frequency range. We prepare and measure the 01/10 coding MS sample, and find that the experimental results in terms of reflectance and RCS reduction are in good agreement with the simulated ones under normal incidence. In addition, under oblique incidence, RCS reduction is suppressed as the angle of incidence increases, but still exhibits RCS reduction effects in a certain frequency range. The designed MS is expected to have valuable potential in applications for stealth field technology.
Keywords: coding metasurface; polarization conversion; energy scattering; RCS reduction
1. Introduction
As a subwavelength artificial composite, metamaterial (MM) has some exotic electromagnetic (EM) properties that are unavailable in nature [1–4]. Research on MM has inspired many novel applications, including perfect lens imaging, inverse doppler effect, abnormal tunneling and many other phenomena [5–10]. As a two-dimensional (2D) planar form of MM [11–15], metasurfaces (MS) are constituted of subwavelength element arrays and have superior performance in tailoring the EM waves, which can be used to control the reflection/refraction wavefront in a smaller size range. By designing an artificial structure on the interface, the propagation and polarization characteristics of the EM wave regulation can be realized [16–21]. Therefore, the MS has prospects for potential application in polarization manipulation [22–29], high-performance antennas [30,31], and so on [32–39].
Radar stealth technology is concerned with reducing the target radar echo signal to achieve stealth, and radar cross section (RCS) area is an important physical quantity for measuring the radar echo capability of a target. RCS can be effectively reduced by using a MS designed with a suitable supper-unit structure [40–42]. As an important branch of MS, phase gradient MS (GMS) can be used to reduce the RCS by introducing an artificial wave vector at an in-plane direction to control the propagation direction of transmitted and reflected wave beams. Wu et al. proposed a GMS based on...
a cruciform structure, which could be applied to RCS reduction at low frequency ranges \[43\]. This designed MS has the defect of a narrow band range.
Recently, the concept of “coding MS” based on GMS was proposed by Cui et al. \[44\], and indicates a new manner for designing MS. The basic units are arranged on a matrix, and the phase responses of elements like 0 and \(\pi\) can be regarded as digitally “0” and “1”. Coding MS can diffuse the energy of EM waves in each direction by optimizing the matrix arrangement. This coding MS exhibits a broadband characteristic compared with traditional GMS, but still has the disadvantage of polarization-sensitivity. Based on the above research, we designed a series of coding MSs, one of which has the characteristics of a wide band range and polarization insensitivity, meaning that it promotes the RCS reduction of incident EM waves in all polarization directions. The traditional absorber can absorb electromagnetic waves and convert them into heat energy, which will be detected by infrared devices. The RCS reduction mechanism of the designed MS is different from the absorber; by designing the structure graphics and coding scheme, the incident waves are reflected irregularly back into free space, which lowers the probability of the MS being detected by infrared devices. Therefore, the scattering energy of each directional beam is small, which can achieve an effective RCS reduction at different angles \[45–52\].
The metal cut-wire structure (rectangular strip structure) is advantageous in terms of its simple structure, the adjustability of its geometric parameters, and has a wide application with respect to the design of MMs and MSs. In this work, we apply it as the basic unit structure for the design of a broadband linear polarization conversion MS. We can obtain the required phase difference by rotating the metal cut-wire structure by 90° along the propagation direction, and the phase response of the two basic units are 0 and \(\pi\), denoted by “0” and “1”, respectively. At the same time, we design a class of 1-bit coding form to focus on the RCS reduction characteristics of the polarization conversion MS. The designed MS presents the characteristics of a broadband range, polarization-insensitivity, and wide incidence angle. For normal \(x/y\)-polarized incidence, both simulations and experiments show that the RCS of the MS is reduced by an average of 10 dB in the frequency range of 6–14 GHz. Meanwhile, at oblique incidence, RCS reduction is suppressed, but still has a certain effect in the above-mentioned frequency range.
2. Design of Basic Unit and Theoretical Analysis
In this paper, we present a basic unit based on a metal cut-wire structure, which we assume to be a “0” unit, as shown in Figure 1a. The whole unit-cell is divided into three functional layers, where the period is \(p = 10\) mm. This design of the three functional layers can form a Fabry-Perot-like resonance cavity, consequently leading to interference of cross-polarization couplings in multi-reflection \[21,22\]. The front layer is the metal cut-wire structure of copper film; the length and width are \(l = 10\) mm and \(w = 1.6\) mm, respectively. The front layer cut-wire structure possesses a symmetric axis at 45° with respect to the \(x\) or \(y\) direction, such that a 90° polarization conversion can be achieved when the incident wave is \(x\)- or \(y\)-polarized. The middle layer dielectric substrate is FR4 film with a thickness of \(h = 3.5\) mm; the side view is shown in Figure 1c. The dielectric constant is 4.3, and the loss tangent is 0.025. The back layer is continuous copper film and has the same thickness as the front metal structure, which is 0.035 mm. We rotated the metal cut-wire counterclockwise by 90° (Figure 1a) to obtain the MS unit structure, as shown in Figure 1b, which is set to “1”.
We used the frequency domain solver on the EM simulation software CST MICROWAVE STUDIO to obtain the co-polarization \(r_{xx}\) and \(r_{yy}\) and cross-polarization \(r_{yx}\) and \(r_{xy}\) reflection coefficients for both incident \(x\)-polarized and \(y\)-polarized waves. As suggested by the reflection coefficients in Figure 1d, the structure is able to achieve efficient linear polarization conversion across a wide frequency range of 6–14 GHz, the cross-polarization reflection coefficients \(r_{yx}\) and \(r_{xy}\) are greater than 0.85, and the co-polarization reflection coefficients \(r_{xx}\) and \(r_{yy}\) are substantially less than 0.3. In addition, at resonance frequencies, the co-polarization reflection coefficients \(r_{xx}\) and \(r_{yy}\) reach a minimum, and the corresponding amplitudes are less than 0.1. The corresponding cross-polarization reflection coefficients \(r_{yx}\) and \(r_{xy}\) reach a maximum, and the amplitudes are greater than 0.9. This
result indicates that the normal incident \( x(y) \)-polarized waves are almost completely converted to \( y(x) \)-polarized waves, or produced approximately 90° linear polarization deflections.
**Figure 1.** (a) The element “0” with an angle of 45° to the x-axis; (b) the element “1” rotated 90° counterclockwise about the z-axis; (c) the side view of the basic unit; (d) the reflection coefficient of elements “0” and “1” under normal \( x \) and \( y \)-polarized incidence; (e) the reflection phase of cross-polarized wave; (f) the linear polarization conversion ratio of \( x \) and \( y \)-polarized wave; (g) cross-polarization reflection phase difference of elements “0” and “1”.
In order to visually reflect the polarization conversion capability of the MS, we define the polarization conversion efficiency as follows [22,23]:
\[
\text{PCR}_x = \frac{|r_{yx}|^2}{|r_{yx}|^2 + |r_{xx}|^2} \quad \text{and} \quad \text{PCR}_y = \frac{|r_{xy}|^2}{|r_{xy}|^2 + |r_{yy}|^2}.
\]
In Figure 1f, the linear polarization conversion ratio of the \( x \) and \( y \)-polarized waves are as high as 90% and reached 99% at resonance frequencies. Figure 1e,g shows the cross-polarization phases of the “0” and “1” units, and the corresponding phase difference, \( \Delta \phi_{10} (\Delta \phi_{xy}) \), in the whole 4–16 GHz range. It is observed that the cross-polarization phase difference \( \Delta \phi_{10} (\Delta \phi_{xy}) = \pm 180° \) can be obtained, which indicates that the phase gradient of the designed MS is 180°.
This basic unit has excellent simulated results; mainly due to the front cut-wire structure layer and metal back layer forming a Fabry-Perot-like resonance cavity, we are able to observe the multiple reflections and transmissions in the Fabry-Perot-like resonance cavity as shown in Figure 2. The transmitted EM waves continue to travel in the dielectric spacer until they encounter the ground plane with a complex propagation phase \( \beta = \sqrt{\varepsilon_r k_0 d} \), where \( k_0 \) is the free space wavenumber, the \( \varepsilon_r \) and \( d \) are the relative permittivity and thickness of the middle dielectric layer. The partial reflection and transmission waves arrive at the air-spacer interface again from the reverse direction after the additional propagation phase \( \beta \) in the dielectric spacer [22]. The incident EM wave prompted multiple reflections in this resonance cavity, including co-polarization wave interference cancellation,
and cross-polarization wave interference superposition. Based on our previous research [22], the cross-polarization and co-polarization reflection coefficients can be expressed as:
\[ r_{yx} = \vec{r}_{yx} + \sum_{j=1}^{\infty} r_{yj} \]
and \( r_{xx} = \vec{r}_{xx} + \sum_{j=1}^{\infty} r_{xj} \). Thus, we simulate the unit-cell structure without the ground plane, and use MATLAB software to calculate the reflection coefficients and polarization conversion ratio for the \( x \)-polarized wave incidence. As shown in Figure 3a,b, the calculated data are in reasonable agreement with the simulated data, further illustrating the working principle of the Fabry-Perot-like resonance cavity. Therefore, we can use the characteristics of the \( \pm 180^\circ \) cross-polarization reflected phase difference based on the metal cut-wire structure elements “0” and “1” and realize polarization conversion by combining different coding combinations.
**Figure 2.** Schematic of multiple reflections and transmissions in a Fabry-Perot-like resonance cavity for polarization conversion.
**Figure 3.** Simulated, calculated (a) reflection coefficients and (b) PCRs of the designed MS.
A traditional MS has a uniform reflection phase when a plane wave impinges on it, leading to a strong scattering pattern. In order to manipulate the reflected beam direction, a phase gradient is introduced into the interface to control the equiphase wavefront. To suppress the normal strong scattering of the MS, the simplest way is to generate a matrix of designed phase distribution. We propose a series of polarization conversion coding MSs composed of “0” and “1” units, and make a hypothesis based on the principle of planar array theory [53,54]; then, we verify this derivation using data from the simulation. For a MS with \( A \times B \) elements, the total scattering field can be expressed as [53]:
\[ E_{total} = EF \cdot AF \]
where $EF$ represents the scattering field of the cell pattern; and $AF$ represents the array factor, which can be expressed as [53]:
$$AF = \sum_{a=1}^{A} \sum_{b=1}^{B} e^{i[(a-1/2)(kd \sin \theta \cos \varphi) + (b-1/2)(kd \sin \theta \sin \varphi) + \Phi(a,b)]}$$
(2)
where $\theta$ is the angle between the incident wave and the $z$-axis along the $XOZ$-plane; $\varphi$ is the angle between incident wave and $x$-axis along the $XOY$-plane; $a$ and $b$ represent the rows and columns of unit cell; $k = 2\pi/\lambda$, $\lambda$ is the wavelength of incident wave; $d$ is the distance between the units; and $\Phi(a,b)$ is the phase difference between the elements. As shown in Figure 4a, we put $01/10$ code into $EF$-fields of the “0” and “1” elements, respectively. The array factor $AF$ expressed as:
$$E_{\text{total}} = EF_{0} \cdot AF_{0} + EF_{1} \cdot AF_{1}$$
(3)
where $AF_{0}$ and $AF_{1}$ are the elements “0” and “1”, respectively; and $EF_{0}$ and $EF_{1}$ are the scattering fields of the “0” and “1” elements, respectively. The array factor $AF_{2 \times 2}$ with a $2 \times 2$ arrangement can be expressed as:
$$AF_{2 \times 2} = \left[ e^{i\left[\frac{1}{2}kd \sin \theta \cos \varphi + \frac{1}{2} kd \sin \theta \sin \varphi + \Phi(1,1)\right]} + e^{i\left[\frac{1}{2}kd \sin \theta \cos \varphi + \frac{1}{2}kd \sin \theta \sin \varphi + \Phi(2,2)\right]} \right]$$
$$+ \left[ e^{i\left[\frac{1}{2}kd \sin \theta \cos \varphi + \frac{1}{2}kd \sin \theta \sin \varphi + \Phi(1,2)\right]} + e^{i\left[\frac{1}{2}kd \sin \theta \cos \varphi + \frac{1}{2}kd \sin \theta \sin \varphi + \Phi(2,1)\right]} \right]$$
(4)
For the $01/10$ coding sequence, $\Phi(1,1)$ and $\Phi(2,2)$ correspond to the phase difference of $0^\circ$; $\Phi(1,2)$ is the $180^\circ$ phase difference of two adjacent cells of column $B$, and $\Phi(1,2) \rightarrow 2kdsin\theta cos\varphi$; $\Phi(2,1)$ is the $180^\circ$ phase difference of two adjacent cells of row $A$, and $\Phi(1,2) \rightarrow 2kdsin\theta cos\varphi$. Further derivation shows that the array factor of coding $01/10$ is:
$$AF_{01/10} = \left[ e^{i\frac{kd}{2}sin\theta \cos\varphi + sin\theta \sin\varphi} + e^{-i\frac{kd}{2}sin\theta \cos\varphi + sin\theta \sin\varphi} \right]$$
$$+ \left[ e^{i\frac{kd}{2}sin\theta \cos\varphi - sin\theta \sin\varphi} + e^{-i\frac{kd}{2}sin\theta \cos\varphi - sin\theta \sin\varphi} \right]$$
(5)
The above formula can be decomposed according to the arrangement of Figure 5a:
$$\begin{align*}
AF_{0} &= e^{i\frac{kd}{2}sin\theta \cos\varphi + sin\theta \sin\varphi} + e^{-i\frac{kd}{2}sin\theta \cos\varphi + sin\theta \sin\varphi} \\
AF_{1} &= e^{i\frac{kd}{2}sin\theta \cos\varphi - sin\theta \sin\varphi} + e^{-i\frac{kd}{2}sin\theta \cos\varphi - sin\theta \sin\varphi}
\end{align*}$$
(6)
$AF_{0}$ and $AF_{1}$ are expressed in the matrices as shown below:
$$\begin{bmatrix}
AF_{0} \\
AF_{1}
\end{bmatrix} = \begin{bmatrix}
e^{i\frac{kd}{2}sin\theta \cos\varphi + sin\theta \sin\varphi} & e^{-i\frac{kd}{2}sin\theta \cos\varphi + sin\theta \sin\varphi} \\
e^{-i\frac{kd}{2}sin\theta \cos\varphi - sin\theta \sin\varphi} & e^{i\frac{kd}{2}sin\theta \cos\varphi - sin\theta \sin\varphi}
\end{bmatrix} \times \begin{bmatrix}
e^{i\frac{kd}{2}sin\theta \cos\varphi} \\
e^{i\frac{kd}{2}sin\theta \cos\varphi}
\end{bmatrix}$$
(7)
We can see that the first half of the matrix corresponds to the arrangement of the $01/10$ coding structure, and the latter part of the matrix corresponds to the $0$–$1$ phase difference distribution.
For the $01/01$ coding sequence, the derivation method is similar to the one for the $01/10$ coding sequence. For $2 \times 2$ coding MS, the area of the unit-cell structure is assumed to be 1. For normal incidence, the angles of incidence are $\theta = \varphi = 0^\circ$; coding $01/10$ corresponds to $AF_{0} = AF_{1} = 2/4 = 1/2$. Similarly, coding $01/01$ corresponds to $AF_{0} = AF_{1} = 1/2$.
Since the unit-cell structure has the characteristic of polarization conversion, the polarization direction of the incident wave is along the $x$-axis. For the normal incident field $EF_{i}$, the incident fields $EF_{0}$ and $EF_{1}$ are expressed as:
$$\begin{align*}
EF_{0} &= r_{x0x0}EF_{i} + r_{y0x0}EF_{i} \\
EF_{1} &= r_{x1x1}EF_{i} + r_{y1x1}EF_{i}
\end{align*}$$
(8)
where \( r_{xx} = r_{xx} e^{j \phi_{xx}} \) is the co-polarization reflection coefficient and \( r_{yx} = r_{yx} e^{j \phi_{yx}} \) is the cross-polarization reflection coefficient. From the previous simulated results, we can see that the units “0” and “1” have the same co-polarization reflection phase, and the cross-polarization reflection phase difference is 180°. Therefore, the total scattering field can be expressed as:
\[
E_{\text{total}} = \frac{1}{2} (r_{x0} E_{f0} e^{j \phi_{x0}} + r_{x1} E_{f1} e^{j \phi_{x1}} + r_{y0} E_{f0} e^{j \phi_{y0}} + r_{y1} E_{f1} e^{j \phi_{y1}} )
\]
\[
= \frac{1}{2} (r_{x0} E_{f0} e^{j \phi_{x0}} + r_{x1} E_{f1} e^{j \phi_{x1}} ) = r_{xx} E_{f} e^{j \phi_{xx}}
\]
The total scattering field of the array structure is \( |E_{\text{total}}| = r_{xx} |E_{f}| \). For a metal plate, the size of the scattering field is \( |E_{f}\text{total}| = |E_{f}| \), and the scattering coefficient is \( r_{xx} = |E_{\text{total}}| / |E_{f}\text{total}| \).
\[
-10 \log |r_{xx}|^2 \geq 10 \text{ dB} \Rightarrow r_{xx} \leq \sqrt{0.1} \approx 0.316
\]
In summary, coding 01/01 and 01/10 have better RCS reduction characteristics under certain conditions compared with coding 00/00 and 11/11. Since the co-polarization and cross-polarization phases of 00/00 and 00/11 are consistent, the cross-polarization components of the scattering field can’t be offset, and RCS can’t be reduced; thus, the corresponding RCS reduction is 0. The next step is to verify this hypothesis by simulation.
**Figure 4.** (a) 2 \times 2 structure arrangement of the 01/10 coding MS; (b) illustration of the 01/10 coding MS; the “0” and “1” indicate the super-cells, and are distinguished by blue and yellow colors; each super-cell consists of 5 \times 5 unit-cells of “0” and “1”, shown in Figure 1a,b.
To meet the periodic boundary conditions required for simulation, we used a 5 \times 5 basic unit as a super-cell and designed a series of coding forms to explore the polarization conversion characteristics of the MS. Figure 5a–c shows the far-field scattering characteristic diagram of each regular coding form at 10 GHz. The energy scattering direction of coding 00/00 or 11/11 is upright, and the normal scattering capability is very strong, suggesting that these coding MSs do not have characteristics of RCS reduction under normal incidence. The energy scattering of 01/01 and 10/10 are the same, diverging to both sides. The energy scattering of 01/10 diverges all around, and the normal scattering capacity is relatively weak. Thus, it can be expected that the 01/01 and 01/10 coding MSs have normal RCS reduction characteristics. Therefore, by controlling the coding method of the MS, it is possible to change the scattering direction of the energy. However, the coding 01/10 has polarization-insensitive properties compared to the coding 01/01. In order to meet the needs of practical applications, we focus on studying the MS of coding 01/10, and the schematic diagram is shown in Figure 4b.
3. Simulation, Experiment and Discussion
Using the frequency domain solver in the EM simulation software CST MICROWAVE STUDIO (2016, CST, Darmstadt, Germany), we perform a numerical simulation of three regular coding MSs, as shown in Figure 4. Firstly, for 01/10 coding MS, the RCS reduction characteristics of a super-cell with \( n \times n \) basic units were discussed numerically; as shown in Figure 6a, the RCS reduction of \( 5 \times 5 \) super-cell increases compared with a \( 3 \times 3 \) or \( 1 \times 1 \) super-cell, and the overall curves shift to the lower frequencies. With an increase in the number of units, the numerical RCS reduction gradually stabilizes. To meet the requirements of test conditions and practical application, a \( 5 \times 5 \) supercell was selected as the basic coding unit for studying the RCS reduction characteristics. Figure 6b shows the RCS reduction in all arrangements under normal incidence, where the MSs of coding 01/01 and 01/10 achieve RCS reduction in a broadband range. In Figure 1d, the reflection coefficient of the basic unit is \( r_{xx} = 0.318 > 0.316 \) at the 7.7 GHz peak, and \( r_{xx} = 0.301 < 0.316 \) at the 12.2 GHz peak. According to Formula (10), we can see that the RCS reduction of the MS should be less than 10 dB around 7.7 GHz. Figure 6b also shows that the MSs of coding 01/01 and 01/10 have dips at the above-mentioned frequency, where RCS reduction is less than 10 dB around 7.7 GHz. For the MSs of coding 00/00 and 11/11, it can be seen that the RCS reduction is essentially zero in our frequency range of interest of 5–15 GHz. This is mainly due to the consistency of co-polarization and cross-polarization phases for all basic units, where the RCS of the target can’t be reduced since the cross-polarization components of the scattering field can’t be offset. These simulated results near perfectly verify the above analysis.
We simulated the RCS reduction of 01/01 and 01/10 coding MSs for the incidence of x-polarized and y-polarized waves, as shown in Figure 6c,d, respectively. In Figure 6c, the RCS reduction amplitude of the 01/01 coding MS is clearly different between the x- and y-polarized waves under normal incidence, indicating that the 01/01 coding MS has a certain polarized selection characteristic. As shown in Figure 6d, the RCS reduction curves of the 01/10 coding MS under x- and y-polarized waves are slightly different, but the trend is basically consistent under normal incidence, indicating that the coding MS exhibits a polarization-insensitive property. The 01/10 coding MS exhibits a near polarization-insensitive property, mainly due to its chessboard arrangement; the direction of electric field and magnetic field are relatively consistent with the incident x- and y-polarized waves. Hence, we can see the reflection coefficients of EM waves are basically the same. However, the 01/01 is asymmetric, and the direction of electric field and magnetic field are not consistent with the incident x- and y-polarized waves under the same incident surface. In addition, the RCS reduction of the 01/10 coding MS is greater than 10 dB on average throughout a wide frequency range of 6–14 GHz and is greater than 20 dB in the frequency range of 10–11 GHz. At 10.5 GHz, the RCS reduction reaches a maximum of 37 dB.
Figure 6. Simulated results of a series of regular coding MSs: (a) RCS reduction of 01/10 coding MS for different super-cell combinations; (b) RCS reduction of the different regular coding MS with 5 × 5 super-cell; (c) RCS reduction of the 01/01 coding MS for normal x- and y-polarized incident waves; (d) RCS reduction of the 01/10 coding MS for normal incident x- and y-polarized waves.
To further study the RCS reduction characteristics of 01/10 coding MS, we study the scattering pattern of the XOZ-plane under normal incidence, as shown in Figure 7a–d; where the scattering characteristics of the 01/10 coding MSs at 6 GHz, 10 GHz, 11 GHz, and 14 GHz are compared with a metal plate with the same size of 400 × 400 mm². According to the law of energy conservation, the main lobe energy is suppressed significantly by enhancing the scattered energy of the side lobes to achieve RCS reduction under normal incidence. The pattern shows that the metal plate has a main lobe throughout the whole band under normal incidence. As shown in Figure 7a,d, relative to the metal plate, the MS has a certain inhibitory effect on the main lobe at 6 and 14 GHz, respectively. As shown in Figure 7b,c, the MS has an obvious inhibitory effect on the main lobe at 10 and 11 GHz, respectively. Based on the above results, it can be suggested that the polarization conversion MS can realize RCS reduction throughout a wide frequency range and can adjust the scattering field dynamically.
Figure 7. Scattering patterns of the 01/10 coding MS and metal plate in the XOZ-plane at (a) 6 GHz; (b) 10 GHz; (c) 11 GHz; and (d) 14 GHz.
In order to further verify the RCS reduction characteristics of the polarization conversion coding MS, we fabricated a MS sample of 01/10 coding and measured it in a microwave anechoic chamber. A 400 × 400 mm² sample was fabricated using traditional printed circuit board (PCB) technology, as shown in Figure 8a. The front and back layers of the sample are covered with copper, and the middle layer is a FR4 substrate with a thickness of 3.5 mm. Using the free space method, we measured the sample in the microwave anechoic chamber (see Figure 8b). The measured sample was fixed in the center of a rotating foam tower, the transmitter and receiver antennas were fixed to the same height, ensuring an angle of 5°. Then, we connected two horn antennas of co-polarization state to two ports of the Agilent Technologies N5244A Vector Analyzer and measured the RCS of the sample.
Figure 8. (a) The fabricated 01/10 coding MS sample; (b) the measurement setup in the microwave anechoic chamber.
The reflectance of simulation and experiment under normal incident x- and y-polarized wave are shown in Figure 9a. Under the conditions of incident different polarization waves, the simulated results under x- and y-polarized incidence are basically coincident, as are the measured results, which reveals the near polarization insensitivity of the proposed MS. However, the results of the simulation and experiments are quite different at the main peak. The main reason behind this deviation may be the error occurring in the production of the sample, such as the inability to precisely control the thickness, or the sample not being completely flat. The simulated and measured reflectance (x- and y-polarization) near 7.7 GHz is slightly greater than −10 dB, which verifies the previous theoretical analysis, and the reflectance (x- and y-polarization) is below −5 dB in the frequency range of 5–14 GHz. Figure 9b shows the RCS reduction of the measured sample at different angles of incidence. At normal incidence, although there is a valley less than 10 dB near the 7.7 GHz, the RCS reduction is basically greater than 10 dB in the frequency range of 6–14 GHz. Under a small oblique incident angle, the RCS reduction curves of oblique incident angles from 0° to 10° are basically the same. When the incident angles reach 20° and 30°, RCS reduction fluctuate around 5 dB in the frequency range of 6–14 GHz.
It can be seen that RCS reduction is suppressed as the angle increases, but still has a certain effect within the above frequency range.
Figure 9. (a) The simulated and measured reflectance results of the sample; (b) RCS reduction of the sample under oblique incident waves from 0° to 30°.
4. Conclusions
In this work, a basic unit based on a metal cut-wire structure was designed, which allowed us to achieve the highly efficient conversion of linear polarization within a broadband frequency range. A hypothesis was proposed based on the principle of planar array theory and was verified by using the simulated data of the MS. Utilizing a super-cell consisting of $5 \times 5$ basic units, we studied the RCS reduction of the polarization conversion MS based on super-coding theory. By simply rotating the metal cut-wire structure, it is possible to realize the characteristics of “0” and “1” basic unit 180° cross-polarization reflected phase differences. This method avoids the need to change the graphic design size and makes it possible to arrange super-coding in an easy and efficient way. We simulated a series of coding forms to generate the RCS reduction graph and scattering pattern under normal incidence, further exploring the relationship of the coding MS with different arrangement modes and the energy scattering direction at the same time. The influence of polarization conversion characteristics for RCS reduction was also proved. Based on the simulated design, we fabricated a 01/10 coding MS sample and measured its reflectance and RCS reduction. The measured results are in good agreement with the simulation, which proves that the polarization conversion MS structure is able to realize the reduction of RCS. Our work proposes a new design as a basis for future studies on RCS reduction of MSs, which has potential application in stealth field technology.
Acknowledgments: This work is supported by the National Natural Science Foundation of China (U1435209, 61605147 and 61701185), and the Natural Science Foundation of Hubei province (Grant No. 2017CFB588).
Author Contributions: Jia Ji Yang and Yong Zhi Cheng conceived and designed the experiments; Jia Ji Yang and Chen Chen Ge performed the experiments; Jia Ji Yang and Yong Zhi Cheng analyzed the data; Yong Zhi Cheng and Rong Zhou Gong contributed reagents/materials/analysis tools; Jia Ji Yang wrote the paper.
Conflicts of Interest: The authors declare no conflict of interest.
References
1. Pendry, J.B.; Schurig, D.; Smith, D.R. Controlling electromagnetic fields. Science 2006, 312, 1780–1782. [CrossRef] [PubMed]
2. Martin, F.; Falcone, F.; Bonache, J.; Marques, R. Miniaturized coplanar waveguide stop band filters based on multiple tuned split ring resonators. IEEE Microw. Wirel. Compon. Lett. 2003, 13, 511–513. [CrossRef]
3. Smith, D.R.; Pendry, J.B.; Wiltshire, M.C.K. Metamaterials and negative refractive index. Science 2004, 305, 788–792. [CrossRef] [PubMed]
4. Engheta, N. Circuits with light at nanoscales: Optical nanocircuits inspired by metamaterials. Science 2007, 317, 1698–1702. [CrossRef] [PubMed]
5. Alù, A.; Engheta, N. Achieving transparency with plasmonic and metamaterial coatings. Phys. Rev. E 2005, 72, 016623. [CrossRef] [PubMed]
6. Monticone, F.; Estakhri, N.M.; Alù, A. Full control of nanoscale optical transmission with a composite metascreen. Phys. Rev. Lett. 2013, 110, 203903. [CrossRef] [PubMed]
7. Shelby, R.A.; Smith, D.R.; Schultz, S. Experimental verification of a negative index of refraction. Science 2001, 292, 79–29. [CrossRef] [PubMed]
8. Pendry, J.B. Negative refraction makes a perfect lens. Phys. Rev. Lett. 2000, 85, 3966. [CrossRef] [PubMed]
9. Xu, H.X.; Tang, S.W.; Ling, X.H.; Luo, W.J.; Zhou, L. Flexible control of highly-directive emissions based on bifunctional metasurfaces with low polarization cross-talking. Ann. Phys. Berl. 2017, 529, [CrossRef] [PubMed]
10. Alrasheed, S.; Fabrizio, E.D. Design and simulation of reflect-array metasurfaces in the visible regime. Appl. Opt. 2017, 56, 3213–3218. [CrossRef] [PubMed]
11. Cheng, Y.Z.; Cheng, Z.Z.; Mao, X.S.; Gong, R.Z. Ultra-thin multi-band polarization-insensitive microwave metamaterial absorber based on multiple-order responses using a single resonator structure. Materials 2017, 10, 1241. [CrossRef] [PubMed]
12. Cheng, Y.Z.; Wu, C.J.; Ge, C.C.; Yang, J.J.; Pei, X.J.; Jia, F.; Gong, R.Z. An ultra-thin dual-band phase-gradient metasurface using hybrid resonant structures for backward RCS reduction. Appl. Phys. B 2017, 123, 143. [CrossRef]
13. Li, Y.F.; Wang, J.; Zhang, J.; Qu, S.B.; Pang, Y.Q.; Zheng, L.; Yan, M.B.; Xu, Z.; Zhang, A.X. Ultra-wide-band microwave composite absorbers based on phase gradient metasurfaces. *Prog. Electromagn. Res.* 2014, 40, 9–18. [CrossRef]
14. Xu, H.X.; Ma, S.J.; Luo, W.J.; Cai, T.; Sun, S.L.; He, Q.; Zhou, L. aberration-free and functionality-switchable meta-lenses based on tunable metasurfaces. *Appl. Phys. Lett.* 2016, 109, 193506. [CrossRef]
15. Fan, Y.C.; Liu, Z.; Zhang, F.L.; Zhao, Q.; Wei, Z.Y.; Fu, Q.H.; Li, J.J.; Gu, C.Z.; Li, H.Q. Tunable mid-infrared coherent perfect absorption in a graphene meta-surface. *Sci. Rep.* 2015, 5, 13956. [CrossRef] [PubMed]
16. Tian, S.; Liu, H.; Li, L. Design of 1-bit digital reconfigurable reflective metasurface for beam-scanning. *Appl. Sci. Basel* 2017, 7, 882. [CrossRef]
17. Lee, S.H.; Choi, M.; Kim, T.T.; Lee, S.; Liu, M.; Yin, X.; Choi, H.K.; Lee, S.S.; Choi, C.G.; Choi, S.Y.; et al. Switching terahertz waves with gate-controlled active graphene metamaterials. *Nat. Mater.* 2012, 11, 936. [CrossRef] [PubMed]
18. Politano, A.; Chiarello, G. Plasmon modes in graphene: Status and prospect. *Nanoscale* 2014, 6, 10927–10940. [CrossRef] [PubMed]
19. Mitrofanov, O.; Viti, L.; Dardanis, E.; Giordano, M.C.; Ercolani, D.; Politano, A.; Sorba, L.; Vitiello, M.S. Near-field terahertz probes with room-temperature nanodetectors for subwavelength resolution imaging. *Sci. Rep.* 2017, 7, 44240. [CrossRef] [PubMed]
20. Politano, A.; Viti, L.; Vitiello, M.S. Optoelectronic devices, plasmonics, and photonics with topological insulators. *APL Mater.* 2017, 5, 035504. [CrossRef]
21. Yang, Q.; Gu, J.; Wang, D.; Zhang, X.; Tian, Z.; Ouyang, C.; Singh, R.; Han, J.; Zhang, W. Efficient flat metasurface lens for terahertz imaging. *Opt. Express* 2014, 22, 25931–25939. [CrossRef] [PubMed]
22. Cheng, Y.Z.; Fang, C.; Mao, X.S.; Gong, R.Z.; Wu, L. Design of an ultrabroadband and high-efficiency reflective linear polarization converter at optical frequency. *IEEE Photonics J.* 2016, 8, 1–9. [CrossRef]
23. Zhao, J.C.; Cheng, Y.Z. Ultra-broadband and high-efficiency reflective linear polarization converter based on planar anisotropic metamaterial in microwave region. *Optik* 2017, 136, 52–57. [CrossRef]
24. Jiang, W.; Xue, Y.; Gong, S.X. Polarization conversion metasurface for broadband radar cross section reduction. *Prog. Electromagn. Res. Lett.* 2016, 62, 9–15. [CrossRef]
25. Li, H.P.; Wang, G.M.; Liang, J.G.; Gao, X.J. Wideband multifunctional metasurface for polarization conversion and gain enhancement. *Prog. Electromagn. Res.* 2016, 155, 115–125. [CrossRef]
26. Song, Y.C.; Ding, J.; Guo, C.J.; Ren, Y.H.; Zhang, J.K. Ultra-broadband backscatter radar cross section reduction based on polarization-insensitive metasurface. *IEEE Antennas Wirel. Propag. Lett.* 2016, 15, 329–331. [CrossRef]
27. Mo, W.C.; Wei, X.L.; Wang, K.J.; Li, Y.; Liu, J.S. Ultrathin flexible terahertz polarization converter based on metasurfaces. *Opt. Express* 2016, 24, 13621–13627. [CrossRef] [PubMed]
28. Jiang, S.C.; Xiong, X.; Hu, Y.S.; Hu, Y.H.; Ma, G.B.; Peng, R.W.; Sun, C.; Wang, M. Controlling the polarization state of light with a dispersion-free metastructure. *Phys. Rev. X* 2014, 4, 021026. [CrossRef]
29. Ding, F.; Wang, Z.; He, S.; Shalaev, V.M.; Kildishev, A.V. Broadband high-efficiency half-wave plate: A supercell-based plasmonic metasurface approach. *ACS Nano* 2015, 9, 4111–4119. [CrossRef] [PubMed]
30. Burokur, S.N.; Daniel, J.P.; Ratajczak, P.; Lustrac, A.D. Tunable bilayered metasurface for frequency reconfigurable directive emissions. *Appl. Phys. Lett.* 2010, 97, 064101. [CrossRef]
31. Burokur, S.N.; Ourir, A.; Daniel, J.P.; Ratajczak, P.; Lustrac, A.D. Highly directive ISM band cavity antenna using a bi-layered metasurface reflector. *Microw. Opt. Technol. Lett.* 2009, 51, 1393–1396. [CrossRef]
32. Dong, G.X.; Shi, H.Y.; Li, W.; He, Y.C.; Zhang, A.X.; Xu, Z.; Wei, X.Y. A multi-band spoof surface plasmon polariton coupling metasurface based on dispersion engineering. *J. Appl. Phys.* 2016, 120, 084505. [CrossRef]
33. Liang, L.J.; Qi, M.Q.; Yang, J.; Shen, X.P.; Zhai, J.Q.; Xu, W.Z.; Jin, B.B. Anomalous terahertz reflection and scattering by flexible and conformal coding metamaterials. *Adv. Opt. Mater.* 2015, 3, 1374–1380. [CrossRef]
34. Liao, Z.; Luo, Y.; Fernandez-Dominguez, A.I.; Shen, X.P.; Maier, S.A.; Cui, T.J. High-order localized spoof surface plasmon resonances and experimental verifications. *Sci. Rep.* 2015, 5, 9590. [CrossRef] [PubMed]
35. Cheng, Y.Z.; Mao, X.S.; Wu, C.J.; Wu, L.; Gong, R.Z. Infrared non-planar plasmonic perfect absorber for enhanced sensitive refractive index sensing. *Opt. Mater.* 2016, 53, 195–200. [CrossRef]
36. Wu, C.J.; Cheng, Y.Z.; Wang, W.Y.; He, B.; Gong, R.Z. Ultra-thin and polarization-independent phase gradient metasurface for high-efficiency spoof surface-plasmon-polariton coupling. *Appl. Phys. Express* 2015, 8, 122001. [CrossRef]
37. Cheng, Y.Z.; Huang, M.L.; Chen, H.R.; Guo, Z.; Mao, X.S.; Gong, R.Z. Ultrathin six-band polarization-insensitive perfect metamaterial absorber based on a cross-cave patch resonator for terahertz waves. *Materials* 2017, 10, 591. [CrossRef] [PubMed]
38. Pors, A.; Ding, F.; Chen, Y.; Radko, I.P.; Bozhevolnyi, S.I. Random-phase metasurfaces at optical wavelengths. *Sci. Rep.* 2016, 6, 28448. [CrossRef] [PubMed]
39. Ding, F.; Pors, A.; Bozhevolnyi, S.I. Gradient metasurfaces: A review of fundamentals and applications. *Rep. Prog. Phys.* 2017, 81, 026401. [CrossRef] [PubMed]
40. Huang, C.; Pan, W.B.; Ma, X.L.; Luo, X.G. Wideband radar cross-section reduction of a stacked patch array antenna using metasurface. *IEEE Antennas Wirel. Propag. Lett.* 2015, 14, 1369–1372. [CrossRef]
41. Liu, Y.; Hao, Y.W.; Li, K.; Gong, S.X. Wideband and polarization independent radar cross section reduction using holographic metasurface. *IEEE Antennas Wirel. Propag. Lett.* 2016, 15, 1028–1031. [CrossRef]
42. Yuan, Y.; Jiang, C.; Liu, L.; Yu, S.; Cui, Z.; Chen, M.; Lin, S.; Wang, S.; Huang, L. Conventional, sensitive and high-throughput method for screening botanic origin. *Sci. Rep.* 2014, 4, 5395. [CrossRef] [PubMed]
43. Wu, C.J.; Cheng, Y.Z.; Wang, W.Y.; He, B.; Gong, R.Z. Design and radar cross section reduction experimental verification of phase gradient meta-surface based on cruciform structure. *Acta Phys. Sin.* 2015, 64, 164102.
44. Cui, T.J.; Qi, M.Q.; Wan, X.; Zhao, J.; Cheng, Q. Coding metamaterials, digital metamaterials and programmable metamaterials. *Light: Sci. Appl.* 2014, 3, e218. [CrossRef]
45. Su, P.; Zhao, Y.J.; Jia, S.L.; Shi, W.W.; Wang, H.L. An ultra-wideband and polarization-independent metasurface for RCS reduction. *Sci. Rep.* 2016, 6, 20387. [CrossRef] [PubMed]
46. Sun, H.Y.; Gu, C.Q.; Chen, X.L.; Li, Z.; Liu, L.L.; Xu, B.Z.; Zhou, Z.C. Broadband and broad-angle polarization-independent metasurface for radar cross section reduction. *Sci. Rep.* 2017, 7, 40782. [CrossRef] [PubMed]
47. Chen, K.; Cui, L.; Feng, Y.J.; Zhao, J.M.; Jiang, T.; Zhu, B. Coding metasurface for broadband microwave scattering reduction with optical transparency. *Opt. Express* 2017, 25, 5571–5579. [CrossRef] [PubMed]
48. Sun, W.J.; He, Q.; Sun, S.L.; Zhou, L. High-efficiency surface plasmon meta-couplers: Concept and microwave-regime realizations. *Light: Sci. Appl.* 2016, 5, e16003. [CrossRef]
49. Xu, H.X.; Tang, S.W.; Ma, S.J.; Luo, W.J.; Cai, T.; Sun, S.L.; He, Q.; Zhou, L. Tunable microwave metasurfaces for high-performance operations: Dispersion compensation and dynamical switch. *Sci. Rep.* 2016, 6, 38255. [CrossRef] [PubMed]
50. Pan, W.B.; Huang, C.; Pu, M.B.; Ma, X.L.; Cui, J.H.; Zhao, B.; Luo, X.G. Combining the absorptive and radiative loss in metasurfaces for multi-spectral shaping of the electromagnetic scattering. *Sci. Rep.* 2016, 6, 21462. [CrossRef] [PubMed]
51. Ma, H.F.; Liu, Y.Q.; Luan, K.; Cui, T.J. Multi-beam reflections with flexible control of polarizations by using anisotropic metasurfaces. *Sci. Rep.* 2016, 6, 39390. [CrossRef] [PubMed]
52. Jia, Y.; Liu, Y.; Guo, Y.J.; Li, K.; Gong, S.X. Broadband polarization rotation reflective surfaces and their applications to RCS reduction. *IEEE Trans. Antennas Propag.* 2016, 64, 179–188. [CrossRef]
53. Balanis, C.A. *Antenna Theory: Analysis and Design*; Wiley: New York, NY, USA, 2005.
54. Zhao, Y.; Cao, X.Y.; Gao, J.; Sun, Y.; Yang, H.H.; Liu, X.; Zhou, Y.L.; Han, T.; Chen, W. Broadband diffusion metasurface based on a single anisotropic element and optimized by the simulated annealing algorithm. *Sci. Rep.* 2016, 6, 23896. [CrossRef] [PubMed]
© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/). | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3487,
1
],
[
3487,
8356,
2
],
[
8356,
10874,
3
],
[
10874,
12797,
4
],
[
12797,
17136,
5
],
[
17136,
20181,
6
],
[
20181,
23502,
7
],
[
23502,
25117,
8
],
[
25117,
27848,
9
],
[
27848,
32104,
10
],
[
32104,
37073,
11
],
[
37073,
40981,
12
]
]
} |
7f8fe667132bcd1943a17c26e5bd972b4e6dbdf7 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 14,
"total-fallback-pages": 0,
"total-input-tokens": 32229,
"total-output-tokens": 10489,
"fieldofstudy": [
"Biology"
]
} | Chapter 6
How Far Are We from Dose On Demand of Short-Lived Radiopharmaceuticals?
Giancarlo Pascali and Lidia Matesic
Abstract PET radiopharmaceuticals are currently produced using a centralized approach, which makes sustainable the distribution to few imaging centers of an only small set of tracers (virtually only $^{18}$F FDG). However, a wider set of structures have demonstrated a potential applicability for imaging in a specific manner several disease condition. In order to allow this wider and more personalized use of PET imaging, the production paradigms need to be changed. In this contribution we will explain how Dose-On-Demand systems can be conceptualized and what are the challenges that are still to be overcome in order for such approach to be of widespread utility.
Keywords Dose On Demand • Microfluidics • PET • Radiochemistry
6.1 Introduction
The clinical production of radiopharmaceuticals or radiotracers for positron-emitting tomography (PET) is currently performed in centralized locations such as commercial radiopharmacies or some dedicated radiochemistry facilities. Generally, these facilities contain a cyclotron to produce the PET radioisotope and laboratories furnished with lead-shielded hot cells containing automated radiosynthesis modules to produce the radiotracer. Quality control equipment is also required to validate and confirm the purity of the radiotracer prior to its dispatch to the imaging centers.
The majority of radiotracer production facilities synthesize $^{18}$F FDG, the gold standard for detecting a variety of cancers. Nowadays, $^{18}$F FDG can be produced in a large batch, making it relatively affordable. Portions can then be dispatched and transported by road or air to the relevant hospital owing to the half-life of fluorine-18 (110 min).
A major challenge in clinical PET radiochemistry is that there are a greater number of hospitals or PET clinics than there are PET radiotracer production facilities. Furthermore, the demand for new clinical PET radiotracers is low due to the cost of production in a centralized location. New PET radiotracers are overwhelmingly used for research purposes only.
To overcome this obstacle, a decentralized approach has been envisaged [1]. Here, scientists could produce their radiotracer of interest in-house, economically and on demand, leading to a concept that we have defined as Dose On Demand (DOD). This short review will cover the important aspects of DOD and detail the journey toward the DOD of short-lived radiopharmaceuticals.
### 6.2 DOD Features
The current production approach of PET radiotracers imposes several limitations and challenges for guaranteeing the most efficient organization of imaging studies [2]. A possible way to improve this situation would require a system for which the type and the quantity of the produced tracer is defined and directly handled by an as final as possible user (e.g., hospital pharmacy, imaging laboratory). This system should implement the reduction to the minimum possible of the amounts of radioactivity and chemicals needed in the preparation, added to an overall simplification of the production process. Such conceptual process can be defined as “Dose On Demand” (DOD) [3, 4, 5]: the operation of producing a radiopharmaceutical in the shortest time possible, using the minimum amount of chemicals and radioactivity strictly needed for the production of the single (or few) imaging dose(s) required.
This approach, exemplified in Fig. 6.1, would provide several benefits to the overall PET community. Firstly, it will hand over flexibility in the application directly to the hospital/imaging center, which can decide on a patient basis which tracer to produce and when; this could also happen in small regional centers, thus not forcing anymore interested patients to commute long distances to the few useful imaging hospitals. This flexibility will also allow the utilization of rare or research tracers to be facilitated, as in this system the small amounts of chemicals and radioactivity needed for a DOD production would be economically sustainable for a single imaging center. Secondly, while a fault in production from a centralized approach will have impact on a large number of patients, a fault in one DOD system will have an impact limited to the patients utilizing those doses only. Lastly, due to the reduction in raw materials needed (as well as related topics, e.g., safety, storage) and the redistribution of running costs over more institutes, the imaging doses will result in a reduced cost and in tracers’ availability to a wider population.
In order to realize a DOD process, few requirements can be envisaged. Firstly, the production needs to be implemented on an automated instrument that can implement preset operations, as well as allowing remote interaction of the operator for minor modifications (i.e., “Automation”). In addition, it has to have
real-time monitoring and audit trail capabilities for monitoring and trending purposes. Secondly, the instrument used needs to be able to handle small aliquots of reagents, from fractions to hundreds of μL (i.e., “Discretization”) with accuracy and repeatability. The handling operations comprise moving, merging, mixing, heating, and similar processes. In other words, the instrument has to be capable to give a defined “chemical history” to any given aliquot of reagents used (that can be different among several aliquots). Thirdly, the processes implemented need to be serially repeatable with minimum operator intervention (“Restarting”). This can be achieved by substituting single-use parts in the system or by cleaning it using validated procedures. A final peculiarity that contributes to achieving DOD is the need to use the minimum amount of chemicals (“Reduction”), which would maximize the atom efficiency of the process, as well as allow an acceptable sacrifice in employing single-use parts or realize a faster cleaning of the system.
If all these requirements can be respected in a process system, a DOD production can be implemented and can be used to produce several doses of the same tracer or of different tracers, using the same system and minimal operator interventions.
### 6.3 Early Examples Conducible to DOD
The possibility to tailor the production of nuclear medicine (NM) tracers as much as possible to the needs of the final user has always been present, and indeed some example of approaches that can be linked to DOD concept can be found in early and current practices.
Generators of radioactive raw nuclides have been widely used (e.g., \(^{99m}\)Tc, \(^{68}\)Ga) \[6\]; the elution of the desired radionuclide is generally done upon demand and followed by simple chemical reactions (generally performed directly in the same NM department) to obtain the final radiopharmaceutical. This approach respects the Restarting requirement, though limited to the raw nuclide production and not to the pharmaceutical preparation; however, it generally does not respect the Discretization nor the Reduction requirements, as it is difficult to separately handle the amount of chemicals needed for a single patient (i.e., in the \(\mu\)Ls range). It sometimes respects the Automation requirement, but in most NM departments, these preparations are performed manually.
The use of very short half-life nuclides can be natively defined as a DOD application, as their handling must be done shortly before their use in imaging. Typical examples are the production of \([^{15}\text{O}]-\text{H}_2\text{O}\) \[7\] and \([^{13}\text{N}]-\text{NH}_3\) \[8\]; shortly after production, the systems utilized can be restarted easily due to their simple setup and the fast decay of process wastes. However, also in this case, these processes cannot be properly defined as DOD since they generally respect the Automation and the Restarting requirements, but neither the Discretization nor the Reduction (i.e., no difference if the production is used for one or several contemporary patients). \(^{11}\)C chemistry \[8\] also falls within the use of very short half-life nuclides as more productions can be run on the same machine on the same day. However, also in this case, the typical systems used can be even less defined as DOD, due to its relatively longer half-life (compared to \(^{15}\)O and \(^{13}\)N).
Currently, the approach that most resembles DOD is represented by the cassette-based systems. In this case, Automation and Restarting requirements can be easily achieved; Discretization and Reduction are generally not pursued, because nowadays these systems are used for single-batch productions, but the principles underlying cassette philosophy could be used to project single-use/single-dose cassettes. In fact, these systems are basically very compact macrofluidic systems; this understanding clarifies how microfluidic concepts can be the ones that should allow a full implementation of DOD in radiopharmaceutical production.
### 6.4 DOD Proof-of-Principle Examples
#### 6.4.1 Minicyclotron/Minichemistry/MiniQC
The Biomarker Generator (BG75), made by ABT Molecular Imaging, is a small (0.37 m × 1.25 m) self-shielded 7.5 MeV cyclotron coupled to an aseptic single-use card-based automated chemical production module and an automated module for quality control. The BG75 was initially used for the DOD production of \([^{18}\text{F}]\)fluoride and \([^{18}\text{F}]\)FDG \[9\]. Using the computer’s software, the operator is able to select whether the \([^{18}\text{F}]\)fluoride or \([^{18}\text{F}]\)FDG is to be produced. For the production of \([^{18}\text{F}]\)fluoride (~1 mCi/min at 5 \(\mu\)A), the process is complete once the product is delivered into the specified vial. Alternatively, for the production of \([^{18}\text{F}]\)FDG, the
software prompts the operator to prepare the tracer-specific Dose Synthesis Card (DSC) and the chemistry and quality control modules while the cyclotron is preparing the $[^{18}\text{F}]$fluoride. The radiosynthesis is then completed on the DSC, including relevant purification, and dispensed into a shielded, sterile syringe or vial. An aliquot of the product is removed for quality control (pH determination, acetonitrile and ethanol residual solvent determination, radiochemical identity and purity, Kryptofix 2.2.2 determination, and a filter integrity test), which is automatically performed by the system, without any operator input.
The BG75 has been able to consistently produce a 10–13 mCi dose of $[^{18}\text{F}]$FDG at 40 min intervals up to six times per day, with products meeting the required USP limits for release [9]. To date, other DOD radiotracers synthesized using the BG75 include Na$[^{18}\text{F}]$F [10] and $[^{18}\text{F}]$FMISO [11, 12].
6.4.2 Continuous Flow Microfluidics
Interestingly, proof-of-concept studies have been recently conducted into the production of $[^{18}\text{F}]$FLT using the cyclotron component of the BG75 system and the Advion NanoTek microfluidic system [13]. Between 70 and 80 mCi of $[^{18}\text{F}]$fluoride were produced by the minicyclotron and the radiosynthesis was subsequently performed under continuous-flow microfluidic conditions to yield $[^{18}\text{F}]$FLT in sufficient quantity and purity for clinical trials. The number of radiochemists using a microfluidic approach has been steadily accumulating in recent years. This may be related to the advantages of microfluidic systems over traditional automated radiochemistry modules, which include a decrease in the amount chemical reagents used, shorter reaction times, greater radiochemical yields, the ability to use solvents under supercritical conditions, and reduced radiation exposure to the operator due to the lower amounts of radioactivity used. The NanoTek Microfluidic Synthesis System by Advion was the first commercially available continuous-flow microfluidic system. The system comprises a concentrator module to azeotropically evaporate the $[^{18}\text{F}]$fluoride from the cyclotron and subsequently reconstitute the isotope into the appropriate solvent; a pump module containing two syringe pumps and loops to store chemical precursors and a reactor module, which contains a syringe pump and loop to house the isotope along with thermostatted slots to store up to four microreactors, where the radiochemical reactions occur.
The previous example of the production of $[^{18}\text{F}]$FLT is the latest in a growing list of radiotracers prepared using the NanoTek system. The first instance was the production of $[^{18}\text{F}]$fallypride for use in micro-PET studies [14]. Initially, the radiochemical optimization of $[^{18}\text{F}]$fallypride was conducted by dispensing 10 μL solutions of the tosyl-fallypride precursor and $[^{18}\text{F}]$fluoride complex into the microreactor at 10 μL/min to obtain 1–1.5 mCi doses of $[^{18}\text{F}]$fallypride. These optimization reactions were performed sequentially and could be considered an early form of DOD. Once the optimal radiochemical conditions were determined, the authors were able to prepare a dose of $[^{18}\text{F}]$fallypride sufficient for human injection (15 mCi) by increasing the volume of the two solutions from 10 μL to
200 μL. The authors also alluded to the fact that multiple high doses of $^{18}$F fallypride could be produced using the same microreactor. Soon after, Pascali et al. [15] described the sequential radiolabeling of ethyl-ditosylate and propyl-ditosylate in the NanoTek system using the same solution of $^{18}$F fluoride complex and swapping the precursor between productions by emptying and refilling the precursor loop with a different substrate. These examples of DOD demonstrated the economical use of the $^{18}$F fluoride solution to yield two radiotracers on the same day. The authors also sequentially prepared several injectable doses of $^{18}$F CB102, a cannabinoid type 2 receptor agonist, for small animal PET imaging, suggesting that freshly prepared doses using a DOD approach were superior to a batch solution to be used over a certain shelf life.
To further evaluate the robustness and reliability of a DOD approach, Pascali et al. were able to produce three sequential doses of three different $^{18}$F fluorocholines with a total processing time of 13–15 min for each dose, including SPE purification [5]. While this example includes a modification to the NanoTek system to incorporate SPE purification, typically, the radiochemical outputs are purified externally to the NanoTek system, particularly HPLC purification. Recent examples include the preparation of $^{18}$F FPEB [16], whereby the radiosynthesis occurred in the NanoTek and the reaction output was sent to a vial preloaded with water and pre-concentrated onto an Oasis HLB Light SPE cartridge to remove DMSO present in the reaction mixture. The cartridge was eluted with acetonitrile and water before being transferred to a GE TRACERlab F$_X$F$_N$ synthesis unit to conduct semi-preparative HPLC purification and formulation. Additionally, the Tau imaging agent, $^{18}$F T807, was produced with the same modifications [17]. Three consecutive >100 mCi productions of $^{18}$F T807 were performed for validation purposes, and $^{18}$F T807 became the first example of human use of a radiopharmaceutical prepared by continuous-flow microfluidics.
The NanoTek system has been modified recently to include HPLC and SPE purification [18]. By utilizing the cable harnessing of the system, a custom-made electrical board was engineered whereby additional switches and analog signals could be added and be controlled by the NanoTek software to activate externally powered devices and record external signals (e.g., detectors), if applicable. This customized system was able to produce 1- or 2-step radiotracers such as $^{18}$F CB102, $^{18}$F fluoroethylcholine [18], $^{18}$F MEL050 (melanin targeting) [19], $^{18}$F fallypride, and $^{18}$F PBR111 (TSPO receptor) [20], in a DOD manner. Similarly, $^{18}$F FMISO has been produced by integrating a HPLC system to the NanoTek through a six-port valve [21]. By fine-tuning the HPLC conditions for $^{18}$F FMISO, the authors were able to eliminate the requirement for SPE.
### 6.4.3 Peptide Labeling
While microfluidic systems have mainly been utilized to radiolabel small molecules, reports of peptide or protein radiolabeling using microfluidics are limited.
Early work in this area featured the direct $^{18}\text{F}$radiolabeling of bombesin derivatives (with 7–8 amino acid residues) that had been modified to incorporate trimethylammonium or triarylsulfonium leaving groups [22]. The peptides could be radiolabeled reproducibly, suggesting a possible DOD approach; however, due to the harsh temperature conditions required for radiolabeling, this method would be unsuitable for protein radiolabeling. An alternative route to radiolabel a peptide is through the use of a prosthetic group as an indirect radiolabeling method. $^{18}\text{F}$SFB [23] and even, the most abundantly used PET tracer, $^{18}\text{F}$FDG [24] have been utilized as prosthetic groups for the radiolabeling of peptides. Although both prosthetic groups were synthesized on macroscale equipment, the subsequent peptide radiolabeling was performed under microfluidic conditions. In each case, the peptide was radiolabeled in a shorter period of time, in higher radiochemical yield (RCY), and using a smaller quantity of the peptide compared to conventional radiolabeling techniques. Only recently has the first microfluidic radiosynthesis of a prosthetic group and the ensuing peptide radiolabeling been reported [25]. Here, the $^{18}\text{F}$F-Py-TEP prosthetic group was prepared in the first microreactor of an Advion NanoTek system from $^{18}\text{F}$fluoride and the corresponding precursor. After exiting the microreactor, the $^{18}\text{F}$F-Py-TEP was transferred to a second microreactor, where it reacted with a model peptide containing free amines. Once again, the peptide coupling was faster than conventional methods and obtained in higher RCY. These accounts all imply that the DOD of radiolabeled peptides for molecular imaging is currently being explored and may be employed in the future.
6.4.4 Solid-Phase Approaches
Although the use of microfluidic conditions is leading to radiochemical reactions being completed in less time than traditional approaches, to further decrease the overall radiochemical processing times, new methods are required to decrease or eliminate the time taken to process and activate the starting $^{18}\text{F}$fluoride. One option is to trap the $^{18}\text{F}$fluoride onto a resin and subsequently perform on-resin radiofluorinations, thus eliminating the need for azeotropic evaporations and re-solubilization of the $^{18}\text{F}$fluoride complex. Reusable polymer-supported phosphazenes have been investigated as suitable resins to perform the $^{18}\text{F}$fluoride trapping and radiofluorination [26]. The PS-$\text{P}_2^{\text{tBu}}$ resin was able to trap >99% of $^{18}\text{F}$fluoride, with no leaching of activity was observed when the column was subsequently dried with helium gas. It was found that substrates with sulfonate leaving groups resulted in the highest RCY when subjected to on-column radiofluorination. The same phosphazene resin could be recycled at least three times using the same substrate, or at least two times using a different substrate, which implies that the DOD production of radiotracers is possible through solid-phase radiofluorination.
Other work in this area includes a continuous-flow system comprising a polystyrene-imidazolium-chloride (PS-Im$^+\text{Cl}^-$) monolith which traps $^{18}\text{F}$fluoride [27]. A solution of base and the relevant precursor could then be flowed through the
PS-Im$^+\left[{ }^{18}\text{F}\right] F^-$ monolith into a preheated microfluidic chip where the radiochemical reaction takes place. The advantage of this method is that the entire process is performed in continuous flow and the microfluidic platform has a very small footprint compared to current processes.
### 6.4.5 Droplet Systems
An interesting extension in the field of microfluidic radiochemistry is through the use of droplets. Also sometimes referred to as segmented flow chemistry, it features droplets (nL-μL) which are separated by an immiscible carrier fluid, similar to oil droplets in water. Droplets can be thought of as individual nano- or microreactors and can be used to aid radiolabeling optimization, whereby each droplet is the result of a predetermined set of reaction parameters. Droplets consisting of approximately 120 nL were formed during the coupling of \( [18\text{F}]\text{FSB} \) with an anti-prostate stem cell antigen diabody [28]. Using a 5 μL sample of the diabody solution was sufficient to screen over 100 different reaction conditions using the droplets, and hence, the optimal reaction conditions were determined rapidly with minimal use of the precious diabody solution. Droplet systems have also been utilized in electrowetting-on-dielectric (EWOD) devices. In EWOD systems, droplets are sandwiched between two plates; the bottom plate consists of electrode pads to manipulate the movement of the droplets throughout the microchip, while the top plate electrically grounds the droplets. The EWOD chip was first used in the synthesis of \( [18\text{F}]\text{FDG} \) [29], but its use has more recently expanded to include \( [18\text{F}]\text{FLT} \) [30], \( [18\text{F}]\text{fallypride} \) [31], and \( [18\text{F}]\text{SFB} \) [32]. While the EWOD chip produces these radiolabeled molecules in comparable, if not greater, RCY than previously, drawbacks of the system include off-chip purification and the potential for radioactivity and volatile side products to escape since the chip is exposed to air. It is envisaged that with advances in technology, the EWOD chip could be further automated, be disposable, and lead to scientists producing their desired radiotracer on demand.
### 6.5 Challenges and Future of DOD
As it can be seen, a perfect DOD system is still not existent, but several data are available demonstrating that such approach should represent the reality in the next future. However, to witness this paradigmatic change, several challenges need to be fronted, and they will represent the future of DOD research in radiopharmaceutical production.
One of the biggest challenges is to understand whether one only system could achieve the desired spread of operations that a DOD process should implement. This should cover not only the production steps but also the switching of chemicals, cleaning, and priming steps. It is very likely that these systems will be based on
micronized approaches (e.g., microfluidics, nanodroplets), but understanding which philosophy they should implement is still under discussion. For example, a system can be projected to implement several different preexisting routes, which would lead to different products in different quantities, or, on the contrary, be represented by a fixed framing to which flexibly interface single-use components/modules (i.e., similar to microcassettes) to build up the desired process. Another possibility might also be represented by the possibility to use the exact same system into which different chemicals are delivered, depending on the production needs. All these options are amenable to deliver a DOD system, but the choice of one or the other will drive the final performance and actual ease/flexibility of use.
Even once the underlying philosophy is clarified, some technical problems are still unsolved or partially addressed. Purification of the finished product represents probably the most important issue, and while there are several excellent systems to perform chemical reactions, there is a notable lack in miniaturization of purification methods or their interfacing with micronized chemistry systems. Some research is now available on micro-chromatographic systems [33], mainly facilitated by the advancements of monolith polymers [34] that can be easily integrated with micron-sized channels/reservoirs [35]. These solutions allow the reduction of inherent void volumes, therefore improving the atom efficiency in the purification process. Also, similar solutions may be useful for the cases in which a simple solid-phase extraction (SPE) would be sufficient to purify the relevant molecule [36, 37]. Polymer chemistry advancement possibly represents the field where useful innovations can have a relevant impact on miniaturized purifications. As an example, molecularly imprinted polymers (MIP) represent a promising approach that would allow to streamline the selective separation of the molecule of interest and its efficient elution [38]. MIP structures are prepared by building the polymer pores around a desired template molecule; once the template is removed, the material acquires selectivity of shape and electronic interaction (i.e., with functional groups of polymer) for the desired molecule [39]. MIP systems are in fact also referred as “synthetic receptors.”
Another innovation that would be generally useful in radiochemistry but particularly applicable to DOD systems (due to their preferred micronized nature) is the use of supported precursors. These systems should be projected in such a way that the labeling reaction is the only event that would make the molecular structure to become free from the solid support bond. In this way, no other complex organic species will be present in the resulting mixture, and only simple filtration and reformulation steps would be required in order to retrieve the radiopharmaceutical. Some systems based on supported sulfonates [40] or triazene [41] have been reported up to date, and patent literature also refers to examples of supported ionic precursors (e.g., iodonium compounds) [42]. However, none of these systems have demonstrated a preferential use compared to traditional methods, probably because of the mismatch between the support active surface and reagent accessibility to it. The use of micronized systems could be beneficial in solving this mismatch, and indeed the use of sulfonate precursors supported on a monolith structure grown directly in a microfluidic chip gave satisfying yields of radiofluorination [43].
A further modification to this approach, which would facilitate the respect of DOD requirements, could be the use of reversibly linked precursors. In this concept, the precursor should form a bond (e.g., covalent coordination) with the support material, and as usual, the labeling reaction should be able to selectively cleave the structure from the support; however, in a second “recycling” step, a change in conditions will allow the recovery of the precursor out of the system and offer a support free to be reversibly functionalized again with a different precursor. A recent paper reported the catalysis by TiO$_2$ nanoparticles in radiofluorinating a tosylate precursor [44]; interestingly enough, the authors suspect that the process is catalyzed due to selective coordination of the tosylate moiety with the titania surface, therefore opening to the idea of a reversible functionalization of metal nanoparticles with several different precursors prior to radiolabeling. Another possibility, drawn from the field of self-healing materials, might be the use of reversible click reactions, a nice example of which is represented by the 1,2,4-triazoline-3,5-dione chemistry (TAD) moieties. This structure reacts in a reversible way (using different temperatures, Fig. 6.2) with indoles through an ene click reaction [45]; however, it undergoes fast Diels-Alder reactions with dienes and is
Fig. 6.2 TAD residue in reversible click-chemistry transformations (Taken from [45])
widely used in biology for its capacity to bind irreversibly tyrosine residues [46]. Such an approach could be very useful in DOD processes aimed at protein labeling, for which radiolabeled prosthetic groups enabling click chemistry are now widely employed [47].
Another important point to clarify is whether DOD systems should produce product vials (as in the traditional approach), a syringe/cartridge dose, or even directly deliver the radiopharmaceutical preparation into the vein of the subject (see Fig. 6.3). Though currently unlikely, the possibility to overcome the concept of product vial is very appealing on the base of flexibility, atom efficiency, and procedure streamlining. Therefore, an outstanding challenge is represented by modifying the regulators’ view [48, 49] on the requirements needed to prepare injectable radiopharmaceuticals for human use, in order to allow easier and more personalized modalities of dose delivery. One of the ways to achieve such result is represented by the change in quality control (QC) paradigms; in fact, the traditional way to produce a separate vial for QC [50] should be overtaken by the possibility of realizing a DOD process whose precise control and monitoring would represent itself a guarantee of good-quality end product.
6.6 Conclusions
Miniaturization and optimization of the biochemical hardware involved have created a substantial personalization of several medical practices. A typical example of this trend that has improved the treatment of diabetic subjects is the current possibility for any person to check their glucose levels using a straightforward handheld system, instead of reaching the nearest hospital and performing a proper blood examination. This level of simplicity, flexibility, and personalization is currently lacking in the important field of radiopharmaceutical production. However, several studies are starting to demonstrate that new chemical technologies (e.g., microfluidics, high-tech polymers) can represent useful tools to achieve what we can define Dose-On-Demand systems. Several challenges are still to be faced before reaching such a useful target in an efficient and affordable way; we however think that the realization of this capability will be the main way to allow the use of rare and disease dedicated tracers whose widespread utilization is currently hindered [51] by the existing radiopharmaceutical production paradigms.
Open Access This chapter is distributed under the terms of the Creative Commons Attribution-Noncommercial 2.5 License (http://creativecommons.org/licenses/by-nc/2.5/) which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.
The images or other third party material in this chapter are included in the work’s Creative Commons license, unless indicated otherwise in the credit line; if such material is not included in the work’s Creative Commons license and the respective action is not permitted by statutory regulation, users will need to obtain permission from the license holder to duplicate, adapt or reproduce the material.
References
1. Keng PY, Esterby M, van Dam RM. Emerging technologies for decentralized production of PET tracers. In: Positron emission tomography – current clinical and research aspects. 2012. doi:10.5772/1280
2. Satyamurthy N. Electronic generators for the production of positron-emitter labeled radiopharmaceuticals where would PET be without them? Clin Positron Imaging. 1999;2:233–53.
3. Wang M-W, Lin W-Y, Liu K, Masterman-Smith M, Kwang-Fu Shen C. Microfluidics for positron emission tomography probe development. Mol Imaging. 2010;9:175–91.
4. Arima V, Pascali G, Lade O, et al. Radiochemistry on chip: towards dose-on-demand synthesis of PET radiopharmaceuticals. Lab Chip. 2013;13:2328–36.
5. Pascali G, Nannavecchia G, Pitzianti S, Salvadori PA. Dose-on-demand of diverse 18F-fluorocholine derivatives through a two-step microfluidic approach. Nucl Med Biol. 2011;38:637–44.
6. Knapp FF (Russ., Mirzadeh S. The continuing important role of radionuclide generator systems for nuclear medicine. Eur J Nucl Med. 1994;21:1151–65.
7. Welch MJ, Kilbourn MR. A remote system for the routine production of oxygen-15 radiopharmaceuticals. J Label Compd Radiopharm. 1985;22:1193–200.
8. Miller PW, Long NJ, Vilar R, Gee AD. Synthesis of 11C, 18F, 15O, and 13N radiolabels for positron emission tomography. Angew Chem Int Ed Engl. 2008;47:8998–9033.
9. Awasthi V, Watson J, Gali H, Matlock G, McFarland A, Bailey J, Anzellotti A. A “dose on demand” biomarker generator for automated production of [(18)F]F(-) and [(18)F]FDG. Appl Radiat Isot. 2014;89:167–75.
10. Awasthi V, Gali H, McFarland A, Anzellotti A. Automated manufacture of Na[18F]F at the University of Oklahoma Health Sciences Center (OUHSC). Dose on-demand and imaging studies. J Nucl Med. 2014;55:1243.
11. Anzellotti A, Bailey J, Ferguson D, McFarland A, Bochev P, Andreev G, Awasthi V, Brown-Proctor C. Automated production and quality testing of [18F]labeled radiotracers using the BG75 system. J Radioanal Nucl Chem. 2015;305:387–401.
12. Anzellotti A, Yuan H. Automated manufacture of [18F]FMISO in the BG 75 system. Synthesis and purification using solid phase extraction. J Nucl Med. 2015;56:1001.
13. Gali H, Nkepang G, Galbraith W, Hammond K, Awasthi V, Collier TL. Preparation of [F-18] FLT using an ABT compact cyclotron and continuous flow microfluidics. J Label Compd Radiopharm. 2015;58 Suppl 1:S362.
14. Lu S, Giamis AM, Pike VW. Synthesis of [F]fallypride in a micro-reactor: rapid optimization and multiple-production in small doses for micro-PET studies. Curr Radiopharm. 2009;2:49–55.
15. Pascali G, Mazzone G, Saccomanni G, Manera C, Salvadori PA. Microfluidic approach for fast labeling optimization and dose-on-demand implementation. Nucl Med Biol. 2010;37:547–55.
16. Liang SH, Yokell DL, Jackson RN, Rice PA, Callahan R, Johnson KA, Alagille D, Tamagnan G, Collier TL, Vasdev N. Microfluidic continuous-flow radiosynthesis of [(18)F] FPEB suitable for human PET imaging. Medchemcomm. 2014;5:432–5.
17. Collier T, Yokell D, Rice P, Jackson R, Shoup T, Normandin M, Brady T, El Fakhri G, Liang S, Vasdev N. First human use of the tau radiopharmaceutical [18F]T807 by microfluidic flow chemistry. J Nucl Med. 2014;55:1246.
18. Pascali G, Berton A, DeSimone M, Wyatt N, Matesic L, Greguric I, Salvadori PA. Hardware and software modifications on the AdVion NanoTek microfluidic platform to extend flexibility for radiochemical synthesis. Appl Radiat Isot. 2014;84:40–7.
19. Matesic L, Kalilnen A, Wyatt NA, Pham TQ, Greguric I, Pascali G. [18F]Fluorination optimisation and the fully automated production of [18F]MEL050 using a microfluidic system. Aust J Chem. 2015;68:69–71.
20. Matesic L, Kallinen A, Greguric I, Pascali G. Dose-on-demand production of multiple 18F-radiotracers on a microfluidic system. J Label Compd Radiopharm. 2015;58 Suppl 1:S22.
21. Zheng M-Q, Collier L, Bois F, Kelada OJ, Hammond K, Ropchan J, Akula MR, Carlson DJ, Kabalka GW, Huang Y. Synthesis of [(18)F]FMISO in a flow-through microfluidic reactor: development and clinical application. Nucl Med Biol. 2015;42:578–84.
22. Selivanova SV, Mu L, Ungersboeck J, Stellfeld T, Anemey SM, Schibli R, Wadsak W. - Single-step radiofluorination of peptides using continuous flow microreactor. Org Biomol Chem. 2012;10:3871.
23. Richter S, Bouvet V, Wuest M, Bergmann R, Steinbach J, Pietzsch J, Neundorf I, Wuest F. (18)F-Labeled phosphopeptide-cell-penetrating peptide dimers with enhanced cell uptake properties in human cancer cells. Nucl Med Biol. 2012;39:1202–12.
24. Bouvet VR, Wuest F. Application of [18F]FDG in radiolabeling reactions using microfluidic technology. Lab Chip. 2013;13:4290–4.
25. Cumming RC, Olberg DE, Sutcliffe JL. Rapid 18 F-radiolabeling of peptides from [18 F]fluoride using a single microfluidics device. RSC Adv. 2014;4:49529–34.
26. Mathiessen B, Zhuravlev F. Automated solid-phase radiofluorination using polymer-supported phosphazenes. Molecules. 2013;18:10531–47.
27. Ismail R, Irribaren J, Javed MR, Machness A, Michael van Dam R, Keng PY. Cationic imidazolium polymer monoliths for efficient solvent exchange, activation and fluorination on a continuous flow system. RSC Adv. 2014;4:25348.
28. Liu K, Lepin EJ, Wang M-W, et al. Microfluidic-based 18F-labeling of biomolecules for immuno positron emission tomography. Mol Imaging. 2011;10:168–76.
29. Keng PY, Chen S, Ding H, et al. Micro-chemical synthesis of molecular probes on an electronic microfluidic device. Proc Natl Acad Sci U S A. 2012;109:690–5.
30. Javed MR, Chen S, Kim H-K, Wei L, Czernin J, Kim C-JCJ, van Dam RM, Keng PY. Efficient radiosynthesis of 3'-deoxy-3'-18F-fluorothymidine using electrowetting-on-dielectric digital microfluidic chip. J Nucl Med. 2014;55:321–8.
31. Javed MR, Chen S, Lei J, Collins J, Sergeev M, Kim H-K, Kim C-J, van Dam RM, Keng PY. High yield and high specific activity synthesis of [18F]fallypride in a batch microfluidic reactor for micro-PET imaging. Chem Commun (Camb). 2014;50:1192–4.
32. Chen S, Javed MR, Kim H-K, Lei J, Lazari M, Shah GJ, van Dam RM, Keng P-Y, Kim C-JCJ. Radiolabelling diverse positron emission tomography (PET) tracers using a single digital microfluidic reactor chip. Lab Chip. 2014;14:902–10.
33. Ali I, Aboul-Enein HY, Gupta VK. Recent advances in silica-based monoliths: preparations, characterizations and applications. J Sep Sci. 2011;34:1945–57.
34. Wouters B, De Vos J, Desmet G, Terryn H, Schoenmakers PJ, Eeltink S. Design of a microfluidic device for comprehensive spatial two-dimensional liquid chromatography. J Sep Sci. 2015;38:1123–9.
35. Zhou D, Chu W, Peng X, McConathy J, Mach RH, Katzenellenbogen JA. Facile purification and click labeling with 2-[18F]fluoroethyl azide using solid phase extraction cartridges. Tetrahedron Lett. 2015;56:952–4.
36. Fedorova O, Kuznetsova O, Stepanova M, Maleev V, Belokon Y, Wester H-J, Krasikova R. A facile direct nucleophilic synthesis of O-(2-[18F]fluoroethyl)-l-tyrosine ([18F]FET) without HPLC purification. J Radioanal Nucl Chem. 2014;301:505–12.
37. Turiel E, Martín-Esteban A. Molecularly imprinted polymers for sample preparation: a review. Anal Chim Acta. 2010;668:87–99.
38. Ismail R, Machness A, van Dam RM, Keng P-Y. Solid-phase [18F]fluorination on a flow-through glass microfluidic chip. In: 16th int. conf. miniaturized syst. chem. life sci. 2012. Okinawa, Japan, p. 629–631.
39. Brown LJ, Bouvet DR, Champion S, et al. A solid-phase route to18F-labeled tracers, exemplified by the synthesis of [18F]2-Fluoro-2-deoxy-D-glucose. Angew Chemie. 2007;119:959–62.
40. Billiet S, De Bruycker K, Driessen F, Goossens H, Van Speybroeck V, Inokuma T, Barbas CF. Facile and stabile linkages through tyrosine: bioconjugation strategies with the tyrosine-click reaction. Bioconjug Chem. 2013;24:520–32.
41. Kim DW. Bioorthogonal click chemistry for fluorine-18 labeling protocols under physiologically friendly reaction condition. J Fluor Chem. 2015;174:142–7.
42. Salvadori AP. Radiopharmaceuticals, drug development and pharmaceutical regulations in Europe. Curr Radiopharm. 2015;1:7–11.
43. Ban H, Nagano M, Gavrilyuk J, Hakamata W, Inokuma T, Barbas CF. Facile and stable linkages through tyrosine: bioconjugation strategies with the tyrosine-click reaction. Bioconjug Chem. 2013;24:520–32. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
1811,
1
],
[
1811,
4945,
2
],
[
4945,
6548,
3
],
[
6548,
9822,
4
],
[
9822,
13251,
5
],
[
13251,
16442,
6
],
[
16442,
19847,
7
],
[
19847,
22786,
8
],
[
22786,
26392,
9
],
[
26392,
27872,
10
],
[
27872,
29156,
11
],
[
29156,
32369,
12
],
[
32369,
36520,
13
],
[
36520,
39299,
14
]
]
} |
f59b453da9bcafb11b7d0f56b782a79dee34963e | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 5,
"total-fallback-pages": 0,
"total-input-tokens": 17341,
"total-output-tokens": 4723,
"fieldofstudy": [
"Business"
]
} | Research on Legal Issues and Legislative Countermeasures of Cross-border E-Commerce in China
TingTing Yu¹,
¹School of Law, Ocean University of China, Qingdao, Shandong, China
Corresponding author. Email: [email protected]
ABSTRACT
In recent years, China's leadership in the field of e-commerce has gradually increased, and import and export retail cross-border e-commerce platforms have been established one after another, driving the scale of China's import and export retail cross-border e-commerce transactions to continue to grow steadily. Based on the analysis of the current legal issues of China's cross-border retail e-commerce exports, this paper further investigates its legislative countermeasures and recommendations, aiming to promote in-depth research in this field, so that the results can better serve the benign development of China's cross-border e-commerce and contribute China's strength to the development of the world economy.
Keywords: Cross-border, e-commerce, legal issues, legislation, countermeasures
1. INTRODUCTION
Along with the development of FTZ, China's cross-border e-commerce has gradually entered a period of rapid development, with a huge expansion from transaction scale, transaction category to transaction amount. The development of cross-border e-commerce has brought tremendous changes both for the country, for enterprises and for consumers. For countries, it has broken down the barriers between countries and made international trade possible to become borderless trade. For enterprises, the development of cross-border e-commerce has broadened the path to international markets and multiplied the potential customer base. For consumers, the development of cross-border e-commerce gives them the opportunity to buy goods from other countries without having to leave home. Cross-border e-commerce is growing rapidly, but compared to traditional online shopping, it involves multiple countries and more complex intermediate links. As more consumers purchase goods from abroad through cross-border e-commerce, a series of problems such as substandard product quality, long delivery time and return service have emerged [2], which are reflected in the inadequate legal policies on cross-border e-commerce. This paper will be based on this further research.
2. ANALYSIS OF THE CURRENT LEGAL SITUATION AND PROBLEMS OF CROSS-BORDER E-COMMERCE IN CHINA
2.1. Pre-trade
2.1.1. Government regulatory mechanism
The current legal regulatory model is not well adapted to the demands of small and medium-sized enterprises, supporting laws and regulations and supervision and management system does not keep pace with the development of the times [3]. First, from the perspective of government services to enterprises, for example, in the customs clearance and release link, enterprises need to declare more items, procedures are also more complex, many of the declared content often cross overlap, but also can not achieve complete paperless customs clearance, customs clearance efficiency is low, the relevant laws and regulations do not make specific provisions in how to promote efficient customs clearance; Second, the regulatory standards of the cross-border market are not determined in the form of law down, for example, the coordination mechanism of customs, commodity inspection, industry and commerce, taxation and other departments is not yet sound, and a mechanism for information sharing and
efficient supervision among competent governments has not yet been established, and there are inconsistencies in the documents of a normative nature formulated by various departments.
2.1.2. Market access and platform regulation
Market access and platform regulation is mainly the legislative regulation of the identity and credit recognition of the market subjects, platform access, platform rule-making, and management of trading subjects [4]. The Measures for the Administration of Network Transactions stipulate the information that natural persons should provide to the platform for transactions through network third-party platforms and the information that legal persons need to disclose on the home page of the network, and provide certain protection for the legitimate rights and interests of consumers [5]. The platforms mainly monitor the transactions in real time according to the rules they set and reasonably resolve the disputes according to the corresponding countermeasures introduced. However, at present, China has not introduced a set of specialized and strict laws and regulations for the management of cross-border platforms.
2.1.3. Information Security Protection
In terms of information security protection, China has not yet introduced a national law on data privacy protection, and the legislation for the protection of personal information on the Internet is mainly scattered in the "Electronic Signature Law", "Tort Liability Law", "Public Security Management Punishment Law" and other sectoral laws.
2.2. Trading session
2.2.1. Payment and Settlement
There are various payment methods for international trade, and with the development of cross-border e-commerce retail model, the use of third-party online payment platforms such as Paypel and Ali Secure Payment is becoming more and more common. Third-party payment refers to the use of non-bank third-party institutions to connect banks, sellers and consumers by using the credit guarantee and technical guarantee functions of payment platforms to realize monetary payment and flow, settlement and other functions for each subject. Cross-border third-party payment is a new industry that has emerged in response to the rapid development of the cross-border e-commerce retail industry, and the payer and receiver of its funds are generally not in the same country [6]. As an important support link for the development of cross-border e-commerce and a highly promising growth point for the financial industry, it has developed relatively rapidly in recent years, and there is a greater need to regulate its unreasonable phenomenon by legal means.
2.2.2. Logistics and Circulation
In general, the logistics mode of cross-border e-commerce retail export should be adapted to the characteristics of "small volume, multi-frequency and low amount" of parcels. At present, among the relevant documents issued in the field of logistics and circulation, "Opinions on Accelerating the Development of E-commerce in Circulation" puts forward the development goals of cross-border e-commerce logistics; "Opinions on Implementing Policies to Support Cross-border E-commerce Retail Exports" puts forward relevant solutions in the field of circulation such as customs, inspection and quarantine; "Announcement on Supervision of Goods and Articles Imported and Exported in Cross-border Trade" The corresponding requirements are put forward for various aspects of import and export supervision of cross-border items, especially in the areas of enterprise registration and filing, cargo management, etc. In view of the analysis of the above prior regulatory documents, there are still many problems that need to be solved in the legal and regulatory aspects.
2.2.3. Taxation
At present, China mainly relies on the Customs Law, the Personal Income Tax Law, the Tax Collection and Administration Law and other laws to implement the taxation of cross-border e-commerce. The Announcement on Matters Relating to the Supervision of Inbound and Outbound Goods and Articles for Cross-border Trade in Electronic Commerce published by China Customs stipulates the conditions for the tax exemption procedures for enterprises' "goods list" and individuals' "articles list" [7]. The Notice on Adjustment of Management Measures for Inbound and Outbound Personal Postal Items stipulates the tax exemption standards for goods sent by individuals to different countries and regions, as well as the return of goods exceeding the limit for non-personal use or customs clearance procedures as required for goods. The Opinions on Implementing Policies Related to Supporting Cross-border E-commerce Retail Exports specifically proposes to develop a new taxation system for Chinese cross-border export enterprises and to develop policies on export tax rebates, value-added tax and tax exemption conditions for the subject by the taxation bureau, the Ministry of Finance and other relevant departments. The Notice on Taxation Policies for Cross-border E-Commerce Retail Exports proposes relevant taxation rules on the conditions for refund and exemption of consumption tax, VAT, etc. for exporters. However, on the whole, these laws tend to be broad and cannot yet effectively adapt to the development of the
new economy, and China has not yet promulgated any tax laws specifically for cross-border e-commerce exports.
2.3. Post-trade
2.3.1. Intellectual Property Disputes
When conducting cross-border transactions, infringement of trademark rights, patent rights and copyrights are often encountered, leading to intellectual property disputes. China, as the original world's major manufacturing production and OEM country, had a lot of copycat products and repeatedly encountered intellectual property infringement problems [8]. The explosive development of cross-border e-commerce conforms to the development trend of the times, but is also related to its relatively low entry threshold, which has led to the uneven quality of many products on the platform and frequent incidents of counterfeiting or causing foreign intellectual property disputes. On the one hand, this reflects the lack of understanding of the concept of branding development of Chinese products and the IPR protection system in the target market when they go abroad. On the other hand, the imperfection of the law in the field of IPR may not be a deterrent for Chinese cross-border exporters and encourage their violations.
2.3.2. After-sales disputes
As disputes in cross-border e-commerce often arise between transaction subjects of two countries, and China currently lacks relevant independent laws to deal with cross-border electronic disputes, when disputes arise, they can generally only be resolved by invoking relevant legal provisions already in place in China. In terms of substantive law, the "Measures for the Administration of Internet Transactions" requires platforms to play a role in dispute resolution, and platforms should build their own dispute resolution and rights maintenance systems, participate in mediating disputes and support consumers to defend their legitimate rights and interests [9]. The Contract Law also contains relevant provisions involving dispute handling, of which Article 126(1) provides that parties to a foreign-related contract may choose for themselves the relevant law applicable to the resolution of the dispute. In terms of procedural law, the Law on the Application of Foreign-related Civil Relations Law provides that the parties in a foreign-related dispute may choose the law to be used by themselves, and also gives some explanations on the exceptions, in particular, the attribution of intellectual property rights in the dispute is also provided for [10]. In terms of after-sales disputes, although the overall law is relatively sound, there is a lack of after-sales protection system for consumers, and the law is unclear about the responsibility of the platform.
3. RESEARCH ON LEGISLATIVE BASIS
3.1. Purpose of Legislation
The purpose of the legislation is to effectively regulate the behavior and the steps of the whole process of the cross-border e-commerce retail export industry, to combat information leakage, infringement of intellectual property rights, tax evasion, counterfeiting and sale of fakes and other behaviors that harm the interests of all parties; to effectively mediate transaction disputes, avoid payment and other risks, to protect all participating parties to enjoy certain rights and actively fulfill the relevant obligations; to jointly maintain market fairness, protect the legitimacy of the relevant interests and claims, and take measures to promote trade facilitation, so that it will develop in a more standardized and rational direction; and to create a good environment for China's cross-border e-commerce retail export industry.
3.2. Legislative Principles
Legislation in the field of cross-border e-commerce retail export in China should uphold the principles of safeguarding national interests, fairness and impartiality, and focusing on efficiency. As the subjects of cross-border trade involve transactions across countries, national interests should be safeguarded without violating international rules, for example, international tax jurisdiction and territority should be adhered to in terms of tax laws. At the same time, each type of group should be treated fairly and impartially, for example, Class B and Class C sellers should be taxed together. In specific operations, attention should be paid to efficiency, for example, the coordination and unification of various departments and data sharing should be enhanced to improve customs clearance efficiency.
4. LEGISLATIVE COUNTERMEASURES RESEARCH AND RECOMMENDATIONS
In 2013, the EU Joint Research Center had pointed out in the Study Report on the Drivers and Impediments of Cross-border E-Commerce released that the absence of relevant laws and regulations would lead to the unavailability of cross-border e-commerce disputes, increase the difficulty of operation and hinder its development, so the relevant legislative work should be carried out in a timely manner. According to the practice of cross-border e-commerce in China and the problems that arise in its development, relevant laws, regulations, rules and treaties should be formulated in various aspects to implement the legislative work, and the relevant provisions of international organizations
such as WTO should be effectively used to resolve disputes. As cross-border e-commerce retail export involves various aspects such as market access, payment, taxation and after-sales, each aspect has its own legal issues specificity, while each aspect also has universal common problems. Therefore, based on the comprehensive analysis of the above-mentioned parts, the following suggestions on the legislation of cross-border e-commerce retail export are proposed.
4.1. Update legal terms
First, a trial single law, revision of the original legal provisions or new judicial interpretation should be introduced. Due to the rapid development of cross-border e-commerce and its special characteristics, consideration can be given to the formulation of a single law to adapt to its development characteristics or through the timely revision of relevant laws, the introduction of new judicial interpretation to deal with the contradiction that the lagging nature of the law can not keep pace with the development of cross-border e-commerce. For example, all the contents related to cross-border e-commerce business (market access and withdrawal of cross-border enterprises, cross-border intellectual property protection, cross-border dispute resolution mechanism, product quality standard determination, cross-border payment and logistics, cross-border data messages and electronic contracts, etc.) can be incorporated into a new cross-border e-commerce law, so that there is a law to follow when problems arise in the field of cross-border e-commerce. Due to the rapid development of cross-border e-commerce, there is always a certain lag in the formulation of laws and regulations, which requires China to continuously amend and improve its legislation on cross-border e-commerce. For example, the legal effect of data messages is clarified in the Contract Law, computer crime-related crimes are added to the Criminal Law, and in the regulation of product quality in the field of cross-border e-commerce exports, the Product Quality Management Law can be amended to provide a legal basis for cross-border product quality disputes by adding and deleting relevant articles. Speeding up the formulation of a single law in the field of cross-border e-commerce and amending the existing regulations is the future development trend of China's cross-border e-commerce legislation.
4.2. Participation in international rule-making
Second, China should actively participate in the formulation of international rules for cross-border e-commerce, and should conform to the international legal framework when dealing with legal disputes. Since cross-border e-commerce involves cross-border trade with various countries around the world, the nature of trade is foreign-related, and international trade rules are bilateral or multilateral, so China needs to take international trade rules into consideration when formulating laws and regulations on cross-border e-commerce, try to conform to the adaptability of trade rules and international treaties, and conform to the laws and regulations under the relevant framework, and constantly strengthen regulatory exchanges and cooperation with various countries and regions. In addition, we should continue to strengthen regulatory exchanges and cooperation with countries and regions, actively participate in multilateral negotiations and international rules and treaties related to cross-border e-commerce, and discuss with countries to promote the development of global cross-border e-commerce international dispute resolution mechanism and the establishment of international regulatory models. At the same time, we should strengthen the interpretation of international law such as the WTO’s Agreement on Trade in Services, improve China's legal system of cross-border trade in services, optimize the international interpretation of existing laws; and draw on the general rules of the United Nations Commission on International Trade Law on e-commerce legislation to develop a legal system suitable for China's cross-border e-commerce.
4.3. Legislation defines rights and obligations
Third, the rights and obligations of each subject should be clarified. Cross-border e-commerce retail export disputes involving all parties involved, so the first thing is to clarify the rights and obligations of the main body of the market. To inspection and quarantine obligations of the subjects, for example, consumers shall have the obligation not to buy goods prohibited from entering or leaving the country; third-party platform shall have the implementation of supervision of e-commerce enterprises and regular reasonable review of goods obligations; enterprises shall have the product information to the relevant departments for the record, when the product problems active recall obligations; fourth-party service providers, especially for sellers to provide customs clearance services such as customs inspection logistics company Should have the obligation to check whether the goods comply with the relevant provisions before sealing the package.
4.4. In conjunction with general law legislation
Fourthly, the principle legal provisions and general legal provisions should be used in conjunction with each other. This requires more principle clauses to be set in the process of e-commerce legislation to be used in conjunction with general clauses to regulate business activities. This is because the principle clauses do not stipulate the specific specification content, but ensure the correctness of the development direction of cross-border e-commerce in the process of its development,
and can play a proper restraining and restricting role in its general development, so the combination of principle clauses and general specific clauses in the future legislation of cross-border e-commerce will be conducive to the legislative work in line with the speed of upgrading network technology and the trend.
4.5. Strengthening Intergovernmental Coordination
Fifth, in terms of market supervision, coordination among governments should be strengthened, an integrated platform for customs clearance should be established, data channels of customs, finance, taxation, commodity inspection and other departments should be opened, the boundaries of responsibilities of each regulatory department should be clarified, effective unified supervision should be carried out, and joint assistance in law enforcement should be provided. Build a cross-border e-commerce industrial park to realize centralized management and “batch shipment and regular declaration”. In terms of trade facilitation measures, a single window can be established to promote the efficiency of customs clearance and implement an “electronic list” system for enterprises, which can be centralized with aggregated electronic customs declarations, VAT invoices and other matters such as foreign exchange clearance and tax refunds. The Hangzhou model can be used as a reference to establish a universal one-stop comprehensive customs clearance service platform to improve the efficiency of law enforcement, promote the efficiency of customs clearance and logistics speed of enterprises in customs clearance and inspection, improve the competitiveness of products, and provide convenience for all parties involved through various measures.
5. CONCLUSION
On the whole, China's cross-border e-commerce started late and developed fast, and the relevant laws and regulations have a certain lag. At present, the law is mainly based on single-line laws, and no special foreign-related laws and regulations have been introduced, and the legal basis is scattered in the relevant laws and regulations, and the legal level of legislation and rules is low, and many disputes are still in the legal vacuum. In the pre-transaction government regulatory mechanism, market access and platform supervision, information security protection, payment, logistics, taxation and post-transaction intellectual property disputes, after-sales disputes and other major aspects of cross-border e-commerce retail exports there are still many legislative shortcomings. This requires the relevant regulatory authorities to take a holistic approach to solve the common problems of legal deficiencies in each link, but also to make up for the legislative deficiencies in each link from a specific perspective, so that the disputes in each area can be based on the law and promote the healthy and orderly development of China's cross-border e-commerce retail exports.
REFERENCES
[1] Xu R. Legal protection of information security in cross-border e-commerce in China [J]. China Business Journal, 2021(03): 21-24.
[2] Zhang G. Legal regulation dilemma and countermeasures of cross-border e-commerce retail import in China[J]. Foreign Economic and Trade Practice, 2020(12): 25-28.
[3] Yao Y. Research on the current situation and countermeasures of legal regulation of cross-border e-commerce [J]. Modern Business, 2020(26): 53-54.
[4] Wei R. Research on the construction of the legal protection mechanism of cross-border e-commerce intellectual property rights in the free trade zone[J]. Legal Expo, 2020(23): 91-92.
[5] Lin Y, Zhang Y. Legal risks of intellectual property rights in cross-border e-commerce development and response[J]. Law and Economy, 2020(06): 7-9.
[6] Tang M. Analysis of the current situation of legal regulation of cross-border e-commerce in China and discussion of countermeasures [J]. China Market, 2020(03): 159-160.
[7] Gao F, Li Q, Li Y. Research and analysis on the legal policy of cross-border e-commerce[J]. Electronic commerce, 2017(11): 36-38.
[8] Tang X. Research on the current situation of legal regulation of cross-border e-commerce in China and countermeasures [J]. Journal of Wuyi University (Social Science Edition), 2017, 19(03): 65-68+94.
[9] Zheng A. An analysis on the statutory taxation of cross-border e-commerce [J]. Development Research, 2016(04): 72-76.
[10] Pei S, Zhai H. Legal risk prevention of cross-border e-commerce under export orientation[J]. Journal of Jilin Radio and Television University, 2016(03): 39-40.. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3456,
1
],
[
3456,
8712,
2
],
[
8712,
13898,
3
],
[
13898,
19523,
4
],
[
19523,
24039,
5
]
]
} |
178e83b0d84d664979ddf866a43a6e36cc06ae65 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 3,
"total-fallback-pages": 0,
"total-input-tokens": 10967,
"total-output-tokens": 2577,
"fieldofstudy": [
"Medicine"
]
} | Calcified Pseudoneoplasm of the Neuraxis (CAPNON)—A Rare Cause for Temporal Lobe Epilepsy: Not all Warrant a Surgical Intervention
Prashanth Raghu, Balaji Jeevanandham, Rajoo Ramachandran, Jeffrey Ralph, Pranesh Paneerselvam
Department of Radiology, Sri Ramachandra Institute of Higher Education and Research, Porur, Chennai, Tamil Nadu, India
Abstract
Epilepsy is a common neurological condition with varied etiological causes, with temporal lobe epilepsy being the most common. Among the varied etiologies of temporal lobe epilepsy, mesial temporal sclerosis is an important one and it presents as intractable epilepsy. However, we describe here a case of intractable temporal lobe epilepsy with a rather rare etiology, calcifying pseudo neoplasm of neuraxis (CAPNON) syndrome. CAPNON is a rare benign lesion that can occur anywhere in the central nervous system. The thought process till date is to excise any intracranial space occupying lesion to relieve pressure and for a better prognosis, which is not questionable. However, we feel in case of CAPNON, wait and watch protocol can be used to a better effect with radiological and clinical follow-up. Above all surgical excision was primarily done due to imaging confusion over CAPNON and this article comes up with few key findings to clinch the radiological diagnosis of CAPNON.
Keywords: Calcifying pseudo neoplasm of neuraxis, cavernoma, epilepsy, pseudoneoplasm
INTRODUCTION
Calcifying pseudoneoplasm of the neuraxis (CAPNON) is a rare entity with only 90 cases being reported in literature till date. Among the reported 90 cases, 54 (60%) are intracranial and 36 (40%) are intraspinal lesions.[1] It was first identified by Miller and erroneously reported as fibro-osseous lesion in 1922. It has been synonymously termed as brain stones, fibro-osseous lesions, and calcifying pseudo tumors.[2] In 1978, it was reported as a distinct entity by Rhodes.[2-4] If intracranial, it can be either intra or extra-axial in location.[2,5] This entity mimics many calcifying intra-axial lesions like ganglioglioma, oligodendroglioma, cavernous malformation, and infection like tuberculosis. Hence, CAPNON should always be considered in the list of differential diagnosis for calcifying intra-axial lesions.[6] Few patients have presented with hallucinations and partial seizures. Here, we illustrate a case presenting with intractable temporal lobe epilepsy and eventually diagnosed as CAPNON on imaging which was conservatively managed with a good clinical outcome.
CASE HISTORY
A male patient came to emergency room with five episodes of seizures since morning involving the right upper and lower limbs with secondary generalization. He was a known case of seizure disorder for the past 5 years and on anti-epileptic medication which includes eptoin and valproate. Based on clinical symptoms, the patient was referred for CT examination.
Non-contrast CT scan revealed multiple thick amorphous calcifications involving the right gangliocapsular region involving the genu and posterior limb of internal capsule, lentiform nucleus, right hippocampus, and right peduncle of midbrain [Figure 1a and b]. The patient was further evaluated with MRI with intravenous Gadolinium...
contrast. The MRI scan revealed an ill-defined intra-axial predominantly T2 hyperintense/T1 hypointense lesion involving the right thalamo capsular region, right medial temporal lobe, and the right peduncle of midbrain [Figure 2a and b]. Multiple discrete areas of blooming noted within the lesion on GRE [Figure 3] with magnitude and phase sequences of SWI showed features suggestive of calcification [Figure 4a and b]. Post-contrast T1-weighted axial and sagittal images shows minimal ill-defined discrete enhancement [Figure 5a and b].
Based on the imaging findings, a diagnosis of CAPNON was made and the patient was advised to continue the anti-epileptic drugs with regular follow-up. On follow-up, EEG was done and found normal. The patient was symptomatically better now. Hence, surgical plan was withheld.
**Discussion**
CAPNON is a rare tumor of the central nervous system. A hypothesis has been suggested that CAPNON may develop as a healing response to a wide range of eliciting factors, which could explain the variations in histopathologic features. Even though the histopathologic features of CAPNON are not clearly understood, a reactive process over a hamartomatous process has been favored.\(^2,7,8\) Interestingly, this reactive process involved in the calcifying pseudoneoplasm is not only constrained to the neuraxis. It can also occur at different other sites, including pleura, breast, and mediastinum.\(^9\)
No sex or age predominance has been mentioned in most of the published articles, however the lesions can occur at any age between 6 and 83 years.\(^2,6,8\) The most common symptoms in intracranial lesions are headache and seizure with only few cases being discovered incidentally.\(^2,8\) Symptoms, when present, are variable and are related to local mass effect, rather than invasive growth. Seizure was the commonest symptom in supratentorial CAPNON (17/36 cases; 47.2%). The seizure types were generalized tonic–clonic seizure (GTCS) in 7 cases (41.1%), focal
impaired awareness seizure (FIAS) in 4 cases (23.5%), focal aware seizure (FAS) in 2 cases (11.8%), and not described in 4 cases.[1]
The lesions are typically sporadic, with few exceptions that have occurred in association with meningioangiomatosis in patients with type 2 neurofibromatosis[9,10] and low-grade glioma.[10]
On CT, generally CAPNON shows an area of solid calcification while on MR imaging the lesions appear hypointense on T1- and T2-weighted sequences.[8] Post-contrast the lesions shows minimal linear internal or rim enhancement.[8] Contrary in our case, the lesion showed areas of T2 hyperintensity, probably because of the presence of chondromyxoid matrix within the areas of calcifications. Perilesional vasogenic edema is not commonly reported and is unexpected in benign lesion.[8] Grabowski et al., in his study, reported vasogenic edema could be because of seizures rather than CAPNON.[8] Similarly, in our case, perilesional edema was present secondary to seizures rather than CAPNON as proposed by Grabowski et al.
The typical histopathological features of CAPNON are (1) presence of palisading spindle to epithelioid cells; (2) presence of chondromyxoid matrix in a nodular pattern; (3) amorphous calcification, psammoma bodies, and osseous metaplasia; (4) foreign-body reaction with giant cells surrounding the calcified region; (5) fibroblastic proliferation.[2,3,11] These pathological presentations are not evident in each case and some lesions may not show all of the above features.[2,11] In our case, the patient was conservatively managed with anti-epileptic medications and is on regular follow-up. Earlier treatment has prevented the patient suffering from future complications like deterioration of memory.
Acknowledgements
The author would like to thank his parents, friends, and teachers for their moral support.
Financial support and sponsorship
Nil.
Conflicts of interest
There are no conflicts of interest.
REFERENCES
1. Tanoue Y, Uda T, Nakajo K, Nishijima S, Sasaki T, Ohata K. Surgically treated intracranial supratentorial calcifying pseudoneoplasms of the neuraxis (CAPNON) with drug-resistant left temporal lobe epilepsy: A case report and review of the literature. Epilepsy Behav Case Rep 2019;11:107-14.
2. Zerehpoosh FB, Samadian M, Azhari VS, Barangi M, Ebrahimzadeh K, Heidary M. A case report of calcifying pseudoneoplasm of the neuraxis: An incidental finding after a car accident. Turk Patoloji Derg 2018;34:265-8.
3. Fletcher AM, Greenlee JJ, Chang KE, Smoker WR, Kirby PA, O’Brien EK. Endoscopic resection of calcifying pseudoneoplasm of the neuraxis (CAPNON) of the anterior skull base with sinonasal extension. J Clin Neurosci 2012;19:1048-9.
4. Rhodes RH, Davis RL. An unusual fibro-osseous component in intracranial lesions. Hum Pathol 1978;9:309-19.
5. Mohapatra I, Manish R, Mahadevan A, Prasad C, Sampath S, Shankar SK. Shankar,Calcifying pseudoneoplasm (fibro osseous lesion) of neuraxis (CAPNON)—A case report. Clin Neuropathol 2010;29:223-6.
6. Aiken AH, Akgun H, Tihan T, Barbaro N, Glastonbury C. Calcifying pseudoneoplasms of the neuraxis: CT, MR imaging, and histologic features. Am J Neuroradiol 2009;30:1256-60.
7. Nonaka Y, Aliabadi HR, Friedman AH, Odere FG, Fukushima T. Calcifying pseudoneoplasms of the skull base presenting with cranial neuropathies: Case report and literature review J Neurol Surg Rep 2012;73:41-7.
8. Merola J, Jain A, Ziad F, Hussain Z. Calcified Pseudoneoplasm of the Neuraxis (CAPNON): A lesson learnt from a rare entity. J Neurol Neurosci 2016. DOI: 10.21767/2171-6625.1000121.
9. Donev K, Scheithauer BW. Pseudoneoplasms of the nervous system. Arch Pathol Lab Med 2010;134:404-16.
10. Higa N, Yokoo H, Hirano H, Yonezawa H, Oyoshi T, Goto Y, et al. Calcifying pseudoneoplasm of the neuraxis in direct continuity with a low-grade glioma: A case report and review of the literature. Neuropathology 2017;37:446-51.
11. Aiken AH, Akgun H, Tihan T, Barbaro N, Glastonbury C. Calcifying pseudoneoplasms of the neuraxis: CT, MR imaging, and histologic features. Am J Neuroradiol 2009;30:1256-60. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3235,
1
],
[
3235,
5233,
2
],
[
5233,
9334,
3
]
]
} |
1bc0a6c5e709f1378d4474b84ccdfebe7241d4d8 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 5,
"total-fallback-pages": 0,
"total-input-tokens": 16787,
"total-output-tokens": 3370,
"fieldofstudy": [
"Chemistry"
]
} | Homogeneous Catalysis
Catalytic Isohypsic-Redox Sequences for the Rapid Generation of C\texttextsubscript{sp3}-Containing Heterocycles
Craig D. Smith, David Phillips, Alina Tirla, and David J. France\textsuperscript{[a]}
Abstract: Cross-coupling reactions catalyzed by transition metals are among the most influential in modern synthetic chemistry. The vast majority of transition-metal-catalyzed cross-couplings rely on a catalytic cycle involving alternating oxidation and reduction of the metal center and are generally limited to forging just one type of new bond per reaction (e.g., the biaryl linkage formed during a Suzuki cross-coupling). This work presents an Isohypsic-Redox Sequence (IRS) that uses one metal to effect two catalytic cycles, thereby generating multiple new types of bonds from a single catalyst source. We show that the IRS strategy is amenable to several widely used transformations including the Suzuki–Miyaura coupling, Buchwald–Hartwig amination, and Wacker oxidation. Furthermore, each of these reactions generates value-added heterocycles with significant sp\texttextsubscript{3}-C (3-dimensional) content. Our results provide a general framework for generating complex products by using a single metal to fulfill multiple roles. By uniting different combinations of reactions in the isohypsic and redox phases of the process, this type of catalytic multiple-bond-forming platform has the potential for wide applicability in the efficient synthesis of functional organic molecules.
Transition metal catalysis is among the most common strategies in organic synthesis for the formation of C–C and C–X bonds.\textsuperscript{[3]} The vast majority of organometallic reactions rely on a single catalytic cycle to generate one new bond.\textsuperscript{[2]} Although the power of transition metal catalysis to effect previously unknown reactions has proved to be tremendously enabling, this “one reaction-one bond” limitation fails to maximize the complexity of the products generated using these methods. Catalytic multiple-bond-forming strategies carry vast potential to impact the “economies of synthesis” through the rapid evolution of molecular complexity.\textsuperscript{[3]} We set out to develop a multiple-bond-forming reaction sequence that would use a single metal to effect multiple catalytic cycles by uniting an isohypsic reaction manifold with more common redox catalytic cycles.\textsuperscript{[4]}
Most transition-metal-catalyzed reactions form a single new bond via a mechanistic cycle that involves alternating oxidative and reductive steps with respect to the metal catalyst. A smaller, but still widely used, set of metal-catalyzed processes occurs without changes in the metal oxidation state, that is, in an isohypsic manifold. Common examples of such catalytic cycles include the conjugate addition of organoborates to \( \alpha,\beta \)-unsaturated carbonyl compounds,\textsuperscript{[6]} Au\textsuperscript{+} or Co\textsuperscript{2+}-catalyzed alkyne activations,\textsuperscript{[7]} metallocarbenoid reactions (e.g., Rh\textsuperscript{+}-catalyzed reactions of \( \alpha \)-diazocarbonyls),\textsuperscript{[8]} and the chain propagation phase of metal-catalyzed alken polymerization.\textsuperscript{[9,8]}
The preference for redox-active catalytic cycles is maintained in the wide field of Pd-catalyzed fine chemical synthesis. Most processes occur by some variant of the well-known iterative sequence of oxidative addition, transmetallation, and reductive elimination. Nevertheless, many isohypsic Pd-catalyzed processes are known, such as the addition of organoborates to activated \( \pi \)-bonds,\textsuperscript{[10]} cyclodimerization processes that terminate by protonation or \( \beta \)-halide elimination,\textsuperscript{[11]} allylic rearrangements of esters or imidates,\textsuperscript{[12]} and the halo-allylation of alkenes,\textsuperscript{[13]} among others.\textsuperscript{[14]}
The mechanistic distinction between redox-active and isohypsic catalysis carries an important consequence from a synthetic perspective, namely that functionality that is inert to the metal oxidation state present in the isohypsic process (such as the aryl halides typically involved in oxidative addition to Pd\textsuperscript{0}) should be tolerated during an isohypsic reaction at a different oxidation state (e.g., Pd\textsuperscript{2+}). Subsequent alteration of the metal oxidation state (for example by the addition of a new reagent) allows for a second catalytic bond formation to occur using the same metal (Figure 1).
This type of transition from isohypsic to redox manifolds is an example of assisted tandem catalysis where one precatalyst effects two distinct catalytic processes using sequential reagent combinations to control the change in mechanism.\textsuperscript{[15]}
Despite its potential for broad utility (based on the number of well elucidated catalytic cycles), this isohypsic-redox strategy has seldom been used in the field of Pd-catalysis, and never in the context of alkene difunctionalization.\textsuperscript{[16]}
We have previously developed a Pd-catalyzed alkene difunctionalization reaction that forms a heterocycle with concomitant creation of an sp\texttextsubscript{3}–sp\texttextsubscript{3} C–C bond (Figure 2b).\textsuperscript{[17]} This methodology was specifically designed to generate heterocycles with significant sp\texttextsubscript{3}–C content, as studies of clinical success...
rates indicate a correlation between the progress of drug candidates through clinical trials and enhanced three-dimensionality.\[^{[18]}\] An isotopic labeling study suggested the alkene heteroallylation process proceeds via an isohypsic mechanism involving a somewhat unusual β-halide elimination step.\[^{[17]}\] Here, we describe the development of an isohypsic-redox sequence (IRS) based on the unification of alkene heteroallylation with transformative Pd-catalyzed redox-active processes such as the Suzuki–Miyaura coupling, Buchwald–Hartwig amination, and both the Wacker and Feringa–Grubbs aldehyde-selective Wacker oxidation protocols (Figure 2c).\[^{[16d, 20]}\] This IRS approach enhances molecular complexity by generating three new bonds in a single process while also forming a heterocycle and a new sp²–sp³ C–C bond.
Our first task in achieving the planned IRS was to identify an appropriate substrate for the alkene heteroallylation reaction that contained a functional handle for use in a diverse array of subsequent redox reactions. As aryl halides are the most commonly used coupling partners in standard Pd-catalyzed processes, bromophenol 1 was selected as our initial test case (Figure 3). Gratifyingly, this alkynyl phenol underwent the desired heteroallylation reaction to generate benzofuran 2 in good yield under our previously optimized conditions without engaging the aryl bromide, as expected by the all PdII catalytic cycle.\[^{[17]}\]
Once the heteroallylation in the presence of an aryl bromide had been demonstrated, we set out to establish our first IRS using the Suzuki–Miyaura cross-coupling, the most common C–C bond-forming reaction used by medicinal chemists.\[^{[21]}\] In this process, we were relying on the well-precedented reduction of PdII to Pd0 by boronic acids to initiate the redox catalytic cycle.\[^{[22]}\] After optimization,\[^{[23]}\] including use of Buchwald dialkylbiaryl phosphine ligands,\[^{[24]}\] we were able to generate the desired biaryl coupling products in good yield through the two catalytic cycles (Figure 4). Substrate scoping studies demonstrated that both electron-withdrawing and electron-donating substituents were tolerated. By modifying the phosphine ligand to XPhos in the case of thiophene (3e),\[^{[25]}\] and PPhos in
\[\text{Figure 1. Overview of the isohypsic-redox sequence (IRS) as an approach to complex molecule synthesis.}\]
\[\text{Figure 2. (a) Widely used cross-coupling strategy. (b) Alkene heteroallylation reaction proceeding through isohypsic mechanism. (c) Postulated isohypsic-redox tandem catalysis.}\]
\[\text{Figure 3. Heteroallylation of alkenyl phenol 1. Isohypsic mechanism tolerates aryl bromide.}\]
\[\text{Figure 4. Tandem heteroallylation–Suzuki coupling. Isolated yields based on 1. *SPhos replaced by XPhos, ^S Phos replaced by PPhos.}\]
the case of pyridine (3f), we were able to effectively couple these heterocycles.
Having demonstrated the capacity to form C–C bonds in the redox phase of the IRS process, we next chose to study C–N bond formation using the Buchwald–Hartwig amination. In this instance, reduction of the Pd(I) was envisaged to occur via a β-hydride elimination from a Pd(IV)-amine complex. Use of a dialkylbiphenyl phosphine was again found to be advantageous in coupling with hexyl amine (Figure 5). In addition to primary amines, the coupling proceeded well with secondary amines to generate morpholine 4c, piperazine 4d, and aniline 4e.
In order to extend the scope of the IRS platform beyond functionalized benzofurans, as well as to make use of the double bond that is installed by the isohypsic heteroallylation, we set out to combine the synthesis of N-containing heterocycles with oxidation of the double bond as a redox step (Figure 6). The Pd(0) generated at the end of the Wacker process would be re-oxidized by an external oxidant to complete a redox cycle. After screening a range of conditions for the standard Wacker oxidation, such as varying the re-oxidant system, we found that benzoquinone was the most effective (5 to 6). A methyl ketone was successfully installed in compounds containing both the isoquinolone and pyrrolopyrazinone ring systems. We then turned our attention to the possible aldehyde-selective Wacker-type alkene oxidation developed by Feringa and Grubbs et al. Using silver nitrite and copper(II) chloride as co-catalysts resulted in formation of the expected aldehyde as the major product in a modest overall yield consistent with the yields reported for these two processes in isolation (5 to 7). Interestingly, the presence of a nitrile ligand (as used in earlier work by Feringa and Grubbs et al.) was found to be essential for the reaction to proceed.
In summary, we have developed a suite of tandem catalytic processes based around the concept of linking the isohypsic (redox neutral) alkene heteroallylation reaction with well-known redox catalytic cycles including the Suzuki–Miyaura, Buchwald–Hartwig, and Wacker transformations. In all cases, one metal is used to effect two different catalytic cycles, thereby providing a strategy for the rapid evolution of molecular complexity in the context of forming 3D heterocycles. Given the number of well-elucidated catalytic cycles, expansion of the IRS concept has vast potential both within the field of Pd-catalysis and beyond.
Experimental Section
Representative procedure: A 4 mL screw-top glass vial was charged with 4-bromo-2-(2'-methylallyl)phenol (1) (45.0 mg, 0.200 mmol), toluene (0.65 mL), allyl chloride (80.0 µL, 1.00 mmol), NaHCO3 (34.0 mg, 0.400 mmol) and Pd(hfacac)2 (5.00 mg, 0.0100 mmol) and the vial was sealed under ambient atmosphere. The resulting mixture was heated to 50 °C by immersion of the entire vial into a pre-heated aluminum block until the substrate had been consumed, as judged by TLC analysis. The reaction mixture was cooled to room temperature and the volatile components were evaporated in vacuo. To the vial was added toluene (0.4 mL), SPhos (8.00 mg, 0.0200 mmol), freshly ground K2PO4 (127 mg, 0.600 mmol) and Pd(hfacac)2 (5.00 mg, 0.0100 mmol) and the vial was sealed under ambient atmosphere. The resulting mixture was heated to 50 °C for 16 h. The reaction mixture was cooled to room temperature then purified by flash chromatography on silica gel (petroleum ether, then petroleum ether/EtOAc; 98:2) to give 3a (36 mg, 68%).
Acknowledgements
Funding from the University of Glasgow and EPSRC to C.D.S. (DTA award ref: EP/J000434/1), D.P. (DTA award ref: EP/K503058/1) and D.J.F. (award ref: EP/I027165/1) is gratefully acknowledged. We also thank GlaxoSmithKline for samples from the New Methodology Expansion Set. Dr. Alistair Boyer (University of Glasgow) and Prof. Nicholas C. O. Tomkinson (University of Strathclyde) provided helpful discussions.
Conflict of interest
The authors declare no conflict of interest.
Keywords: heteroaromatics · homogeneous catalysis · isohypsic reaction · palladium tandem catalysis | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
0,
1
],
[
0,
5503,
2
],
[
5503,
8357,
3
],
[
8357,
11907,
4
],
[
11907,
12501,
5
]
]
} |
bccc6359f922bd44ea669ed6ba9123809bd46fd1 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 8,
"total-fallback-pages": 0,
"total-input-tokens": 25132,
"total-output-tokens": 9008,
"fieldofstudy": [
"Medicine",
"Biology"
]
} | THE PREVALENCE OF THYROID DYSFUNCTION IN ELDERLY CARDIOLOGY PATIENTS WITH MILD EXCESSIVE IODINE INTAKE IN THE URBAN AREA OF SÃO PAULO
Glaucia C. Duarte,1 Eduardo K. Tomimori,1 Rosalinda Y. A. Camargo,1 Ileana G.S. Rubio,1 Mauricio Wajngarten,II Amanda G. Rodrigues,II Meyer Knobel,1 Geraldo Medeiros-Neto1
doi: 10.1590/S1807-59322009000200011
OBJECTIVES: To evaluate the prevalence of thyroid dysfunction in elderly cardiac patients in an outpatient setting.
SUBJECTS AND METHODS: A total of 399 consecutive patients (268 women, age range 60–92 years) who were followed at Heart Institute were evaluated for thyroid dysfunction with serum free T4, TSH, anti-Peroxidase antibodies, urinary iodine excretion measurements and thyroid ultrasound.
RESULTS: Hyperthyroidism (overt and subclinical) was present in 29 patients (6.5%), whereas hypothyroidism (overt and subclinical) was found in 32 individuals (8.1%). Cysts were detected in 11 patients (2.8%), single nodules were detected in 102 (25.6%), and multinodular goiters were detected in 34 (8.5%). Hashimoto’s thyroiditis was present in 16.8% patients, most of whom were women (83.6%). The serum TSH increased with age and was significantly higher (p= <0.01) in patients, compared to the normal control group. No significant differences in serum TSH and free T4 values were observed when patients with atrial fibrillation (AF) were compared with those without arrhythmia. The median urinary iodine levels were 210 µg/L (40–856 µg/L), and iodine levels were higher in men than in women (p<0.01). Excessive iodine intake (urinary iodine >300 µg/L) was observed in one-third of patients (30.8%).
CONCLUSIONS: Elderly patients have a higher prevalence of both hypo- and hyperthyroidism as well as thyroid nodules when compared with the general population. About one-third of the older patients had elevated urinary secretion of iodine and a higher prevalence of chronic Hashimoto’s thyroiditis. It is recommended that ultrasonographic studies, tests for thyroid function and autoimmunity should be evaluated in elderly patients.
KEYWORDS: Iodine intake; Thyroid dysfunction; Cardiologic patients; Elderly patients; Urinary iodine.
INTRODUCTION
Thyroid diseases are common clinical problems associated with aging.1-5 In the elderly, a reduction in the activity of the hypothalamus-hypophysis-thyroid axis is expected and is associated with an inadequate response of pituitary hormone release after provocative tests. The thyroid gland undergoes anatomical and physiological adaptations in association with age (changes in weight, iodine uptake and hormone synthesis), which provide evidence of a reduction in thyroid function.4 In humans, there is also an age-dependent increase in thyroid volume and nodularity. The degree of this increase depends upon several factors, including iodine intake, and its detection depends on the methods (clinical, echographic or pathological) employed for evaluation.5 It is also relevant that the curve of the relationship between the iodine intake level
1Thyroid Unit, Division of Endocrinology, Hospital das Clinicas, University of Sao Paulo Medical School, Sao Paulo, Brazil
2Geriatric Cardiology Unit, Heart Institute, Hospital das Clinicas, University of Sao Paulo Medical School, Sao Paulo, Brazil
Email: [email protected]
Tel.: 55 11 3064-6835
Received for publication on August 08, 2008
Accepted for publication on October 30, 2008
and the occurrence of thyroid diseases within a population is U-shaped, with an increase in risk with both low and high iodine intake levels. Thus, interpreting data on the prevalence of thyroid disease in a specific area must take into account the influence of the iodine intake in that region. A study conducted in Denmark showed that aging is accompanied by a high prevalence of goiter, mainly in areas of low iodine intake. Also, a recent survey showed that the prevalence of goiter in the elderly population reaches figures as high as 74% in patients aged 55–75 years and 54% in patients aged 76–84 years, with a prevalence of nodular goiter of 25% and 21%, respectively. Another study comprised of 634 endocrinology outpatients with goiter (age range 55–91, 544 women) indicated that nontoxic and toxic multinodular goiter were the predominant thyroid abnormalities. Furthermore, dietary iodine may increase the risk of chronic autoimmune thyroiditis and for hyperthyroidism in the elderly.
Gender also seems to be an important determinant. For example, hypothyroidism is far more common among elderly women than men, particularly in the presence of thyroid antibodies. However, evaluating thyroid function in the elderly is complicated by an increased prevalence of nonthyroidal illness and autoimmune subclinical hypothyroidism. We have previously demonstrated an elevated urinary iodine concentration (>300 µg/L) in 53% of 844 schoolchildren in São Paulo. This observation led us to infer that the population of this large Brazilian state could be exposed to a mildly excessive daily iodine intake. We assumed that the adult population, especially individuals aged 60 years and older, could also be ingesting high levels of iodine.
The purpose of this cross-sectional study was to evaluate the thyroid parameters of function, autoimmunity, ultrasonographic aspects and size in older subjects from an urban area in São Paulo with a mild iodine excess.
SUBJECTS AND METHODS
Patients
Subjects were recruited from a random sample of outpatients attending the cardiology follow-up clinic at the Geriatric Cardiology Unit, Heart Institute, Hospital das Clinicas, University of Sao Paulo Medical School, in São Paulo, Brazil.
Participants were aged 60 years and older and had no history of thyroid or liver disease, thyroid surgery, radioactive iodine therapy, had not been evaluated with radiologic tests using contrast media, were not on iodine-containing vitamin/mineral preparations, antithyroid drugs or thyroid hormone. A detailed clinical history, review of previous case record, and a clinical examination were performed for each participant.
Enrollment included 400 consecutive patients (268 women). One patient was excluded due to non-compliance with the study protocol. The final study population was comprised of 399 patients with a mean age of 73.3 ± 7.5 years (range 60–92 years, median 73 years). Among the participants, 46 had atrial fibrillation (AF). The population was subdivided into groups according to the following age ranges: 60-70 years (n=138), 71-80 years (n=175), and 81-92 years (n=75).
A control group of 320 normal subjects (age range 18–60 years, 213 women) from the metropolitan area of São Paulo, Brazil was selected. Individuals in this group had a normal thyroid upon physical examination and ultrasound, negative thyroid peroxidase auto-antibodies (TPOAb), and no history of either past or present thyroid disease or treatment with thyroid hormone.
The criteria for a normal thyroid ultrasound were a homogeneous echogenic pattern throughout the gland, absence of nodules and cysts, and an absence of diffuse or heterogeneous abnormalities. Normal levels of urinary iodine excretion were between 100 and 299 µg/L.
In this reference cohort, serum thyroid-stimulating hormone (TSH) levels ranged from 0.6 to 3.7 mU/L with a mean ± SD of 1.8 ± 0.8 mU/L, and serum free T4 (FT4) ranged from 0.87 to 1.6 ng/dL (mean ± SD 1.16 ± 0.2 ng/dL). Thyroid volume ranged from 6 to 14.2 mL in women and 7 to 14.9 mL in men.
Design
Patients were asked about their present and past history of thyroid disease; presence of neck enlargement over the past year; use of medications; local symptoms, such as dyspnea, dysphagia, dysphonia, pain or discomfort; and general manifestations, such as weight or appetite loss, depression, apathy or irritability, palpitations, and muscle weakness. The majority of these patients had hypertension, mild left ventricular dysfunction, and chronic fatigue, but otherwise was in relatively good health. Standard clinical, analytical and morphological (ultrasonography) procedures were used for etiological diagnoses. Serum TSH, FT4 and TPOAb were obtained in all patients. Nonfasting spot urine and household salt samples were collected for iodine determination.
The study was approved by Hospital das Clínicas Research Ethics Committee. All participants agreed to participate in the study and signed an informed consent in accordance with institutional requirements.
METHODS
Fasting blood samples were drawn and stored at -20°C for a variety of assessments, including thyroid tests.
**Thyroid function tests:** TSH (reference range 0.5–4.0 µU/mL, functional detection limit 0.03 µU/mL), FT4 (reference range 0.9–1.7 ng/dL), and TPOAb (reference range <35 U/mL) were measured by solid-phase fluorimunoassays (AutoDELPHIA hTSH Ultra; AutoDELPHIA free thyroxine (FT4); and AutoDELPHIA TPOAb, PerkinElmer Life and Analytical Sciences, Wallac Oy, Finland).
**Echographic studies:** Thyroid ultrasonographic studies were conducted by the same investigators (GD and EKT) using a portable apparatus (7.5 mHz transducer; GE Medical Systems, USA). Thyroid volume was calculated according to the formula width X length X thickness X 0.524 for each lobe and added to the isthmus volume. Goiter was diagnosed if the thyroid volume was above 18 mL for men or 16 mL for women, as previously reported. Considering that the degree of hypoechoegenicity is associated with the appearance of thyroid dysfunction in patients with thyroid autoimmunity, we analyzed the echogenic pattern (hyperechoic, isoechoic, hypoechoic) of the thyroid in comparison to the neck muscles. The results were expressed as follows: grade 1 (normal and hyperechoic), grade 2 (slightly hyperechoic), grade 3 (mildly hypoechoic), and grade 4 (extensively hypoechoic).
Additional ultrasound structural focal abnormalities were described as cysts (circumscribed areas of greatly reduced or absent echogenicity), single nodules or multiple nodules.
**Urinary iodine excretion**
Urinary Iodine concentration was assayed using a modified Sandell-Kolthoff method in the nonfasting urine samples. Urine was collected between 8 and 12 PM in Monovette plastic syringes, which do not interfere with iodine determination, and frozen at -20°C within 30 min of collection. Concentration of iodine was expressed as µg of iodine per L of urine.
**Concentration of iodine in salt samples**
The current legal concentration of iodine in salt for human use is 20–60 mg of iodine/kg of salt (National Agency for Sanitary Surveillance, March 2003). The iodine content in salt samples were assayed at the end of the study and presented as mg of iodine per kg of salt (mg I/kg).
**Diagnostic criteria**
Thyroid dysfunction was defined according to the following laboratory reference values: overt hyperthyroidism (TSH <0.1 µU/mL and FT4 >1.7 ng/dL), subclinical hyperthyroidism (TSH>0.4 µU/mL and FT4 >1.7 ng/dL), overt hypothyroidism (TSH >4.0 µU/mL and FT4 <0.7 ng/dL) and subclinical hypothyroidism (TSH >4.0 µU/mL and FT4 >0.7 ng/dL). Serum TSH levels between 0.1–0.5 µU/mL were considered suppressed.
The diagnosis of autoimmune thyroid disease was defined by the following parameters: (1) a marked absence of echoes within the limits of the thyroid on ultrasound (grades 3 and 4, or marked hypoechoegenicity) or the presence of an atrophic gland (volume smaller than 4 mL) and (2) the presence of TPOAb (>35 U/mL) associated with echographic pattern of marked hypoechoegenicity or thyroid atrophy.
**Statistical analysis**
Quantitative variables are presented as the means ± standard deviation, and the qualitative variables are presented as proportions. The chi-square test was used to compare two group means for normally distributed data, except for TSH, FT4, thyroid volume, urinary iodine and TPOAb, in which non-parametric statistics were used with the Mann-Whitney test. The chi-square test was used to compare proportions between independent groups. The level of significance was set to 5%. Statistical analysis was conducted using SAS 8.0 software (Statistical Analysis System, Cary, NC, USA) and Minitab version 14 (Minitab, Inc., State College, PA, USA).
**RESULTS**
**Prevalence of thyroid dysfunction**
Among the 399 subjects, three (0.8%) had overt hyperthyroidism (two women, age range 66–79 years), and 26 (6.5%) had subclinical hyperthyroidism (21 women, age range 60-85 years).
Overt hypothyroidism was present in 17 individuals (4.3%, 11 women, age range 61–91 years), whereas subclinical hypothyroidism was found in 15 patients (8.1%, 10 women, age range 70–86 years).
**Serum TSH**
The mean TSH value was $2.1 ± 3.1$ µU/mL (median 1.5 µU/mL, range 0.03–46.1 µU/mL) (Table 1). TSH values were similar in men and women (mean $2.14 ± 3.68$ µU/mL versus $1.83 ± 1.39$ µU/mL, respectively, $p=0.647$). The
The prevalence of thyroid dysfunction in elderly cardiology patients
Duarte GC et al.
The average mean serum TSH value was significantly higher in the studied population when compared with the normal reference group (1.8 ± 0.8 µU/mL, p=0.02) although the biological importance of such a small difference is debatable. When subdivided according to the age ranges, the mean TSH values were as follows: 60–70 years, 1.66 ± 1.18 µU/mL; 71–80 years, 2.04 ± 2.78 µU/mL; and 81–92 years, 2.61 ± 5.3 µU/mL. There was a positive and significant correlation between the serum TSH levels and increasing age (p= 0.01) (Figure 1). Significant difference in the TSH values was observed among age groups (p=0.02) (Figure 1) but not between genders (p=0.36).
positive TPOAb values were found in 36 patients (9%), with similar frequencies in women (9.7%, n=26) and men (7.6%, n=10, p=0.498) (Table 1).
Ultrasonographic abnormalities
Hypoechogenicity (grades 3 and 4) was present in 67 patients (16.8%) and was considered indicative of chronic autoimmune thyroiditis. Fifty-six of the 268 women (20.9%) presented with marked hypoechogenicity, compared with 11 out of the 131 men (8.4%, p<0.001). Thyroid cysts were detected in 11 subjects (2.8%, nine women), single nodules were found in 102 subjects (79 women, 25.6%), and multiple nodules were found in 34 individuals (8.5%, 24 women) (Figure 2).
The thyroid volumes of the studied subjects are shown in Table 1. Thyroid enlargement was found in 45 patients
Table 1 - Demographic aspects, thyroid laboratory tests, thyroid volume, and urinary iodine concentration in patients with and without atrial fibrillation
| | Total Cohort (n=399) | Patients with atrial fibrillation (n = 46) | Patients without atrial fibrillation (n = 353) | Significance p |
|-------------------------|----------------------|--------------------------------------------|-----------------------------------------------|----------------|
| Sex | Men | 268 | 15 | 116 | 0.973 |
| | Women | 131 | 31 | 237 | |
| Age, years | Mean ± SD | 73.3 ± 7.5 | 76.6 ± 6.3 | 72.8 ± 7.5 | 0.001 |
| TSH (µU/mL) | Mean ± SD | 2.1±3.1 | 1.94±1.25 | 2.08±3.2 | 0.345 |
| FT4 (ng/dL) | Mean ± SD | 1.1 ± 02 | 1.11 ± 0.23 | 1.11 ± 0.20 | 0.982 |
| TPOAb (U/mL) | Ab + | (9.9%) 36/363 | (9.5%) 4/42 | (9.9%) 32/321 | 0.934 |
| Thyroid volume (mL) | Mean ± SD | 12.86 ± 17.86 | 12.94 ± 8.74 | 12.85 ± 18.73 | 0.459 |
| Urinary iodine (µg/L) | Median | 210.0 (40-856) | 216.1 (94-832) | 208.1 (40-858) | 0.497 |
Values in parentheses represent ranges.
Figure 1 - TSH levels within age groups. Note the increasing TSH values with age in men and women (p<c0.02)
FT4
The mean serum FT4 level was 1.11 ± 0.21 ng/dL (median 1.1 ng/dL, range 0.1– 2.4 ng/dL) (Table 1). There
The prevalence of thyroid dysfunction in elderly cardiology patients
Duarte GC et al.
(11.3%), with a significant difference between the mean thyroid volume for women (13.0 ± 21.4 mL, range 3.3–320 mL) compared with men (12.5 ± 6.0 mL, range 2.7–54.2 mL, p=0.001). There was no difference in thyroid volume between the normal reference group and the studied cohort (p=0.813). A significantly increased thyroid volume was seen in the participants with single nodules (11.4 mL, range 3.9–320 mL) and multiple nodules (12.9 mL, range 4.1–122 mL) compared with patients without nodules (10.4 mL, range 2.7–35 mL, p<0.01) (Figure 2).
Urinary and household salt iodine concentrations
The urinary iodine concentration of the studied population is shown in Figure 3. In 13 of the patients (3.3%), urinary iodine levels were below 100 µg/L, whereas in 116 patients (30.8%), the values were above 300 µg/L. Overall, the studied population had a median urinary iodine excretion of 210 µg/L (range 40–856 µg/L). Male subjects presented with a mean urinary iodine concentration that was significantly higher than women (279.0 ± 142.0 µg/L versus 243.0 ± 130.0 µg/L, respectively, p=0.006) (Figure 3). The studied population had a significantly lower median urinary iodine concentration compared with a population of 1,017 adults from São Paulo.\textsuperscript{16} The median urinary iodine concentration, according to age range quartiles, were as follows: 60–70 years, 243 µg/L (range 40–856 µg/L); 71–80 years, 203 µg/L, range 84–832 µg/L; and 81–92 years, 203 µg/L, range 100–594 µg/L. No significant difference was observed among the age groups (p=0.19).
The mean iodine content in the 97 household salt samples that were evaluated was 35.6 ± 8.9 mg I/kg (range 23.8–81.2 mg I/kg). These values are in accordance with the current legal recommendation.
Relationship between atrial fibrillation (AF), thyroid abnormalities and urinary iodine
Elderly patients with atrial fibrillation (n=46) were significantly older than those without arrhythmia (mean age 76.6 ± 6.3 years versus 72.8 ± 7.5 years, p=0.001). There was no relationship between thyroid function, thyroid size, gender and urinary iodine concentration between patients with or without atrial fibrillation (Table 1). Three out of 46 patients (6.5%) with AF presented with subclinical hyperthyroidism, and five of these patients (10.9%) had subclinical hypothyroidism.
DISCUSSION
Thyroid dysfunction in elderly individuals often occurs unnoticed, and methods for accurate detection may be controversial.\textsuperscript{17} Therefore, screening for thyroid nodular disease, hyperthyroidism and hyperthyroidism is recommended in older patients\textsuperscript{18}. Screening for overt hyperthyroidism usually reveals patients with subclinical forms of the disease, e.g., subclinical hypothyroidism (elevated serum TSH and normal FT4 levels) and subclinical hyperthyroidism (suppressed serum TSH and normal FT4 levels).
Subclinical thyroid dysfunction has been associated with a range of serious clinical outcomes and increased risk of progression to overt thyroid dysfunction. Environmental factors, such as sustained and elevated levels of nutritional iodine, have been associated with hyperthyroidism in the elderly, a population with increased prevalence of nodular thyroid disease.\textsuperscript{19} Nodular disease, both with the presence of single (cystic or solid) or multiple nodules, may be present in up to 50% of individuals above the age of 55 years.\textsuperscript{5} Iodine excess is also frequently associated with an increased prevalence of chronic autoimmune thyroid disease, which affects mostly women.\textsuperscript{20}
Another feature of increasing age is a decrease in the hypothalamic-pituitary response to low serum FT4 and total T4 levels. Serum TSH levels are lower among the
elderly compared to younger patients with the same degree of thyroid failure. A slight increase in serum TSH may indicate hypothyroidism in older adults, compared to younger individuals. This event is also observed among elderly patients with thyroid cancer who need a longer time period to reach sufficiently high TSH values to allow for an effective radioiodine ablation following thyroid hormone withdrawal.
Atrial fibrillation is an important dysrhythmia that represents an independent risk factor for other cardiovascular events and stroke. Low serum TSH is considered a risk factor for atrial fibrillation in older patients. More recently, Gammage et al. found a risk factor between the serum FT4 concentration and atrial fibrillation. Even in euthyroid subjects with normal serum TSH levels, the serum FT4 concentration was independently associated with atrial fibrillation in their study.
In the present study, we studied an elderly population from cardiac outpatient clinics. The majority of these patients had mild hypertension, chronic fatigue, and possibly an early stage of mild cardiac failure; otherwise, these patients were in relatively good health. Among these patients, 11.5% presented with atrial fibrillation. We compared the thyroid function tests and autoantibodies, thyroid volume and urinary excretion of iodine in both groups of patients with and without atrial fibrillation, and we could not find a significant difference between genders, with respect to the serum TSH, serum FT4, TPOAb, thyroid volume or urinary iodine concentration. Thus, we could not associate the presence of atrial fibrillation with thyroid dysfunction in this population.
Serum TSH values increased according to age in both genders. This increase attained statistical significance among age groups (Figure 2) but not between genders. Moreover, there was a positive and significant correlation of serum TSH and increasing age. Overall, the serum FT4 concentrations were similar in both genders, but the levels were significantly lower in the elderly when compared with the normal reference group. This observation confirms that the serum FT4 values have a tendency to be lower in the elderly as compared with a normal, younger adult population.
Severe hypochoegenicity, as seen in the ultrasonographic studies of this population, is an excellent method to determine the presence of chronic autoimmune thyroid disease. Indeed, Raber et al. have found that grades 3 and 4 hypochoegenicity have a higher predictive value for the diagnosis of Hashimoto’s thyroiditis when compared with the presence of thyroid autoantibodies (TPOAb). In our study, 17% of the patients filled the criteria for a marked hypoechoic pattern, but only half of these patients (9%) had a positive TPOAb. We may assume that the presence of both hypochoegenicity and TPOAb positivity is compatible with a diagnosis of Hashimoto’s thyroiditis in the elderly. However, the presence of an isolated marked hypochoegenicity should be considered as indicative of chronic autoimmune thyroid disease, even in the absence of TPOAbs. One possible explanation for this indication would be a presumably low response of the immunologic system to auto-antigens (or antigens in general) that may be present in elderly patients.
In a study conducted in São Paulo in 1994, a relatively higher prevalence of chronic autoimmune thyroid disease was observed in the elderly population when compared with a normal control population. As the population in this area experienced a period of five years (1998–2003) of excessive iodine intake due to the heavy addition of potassium iodate to table salt (40-100 ppm), it is possible that this environmental factor could have triggered a higher prevalence of Hashimoto’s thyroiditis. As observed in other countries, a sustained and prolonged excessive iodine intake may be linked to the increased prevalence of chronic autoimmune thyroid disease.
Thyroid volume significantly increased with advanced age, probably a result of the presence of single or multinodular thyroid disease. This increase is clearly shown in Figure 2, which shows that patients with multinodular goiter had increased thyroid volumes as compared with the echographically normal glands.
Hyperthyroidism (overt and subclinical) was present in 7.3% of the patients, an increased prevalence when compared with controls (3.3%). Excessive production of thyroid hormones may be due to Graves’ disease (TPOAb and TRAb positivity) or autonomous nodules in patients with single or multinodular thyroid disease. We did not perform TRAb studies in this cohort, and we assume that the relatively high prevalence of hyperthyroidism may be caused by a combination of high nutritional iodine and the presence of thyroid nodules.
Urinary iodine excretion was above 251 µg/L of urine in about one-third of our patients. Men had a higher iodine excretion than women did. The median urinary iodine was similar among the various age groups. Thus, during the period of the completion of this study (2007), elderly patients had a relatively normal nutritional intake, although it is possible that a higher iodine intake occurred during the five-year period of excessive iodine intake (1998-2003). There was no relationship between the urinary iodine concentrations and the presence of atrial fibrillation.
Overt hypothyroidism was present in 4.3% of the population, and subclinical hypothyroidism was confirmed in an additional 8.1%. Both prevalences were relatively higher compared with other similar population studies. Most of these patients had evidence of marked hypochoegenicity on
thyroid ultrasound, positive TPOAb or both. Thus, the results of our study corroborate previous findings of a high prevalence of autoimmune thyroid disease in older individuals.
In conclusion, we have confirmed that thyroid dysfunction is highly prevalent in the elderly who present with both hypo- and hyperthyroidism. Autoimmunity (Hashimoto’s thyroiditis) is probably the most common cause for decreased thyroid function and may be more prevalent when environmental iodine is excessive. Atrial fibrillation had no relationship with any of the studied thyroid parameters. Urinary iodine excretion was relatively high in this population without relation to increasing age. The serum TSH value (but not FT4) increased with age. Thyroid nodular disease (either single or multiple nodules) was highly prevalent (36.8%), confirming that nodular thyroid disease is common in the elderly. Taking into account all these facts, we suggest that elderly patients should be screened for thyroid dysfunction, particularly in an environment with relatively high iodine intake.
ACKNOWLEDGEMENTS
We gratefully acknowledged Adolfo Lutz Institute, a Public Health Laboratory in São Paulo (Brazil) for urinary iodine studies, and Prof. Celia Colli, Dept of Food and Experimental Nutrition, University of São Paulo, Pharmaceutical Sciences School, São Paulo, Brazil, who analyzed salt iodine content.
REFERENCES
1. Tunbridge WM, Evered DC, Hall R, Appleton D, Brewis M, Clark F, et al. The spectrum of thyroid disease in a community: the Wickham survey. Clin Endocrinol. 1977;7:481-93.
2. Canaris GJ, Manowitz NR, Mayor G, Ridgway EC. The Colorado thyroid disease prevalence study. Arch Intern Med. 2000;160:526-34.
3. Hollowell JG, Stachling NW, Flanders WD, Hannon WH, Gunter EW, Spencer CA, et al. Serum TSH, T4, and thyroid antibodies in the United States population (1988 to 1994). J Clin Endocrinol Metab. 2002;87:489-99.
4. Weisel M. Disturbances of thyroid function in the elderly. Wien Klin Wochenschr. 2006;118:16-20.
5. Mariotti S, Franceschi C, Cossarizza A, Pinchera A. The aging thyroid. Endocr Rev. 1995;16:686-715.
6. Laurberg P, Bülow Pedersen I, Knudsen N, Ovesen L, Andersen S. Environmental iodine intake affects the type of nonmalignant thyroid disease. Thyroid. 2001;11:457-69.
7. Laurberg P, Pedersen KM, Vestergaard H, Sigurdsson G. High incidence of multinodular toxic goiter in the elderly population in a low iodine intake area vs. high incidence of Graves’ disease in the young in a high iodine intake area: comparative surveys of thyrotoxicosis epidemiology in East-Jutland Denmark and Iceland. J Intern Med. 1991;229:415-20.
8. Cavaliere R, Antonangeli L, Vitti P, Pinchera A, Aghini-Lombardi F. The aging thyroid in a mild to moderate iodine deficient area of Italy. J Endocrinol Invest. 2002;25:66-8.
9. Diez JJ. Goiter in adult patients aged 55 years and older: Etiology and clinical features in 634 patients. J Gerontol. 2005;60:920-3.
10. Bagchi N, Brown TR, Parish RF. Thyroid dysfunction in adults over age 55 years. A study in an urban US community. Arch Intern Med. 1990;150:785-7.
11. Chiovato L, Mariotti S, Pinchera A. Thyroid diseases in the elderly. Baillieres Clin Endocrinol Metab 1997;11:251-70.
12. Duarte GC, Tomimori EK, Bariolli RA, Ferreira JE, Catarino RM, Camargo RYA, et al. Avaliação ultra-sonográfica da tireoide e determinação da iodúria em escolares de diferentes regiões do estado de São Paulo. Arq Bras Endocr Metab. 2004;49:842-6.
13. Tomimori EK, Pedrinola F, Cavaliere H, Knobel M, Medeiros-Neto G. Prevalence of incidental thyroid disease in a relatively low iodine intake area. Thyroid. 1995;5:273-6.
14. Mazzotti G, Sorvillo F, Iorio S, Carbone A, Romeo A, Piscopo M, et al. Grey-scale analysis allows a quantitative evaluation of thyroid echogenicity in the patients with Hashimoto’s thyroiditis. Clin Endocrinol. 2003;59:223-9.
15. Als C, Helbling A, Peter K, Haldimann M, Zimmerli B, Gerber H. Urinary iodine concentration follows a circadian rhythm: a study with 3023 spot urine samples in adults and children. J Clin Endocrinol Metab. 2000;85:1367-9.
16. Camargo YA, Tomimori EK, Neves SC, Rubio IGS, Galrão AL, Knobel M, et al. Thyroid and the environment: exposure to excessive nutritional iodine increases the prevalence of thyroid disorders in São Paulo, Brazil. Eur J Endocrinol. 2008;159:293-9.
17. Camargo RY, Pellanda L, Porta V, Maciel, P, Furquim A, Schaab B. Preoperative assessment of thyroid nodules: role of ultrasonography and fine needle aspiration biopsy followed by cytology. Clinics. 2007;62:411-8.
18. Gauskloo J, van Exel E, de Craen AJM, Meinders AE, Frolich M & Westendorp RGJ. Thyroid status, disability and cognitive function, and survival in old age. JAMA. 2004;292:2591-9.
19. Stanbury JB, Ernman AE, Bourdoux P, Todd C, Oken E, Tonglet R, et al. Iodine-induced hyperthyroidism: occurrence and epidemiology. Thyroid. 1998;8:83-100.
20. Weetman AP. Autoimmune thyroid disease: propagation and progression. Eur J Endocrinol. 2003;148:1-9.
21. Carlé A, Laurberg P, Pedersen IB, Perrild H, Ovesen L, Rasmussen LB, et al. Age modifies the pituitary TSH response to thyroid failure accompanied by a high prevalence of goiter, mainly in areas of low iodine intake. Thyroid. 2007;7:139-44.
22. Gammage MD, Parle JV, Holder RL, Roberts LM, Hobbs FDR, Wilson S, et al. Association between serum free thyroxine concentration and atrial fibrillation. Arch Intern Med. 2007;167:928-34.
23. Sawin CT, Geller A, Wolf PA, Belanger AJ, Baker E, Bacharach P, et al. Low serum thyrotropin concentrations as a risk factor for atrial fibrillation in older persons. N Engl J Med. 1994;331:1249-52.
24. Raber W GA, Nowotny P, Vierhepper H. Thyroid ultrasound versus antithyroid peroxidase antibody determination: a cohort study of four hundred fifty-one subjects. Thyroid. 2002;12:725-31.
25. Weetman AP, McGregor MA. Autoimmune thyroid disease: further developments in our understanding. Endocr Reviews.1994;15:788-829.
26. Zois C, Stavrou I, Kalogerou C, Svarna E, Dimoliatis I, Seferiadis K, et al. High prevalence of autoimmune thyroiditis in schoolchildren after elimination of iodine deficiency in northwestern Greece. Thyroid. 2003;13:485-9.
27. Teng W SZ, Teng X, Guan H, Li Y, Teng D, Jin Y, et al. Effect of iodine intake on thyroid diseases in China. N Engl J Med. 2006;354:2783-93. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3434,
1
],
[
3434,
8470,
2
],
[
8470,
12883,
3
],
[
12883,
16534,
4
],
[
16534,
20398,
5
],
[
20398,
26037,
6
],
[
26037,
30973,
7
],
[
30973,
32418,
8
]
]
} |
a1c81df38b21015b88ecd69943fde0e2e486ff92 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 24,
"total-fallback-pages": 0,
"total-input-tokens": 91052,
"total-output-tokens": 21124,
"fieldofstudy": [
"Computer Science"
]
} | Cell-average based neural network method for hyperbolic and parabolic partial differential equations
Changxin Qiu\textsuperscript{a}, Jue Yan\textsuperscript{a,1,*}
\textsuperscript{a}Department of Mathematics, Iowa State University, Ames, IA 50011, USA
Abstract
Motivated by finite volume scheme, a cell-average based neural network method is proposed. The method is based on the integral or weak formulation of partial differential equations. A simple feed forward network is forced to learn the solution average evolution between two neighboring time steps. Offline supervised training is carried out to obtain the optimal network parameter set, which uniquely identifies one finite volume like neural network method. Once well trained, the network method is implemented as a finite volume scheme, thus is mesh dependent. Different to traditional numerical methods, our method can be relieved from the explicit scheme CFL restriction and can adapt to any time step size for solution evolution. For Heat equation, first order of convergence is observed and the errors are related to the spatial mesh size but are observed independent of the mesh size in time. The cell-average based neural network method can sharply evolve contact discontinuity with almost zero numerical diffusion introduced. Shock and rarefaction waves are well captured for nonlinear hyperbolic conservation laws.
Keywords: Machine learning neural network; Finite volume methods; Nonlinear hyperbolic conservation laws.
1. Introduction
In this paper, we develop cell-average based neural network (CANN) method solving time dependent hyperbolic and parabolic partial differential equations (PDEs)
\[ u_t + f(u)_x = \mu u_{xx}. \] (1.1)
Our general idea is to follow available numerical schemes to build up neural network methods. We consider to combine the powerful machine learning mechanism of neural networks and the principles of classical numerical methods to develop suitable neural network methods for partial differential equations (1.1).
Machine learning with neural networks have achieved tremendous success in image classification, text, videos and speech recognition \([1, 2, 3, 4]\) for the last three decades. In the last few years, connections between differential equations and machine learning have been established, i.e. \([5, 6, 7, 8, 9, 10, 11]\). Very recently, machine learning neural networks have also been explored to directly solve partial differential equations, for which a better solver may be obtained or it can assist to improve the performance of current numerical methods.
*corresponding author
Email addresses: [email protected] (Changxin Qiu), [email protected] (Jue Yan)
\textsuperscript{1}Research work of the author is supported by National Science Foundation grant DMS-1620335 and Simons Foundation grant 637716.
Nonlinear convection diffusion equation (1.1) may not be complex and hard to solve in general. It can be used as a prototype or model equation for the more complicated Euler and Navier-Stokes equations for fluid dynamics. There exist quite a few numerical methods successfully developed for (1.1). Several numerical challenges are still present. For example, a lot of ongoing efforts are toward investigating implicit or semi-implicit methods to obtain efficient solvers. It turns out, once well trained, our cell-average or finite volume based neural network method can be relieved from the explicit scheme CFL restriction and can adapt large time step size (i.e. $\Delta t = 4\Delta x$) for solution evolution, even being implemented as an explicit method.
1.1. Related works
Depending on the ways of neural networks applied and the goals, neural network methods can be roughly classified into two groups. One group is to design and apply neural network methods to directly solve partial differential equations. The second group is to have neural networks applied to assist and improve available numerical methods.
One popular class is to find best solution representation in terms of neural networks. With network input vector as $x$ and $t$, such methods have the advantage of automatic differentiation, mesh free and can be applied to solve many types of PDEs. Approximation power of neural networks [12, 13, 14, 15, 16] are explored with these methods. For solving PDEs, we have the early works of [17, 18], the works of [19, 20] and the popular physics-informed (PINN) methods of [21, 22, 23, 24]. To improve the PINN efficiency on larger domain, an extreme and distributed network is considered in [25]. Application to incompressible Navier-Stokes equations is studied in [26]. We also have the works of [27] and [28, 29], for which weak formulations are applied in the loss function in stead of PDEs explicitly enforced on collocation points. Comparison to reduced basis method is studied in [30] and performance comparison to finite volume and discontinuous Galerkin methods are considered in [31]. We refer to [32, 33] for PINN convergence studies.
Neural network methods have been found rather successful for solving high dimensional PDEs. We refer to the early work of [34] and other discussions in [35, 36, 37, 38]. Other studies include designing specific networks for elliptic type PDEs or solving inverse problems, see [39, 40, 41]. In [42], convolutional networks are used for estimating the mean and variance for uncertainty quantification. We also have the works of [43] and [44], for which method of lines approach is explored with Fourier basis considered and Residual networks applied to evolve the dynamical system. Different to PINN or related methods in which a global solution with neural networks is sought, our method is similar to finite volume scheme thus is mesh dependent and is a local solver.
The other group is to combine networks with classical numerical methods for performance improvement. We have early result of [45] applying neural networks as trouble-cell indicator. Deep reinforcement network in [46] is explored to estimate the weights and enhance WENO schemes performance. In [47], network is applied for identifying suitable amount of artificial viscosity added. Neural network of [48] is applied to speed up iterative solver for elliptic type PDEs. Furthermore, we have WENO schemes augmented with convolutional networks for shock detection in [49] and DG methods with imaging edge detection technique with convolutional network explored for strong shock detection in [50]. We further have the work of [51] in which network is applied estimating total variation bounded constants for DG methods.
1.2. Motivation and our approach
The cell-average based neural network method is closely related to finite volume scheme. Let’s review first order upwind finite volume method for linear advection equation
$$u_t + u_x = 0,$$ \hspace{1cm} (1.2)
which motivates the design of our cell-average based method. Integrating the advection equation (1.2) over one computational cell $I_j = [x_{j-1/2}, x_{j+1/2}]$, with cell average notation $\bar{u}_j = \frac{1}{\Delta x} \int_{I_j} u(x, t) \, dx$ and numerical flux $\hat{u}$ introduced at $x_{j+1/2}$, we obtain $\frac{\partial \bar{u}_j}{\partial t} + \frac{\hat{u}_{j+1/2} - \hat{u}_{j-1/2}}{\Delta x} = 0$. This is the starting point of finite volume schemes. Adapting upwind for the numerical flux and forward Euler for time discretization, we have the following first order upwind finite volume method
$$
\frac{\bar{u}_j^{n+1} - \bar{u}_j^n}{\Delta t} + \frac{\bar{u}_j^n - \bar{u}_{j-1}^n}{\Delta x} = 0.
$$
(1.3)
For cell-average based neural network method, we consider to integrate the advection equation (1.2) over rectangle box of $I_j \times (t_n, t_{n+1})$, which covers both the computational cell in space and the sub interval in time. The integral or weak formulation of the advection equation is obtained as
$$
\bar{u}_j(t_{n+1}) = \bar{u}_j(t_n) - \frac{1}{\Delta x} \int_{t_n}^{t_{n+1}} \int_{I_j} u_x \, dx \, dt.
$$
(1.4)
Exact solution of (1.2) satisfies the above integral formulation. With extra regularity on the solution, the above integral format is equivalent to the advection equation (1.2) given in differentiation format. A simple feed forward neural network is adapted to approximate spatial variable related term $N(\overrightarrow{V}_j^m; \Theta) \approx \int_{t_n}^{t_{n+1}} \int_{I_j} u_x \, dx \, dt$. Here $\overrightarrow{V}_j^m$ is the network input vector and $\Theta$ is the network parameter set. And we obtain the following cell-average based neural network method for (1.2)
$$
\bar{v}_j^{n+1} = \bar{v}_j^n + N(\overrightarrow{V}_j^m; \Theta^\star), \quad \forall j \quad \text{and} \quad \forall n.
$$
(1.5)
Setting $\bar{v}_j^n = \bar{u}_j(t_n)$ and by minimizing the difference between network output $\bar{v}_j^{n+1}$ and the target $\bar{u}_j(t_{n+1})$, optimal parameter set $\Theta^\star$ can be obtained through offline supervised learning.
In a word, through intensive training we force the neural network to learn the solution average evolution between two neighboring time steps. The functionality of the parameter set $\Theta^\star$ is similar to the scheme definition of upwind finite volume method (1.3), which are the coefficients of cell-average based neural network scheme (1.5). For nonlinear convection diffusion equation (1.1), same framework is applied that is summarized below
$$
\bar{u}_j(t_{n+1}) = \bar{u}_j(t_n) - \frac{1}{\Delta x} \int_{t_n}^{t_{n+1}} \int_{I_j} \{f(u)_x - \mu u_{xx}\} \, dx \, dt \approx \bar{v}_j^{n+1} = \bar{v}_j^n + N(\overrightarrow{V}_j^m; \Theta^\star).
$$
Notice that we do not discretize or approximate differential terms involving spatial variable. In stead, having the neural network handle all spatial variable related differentiation and integration approximation is the major idea of our method. In Figure 1 we list two simulations with our CANN method. Shock and contact discontinuity are both sharply captured.
Below we summarize the results and features of cell-average based neural network method. Some are quite outstanding and are not common to classical numerical methods.
- Adapt any time step size, i.e. $\Delta t = 8 \Delta x$, even being an explicit scheme
- For Heat equation, errors are independent of time step size $\Delta t$.
- Introduce almost zero artificial numerical diffusion for contact discontinuity evolution
Due to some mysterious reason, the neural network method is able to catch up solution information around the next time level $t_{n+1}$ thus allows large time step size evolution. Different to classical
numerical methods for which we design a scheme first, here the network itself finds the best scheme for us.
The organization of the article is the following. In section §2.1, we introduce the definition of cell-average based neural network method and highlight its connection to finite volume schemes. In section §2.2, we emphasize the training process and summarize the learning data difference between linear and nonlinear PDEs. In Section §2.3, we list the major results of cell-average based neural network method. Sequence of numerical examples are presented in section §3. Final conclusion remarks are given in section §4.
2. Neural network solver
2.1. Problem setup, motivation and cell-averaged neural network method
We consider to develop finite volume or cell-average based neural network (CANN) method solving partial differential equations (PDEs)
\[ u_t = L(u), \quad (x,t) \in (a,b) \times \mathbb{R}^+ . \]
(2.1)
Here \( t \) and \( x \) denote the time and spatial variables and \( (a,b) \) is the spatial domain. Differentiation operator \( L \) is introduced to represent a generic first order hyperbolic or second order parabolic differentiation operator. For example, we have \( L(u) = (u^2)_x \) for the inviscid Burgers equation and \( L(u) = u_{xx} \) for the Heat equation.
Our neural network method is mesh dependent and motivated by finite volume method. Once well trained, the cell-average based neural network method can be applied solving PDEs (2.1) as a regular finite volume scheme. Having a uniform partition of \( (a,b) \) into \( J \) cells and \( \Delta x = \frac{b-a}{J} \) is adapted as the cell size. Denoting \( x_{1/2} = a, \ x_{J+1/2} = b \), we have \( [a,b] = \bigcup_{j=1}^{J} I_j \) with \( I_j = [x_{j-1/2}, x_{j+1/2}] \) as one computational cell. Furthermore, we have partition in time and adapt \( \Delta t \) for the time step size and we have \( t_n = n \times \Delta t \) with \( t_0 = 0 \).
Now integrate the partial differential equation (2.1) over the computational cell \( I_j \) and time interval \( [t_n, t_{n+1}] \), we have
\[ \int_{t_n}^{t_{n+1}} \int_{I_j} u_t \, dx \, dt = \int_{t_n}^{t_{n+1}} \int_{I_j} L(u) \, dx \, dt. \]
(2.2)
With the definition of cell average \( \bar{u}_j(t) = \frac{1}{\Delta x} \int_{I_j} u(x,t) \, dx \), equation (2.2) can be integrated out as
\[ \bar{u}_j(t_{n+1}) - \bar{u}_j(t_n) = \frac{1}{\Delta x} \int_{t_n}^{t_{n+1}} \int_{I_j} L(u) \, dx \, dt. \]
(2.3)
The equation above is nothing but the integral format or a weak formulation of (2.1). It is equivalent to the original partial differential equation (2.1) with extra differentiation condition applied. The integral format (2.3) is the starting point at which we design our neural network method. The idea of cell-average based neural network method is to apply a simple fully connected network $N(\cdot; \Theta)$ to approximate the right hand side of (2.3)
$$N(\cdot; \Theta) \approx \frac{1}{\Delta x} \int_{t_n}^{t_{n+1}} \int_{I_j} L(u) \, dx \, dt,$$
(2.4)
where $\Theta$ denotes the network parameter set of all weight matrices and biases.
Given the solution averages $\{\bar{u}^n_j\}$ at time level $t_n$, we apply following neural network to approximate the solution average $\bar{u}^{n+1}_j$ at next time level $t_{n+1}$
$$\bar{v}^{out}_j = \bar{v}^{in}_j + N(\bar{V}^{in}_j; \Theta).$$
(2.5)
With $\bar{v}^{in}_j = \bar{u}^n_j$ and comparing (2.5) and the integral format (2.3) of the PDEs, we have
$$\bar{v}^{out}_j \approx \bar{u}^{n+1}_j.$$
Vector $\bar{V}^{in}_j$ as the input vector of the network $N(\bar{V}^{in}_j; \Theta)$ is an important component that should be carefully chosen, see Figure 2.1. Its general format is given as
$$\bar{V}^{in}_j = [\bar{u}^{n}_{j-p}, \bar{u}^{n}_{j-1}, \bar{u}^{n}_j, \bar{u}^{n}_{j+1}, \ldots, \bar{u}^{n}_{j+q}]^T,$$
(2.6)
where we include the left $p$ cell averages and right $q$ cell averages of $\bar{u}^n_j$ in the input vector. The suitable stencil or the $p$ and $q$ values in (2.6) determine the effectiveness of the neural network method approximating the solution average $\bar{u}^{n+1}_j$ at the next time level.
Let’s use advection equation $u_t + u_x = 0$ as an example to demonstrate the relation between finite volume method and the cell-averaged neural network method. Forward Euler upwind finite volume scheme can be rewritten in terms of vector multiplication format as
$$\bar{u}^{n+1}_j - \bar{u}^n_j = \frac{\Delta t}{\Delta x} (\bar{u}^n_j - \bar{u}^n_{j-1}) = [\frac{\Delta t}{\Delta x} \ - \frac{\Delta t}{\Delta x}] \begin{bmatrix} \bar{u}^n_j \\ \bar{u}^n_{j-1} \end{bmatrix}.$$
Now consider the cell-average based neural network method for $u_t + u_x = 0$. We adapt upwind mechanism and choose input vector as $\bar{V}^{in}_j = [\bar{u}^n_j, \bar{u}^n_{j-1}]^T$, see illustration in Figure 2.1. The goal of the neural network is to obtain optimal parameter set $\Theta$ through minimizing the error between
network output $\bar{v}^{out}_j$ and the given target $\bar{u}^{n+1}_j$. The parameter set $\Theta$ is similar to the coefficients of a finite volume scheme, for example the $\frac{\Delta t}{\Delta x}$ before $\bar{u}^n_j$ and $-\frac{\Delta t}{\Delta x}$ before $\bar{u}^{n+1}_j$ of the upwind scheme. We highlight the network parameter set $\Theta$ interact with input vector $\bar{V}^{in}_j$ in a nonlinear fashion, even for a linear differential equation. Thus neural network method behaves differently to finite volume schemes. Once well trained, the optimal parameter set $\Theta^*$ uniquely determines the \textbf{scheme definition} of a neural network method. We then apply this neural network solver at any cell location (index $j$) and at any time level $t_n$ to evolve the solution average to next time level $t_{n+1}$ as an explicit scheme.
Motivated by the well established finite volume schemes for hyperbolic PDEs, we adapt upwind or method of characteristic mechanism to choose suitable network input vector $\bar{V}^{in}_j$ for hyperbolic PDEs. And we adapt a symmetric mechanism similar to central scheme for Heat equation to choose suitable network input vector for parabolic PDEs. The guideline of picking a suitable stencil ($p$ and $q$ in (2.6)) for the network input vector $\bar{V}^{in}_j$ is summarized below.
\textbf{Guideline on the choice of network input vector $\bar{V}^{in}_j$}:
- Hyperbolic PDEs: include the characteristic or domain of dependence
- Parabolic PDEs: a symmetric stencil mechanism
In this paper, we consider a standard fully connected neural network with $M$ ($M \geq 3$) layers. The input and output vectors of the network are the first and last layer. Among the total $M$ layers, the interior ($M-2$) are the hidden layers. Thus the minimum structure of the neural network involves one hidden layer with $M = 3$. We have $n_i$ ($i = 1, \cdots , M$) denote the number of neurons in each layer. The first layer is the input vector with $n_1 = p + q + 1$ as its dimension and the last layer is the output vector with $n_M = 1$ as its dimension. The abstract goal of machine learning is to find a function $\mathcal{N}: R^{p+q+1} \to R^1$ such that $\mathcal{N}(\cdot; \Theta)$ accurately approximates $\frac{1}{\Delta x} \int_{t_{n+1}}^{t_n} \int_{I_j} \mathcal{L}(u) \, dx \, dt$, the right hand side of (2.3). The optimal parameter set $\Theta$ of the network $\mathcal{N}(\cdot; \Theta)$ will be obtained by training the network intensively over the given data set.
Learning data are collected in the form of pairs. Each pair refers to the solution averages at two neighboring time level. The training data set is denoted as
$$S = \left\{ \left( \bar{u}^n_j, \bar{u}^{n+1}_j \right) , \; j = 1, \cdots , J \right\}_{n=0}^m , \tag{2.7}$$
which are solution averages obtained from highly accurate numerical method for the PDE. We highlight the training data pairs are solution averages collected over the spatial domain and from time levels $t_0$ to $t_{m+1}$.
The fully neural network $\mathcal{N}(\cdot; \Theta)$ consists of $M$ layers. Each two consecutive layers is connected with an affine linear transformation and a point-wise nonlinear activation function. The function or the mapping $\mathcal{N}(\cdot; \Theta)$ is a composition of following operators,
$$\mathcal{N}(\cdot; \Theta) = (\sigma_M \circ W_{M-1}) \circ \cdots \circ (\sigma_2 \circ W_1) , \tag{2.8}$$
where $\circ$ stands for operator composition. We have $W_i$ denoting the linear transformation operator or the weight matrix connecting the neurons from $i$-th layer to $(i+1)$-th layer. The parameter set is further augmented with the biases vectors. We have $\sigma_i: R \to R$ as the activation function ($i \geq 2$), which is applied to each neuron of the $i$-th layer in a component-wise fashion. In this paper we apply $\tanh(x)$ function as the activation function. Specifically $\sigma_i = \tanh(x)$ is applied between all layers, except to the output layer for which we have $\sigma_M(x) = x$.
6
Definition 1. A cell-average based neural network method is uniquely determined by the following four components: (1) the choice of spatial mesh size $\Delta x$; (2) the choice of time step size $\Delta t$; (3) the choice of network input vector $\overrightarrow{V}_{jn}$ of (2.6); and (4) the number of hidden layers and neurons per layer the corresponding structure of the neural network.
We further highlight that our cell-average based neural network method, once well trained, will be implemented as a regular explicit finite volume scheme. Notice the cell-average based neural network method is designed to approximate the weak or integral format of the partial differential equations (2.3), not the original partial differential equations (2.1) given in differentiation format.
2.2. Training Process
In this section, we discuss how to train the network to obtain optimal parameter set $\Theta^*$ such that the neural network (2.5) can accurately approximate the solution average evolution $\bar{u}_{jn} \rightarrow \bar{u}_{jn+1}$. This is achieved by applying $\bar{v}_{jn} = \bar{u}_{jn}$ in (2.5) to obtain network output $\bar{v}_{jn}^{out}$, comparing with the target $\bar{u}_{jn+1}$, and then looping among the data set $S$ of (2.7) to minimize the error or the squared loss function
\[
L_{j,n}(\Theta) = (\bar{v}_{jn}^{out} - \bar{u}_{jn+1})^2, \tag{2.9}
\]
for all $j = 1, \cdots, J$ and for all $n = 0, \cdots, m$. This choice of loss function defined over one single data pair is referred as the stochastic or approximate gradient descent method. Notice for $j = 1$ or $j = J$ or those close to boundary cells, the stencil or network input vector $\overrightarrow{V}_{jn}$ of (2.6) requires averages values of $p$ cells to the left and $q$ cells to the right of the current cell. In this paper we only consider Dirichlet or periodic boundary conditions. Thus we simply copy solution averages from inside the domain implementing periodic boundary condition. And we assign exact solution values to those out of domain ghost cells solving Burgers equation Riemann problems.
Recall we have $S$ of (2.7) denoting the training data set, which include solution averages spread over the spatial domain with index $j$ and up to time levels $t_{m+1}$. Numerical tests show that for linear partial differential equations, training data of $\left(t_0, t_1\right)$ with $m = 0$ in $S$ is sufficient for obtaining optimal parameter set $\Theta$. For nonlinear partial differential equations, i.e. the Burgers equation, multiple time levels ($m > 0$ in $S$) data are necessary to guarantee the neural network effectively learning the solution evolution mechanism, for example capturing the right shock speed.
As discussed in section 2.1, the network parameter set $\Theta$ behave more or less as the coefficients of a finite volume scheme. Parameter set $\Theta$ is independent of at what cell location and at what
time level the neural network method is applied. Thus we should minimize the loss function over all spatial index \( j \) and at all time levels in \( S \) to obtain the optimal parameter set \( \Theta^* \) or the coefficients of the neural network method.
One epoch of iteration is defined as the following. The weights and biases are first randomly generated from normal distribution around zero. The parameter set \( \Theta \) are updated sequentially through space and time with all data pairs applied \textit{once} in the training set \( S \). In a word, we start with first time level pair \((t_0, t_1)\) and go through all spatial cells and iteratively update the parameter set \( \Theta^{j,t_0} \) through the gradient descent direction of (2.9). Then we move on to the next time level pairs \((t_n, t_{n+1}) \) \((n \geq 1)\) and go through all spatial cells again updating the parameter set \( \Theta^{j,t_n} \). Below is an illustration of one epoch definition, for which we use solution averages pairs in the training set \( S \) once iterating the network parameter, see Figure 3.
\[
\Theta_{\text{old}}^{j,t_n} \rightarrow \Theta_{\text{new}}^{j,t_n}, \quad j = 1, \cdots, J, \quad n = 0, \cdots, m.
\]
The superscripts \( j \) and \( t_n \) notations are added to specify the iteration or update is processed at what cell and at what time level. They all refer to the update of the same optimized parameter set \( \Theta^* \). Learning rate \( \alpha = 0.01 \) or \( \alpha = 0.001 \) is applied, when updating through stochastic gradient descent method.
In this paper, we have integer \( K \) introduced denoting the total number of epochs iterations involved in training. Instead of iterating through one epoch defined above, at each time level we run total \( K \) iterations to update the parameter set \( \Theta \) with one iteration defined as the one round of updating over spatial domain (index \( j \)). We then continue the iteration of \( \Theta \) to the next time level, which can be summarized as
\[
\Theta^{j,t_n,i} \rightarrow \Theta^{j,t_n,i+1}, \quad j = 1, \cdots, J, \quad i = 1, \cdots, K, \quad n = 0, \cdots, m. \tag{2.10}
\]
In the end, the parameter set \( \Theta \) have been iteratively updated with total \( J \times K \times (m + 1) \) times. Specifically for the last time level pair \((t_m, t_{m+1})\), we record the squared \( L_2 \) error defined below corresponding to iteration
\[
L_2^2(t_{m+1}) = \sum_{j=1}^{J} \Delta x L_{j,t_{m}}(\Theta) = \sum_{j=1}^{J} (\bar{v}_{j_{\text{out}}}^m - \bar{u}_{j_{\text{in}}}^{m+1})^2 \Delta x. \tag{2.11}
\]
We output the squared \( L_2 \) error of (2.11) corresponding to iteration index \( i = 1, \cdots, K \) to demonstrate the effectiveness of cell-average based neural network method. Now we conclude the section with comment on the minimum size of training data set \( S \) of (2.7), which are purely lab results observed from numerical tests.
\textbf{Remark 1.} For linear partial differential equations, one time level solution averages in the training set \( S \) corresponding to \((t_0, t_1)\) or \( m = 0 \) in (2.7), is sufficient for obtaining an effective neural network. For nonlinear partial differential equations, it is necessary to include multiple time levels \((m > 0)\) of solution averages in the training set \( S \) to have the neural network learn the evolution mechanism successfully.
For example in the numerical section when we solve inviscid Burgers’ equation with smooth \( \sin(x) \) initial in Example 3.3.1, we have solution averages data pairs up to \( t = 2 \) included in the training set \( S \). This is because a shock starts to develop at \( t = 1 \). We need to have more time levels solution averages included in the training set to make sure the neural network solver be able to learn the shock capturing mechanism.
2.3. Implementation and summary of cell-average based neural network method
With the optimal weights and biases $\Theta^*$ obtained and the neural network $N(\vec{V}_j^n; \Theta^*)$ well defined and available, the cell-average based neural network method can be implemented as a regular explicit finite volume scheme as below
$$\bar{v}_j^{n+1} = \bar{v}_j^n + N(\vec{V}_j^n; \Theta^*), \quad \forall j = 1, \cdots, J, \quad \forall n = 0, 1, 2, \cdots.$$ \hspace{1cm} (2.12)
Again, with the spatial and time step sizes of $\Delta x$ and $\Delta t$ and the previously chosen network input vector of $\vec{V}_j^n = [\bar{v}_{j-p}^n, \cdots, \bar{v}_{j-1}^n, \bar{v}_j^n, \bar{v}_{j+1}^n, \cdots, \bar{v}_{j+q}^n]^T$, together with the network optimal parameter set $\Theta^*$, we have a complete definition of a neural network method.
For cell-average based neural network method (2.12), we assign ghost cell values and apply boundary conditions as a regular finite volume method. Again, we only consider Dirichlet or periodic boundary conditions in this paper. A generic stencil include $p$ cells to the left and $q$ cells to the right to evolve current cell to the next time level. Similar to the training process discussed in section 2.2, we simply copy solution averages from inside the domain for periodic boundary conditions. And we assign exact solution values to those out of domain ghost cells for Burgers’ equation Riemann problems, i.e. for Dirichlet boundary conditions.
One amazing result is that cell-average based neural network method can be relieved from the CFL restriction on time step size, especially for parabolic PDEs. With explicit discretization in time, classical numerical method requires time step size to be as small as $\Delta t \approx (\Delta x)^2$, which is very expensive for multi-dimensional problems. It turns out neural network method can adapt to any time step size $\Delta t$ that is independent of spatial size $\Delta x$, and still a stable method can be obtained. Once well trained, the neural network method can efficiently and accurately evolve the solution forward in time as an explicit method.
Recall our neural network method is based on the following integral format of the partial differential equation $u_t + \mathcal{L}(u) = 0$
$$\bar{u}_j(t_{n+1}) - \bar{u}_j(t_n) = \frac{1}{\Delta x} \int_{t_n}^{t_{n+1}} \int_{I_j} \mathcal{L}(u) \, dx \, dt,$$
with $\mathcal{L}$ as the differentiation operator on the spatial variable. It is equivalent to the PDEs given in differentiation format. Due to some mysterious reason, the neural network method is able to catch up solution information around time level $t_{n+1}$ thus allows large time step size evolution as an implicit method. Below we summarize the major result of our CANN method.
**Lemma 1.** Even being an explicit scheme, cell-average based neural network method (2.12) can adapt to any time step size $\Delta t$ for hyperbolic and parabolic PDEs (2.1). With $\Delta t = \Delta x$, first order of accuracy is obtained with neural network method (2.12) for linear advection and linear convection diffusion equations. For Heat equation, neural network method errors depend only on spatial mesh size $\Delta x$ and are independent of time step size $\Delta t$.
Besides accuracy and error behavior studies, we also carry our a series of numerical tests with cell-average based neural network method. Now we list the advantages of our CANN method that are not common for classical numerical methods.
- Allow large time step size evolution, i.e. $\Delta t = 8\Delta x$ for linear hyperbolic PDEs.
- For Heat equation, errors are independent of time step size. Similar errors are obtained for $\Delta t = \Delta x$, $\Delta t = 2\Delta x$ and $\Delta t = 4\Delta x$ when same spatial mesh $\Delta x$ is adapted.
• Introduce almost zero artificial numerical diffusion for contact discontinuity propagation.
• Introduce little numerical dissipation and dispersive errors after long time run.
Remark 2. Once one cell-average based neural network is well trained and available, it can be applied to solve same PDE associated with different initials and over different domains.
Remark 3. It remains unknown how to choose a suitable neural network architecture in terms of number of hidden layers and neurons per layer. Numerical tests show one or two hidden layers with a few neurons work well. We further mention that some network structure may lead to extremely small errors and can be even regarded as a perfect solver, for which regular order of convergence over refined mesh error analysis does not hold anymore.
3. Numerical Example
In this section, we carry out a series of numerical tests to check out the accuracy and capability of cell-average based neural network methods. We start with linear advection equation, Heat equation and convection diffusion equations to investigate if or not order of convergence can be observed. Then we move on to nonlinear hyperbolic conservation law to test the capability of neural network method capturing shock and rarefaction waves propagation.
Over the section we adapt $T$ as a generic final time at where we compute the errors and orders. As mentioned previously, time step size $\Delta t$ is chosen before and is a part of the definition of a neural network method (2.12). Thus we always have $T$ as an integer multiple times of $\Delta t$. Below we list the $L_2$ and $L_\infty$ errors formula as used in finite volume methods.
\[
Error_{L_2}(T) = \sqrt{\sum_{j=1}^{J} (\bar{v}_j(T) - \bar{u}_j(T))^2 \Delta x}
\]
\[
Error_{L_\infty}(T) = \max_{j=1}^{J} |\bar{v}_j(T) - \bar{u}_j(T)|
\]
Again, we have $\bar{v}_j(T)$ denote our CANN method (2.12) solution on cell $j$ and at final time $T$. We have $\bar{u}_j(T)$ denote either the exact solution average or the reference solution average on cell $j$ and at time $T$ that is obtained from a highly accurate numerical method.
As marked down in Definition 1, four components of $\Delta x$, $\Delta t$, network input vector $\vec{V}_j^{in}$ and the structure of neural network (number of hidden layers and neurons per layer) together identify one neural network solver (2.12). Most of the time we follow the principle listed in section 2.1 to choose suitable stencil width or the network input vector of $\vec{V}_j^{in}$. We highlight that for linear PDEs one time level $(t_0, t_1)$ solution average data pair is applied training the network and for nonlinear PDEs multiple time levels solution average data pair are needed to obtain an effective neural network solver, see Remark 1. We also mention that squared $L_2$ error of (2.11) around $10^{-8}$ or smaller is used as stop condition in training.
3.1. Linear advection equation
In this subsection, we focus on linear hyperbolic equation of
\[
u_t + u_x = 0.
\]
Even the above linear equation is a simple model, quite a few problems can be tested to evaluate the capability of a numerical method. We consider three problems for advection (3.3). One is about
smooth function evolution and we check if order of convergence can be observed. Then we study the contact discontinuity propagation problem. It is not a trivial test, since numerical methods tend to either generate smeared out approximations or numerical oscillations around the discontinuity. For the third test, we have the neural network method simulate wave propagation after long time run. Neural network method is quite stable and produces little dispersive and dissipation errors after long time simulation.
Example 3.1.1. smooth wave propagation
In this example we solve (3.3) with initial condition $u(x,0) = \sin(x)$. Spatial domain is taken as $D = [0, 2\pi]$ and the exact solution $u(x,t) = \sin(x-t)$ is a smooth wave propagating from left to right. We consider a simple neural network with input vector of
$$\vec{V}_j^m = \begin{bmatrix} \bar{u}_{j-1}^n, \bar{u}_j^n \end{bmatrix}^T,$$
that is similar to the upwind finite volume scheme. The network picked consists of 2 hidden layers with 5 neurons per layer. Total iterations of $K = 5 \times 10^5$ is applied to train the networks. Periodic boundary conditions are applied. We carry out two accuracy tests, for which we compute the $L_2$ and $L_\infty$ errors of (3.1) and (3.2) at final time $T = \pi$.
Case I:
For the first case study we mimic the accuracy check of a standard numerical method. We consider four spatial meshes of $\Delta x = \pi/10, \pi/20, \pi/40, \pi/80$. For each cell size $\Delta x$, time step size $\Delta t = \Delta x$ is taken correspondingly. The four well trained neural network solvers are very similar to each other except the mesh size. All four network solvers are used to solve the smooth wave propagation to final time $T = \pi$ and $L_2$ and $L_\infty$ errors are computed and listed in Table 1.
| $\Delta x$ | $L_2$ | order | $L_\infty$ | order |
|------------|-----------|-------|------------------|-------|
| $\pi/10$ | $1.8756e^{-2}$ | | $1.0237e^{-2}$ | |
| $\pi/20$ | $8.0830e^{-3}$ | $1.21$| $4.7403e^{-3}$ | $1.11$|
| $\pi/40$ | $1.5547e^{-3}$ | $2.39$| $9.7037e^{-4}$ | $2.29$|
| $\pi/80$ | $6.3500e^{-4}$ | $1.28$| $3.9838e^{-4}$ | $1.28$|
Table 1: Errors and orders of neural network methods for Example 3.1.1 (Case I), $\Delta t = \Delta x$
Case II:
For this case we keep the spatial mesh size $\Delta x = \pi/40$ fixed, but consider three different time step sizes of $\Delta t = 2\Delta x, \Delta t = 5\Delta x$ and $\Delta t = 8\Delta x$. The three neural network solvers apply same upwind like network input vector of (3.5). The only difference is the time step size. We list the $L_2$ and $L_\infty$ errors computed at final time $T = \pi$ in Table 2. Three network solvers give similar errors and it is not clear if the network errors relate to $\Delta t$. For all three time step sizes, i.e. $\Delta t = 8\Delta x$, the principle of having domain of dependence included is not followed. We simply apply (3.5) as the input vector that is like the upwind scheme. These settings conflict with method of characteristic. But all neural network solvers work well and give errors similar to those in case I.
Example 3.1.2. contact discontinuity
In this example, we solve advection equation (3.3) with initial condition
$$u(x,0) = \begin{cases} 1 & x \leq 0, \\ 2 & x > 0, \end{cases}$$
Table 2: Errors and orders of neural network methods for Example 3.1.1 (Case II), $\Delta x = \pi/40$ fixed with varying $\Delta t$ over domain $D = [-1, 4]$. Periodic boundary condition is applied. The contact discontinuity is initially located at $x = 0$, which moves back into the domain after $t > 4$. Same network input vector of (3.5) that is similar to upwind scheme is considered. One hidden layer of 10 neurons is the chosen network structure and a total of $K = 5 \times 10^5$ iterations are applied training the network. We have $\Delta x = \Delta t = \frac{1}{20}$. Snapshots of our CANN method simulation are presented in Figure 4. The contact discontinuity is sharply resolved, even after one period of evolution at $t = 5$. No oscillation is observed and there is almost no artificial diffusion introduced with the neural network method. Numerical simulation with CANN method behaves better than many numerical methods that tend to generate smeared out simulations.
**Example 3.1.3. long time evolution**
In this example we solve advection equation (3.3) with initial condition
$$u(x, 0) = \sin(x),$$
over domain $D = [0, 2\pi]$. We choose $\Delta x = \frac{2\pi}{100}$ and $\Delta t = 4\Delta x$. Long time simulation with CANN method is considered. To include the domain of dependence, following neural network input vector is taken
$$\vec{V}_j^n = \left[ \bar{u}^n_{j-6}, \bar{u}^n_{j-5}, \bar{u}^n_{j-4}, \bar{u}^n_{j-3}, \bar{u}^n_{j-2}, \bar{u}^n_{j-1}, \bar{u}^n_{j} \right]^T.$$ (3.5)
The chosen network is with 1 hidden layer and 10 neurons. A total of $K = 5 \times 10^5$ iterations are applied. We compute $L_2$ and $L_\infty$ errors at different time locations, which are listed in Table
3. After several periods, neural network method is still able to accurately capture the solution evolution and gives small dispersive and dissipation errors.
\[ t = 4\pi/5 \quad t = 2\pi \quad t = 4\pi \quad t = 8\pi \]
| | \[L_2\] | \[L_\infty\] |
|----------|-------------|--------------|
| \[t = 4\pi/5\] | \[9.1869e^{-14}\] | \[1.2620e^{-13}\] |
| \[t = 2\pi\] | \[1.0045e^{-12}\] | \[1.0600e^{-12}\] |
| \[t = 4\pi\] | \[1.8625e^{-10}\] | \[1.7158e^{-10}\] |
| \[t = 8\pi\] | \[7.1709e^{-6}\] | \[5.3325e^{-6}\] |
Table 3: \(L_2\) and \(L_\infty\) errors (Example 3.1.3) after long time simulation with neural network method
3.2. Linear convection diffusion equation
In this section, we solve linear convection diffusion equation with cell-average based neural network method. It is further confirmed that we can be relieved from explicit scheme small time step size restriction. For Heat equation, numerical tests show the errors only relate to spatial mesh size \(\Delta x\) and is independent of time step size \(\Delta t\).
Example 3.2.1. Heat equation
In this example, we consider solving the Heat equation
\[ u_t = u_{xx}, \]
with initial \(u(x,0) = \sin(\pi x)\) and over domain \(D = [0,1]\). Exact solution is \(u(x,t) = e^{-\pi^2 t} \sin(\pi x)\).
Different to hyperbolic PDEs, we have infinite speed of propagation for Heat equation. Here we follow a similar to central scheme symmetric mechanism to pick up the network input vector as
\[ \vec{V}_j^m = [\vec{u}_{j-3}^n, \vec{u}_{j-2}^n, \vec{u}_{j-1}^n, \vec{u}_j^n, \vec{u}_{j+1}^n, \vec{u}_{j+2}^n, \vec{u}_{j+3}^n]^T. \]
(3.6)
The chosen network is with 2 hidden layers and 15 neurons per layer. Iterations are run up to \(K = 10^5\) times to optimize the network parameter set. Periodic boundary conditions are applied. Final time \(T = 0.1\) is adapted to compute \(L_2\) and \(L_\infty\) errors of (3.1) and (3.2). We carry out three accuracy tests. For the first and third tests, we always choose \(\Delta t = \Delta x\) and have \(\Delta x\) refined to check out order of convergence. In the second test we have \(\Delta x\) fixed and choose \(\Delta t\) as a multiple times of \(\Delta x\). Same network structure is applied, but different choices of \(\Delta t, \Delta x\) or network input vector \(\vec{V}_j^m\) are considered in each test.
Case I:
For this case, we investigate the convergence order of neural network method. We consider four settings of \(\Delta x = \frac{1}{40}, \frac{1}{80}, \frac{1}{160}, \frac{1}{320}\). For each cell size \(\Delta x\), we choose \(\Delta t = \Delta x\) correspondingly. The four network solvers are similar to each other except the mesh size. We solve the Heat equation with each neural network solver to \(T = 0.1\) and compute the \(L_2\) and \(L_\infty\) errors. In Table 4 we list all errors and orders. Clean first order of convergence is observed with CANN method.
| \(\Delta x\) | \(L_2\) | \[order\] | \(L_\infty\) | \[order\] |
|---------------|-------------------|-----------|------------------------|-----------|
| \(1/40\) | \(8.6949e^{-3}\) | 2.0873e^{-2} | \(1.2620e^{-13}\) | 0.94 |
| \(1/80\) | \(4.5270e^{-3}\) | 0.94 | \(1.4104e^{-2}\) | 0.57 |
| \(1/160\) | \(2.4736e^{-3}\) | 0.87 | \(7.2650e^{-3}\) | 0.96 |
| \(1/320\) | \(1.2894e^{-3}\) | 0.94 | \(3.7860e^{-3}\) | 0.94 |
Table 4: Errors and orders of neural network methods for Heat equation (Case I), \(\Delta t = \Delta x\)
Case II:
For this case we fix the spatial mesh size $\Delta x = 1/160$ but apply three time step sizes of $\Delta t = \Delta x$, $\Delta t = 2\Delta x$ and $\Delta t = 4\Delta x$. We compute the $L_2$ and $L_\infty$ errors with each neural network solver at final time $T = 0.1$. Errors are listed in Table 5. The three network solvers give similar errors. The accuracy of CANN method seems to be independent of time step size $\Delta t$.
| $\Delta t$ | $L_2$ | $L_\infty$ |
|--------------|-----------|--------------|
| $4\Delta x$ | $2.1981e^{-3}$ | $6.8272e^{-3}$ |
| $2\Delta x$ | $2.4969e^{-3}$ | $7.2399e^{-3}$ |
| $\Delta x$ | $2.4736e^{-3}$ | $7.2650e^{-3}$ |
Table 5: Errors of neural network methods for Heat equation (Case II), $\Delta x = 1/160$ fixed, different $\Delta t$
Case III:
Motivated by infinite speed of propagation, we wonder if the error and accuracy may be improved with increased stencil width of the network input vector, the $p$ and $q$ values of (2.6). Three spatial mesh sizes of $\Delta x = 1/40, 1/80, 1/160$ with $\Delta t = \Delta x$ are studied. Different to case I, we gradually increase the input vector stencil width with refined mesh. We have $p = q = 2$ for $\Delta x = 1/40$, $p = q = 4$ for $\Delta x = 1/80$ and $p = q = 8$ for $\Delta x = 1/160$. The three networks use same spatial cells to evolve the solution average. Errors are listed in Table 6. There is no sign of improvement with wider stencil included.
| $\Delta x$ | $L_2$ | $L_\infty$ |
|--------------|-----------|--------------|
| $1/40$ | $7.1179e^{-3}$ | $1.8046e^{-2}$ |
| $1/80$ | $6.3502e^{-3}$ | $1.8319e^{-2}$ |
| $1/160$ | $6.4138e^{-3}$ | $2.1510e^{-2}$ |
Table 6: Errors of neural network methods for Heat equation (Case III), $\Delta t = \Delta x$, wider stencil on refined mesh
Example 3.2.2. Linear convection diffusion equation
In this subsection, we consider linear convection-diffusion equation
$$u_t = u_{xx} + u_x, \quad x \in D, \quad t \geq 0,$$
with initial $u(x, 0) = \sin(x)$. Spatial domain is $D = [0, 2\pi]$. We further check whether the first order of accuracy can be obtained. We choose same network input vector of (3.6) as for Heat equation. Periodic boundary conditions are considered. Exact solution is available with $u(x, t) = e^{-t} \sin(x+t)$. The picked network structure involves 1 hidden layer and 15 neurons. Number of iterations is taken as $K = 5 \times 10^6$. We adapt final time $T = \pi/4$ to compute all errors and orders.
For the order of convergence test, four spatial mesh sizes of $\Delta x = \frac{\pi}{40}, \frac{\pi}{80}, \frac{\pi}{160}, \frac{\pi}{320}$ are studied. For each $\Delta x$, we choose $\Delta t = \Delta x$ correspondingly. Again, the four neural network solvers are similar to each other except the mesh size. In Table 7, we list the $L_2$ and $L_\infty$ errors. Roughly first order of convergence is observed.
The second test is similar to the Case II of Example 3.2.1. We fix the spatial size $\Delta x = \pi/160$ and vary the time step size from $\Delta t = \Delta x$, $\Delta t = 2\Delta x$ to $\Delta t = 4\Delta x$. The computed $L_2$ and $L_\infty$ errors of the three network solvers are listed in Table 8. The errors seem to be related to time step size $\Delta t$, even the relationship is not clear. Again, the cell-average based neural network allows large time step. Notice the choice of $\Delta t = 4\Delta x$ involves a time step size roughly 200 times bigger than the regular CFL restriction of $\Delta t \approx \Delta x^2$.
Table 7: Errors and orders of neural network methods for linear convection diffusion equation, $\Delta t = \Delta x$.
| $\Delta x$ | $L_2$ | order | $L_{\infty}$ | order |
|------------|-------|-------|--------------|-------|
| $\pi/40$ | $5.3013e^{-3}$ | | $3.2397e^{-3}$ | |
| $\pi/80$ | $1.3801e^{-3}$ | 1.94 | $7.5132e^{-4}$ | 2.11 |
| $\pi/160$ | $5.0091e^{-4}$ | 1.46 | $2.6084e^{-4}$ | 1.52 |
| $\pi/320$ | $2.6771e^{-4}$ | 0.91 | $1.5045e^{-4}$ | 0.89 |
Table 8: Errors of neural network methods for linear convection diffusion equation, $\Delta x = \pi/160$ fixed, different $\Delta t$.
| $\Delta t$ | $L_2$ | $L_{\infty}$ |
|------------|-------|--------------|
| $4\Delta x$ | $1.0249e^{-3}$ | $5.8832e^{-4}$ |
| $2\Delta x$ | $8.2225e^{-4}$ | $4.2640e^{-4}$ |
| $\Delta x$ | $5.0091e^{-4}$ | $2.6084e^{-4}$ |
| $\Delta x/2$ | $3.7288e^{-4}$ | $2.0151e^{-4}$ |
### 3.3. Nonlinear hyperbolic equation
In this section, we investigate the effectiveness of neural network method solving nonlinear conservation laws. We consider the inviscid Burgers’ equation
$$u_t + \left(\frac{u^2}{2}\right)_x = 0, \quad (x,t) \in D \times R^+.$$ (3.7)
Four benchmark problems are studied to illustrate the computational challenges of nonlinear hyperbolic PDEs. We first consider the smooth initial $\sin(x)$ wave which develops into a shock after finite time evolution. Then we study two Riemann problems with two piece wise constants initials that will develop into a shock and a rarefaction wave. Last example involves three piece wise constants initial that will develop into the interaction between rarefaction wave and shock. In all four examples, Dirichlet boundary conditions are applied.
For nonlinear problems, multiple time levels of data pairs are necessary and applied training the neural network. For all four examples, time step size $\Delta t = 0.1$ is taken. We roughly have $\Delta t \approx 2\Delta x$. Solution cell average values up to $t = 2.0$, as listed below
$$S = \left\{ \left(\bar{u}_j^n, \bar{u}_{j+1}^{n+1}\right), \ j = 1, \cdots, J \right\}_{n=0}^{m=19},$$ (3.8)
are included in the training data set. So we have twenty time levels of solution average pairs $(\bar{u}_j^n, \bar{u}_{j+1}^{n+1})$ used in training the network. Spatial domain is either $D = [0, 2\pi]$ or $D = [-1, 5]$. Mesh size of $\Delta x = \frac{2\pi}{100}$ for example 3.3.1 and $\Delta x = \frac{6}{100}$ for other three examples are applied.
**Example 3.3.1. Smooth initial sine wave**
We first consider the inviscid Burgers equation (3.7) associated with smooth initial value of
$$u(x, 0) = \sin(x),$$
over domain $D = [0, 2\pi]$. Zero boundary conditions $u(0, t) = u(2\pi, t) = 0$ and its zero extension to out of domain ghost cells are applied. With $\Delta t = 0.1$ and $\Delta x = 2\pi/100$ and characteristic speed less than one, the network input vector is taken as
$$\bar{V}_j^m = \left[ \bar{u}_j^{n-3}, \bar{u}_j^{n-2}, \bar{u}_j^{n-1}, \bar{u}_j^n, \bar{u}_{j+1}^n, \bar{u}_{j+2}^n, \bar{u}_{j+3}^n \right]^T.$$
This stencil includes the characteristic and the domain of dependence. Training data pairs $(\bar{u}_j^n, \bar{u}_{j+1}^n)$ are obtained from highly accurate discontinuous Galerkin method. The network itself contains 2 hidden layers with 8 neurons per layer. The network training is conducted for up to $K = 10^5$ iterations. We also output the squared $L_2$ training error of (2.11) corresponding to
iteration at the very last time level \( t = 2 \), see Fig 5.
Well trained network is applied to simulate solution evolution to \( T = 3 \). Before and after shock developed are illustrated in Fig 6. The cell-average based neural network results are comparable to those obtained with classical numerical methods. Shock evolution is sharply captured and no oscillation is generated with the neural network method.
**Example 3.3.2. Single shock propagation**
In this example, we consider a Riemann problem of (3.7) with piece wise constant initial
\[
u(x,0) = \begin{cases}
1, & x < 0, \\
0, & \text{otherwise}.
\end{cases}
\]
Computational domain is \( D = [-1,5] \). Dirichlet boundary conditions of \( u(-1,t) = 1, u(5,t) = 0 \) and its out of domain extensions are applied. Cell size \( \Delta x = 0.06 \) and time step size \( \Delta t = 0.1 \) are taken. The network input vector of (2.6) is chosen as
\[
\overrightarrow{V_j}^n = \left[ \bar{u}_{j-4}^n, \bar{u}_{j-3}^n, \bar{u}_{j-2}^n, \bar{u}_{j-1}^n, \bar{u}_j^n, \bar{u}_{j+1}^n, \bar{u}_{j+2}^n \right]^T.
\]
This is a biased choice which includes more points to the left. Again the choice covers the domain of dependence. Training data are obtained from the exact solution
\[
u(x,t) = \begin{cases}
1, & x \leq \frac{1}{2}t, \\
0, & \text{otherwise}.
\end{cases}
\]
(3.9)
The neural network picked contains 1 hidden layer of 8 neurons. Network training is conducted for up to \( K = 10^5 \) iterations. The squared \( L_2 \) training error of (2.11) to iteration at \( t = 2 \) are output in Fig 5. After well trained, the neural network is applied solving the Riemann problem to \( T = 8.0 \). Screen shots of \( t = 0, t = 1.5, t = 3.0 \) and \( t = 6.0 \) are shown in Figure 7. The neural network method can accurately and sharply capture the shock evolution. Notice the neural network simulation at \( t = 8 \) is way later than the training data set of \( t = 2 \).
Example 3.3.3. *Rarefaction wave*
In this example, we consider a Riemann problem of (3.7) with piece wise constant initial
\[ u(x, 0) = \begin{cases} 0, & x < 0, \\ 1, & \text{otherwise}. \end{cases} \]
Domain is \( D = [-1, 5] \). Dirichlet boundary condition \( u(-1, t) = 0 \), \( u(5, t) = 1 \) and its out of domain extension are applied. Cell size \( \Delta x = 0.06 \) and time step size \( \Delta t = 0.1 \) are chosen. Network input vector is taken as
\[ \vec{V}^{in}_{j} = [\bar{u}^{n}_{j-2}, \bar{u}^{n}_{j-1}, \bar{u}^{n}_{j}, \bar{u}^{n}_{j+1}]^{T}. \]
The choice still includes the domain of dependence. Training data are generated from exact solution
\[ u(x, t) = \begin{cases} 0, & x < 0, \\ \frac{x}{t}, & 0 \leq x \leq t, \\ 1, & \text{otherwise}. \end{cases} \]
The network contains 2 hidden layers with 8 neurons per layer. Iteration steps up to \( K = 10^5 \) are applied. The squared \( L_2 \) training error of (2.11) corresponding to iteration at \( t = 2 \) are also included in Figure 5. After well trained, the neural network solver is applied solving this Riemann problem up to \( T = 4.0 \). Screen shots are shown in Figure 8. The neural network method can open up and well resolves the rarefaction wave evolution.
Example 3.3.4. *Interaction of rarefaction and shock waves*
In this example, we consider solving (3.7) with three piece wise constants initial
\[ u(x, 0) = \begin{cases} 0, & x < 0, \\ 1, & 0 \leq x \leq 1, \\ 0, & \text{otherwise}. \end{cases} \]
Spatial domain is set as $D = [-1, 5]$. Dirichlet boundary conditions $u(-1, t) = u(5, t) = 0$ and its out of domain extension are applied. Network input vector is taken as
$$
\vec{V}^n_j = \begin{bmatrix}
\bar{u}^n_{j-4}, \bar{u}^n_{j-3}, \bar{u}^n_{j-2}, \bar{u}^n_{j-1}, \bar{u}^n_{j}, \bar{u}^n_{j+1}, \bar{u}^n_{j+2}
\end{bmatrix}^T.
$$
Training data pairs $(\bar{u}^n_j, \bar{u}^n_{j+1})$ are generated from the exact solution. Here the rarefaction wave travels faster than the shock and it merges into the shock wave after a while. Before the meet at $t = 2$, the exact solution is given as
$$
u(x, t) = \begin{cases}
0, & x < 0, \\
\frac{x}{t}, & 0 \leq x < t, \\
1, & t \leq x \leq 1 + \frac{t}{2}, \\
0, & \text{otherwise}.
\end{cases}
$$
After $t > 2$, the rarefaction wave runs into the shock wave. The solution is composed with zero state to the left, followed by a rarefaction wave and connected with a shock to the right with which the shock speed slows down as time evolves. The shock speed is determined by the following formula
$$
\sigma'(t) = \frac{1}{2} \left( \frac{\sigma(t)}{t} \right)^2 = \frac{\sigma(t)}{2t}.
$$
With initial value $\sigma(2) = 2$, we obtain $\sigma(t) = \sqrt{2t}$. For $t > 2$, the exact solution is given by
$$
u(x, t) = \begin{cases}
0, & x < 0, \\
\frac{x}{t}, & 0 \leq x \leq \sqrt{2t}, \\
0, & \sqrt{2t} < x.
\end{cases}
$$
Figure 9: rarefaction wave and shock wave interaction (Example 3.3.4) with neural network method
The picked network contains 2 hidden layers with 8 neurons per layer. Training is conducted up to $K = 10^5$ iterations. Squared $L_2$ training error of (2.11) corresponding to iteration at $t = 2$.
is listed in Figure 5. After well trained, neural network method is applied with its solution screen shots before and after waves interaction shown in Figure 9. Recall the training set only include solution pairs up to \( t = 2 \) for which the rarefaction wave has not met the shock wave yet. The neural network method is capable of accurately capturing shock propagation after the interaction. We mention that we also check out the \( L_2 \) errors, which are around \( O(10^{-3}) \) for all four examples, at a time when solutions all develop singularities. Recall the cell size is around \( \Delta x \approx 0.06 \).
### 3.4. Nonlinear convection diffusion equation
In this section, we investigate the viscous Burgers’ equation
\[
u_t + \left( \frac{u^2}{2} \right)_x = \mu u_{xx}, \quad (x, t) \in D \times R^+, \tag{3.10}
\]
with \( \mu = 0.1 \). Zero boundary conditions of \( u(0, t) = u(2\pi, t) = 0 \) are applied. Same time step size \( \Delta t = 0.1 \) and cell size \( \Delta x = \frac{2\pi}{100} \) as in Example 3.3.1 are taken. The cell-average based neural network method can adapt large time step size \( \Delta t \). We use this setting to simply illustrate the effectiveness and efficiency of neural network method solving nonlinear convection diffusion equations. The network input vector is taken the same as that of Example 3.3.1
\[
\overrightarrow{V}_j^n = [\overline{u}_{j-3}^n, \overline{u}_{j-2}^n, \overline{u}_{j-1}^n, \overline{u}_j^n, \overline{u}_{j+1}^n, \overline{u}_{j+2}^n, \overline{u}_{j+3}^n]^T.
\]
Training data \( \{\overline{u}_j^n, \overline{u}_j^{n+1}\} \) are generated from the refined mesh highly accurate discontinuous Galerkin method. With time step \( \Delta t = 0.1 \) and up to \( t = 2.0 \), twenty time levels of solution averages are applied in the training data set. The network structure contains 2 hidden layers of each with 8 neurons. The network training is conducted for up to \( K = 10^5 \) iterations.
Solution evolution run by neural network method is shown in Figure 10. The results are comparable to those obtained from classical numerical methods.
4. Conclusions
A finite volume or cell-average based neural network method is developed for hyperbolic and parabolic partial differential equations. Classical numerical methods and principles motivate and guide the design of machine learning neural network methods. It is found the cell-average based neural network method can be relieved from the explicit scheme CFL restriction and is able to evolve the solution forward in time with large time step size. The method is able to accurately capture shock and rarefaction waves. We believe this is the right direction to explore neural networks solvers for partial differential equations, which is built upon solution properties and successful numerical methods.
References
[1] Y. LeCun, Y. Bengio, Convolutional networks for images, speech, and time-series, The handbook of brain theory and neural networks (1995).
[2] Y. Bengio, Learning Deep Architectures for AI, Found. Trends Mach. Learn. 2 (2009) 1–127.
[3] A. Krizhevsky, I. Sutskever, G. Hinton, Imagenet classification with deep convolutional neural networks, Advances in Neural Information Processing Systems 25 (2012) 1097–1105.
[4] Y. LeCun, Y. Bengio, G. Hinton, Deep learning, Nature 521 (2015) 436–444.
[5] W. E, A proposal on machine learning via dynamical systems, Commun. Math. Stat. 5 (2017) 1–11.
[6] P. Chaudhari, A. Oberman, S. Osher, S. Soatto, G. Carlier, Deep relaxation: partial differential equations for optimizing deep neural networks, 2017. arXiv:1704.04932.
[7] S. H. Rudy, S. L. Brunton, J. L. Proctor, J. N. Kutz, Data-driven discovery of partial differential equations, Science Advances 3 (2017) e1602614.
[8] B. Chang, L. Meng, E. Haber, L. Ruthotto, D. Begert, E. Holtham, Reversible architectures for arbitrarily deep residual neural networks, in: Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence, (AAAI-18), 2018, AAAI Press, 2018, pp. 2811–2818.
[9] Z. Long, Y. Lu, B. Dong, PDE-Net 2.0: learning PDEs from data with a numeric-symbolic hybrid deep network, J. Comput. Phys. 399 (2019) 108925, 17.
[10] L. Ruthotto, E. Haber, Deep neural networks motivated by partial differential equations, J. Math. Imaging Vision 62 (2020) 352–364.
[11] J. He, J. Xu, MgNet: a unified framework of multigrid and convolutional neural network, Sci. China Math. 62 (2019) 1331–1354.
[12] G. Cybenko, Approximation by superpositions of a sigmoidal function, Mathematics of Control, Signals and Systems 2 (1989) 303–314.
[13] K. Hornik, M. Stinchcombe, H. White, Universal approximation of an unknown mapping and its derivatives using multilayer feedforward networks, Neural Networks 3 (1990) 551 – 560.
[14] A. R. Barron, Universal approximation bounds for superpositions of a sigmoidal function, IEEE Transactions on Information Theory 39 (1993) 930–945.
[15] M. Leshno, V. Y. Lin, A. Pinkus, S. Schocken, Multilayer feedforward networks with a nonpolynomial activation function can approximate any function, Neural Networks 6 (1993) 861–867.
[16] A. Pinkus, Approximation theory of the mlp model in neural networks, Acta Numer. 8 (1999) 143–195.
[17] I. Lagaris, A. Likas, D. Fotiadis, Artificial neural networks for solving ordinary and partial differential equations, IEEE transactions on neural networks 95 (1998) 987–1000.
[18] K. Rudd, S. Ferrari, A constrained integration (cint) approach to solving partial differential equations using artificial neural networks, Neurocomputing 155 (2015) 277–285.
[19] J. Berg, K. Nyström, A unified deep artificial neural network approach to partial differential equations in complex geometries, Neurocomputing 317 (2018) 28–41.
[20] J. Sirignano, K. Spiliopoulos, DGM: a deep learning algorithm for solving partial differential equations, J. Comput. Phys. 375 (2018) 1339–1364.
[21] M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics informed deep learning (part I): data-driven solutions of nonlinear partial differential equations (2017). arXiv:1711.10561.
[22] M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics informed deep learning (part II): data-driven discovery of nonlinear partial differential equations, arXiv abs/1711.10566 (2017).
[23] M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, Journal of Computational Physics 378 (2019) 686–707.
[24] L. Lu, X. Meng, Z. Mao, G. E. Karniadakis, Deepxde: A deep learning library for solving differential equations, SIAM Review 63 (2021) 208–228.
[25] V. Dwivedi, B. Srinivasan, Physics informed extreme learning machine (PIELM)–a rapid method for the numerical solution of partial differential equations, Neurocomputing 391 (2020) 96–118.
[26] X. Jin, S. Cai, H. Li, G. E. Karniadakis, Nsfnets (navier-stokes flow nets): Physics-informed neural networks for the incompressible navier-stokes equations, Journal of Computational Physics 426 (2021) 109951.
[27] Y. Zang, G. Bao, X. Ye, H. Zhou, Weak adversarial networks for high-dimensional partial differential equations, Journal of Computational Physics 411 (2020) 109409.
[28] Z. Cai, J. Chen, M. Liu, Least-squares relu neural network (lsnn) method for linear advection-reaction equation, Journal of Computational Physics (2021) 110514.
[29] Z. Cai, J. Chen, M. Liu, Least-squares relu neural network (lsnn) method for scalar nonlinear hyperbolic conservation law, 2021. arXiv:2105.11627.
[30] G. Kutyniok, P. Petersen, M. Raslan, R. Schneider, A theoretical analysis of deep neural networks and parametric pdes, Constructive Approximation (2021). URL: https://doi.org/10.1007/s00365-021-09551-4.
[31] C. Michoski, M. Mišosavljević, T. Oliver, D. R. Hatch, Solving differential equations using deep neural networks, Neurocomputing 399 (2020) 193–212.
[32] Y. Shin, J. Darbon, G. Em Karniadakis, On the convergence of physics informed neural networks for linear second-order elliptic and parabolic type pdes, Communications in Computational Physics 28 (2020) 2042–2074.
[33] F. Laakmann, P. Petersen, Efficient approximation of solutions of parametric linear transport equations by ReLU DNNs, Adv. Comput. Math. 47 (2021) Paper No. 11, 32.
[34] C. Beck, W. E, A. Jentzen, Machine learning approximation algorithms for high-dimensional fully nonlinear partial differential equations and second-order backward stochastic differential equations, J. Nonlinear Sci. 29 (2019) 1563–1619.
[35] Q. Chan-Wai-Nam, J. Mikael, X. Warin, Machine learning for semi linear PDEs, J. Sci. Comput. 79 (2019) 1667–1712.
[36] H. Pham, X. Warin, M. Germain, Neural networks-based backward scheme for fully nonlinear pdes, SN Partial Differ. Equ. Appl. 2 (2021).
[37] H. Martin, J. Arnulf, K. Thomas, N. T. Anh, A proof that rectified deep neural networks overcome the curse of dimensionality in the numerical approximation of semilinear heat equations, SN Partial Differ. Equ. Appl. 1 (2020) 10.
[38] K. O. Lye, S. Mishra, D. Ray, Deep learning observables in computational fluid dynamics, Journal of Computational Physics 410 (2020) 109339.
[39] Y. Fan, L. Lin, L. Ying, L. Zepeda-Núñez, A multiscale neural network based on hierarchical matrices, Multiscale Model. Simul. 17 (2019) 1189–1213.
[40] Y. Khoo, J. Lu, L. Ying, Solving parametric pde problems with artificial neural networks, European Journal of Applied Mathematics (2020) 1–15.
[41] Y. Li, J. Lu, A. Mao, Variational training of neural network approximations of solution maps for physical models, J. Comput. Phys. 409 (2020) 109338.
[42] N. Winovich, K. Ramani, G. Lin, ConvPDE-UQ: convolutional neural networks with quantified uncertainty for heterogeneous elliptic partial differential equations on varied domains, J. Comput. Phys. 394 (2019) 263–279.
[43] K. Wu, D. Xiu, Data-driven deep learning of partial differential equations in modal space, Journal of Computational Physics 408 (2020) 109307.
[44] T. Qin, Z. Chen, J. D. Jakeman, D. Xiu, Data-driven learning of nonautonomous systems, SIAM Journal on Scientific Computing 43 (2021) A1607–A1624.
[45] D. Ray, J. S. Hesthaven, An artificial neural network as a troubled-cell indicator, J. Comput. Phys. 367 (2018) 166–191.
[46] Y. Wang, Z. Shen, Z. Long, B. Dong, Learning to discretize: Solving 1d scalar conservation laws via deep reinforcement learning, Communications in Computational Physics 28 (2020) 2158–2179.
[47] N. Discacciati, J. S. Hesthaven, D. Ray, Controlling oscillations in high-order discontinuous Galerkin schemes using artificial viscosity tuned by neural networks, J. Comput. Phys. 409 (2020) 109304, 30.
[48] J. Hsieh, S. Zhao, S. Eismann, L. Mirabella, S. Ermon, Learning neural PDE solvers with convergence guarantees, arXiv abs/1906.01200 (2019). URL: http://arxiv.org/abs/1906.01200. arXiv:1906.01200.
[49] Z. Sun, S. Wang, L.-B. Chang, Y. Xing, D. Xiu, Convolution neural network shock detector for numerical solution of conservation laws, Communications in Computational Physics 28 (2020) 2075–2108.
[50] A. D. Beck, J. Zeifang, A. Schwarz, D. G. Flad, A neural network based shock detection and localization approach for discontinuous galerkin methods, Journal of Computational Physics 423 (2020) 109824.
[51] X. Yu, C.-W. Shu, Multi-layer perceptron estimator for the total variation bounded constant in limiters for discontinuous galerkin methods, La Matematica: Official Journal of the Association for Women in Mathematics (submitted). | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2835,
1
],
[
2835,
6831,
2
],
[
6831,
10589,
3
],
[
10589,
13064,
4
],
[
13064,
15563,
5
],
[
15563,
19600,
6
],
[
19600,
22633,
7
],
[
22633,
26492,
8
],
[
26492,
30317,
9
],
[
30317,
33526,
10
],
[
33526,
36879,
11
],
[
36879,
38700,
12
],
[
38700,
42249,
13
],
[
42249,
45826,
14
],
[
45826,
48871,
15
],
[
48871,
49272,
16
],
[
49272,
51310,
17
],
[
51310,
52906,
18
],
[
52906,
54584,
19
],
[
54584,
56801,
20
],
[
56801,
59619,
21
],
[
59619,
62635,
22
],
[
62635,
65426,
23
],
[
65426,
66270,
24
]
]
} |
044970c3b1d3fe3bf1d62d1da6676f63ba6ff3f5 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 20,
"total-fallback-pages": 0,
"total-input-tokens": 53613,
"total-output-tokens": 17946,
"fieldofstudy": [
"Chemistry"
]
} | Recent Advances in Photoacoustic Imaging for Deep-Tissue Biomedical Applications
Sheng Wang1,2,*, Jing Lin*, Tianfu Wang1, Xiaoyuan Chen3,5, Peng Huang1,5
1. Guangdong Key Laboratory for Biomedical Measurements and Ultrasound Imaging, Department of Biomedical Engineering, School of Medicine, Shenzhen University, Shenzhen 518060, China;
2. Key Laboratory of Optoelectronic Devices and Systems of Ministry of Education and Guangdong Province, College of Optoelectronic Engineering, Shenzhen University, Shenzhen 518060, China.
3. Laboratory of Molecular Imaging and Nanomedicine, National Institute of Biomedical Imaging and Bioengineering, National Institutes of Health, Bethesda, Maryland 20892, United States.
* S.W. and J.L. contributed equally to this work.
Corresponding authors: [email protected], [email protected].
© Ivyspring International Publisher. Reproduction is permitted for personal, noncommercial use, provided that the article is in whole, unmodified, and properly cited. See http://ivyspring.com/terms for terms and conditions.
Received: 2016.07.04; Accepted: 2016.08.16; Published: 2016.10.07
Abstract
Photoacoustic imaging (PAI), a novel imaging modality based on photoacoustic effect, shows great promise in biomedical applications. By converting pulsed laser excitation into ultrasonic emission, PAI combines the advantages of optical imaging and ultrasound imaging, which benefits rich contrast, high resolution and deep tissue penetration. In this paper, we introduced recent advances of contrast agents, applications, and signal enhancement strategies for PAI. The PA contrast agents were categorized by their components, mainly including inorganic and organic PA contrast agents. The applications of PAI were summarized as follows: deep tumor imaging, therapeutic responses monitoring, metabolic imaging, pH detection, enzyme detection, reactive oxygen species (ROS) detection, metal ions detection, and so on. The enhancement strategies of PA signals were highlighted. In the end, we elaborated on the challenges and provided perspectives of PAI for deep-tissue biomedical applications.
Key words: Photoacoustic imaging
1. Introduction
Optical imaging techniques, due to their high sensitivity, high temporal resolution and short acquisition time, have received tremendous attention in the field of biomedical applications [1-4]. However, one major challenge of optical imaging techniques is the strong light absorption/scattering of skin, tissue, blood, and so on, leading to limited tissue penetration [5-7]. Conventional optical imaging does not allow for the diagnosis of deep-seated lesions. Therefore, it is highly desirable to develop a novel optical imaging technique to overcome the above bottleneck for deep-seated tissues.
Near-infrared (NIR) optical imaging techniques (e.g. diffuse optical tomography, NIR fluorescence imaging) show great potential to overcome the limitation of tissue penetration for deep tissues imaging because NIR light (700-2,500 nm) can pass through biological tissues (e.g. skin) more efficiently than visible light [8-14]. Particularly, photoacoustic imaging (PAI), which is based on photoacoustic effect, is a promising biomedical imaging modality for deep-tissue applications [15-17]. So far, PAI has been employed to visualize hierarchical biostructures from organelles, cells to organs. As shown in Figure 1, in principle, when laser pulses are applied, some of the delivered optical energy will be absorbed by contrast agents or biological tissues and converted into heat, then a broadband ultrasonic emission will be generated due to the heat-induced transient thermoelastic expansion, which can be detected by an acoustic detector and analyzed to reconstruct PA images. PAI combines the high selectivity of optical
imaging and the deep-tissue penetration of ultrasonic imaging, so it could overcome the limitations of conventional optical imaging techniques [18].
In this paper, we will introduce recent advances of contrast agents, applications, and signal enhancement strategies for PAI. The challenges and outlooks of PAI for deep-tissue biomedical applications are also included.
2. PA contrast agents
2.1 Endogenous PA contrast agents
Some endogenous substances with special light absorption such as hemoglobin (Hb) [19], melanin [20], and lipid [21], have been used as PA contrast agents for biomedical applications (e.g. anatomical, functional and metabolic studies) [22-25]. For example, Hb is an iron-containing metalloprotein in red blood cells that carries oxygen throughout the body. According to the different light absorption spectra of oxyhemoglobin and deoxyhemoglobin, PAI is able to measure the total Hb concentration (C_{Hb}) and oxygen saturation (sO2) [26]. Melanin, a group of natural pigments occurring in the hair, skin, and iris of the eye in most organisms, is a classical endogenous PA contrast agent with strong light absorption in both UV/vis and NIR regions [27]. Besides these agents, genetically encoded probes can also provide good PA contrast by expressing NIR-absorbing proteins, which endows PAI the ability to monitor and image gene expression [28].
2.2 Exogenous PA contrast agents
The use of exogenous PA contrast agents promises PAI with high sensitivity, specificity and signal-to-background ratio over endogenous PA contrast agents. To date, a variety of inorganic/organic PA contrast agents with strong NIR absorbance and high photothermal conversion efficiency have been explored for deep-tissue PAI. According to their structures, they are mainly categorized into inorganic and organic PA contrast agents (Table 1).
2.2.1 Inorganic PA contrast agents
Metallic and semimetallic nanomaterials [29-33], especially gold nanomaterials [34], can achieve high light-to-heat conversion through localized surface plasmon resonance (LSPR), which occurs when the frequency of incident photons matches the frequency of electrons on the surface of these nanomaterials, leading to a strong optical absorption. By controlling the size/morphology of gold nanomaterials, their LSPR absorbance can be engineered into the NIR region. Moreover, gold nanomaterials have the following merits of relative inertness, prominent biocompatibility, and excellent plasmonic property. So far, various gold nanostructures (nanorods [35-38], nanostars [39, 40], nanocages [41, 42], nanoshells [43], nanovesicles [44], nanoflowers [45, 46]) with LSPR peaks in the NIR region have been explored as contrast agents for PAI as well as theranostics. However, some gold nanomaterials with suboptimal photothermal stability could be melted under pulsed laser irradiation and thus lose their NIR absorbance.
Carbon-based nanomaterials, such as carbon nanotubes [47, 48], graphenes [49-55], carbon dots [56, 57], have been widely used in biomedical applications. Carbon nanomaterials usually show broad absorbance from UV to NIR regions. In addition, a large amount of compounds/materials are easy to integrate with carbon-based nanomaterials for PAI and other functions [58-61]. However, one disadvantage of carbon-based nanomaterials is their heterogeneity. For example, carbon nanotubes are a mixture of nanotubes with different lengths and diameters.
Transition metal chalcogenides (TMC)-based nanomaterials, such as copper sulfide (CuS), tungsten sulfide (WS₂), molybdenum sulfide (MoS₂) and so on, due to energy band transitions, are a class of semiconductor nanocrystals with strong NIR absorption [62-69]. Among them, CuS nanoparticles, due to their good photothermal effect and low cost, have been extensively studied for PAI [70-73]. Despite the high photothermal conversion efficiency and good photothermal stability, TMC-based nanomaterials usually contain heavy metal elements, which limits their potential for clinical translation.
Although a lot of inorganic PA contrast agents exhibited excellent PAI performance in animal experiments, the intrinsic non-biodegradability of these agents may lead to relative long retention time and potential long-term toxicity, which limits their clinical translation [74]. Recently, black phosphorus (BP) nanomaterials with broad absorption in both UV and NIR regions, due to its excellent biodegradable property, has been used in PAI [75]. Compared with other inorganic PA contrast agents, BP nanomaterials are composed of element phosphorus without any heavy metals. Furthermore, BP nanomaterials can be fully degraded into biocompatible ions, which can be absorbed, metabolized, and cleared by the body [76].
2.2.2 Organic PA contrast agents
Besides inorganic PA contrast agents, various NIR-absorbing organic agents including porphyrin- and cyanine-based dyes, perylene-diimide (PDI) derivatives and semiconducting polymers, due to their good biodegradability and biocompatibility, have been widely developed for PAI [77-79].
In the past decades, a plethora of small molecule organic dyes have been synthesized for bioimaging [80-82]. Compared with inorganic nanomaterials, these dyes show good biocompatibility and biodegradability. Among them, indocyanine green (ICG), has been approved by the US Food and Drug Administration (FDA) and used in the clinic for a long time [83]. However, most of organic dyes suffer from poor aqueous solubility, poor photothermal stability and short bloodstream circulation half-life. To address these issues, many nanocarriers including micelles, liposomes and proteins, have been employed as the
---
**Table 1. Examples of PA contrast agent explored in PAI.**
| Materials | Types of nanoagents | Advantages (+)/Disadvantages (-) | Ref. |
|----------------------------|-------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------|----------|
| Inorganic Metallic nanomaterials | Au nanorods; Au nanostars; Au nanocages; Au nanoshell; Au nanovesicles; Au nanoflowers; Ag nanoplates; Palladium nanoplates; antimony nanoparticles | (+) tunable physiochemical properties; chemically inert element with reasonable biocompatibility; able to carry cargoes. | [29-34, 37, 38, 40-45] |
| Carbon-based nanomaterials | Carbon nanotubes; Graphenes; Carbon dots | (-) non-biodegradability; suboptimal photothermal stability | [47, 54, 55, 57] |
| TMC | CuS; WS₂; MoS₂; FeS; Bi₂S₆; CuSe; CoSe₂; BiSe | (+) able to carry cargoes; good photothermal stability. | [62-69] |
| Organic Dyes | porphyrin- and cyanine-based dyes, e.g. ICG, IR780, IR825, etc. | (-) non-biodegradability; contain heavy metal elements | [77-79, 83-86] |
| Polymer-based nanomaterials | Polyethylene glycol, Polyamide, Polyethylene glycol; Semiconducting polymers | (+) good biocompatibility/biodegradability. | [87-94] |
http://www.thno.org
vehicles for delivering organic dyes [79, 84-86].
A series of conducting and semiconducting polymers with conjugated structures, such as polypyrrole, polyaniline, polydopamine and so on, have also been developed as PA contrast agents [87-91]. Compared with small molecule organic dyes, polymer-based PA contrast agents showed much better photothermal stability. During the synthesis process of these polymers, many functional moieties, such as drugs (doxorubicin, sorafenib), contrast agents (64Cu, Fe3O4, gadolinium, tantalum oxide), and so on, can be integrated for PAI with other imaging modalities and therapeutics [91-97]. Many in vitro and in vivo studies have demonstrated biocompatibility of these polymer-based nanomaterials [98, 99], but till now their biodegradation behaviors are still not studied in details.
3. Biomedical applications of PAI
Owing to the merits of rich contrast, high resolution and deep penetration depth, PAI has received tremendous attention as a promising noninvasive and nonionizing technique which enables multiscale and multicontrast visualization in a wide range of biomedical applications. In this section, we will summarize the biomedical applications of PAI.
3.1 Deep tumor imaging and therapy monitoring
Recently, a phosphorus phthalocyanine dye with intense absorption at 1000 nm was employed as PA contrast agent for deep PAI beyond 10 cm of chicken breast phantom and through a 5 cm human arm [100]. These results suggested that NIR light-induced PAI is a promising technique for deep-tissue diagnosis. For tumor imaging, the PA contrast agents should be tumor specific either through passive targeting or active targeting. The leaky tumor vasculature combined with poor lymphatic drainage of tumors may allow nano-sized PA contrast agents (< 100 nm) to effectively accumulate in the tumor tissue in a passive targeting manner through the so-called enhanced permeability and retention (EPR) effect [101]. In addition, with the conjugation of targeting ligands (e.g. antibodies, peptides, aptamers), PA contrast agents can actively bind to overexpressed receptors on the tumor cell membranes [102]. In a recent study, Cheng et al. developed perylene-diimide (PDI)-based nanoparticles as highly efficient PA contrast agents for in vivo deep brain tumor imaging [18]. As shown in Figure 2A, PDI molecules were encapsulated into the micelles composed of amphiphilic DSPE-mPEG. The resulting PDI-based nanoparticles with an average particle size of 48 nm were then used for detection of deep orthotopic brain tumor with high PA contrast and sensitivity (Figure 2B). Furthermore, by constructing the 3D PA image, the spatial distribution of PDI-based nanoparticles in tumor was clearly mapped. As shown in Figure 2C, the PDI-based nanoparticles were localized around the injection position of tumor cells.
Besides tumor imaging, PAI also can be used to evaluate the therapeutic efficacy of chemotherapy, photodynamic therapy, radiation therapy and antiangiogenic therapy [103-106]. For example, Hasan et al. used PAI to monitor the progress of photodynamic therapy (PDT) [104]. PDT is a clinical technique which exploits photosensitizers and light excitation. Upon laser irradiation, photosensitizers consume oxygen to generate cytotoxic reactive oxygen species (ROS), thus causing changes in $sO_2$ [107]. By measuring the level of oxyhemoglobin and deoxyhemoglobin with PAI, the $sO_2$ can be monitored and used as a marker for predicting PDT treatment efficacy. On PDT treatment, the drug-light-interval (DLI, defined as the interval between photosensitizer administration and light irradiation) plays an important role. The previous studies demonstrated that PDT with 1 h DLI was more effective when compared to PDT with 3 h DLI [108]. As shown in Figure 3, in the 1 h DLI group, clear hypoxic areas can be observed at both 6 and 24 h post-PDT; while no statistical change in $sO_2$ value in the 3 h DLI group. These results demonstrated that PAI can be used to monitor the response to PDT by measuring $sO_2$ change.
3.2 Metabolic imaging
PAI is also suitable for anatomical, functional and metabolic imaging, by measuring the contents of hemoglobin concentration ($C_{Hb}$), oxygen partial pressure (pO2) and oxygen saturation ($sO_2$) [19, 22, 23]. For example, as shown in Figure 4A, the PA image of $C_{Hb}$ reveals the vascular anatomy down to single capillaries in a living mouse ear [19]. The in vivo real-time oxygen unloading process can also be monitored by measuring the $sO_2$. As shown in Figure 4B, the $sO_2$ map of a capillary loop in a human finger cuticle indicated that most oxygen is unloaded from hemoglobin at the tip of the loop [24]. Based on the measurements of vessel diameter, total $C_{Hb}$, $sO_2$ and blood flow velocity, the metabolic rate of oxygen can be computed, which is helpful to understand metabolism-related pathogenic mechanisms (Figure 4C) [25]. With the help of a NIR dye-labeled 2-deoxyglucose (IRDye800-2DG), PAI can be used for in vivo tumor glucose metabolism imaging [109].
786-O kidney tumors (Figure 4D), the tumor-bearing kidney of mice displayed higher IRDye800-2DG uptake than the health kidney. At the most metabolically active site in the tumor, the tumor-to-normal kidney contrast for IRDye800-2DG uptake was approximately 3.3.
3.3 Imaging tumor microenvironment
As we know, the pH change, enzyme activity, reactive oxygen species (ROS) level, and metal ion concentration, play very important roles in life science. Aberrant behaviors of these parameters are associated with many diseases including cancer, inflammation, cardiovascular diseases, Alzheimer’s disease, Wilson disease, and so on [110-113]. In the past decades, many fluorescent probes have been developed for in vivo detections [114-118]. Encouraged by the absorption changes of some PA contrast agents at different conditions, PAI show great potential to monitor or track the environment changes based on the readout intensity of PA signals [119-126]. Compared to traditional detection methods (e.g. fluorescence imaging), PAI promises in vivo quantitative detection with deep tissue penetration.
Figure 2. (A) Schematic illustration showing the structure of perylene-dimide (PDI) nanoparticles and the in vivo PAI process of brain tumor. (B) The ultrasonic (US), PA and their overlay images of brain coronal sections at 2 d postinjection of PDI nanoparticles. (C) Photographic transverse image (left) and PA sagittal 3D image (right) of tumor-bearing brain. The red dot is the injection point of cells. Reproduced with permission [18]. Copyright 2015 Wiley-VCH.
**Figure 3.** (A) PA images of blood oxygen saturation (StO2) in 1 h drug-light-interval (DLI) group (up) and 3 h DLI group (down) at various time points (pre-PDT, post-PDT, 6 h and 24 h post-PDT). Blue and red signals represent hypoxic and oxygenated regions, respectively. (B) Mean StO2 values in 1 h DLI group and 3 h DLI group. Error bars indicate standard deviation. Reproduced with permission [104]. Copyright 2015 Ivyspring International Publisher.
**Figure 4.** (A) In vivo PA image of relative CHb in a living mouse ear, revealing the vascular anatomy. Reproduced with permission [19]. Copyright 2011 Optical Society of America. (B) In vivo PA images of CHb and sO2 in human finger cuticle. Reproduced with permission [24]. Copyright 2013 Biophysical Society. (C) In vivo PA images of CHb, sO2, and blood flow speed in a nude mouse ear. Reproduced with permission [25]. Copyright 2015 Optical Society of America. (D) In vivo PA images of orthotopically implanted 786-O kidney tumors. Left: anatomical image; middle: total hemoglobin concentration (HbT) image; right: IRDye800-2DG concentration (DG) image; HK: healthy kidney; HM: hypermetabolic. Reproduced with permission [109]. Copyright 2012 SPIE.
3.3.1 pH detection
Since many diseases such as cancer are always accompanied by pH change at the microenvironment [110], the in vivo real-time pH detection may be a potential indicator for early recognition, diagnosis, monitoring, and prognosis of diseases. In order to detect pH in vivo, recently, some organic dyes with pH-responsive NIR absorption shift have been exploited as PA contrast agents for PA pH detection [119-121]. For example, Liu et al. developed a pH-responsive nanoprobe based on albumin-benzo[a]phenoxazine (BPOx)-IR825 complexes (HSA-BPOx-IR825) for in vivo ratiometric PA pH imaging (Figure 5A) [119]. Along with the decrease of pH, the IR825 absorbance peak at 825 nm showed no obvious change; however, the absorbance at 670-680 nm significantly increased through the protonation and intramolecular charge-transfer of BPOx (Figure 5B). Therefore, the PA signal intensity at 680 nm significantly increased while PA signal intensity at 825 nm showed negligible change (Figure 5C). In the pH range of 5.0-7.0, the ratiometric PA signal (PA_{680 nm}/PA_{825 nm}) showed excellent pH dependence, which is suitable for PA imaging of tumor microenvironment (Figure 5D). In vivo pH detection of tumor and muscle was then carried out on a mouse model via the local administration of HSA-BPOx-IR825. As shown in Figure 5E and F, in the tumor region, the PA intensity at 680 nm was much stronger than that at 825 nm, resulting in relatively high PA signal ratio (PA_{680 nm}/PA_{825 nm} = 2.35). In contrast, the PA signal ratio in the muscle region (PA_{680 nm}/PA_{825 nm} = 1.25) was obviously lower than that in the tumor. As shown in Figure 5G, the pH values of tumor and muscle were determined to be ≈6.7 and >7.0 based on the standard calibration curve in Figure 5D, which demonstrated that the HSA-BPOx-IR825 could be used as a PA contrast agent for pH detection in vivo. In another study, as shown in Figure 6A, Pu et al. developed an activatable PA nanoprobe (SON), which consists of a semiconducting oligomer (SO, inert PA matrix) and a boron-dipyromethene dye (pH-BDP, pH indicator) [122]. Under tumor acidic environment, the hydroxyl group of pH-BDP can undergo protonation, thus leading to the absorption changes of SON. With the pH decrease, the absorption peak at 750 nm decreased significantly, while the absorption peak at 680 nm remained unchanged (Figure 6B). Consequently, the...
PA intensity at 750 nm also dropped along with the decrease of pH (Figure 6C and D). Therefore, the ratiometric PA signals (PA\textsubscript{680 nm}/PA\textsubscript{750 nm}) were used to quantify pH (Figure 6E). For \textit{in vivo} pH detection, SON was locally injected into tumor and muscle in living nude mice. Then PA images were recorded under the excitation of 680 and 750 nm lasers, respectively. As shown in Figure 6F, relatively high PA signals at both 680 and 750 nm were detected in the muscle; while in the tumor site, the PA intensity at 750 nm was obviously lower than that at 680 nm. To minimize the interference of endogenous PA contrast agents (e.g. hemoglobin), the PA intensity increments (PA intensity after injection of samples subtracted by the tissue background intensity before injection) at 680 and 750 nm (ΔPA\textsubscript{680 nm} and ΔPA\textsubscript{750 nm}) were used to evaluate the \textit{in vivo} pH. Quantification of the ratiometric PA intensity increments (ΔPA\textsubscript{680 nm}/ΔPA\textsubscript{750 nm}) showed that the ΔPA\textsubscript{680 nm}/ΔPA\textsubscript{750 nm} in the tumor was higher than that in the muscle, suggesting a lower pH value in the tumor region (Figure 6G).
3.3.2 Enzyme detection
Matrix metalloproteinases (MMPs), a family of zinc-dependent endopeptidases that degrade proteins in the extracellular matrix, not only take action in tumor angiogenesis, but also are involved in multiple signaling pathways [127]. To detect MMPs \textit{in vivo}, Liu et al. conjugated NIR-absorbing CuS nanoparticles with a black hole quencher 3 (BHQ3) via a MMP-cleavable peptide linker (Figure 7A) [124]. As shown in Figure 7B, the as-prepared CuS-peptide-BHQ3 (CPQ) showed PA signals at 680
nm (owing to BHQ3 absorbance) and 930 nm (owing to CuS absorbance). In the presence of MMPs, the BHQ3 would be released from the CuS nanoparticles due to the cleavage of peptide linker, leading to the decrease of absorption peak at 630 nm (Figure 7C). In the tumor microenvironment, the small molecule BHQ3 would be rapidly metabolized after cleavage, while the CuS nanoparticles would be retained in the tumor. The ratiometric PA signals ($PA_{680\,\text{nm}}/PA_{930\,\text{nm}}$) were used as an \textit{in vivo} indicator of MMPs activity (Figure 7D). Xing \textit{et al.} developed a MMP2 antibody conjugated gold nanorod (AuNR-Abs), which can specifically target MMP2 to achieve PAI detection of MMP2 in atherosclerotic plaques [123]. The results demonstrated the feasibility of quantitative PA detection of MMP2 in atherosclerotic plaques.
3.3.3 ROS detection
Recently, Rao \textit{et al.} reported an \textit{in vivo} real-time PA ROS detection method [125]. As shown in Figure 8A, IR775S, a cyanine dye that can sense specific ROS-mediated oxidation [128], was loaded into a semiconducting polymer nanoparticle (SPN) composed of poly(cyclopentadithiophene-alt-benzothiadiazole) (SP1). The PA spectrum of obtained ratiometric photoacoustic probe (RSPN) showed three PA peaks at 700, 735 and 820 nm, respectively. In the presence of ONOO$^-$ and ClO$^-$, the PA peak of IR775S at 820 nm decreased significantly due to the ROS-mediated decomposition of IR775S (Figure 8B); while the peak of SP1 at 700 nm had negligible change (Figure 8C). Thus the ROS level can be evaluated according to the change of ratiometric PA signals ($PA_{700\,\text{nm}}/PA_{820\,\text{nm}}$) (Figure 8D). In order to build an \textit{in vivo} model, zymosan, a structural polysaccharide that can simulate the generation of ROS, was injected intramuscularly into the thigh of living mice. After 20 min, RSPN was injected into the same location and the PA signals at both 700 nm and 820 nm were monitored simultaneously. As shown in Figure 8E, for the mice without zymosan treatment, the PA signals at both 700 nm and 820 nm remained nearly unchanged over time. In contrast, the PA signal at 820 nm for the mice with zymosan treatment significantly decreased over time, resulting in an increased $PA_{700\,\text{nm}}/PA_{820\,\text{nm}}$ value (Figure 8F). Thus, RSPN can be used as a PA probe for \textit{in vivo} effective detection of ROS.
\begin{figure}[h]
\centering
\includegraphics[width=\textwidth]{figure7.png}
\caption{(A) Mechanism of CuS-peptide-BHQ3 (CPQ) for MMP detection by using PAI. (B) Absorption spectra of CuS nanoparticles and CPQ. (C) Time-dependent absorbance spectra of CPQ incubated with matrix metalloproteinase-13 (MMP-13). (D) In vivo PA images showing CPQ response to MMP-13. Reproduced with permission [124]. Copyright 2014 Ivyspring International Publisher.}
\end{figure}
Figure 8. (A) Mechanism of ratiometric photoacoustic probe (RSPN) for reactive oxygen species (ROS) detection. (B) The structure change of IR775 in response to ROS. (C) PA spectra of RSPN in the absence and presence of ROS. (D) Ratio of PA amplitude (PA_{700}/PA_{820}) after ROS treatment. (E) PA images of saline (i) and zymosan (ii) treated thigh regions in a murine model of acute oedema. (F) Ratiometric PA signals (700/820) as a function of time post-injection of RSPN. Reproduced with permission [125]. Copyright 2014 Nature.
3.3.4 Metal ion detection
PAI also can be used to detect metal ions, such as copper (Cu^{2+}) and lithium (Li^{+}). For example, Chan et al. developed a PA probe for the chemoselective visualization of Cu^{2+}, which is a crucial metal ion in chronic nervous diseases (e.g. Alzheimer’s disease) [126]. As shown in Figure 9A, the acoustogenic probe for Cu^{2+}-2 (APC-2) contains a 2-picolinic ester sensing module that can specifically respond to Cu^{2+}. In the absence of Cu^{2+}, the PA signal of the probe at 767 nm will be lower than that at 697 nm, leading to a small ratiometric PA_{767 nm}/PA_{697 nm}. In the presence of Cu^{2+}, the sensitive module hydrolyzed, a stronger PA signal at 767 nm and a weaker PA signal at 697 nm will be detected, resulting in a large ratiometric PA_{767 nm}/PA_{697 nm} (Figure 9B). Based on the ratiometric PA_{767 nm}/PA_{697 nm} change, the Cu^{2+} can be detected (Figure 9C). Due to the high selectivity and the deep-tissue penetration of PAI, this PA probe shows great potential for in vivo Cu^{2+} detection. In another study, Clark et al. developed a PA nanosensor for in vivo Li^{+} detection [129]. As shown in Figure 9D, the nanosensor is mainly composed of a lithium selective crown ether ionophore, lithium ionophore VI (L), and a chromoionophore (CH^{+}). Li^{+} is recognized and extracted into the hydrophobic polymer core of the nanosensor, leading to the deprotonation of CH^{+}, which changes the optical properties of the
nanosensor. As a result, the photoacoustic amplitude at 515 nm increase and the photoacoustic amplitude at 660 nm decreases (Figure 9E). For in vivo detection of Li⁺, the nanosensor was injected into the skin of mice and imaged with PA tomography (Figure 9F). Upon intraperitoneal administration of lithium (38 mg/kg), the ratiometric PA₅₁₅ nm/PA₆₆₀ nm of the nanosensor increased by 25% in 14 min (Figure 9G). To date, many organic dyes that can respond to metal ions and realize absorbance change have been developed for fluorescence detection [130-132], which could be utilized for PA metal ions detection.
Figure 9. (A) Mechanism of PAI for Cu²⁺ detection. (B) Normalized absorbance spectra of acoustogenic probe for Cu²⁺-2 (APC-2) and its hydrolyzed product (20). (C) PA signal responses of APC-2 to various metal ions. Reproduced with permission [126]. Copyright 2015 American Chemical Society. (D) The principle of PAI for Li⁺ detection. (E) PA signal responses of nanosensor to Li⁺. (F) In vivo dual wavelength PA images of mice. (G) Ratiometric PA signals (515/660) as a function of time post-injection of Li⁺. Reproduced with permission [129]. Copyright 2015 American Chemical Society.
4. Strategies to enhance PA signals
The PA signal enhancement strategies are critical for improving the sensitivity of both diagnosis and detection. Different strategies for PA signal enhancement, such as synergistic effect, plasmonic coupling effect, self-assembly, photoinduced electron transfer, as well as background signal deduction will be summarized in this section.
4.1 Synergistic effect
A simple way to amplify the PA signal is the combination of two or more PA contrast agents together to achieve higher absorbance. Various nanocomposites have been developed for the amplification of PA signals, such as dye-graphene [133], gold-graphene [134, 135]. The PA signal amplification mechanisms of these nanocomposites mainly include absorption superposition [136], interaction of different agents [137] and reduced heat transfer resistance [135]. For example, Gambhir et al. developed a family of composites based on single-walled carbon nanotubes (SWNTs) and organic dyes for highly sensitive PAI [136, 138]. As shown in Figure 10A, compared with plain SWNTs, the indocyanine green (ICG)-coated SWNTs exhibited a 20-fold higher absorbance due to the absorption superposition of both dyes and SWNTs. Thus the ICG-coated SWNTs showed higher PA contrast and can easily detect ~20 times fewer cancer cells. Lim and
co-workers reported a hybrid PA contrast agent (rGO-AuNR), composed of gold nanorods and reduced graphene oxide (rGO), for sensitive PAI [135]. GO-coated gold nanorods (GO-AuNRs) were prepared through simple electrostatic interactions. After chemical or optical reduction, rGO-AuNRs were successfully prepared (Figure 10B). In this case, rGO can not only enhance NIR light absorption via absorption superposition, but also reduce heat transfer resistance from the gold nanorods to the ambient signal-generating medium. The results demonstrated that the rGO-AuNRs exhibit significantly amplified PA signal intensity when compared to AuNRs and GO-AuNRs (Figure 10C and D). Chen et al. reported a hybrid rGO-loaded gold nanorod vesicle (rGO-AuNRVe) for sequential drug transfer resistance from the gold nanorods to the gold nanorod vesicle can concentrate the electromagnetic radiations, resulting in the absorption superposition, but also reduce heat absorption superposition, but also reduce heat transfer resistance from the gold nanorods to the ambient signal-generating medium. The results demonstrated that the rGO-AuNRs exhibit significantly amplified PA signal intensity when compared to AuNRs and GO-AuNRs (Figure 10C and D). Chen et al. reported a hybrid rGO-AuNRVe for sequential drug release and enhanced photothermal and PA effect (Figure 10E) [137]. Compared with the simple mixture of AuNRVe and rGO, the rGO-AuNRVe showed remarkably amplified PA signal at the same OD808 value (Figure 10F and G). It is believed that the cavity of gold nanorod vesicle can concentrate the electromagnetic radiations, resulting in the absorption efficiency enhancement of the encapsulated rGO [139].
4.2 Plasmonic coupling effect
Plasmonic nanostructures, such as gold nanomaterials, have been actively studied as PA contrast agents. Since the size of the nanomaterials are much smaller than the wavelength of light, the lateral and vertical finite spacing will lead to strong interaction between the neighboring elements. Compared to an individual element, the optical properties of plasmonic nanostructures will be changed substantially [140]. With the distance decrease of adjacent gold nanoparticles, the plasmonic coupling effect would be significantly enhanced, which promises the PA signal amplification. For example, Chen et al. developed a plasmonic biodegradable gold nanovesicle (BGV) by the self-assembly of polymer-tethered gold nanoparticles [141]. As shown in Figure 11A, in the formation process of BGV, the distance (d) between adjacent gold nanoparticles reduced, leading to an ultrastrong plasmonic coupling effect. Plasmonic coupling between gold nanoparticles can generate enhanced electromagnetic field, resulting in enhanced photothermal conversion efficiency and thus amplified PA signals (Figure 11B and C). In a more recent study, they further demonstrated that the chain gold vesicle, self-assembled through a stepwise hierarchical self-assembly process, had stronger plasmonic coupling effect than the non-chain vesicle, and therefore achieved PA signal amplification (Figure 11D and E) [142]. Based on the same principle, the developed plasmonic gold bellflowers (GBFs) which also amplified the PA signals because of the presence of strong plasmonic coupling effect among the multiple-branched petals (Figure 11F) [45]. The PA signal intensities of GBFs were much stronger than that of gold nanorods (GNRs) and gold nanostars (GNSs) at the same OD880 value (Figure 11G). This PA enhancement strategy is suitable for plasmonic metal nanomaterials, such as gold, silver, and so on.
4.3 Self-assembly
A J-aggregate is a type of nanoassembly when dye molecules aggregate under the influence of certain conditions. Compared to dye monomer, J-aggregate shows red-shifted absorption wavelength and higher absorption coefficient [143]. Therefore, the use of a self-assembly nanostructure is another strategy to amplify the PA signals of organic dyes [144-146]. For example, Chen et al. reported an in situ tumor-specific supramolecular self-assembly (ICG doped nanofiber) for enhanced PAI [147]. The mixture of ICG and an alkaline phosphatase (ALP)-responsive peptide forms micelles, which can efficiently self-assemble into nanofibers in the ALP-rich tumor macroenvironment (Figure 12A). As shown in Figure 12C and D, the PA signal of nanofibers in tumor site was much stronger than that of ICG. This excellent performance of nanofibers in PAI was mainly attributed to their unique in situ self-assemble features: (i) the J-aggregate structure of ICG in nanofibers induced an amplified PA signal (Figure 12B), and (ii) the assembly induced enhanced tumor retention. Based on a similar strategy, Wang and co-workers designed an enzyme-responsive small-molecule precursor 1, which had three functional segments: purpurin 18 (P18) as the NIR dye, enzyme-responsive peptide (PLGVRG) as the linker, and RGD peptide as the targeting ligand [144]. As shown in Figure 12E, in the presence of gelatinase, which is overexpressed in tumor, the PLGVRG linkers was cleaved, and then the residues self-assembled into nanofibers due to the enhanced hydrophobicity and reduced steric hindrance. Compared to 2 (P18-PMGMRGRGD without responsiveness) and 3 (P18-PLGVRGRDG without targeting), 1 showed a significantly amplified PA signal, which was attributed to the assembly induced signal amplification and assembly induced retention effect (Figure 12F). This similar strategy was also used for bacterial infection detection [145]. This self-assembly PA enhancement strategy is suitable for organic dyes by forming ordered arrangement, such as J-aggregates.
4.4 Photoinduced electron transfer (PET)
PET is a transfer process of excited state electron, in which excited electron is transferred from donor to acceptor. This strategy is able to enhance the nonradiative heat generation, which can be used for PA signal amplification [148]. Pu et al. developed a fullerene doped semiconducting polymer nanoparticle (SPN-F) that can amplify the PA signal through the intraparticle photoinduced electron transfer [148]. As shown in Figure 13A, the lowest unoccupied molecular orbital (LUMO) and highest occupied molecular orbital (HOMO) are -3.5 and -4.9 eV for semiconducting polymer and -4.3 and -6.5 eV for fullerene, respectively. Therefore, the semiconducting polymer and fullerene play their roles as electron donor and acceptor, respectively, leading to quenched fluorescence and enhanced nonradiative heat generation upon light irradiation, ultimately achieve amplified PAI signal. As shown in Figure 13B and C, with increasing amount of fullerene doped (SPN-Fx, x: the weight percentages of fullerene), the PA intensity obviously increased. The amplified PA signals of SPNs have also been demonstrated in vivo. As shown in Figure 13D and E, compared to SPN-F0 injected mice, the mice injected with SPN-F20 exhibited much higher PA signals in the tumor at all the time points.
4.5 Background signal deduction
Besides the above mentioned strategies for PA signal amplification, the strategy of PA background signal deduction is another way to indirectly enhance the PA signal. For example, Wang et al. reported a reversibly switchable bacterial phytochrome based PA probe for in vivo tumor detection with substantially decreased background signals [149]. As shown in Figure 14A, upon light illumination,
rhodopseudomonas palustris (BphP1) can be reversibly switched between Pfr (ON-state) and Pr (OFF-state). Because the PA background signal of vascular absorbers remains relatively constant, this switchable property of PA probe allows to subtract the two obtained PA images under individual states, and completely remove the background signals (Figure 14B and C), thus achieving enhanced detection sensitivity and improved spatial resolution (Figure 14D). In another study, Gao et al. reported iron oxide and gold-coupled core-shell nanoparticles (MNP-Au) for magnetomotive PAI (mmPAI) [150]. As shown in Figure 14E, a pulsed magnetic field is applied during PA signal acquisition. MNP-Au particles move and create a moving source when the field is on, and then return to their original positions when the field is off. In contrast, the non-magnetic background PA sources do not move during this process. Therefore, coherent motion processing of a PA image sequence can identify sources related to MNP-Au and reject all diffuse and localized background signals. Then both conventional PAI and mmPAI were performed using gold nanorod (Au NR) to mimic strong background tissue signals. As shown in Figure 14F, compared to conventional PAI, the Au NR mimicking a strong background signal is almost completely suppressed in mmPAI.
Figure 12. (A) Illustration showing the in situ formation of nanofibers. (B) PA intensity changes of ICG (1) and nanofiber (5) at different ICG concentrations. PA images (C) and PA intensities (D) of tumor tissues after intravenous injection of samples. Reproduced with permission [147]. Copyright 2015 American Chemical Society. (E) Illustration showing the structure of compound 1 and the in situ formation of nanofibers. (F) PA signal intensities of tumor sites after intravenous injection of samples. Reproduced with permission [144]. Copyright 2015 Wiley-VCH.
5. Conclusion and outlook
PAI, as a promising real-time imaging technique with high spatial resolution and tissue penetration depth, has been widely studied. This paper briefly summarized recent advances in PAI for biomedical applications, such as deep tumor imaging, therapy monitoring, metabolic imaging, pH detection, enzyme detection, ROS detection, metal ions detection, and so on. To improve PAI sensitivity, various exogenous contrast agents, such as metallic nanomaterials, semimetallic nanomaterials, carbon-based nanomaterials, transition metal chalcogenides, organic dyes, polymer-based nanomaterials, have been developed for PAI in recent years. In order to enhance PA signals, different strategies which are based on synergistic effect of two or more PA contrast agents, plasmonic coupling effect of plasmonic metal nanomaterials, self-assembly of organic dyes, photoinduced electron transfer and background signal deduction have been reported. The exploitation of these strategies not only can improve the PA sensitivity, but also can reduce the dose of PA contrast agents.
Although various PA contrast agents have demonstrated their high PA efficiency and unnoticeable toxicity in pre-clinical studies, they have not been approved for clinical use. In the future, the following directions should be taken into account: 1) Fabrication of activatable PA contrast agents that sense specific stimuli of interest followed by PA signal change in a predictable manner, such as a shift in the absorption peak. By analyzing the differential PA signals, the surrounding PA noises can be reduced and then the selectivity and sensitivity of PAI can be improved. 2) Optimization of targeting PA contrast agents to actively target the tumor, improve the tumor accumulation rate, and reduce the reticuloendothelial system (RES) retention by using specific biomarkers (such as peptides, antibodies, and so on). 3)
Development of new strategies that can further enhance the PA signals of PA contrast agents. The current strategies to amplify the PA signals mainly focus on the improvement of NIR adsorption and photothermal effect by exploiting plasmonic coupling effect, self-assembly or photoinduced electron transfer. 4) Development of PA-based multi-modal contrast agents that can provide complementary information by combining information from various imaging modalities and overcome the weaknesses of each imaging modality. 5) Systematic assessment of the stability, toxicity, biocompatibility, biodegradability, immunogenicity and pharmacokinetics of PA contrast agents that would be of great help to accelerate their clinical translation for PAI.
Acknowledgment
This work was supported by the National Science Foundation of China (81401465, 51573096), and the Intramural Research Program (IRP) of the NIBIB, NIH.
Competing Interests
The authors have declared that no competing interest exists.
References
1. Jir Y, Gao X. Plasmonic fluorescent quantum dots. Nat Nanotechnol. 2009; 4: 571-6.
2. Wang Y, Zhou K, Huang G, Hensley C, Huang X, Ma X, et al. A nanoparticle-based strategy for the imaging of a broad range of tumours by nonlinear amplification of microenvironment signals. Nat Mater. 2013; 12: 204-12.
3. Zvezdevskiy P, Gao X. Multifunctional quantum dots for personalized medicine. Nano Today. 2009; 4: 414-28.
4. Nguyen QT, Tsien RY. Fluorescence-guided surgery with live molecular navigation—a new cutting edge. Nat Rev Cancer. 2013; 13: 653-62.
5. Li J, Zhang X, Jiang S. Multicolor core/shell-structured upconversion fluorescence nanoparticles. Adv Mater. 2008; 20: 4765-79.
6. Wang H, Liu Z, Wang S, Dong C, Gong X, Zhao P, et al. MC540 and upconverting nanocrystal loaded polymeric liposome for near-infrared light-triggered photothermodynamic therapy and cell fluorescence imaging. ACS Appl Mater Interfaces. 2014; 6: 3219-25.
7. Park YL, Kim HM, Kim JH, Moon KC, Yoo B, Lee KT, et al. Theranostic probe based on lanthanide-doped nanoparticles for simultaneous in vivo dual-modal optical and photoacoustic imaging. Adv Mater. 2012; 24: 5759-63.
8. Yang W, Gao W, Gong X, Zhang B, Wang S, Chen N, et al. Facile synthesis of Gd-Cu-In-S/ZnS bimodal quantum dots with optimized properties for tumor targeted fluorescence/MR imaging. ACS Appl Mater Interfaces. 2015; 7: 24578-85.
9. Guo W, Sun X, Jacobson O, Yan X, Min K, Srivatsan A, et al. Intrinsically radioactive (ps)Cu(CuBr)2/ZnS quantum dots for PET and optical imaging: improved intrinsic stability and controllable Cerenkov luminescence. ACS Nano. 2015; 9: 1888-95.
10. Guo W, Chen N, Tu Y, Dong C, Zhang B, Hu C, et al. Synthesis of Zn-Cu-In-S/ZnS core/shell quantum dots with inhibited blue-shift photoluminescence and applications for tumor targeted bioimaging. Theranostics. 2015; 5: 90-102.
11. Hong G, Robinson JT, Zhang Y, Diao S, Antaris AL, Wang Q, et al. In vivo fluorescence imaging with Ag3S quantum dots in the second near-infrared region. Angew Chem Int Ed. 2012; 51: 9818-21.
12. Zhang Y, Hong G, Chen C, Li F, Dai H, Wang Q. Ag3S quantum dot: a bright and biocompatible fluorescent nanoprobe in the second near-infrared window. ACS Nano. 2012; 6: 3695-702.
13. Li C, Li F, Zhang Y, Zhang W, Zhang XE, Wang Q. Real-time monitoring of surface chemistry-dependent in vivo behaviors of protein nanocages via incorpating an Au@Ag3S quantum dot. ACS Nano. 2015; 9: 12555-63.
14. Wang LV, Hu S. Photoacoustic tomography: in vivo imaging from organelles to organs. Science. 2012; 335: 1458-62.
15. Kim C, Favazza C, Wang LV. In vivo photoacoustic tomography of chemically: high-resolution functional and molecular optical imaging at new depths. Chem Rev. 2010; 110: 2756-82.
16. Nie L, Chen X. Structural and functional photoacoustic molecular tomography aided by emerging contrast agents. Chem Soc Rev. 2014; 43: 7132-70.
17. Lin J, Wang X, Huang F, Wang Z, Chen S, Niu G, et al. Photosensitizer-loaded gold-loaded vesicles with strong plasmonic coupling effect for imaging-guided photothermal/photodynamic therapy. ACS Nano. 2015; 7: 3320-9.
18. Fan Q, Cheng K, Yang Z, Zhang R, Yang M, Hu X, et al. Protein-dimerized near-infrared photosensitizer-loaded nanoparticles for highly efficient photothermal agents for deep brain tumor imaging in living mice. Adv Mater. 2017; 29: 843-7.
19. Hu S, Maslov K, Wang LV. Second-generation optical-resolution photoacoustic microscopy with improved sensitivity and speed. Opt Lett. 2011; 36: 1334-6.
20. Schwarz M, Buehler A, Aguirre J, Ntziochristos V. Three-dimensional multispectral optoacoustic tomography reveals melanin and blood oxygenation in human skin in vivo. J Biophotonics. 2016; 9: 55-60.
21. Wang B, Su JL, Amiran J, Litovsky SH, Smalling B, Emelianov S. Detection of lipid in atherosclerotic vessels using ultrasound-guided spectroscopic intravascular photoacoustic imaging. Opt Express. 2010; 18: 4889-97.
22. Favazza CP, Jassim G, Cornelius LA, Wang LV. In vivo photoacoustic microscopy of human cutaneous microvasculature and a nevus. J Biomed Opt. 2011; 16: 016015.
23. Ashkenazi S, Huang S-W, Horvath T, Koo Y-EI, Kopelman R. Photothermal probing of fluorohere excited state lifetime with application to oxygen sensing. J Biomed Opt. 2008; 13: 034023.
24. Hu S, Wang LV. Optical-resolution photoacoustic microscopy: auscultation of biological systems at the cellular level. Biophys J. 2013; 105: 841-7.
25. Ning B, Kennedy MJ, Dixon AJ, Sun N, Cao R, Soetikno BT, et al. Simultaneous photoacoustic microscopy of microvascular anatomy, oxygen saturation, and blood flow. Opt Lett. 2015; 40: 910-3.
26. Laufer J, Johnson P, Zhang E, Treeby B, Cox B, Pedley B, et al. In vivo preclinical imaging of tumor vasculature development and leakage. J Biomed Opt. 2008; 13: 034023.
27. Oh JT, Li ML, Zhang HF, Maslov K, Stoica G, Wang LV. Three-dimensional imaging of skin melanoma in vivo by dual-wavelength photoacoustic microscopy. J Biomed Opt. 2006; 11: 034022.
28. Fauth H, Gao X, Niu J, Xiao J, Yao J, Wang LV, Verkhusha VV. Deep-tissue photoacoustic tomography of a genetically encoded near-infrared fluorescent probe. Angew Chem Int Ed. 2012; 51: 1448-51.
56. Huang, P., Lin, J., Wang, X., Wang, Z., Zhang, C., He, M., et al. Light-triggered theranostics based on photosensitizer-conjugated carbon dots for simultaneous enhanced fluorescence imaging and photothermal therapy. Adv. Mater. Sci. 2012; 24: 5041-11.
57. Ge, J., Jia, Q., Liu, W., Guo, L., Liu, Q., Lan, M., et al. Red-emissive carbon dots for fluorescence, photothermal, and thermal theranostics in living mice. Adv. Mater. 2015; 27: 4169-77.
58. Huang, J., Sun, W., Zhu, Q., Burns NA, Khan, SA, Mo, R., et al. Furin-mediated sequential delivery of anticancer cytokine and small-molecule drug shuttled by graphene. Adv Mater. 2015; 27: 1021-8.
59. Tian, B., Wang, C., Zhang, S., Feng, L., Liu, Z. Photothermally enhanced photothermal therapy delivered by nano-graphene oxide. ACS Nano. 2011; 5: 7000-9.
60. Liu, Z., Robinson JT, Sun, X., Dai, H. PEGylated nanographene oxide for delivery of water-insoluble cancer drugs. J Am Chem Soc. 2008; 130: 10876-7.
61. Rong, Y., Pan, X., Srivastava A, Käsemann T, Yao, X., Wang, F. et al. Photosensitizer loaded nano-graphene for multimodality imaging guided photothermal tumor therapy. Theranostics. 2014: 4: 239-39.
62. Cheng, L., Liu, J., Gu, X., Gong, H., Shi, X., Liu, T., et al. PEGylated WS0 nanosheets as a multifunctional theranostic agent for in vivo dual-modal CT/photothermal imaging guided photothermal therapy. Adv. Mater. 2014; 26: 1886-93.
63. Yin, W., Yan, L., Yu, J., Tian, G., Zhou, L., Zheng, X., et al. High-throughput synthesis of single-layer MoS2 nanosheets as a near-infrared photothermal-triggered drug delivery for effective cancer therapy. ACS Nano. 2014, 8: 6922-33.
64. Yang, K., Yang, G., Chen, L., Cheng, L., Wang, L., Ge, C., et al. Fe3O4@ polydopamine core-shell nanocomposites for intracellular mRNA detection and imaging-guided photothermal therapy. ACS Nano. 2014; 8: 8769-83.
65. Wang, X., Hu, C., Wang, S., Sun, C., Liu, J., et al. In vitro and in vivo near-infrared photothermal therapy of cancer using polypropylene organic nanoparticles. Adv Mater. 2012; 24: 5586-92.
66. Pu, K., Mei J, Jokser J, Hong, G., Antalis AL, Chattopadhyay N, et al. Dikephotoporphyrin-oxide-based micromolding polymer nanoparticles for in vivo photothermal imaging. Adv Mater. 2015; 27: 5184-90.
67. Lin, J., Wang, M., Hu, H., Yang, W., Wen, B., Wang, Z., et al. Multimodal-imaging-guided cancer phototherapy by versatile biomimetic theranostics with UV and gamma-irradiation protection. Adv Mater. 2016; 28: 3279-93.
68. Song, X., Liang, C., Gong, H., Chen, Q., Wang, C., Liu, Z. Photonsensitizer-conjugated albumin-polypropylene nanoparticles for imaging-guided in vivo photothermal therapy. Ultra Nano Mater. 2015; 11: 109-21.
81. Tan, X., Luo, S., Wang, D., Su, Y., Cheng, T., Shi, C. A NIR heptamethine dye with intrinsic cancer targeting, imaging and photostimulating properties. Biomaterials. 2012; 33: 2239-50.
87. Lim C-K, Shin J, Lee Y-D, Kim J, Oh KS, Yuk SH, et al. Porphyrin-conjugated nanoscopic polymeric systems for tumor-homing near-infrared phototherapy of cancer. Theranostics. 2012; 2: 871-9.
88. Song, X., Chen Q., Liu Z. Recent advances in the development of organic photothermal nano-agents. Nano Research. 2015; 8: 340-56.
89. Rong, P., Huang, P., Liu, Z., Lin, J., Jin, A., Ma, Y., et al. Protein-based photothermal theranostics for imaging-guided cancer therapy. Nanoscale. 2015; 7: 16330-6.
90. Chen, Q., Liang, C., Wang, C., Liu Z. An imagable and photothermal “alizarin-like” combination nanomaterial for treatment of subcutaneous and metastatic breast tumors. Adv Mater. 2015; 27: 903-10.
91. Huang, P., Rong, P., Jin, A., Yan, X., Zhang MG, Liu, J., et al. Dye-loaded ferritin nanoscale for multimodal imaging and photothermal therapy. Adv Mater. 2015; 27: 1021-8.
92. Yang, J., Choi, J., Bang, D., Kim, E., Lim E-K, Park H., et al. Convertable organic nanoparticles for near-infrared photothermal ablation of cancer cells. Angew Chem Int Ed. 2011; 50: 441-4.
93. Tian, B., Wang, C., Liu, Y., Liu, L., Lin, L., et al. Multimodal imaging – guided photothermal therapy of tumor. ACS Nano. 2014; 8: 12925-35.
94. Sheng, Z., Hu, D., Zheng, M., Zhao, P., Liu H., Gao D., et al. Protein-based photothermal contrast agents. Nat Mater. 2011; 10: 324-32.
95. Rich LJ, Seshadri M. Photoacoustic monitoring of tumor and normal tissue in small animals with whole-body photoacoustic computed tomography. J Biomed Opt. 2012; 17: 076012.
96. Bohndiek SE, Sasportas LS, Machtaler S, Jokerst JV, Hori S, Gambhir SS. Targeting regimens enhances the efficacy of photodynamic therapy. Int J Radiat Oncol Biol Phys. 2005; 61: 1216-26.
97. Zhang R, Fan Q, Yang M, Cheng K, Liu X, Zhang L, et al. Engineering melanin nanoparticles as an efficient drug-delivery system for remotely controlled cancer therapy with synergistic antitumor effect. ACS Nano. 2013; 7: 6782-95.
98. Liu, J., Li, Y., Li, X., Jing, Z., Deng, Z., Yue, X., et al. PEGylated polypropylene nanoparticles conjugating gadolinium chelates for dual-modal MRI/photothermal imaging guided photothermal therapy of cancer. Adv Funct Mater. 2015; 25: 1451-62.
99. Wang, C., Xu, H., Liang, C., Liu, Y., Li, Z., Yang, G., et al. Iron oxide@ polypropylene nanoparticles as a multifunctional drug carrier for remotely controlled cancer therapy with synergistic antitumor effect. ACS Nano. 2013; 7: 6782-95.
100. Chen, B., Pogue BW, Hoopes PJ, Hasan T. Combining vascular and cellular imaging – guided photothermal nano-agents. Nano Research. 2015; 8: 340-54.
101. Peer D, Karp JM, Hong S, Farokhzad OC, Margalit R, Langer R. Nanocarriers as a precision nanomedicine for photothermal therapy. Biomaterials. 2012; 33: 903-10.
102. Peer D, Karp JM, Hong S, Farokhzad OC, Margalit R, Langer R. Nanocarriers as a precision nanomedicine for photothermal therapy. Biomaterials. 2012; 33: 903-10.
103. Bohndiek SE, Sasportas LS, Machtaler S, Jokerst JV, Hori S, Gambhir SS. Targeting regimens enhances the efficacy of photodynamic therapy. Int J Radiat Oncol Biol Phys. 2005; 61: 1216-26.
104. Chen B, Pogue BW, Hoopes PJ, Hasan T. Combining vascular and cellular imaging–guided photothermal therapy. ACS Nano. 2014; 8: 12250-8.
105. Bohndiek SE, Sasportas LS, Machtaler S, Jokerst JV, Hori S, Gambhir SS. Targeting regimens enhances the efficacy of photodynamic therapy. Int J Radiat Oncol Biol Phys. 2005; 61: 1216-26.
106. Peer D, Karp JM, Hong S, Farokhzad OC, Margalit R, Langer R. Nanocarriers as a precision nanomedicine for photothermal therapy. Biomaterials. 2012; 33: 903-10.
107. Peer D, Karp JM, Hong S, Farokhzad OC, Margalit R, Langer R. Nanocarriers as a precision nanomedicine for photothermal therapy. Biomaterials. 2012; 33: 903-10.
108. Chen B, Pogue BW, Hoopes PJ, Hasan T. Combining vascular and cellular imaging–guided photothermal therapy. ACS Nano. 2014; 8: 12250-8.
109. Chatni MR, Xia J, Sohn R, Maslov K, Guo Z, Zhang Y, et al. A NIR heptamethine dye with intrinsic cancer targeting, imaging and photostimulating properties. J Am Chem Soc. 2011; 133: 11376-82.
110. Chew, E., Tan, C., Tan, M., Chua, C., Siew, M., et al. Encapsulating tantalum oxide into polystyrene microparticles as a multifunctional drug carrier for remotely controlled cancer therapy with synergistic antitumor effect. ACS Nano. 2013; 7: 6782-95.
111. Tan, X., Luo, S., Wang, D., Su, Y., Cheng, T., Shi, C. A NIR heptamethine dye with intrinsic cancer targeting, imaging and photostimulating properties. Biomaterials. 2012; 33: 2239-50.
112. Lim C-K, Shin J, Lee Y-D, Kim J, Oh KS, Yuk SH, et al. Porphyrin-conjugated nanoscopic polymeric systems for tumor-homing near-infrared phototherapy of cancer. Theranostics. 2012; 2: 871-9.
113. Song, X., Chen Q., Liu Z. Recent advances in the development of organic photothermal nano-agents. Nano Research. 2015; 8: 340-56.
112. Coakley RJ, Taggart C, McElvany NG, O'Neill SJ. Cytosolic pH and the inflammatory microenvironment modulate human cell death in neutrophils after phagocytosis. Blood. 2002; 100: 3838-91.
113. Bandmann O, Weiss KH, Kaler SG. Wilson's disease and other neurological copper disorders. Lancet Neurol. 2015; 14: 103-13.
114. Lippert AR, De Bittner GCV, Chang CJ. Boronate oxidation as a bioorthogonal reaction approach for studying the chemistry of hydrogen peroxide in living systems. Acc Chem Res. 2011; 44: 793-804.
115. Li D, Shi Y, Tian X, Wang M, Huang B, Li F, et al. Fluorescent probes with dual-mode for rapid detection of SO4 radicals in living cells: ratiometric and two-photon fluorescent sensors. Sens Actuators B. 2016; 233: 1-6.
116. Chen Q, Liu X, Zeng J, Cheng Z, Liu Z. Albumin-based nanoprobe for Cu2+ imaging in alkaline environment and the application for Cu2+ detection in living cells. ACS Nano. 2015; 9: 2711-9.
117. Guha S, Shaw GK, Mitcham TM, Bouchard RR, Smith BD. Croconaine acrylate for acid photothermal heating and ratiometric photoacoustic imaging of acidic pH. Chem Commun. 2016; 52: 120-3.
118. Miao Q, Lyu Y, Ding D, Pu K. Semiconducting oligomer nanoparticles as an activatable photoacoustic probe with amplified brightness for in vivo imaging of pH. Adv Mater. 2016; 28: 3662-8.
119. Qin H, Zhao Y, Zhang J, Pan X, Yang S, Xing D. Inflammation-targeted gold nanorods for intravascular photoacoustic imaging detection of matrix metalloproteinase-2 (MMP-2) in atherosclerotic plaques. Nanomed Nanotechnol. 2016; 12: 1765-74.
120. Chen Q, Liu X, Zeng J, Cheng Z, Liu Z. Albumin-NIR dye self-assembled nanoparticles for photoacoustic imaging and pH-responsive photothermal therapy effective for large biomaterials. Biomedm. 2016; 98: 23-30.
121. Guha S, Shaw GK, Mitcham TM, Bouchard RR, Smith BD. Croconaine acrylate for acid photothermal heating and ratiometric photoacoustic imaging of acidic pH. Chem Commun. 2016; 52: 120-3.
122. Miao Q, Lyu Y, Ding D, Pu K. Semiconducting oligomer nanoparticles as an activatable photoacoustic probe with amplified brightness for in vivo imaging of pH. Adv Mater. 2016; 28: 3662-8.
123. Qin H, Zhao Y, Zhang J, Pan X, Yang S, Xing D. Inflammation-targeted gold nanorods for intravascular photoacoustic imaging detection of matrix metalloproteinase-2 (MMP-2) in atherosclerotic plaques. Nanomed Nanotechnol. 2016; 12: 1765-74.
124. Yang K, Zhu L, Nie L, Sun X, Cheng L, Wu C, et al. Pathology-condition-driven construction of supramolecular nanoassemblies for bacterial infection detection. Adv Mater. 2016; 28: 254-62.
125. Huang P, Gao Y, Lin J, Hu H, Liao H-S, Yan X, et al. Tumor-specific formation of enzyme-instructed supramolecular self-assemblies as cancer theranostics. ACS Nano. 2015; 9: 9517-27.
126. Lui N, Giessen H. Coupling effects in optical metamaterials. Angew Chem Int Ed. 2010; 49: 9836-8.
127. Huang P, Lin J, Li W, Rong P, Wang Z, Wang S, et al. Biodegradable gold nanovesicles with an ultrastable plasmonic effect for photoacoustic imaging and photothermal therapy. Angew Chem Int Ed. 2013; 52: 13958-64.
128. Lyu Y, He J, Yang K, Yi C, Liu Y, Nie L, et al. Folding up of gold nanoparticle strings into plasmmonic vesicles for enhanced photoacoustic imaging. Angew Chem Int Ed. 2015; 127: 16035-8.
129. Wurthner F, Kaiser TE, Saha-Moller CR. J-aggregates: from serendipitous discovery to supramolecular engineering of functional dye materials. Angew Chem Int Ed. 2011; 50: 3570-410.
130. Zhao D, Qi GB, Zhao YX, Qiao SL, Yang C, Wang H. In situ formation of nanofibers from purpurin18-peptide conjugates and the assembly induced retention effect in tumor sites. Adv Mater. 2015; 27; 6125-30.
131. Li LL, Ma HL, Qi GB, Zhang D, Yu F, Hu Z, et al. Photoselective-condition-driven construction of supramolecular nanosystems for bacterial infection detection. Adv Mater. 2016; 28: 254-62.
132. Moon H, Kumar D, Kim H, Sim C, Chang J-H, Kim J-M, et al. Pathological-condition-driven construction of supramolecular nanosystems for bacterial infection detection. Nat Materials. 2016; 8: 12618-25.
133. Bandmann O, Weiss KH, Kaler SG. Wilson's disease and other neurological copper disorders. Lancet Neurol. 2015; 14: 103-13.
134. Lippert AR, De Bittner GCV, Chang CJ. Boronate oxidation as a bioorthogonal reaction approach for studying the chemistry of hydrogen peroxide in living systems. Acc Chem Res. 2011; 44: 793-804.
135. de la Zerda A, Liu Z, Bodapati S, Teed R, Vaithilingam S, Khuri-Yakub BT, et al. Ultrasensitive sensitivity carbon nanotube agents for photoacoustic molecular imaging in living mice. Nano Lett. 2010; 10: 2168-72.
136. de la Zerda A, Liu Z, Bodapati S, Teed R, May SY, Tabakman SM, Liu Z, et al. Family of enhanced photoacoustic imaging agents for high-sensitivity and multiplexing studies in living mice. ACS Nano. 2012; 6: 4694-701.
137. Lyu Y, Fang Y, Miao Q, Zhen X, Ding D, Pu K. Intraparticle molecular orbital reaction approach for studying the chemistry of hydrogen peroxide in living systems. Acc Chem Res. 2011; 44: 793-804.
138. de la Zerda A, Liu Z, Bodapati S, Teed R, May SY, Tabakman SM, Liu Z, et al. Family of enhanced photoacoustic imaging agents for high-sensitivity and multiplexing studies in living mice. ACS Nano. 2012; 6: 4694-701.
139. Zhang P, Guo Y. Surface-enhanced Raman scattering inside metal nanoshells. J Am Chem Soc. 2009; 131: 3808-9. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3806,
1
],
[
3806,
5733,
2
],
[
5733,
11365,
3
],
[
11365,
16440,
4
],
[
16440,
18005,
5
],
[
18005,
19216,
6
],
[
19216,
21630,
7
],
[
21630,
23412,
8
],
[
23412,
26301,
9
],
[
26301,
28319,
10
],
[
28319,
29516,
11
],
[
29516,
30837,
12
],
[
30837,
36498,
13
],
[
36498,
38248,
14
],
[
38248,
40139,
15
],
[
40139,
42054,
16
],
[
42054,
43045,
17
],
[
43045,
48275,
18
],
[
48275,
56102,
19
],
[
56102,
61515,
20
]
]
} |
e0d3d21ce06035549863810066719f1f448d063b | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 21,
"total-fallback-pages": 0,
"total-input-tokens": 66277,
"total-output-tokens": 21042,
"fieldofstudy": [
"Biology"
]
} | (Phospho)proteomic Profiling of Microsatellite Unstable CRC Cells Reveals Alterations in Nuclear Signaling and Cholesterol Metabolism Caused by Frameshift Mutation of NMD Regulator UPF3A
Malwina Michalak, Eva-Maria Katzenmaier, Nina Roeckel, Stefan M. Woerner, Vera Fuchs, Uwe Warnken, Yan P. Yuan, Peer Bork, Gabriele Neu-Yilik, Andreas Kulozik, Magnus von Knebel Doeberitz, Matthias Kloor, Jürgen Kopitz and Johannes Gebert
1 Department of Applied Tumor Biology, Institute of Pathology, Heidelberg University Hospital, Im Neuenheimer Feld 224, 69120 Heidelberg, Germany; [email protected] (M.M.); [email protected] (E.-M.K.); [email protected] (N.R.); [email protected] (V.F.); [email protected] (M.v.K.D.); [email protected] (M.K.); [email protected] (J.K.)
2 Molecular Medicine Partnership Unit, Medical Faculty of the University of Heidelberg and European Molecular Biology Laboratory, 69120 Heidelberg, Germany; [email protected] (S.M.W.); [email protected] (P.B.); [email protected] (G.N.-Y.); [email protected] (A.K.)
3 Department of Pediatric Oncology, Hematology and Immunology, Children’s Hospital, University of Heidelberg, Im Neuenheimer Feld 430, 69120 Heidelberg, Germany
4 Department of Internal Medicine I, Endocrinology and Metabolism, University of Heidelberg, Im Neuenheimer Feld 410, 69120 Heidelberg, Germany
5 Clinical Cooperation Unit Neurooncology, DKFZ (German Cancer Research Center), Im Neuenheimer Feld 280, 69120 Heidelberg, Germany; [email protected]
6 Structural and Computational Biology Unit, European Molecular Biology Laboratory, Meyerhofstraße 1, 69117 Heidelberg, Germany; [email protected]
7 Max-Delbrück-Centre for Molecular Medicine, Robert-Rössle-Straße 10, 13125 Berlin, Germany
8 Clinical Cooperation Unit Applied Tumor Biology, DKFZ (German Cancer Research Center) Heidelberg, Im Neuenheimer Feld 280, 69120 Heidelberg, Germany
* Correspondence: [email protected]; Tel.: +49-6221-564223
† Current address: Roche Diagnostics GmbH, MMQXAD, Nonnenwald 2, 82377 Penzberg, Germany.
Received: 9 June 2020; Accepted: 20 July 2020; Published: 23 July 2020
Abstract: DNA mismatch repair-deficient colorectal cancers (CRCs) accumulate numerous frameshift mutations at repetitive sequences recognized as microsatellite instability (MSI). When coding mononucleotide repeats (cMNRs) are affected, tumors accumulate frameshift mutations and premature termination codons (PTC) potentially leading to truncated proteins. Nonsense-mediated RNA decay (NMD) can degrade PTC-containing transcripts and protect from such faulty proteins. As it also regulates normal transcripts and cellular physiology, we tested whether NMD genes themselves are targets of MSI frameshift mutations. A high frequency of cMNR frameshift mutations in the UPF3A gene was found in MSI CRC cell lines (67.7%), MSI colorectal adenomas (55%) and carcinomas (63%). In normal colonic crypts, UPF3A expression was restricted to single chromogranin A-positive cells. SILAC-based proteomic analysis of KM12 CRC cells revealed UPF3A-dependent down-regulation of several enzymes involved in cholesterol biosynthesis. Furthermore, reconstituted UPF3A expression caused alterations of 85 phosphosites in 52 phosphoproteins. Most of them (38/52, 73%) reside in nuclear phosphoproteins involved in regulation of gene expression and RNA splicing. Since UPF3A mutations can modulate the (phospho)proteomic signature and expression of enzymes involved in
cholesterol metabolism in CRC cells, UPF3A may influence other processes than NMD and loss of UPF3A expression might provide a growth advantage to MSI CRC cells.
**Keywords:** nonsense-mediated RNA decay; coding mononucleotide repeats; DNA mismatch repair; MSI tumorigenesis; UPF3A
1. Introduction
Microsatellite instability (MSI) is a genetic phenotype characterized by the accumulation of insertion/deletion mutations in short repetitive sequences. This is observed in the majority of tumors associated with hereditary nonpolyposis colorectal cancer (HNPCC, Lynch syndrome) but also occurs in approximately 15% of sporadic colorectal carcinomas. At the molecular level, MSI is caused by functional inactivation of the cellular DNA mismatch repair (MMR) system. In HNPCC-associated tumors, MMR deficiency arises due to germline and somatic mutations in one of several DNA MMR-genes (MLH1, MSH2, MSH6, and PMS2) whereas epigenetic silencing of the MLH1 gene accounts for the development of sporadic MSI tumors. In contrast to their microsatellite stable counterparts, colorectal cancers with high level of microsatellite instability (MSI) show distinct clinico-histopathological features including predominant proximal localization, strong lymphocytic infiltration, better prognosis and altered chemoresponsiveness [1–3].
Frameshift mutations in coding region microsatellites (coding mononucleotide repeats, cMNRs) of specific genes are thought to contribute to these clinico-histopathological features. Accordingly, many studies have been performed to predict, identify, and validate target genes potentially involved in MSI tumorigenesis [4–6]. Examples include cMNR-harboring genes encoding proteins of major cellular pathways such as signal transduction (TGFBR2, ACVR2), apoptosis (BAX), and transcription (TCF4). At the transcriptional level these cMNR insertion/deletion mutations lead to shifts in the translational reading frames. These alternative reading frames often contain premature translation termination codons (PTCs) and potentially induce the synthesis of truncated and/or functionally altered proteins. Several cellular control mechanisms suppress the expression of such transcripts, the best characterized of these being the nonsense-mediated mRNA decay pathway (NMD). NMD recognizes mRNAs with PTCs located more than 50–55 base pairs upstream of the last exon-exon junction and initiates their degradation. This task is accomplished by several proteins namely the Upf (Up-frameshift) proteins UPF1, UPF2, UPF3A, and UPF3B, the SMG (Suppressor with Morphological effect on Genitalia) factors SMG1, SMG5, SMG6, SMG7, SMG8, and SMG9 as well as structural and peripheral components of the exon junction complex (Y14, MAGOH, RNPS1, elf4AIII, CASC3, P29) [7,8]. The phosphoinositide 3-kinase (PI3K) such as SMG1 forms together with SMG8 and SMG9 a kinase complex that specifically regulates phosphorylation of UPF1 at several N-and C-terminal S/T-Q-sites. SMG5, SMG6, and SMG7 have been reported to be involved in dephosphorylation of UPF1. However, they also initiate decay of NMD target mRNAs. SMG6 is an endonuclease that cleaves PTC-containing mRNAs near the PTC. After translation termination at PTC the SMG5/7 complex recruits after translation termination at a PTC the cytoplasmic 3′-poly(A)-tail deadenylase CCR4/NOT complex which in turn initiates decapping and enables both 5′ and 3′ exonucleolytic decay of the RNA body [9].
Among these NMD pathway components UPF3 stands out because it exists in two paralogs, UPF3A and UPF3B that differ in their ability to regulate NMD and translational efficiency [10]. UPF3B is a bona fide NMD activator that has been reported to regulate translation termination and modulate ribosome recycling [11]. In contrast, its sister paralog UPF3A is a powerful NMD repressor thereby acting as a regulator of gene expression [12].
In addition to acting as a quality control system that rids cells of aberrant mRNAs with crippled protein coding potential, numerous studies revealed that NMD constitutes an important post-transcriptional layer of gene expression control involved in the regulation of many different biological pathways [13]. In particular, NMD targets at least 10% of normal mammalian mRNAs to modulate appropriate cellular responses—adaptation, differentiation, or death—to environmental changes [9].
Both functions of NMD, to eliminate aberrant mRNAs and to modulate gene expression have been shown to affect the outcome of several diseases. For example, NMD can prevent the expression of potentially deleterious proteins that might confer a dominant negative phenotype, as was shown for β-thalassemia [14]. Germline mutations in the NMD factor gene UPF3B that abrogate normal UPF3B function cause various forms of intellectual disability and other mental disorders [15–18]. Furthermore, considering the accumulation of cMNR-mutated mRNAs in MSI tumor cells, it is not surprising that NMD plays a significant role in modulation of this cancer phenotype [19,20]. By suppressing the expression of mutated proteins NMD is believed to aid tumor cells to escape the immune system [20]. On the other hand, mutations that disrupt NMD functions are commonly observed in pancreatic cancer [21]. Furthermore, reduction in NMD activity can affect the clinical outcome of hepatocellular carcinomas [22]. It is, therefore, reasonable to assume that cMNR mutations in NMD effector genes might be under positive or negative selection pressure, depending on their effect on NMD efficiency [23].
To test this hypothesis, we aimed to identify coding region microsatellites in NMD effector genes that might be targets of frameshift mutations and explore their impact on the molecular phenotype of MSI tumor cells as defined by their proteomic and phosphoproteomic profile.
2. Results
2.1. Identification of NMD-Related Genes Harboring cMNRs
We searched our human cMNR database for potential MSI target genes encoding proteins of the NMD pathway. Among 14 NMD-associated genes with coding region microsatellites we excluded eight genes (CASC3, EIF4A3, MAGOH, PYM, RNPS1, SMG6, UPF1, Y14) from further analyses because they contained repeats with a maximum length of six mononucleotides that are known to exhibit low mutation frequencies [24]. In contrast, coding microsatellites of increased repeat length (≥7 mononucleotides) were detected in the remaining six NMD-associated genes SMG1 (T7, A7), SMG5 (C7), SMG7 (A9), UPF2 (A7), UPF3B (T7) and UPF3A (A7, A9). Although the UPF3A gene contained two cMNRs only the A9 repeat was considered for subsequent analyses because it was retained in all UPF3A splice variants. Coding repeat mutations are only of functional relevance if affected genes are expressed in the target tissue. Our search for expression data in several databases revealed that these six cMNR-harboring candidate genes are expressed in colon epithelial cells [25,26].
2.2. cMNR Frameshift Mutations in NMD-Associated Genes
We next investigated the cMNR frameshift mutation frequency of these six NMD-associated genes in MSI colorectal cancer cell lines (n = 30). PCR fragment length analysis revealed cMNR mutations in SMG1, SMG7, and UPF3A. Particularly low frequencies of heterozygous mutations occurred in the first A7 repeat of SMG1 (11%; 2/18) and in the A9 repeat of SMG7 (15%; 3/19) (Table 1). These mutation frequencies are well within the expected range for repeats of this type and length. In contrast, a high frequency of 1- or 2-bp deletions was detected in the A9 repeat in the UPF3A gene (67.7%; 21/31). The majority of these somatic mutations did affect only one allele, whereas several MSI cell lines also showed biallelic mutations in the UPF3A coding repeat (19%; 6/31; Supplementary Table S1). However, with the exception of a single normal colon mucosa specimen (1/101) cMNR mutations were detected neither in several control samples including microsatellite stable (MSS) CRC cell lines (0/20) nor in the peripheral blood of healthy donors (0/68).
These results suggest a positive selection for UPF3A mutations at least in cultured cell lines. Therefore, we extended our analysis to primary MSI tumors associated with Lynch Syndrome including colorectal adenomas (n = 27) and carcinomas (n = 101) as well as cancers of the endometrium (n = 13), stomach (n = 13) and upper urinary tract (n = 11). Although UPF3A cMNR frameshift mutations were observed in each MSI tumor entity examined, frequencies varied considerably among different tissues. In particular, a high frequency of cMNR frameshift mutations occurred in MSI colorectal adenomas (55%, 15/27) and carcinomas (61%, 62/101) whereas much lower frequencies were found in MSI stomach (5/13, 38.5%), endometrial (3/13, 23.1%), and urothelial (1/11, 9.1%) tumors (Table 2). The high UPF3A mutation frequency in MSI colorectal carcinomas was independently confirmed in an additional set of 78 CRCs (Table 2, validation set). The UPF3A mutation status of these tumors did not correlate with clinicopathological features. Overall, UPF3A represents the only NMD-related gene genetically altered in more than 50% of MSI colorectal cancer cell lines and MSI primary colorectal tumors. These mutation data in conjunction with our statistical model [27] are highly predictive for positive selection of UPF3A mutations. In addition, biallelic mutations in several MSI CRC cell lines and the occurrence of mutations in pre-neoplastic lesions provide strong evidence for a likely contribution of UPF3A mutations to MSI tumorigenesis.
Table 1. Frameshift mutation frequencies in NMD-related genes.
| Hugo ID | Role in NMD | cMNR | Position | Mutated |
|---------|-------------|------|----------|---------|
| SMG1 | UPF1 kinase | T7 | 1339 | 2/18 (11%) |
| | | T7 | 9018 | 0/19 |
| | | A7 | 10694 | 0/19 |
| SMG5 | Promotes UPF1 dephosphorylation | C7 | 12 | 0/19 |
| SMG7 | Promotes UPF1 dephosphorylation | A9 | 2273 | 3/19 (15%) |
| UPF2 | Binds UPF3A/B, recruits UPF1 | A7 | 355 | 0/16 |
| UPF3A | Binds EJC, recruits UPF2 | A7 | 486 | 0/19 |
| UPF3B | Binds EJC, recruits UPF2 | A9 | 790 | 18/23 (78%) |
Position of the first base in the repeat within the coding sequence, based on the ENSEMBL transcripts ENST00000276201 (SMG1), OTTHUMT00000046308 (SMG5), ENST00000367537 (SMG7), ENST00000262803 (UPF2), ENST00000375299 (UPF3A) and ENST00000276201 (UPF3B). UPF3A splice variant represented by ENSEMBL transcript ENST00000351487 lacks the A7 repeat while the A9 repeat starts at coding base pair 691.
Table 2. Frameshift mutation frequencies in different tumor entities for the A9-repeat of UPF3A.
| Tissue | Colorectal | Endometrial | Gastric | Urothelial |
|--------------|------------|-------------|---------|------------|
| Stage | Cancer | Adenomas | Cell Lines | Cancer | Cancer | Cancer |
| Mut | 62/101 (61%) | 15/27 (55%) | 21/31 (67%) | 3/13 (23%) | 5/13 (38%) | 1/11 (9%) |
| Biall. Mut | 51/78 (65%) | – | – | 6/31 (19%) | – | – |
Test set; Validation set.
2.3. UPF3A Protein Expression in MSI Colorectal Cancer Cell Lines
To determine if the mutational status affects UPF3A expression, we performed Western Blot analyses of protein extracts from 30 MSI colorectal cancer cell lines that differ in their UPF3A allele status. As a control we used the colorectal cancer cell line SW948 that has an intact DNA mismatch repair system and lacks the high load of frameshift mutations and truncated proteins. In SW948 cells, protein bands in the expected size range of 56 kDa to 54 kDa were detected by a polyclonal antibody directed against the C-terminus of UPF3A (Figure 1). UPF3A could not be detected in cell lysates of all MSI colorectal cancer cell lines with homozygous A9 cMNR frameshift mutations (UPF3A−/−; 6/30, 20%). Heterogeneous protein expression patterns were observed in MSI cell lines that were either homozygous wild type (UPF3A+/+; 10/30, 33%) or heterozygous mutant for the UPF3A
A9 coding repeat (UPF3A⁺/−; 14/30, 47%). Despite the presence of at least one wild type allele, high variation in UPF3A protein expression was observed including complete loss of expression (5/14, 36%) as well as expression of proteins of different sizes detected by the UPF3A antibody (9/14, 64%). We also analyzed UPF3B expression in these cell lines because UPF3A and UPF3B have been proposed to share functional similarity and might regulate each other’s activity [17]. However, no clear correlation between UPF3A mutation status and UPF3B expression could be observed (Supplementary Figure S1).
**Figure 1.** Western Blot analysis of UPF3A protein expression in colorectal cancer cell lines. SW948 cells (MSS) served as UPF3A expression control while ß-actin was used as loading control. The status of wildtype (+) and mutated (−) cMNR alleles for each cell line is indicated.
2.4. Expression of UPF3A in Normal Colon Epithelium and Colorectal Tumors
Information about UPF3A expression in normal colon mucosa is very limited. We therefore performed UPF3A immunohistochemistry on formalin-fixed paraffin-embedded tissue specimens of normal colon epithelium and tumor tissue. Interestingly, strong immunohistochemical staining was found in single cells within the epidermal walls of colon crypts while the surrounding tissue showed very little or no staining (Figure 2A). In contrast, tumor tissues that carry UPF3A mutations have lost UPF3A protein expression (Figure 2B). Since the UPF3A staining pattern in normal colonic crypt epithelial cells was reminiscent of the staining pattern known for enteroendocrine cells we performed double immunofluorescence analysis for comparing expression of UPF3A with that of chromogranin A (CHGA), a marker for these terminally differentiated cells (Figure 2C; Supplementary Figure S2). In normal colonic crypts a punctuate pattern of double labeled immunofluorescent cells was observed. In some cells only one of both proteins was expressed. At the subcellular level both proteins showed a slightly different distribution. While UPF3A was expressed basally and thus directed toward the epidermis, CHGA expression overlaps with this site but also extends toward the luminal side of the crypt epithelium. In most labeled cells UPF3A and CHGA expression was predominantly observed in the cytoplasm. These results suggest that UPF3A is expressed in some subset of enteroendocrine cells.
2.5. Characterization of KM12-UPF3A Model Cell Line
Because NMD regulates normal and pathological cellular physiology, perturbation of NMD, for example due to loss of expression of an NMD effector would be expected to alter expression profiles of affected cells. To analyze UPF3A-specific effects, we generated a genetically modified MSI colon cancer cell line KM12-UPF3A (UPF3A−) that confers doxycycline (dox)-regulated wild type (WT) UPF3A expression in an isogenic background. RT-PCR analysis confirmed expression of the endogenous UPF3A mutant (A8 repeat) as well as the transgenic UPF3A WT transcript (A9 repeat) in these cells. At the protein level, KM12-UPF3A cells showed dox-inducible expression of WT UPF3A recognized by the UPF3A antibody on Western blots upon induction and time-course analysis (Figure 3). However, a truncated UPF3A protein predicted to be encoded by the endogenous frameshift mutant UPF3A transcript could not be detected, although the antibody used for immunoblotting recognizes an N-terminal UPF3A epitope. Doxycycline itself did not cause any unspecific effects because KM12-Tet-on control cells failed to express any UPF3A protein, both in the absence or presence of doxycycline. Thus, our dox-inducible KM12-UPF3A model system represents a versatile tool to identify MSI-specific molecular and cellular alterations associated with UPF3A expression in an isogenic background.
Figure 2. (A,B) Immunohistochemical staining of UPF3A in normal colon crypts (A) and MSI colorectal tumor (UPF3A mutated) (B). Positively stained cells are indicated by arrows. (C) Double immunofluorescence staining of chromogranin A (CHGA; green), a marker of endocrine cells as well as UPF3A (red) in normal colon crypts. Cell nuclei were counterstained with 4′,6-diamidino-2-phenylindole (DAPI). Overlay of the fluorescence signals confirms colocalization of both proteins (yellow and orange) in individual cells.
Figure 3. Time course of doxycycline-induced UPF3A protein expression. KM12-UPF3A cells were grown in the absence (−) or presence of doxycycline (0.5 µg/mL) for the indicated times and total cellular protein (50 µg) was analyzed by Western blotting. Human β-actin was used as loading control.
2.6. UPF3A Induces Proteomic and Phosphoproteomic Changes
To investigate the consequences of UPF3A frameshift mutations on the proteomic landscape of CRC cells, UPF3A-proficient (+dox, pUPF3A), and UPF3A-deficient (dox, dUPF3A), cells were compared by analyzing global protein expression as well as associated phosphorylation changes. Combining proteomic and phosphoproteomic analyses brought the advantage of a deeper understanding of the cell phenotype. Consequently, alterations in protein expression were taken into account in our protein phosphorylation analysis, another advantage of the proposed strategy. SILAC-based proteomic and phosphoproteomic analyses were performed in our KM12-UPF3A model cell line (Figure 4). A total of 1298 proteins were identified and quantified in at least two biological replicates with at least two unique peptides (Supplementary Table S2; Supplementary Figure S3A), including a subset of 35 proteins that were regulated (fold change > 1.5) in a UPF3A-dependent manner (Table 3; Table 4).
**Table 3.** Summary of the proteome and phosphoproteome profiling upon UPF3A expression.
| Proteins | Total # |
|----------|---------|
| Identified and quantified | 1298 |
| Regulated * | 35 |
**Phosphosites**
| Identified and quantified | 2248 |
| Unknown $^\S$ | 27 |
| Corrected for protein expression | 779 |
| Regulated sites (phosphoproteins) | 85 (52) |
| Known regulation $^\S$ | 7 |
* >1.5-fold in at least two biological replicates; $^\S$ According to PhosphoSitePlus.
**Table 4.** UPF3A-induced protein regulation (>1.5 fold in at least two biological replicates).
| Protein Name | Gene Name | Ratio pUPF3A/dUPF3A |
|--------------|-----------|---------------------|
| GTPase-activating protein and VPS9 domain-containing protein 1 | GAPVDI | 2.19 |
| DNA (cytosine-5)-methyltransferase 1 | DNMT1 | 1.88 |
| Tumor protein D52 | TPD52 | 1.70 |
| COP9 signalosome complex subunit 3 | COP53 | 1.65 |
| U1 small nuclear ribonucleoprotein 70 kDa | SNRNP70 | 1.64 |
| DNA replication licensing factor MCM7 | MCM7 | 1.49 |
| Nucleolar protein 56 | NOP56 | −1.51 |
When these regulated proteins were mapped and visualized by STRING database in an interaction network (Figure 5) and compared with Gene Ontology enrichment analysis, two clusters became apparent: one cluster showed significant enrichment of proteins involved in cholesterol metabolism whereas another cluster centered around proteins with oxidoreductase activity.
In parallel, 2248 phosphorylation sites were identified with high probability (>0.75) and quantified in at least two biological replicates. To each phosphosite corresponding information from PhosphoSite Plus database was aligned which revealed 27 yet unknown phosphosites (Supplementary Table S3; Supplementary Figure S3B). To avoid false identification of phosphorylation changes due to alterations in whole protein expression, 779 phosphosites have been matched and normalized to their protein expression levels. Among them, 85 phosphosites, located on 52 phosphoproteins, were found to be regulated (>1.5 fold change) in at least two biological replicates (Table 3). The top 10 up- and down-regulated phosphosites are listed in Table 5 and detailed information about each analyzed phosphosite is available in Supplementary Table S4. Interestingly, some phosphosites have been identified on both mono and multiply phosphorylated peptides, sometimes with contrary regulation. One example is the CTNND1 protein. In UPF3A-proficient cells, its S349 site is hypophosphorylated together with two other phosphosites (S346 and S352), but remains hyperphosphorylated when the two adjacent phosphosites have lost their phosphorylation status (Supplementary Figure S4). Similar effects were observed for phosphosites in proteins LMNB1 (S23) and SFSR9 (S213 and S216).
When regulated phosphoproteins were analyzed by Gene Ontology Enrichment and interaction network analysis, a significant proportion of regulated phosphoproteins (75%) comprised nuclear proteins specifically involved in RNA splicing and positive regulation of gene expression (Figure 6).
**Figure 5.** Proteomic data analysis: (A) Interaction network of regulated proteins (generated by STRING v11.0). The connecting lines between protein nodes represent protein-protein interactions and the thickness of the edge indicates interaction score (minimum interaction score = 0.4). Coloring of proteins is based on further enrichment analysis. Proteins marked in red are involved in regulation of the cholesterol biosynthetic process. Proteins associated with oxidoreductase activity are marked in blue. (B,C) Enrichment analysis of regulated proteins (performed in STRING v11.0). Graphs are showing the most enriched Gene Ontology (GO) Biological Processes (B) and Molecular Functions (C) among the regulated proteins with observed protein count in each category and calculated p values corrected for multiple testing (Benjamini and Hochberg).
**Table 5.** Top 10 up-regulated and top 10 down-regulated phosphopeptides upon UPF3A expression.
| Protein Name | Gene Name | Phosphosite | Ratio pUPF3A/dUPF3A |
|-------------------------------------|-----------|-------------|---------------------|
| Lamin-B1 * | LMNB1 | T20, S23 | 2.39 |
| Nucleophosmin | NPM1 | S254 | 2.35 |
| Catenin delta-1 § | CTNND1 | S349 | 2.15 |
| Cell division cycle 5-like protein | CDC5L | S303 | 2.05 |
| Multifunctional protein ADE2; | | | |
| Phosphoribosylaminomimidazole-succinocarboxamide synthase | | | |
| Neuroblast differentiation-associated protein AHNAK | AHNAK | S5763 | 1.96 |
| General transcription factor IIF subunit 1 | GTF2F1 | S224 | 1.95 |
| Importin subunit alpha-3 | KPNA4 | S60 | 1.88 |
| MARCKS-related protein | MARCKSL1 | S104 | 1.84 |
| Sister chromatid cohesion protein PDS5 homolog A | PDS5A | S1206 | 1.77 |
| Serine/arginine repetitive matrix protein 2 | SRRM2 | S377 | −2.27 |
| Tumor protein D52 | TPDS2 | S223 | −2.30 |
| Tumor protein D54 | TPDS2L2 | S19 | −2.71 |
| Serine/arginine repetitive matrix protein 2 * | SRRM2 | S876 | −2.90 |
| RNA-binding protein 14 | RBM14 | S618 | −3.20 |
| Serine/arginine-rich splicing factor 9 | SRSF9 | S211, S216 | −3.41 |
| Nuclear mitotic apparatus protein 1 | NUMA1 | S1969 | −3.43 |
| Catenin delta-1 | CTNND1 | S230 | −3.58 |
| Catenin delta-1 § | CTNND1 | S346, S349, S352 | −3.74 |
| Ras-related protein Rab-7a | RAB7A | S72 | −5.34 |
* Mean of ratios from three biological replicates; § Phosphosite (S349) identified on both mono and multiple phosphorylated peptides with different expression levels.
3. Discussion
In microsatellite unstable cells, frameshift mutations in coding microsatellites occur at high frequency often introducing premature translation termination codons into the affected transcripts which thus become potential targets for NMD. As genes encoding NMD factors may contain themselves coding mononucleotide repeats (cMNRs) and as NMD has been shown to be an important modulator of MSI tumorigenesis [19,20], we sought to address the potential link between MSI and NMD by analyzing cMNR mutation rates in NMD factor genes. Focusing only on cMNRs of increased length (>7 nt) we found overall very low mutation frequencies in NMD factor genes with the notable exception of the A9 repeat in UPF3A. This indicates that mutations in the overall NMD pathway are not positively selected for, possibly emphasizing the role this pathway plays in preventing an immune reaction against the emerging tumor. This appears to be especially likely as MSI tumors frequently show strong...
lymphocyte infiltration and a positive selection for mutations in β2-microglobulin thereby facilitating immune evasion [28].
UPF3A is one of two human paralogs with homology to the *Saccharomyces cerevisiae* Upf3 protein [29,30] and it has been reported as a potent inhibitor of NMD that can stabilize several substrate mRNAs [10,30,31]. Loss of UPF3A will hence destabilize these transcripts with biological implications if encoded proteins are involved in growth, differentiation, or apoptosis [17]. Therefore, the observed high frequency of *UPF3A* mutations we found in MSI colorectal cancer might actually result from a selective pressure to enhance NMD efficiency. However, preliminary analysis of physiological NMD targets in dUPF3A and pUPF3A MSI CRC cell lines did not reveal such UPF3A-specific effects on NMD efficiency. Although this needs to be corroborated by more detailed experiments it rather suggests that UPF3A may play a role in other processes than NMD.
It is well known that genetic alterations of NMD factors appear to be associated with neuro-developmental disorders. For example, single allele *UFP2* deletions and mutations of *UPF3B* have been identified in patients with intellectual disability (ID) [32]. Similarly, other NMD genes such as *UPF3A*, *SMG6*, *EIF4A3*, and *RNPS1* are frequently deleted and/or duplicated in these patients [33]. It has also been reported that UPF3A is decreased in cells of patients with atrophic subcortical lateral sclerosis [34]. How altered UPF3A function might contribute to such diverse disease pathologies such as neuronal defects and colon cancer is difficult to reconcile. However, in this context, it is interesting to note that a recent transgenic mouse study uncovered direct physical contact between enteroendocrine cells and neurons innervating the small intestine and colon [35]. Since our data revealed UPF3A expression predominantly in enteroendocrine cells of normal human mucosa, one might speculate that impaired UPF3A function and affected NMD target transcripts disrupt this gut-brain chemosensory circuit leading to abnormal enteroendocrine cell physiology and failure to respond to changes in gut microbiota.
We show for the first time that *UPF3A* frameshift mutations are frequent in MSI colorectal cancer cell lines, primary cancers and adenomas but also occur in MMR-deficient tumors of other organs, albeit at lower frequency. In MSI CRC cell lines, loss of UPF3A expression was associated with biallelic A9 cMNR mutations. However, we did not detect any biallelic *UPF3A* mutations in primary tumor tissues. Despite enrichment of tumor cells by microdissection we cannot exclude that residual inflammatory and connective tissue cells may have confounded mutation analysis. In samples that were found to be heterozygous for frameshift mutations in the A9 coding repeat, additional inactivating point mutations in the remaining *UPF3A* coding sequence also cannot be excluded. Alternatively, haploinsufficiency might impair normal UPF3A function. In fact, even a 50% decrease of *Upf3a* expression in heterozygous mice (*Upf3a*+/−) has been reported to be sufficient to cause alterations in NMD substrate levels and defects in spermatocytes [12]. These authors also observed that *Upf3a*−/− homozygosity causes embryonic lethality. How complete and/or partial UPF3A loss might contribute to MSI tumorigenesis remains unresolved as long as cell- and tissue-specific normal and aberrant UPF3A expression patterns and their impact on NMD substrates have not been completely elucidated. Our immunohistochemical staining data in normal colon epithelial cells show that UPF3A is not ubiquitously expressed. Instead, overlap of UPF3A expression with staining of the enteroendocrine cell (EEC) marker chromogranin A suggests that in the colon epithelium its expression is restricted to this cell type. Which specific EEC subtype actually is affected by partial or complete loss of UPF3A warrants further investigation.
Apart from these potential target cells of UPF3A expression in the colon epithelium, our proteomics data also uncovered a potential link between UPF3A expression and the cholesterol metabolic pathway. In particular, reconstituted UPF3A expression in our MMR-deficient CRC model cell line was associated with down-regulation of proteins involved in cholesterol biosynthesis and hence UPF3A-deficiency leads to its up-regulation. Cholesterol is an essential building block of cell membranes by modulating membrane fluidity and functions, such as transmembrane signal transduction and interaction with the extracellular matrix. Cellular cholesterogenesis correlates with cell proliferation rates, while suppression
of cholesterol biosynthesis inhibits cell growth [36–38]. Especially fast growing tumor cells require increased amounts of cholesterol as essential components for membrane buildup, as well as for synthesis of signaling molecules. Accordingly, up-regulation of enzymes for cholesterol biosynthesis causing an increase of cellular cholesterol production is essential for tumorigenesis and tumor progression [39–42]. Furthermore, analysis of the expression of cholesterol synthesis genes in diverse cancers using the Cancer Genome Atlas (TCGA) also indicated deregulation of cholesterol homeostasis as an important factor in cancer development [43]. Key enzymes of cholesterol biosynthesis are considered to modulate lipid raft structures thereby enhancing raft-associated prometastatic signaling [44]. Several studies specifically elucidated a role of the cholesterol biosynthetic pathway in colon tumorigenesis and progression [45–49] and inhibition of cholesterol synthesis has been suggested for CRC treatment [50,51]. In summary, metabolic reprogramming toward increased cholesterol synthesis appears to be functional in CRC. We for the first time obtained evidence that UPF3A might be involved in this reprogramming process. In this context, loss of UPF3A expression might provide a growth advantage for dMMR CRC tumor cells. However, further relevant experiments should be provided to support this hypothesis.
Our phosphoproteomics data also highlight the impact of UPF3A on the phosphorylation status of nuclear-related proteins. For example, phosphorylation of several Ser residues in the Armadillo family protein CTNND1/p120 were affected by UPF3A expression even in opposing directions. It has been reported that CTNND1 is highly phosphorylated and shuttles between the cytoplasm and nucleus where it can interact with transcriptional activators (β-catenin) and repressors (Kaiso) thereby regulating gene expression [52]. Likewise, CTNND1 is known to be required for nuclear translocation of E-cadherin which in turn regulates β-catenin activity, thereby promoting increased expression of downstream genes and accelerating colorectal tumor growth and migration [53]. To understand how UPF3A activity contributes to the biology of normal and cancerous colon cells more detailed molecular studies are warranted.
Overall, combining computational and mutational screening with inducible gene expression and phospho/proteomic analyses identified UPF3A as a frequent target of frameshift mutations and an important modulator of expression and phosphorylation of proteins involved in cholesterol biosynthesis, redox reactions, and splicing. As a versatile and general approach, it can be applied to any gene and protein of interest.
4. Materials and Methods
4.1. Database Analyses
Genes involved in NMD were chosen based on the current literature. Candidate genes containing cMNR sequences of at least seven nucleotides in length were identified from seltarbase.org [27]. Existing expression data of candidate genes in colon tissue were obtained from proteinatlas.org [54], biogps.org [25], GeneSapiens system [55] and the genecards.org [26].
4.2. Cancer Cell Lines and Human Tissue
Colorectal cancer cell lines were grown under standard conditions in DMEM (Dulbecco’s Modified Eagle Medium) medium (Invitrogen, Carlsbad, CA, USA) supplemented with 10% FCS in the presence of 100 µg/mL penicillin and 100 µg/mL streptomycin (PAA Laboratories GmbH, Cölbe, Germany). Most cell lines and features have been described previously [4,56]. Additional cell lines were provided by: German Cancer Research Center Tumorbank (DLD 1, HCT 15; Heidelberg, Germany), Ludwig Cancer Research Institute (LIM 1215, LIM 405, LIM 2412, LIM 2537; Melbourne, Australia), M. Schwab (HDC9, HDC 108, HDC 135, HDC143; German Cancer Research Center, Heidelberg, Germany), W. Bodmer (LS411 and GP2D; University of Oxford, UK), S. Michel (K073; Heidelberg University, Germany) and M. Linnebacher (HROC24; University of Rostock, Germany).
Cell growth was determined by the CellTiter 96® AQueous One Solution Cell Proliferation Assay (Promega, Walldorf, Germany) or by cell counting in a Neubauer hemocytometer. Human tissues
were obtained from the local tissue bank established within the German Collaborative Group on HNPCC. Two sets of MSI CRCs were analyzed: a test set \((n = 101)\) and a validation set \((n = 78)\). Clinicopathological features of the validation set are indicated in Supplementary Table S5. Informed consent was obtained from all patients and the study protocol was approved by the local Ethics Committee (Nr. 220/2002, 18 February 2011). For all tissue samples MSI status has been determined previously based on the National Cancer Institute/ICGHNPC reference marker panel [57] and CAT25 as an additional mononucleotide marker [58]. MSI is defined by instability in at least 30% of tested markers.
4.3. Nucleic Acid Isolation, Analysis, and RT-PCR
Genomic DNA was isolated using the DNeasy Tissue Kit (Qiagen, Hilden, Germany). RNA was isolated using the RNeasy Mini Kit (Qiagen). cDNA synthesis was performed with Superscript II Reverse Transcriptase (Life Technologies, Carlsbad, CA, USA) according to the manufacturer’s instructions. For analysis of UPF3A transcripts, fragments of UPF3A cDNA containing the A7 and A9 repeats were PCR-amplified and separated on 2% agarose gels. More detailed analysis of transcript isoforms was performed by cloning different-sized PCR fragments into the pCR2.1 TOPO vector (Life Technologies) and subsequent sequencing. All primers are listed in Supplementary Table S6.
4.4. Generation of a Dox-Inducible UPF3A Expression Plasmid
Full-length UPF3A cDNA was PCR-amplified using primers extended by restriction sites NheI and NotI, digested, and cloned into NheI/NotI-digested plasmid pFH-IRESneo [59] resulting in plasmid pFH-IRESneo-UPF3A. From this plasmid the FLAG-HA-UPF3A cDNA sequence was subcloned into KpnI/NotI-digested pTRE-Tight-BI-DsRed-Express vector (Clontech, Mountain View, CA, USA) to yield the final expression vector pTRE-Tight-BI-DsRed-Express-UPF3A.
4.5. Stably Transfected Cells
KM12-TET-on colorectal tumor cells were generated by transfecting \(10^7\) KM12 cells [60] with plasmid pN1pβactin-rTA2S-M2-IRES-EG [61] using electroporation (Amaxa Nucleofector 1, program T-20, solution V, 5 µg plasmid DNA). Stable G-418 resistant clones (400 µg/mL) were selected for 4 weeks. Clones were screened for long-term and persistent EGFP (enhanced green fluorescent protein) expression (>9 months) by fluorescence microscopy. EGFP+ clones were screened for rta activity by transient transfection of a dox-inducible (final dox concentration: 0.5 µg/mL) luciferase reporter gene plasmid [62]. After subcloning by limiting dilution, one KM12-TET-on clone was transected with plasmid pTRE-Tight-BI-DsRed-Express-UPF3A as described above and stable clones were selected in the presence of hygromycin B (50 µg/mL). Candidate clones were screened for dox-inducible (0.5 µg/mL) expression of DsRed protein by fluorescence microscopy. Finally, one G418- resistant, Hyg-resistant, EGFP-positive, DsRed-positive clone was obtained that showed long-term (>1 year) and persistent dox-inducible UPF3A expression as confirmed by Western blot analysis.
4.6. Coding Microsatellite Frameshift Mutation Analysis
Genomic DNA was isolated using the DNeasy Tissue kit (Qiagen). Frameshift mutations were analyzed as described previously [4]. Primers were designed to obtain short amplicons of about 100 bp (Supplementary Table S6) to allow robust amplification from different types of tissue. PCR fragments were analyzed on an ABI3100 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA).
4.7. Western Blot Analysis
Soluble protein was extracted by lysing cell pellets in 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1% Triton X-100, 1% sodium deoxycholate, 0.1% SDS, 0.1 mM CaCl2, and 0.01 mM MgCl2, followed by sonication (Bandelin-Sonopuls, Bandelin electronic GmbH & Co. KG, Berlin, Germany) and
ultracentrifugation (100,000× g, 15 min, 4 °C in a Beckman TLA 100.2 rotor, Beckman Coulter, Indianapolis, IN, USA.). Protein concentration was determined by Lowry assay [63], 65 µg of protein were subjected to Western Blot analysis as described in [64] using the primary antibodies anti-UPF3A (1:5000, rabbit-monospecific, HPA018325, Sigma-Aldrich, Saint Louis, MO, USA) and anti-UPF3B (1:2500, rabbit-monospecific, HPA001800, Sigma-Aldrich), and the secondary antibody anti-Rabbit IgG, HRP Conjugate (Promega). Visualization was performed with Western Lighting Chemiluminescence Reagent Plus (PerkinElmer, Rodgau, Germany) on Kodak BioMax light films (Sigma-Aldrich).
4.8. Immunohistochemistry
Paraffin blocks were cut into 2-µm-thick sections. Deparaffinization and tissue staining were performed according to standard protocols [65] using anti-UPF3A primary antibody (1:300 1% horse serum/PBS-T, 2h or overnight at RT; Sigma-Aldrich) followed by secondary biotinylated anti-rabbit antibody (1:50 in 1% horse serum/PBS-T, 30 min at RT; Vector Laboratories, Burlingame, CA, USA).
4.9. Immunofluorescence Staining and Imaging
For immunofluorescence staining, 3-µm sections were deparaffinized, rehydrated and boiled in Epitope Retrieval Solution (CINtec® PLUS Cytology, Roche, Basel, Switzerland) for 10 min. After rinsing with deionized water, slides were washed twice in PBS for 5 min and once with deionized water. Immunofluorescence staining was performed as described before [66] with following primary antibodies (in 3% BSA/PBS, overnight, 4 °C): anti-UPF3A (1:1000, ScyTek Laboratories, Logan, UT, USA) and secondary antibodies (in 3% BSA/PBS, 30 min at 37 °C followed by 30 min incubation at RT): Alexa Fluor 488 labeled anti-mouse IgG (1:800, Thermo Fisher Scientific, Waltham, MA, USA) and Alexa Fluor 594 labeled anti-rabbit IgG (1:200, Thermo Fisher Scientific). Subsequently, counterstaining of the nuclei with DAPI (1:500, Thermo Fischer Scientific) was carried out in 3% BSA/PBS for 30 min in a humid chamber at RT. Finally, slides were embedded using DAKO Fluorescence Mounting Medium (DAKO, Hamburg, Germany). For colocalization studies both primary as well as secondary antibodies were applied at the same time. Immunofluorescence analysis was carried out using an Olympus AX 70 (40× magnification) microscope. No fluorescence staining was observed on control slides. For further colocalization studies confocal laser scanning microscopy was conducted equipped with a Plan-Apochromat 63×/1.40 Oil DIC objective, a UV diode, Argon and Helium-Neon lasers with emissions at respective wavelengths: 405 nm (DAPI), 488 nm (FITC) and 594 nm (TRITC), as well as reflected light photomultiplier tubes. Image acquisition and processing was performed using Leica LAS AF and ImageJ software [67]. Background was subtracted with constant settings using ImageJ’s Rolling ball background subtraction.
4.10. SILAC Labeling and Protein Extraction
SILAC labeling was performed as described before [64,68]. Briefly, two cell populations of generated KM12-UPF3A clone were cultured separately in ‘heavy medium (containing L-[13C6,15N4] arginine (R10); L-[13C6,15N2] lysine (K8)) or ‘light’ medium (containing L-[12C6,14N4] arginine (R0); L-[12C6,14N2] lysine (K0)) (Silantes, München, Germany) at 37 °C in a 5% CO2 atmosphere for 8 days. To avoid arginine-to-proline conversion, the medium was additionally supplemented with L-proline (Sigma-Aldrich) to a final concentration of 200 µg/mL [69]. The saturated incorporation was confirmed as described below and shown at Supplementary Figure S4. ‘Heavy’ labeled cells were then treated for 24 h with doxycycline (500 ng/mL) to induce UPF3A expression, whereas control cell populations were exposed to dox-free medium. The experiment was performed in triplicate. For protein extraction, cells were suspended in a radioimmunoprecipitation assay buffer (RIPA), containing 50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS supplemented with 1% DTT and fresh protease (cOmplete Mini; Roche) and phosphatase (PhosSTOP, Roche) inhibitors, and treated with benzonase (125 U; Merck Millipore, Burlington, MA, USA) on an orbital shaker (at 300 rpm) on ice for 1 h. After centrifugation at 13 000 rpm for 30 min at 4 °C protein concentration
of the extracts was measured by using 2D Quant Kit reagents (GE Healthcare, Chicago, IL, USA) according to the manufacturer’s instructions.
4.11. Tryptic Digestion
Protein lysates from both culture conditions (‘heavy’ and ‘light’) were mixed in a 1:1 ratio based on their protein concentration. Quantitative protein precipitation using a methanol-chloroform-water mixture [70] was performed in order to remove reagents, especially protease inhibitors, prior to tryptic digestion. Enzymatic digestion was performed in low (5 µg) and high (300 µg) protein amount samples as described in [64] with 50 ng or 3 µg trypsin in 40 mM NH₄HCO₃ solution overnight at 37 °C (for high protein amount with constant shaking on a Thermomixer (500 rpm)). After digestion, 5 µg protein samples were subjected to shot-gun mass spectrometry analysis, while 300 µg protein samples underwent phosphopeptide enrichment for phosphorylation analysis. In parallel, 5 µg of the protein lysate of ‘heavy’ labeled cells after 8 days of culture was subjected to the analysis of the amino acids incorporation and underwent the same tryptic digestion procedure.
4.12. Phosphopeptide Enrichment
IMAC material was prepared as described in [64] from Ni-NTA silica material contained in 6 spin columns (Ni-NTA Spin Columns, Qiagen).
To clean and concentrate peptide mixtures after tryptic digestion, StageTip procedure [71] was applied as described before [64] using C18 material and the reversed phase material (Oligo™ R3, Applied Biosystems) packed into a pipette tip (volume up to 200 µL). Briefly, binding was performed with 2.5% formic acid followed by washing with 2.5% formic acid and elution with 2 times 100 µL of 0.6% acetic acid in 80% acetonitrile.
Each sample was then diluted with 0.6% acetic acid to a final concentration of 60% acetonitrile and added to the 300 µg of prepared IMAC material that was washed thrice with 100 µL of 0.6% acetic acid in 60% acetonitrile before use. Samples were vortexed briefly and incubated for 1.5 h on a rotator. After centrifugation the supernatant from each sample was transferred to the 3 mg of prepared and freshly washed IMAC material (as described above) and incubated on a rotator. After 1.5 h of incubation all IMAC material was washed three times with 100 µL 0.6% acetic acid in 60% acetonitrile. Elution of the phosphopeptides from the IMAC material was performed twice with 40 µL 1% NH₃ and 5 min incubation with occasional vortexing. The final solution was processed by StageTip purification as described above with volume adjustment for lower peptide amount to 10 µL pipette tip (maximal solution volume: 50 µL). Each solution was dried completely in a vacuum centrifuge and frozen. Prior to nanoLC-ESI-MS/MS analysis peptides were redissolved in 5 µL in 2.5% hexafluoroisopropanol/0.1% TFA by sonication for 5 min.
4.13. LC-MS/MS
Peptides from tryptic digestion were separated using the Dionex UltiMate 3000 nanoUPLC system as described before [72]. Peptides were trapped on an Acclaim Pepmap 100 column (100 µm × 20 mm, particle size 5 µm). The liquid chromatography separation was performed on a C18 column (75 µm × 50 cm, particle size 2 µm) with a flow rate of 300 nL/min using a 2 h gradient of solvent A (99.9% water, 0.1% formic acid) and solvent B (80% acetonitrile, 19.9% water, 0.1% formic acid) in the following sequence: 2 min at 2% B, from 2 to 8% B in 1 min, from 8 to 25% B in 80 min, from 25 to 40% B in 10 min, from 40 to 95% B in 1 min, 5 min at 95% B, from 95 to 2% B in 1 min, and 20 min at 2% B. The nanoUPLC system was coupled online to a Q Exactive HF-X Hybrid Quadrupole-Orbitrap mass spectrometer (Thermo Fisher Scientific). The following parameters were set: ESI voltage 2200 V; capillary temperature 275 °C, normalized collision energy 35 V. Data were acquired by scan cycles of one FTMS (Fourier-transform mass spectrometry) scan with a resolution of 120,000 at m/z 200 and a range from 300 to 2000 m/z in parallel with ten MS/MS scans in the ion trap of the most abundant precursor ions.
Peptides after phosphopeptide enrichment and peptides for incorporation analysis were analyzed by a linear ion trap quadrupole LTQ Orbitrap-XL mass spectrometer (Thermo Fisher Scientific) coupled to a nanoAcquity ultra high-performance liquid chromatography (UPLC) system (Waters) as described in [64,72]. Peptides were separated on a BEH C18 (100 μm × 100 mm, particle size 1.7 μm), analytical column at a constant flow of 0.4 μL/min using a 3h stepped linear gradient of solvent C (98.9% water, 1% acetonitrile, 0.1% formic acid) and solvent D (99.9% acetonitrile and 0.1% formic acid) in the following sequence from 0 to 4% D in 1 min, from 4 to 25% D in 139 min, from 25 to 40% D in 15 min, from 40 to 85% D in 10 min, 5 min at 85% D, from 85 to 4% D in 2 min, and 15 min at 4% D. The Orbitrap was operated with the following parameters: ESI voltage 2000 V, capillary temperature 200 °C, normalized collision energy 35 V. Data were acquired using XCalibur (version 2.0.7; Thermo Fisher Scientific) by scan cycles of one FTMS scan with a resolution of 60,000 at m/z 400 and a range from 300 to 2000 m/z in parallel with six MS/MS scans in the ion trap of the most abundant precursor ions.
4.14. Protein and Phosphopeptide Identification and Quantification
The MS files were processed with the MaxQuant software (version 1.6.2.6) [73] and searched with Andromeda search engine [74] against the human SwissProt database (download: 2019.03.01, 20,412 entries) [75]. Enzyme specificity was set to that of trypsin, allowing for cleavage N-terminal to proline residues and up to two missed cleavage sites (for proteome) and up to four missed cleavage sites (for phosphopeptides). A minimum peptide length of seven amino acids was required. Carbamidomethylation (C) was set as fixed modification, whereas oxidation (M), deamidation (NQ), protein N-terminal acetylation and if necessary, phosphorylation (STY) were considered to be variable modifications. No labeling or double SILAC labeling was defined according to a maximum of 3 or 5 labeled amino acids. Mass tolerances were defined for precursor and fragmented ions as follows: MS first search—20 ppm, MS main search—6 ppm and MS/MS—0.5 Da. The false discovery rates (FDRs) at the protein and peptide level were set to 1%. SILAC-based quantification was based on unique and razor peptides only, and a minimum of two ratio counts was required. Peptide ratios were calculated and normalized for each arginine- and/or lysine-containing peptide as described [73]. In addition, the “match between the runs” feature was implemented with default settings to increase the number of quantified peptides.
Incorporation efficiency of SILAC labeling was analyzed using written R script as described in [76] for all peptides and for peptides with each isotope separately as shown in Supplementary Figure S4.
Further data analysis was performed in Perseus (version 1.6.1.3) software. Matches to the reverse database proteins identified by one site only in modified peptides and common contaminants (KRT2 and KRT10) were removed from the MaxQuant output. Exclusively phosphosites quantified in at least 2 (out of 3) replicates and with localization probability higher than 0.75 were subjected to further analysis. Only proteins identified with at least two unique peptides and quantified in at least 2 (out of 3) biological replicates were considered for the subsequent analysis. Obtained phosphopeptides ratios were corrected for differential protein expression by dividing by the matched protein ratios. Proteins and phosphosites changed by >1.5-fold in at least two biological replicates were considered regulated. In addition, to each identified and quantified phosphosite information from PhosphoSite Plus database [77] were assigned, including known and regulatory phosphosites.
4.15. Data Analysis
Global interaction network of regulated proteins and phosphoproteins was predicted in STRING v11.0 [78]. Each protein-protein interaction (PPI) has a combined score (edge score), which represents the reliability of the interaction between proteins. The PPI interactions with a combined score (0: lowest confidence; 1: highest confidence) larger than 0.4 were used for network visualization. In addition, enrichment analysis of regulated proteins and phosphoproteins were performed also in STRING v11 for Gene Ontology Biological Processes (GOBP), Cellular Compartments (GOCOCC) and Molecular Function (GOMF). Multiple hypothesis testing was controlled by using a Benjamini-Hochberg
FDR. Visible clusters on PPI maps were assigned to enriched ontologies by color coding the nodes (proteins and phosphoproteins).
**Supplementary Materials:** Supplementary materials can be found at [http://www.mdpi.com/1422-0067/21/15/5234/s1](http://www.mdpi.com/1422-0067/21/15/5234/s1).
**Author Contributions:** Conceptualization, J.K. and J.G.; Data curation, M.M., E.-M.K., J.K. and J.G.; Funding acquisition, J.K. and J.G.; Methodology, M.M., E.-M.K., N.R., V.F., U.W. and M.K.; Software, S.M.W., Y.P.Y. and P.B.; Supervision, A.K., M.v.K.D., J.K. and J.G.; Writing—Original draft, M.M., J.K. and J.G.; Writing—Review & editing, M.M., Y.P.Y., P.B., G.N.-Y., M.K., J.K. and J.G. All authors have read and agreed to the published version of the manuscript.
**Funding:** This research was supported in part by the DFG (Nr. 592/6-2), German Cancer Aid (Nr. 110121) and Sander Stiftung.
**Acknowledgments:** The expert technical assistance of Marcel Karl, Sigrun Himmelsbach and Beate Amtor is gratefully appreciated. We thank N. Gehring for helpful discussions and D. Krunic and the DKFZ Light Microscopy facility for the support with image acquisition and data analysis.
**Conflicts of Interest:** The authors declare no conflict of interest.
**Abbreviations**
- **cMNR** Coding mononucleotide repeat
- **CRC** Colorectal cancer
- **Dox** Doxycycline
- **dUPF3A** UPF3A-deficient
- **FTMS** Fourier-transform mass spectrometry
- **MMR** DNA mismatch repair
- **MS** Mass spectrometry
- **MSI** Microsatellite instability
- **MSS** Microsatellite stability
- **NMD** Nonsense-mediated RNA decay
- **PPI** Protein-protein interaction
- **PTC** Premature termination codon
- **pUPF3A** UPF3A-proficient
- **SILAC** Stable isotope labeling with amino acids in cell culture
- **UPLC** Ultra high-performance liquid chromatography
- **WT** Wild type
**References**
1. Buckowitz, A.; Knaebel, H.; Benner, A.; Bläker, H.; Gebert, J.; Kienle, P.; von Knebel Doeberitz, M.; Kloos, M. Microsatellite instability in colorectal cancer is associated with local lymphocyte infiltration and low frequency of distant metastases. *Br. J. Cancer* 2005, 92, 1746–1753. [CrossRef] [PubMed]
2. Lynch, H.T.; Smyrk, T.; Lynch, J.F. Molecular genetics and clinical-pathology features of hereditary nonpolyposis colorectal carcinoma (Lynch syndrome). *Oncology* 1998, 55, 103–108. [CrossRef] [PubMed]
3. Popat, S.; Hubner, R.; Houlston, R. Systematic review of microsatellite instability and colorectal cancer prognosis. *J. Clin. Oncol.* 2005, 23, 609–618. [CrossRef] [PubMed]
4. Woerner, S.M.; Gebert, J.; Yuan, Y.P.; Sutter, C.; Ridder, R.; Bork, P.; von Knebel Doeberitz, M. Systematic identification of genes with coding microsatellites mutated in DNA mismatch repair-deficient cancer cells. *Int. J. Cancer* 2001, 93, 12–19. [CrossRef]
5. Mori, Y.; Yin, J.; Rashid, A.; Leggett, B.A.; Young, J.; Simms, L.; Kuehl, P.M.; Langenberg, P.; Meltzer, S.J.; Stine, O.C. Instabilotyping: Comprehensive identification of frameshift mutations caused by coding region microsatellite instability. *Cancer Res.* 2001, 61, 6046–6049. [CrossRef] [PubMed]
6. Duval, A.; Gayet, J.; Zhou, X.-F.; Iacopetta, B.; Thomas, G.; Hamelin, R. Frequent frameshift mutations of the TCF-4 gene in colorectal cancers with microsatellite instability. *Cancer Res.* 1999, 59, 4213–4215. [CrossRef]
7. Holbrook, J.A.; Neu-Yilik, G.; Hentze, M.W.; Kulozzik, A.E. Nonsense-mediated decay approaches the clinic. *Nat. Genet.* 2004, 36, 801–808. [CrossRef]
8. Isken, O.; Maquat, L.E. The multiple lives of NMD factors: Balancing roles in gene and genome regulation. Nat. Rev. Genet. 2008, 9, 699–712. [CrossRef]
9. Kurosaki, T.; Popp, M.W.; Maquat, L.E. Quality and quantity control of gene expression by nonsense-mediated mRNA decay. Nat. Rev. Mol. Cell Biol. 2019, 20, 406–420. [CrossRef]
10. Kunz, J.B.; Neu-Yilik, G.; Hentze, M.W.; Kulozik, A.E.; Gehring, N.H. Functions of hUpf3a and hUpf3b in nonsense-mediated mRNA decay and translation. RNA 2006, 12, 1015–1022. [CrossRef]
11. Neu-Yilik, G.; Raimondeau, E.; Eliseev, B.; Yeramala, L.; Amthor, B.; Deniaud, A.; Huard, K.; Kerschgens, K.; Hentze, M.W.; Schaffitzel, C. Dual function of UPF3B in early and late translation termination. EMBO J. 2017, 36, 2968–2986. [CrossRef]
12. Shum, E.Y.; Jones, S.H.; Shao, A.; Dumdie, J.; Krause, M.D.; Chan, W.-K.; Lou, C.-H.; Espinoza, J.L.; Song, H.-W.; Phan, M.H. The antagonistic gene paralogs Upf3a and Upf3b govern nonsense-mediated RNA decay. Cell 2016, 165, 382–395. [CrossRef] [PubMed]
13. Nasif, S.; Contu, L.; Mühlemann, O. Beyond quality control: The role of nonsense-mediated mRNA decay (NMD) in regulating gene expression. Semin. Cell Dev. Biol. 2018, 75, 78–87. [CrossRef]
14. Hall, G.; Thein, S. Nonsense codon mutations in the terminal exon of the beta-globin gene are not associated with a reduction in beta-mRNA accumulation: A mechanism for the phenotype of dominant beta-thalassemia. Blood 1994, 83, 2031–2037. [CrossRef] [PubMed]
15. Tarpey, P.S.; Raymond, F.L.; Nguyen, L.S.; Rodriguez, J.; Hackett, A.; Vandeure, L.; Smith, R.; Shoubridge, C.; Edkins, S.; Stevens, C. Mutations in UPF3B, a member of the nonsense-mediated RNA decay complex, cause syndromic and nonsyndromic mental retardation. Nat. Genet. 2007, 39, 1127–1133. [CrossRef] [PubMed]
16. Addington, A.; Gauthier, J.; Piton, A.; Hamdan, F.; Raymond, A.; Gogtay, N.; Miller, R.; Tossell, J.; Bakalar, J.; Germain, G. A novel frameshift mutation in UPF3B identified in brothers affected with childhood onset schizophrenia and autism spectrum disorders. Mol. Psychiatry 2011, 16, 238–239. [CrossRef]
17. Chan, W.-K.; Bhalla, A.D.; Le Hir, H.; Nguyen, L.S.; Huang, L.; Georz, J.; Wilkinson, M.F. A UPF3-mediated regulatory switch that maintains RNA surveillance. Nat. Struct. Mol. Biol. 2009, 16, 747–753. [CrossRef]
18. Laumonnier, F.; Shoubridge, C.; Antar, C.; Nguyen, L.; Van Esch, H.; Kleefta, T.; Brailat, S.; Fryns, J.; Hamel, B.; Chelly, J. Mutations of the UPF3B gene, which encodes a protein widely expressed in neurons, are associated with nonspecific mental retardation with or without autism. Mol. Psychiatry 2010, 15, 767–776. [CrossRef]
19. El-Bchiri, J.; Buhrad, O.; Penard-Lacronique, V.; Thomas, G.; Hamelin, R.; Duval, A. Differential nonsense mediated decay of mutated mRNAs in mismatch repair deficient colorectal cancers. Human Mol. Genet. 2005, 14, 2435–2442. [CrossRef]
20. El-Bchiri, J.; Guilloux, A.; Dartigues, P.; Loire, E.; Mercier, D.; Buhrad, O.; Sobbani, I.; de la Grange, P.; Auboeuf, D.; Praz, F.; et al. Nonsense-Mediated mRNA Decay Impacts MSI-Driven Carcinogenesis and Anti-Tumor Immunity in Colorectal Cancers. PLoS ONE 2008, 3, e2583. [CrossRef] [PubMed]
21. Liu, C.; Karam, R.; Zhou, Y.; Su, F.; Ji, Y.; Li, G.; Xu, G.; Lu, L.; Wang, C.; Song, M. The UPF1 RNA surveillance gene is commonly mutated in pancreatic adenosquamous carcinoma. Nat. Med. 2014, 20, 596–598. [CrossRef] [PubMed]
22. Chang, L.; Li, C.; Guo, T.; Wang, H.; Ma, W.; Yuan, Y.; Liu, Q.; Ye, Q.; Liu, Z. The human RNA surveillance factor UPF1 regulates tumorigenesis by targeting Smad7 in hepatocellular carcinoma. J. Exp. Clin. Cancer Res. 2016, 35, 8. [CrossRef] [PubMed]
23. Popp, M.W.; Maquat, L.E. Nonsense-mediated mRNA decay and cancer. Curr. Opin. Genet. Dev. 2018, 48, 44–50. [CrossRef] [PubMed]
24. Woerner, S.M.; Benner, A.; Sutter, C.; Schiller, M.; Yuan, Y.P.; Keller, G.; Bork, P.; Doeberitz, M.V.K.; Gebert, J.F. Pathogenesis of DNA repair-deficient cancers: A statistical meta-analysis of putative Real Common Target genes. Oncogene 2003, 22, 2226–2235. [CrossRef]
25. Su, A.I.; Wiltshire, T.; Batalov, S.; Lapp, H.; Ching, K.A.; Block, D.; Zhang, J.; Soden, R.; Hayakawa, M.; Kreiman, G.; et al. A gene atlas of the mouse and human protein-encoding transcriptomes. Proc. Natl. Acad. Sci. USA 2004, 101, 6062–6067. [CrossRef]
26. Safran, M.; Dalah, I.; Alexander, J.; Rosen, N.; Iny Stein, T.; Shmoish, M.; Nativ, N.; Bahir, I.; Doniger, T.; Krug, H.; et al. GeneCards Version 3: The human gene integrator. Database 2010, 2010. [CrossRef]
49. Notarnicola, M.; Messa, C.; Cavallini, A.; Bifulco, M.; Tecce, M.F.; Eletto, D.; Di Leo, A.; Montemurro, S.; Laezza, C.; Caruso, M.G. Higher farnesyl diphosphate synthase activity in human colorectal cancer inhibition of cellular apoptosis. *Oncoology* 2004, 67, 351–358. [CrossRef]
50. Bardou, M.; Barkun, A.; Martel, M. Effect of statin therapy on colorectal cancer. *Gut* 2010, 59, 1572–1585. [CrossRef]
51. Murai, T. Cholesterol lowering: Role in cancer prevention and treatment. *Biol. Chem.* 2015, 396, 1–11. [CrossRef] [PubMed]
52. Schackmann, R.C.; Tenhagen, M.; van de Ven, R.A.; Derksen, P.W. p120-catenin in cancer–mechanisms, models and opportunities for intervention. *J. Cell Sci.* 2013, 126, 3515–3525. [CrossRef]
53. Zhao, Y.; Yu, T.; Zhang, N.; Chen, J.; Zhang, P.; Li, S.; Luo, L.; Cui, Z.; Qin, Y.; Liu, F. Nuclear E-cadherin acetylation promotes colorectal tumorigenesis via enhancing β-catenin activity. *Mol. Cancer Res.* 2019, 17, 655–665. [CrossRef] [PubMed]
54. Uhlen, M.; Oksvold, P.; Fagerberg, L.; Lundberg, E.; Jonasson, K.; Forsberg, M.; Zwahlen, M.; Kampf, C.; Wester, K.; Hoerger, S. Towards a knowledge-based human protein atlas. *Nat. Biotechnol.* 2010, 28, 1248–1250. [CrossRef] [PubMed]
55. Uhlen, M.; Oksvold, P.; Fagerberg, L.; Lundberg, E.; Jonasson, K.; Forsberg, M.; Zwahlen, M.; Kampf, C.; Wester, K.; Hoerger, S. Towards a knowledge-based human protein atlas. *Nat. Biotechnol.* 2010, 28, 1248–1250. [CrossRef] [PubMed]
56. Kilipinen, S.; Autio, R.; Ojala, K.; Iljin, K.; Bucher, E.; Sara, H.; Pisto, T.; Saarela, M.; Skotheim, R.I.; Björkman, M. Systematic bioinformatic analysis of expression levels of 17,330 human genes across 9783 samples from 175 types of healthy and pathological tissues. *Genome Biol.* 2008, 9, R139. [CrossRef] [PubMed]
57. Boland, C.R.; Thibodeau, S.N.; Hamilton, S.R.; Sidransky, D.; Eshleman, J.R.; Burt, R.W.; Meltzer, S.J.; Kinzler, K.W.; Vogelstein, B. A genetic model for colorectal tumorigenesis. *Proc. Natl. Acad. Sci. USA* 1998, 95, 11440–11445. [CrossRef] [PubMed]
58. Findeisen, P.; Kloor, M.; Merx, S.; Sutter, C.; Woerner, S.M.; Dostmann, N.; Benner, A.; Donudog, B.; Pawliita, M.; Dippold, W.; et al. T25 Repeat in the 3’ Untranslated Region of the CASP2 Gene: A Sensitive and Specific Marker for Microsatellite Instability in Colorectal Cancer. *Cancer Res.* 2005, 65, 8072–8078. [CrossRef]
59. Martínez, E.; Palhan, V.B.; Tjernberg, A.; Lymar, E.S.; Gamper, A.M.; Kundu, T.K.; Chait, B.T.; Di Leo, A.; Montemurro, S.; Laezza, C.; Caruso, M.G. Higher farnesyl diphosphate synthase activity in human colorectal cancer inhibition of cellular apoptosis. *Oncoology* 2004, 67, 351–358. [CrossRef]
60. Welman, A.; Barraclough, J.; Dive, C. Generation of cells expressing improved doxycycline-regulated reverse transcriptional transactivator rtTA2S-M2. *Nat. Protoc.* 2006, 1, 803–811. [CrossRef] [PubMed]
61. Welman, A.; Barraclough, J.; Dive, C. Generation of cells expressing improved doxycycline-regulated reverse transcriptional transactivator rtTA2S-M2. *Nat. Protoc.* 2006, 1, 803–811. [CrossRef] [PubMed]
62. Gossen, M.; Bujard, H. Tight control of gene expression in mammalian cells by tetracycline-responsive promoters. *Proc. Natl. Acad. Sci. USA* 1992, 89, 5547–5551. [CrossRef] [PubMed]
63. Lowry, O.H.; Rosebrough, N.J.; Farr, A.L.; Randall, R.J. Protein measurement with the Folin phenol reagent. *J. Biol. Chem.* 1951, 193, 265–275. [PubMed]
64. Michalak, M.; Warnken, U.; Schnöller, M.; Gabius, H.J.; Kopitz, J. Detection of malignancy-associated phosphoprotein changes in human colorectal cancer induced by cell surface binding of growth-inhibitory galectin-4. *IUBMB Life* 2019, 71, 364–375. [CrossRef] [PubMed]
65. Kloor, M.; Sutter, C.; Wentzensen, N.; Cremer, F.W.; Buckowitz, A.; Keller, M.; von Knebel Doeberitz, M.; Gebert, J. A large MSH2 Alu insertion mutation causes HNPCC in a German kindred. *Human Genet.* 2004, 115, 432–438. [CrossRef]
66. Katzenmaier, E.; Stark, H.-J.; Gebert, J.; Kopitz, J. Galectin-12 colocalizes with splicing factor-rich speckles and shuttles between the nucleus and cytoplasm in colon cancer cells. *J. Mol. Biochem.* 2018, 7, 28–40.
67. Schindelin, J.; Arganda-Carreras, I.; Frise, E.; Kaynig, V.; Longair, M.; Pietzsch, T.; Preibisch, S.; Rueden, C.; Saalfeld, S.; Schmid, B. Fiji: An open-source platform for biological-image analysis. *Nat. Methods* 2012, 9, 676–682. [CrossRef]
68. Michalak, M.; Warnken, U.; Andre, S.; Schnolzer, M.; Gabius, H.-J.; Kopitz, J. Detection of proteome changes in human colon cancer induced by cell surface binding of growth-inhibitory human galectin-4 using quantitative SILAC-based proteomics. *J. Proteome Res.* **2016**, *15*, 4412–4422. [CrossRef]
69. Lößner, C.; Warnken, U.; Pscherer, A.; Schnölzer, M. Preventing arginine-to-proline conversion in a cell-line-independent manner during cell cultivation under stable isotope labeling by amino acids in cell culture (SILAC) conditions. *Anal. Biochem.* **2011**, *412*, 123–125. [CrossRef]
70. Wessel, D.M.; Flügge, U. A method for the quantitative recovery of protein in dilute solution in the presence of detergents and lipids. *Anal. Biochem.* **1984**, *138*, 141–143. [CrossRef]
71. Rappsilber, J.; Mann, M.; Ishihama, Y. Protocol for micro-purification, enrichment, pre-fractionation and storage of peptides for proteomics using StageTips. *Nat. Protoc.* **2007**, *2*, 1896. [CrossRef] [PubMed]
72. Fricke, F.; Michalak, M.; Warnken, U.; Hausser, I.; Schnölzer, M.; Kopitz, J.; Gebert, J. SILAC-Based Quantification of TGFR2-Regulated Protein Expression in Extracellular Vesicles of Microsatellite Unstable Colorectal Cancers. *Int. J. Mol. Sci.* **2019**, *20*, 4162. [CrossRef] [PubMed]
73. Cox, J.; Mann, M. MaxQuant enables high peptide identification rates, individualized ppb-range mass accuracies and proteome-wide protein quantification. *Nat. Biotechnol.* **2008**, *26*, 1367–1372. [CrossRef] [PubMed]
74. Cox, J.; Neuhauser, N.; Michalski, A.; Scheltema, R.A.; Olsen, J.V.; Mann, M. Andromeda: A peptide search engine integrated into the MaxQuant environment. *J. Proteome Res.* **2011**, *10*, 1794–1805. [CrossRef] [PubMed]
75. The UniProt Consortium. UniProt: The universal protein knowledgebase. *Nucleic Acids Res.* **2017**, *45*, D158–D169. [CrossRef]
76. Stohr, G.; Andreas, T. Quantitative LC-MS of Proteins. In *Protein and Peptide Analysis by LC-MS: Experimental Strategies*; Letzel, T., Ed.; Royal Society of Chemistry: London, UK, 2011.
77. Hornbeck, P.V.; Zhang, B.; Murray, B.; Kornhauser, J.M.; Latham, V.; Skrzypek, E. PhosphoSitePlus, 2014: Mutations, PTMs and recalibrations. *Nucleic Acids Res.* **2015**, *43*, D512–D520. [CrossRef]
78. Szklarczyk, D.; Gable, A.L.; Lyon, D.; Junge, A.; Wyder, S.; Huerta-Cepas, J.; Simonovic, M.; Doncheva, N.T.; Morris, J.H.; Bork, P. STRING v11: Protein—Protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets. *Nucleic Acids Res.* **2019**, *47*, D607–D613. [CrossRef]
© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/). | 2025-03-06T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3672,
1
],
[
3672,
7563,
2
],
[
7563,
11721,
3
],
[
11721,
15736,
4
],
[
15736,
19859,
5
],
[
19859,
20670,
6
],
[
20670,
23062,
7
],
[
23062,
25075,
8
],
[
25075,
28104,
9
],
[
28104,
29097,
10
],
[
29097,
33813,
11
],
[
33813,
38015,
12
],
[
38015,
41853,
13
],
[
41853,
46178,
14
],
[
46178,
50230,
15
],
[
50230,
54759,
16
],
[
54759,
58272,
17
],
[
58272,
62853,
18
],
[
62853,
62853,
19
],
[
62853,
67294,
20
],
[
67294,
70162,
21
]
]
} |
f3c08205f2f7c4e75b56a3c17b2d37061a9735ab | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 11,
"total-fallback-pages": 0,
"total-input-tokens": 58367,
"total-output-tokens": 13919,
"fieldofstudy": [
"Mathematics"
]
} | We study the Lasry-Lions approximation using the kernel determined by the fundamental solution with respect to a time-dependent Tonelli Lagrangian. This approximation process is also applied to the viscosity solutions of the discounted Hamilton-Jacobi equations.
1. Introduction
The method of Lasry-Lions regularization is a kind of variational approximation which is a generalization of the Moreau-Yosida approximation in convex analysis, see, for instance [19] and [1]. Beyond the analytic aspect of such regularization using the standard kernel \(|x - y|/2t\) (\(x, y \in \mathbb{R}^n\) and \(t > 0\)), more dynamical aspect of this method has already been studied widely in the past decade, especially with an emphasis on the weak KAM theory and Mather theory:
- An explanation of such a method using the fundamental solution of the associated Hamilton-Jacobi equations instead of the quadratic kernels was first given by Bernard ([2]). In the context of weak KAM theory, this method is closely connected to the Lax-Oleinik operators \(T_{s,t}^\pm\) ([4], [5] and [18]).
- Ilmanen’s lemma on insertion of \(C^{1,1}\) functions between a semiconvex function less than a semiconcave function ([3] and [17]).
- In [14], the authors also obtained the limiting behavior of the derivatives of the approximating sequence and the relation between the regular and singular dynamics of the associated Hamiltonian dynamical systems.
- There also exists a connection to the theory of generalized characteristics by the recent work on global propagation of singularities of the viscosity solutions of Hamilton-Jacobi equations ([9]), and [6, 7, 10] for more about the singularities propagation of weak KAM solutions.
- An application of standard Lasry-Lions approximation to the minimal homoclinic orbits ([8]).
Let \(H: \mathbb{R}^n \times \mathbb{R}^n \to \mathbb{R}\) be a Tonelli Hamiltonian (i.e., \(H = H(x,p)\) is of \(C^2\) class and it is strictly convex in \(p\) and uniformly superlinear in \(p\)), and let \(L: \mathbb{R}^n \times \mathbb{R}^n \to \mathbb{R}\) be the associated Tonelli Lagrangian. In this paper, we extend the Lasry-Lions regularization procedure to the viscosity solution of the discounted Hamilton-Jacobi equation
\[
\lambda u^\lambda(x) + H(x, Du^\lambda(x)) = 0, \quad x \in \mathbb{R}^n
\]
with a discount factor $\lambda > 0$. The associated dynamical system is dissipative system and it is a very special kind of contact type Hamiltonian systems (see, for instance, [21], [22], [11] and [23]). In fact, by defining a new Hamiltonian $H^\lambda(t, x, p) = e^{\lambda t}H(x, e^{-\lambda t}p)$, this equation can be reduced to a time-dependent evolutionary Hamilton-Jacobi equation
\[
D_t v + H^\lambda(t, x, D_xv) = 0.
\]
Therefore, one can define a kind of intrinsic Lasry-Lions regularization with respect to $u^\lambda$ as
\[
\hat{T}_t u^\lambda(x) = \sup_{y \in \mathbb{R}^n} \{u^\lambda(y) - A^\lambda_{s,t}(x, y)\},
\]
where $A^\lambda_{s,t}(x, y)$ is the fundamental solution with respect to the time-dependent Lagrangian $L^\lambda(t, x, v) = e^{\lambda t}L(x, v)$.
The main result of this paper clarifies the approximation property of this kind of Lasry-Lions regularization. We obtain not only the uniform convergence of $\hat{T}_t u^\lambda$ to $u^\lambda$ but also the limit of $DT_t u^\lambda$ as $t \to 0^+$. It is worth noting that the latter is closely connected to the intrinsic explanation of the propagation of singularities along generalized characteristics of the associated viscosity solutions ([9]).
To study the aforementioned intrinsic approximation, we need the regularity properties of the fundamental solutions $A^\lambda_{s,t}(x, y)$, which is the least action of the absolutely continuous curve connecting $x$ to $y$ from time $s$ to $t$. The required regularity result is a generalization of the relevant result in [9].
The paper is organized as follows: In section 2, we briefly review some fundamental facts of semiconcave functions and Tonelli’s theory in the calculus of variation. In section 3, we prove a Lasry-Lions approximation result for the time-dependent Lagrangians, then discuss this approximation method in a model of discounted Hamilton-Jacobi equations and its connection to the propagation of singularities of associated viscosity solutions.
Acknowledgments This work was partially supported by the Natural Scientific Foundation of China (Grant No. 11631006, No. 11501290, No. 11471238), and the National Basic Research Program of China (Grant No. 2013CB834100).
2. Viscosity Solutions and Semiconcave Functions
In this section, we briefly review some basic properties of semiconcave functions and the viscosity solutions of Hamilton-Jacobi equations.
Let $\Omega \subset \mathbb{R}^n$ be a convex open set. A function $u : \Omega \to \mathbb{R}^n$ is semiconcave (with linear modulus) if there exists a constant $C > 0$ such that
\[
\lambda u(x) + (1 - \lambda)u(y) - u(\lambda x + (1 - \lambda)y) \leq \frac{C}{2} \lambda(1 - \lambda)|x - y|^2
\]
for any $x, y \in \Omega$ and $\lambda \in [0, 1]$. The constant $C$ that satisfies the above inequality is called a semiconcavity constant of $u$ in $\Omega$. A function $u$ is said to be locally semiconcave if for each $x \in \Omega$ there exists an open ball $B(x, r) \subset \Omega$ such that $u$ is a semiconcave function on $B(x, r)$.
If $D \subset \mathbb{R}^n \rightarrow \mathbb{R}$ be a continuous function, $x \in \Omega$, the following closed convex sets
\[ D^+ u(x) = \{ p \in \mathbb{R}^n : \limsup_{y \rightarrow x} \frac{u(y) - u(x) - \langle p, y - x \rangle}{|y - x|} \leq 0 \} \]
\[ D^- u(x) = \{ p \in \mathbb{R}^n : \liminf_{y \rightarrow x} \frac{u(y) - u(x) - \langle p, y - x \rangle}{|y - x|} \geq 0 \} \]
are called the \textit{superdifferential} and \textit{subdifferential} of $u$ at $x$ respectively.
**Definition 2.2.** Let $u : \Omega \subset \mathbb{R}^n \rightarrow \mathbb{R}$ be locally Lipschitz. We call a vector $p \in \mathbb{R}^n$ a \textit{limiting differential} of $u$ at $x$ if there exists a sequence $\{ x_k \} \subset \Omega \setminus \{ x \}$ such that $u$ is differentiable at $x_k$ for all $k \in \mathbb{N}$ and
\[ \lim_{k \rightarrow \infty} x_k = x, \quad \text{and} \quad \lim_{k \rightarrow \infty} Du(x_k) = p. \]
The set of all limiting differentials of $u$ at $x$ is denoted by $D^+ u(x)$.
**Proposition 2.3** ([12]). Let $u : \Omega \rightarrow \mathbb{R}$ be a semiconcave function and $x \in \Omega$. Then the following properties hold:
1. $D^+ u(x)$ is a nonempty closed convex set in $\mathbb{R}^n$ and $D^+ u(x) \subset \partial D^+ u(x)$, where $\partial D^+ u(x)$ denotes the topological boundary of $D^+ u(x)$.
2. The set-value function $x \mapsto D^+ u(x)$ is upper semi-continuous.
3. $D_+ u(x) = \text{co} D^+ u(x)$.
4. If $D^+ u(x)$ is a singleton, then $u$ is differentiable at $x$. Moreover, if $D^+ u(x)$ is a singleton for every point in $\Omega$, then $u \in C^1(\Omega)$.
Recall that a continuous real-valued function $u$ on $(0, +\infty) \times \mathbb{R}^n$ is called a \textit{viscosity supersolution} (resp. \textit{viscosity subsolution}) of the Hamilton-Jacobi equation
\[ D_t u + H(t, x, D_x u) = 0 \]
if for any $(t, x) \in (0, +\infty) \times \mathbb{R}^n$
\[ p_t + H(t, x, p_x) \geq 0 \quad (\text{resp.} \leq 0), \quad \forall (p_t, p_x) \in D^- u(t, x) \quad (\text{resp.} D^+ u(t, x)). \]
A continuous function $u$ is called a \textit{viscosity solution} of the equation if it is both a viscosity subsolution and a viscosity supersolution.
In this paper, we concentrate on Lagrangians on Euclidean configuration space $\mathbb{R}^n$. We say that a function $\theta : [0, +\infty) \rightarrow [0, +\infty)$ is \textit{superlinear} if $\theta(r)/r \rightarrow +\infty$ as $r \rightarrow +\infty$.
**Definition 2.4.** A function $L : \mathbb{R} \times \mathbb{R}^n \times \mathbb{R}^n \rightarrow \mathbb{R}$ is called a (time-dependent) \textit{Tonelli Lagrangian} if $L$ is a function of class $C^2$ satisfying the following conditions:
1. $L_{vv}(t, x, v)$ is positive definite for all $(t, x, v) \in \mathbb{R} \times \mathbb{R}^n \times \mathbb{R}^n$.
2. There exist two superlinear functions $\theta, \bar{\theta} : [0, +\infty) \rightarrow [0, +\infty)$ and a constant $c_0 \geq 0$ such that
\[ \bar{\theta}(|v|) \geq L(t, x, v) \geq \theta(|v|) - c_0, \quad (t, x, v) \in \mathbb{R} \times \mathbb{R}^n \times \mathbb{R}^n. \]
3. There exists a constant $c > 0$ such that
\[ |L_t(t, x, v)| \leq c(1 + L(t, x, v)), \quad (t, x, v) \in \mathbb{R} \times \mathbb{R}^n \times \mathbb{R}^n. \]
Let $L$ be a Tonelli Lagrangian and let $H$ be the associated Hamiltonian. Given $x \in \mathbb{R}^n$, $y \in B(x, R)$ with $R > 0$, and $s < t$, we define
\[ \Gamma_{x,y}^{s,t} = \{ \xi \in W^{1,k}([s,t], \mathbb{R}^n) : \xi(s) = x, \xi(t) = y \}, \]
and
\begin{equation}
A_{s,t}(x, y) = \inf_{\xi \in \Gamma_{s,t}^{x,y}} \int_s^t L(\tau, \xi(\tau); \dot{\xi}(\tau)) d\tau.
\end{equation}
The existence the minimizers in (2.4) is a well known result in Tonelli’s theory, (see, for instance, [12]). We call $\xi \in \Gamma_{s,t}^{x,y}$ a minimizer for $A_{s,t}(x, y)$ if
\[ A_{s,t}(x, y) = \int_s^t L(\tau, \xi(\tau); \dot{\xi}(\tau)) d\tau. \]
It is well known that such a minimizer $\xi$ must be of class $C^2$.
It is known that, for any $t_0 \in \mathbb{R}$, $x_0 \in \mathbb{R}^n$, the function $u(t, x) = A_{t_0,t}(x_0, x)$ is called a fundamental solution of the Hamilton-Jacobi equation
\begin{equation}
D_t u(t, x) + H(t, x, D_x u(t, x)) = 0 \quad x \in \mathbb{R}^n, t > t_0.
\end{equation}
When considering the Cauchy problem with initial condition $u(t_0, x) = u_0(x)$ with $u_0 \in \text{Lip}(\mathbb{R}^n)$, the associated unique viscosity solution has the following representation:
\begin{equation}
A_{s,t}(x, y) = \sup_{\xi \in \Gamma_{s,t}^{x,y}} \left\{ u_0(y) - A_{s,t}(x, y) \right\}, \quad x \in \mathbb{R}^n, t > t_0.
\end{equation}
Let us recall the Lax-Oleinik operators for time-dependent Lagrangians. For any $s < t$, we define
\begin{align}
T^+_n t u_0(x) &:= \sup_{y \in \mathbb{R}^n} \{ u_0(y) - A_{s,t}(x, y) \}, \\
T^-_n t u_0(x) &:= \inf_{y \in \mathbb{R}^n} \{ u_0(y) + A_{s,t}(x, y) \}.
\end{align}
Therefore, $u(t, x) = T_{t_0,t} u_0(x)$ is the unique viscosity solution of (2.2) with the initial condition $u(t_0, x) = u_0(x)$. For any $t_1 > t_0$, it is well known that $u_1(x) = u(t_1, x) = T_{t_0,t_1} u_0(x)$ is a locally semiconcave function (see [12]).
3. LASRY-LIONS APPROXIMATION FOR DISCOUNTED EQUATIONS
3.1. Positive type Lax-Oleinik Operators in time-dependent case. In [6], the authors studied the intrinsic relation between propagation of singularities and the procedure of sup-convolution. In this section, we concentrate on the case of sup-convolution $T^+_n t u$ with $u$ a locally semiconcave function.
Let $u : \mathbb{R}^n \to \mathbb{R}$ be a locally semiconcave function. Fixed $x \in \mathbb{R}^n$, $t_0 > 0$, $\kappa > 0$ and $0 < T < 1$. For any $t \in [t_0, t_0 + T]$, we define the local barrier function $\psi^{x,t}_{t_0,t} : B(x, \kappa(t-t_0)) \to \mathbb{R}$ as
\[ \psi^{x,t}_{t_0,t}(y) := u(y) - A_{t_0,t}(x, y). \]
Now we need the following condition:
\[ (M) \quad \psi^{x,t}_{t_0,t} \text{ attains a unique maximum point in } B(x, \kappa(t-t_0)). \]
The following result shows condition (M) is satisfied if $u \in \text{Lip}(\mathbb{R}^n, \mathbb{R})$. It is a slight generalization of Lemma 3.1 in [9].
**Lemma 3.1.** Suppose $L$ is a Tonelli Lagrangian and let $u_0 \in \text{Lip}(\mathbb{R}^n, \mathbb{R})$. Then, the supremum in (2.4) is attained for every $(t, x) \in (t_0, +\infty) \times \mathbb{R}^n$. Moreover,
there exists a constant $\kappa_0 > 0$, depending only on $\text{Lip}(u_0)$, such that, for any $(t, x) \in (t_0, +\infty) \times \mathbb{R}^n$ and any maximum point $y_{t,x}$ of $\psi_{t,x}^\tau(y)$, we have
$$|y_{t,x} - x| \leq \kappa_0 (t - t_0).$$
If $\xi : [t_0, t] \to \mathbb{R}^n$ is the unique minimizer for $A_{t_0,t}(x, y)$, we define the associated dual arc $p_t$ as
$$p_t(s) = L_v(s, \xi_t(s), \dot{\xi}_t(s)), \quad s \in [t_0, t].$$
**Theorem 3.2.** Suppose $u : \mathbb{R}^n \to \mathbb{R}$ is a locally semiconcave function and $L$ is a Tonelli Lagrangian. If condition $[\mathcal{M}]$ is satisfied, then $T_{t_0,t}^+ u$ is of class $C^{1,1}_{\text{loc}}$ for all $t \in [t_0, t_0 + T]$. Moreover, $\lim_{t \to t_0^+} DT_{t_0,t}^+ u(x) = q_x$, where $q_x$ is the unique element of $D^+ u(x)$ such that
$$H(t_0, x, q_x) = \min_{p \in D^+ u(x)} H(t_0, x, p)$$
**Proof.** Fix $(t_0, x) \in \mathbb{R} \times \mathbb{R}^n$, we have that $\tilde{\psi}_{t_0,t}^\tau$ attains the maximum at $y_t \in B(x, \kappa(t - t_0))$ for each $t \in [t_0, t_0 + T]$ by condition $[\mathcal{M}]$. Let $\xi_t \in \Gamma_{t_0,t}^+$ be the minimizer for $A_{t_0,t}(x, y_t)$, by $[\text{A.2}]$, we have
$$L_v(t, \xi_t(t), \dot{\xi}_t(t)) = D_y A_{t_0,t}(x, y_t) \in D^+ u(y_t),$$
since $y_t$ is a maximizer of $\psi_{t_0,t}^\tau$. Moreover, the family $\{\xi_t(\cdot)\}_{t \in [t_0, t_0 + T]}$ is equi-Lipschitz by Lemma 3.1 and Proposition A.2. Let $v_t := (\xi_t(t) - x)/(t - t_0)$, we obtain
$$\left|\frac{\xi_t(t) - x}{t - t_0} - \dot{\xi}_t(t_0)\right| \leq \frac{1}{t - t_0} \int_{t_0}^{t} |\xi_t(s) - \dot{\xi}_t(t_0)| ds \leq \frac{C_1}{t - t_0} \int_{t_0}^{t} (s - t_0) ds = \frac{C_1}{2} (t - t_0).$$
Thus, we have
$$v_0 := \lim_{t \to t_0^+} v_t = \lim_{t \to t_0^+} \dot{\xi}_t(t_0).$$
Since $u$ is a locally semiconcave function, for any $y \in B(x, \kappa(t - t_0))$, $p_x \in D^+ u(x)$ and $p_y \in D^+ u(y)$, we have $([12] \text{ Proposition 3.3.10})$
$$\langle p_y - p_x, y - x \rangle \leq C_2 |y - x|^2.$$
Taking any $t_k \to t_0$, we have
$$p_{y_k} = L_v(t_k, \xi_{t_k}(t_k), \dot{\xi}_{t_k}(t_k)) \in D^+ u(y_k).$$
Then, for any $p_x \in D^+ u(x)$
$$\langle p_x - L_v(t_k, \xi_{t_k}(t_k), \dot{\xi}_{t_k}(t_k)), v_{t_k} \rangle + C_2 (t_k - t_0)|v_{t_k}|^2 \geq 0.$$
Taking the limit in the above inequality as $k \to \infty$ we obtain
$$\langle p_x, v_0 \rangle \geq \langle L_v(t_0, x, v_0), v_0 \rangle = \langle q_x, v_0 \rangle, \quad \forall p_x \in D^+ u(x),$$
where $q_x := L_v(t_0, x, v_0) \in D^+ u(x)$ by the upper semicontinuity of $x \mapsto D^+ u(x)$. Thus, for all $p_x \in D^+ u(x)$,
$$H(t_0, x, p_x) \geq \langle L_v(t_0, x, v_0), v_0 \rangle - L(t_0, x, v_0) = H(t_0, x, q_x),$$
$^1$\text{Lip}(u_0) stands for the least Lipschitz constant of $u_0$.
and \( q_x \) is the unique minimum point of \( H(t_0, x, \cdot) \) on \( D^+ u(x) \). The uniqueness of \( p_x \) implies the uniqueness of \( v_0 \) since \( L_v(t_0, x, \cdot) \) is injective, and we have
\[
\lim_{t \to t_0^+} DT_{t_0,t}^+ u(x) = \lim_{t \to t_0^+} L_v(t_0, \xi(t_0), \xi(t_0)) = q_x.
\]
This completes the proof of the theorem. \( \square \)
### 3.2. Lasry-Lions regularization on discounted equations
For \( \lambda > 0 \), we consider the Hamilton-Jacobi equations with discount factors,
\[
(HJ_d) \quad \lambda u^\lambda(x) + H(x, Du^\lambda(x)) = 0, \quad x \in \mathbb{R}^n.
\]
Multiplying \( e^{\lambda t} \) in \((HJ_d)\) and defining \( v^\lambda(t, x) = e^{\lambda t} u^\lambda(x) \), one can check that \( v = v^\lambda \) is a viscosity solution of
\[
(HJ_e) \quad D_t v + H^\lambda(t, x, D_x v) = 0
\]
with \( H^\lambda(t, x, p) = e^{\lambda t} H(x, e^{-\lambda t} p) \), if \( u^\lambda \) is a viscosity solution of \((HJ_d)\). The associated Lagrangian \( L^\lambda \) with respect to \( H^\lambda \) has the form
\[
L^\lambda(t, x, v) = e^{\lambda t} L(x, v).
\]
**Proposition 3.3.** \( u^\lambda(x) \) is a viscosity solution of \((HJ_d)\) if and only if \( v^\lambda(t, x) \) is a viscosity solution of \((HJ_e)\).
**Proof.** It is not hard to check that \( u^\lambda \) is a locally semiconcave function if and only if so is \( v^\lambda \) when restricted to any compact time interval. The local semiconcavity properties of viscosity solutions of \((HJ_d)\) and \((HJ_e)\) are well known results, see, for instance, [12]. Thus, our conclusion is a direct consequence of Proposition 5.3.1 in [12]. \( \square \)
Now, one can define a kind of **intrinsic Lasry-Lions regularization** with respect to \( u^\lambda \) as follows: let \( u^\lambda \) be a viscosity solution of the discounted Hamilton-Jacobi equation \((HJ_d)\), define
\[
\hat{T}_t u^\lambda(x) = \sup_{y \in \mathbb{R}^n} \{ u^\lambda(y) - A_{0,t}^\lambda(x, y) \} = T_{0,t}^+ v^\lambda(0, x),
\]
where \( A_{0,t}^\lambda(x, y) \) is the fundamental solution with respect to the Lagrangian \( L^\lambda \).
**Theorem 3.4.** If \( u^\lambda \) is a viscosity solution of the discounted Hamilton-Jacobi equation \((HJ_d)\), and \( \hat{T}_t u^\lambda \) is the associated intrinsic Lasry-Lions regularization, then we have \( \hat{T}_t u^\lambda \) is of class \( C^{1,1}_{loc} \) and \( \hat{T}_t u^\lambda \) tends to \( u^\lambda \) uniformly as \( t \to 0^+ \). Moreover, there exists an unique \( q_x^\lambda \in D^+ u^\lambda(x) \) such that
\[
H(x, q_x^\lambda) = \min_{p \in D^+ u^\lambda(x)} H(x, p)
\]
and \( \lim_{t \to 0^+} D \hat{T}_t u^\lambda(x) = q_x^\lambda \).
**Proof.** Notice that \( L^\lambda \) satisfies conditions (L1)-(L3) for any fixed \( \lambda > 0 \). The \( C^{1,1} \) regularity of \( \hat{T}_t u^\lambda \) is a direct consequence of the \( C^{1,1} \) regularity of \( A_{0,t}(x, \cdot) \), (A.6) and condition (M) which holds by a slight generalization of Lemma 3.1 in [9], since \( v^\lambda(0, \cdot) = u^\lambda \) is semiconcave and Lipschitz. Now, fix \( T > 0 \) as in Theorem 3.2, then for any \( t \in (0, T] \), there exists a unique maximizer \( y_{t,x} \) of \( u^\lambda(\cdot) - A_{0,t}^\lambda(x, \cdot) \), and \( \lim_{t \to 0^+} y_{t,x} = x \), therefore \( \hat{T}_t u^\lambda \) tends to \( u^\lambda \) uniformly as \( t \to 0^+ \).
Applying Theorem 3.2 to the solution \( v^\lambda \) of (HJ\( \lambda \)), there exists a unique \( q^\lambda_x \in D^+ v^\lambda(0, x) = D^+ u^\lambda(x) \) such that
\[
\lim_{t \to t_0^+} DT_{0,t}^+ v^\lambda(0, x) = q^\lambda_x \in D^+ v^\lambda(0, x) = D^+ u^\lambda(x)
\]
and
\[
H^\lambda(0, x, q^\lambda_x) = \min_{p \in D^+_0 v^\lambda(0, x)} H^\lambda(0, x, p).
\]
which is equivalent to (3.4). \( \square \)
Let \( \lambda > 0 \), a recent work by Davini, et al. (15) shows the unique solution \( u^\lambda \) of (HJ\( \lambda \)) converges uniformly, as \( \lambda \to 0^+ \), to a certain viscosity solution of the stationary Hamilton-Jacobi equation
\[
(HJ_\lambda) \quad H(x, Du(x)) = 0,
\]
when 0 is Mańe’s critical value. Comparing to the results in [14], there exists a unique \( q_x \in D^+ u(x) \) such that
\[
(3.5) \quad H(x, q_x) = \min_{p \in D^+ u(x)} H(x, p),
\]
\[
\lim_{\lambda \to 0^+} DT_t^+ u(x) = q_x. \quad \text{Therefore, one can raise the following problem:}
\]
**Problem:** For \( q^\lambda_x \) and \( q_x \) defined in (3.4) and (3.5), does \( \lim_{\lambda \to 0^+} q^\lambda_x = q_x \)?
**Remark 3.5.** To answer Problem above, a possible systematic approach will be based on the recent works [21] and [22]. Moreover, one can understand such a problem as follows (11 and 23): We suppose \( L \) is a function of \( C^2 \) class and it satisfies the following conditions:
(L1) \( L_{vv}(x, r, v) > 0 \) for all \( (x, r, v) \in \mathbb{R}^n \times \mathbb{R} \times \mathbb{R}^n \);
(L2) For each \( r \in \mathbb{R} \), there exist two superlinear and nondecreasing function \( \overline{\theta}_r, \theta_r : [0, +\infty) \to [0, +\infty) \), \( \theta_r(0) = 0 \) and \( c_r > 0 \), such that
\[
\overline{\theta}_r(|p|) \geq L(x, r, v) \geq \theta_r(|p|) - c_r, \quad (x, v) \in \mathbb{R}^n \times \mathbb{R}^n.
\]
(L3) There exists \( K > 0 \) such that
\[
|L_r(x, r, v)| \leq K, \quad (x, r, v) \in \mathbb{R}^n \times \mathbb{R} \times \mathbb{R}^n.
\]
Fix \( x, y \in \mathbb{R}^n \), \( u \in \mathbb{R} \) and \( t > 0 \). Define \( \Gamma^t_{x,y} = \{ \xi \in AC([0, t], \mathbb{R}^n) : \xi(0) = x, \xi(t) = y \} \). Let \( \xi \in \Gamma^t_{x,y} \), we consider the Carathéodory equation
\[
(3.6) \quad \dot{u}_\xi(s) = L(\xi(s), u_\xi(s), \dot{\xi}(s)), \quad \text{a.e.} \ s \in [0, t]
\]
with initial conditions \( u_\xi(0) = u \). We define
\[
(3.7) \quad A(t, x, y, u) = u + \inf_{\xi} \int_0^t L(\xi(s), u_\xi(s), \dot{\xi}(s)) \, ds,
\]
where the infimum is taken over of \( \xi \in \Gamma^t_{x,y} \) and \( u_\xi : [0, t] \to \mathbb{R}^n \) is a absolutely continuous curve determined by (3.6). In the case of discounted equations, \( L(x, u, v) = -\lambda u + L(x, v) \). One can define the negative type Lax-Oleinik operator \( T^-_t : C(\mathbb{R}^n, \mathbb{R}) \to C(\mathbb{R}^n, \mathbb{R}) \) for any \( t > 0 \):
\[
(3.8) \quad (T^-_t \phi)(x) = \inf_{y \in \mathbb{R}^n} A(t, y, x, \phi(y)).
\]
It is not very difficult to show the fundamental solution \( A(t, x, y, u) \) is locally semiconcave with constants depending on \(|x - y|/t \) and \( \lambda \). Thus, the key point of the
uniform semiconcavity of $u^\lambda$ is to show that there exists $\kappa_0 > 0$ independent of $x$ such that the minimum of $A(t, \cdot, x, \phi(\cdot))$ on $\mathbb{R}^n$ is attained and all the minimum points is contained in $B(x, \kappa_0 t)$. If $\phi = u^\lambda$ is a Lipschitz weak KAM solution of the equation $H(x, u(x), Du(x)) = 0$ with respect to $H(x, u, p) = \lambda u + H(x, p)$, then this gives a uniform bound of the velocity all of backward calibrated curves which leads to the uniform constants in the associated semiconcavity estimate. We will answer Problem above in a much more general context in the future.
3.3. Connection to singularities. The intrinsic Lasry-Lions regularization is closely connected to the propagation of singularities of the solution $u^\lambda$ of (HJ). It is obvious that $u^\lambda$ shares the singularities of $v^\lambda$ in (HJ). In this section, we suppose that 0 is Mañe’s critical value.
Recall that a point $(t, x) \in \mathbb{R} \times \mathbb{R}^n$ is called a singular point of a semiconcave function $u(t, x)$ if $D^+ u(t, x)$ is not a singleton. The set of all singular points of $u$ is denoted by $\text{Sing}(u)$. It is obvious that $(t, x) \in \text{Sing}(v^\lambda)$ if and only if $x \in \text{Sing}(u^\lambda)$.
Using the representation formula of $v^\lambda$ (see, for instance, [15 Proposition 3.5]), for any $\tau, x \in \mathbb{R}$, we obtain that
$$v^\lambda(\tau, x) = v^\lambda(\tau - t, \gamma_x(\tau - t)) + \int_{\tau - t}^\tau L^\lambda(s, \gamma_x(s), \dot{\gamma}_x(s)) \, ds,$$
where the infimum is taken over all absolutely continuous curves $\gamma : (-\infty, \tau) \to \mathbb{R}^n$, with $\gamma(\tau) = x$. Moreover, there exists a Lipschitz and $C^2$ curve $\gamma_x : (-\infty, \tau) \to \mathbb{R}^n$, with $\gamma_x(\tau) = x$, such that, for any $t > \tau$,
$$v^\lambda(\tau, x) = v^\lambda(\tau - t, \gamma_x(\tau - t)) + \int_{\tau - t}^\tau L^\lambda(s, \gamma_x(s), \dot{\gamma}_x(s)) \, ds.$$
(3.10)
It is clear that $v^\lambda$ is differentiable at $(\tau - t, \gamma_x(\tau - t))$ for all $t > \tau$, and $\gamma_x$ is an extremal of the associated Euler-Lagrange equation with respect to $L^\lambda$. As in the classical weak KAM theory, it is not difficult to prove that $(\tau, x)$ is a differentiable point of $v^\lambda$ if and only if there exists a unique $\gamma_x$ satisfying (3.10) (see also [12 Theorem 6.4.9]).
**Theorem 3.6.** Let $x_0 \in \text{Sing}(u^\lambda)$, then any maximizer $y_{t_1, x_0}$ of $v^\lambda(t_0, \cdot) - A^\lambda_{0, t_1}(x_0, \cdot)$ is contained in $\text{Sing}(u^\lambda)$ for all $t > 0$ and there exists $t_1 > 0$ such that the map $t \mapsto y_{t, x_0}$, the maximizers with respect to $v^\lambda(t_0, \cdot) - A^\lambda_{0, t}(x_0, \cdot)$ for $0 < t < t_1$, is continuous. Moreover, the right derivative $\frac{d}{dt} y_{t_0, x_0} |_{t=0^+}$ exists and it is equal to $q^\lambda_{x_0}$ as in Theorem 3.4, i.e., $q^\lambda_{x_0}$ is the unique element in $D^+ v^\lambda(t_0, x_0)$ such that
$$H(x_0, q^\lambda_{x_0}) = \min_{p \in D^+ v^\lambda(t_0, x_0)} H(x_0, p).$$
(3.11)
**Proof.** Fix $t_0 \in \mathbb{R}$, then $(t_0, x_0) \in \text{Sing}(v^\lambda)$ since $x_0 \in \text{Sing}(u^\lambda)$. For any $t > 0$ and $y_{t, x_0} \in \text{arg max} \{v^\lambda(t_0, \cdot) - A^\lambda_{0, t}(x_0, \cdot)\}$ (which is nonempty since $v^\lambda(t_0, \cdot)$ is Lipschitz and Lemma 3.1), suppose $y_{t, x_0}$ is a differentiable point of $v^\lambda(t_0, \cdot)$. Thus
$$0 \in D^+ \{v^\lambda(t_0, \cdot) - A^\lambda_{0, t}(x_0, \cdot)\}(y_{t, x_0}) = D_y v^\lambda(t_0, y_{t, x_0}) - D^- \{A^\lambda_{0, t}(x_0, \cdot)\}(y_{t, x_0}),$$
equivalently, $D_y v^\lambda(t_0, y_{t, x_0}) \in D^- \{A^\lambda_{0, t}(x_0, \cdot)\}(y_{t, x_0})$. It follows that $A^\lambda_{0, t}(x_0, \cdot)$ is differentiable at $y_{t, x_0}$ and
$$p_{t, x_0} = D_y v^\lambda(t_0, y_{t, x_0}) = D_y A^\lambda_{0, t}(x_0, y_{t, x_0}).$$
since $A^\lambda_0(x_0,\cdot)$ is locally semiconcave (see, for instance, \[12\]). Therefore, there exists two $C^2$ curves $\xi_{t,x_0} : [0, t] \to \mathbb{R}^n$ and $\gamma_{x_0} : (-\infty, t] \to \mathbb{R}^n$ such that $\xi_{t,x_0}(0) = x_0$, $\gamma_{x_0}(t) = \gamma_{t,x_0}(t) = y_{t,x_0}$ and $p_{t,x_0} = L_v(\gamma_{x_0}(t), \dot{\gamma}_{x_0}(t)) = L_v(\xi_{t,x_0}(t), \dot{\xi}_{t,x_0}(t))$.
Since $\xi_{t,x_0}$ and $\gamma_{x_0}$ has the same endpoint condition at $t$, then they coincide on $[0, t]$. Thus, $x_0 = \gamma_{x_0}(0)$ and $(0, x_0)$ is a differentiable point of $v^\lambda$ since $\gamma_{x_0}$ is a backward calibrated curve by \[10\]. On the other hand, $(0, x_0)$ is contained in $\text{Sing}(v^\lambda)$ since $(0, x_0) \in \text{Sing}(v^\lambda)$. This leads to a contradiction.
To prove \[11\], we need a slight modification of Theorem 3.2. Notice that $v^\lambda(t, \cdot)$ is equi-Lipschitz and equi-semiconcave for $t \in [0, t_1]$. By the regularity properties of the fundamental solutions, $v^\lambda(t_0, \cdot) - A^\lambda_0(t_0, \cdot)$ is strictly concave for $t \in (0, t_2]$, where $t_2 \leq t_1$ is determined by Proposition A.4 and the semiconcavity of $u^\lambda$. Therefore, $t \mapsto y_{t,x_0}$ is a continuous selection since the function $v^\lambda(t_0, \cdot) - A^\lambda_0(t_0, \cdot)$ is continuous. By the same argument as in the proof of Theorem 3.2 we obtain that
$$\frac{d}{dt} y_{t,x_0} |_{t=0^+} = q^\lambda_{x_0}$$
with $q^\lambda_{x_0}$ satisfying \[11\].
\[APPENDIX A. REGULARITY PROPERTIES OF FUNDAMENTAL SOLUTIONS\]
Here we collect some relevant regularity results with respect to the fundamental solutions of \[2.2\] on $\mathbb{R}^n$. The proofs of these regularity results are similar to those in \[9\] in the time-independent case. The difference is that we need an extra condition (L3) to ensure the uniform Lipschitz estimate of the minimizers in the relevant Tonelli-like variational problem. We omit the proof.
**Proposition A.1.** Let $a \leq s < t \leq b$, $R > 0$ and suppose $L$ satisfies condition (L1)-(L3). Given any $x \in \mathbb{R}^n$ and $y \in B(x, R)$, let $\xi \in \Gamma^s_{\infty} x, y$ be a minimizer for $A_{s,t}(x, y)$ and let $p(\cdot)$ be the dual arc. Then we have that
$$\sup_{\tau \in [s, t]} |\dot{\xi}(\tau)| \leq \kappa_T(R/(t-s)), \quad \sup_{\tau \in [s, t]} |p(\tau)| \leq \kappa_T(R/(t-s))$$
and
$$\sup_{\tau \in [s, t]} |\xi(\tau) - x| \leq \kappa_T(R/(t-s)),$$
where $\kappa_T : (0, \infty) \to (0, \infty)$ is nondecreasing and $T = b - a$.
Fix $x \in \mathbb{R}^n$ and suppose $R > 0$ and $L$ is a Tonelli Lagrangian. For any $a \leq s < t \leq b$, $T = b - a$ and $y \in B(x, R)$, let $\xi \in \Gamma^s_{\infty} x, y$ be a minimizer for $A_{s,t}(x, y)$ and let $p$ be its dual arc. Then there exists a nondecreasing function $\kappa_T : (0, \infty) \to (0, \infty)$ such that
$$\sup_{\tau \in [s, t]} |\dot{\xi}(\tau)| \leq \kappa_T(R/(t-s)), \quad \sup_{\tau \in [s, t]} |p(\tau)| \leq \kappa_T(R/(t-s)),$$
by Proposition A.1. Now, $a < b$, $x \in \mathbb{R}^n$ and $\lambda > 0$ define compact sets
$$K_{a,b,x,\lambda} := [a, b] \times B(x, \kappa(4\lambda)) \times B(0, \kappa(4\lambda)) \subset \mathbb{R} \times \mathbb{R}^n \times \mathbb{R}^n,$$
$$K^*_{a,b,x,\lambda} := [a, b] \times B(x, \kappa(4\lambda)) \times \overline{B}(0, \kappa(4\lambda)) \subset \mathbb{R} \times \mathbb{R}^n \times (\mathbb{R}^n)^*.$$
Suppose there exists a constant $C_T = \text{Proposition A.4.}$
Then any minimizer $\xi \in \Gamma_{x,y+z}^{s,t+h}$ for $A_{s,t+h}(x,y+z)$ and corresponding dual arc $p$ satisfy the following inclusions
$$
\{(\tau, \xi(\tau), \dot{\xi}(\tau)) : \tau \in [s, t + h]\} \subset K_{s,s+1,x,\lambda},
$$
$$
\{(\tau, \xi(\tau), p(\tau)) : \tau \in [s, t + h]\} \subset K_{s,s+1,x,\lambda}^*.
$$
**Proposition A.3.** Suppose $L$ is a Tonelli Lagrangian. Then for any $\lambda > 0$ there exists a constant $C_\lambda > 0$ such that for any $x \in \mathbb{R}^n$, $s < t$ with $T = t - s < 2/3$, $y \in B(x, \lambda T)$, and $(h, z) \in \mathbb{R} \times \mathbb{R}^n$ satisfying $|h| < T/2$ and $|z| < \lambda T$ we have
$$
A_{s,t+h}(x, y + z) + A_{s,t-h}(x, y - z) - 2A_{s,t}(x, y) \leq \frac{C_\lambda}{T} (|h|^2 + |z|^2).
$$
Consequently, $(t, y) \mapsto A_{s,t}(x, y)$ is locally semiconcave in $(0,1) \times \mathbb{R}^n$, uniformly with respect to $x$ and $s$.
**Proposition A.4.** Suppose $L$ is a Tonelli Lagrangian and, for any $\lambda > 0$, there exists $T_\lambda > 0$ such that for any $x \in \mathbb{R}^n$, $s < t$, the function $(t, y) \mapsto A_{s,t}(x, y)$ is semiconvex on the cone
$$
S_\lambda(x, T_\lambda) := \{(t, y) \in \mathbb{R} \times \mathbb{R}^n : T = t - s < T_\lambda, |y - x| < \lambda T\},
$$
and there exists a constant $C''_{\lambda} > 0$ such that for all $(t, y) \in S_\lambda(x, T_\lambda)$, all $h \in [0, T/2)$, and all $z \in B(0, \lambda T)$ we have that
$$
A_{s,t+h}(x, y + z) + A_{s,t-h}(x, y - z) - 2A_{s,t}(x, y) \geq -\frac{C''_{\lambda}}{T} (h^2 + |z|^2).
$$
Moreover, there exists $T''_{\lambda} \in (0, T'_\lambda]$ and $C'''_{\lambda} > 0$ such that for all $T \in (0, T''_{\lambda})$ the function $A_{s,t}(x, \cdot)$ is uniformly convex on $B(x, \lambda T)$ and for all $y \in B(x, \lambda T)$ and $z \in B(0, \lambda T)$ we have that
$$
A_{s,t}(x, y + z) + A_{s,t}(x, y - z) - 2A_{s,t}(x, y) \geq \frac{C'''_{\lambda}}{T} |z|^2.
$$
**Proposition A.5.** Suppose $L$ is a Tonelli Lagrangian and, for any $\lambda > 0$, there exists $T'_{\lambda} > 0$ such that for any $x \in \mathbb{R}^n$ the functions $(t, y) \mapsto A_{s,t}(x, y)$ and $(t, y) \mapsto A_{s,t}(y, x)$ are of class $C^{1,1}_{\text{loc}}$ on the cone $S_\lambda(x, T'_{\lambda})$ defined in (A.2). Moreover, for all $(t, y) \in S(x, T'_{\lambda})$
$$
D_y A_{s,t}(x, y) = L_v(t, \xi(t), \dot{\xi}(t)),
$$
$$
D_x A_{s,t}(x, y) = -L_v(s, \xi(s), \dot{\xi}(s)),
$$
where $\xi \in \Gamma_{x,y}^{s,t}$ is the unique minimizer for $A_{s,t}(x, y)$.
**References**
[1] Attouch, H.; Azé, D., Approximation and regularization of arbitrary functions in Hilbert spaces by the Lasry-Lions method. Ann. Inst. H. Poincaré Anal. Non Linéaire 10 (1993), no. 3, 289–312.
[2] Bernard, P., Existence of $C^{1,1}$ critical sub-solutions of the Hamilton-Jacobi equation on compact manifolds. Ann. Sci. École Norm. Sup. (4) 40 (2007), no. 3, 445–452.
[3] Bernard, P., Lasry-Lions regularization and a lemma of Ilmanen. Rend. Semin. Mat. Univ. Padova 124 (2010), 221–229.
[4] Bernard, P., *The dynamics of pseudographs in convex Hamiltonian systems*. J. Amer. Math. Soc. 21 (2008), no. 3, 615–669.
[5] Bernard, P., *The Lax-Oleinik semi-group: a Hamiltonian point of view*. Proc. Roy. Soc. Edinburgh Sect. A 142 (2012), no. 6, 1131–1177.
[6] Cannarsa, P., Chen, C, Cheng, W., *Global generalized characteristics for Cauchy problems*, preprint, 2016.
[7] Cannarsa, P.; Cheng, W.; Zhang, Q., *Propagation of singularities for weak KAM solutions and barrier functions*. Comm. Math. Phys. 331 (2014), no. 1, 1–20.
[8] Cannarsa, P., Cheng, W., *Homoclinic orbits and critical points of barrier functions*, Nonlinearity 28 (2015), no. 6, 1823–1840.
[9] Cannarsa, P.; Cheng, W., *Generalized characteristics and Lax-Oleinik operators: global theory*. Calc. Var. Partial Differential Equations 56 (2017), no. 5, 56:125.
[10] Cannarsa, P.; Cheng, W.; Fathi, A. *On the topology of the set of singularities of a solution to the Hamilton-Jacobi equation*. C. R. Math. Acad. Sci. Paris 355 (2017), no. 2, 176–180.
[11] Cannarsa, P., Cheng, W.; Yan, J. *Herglotz’ generalized variational principle and contact type Hamilton-Jacobi equations*, preprint, 2017.
[12] Davini, A.; Fathi, A.; Iturriaga, R.; Zavidovique, M., *Convergence of the solutions of the discounted Hamilton-Jacobi equation*. Invent. Math. 206 (2016), no. 1, 29–55.
[13] Fathi, A., *Weak KAM theorem in Lagragian dynamics*, to be published by Cambridge University Press.
[14] Fathi, A.; Zavidovique, M., *Ilmanen’s lemma on insertion of \(C^{1,1}\) functions*. Rend. Semin. Mat. Univ. Padova 124 (2010), 203–219.
[15] Lasry, J.-M.; Lions, P.-L., *A remark on regularization in Hilbert spaces*. Israel J. Math. 55 (1986), no. 3, 257–266.
[16] McEneaney, W.M.; Dower, P.M., *The principle of least action and fundamental solutions of mass-spring and N-body two-point boundary value problems*. SIAM J. Control Optim. 53 (2015), no. 5, 2898–2933.
[17] Wang, K.; Wang, L.; Yan, J. *Implicit variational principle for contact Hamiltonian systems*. Nonlinearity 30 (2017), no. 2, 492–515.
[18] Zhao, K.; Cheng, W. *On the vanishing contact structure for viscosity solutions of contact type Hamilton-Jacobi equations I: Cauchy problem*, preprint, arXiv:1801.06088. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2321,
1
],
[
2321,
5378,
2
],
[
5378,
8892,
3
],
[
8892,
11747,
4
],
[
11747,
14552,
5
],
[
14552,
17949,
6
],
[
17949,
21109,
7
],
[
21109,
25075,
8
],
[
25075,
28525,
9
],
[
28525,
31600,
10
],
[
31600,
33853,
11
]
]
} |
73077f72ac141a3b68a5a42aed9967fec52ac4df | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 11,
"total-fallback-pages": 0,
"total-input-tokens": 32618,
"total-output-tokens": 13267,
"fieldofstudy": [
"Medicine"
]
} | Etiology and Pathophysiology of Tinnitus - A Systematic Review
Sujoy Kumar Makar
ABSTRACT
Introduction: Prevalence of tinnitus range from 7.1% to 14.6% (National Center for Health Statistics, 2016), but the mechanisms responsible for the development of this abnormal sensory state remain poorly understood.
Objectives: To determine the evidence for different etiologies and pathophysiology of tinnitus identified by clinical diagnostic tests in the adult population.
Study Design: Systematic literature review.
Methods: Review of data base using PRISMA guidelines: Google Scholar, Medline, Springer Link, Pubmed. In addition, manual reference search of identified papers. Randomized controlled trials, case control study, prospective cohort studies, and retrospective reviews of consecutive patients in which clear data were reported with respect to etiology and pathophysiology of tinnitus.
Results: Sixty seven articles met the inclusion criteria. The papers searched recent studies from 2004 to 2018 for different etiologies such as noise exposure, age, ototoxic drugs, hearing loss among patients with tinnitus. Multiple pathophysiology were identified, including inner ear pathology, auditory nerve synchronisation, central nervous system anomalies and limbic and autonomous nervous system problems. The group of papers evaluated tinnitus patients with specific diagnostic tests such as pure tone audiometry, Immitance audiometry, otoacoustic emission, Auditory brainstem response and diagnostic imaging of fMRI, MRI and PET study.
Conclusions: The results indicate a high level of heterogeneity between the studies for all the assessed areas. These results support the need for greater stratification of the tinnitus population and the importance of a standardized Puretone audiometry with extended high frequency, OAE, ABR and diagnostic imaging (fMRI, MRI & PET) method to make comparisons between studies possible. Diagnostic imaging is an important useful method for identification of intracranial pathology that can present with tinnitus as a primary symptom. Establishment of a direct causal link between tinnitus and these etiologies and pathophysiology remains elusive.
Keywords: Tinnitus; Loudness; Etiology; pathophysiology.
Department of Audiologist and Speech-Language Pathologist, Ali Yavar Jung National Institute of Speech and Hearing, India
*Send correspondence to:
Sujoy Kumar Makar
Department of Audiologist and Speech-Language Pathologist, Ali Yavar Jung National Institute of Speech and Hearing, India. E-mail: [email protected], Phone: +91302040753
Paper submitted on February 18, 2021; and Accepted on April 24, 2021
INTRODUCTION
Tinnitus is the perception of a continuous or intermittent sound in the absence of external acoustic stimulation. According to Bhatt, Lin and Bhattacharyya, prevalence of tinnitus in the United States is approximately 1 in 10 adults. The overall prevalence of tinnitus was 13.5% in patients aged below 50 years and 34.4% in patients aged above 67 years. The incidence was 27.8% in Sweden as reported by (2017). Several risk factors for the development of tinnitus are noise (19.6%), otoxicity (16.8%), presbycusis (16.3%) and increasing age (16.3%) as reported (2014). It is also possible to have severe tinnitus with no evidence of any aural pathology. Further, tinnitus sufferers want to know how their tinnitus is generated and whether it is curable. However, till date the literature is not able to fully explain about the pathophysiology of subjective tinnitus and cannot assure patients about the prognosis of tinnitus. It is essential to have evidence based on its underlying aetiology and pathophysiology to treat tinnitus effectively; once the aetiology and pathophysiology is known, the disorder can be treated. However, in case of subjective tinnitus, it is difficult to identify a single origin of tinnitus. Hence it is difficult to treat it completely till date. Tinnitus is of interest to audiologists because it comes under their professional domain and it creates problems in human health. Therefore, it is necessary for the audiologists to understand the evidence based pathophysiology of tinnitus to explain changes of brain activity in tinnitus patients for better management and prevention of tinnitus. Review of literature is an integral part of research. This involves identification and analysis of documents containing information related to the research problem, with the purpose of providing context for the research and its justification, identifying the areas which have been already covered as well as the research gaps. Studied a systematic review to assess the scientific evidence on the associations between symptoms of depression and tinnitus. A systematic review of tinnitus prevalence and severity. However, there is a need to study how much high-level evidence exists for the aetiology and complex pathophysiology of tinnitus. To that end, this article provides a broad-based review of what is presently known about aetiology, involvement of cochlea, auditory nerve, auditory cortex and somatosensory systems in tinnitus patients and its clinical implications. The summary of fundamental information has relevance to both clinical and research arenas. The research question is “Does high level evidence exist to support aetiology and pathophysiology of tinnitus?” The strength of present systematic review search examines etiology and pathophysiology of tinnitus as well as validated measures to find out aetiology in the auditory pathway.
MATERIALS AND METHODS
A systematic review of the literature was performed including searches by using PRISMA guidelines the following electronic data base: Google Scholar, Medline, Springer Link, Pubmed. The key words were: Causes of tinnitus, tinnitus, cochlear tinnitus, neural tinnitus, origin of tinnitus, neurophysiological and psycho-physical dimension of tinnitus, origin of tinnitus, cortical tinnitus, somatosensory tinnitus, neural plasticity changes in tinnitus patients, epidemiologic studies, risk factors, cohort, case control, randomized controlled trial or controlled clinical, etc. Articles were reviewed using a prior determined selection criterion. Inclusion criteria included: randomized controlled trials, prospective cohort studies, retrospective reviews. Exclusion criteria included: prospective or retrospective study, comments, practice guidelines, editorials letters, book chapters, except that references were reviewed. This systematic review is a humble attempt to identify the origin and pathophysiology of tinnitus. Accessed 425 publications, and dotted down to 57 original papers and 10 review papers identifying the latest procedures used to cite the origin and pathophysiology. Studies were selected that used PTA, IA, OAE, ABR, (f MRI), (r CBF), PET, MEG for investigation. The review gives importance to neurophysiological and psycho-physical dimension of tinnitus. Thus the composition of the processes involved in tinnitus is included in the review. Journal titles were independently reviewed, and articles were included if researcher felt they were relevant. Selected abstracts were then reviewed using the predetermined inclusion and exclusion criteria. Articles that were considered relevant or uncertain relevance were retrieved as full text articles. The full text papers were reviewed and data extraction performed independently.
RESULTS
Out of the 67 studies that met the search criteria, 54 were research studies, 10 were review studies and 3 case study report (Figure 1). A descriptive summary including the type of study, research design, sample size, measurement and result are provided in Tables 1 and 2. Several combined objective studies suggest that dysfunction starts in the cochlea and then a weak imbalance of neural activity is generated in the central pathway; this is noticed at low signal level in the auditory systems and being a new signal it is enhanced by sub-cortical centres, transferred to the auditory cortex and perceived as an abnormal sound tinnitus. Longer duration involvement of auditory system in tinnitus patients affect limbic system and autonomic nervous system also. Thus this evidence supports the viewpoint that in longer duration tinnitus there is involvement of whole brainstem; also that multiple feature tinnitus are a result of abnormal activity within the central auditory pathway. These abnormalities which are due to abnormal activity in IC, CN, MSOC and brainstem, lead to changes in tonotopic organisation of auditory maps. Thus original abnormalities trigger secondary abnormalities and this could explain why pitch, loudness and RI of tinnitus change in longer duration tinnitus. Cochlear-type tinnitus is suggested to result from aberrant activity generated at the cochlear nerve level. The outer hair cells regulate the endocochlear potential that contribute to enhancement of cochlear spontaneous
activity. A reduction in cochlear activity produces tinnitus-related plastic changes, namely cortical reorganisation, thalamic neuron hyperpolarisation, facilitation of non-auditory limbic inputs and increase in central gain. These central changes are associated with abnormal patterns of spontaneous activity in the auditory pathway, i.e. hypersynchrony activity. The somatosensory system, and the limbic and autonomic nervous systems are also involved in tinnitus generation/manipulation 26-30.
Etiology and Hearing loss: Sensorineural hearing loss is commonly accompanied with tinnitus. Some researchers believe that subjective tinnitus cannot exist without hearing loss (American Tinnitus Association, 2019). Subjective tinnitus has no identifiable cause other than hearing loss (ASHA, 2019). Even individuals with tinnitus and normal hearing show a significant hearing loss at extended higher frequency 10,000 to 20,000 Hz 31. Reported that there was significant differences of high frequency threshold between tinnitus ear and non-tinnitus ear (P < 0.01); also significant differences of high frequency threshold between tinnitus ear and non-tinnitus ear in each group (P < 0.01). 32 suggested that in patients with unilateral tinnitus, hearing threshold (0.125-8 kHz) of tinnitus ear and contralateral ear difference was not statistically significant (P < 0.05), but in extended high frequency (> 10 kHz) the difference between two ears was statistically significant (P > 0.05). Shim et al (2009) reported that patients with tinnitus who have normal hearing below 8 kHz have decreased hearing ability at extended high-frequencies at 10 kHz, 12 kHz, 14 kHz, and 16 kHz. statistically significant differences (p<0.01) were found between the determination of the frequency of tinnitus made with conventional and high-frequency audiometers, as well as a correlation between high-frequency tinnitus and distress expressed by patients. 33-35 reported that the frequency of tinnitus is an important factor in tinnitus group: between 8,000 Hz to 20,000 Hz, with the increases in frequency, hearing loss also increase. Many studies have represented tinnitus as a threshold phenomenon for which any one factor, such as chronic progressive hearing loss is insufficient to elicit its emergence--two or more trigger factors (i.e., psychosocial stress, noise exposure, and somatic factors) can act synergistically to produce symptomatic tinnitus 36-40.
Noise: According to ASHA (2019) loud noise exposure might cause tinnitus (ASHA, 2019). Population-based data indicate that excessive noise exposure represents the second most common cause of tinnitus 41-43 offer a review of several studies suggested that noise trauma is the single most unique cause of tinnitus (18%), followed by head and neck trauma (8%), whereas drugs (most often salicylate) only account for 2% of known incidents of tinnitus. Small temporary changes in the outer hair cells (OHCs) following noise exposure can also trigger the emergence of tinnitus by increasing the gain of the central auditory system 44. Noise induced hearing loss (NIHL) is most prevalent cause of tinnitus 45. Many environmental factors can cause tinnitus, mostly related to the effect of noise on the auditory system and subsequent damage to the microstructures in the cochlea 46-48. According to American Tinnitus Association 2019, exposure to loud noises in a single traumatic experience or over time can
---
**Figure 1:** Systematic literature review using PRISMA guidelines step by step.
### Table 1: Summary of Etiology and Pathphysiology of Tinnitus studies utilizing a diagnostic algorithm.
| Etiology / pathophysiology | Etiology | Research design | Samples size | Test performed | Results |
|-----------------------------|----------|-----------------|--------------|----------------|---------|
| Noise and other causes | Bhatt, Lin, & Bhattacharya (2016) | Cross-sectional analysis | n= 75 764 | Survey questionnaire, Tinnitus questionnaire | Loud noises at work (odds ratio, 3.3; 95% CI, 2.9-3.7) and recreational noise (odds ratio, 2.6; 95% CI, 2.3-2.9). Years of work-related noise exposure correlated with increasing prevalence of tinnitus (r = 0.13; 95% CI, 0.10-0.16). |
| | Samarei & Fatholahi (2004) | Cross-sectional | n=184 | PTA, OE, questionnaire | Most common causes of tinnitus were noise (19.6%), ototoxicity (16.8%) and presbycusis (16.3%). |
| | Kujawa & Liberman (2009). | Experimental study | N=128 (mice) | ABR, ECOGG, DPOAE, cochlear mapping software | Damage to the hair cells has progressive consequences that are considerably more widespread than are revealed by conventional threshold testing. This primary neurodegeneration should add to difficulties in hearing in noisy environments, and could contribute to tinnitus, commonly associated with inner ear damage. |
| | Noreña & Eggermont (2006) | Experimental study | Animal study (cat) | | Noise-induced hearing loss induces reorganization of the tonotopic map in auditory cortex and increases spontaneous firing rate and neural synchrony. This is interpreted as an absence of putative neural signs of tinnitus. |
| Hearing loss and tinnitus | López-González et al., (2012) | Purposive sampling | n=47 | PTA with extended high frequency | Statistically significant differences reported between the determination of the frequency of tinnitus made with conventional and high-frequency audiometers, as well as a correlation between high-frequency tinnitus and distress expressed by patients. |
| | Tang, Ji & Liu (2011) | Case control study | n=200 | ENT, IA, PTA with high frequency | In patients with unilateral tinnitus, the difference in hearing threshold of tinnitus ear and contralateral ear (0.125 to 8 kHz), was statistically significant, but in extended high frequency (> 10 kHz), the difference between two ears was not statistically significant. |
| | Yildirim, Berkiten, Kuzdere & Ugras (2010) | Case control study | n=154 | PTA, IA, OE, tinnitus pitch and loudness matching. | There were significant hearing loss related to age and frequency from 8,000-20,000 Hz of the patients with normal hearing in 250-4,000 Hz frequency range. |
| | Shargorodsky, Curhan & Farwell (2010) | Cross-sectional study | n=14,178 | Survey questionnaire | The frequency between 8,000 Hz to 20,000 Hz is an important factor in tinnitus group. As the frequency increases, hearing loss increases. |
| | Shim et al., (2009) | Case control study | n=510 | High frequency PTA, ABR, tinnitus pitch & loudness | The significant association between tinnitus and age (p <.01), (60 to 69 years), smoking (p <.01), hypertension (p <.01), diabetes mellitus (p <.01) suggest that vascular disease might have a greater contribution to the etiology of tinnitus. Patients with tinnitus who have normal hearing below 8 kHz have decreased hearing ability at extended high-frequencies at 10 kHz, 12 kHz, 14 kHz, and 16 kHz. |
Cai & Tang, (2004) Case control study n=78 PTA, IA, tinnitus pitch and loudness
Sindhusake et al., (2003) Cross-sectional study n=2145 PTA, TEOAE, SOAE
Cochlea Makar, Mukundan and Gore, (2017) Case-control study n=60 PTA, IA, ABR, DPOAE, Tinnitus, pitch and loudness
Table 2: Charting the data of systematic review for studies.
| Authors and Year | Title | Methods | Results |
|------------------|-------|---------|---------|
| Henry, Roberts, Caspary, Theodoroff & Salvi (2014) | Underlying Mechanisms of Tinnitus: Review & Clinical Implications | Review | Tinnitus is a pathology involving neuroplastic changes in central auditory structures that take place when brain is deprived of its normal input by pathology in the cochlea. Cochlear damage leads to a reorganization of the pathways in the central auditory system. Reduction in auditory nerve input leads to disinhibition of the DCN and an increase in spontaneous activity in the central auditory system, which is experienced as tinnitus. Loss of input in the lower lesion frequency range in cochlea leads to an over representation of lesion-edge frequencies, which causes hyperactivity and possible burst-firing in central auditory pathways, constituting the initial tinnitus signal. Under normal circumstances, the tinnitus signal is cancelled out at the level of the thalamus by an inhibitory feedback loop originating in paralimbic structures. If the limbic regions are compromised, this "noise-cancellation" mechanism breaks down, and chronic tinnitus results. Tinnitus generators are theoretically located in the auditory pathway, and such generators and various mechanisms occurring in the peripheral auditory system and central auditory system have been explained in terms of the auditory plasticity theory, the crosstalk theory, the somatosensory, limbic and autonomic nervous systems. Cochlear-type tinnitus is suggested to result from aberrant activity generated at the cochlear nerve level. The outer hair cells regulate the endocochlear potential that contribute to enhancement of cochlear spontaneous activity. A reduction in cochlear activity leads to tinnitus-related plastic changes, namely cortical reorganisation, thalamic neuron hyperpolarisation, facilitation of non-auditory inputs and/or increase in central gain. These central changes are associated with abnormal patterns of spontaneous activity in the auditory pathway, i.e. hypersynchrony activity. The neuroimaging methods fMRI and PET measure signals that presumably reflect the firing rates of multiple neurons and are assumed to be sensitive to changes in the level of neural activity. The general trend emerging from the neuroimaging studies, is that tinnitus in humans may correspond to enhanced neural activity across several centers of the central auditory system. Non-auditory areas including the frontal areas, the limbic system and the cerebellum seems associated with the perception of tinnitus. ABR findings suggest that the longer latency and reduced amplitude of wave I for the tinnitus group with normal hearing compared to matched controls was the most consistent finding across studies. Tinnitus stemming from imbalances in the excitatory and inhibitory inputs to auditory neurons. Such changes occur at multiple levels of the auditory system and involve a combination of interacting phenomena that are triggered by loss of normal input from the inner ear. This loss sets in motion a number of plastic readjustments in the higher level central auditory system. |
| Lockwood, Salvi & Burkad (2002) | Tuning Out the Noise: Limbic-Auditory Interactions in Tinnitus | Review | |
| Rauschecker, J. P., Leaver, A. M., & Mühlau, M. (2010). | Tinnitus: Characteristics, Causes, Mechanisms, and Treatments | Review | |
| Han, Lee, Kim, Lim, & Shin (2009) | Revisiting the Cochlear and Central Mechanisms of Tinnitus and Therapeutic Approaches | Review | |
| Noreña (2015) | Neural activity underlying tinnitus generation: Results from PET and fMRI | Review | |
| Lanting, De Kleine, Van Dijik (2009) | Auditory Brainstem Responses in Tinnitus: A Review of Who, How, and What? | Review | |
| Milloy, Fournier, Benoit, Noreña, & Koravand (2017). | Tinnitus: Models and mechanisms | Review | |
| Kaltenbach (2011). | | | |
damage the auditory system and result in hearing loss and sometimes, tinnitus as well.
**Age:** Any pathologic lesion in the auditory pathway or any reduction in auditory nerve function due to ageing has the potential to produce tinnitus. Bilateral subjective tinnitus can be associated with presbycusis. The significant association between tinnitus and old age (60 to 69 years) also suggest that vascular disease might have a greater contribution to the etiology of tinnitus. Net down regulation of functional inhibition may result from production of plastic maladaptive compensatory changes due to partial deafferentation of the central auditory system caused due to aging. In old age hearing often deteriorates, typically starting around the age of 60. This form of hearing loss tends to be in both ears and involves the sensory loss of high-frequency sounds. Age-related hearing loss explains, in part, why tinnitus is so prevalent among seniors (American Tinnitus Association, 2019).
**Psychological Status:** Excessive stress might cause tinnitus according to ASHA (2019) about 75% of new cases reported to tinnitus clinics are related to emotional stress as the trigger factor. A Study on the risk factors for developing tinnitus reported in population based studies including psychological status of the affected individuals, suggested there are clear association between the psychological state of the individual and tinnitus; in particular, the condition is more often experienced by depressive patients (19% vs. 39.5% of depressive patients). Tinnitus is more often associated with hearing disorders (20% vs. 30–37%). These data are based on a study including over 14,000 respondents with an average tinnitus prevalence of 25.3% in the individual subgroups. The significant association between tinnitus and hypertension (p<.01) suggest that vascular disease might have a greater contribution to the etiology of tinnitus. Reported the changes in the levels of neurosteroids in the central nervous system associated with depression could be a leading cause of tinnitus. In their conclusion, American Tinnitus Association (2019) suggested psychiatric disorders such as depression, anxiety and stress are trigger factors for tinnitus.
**Ototoxicity:** Bilateral subjective tinnitus can be associated with ototoxicity, side-effect of some oral medications, such as salicylates, nonsteroidal anti-inflammatory drugs, aminoglycoside antibiotics, loop diuretics, and chemotherapy agents. Tinnitus is a potential short-lived side-effect of many prescription medications and if the patient stops taking the medication, the tinnitus symptoms typically recede. There are some ototoxic drugs known to cause more permanent tinnitus symptoms, such as, non-Steroidal Anti-Inflammatory Drugs (NSAIDs), certain antibiotics, cancer medications, Water pills and diuretics, Quinine-based medications (American Tinnitus Association, 2019).
**Medical problem:** Neurologic causes include head injury, multiple sclerosis, vestibular schwannoma and cerebellopontine-angle tumors. Infectious causes include otitis media, meningitis, syphilis, and other infections that affect hearing. Tinnitus is a symptom of medical conditions such as, metabolic disorders, hypothyroidism, anemia, autoimmune disorders, lyme disease, fibromyalgia, blood vessel disorders, high blood pressure, atherosclerosis; traumatic brain injury caused by concussive shock, could damage the brain’s auditory processing areas and generate tinnitus symptoms. Vestibular disorders such as acoustic neuroma, vestibular schwannoma and other tumorous growths (American Tinnitus Association, 2019) are associated with tinnitus. Recently ASHA (2019) reported that Ménière’s disease, migraine, head injury, drugs or medicines that are toxic could also be linked to tinnitus. Shargorodsky, reported a significant association between tinnitus and diabetes mellitus (p<01) suggestive of vascular disease having a greater contribution to the cause of tinnitus.
**Somatosensory cause:** Several researchers attempted to study the connection between auditory and somatosensory system, and reported modification in the loudness and pitch of tinnitus via somatic maneuvers such as jaw clenching or tensing their neck muscles both the firing rates and temporal response patterns to the sound can be modulated by trigeminal stimulation preceded by an acoustic stimulus. This bimodal integration is replicated in neurons of the IC and this receives converging inputs from both the DCN and somatosensory nuclei. Another example of somatic tinnitus is that damage to the muscles, ligaments, or cartilage in the temper mandibular joint disorder (TMJ), which shares some ligaments and nerve connections with middle ear. In many scenarios, fixing the TMJ disorder will alleviate tinnitus symptoms (American Tinnitus Association, 2019).
**PATHOPHYSIOLOGY**
The pathophysicsiology of subjective tinnitus is poorly understood so are the neuroplastic changes in central auditory structures that take place when brain is deprived of its normal input by pathology in the cochlea.
**Cochlear pathophysiology:** According to Job, there are evidence of cochlear outer hair cell dysfunctions in participants susceptible to tinnitus due to noise. In almost all situations OHCs are damaged more than IHCs, which results in the disinhibition of neurons in the dorsal cochlear nuclei (DCNs). Therefore, there will be an area within organ of corti where OHCs are affected but IHCs are intact. This would affect coupling between the tectorial membrane and the basilar membrane, to the extent that the tectorial membrane might directly impinge upon the cilia of the IHCs, thus causing them to depolarise. The role that increased neural activity in the auditory periphery may have in tinnitus generation can be explained. When OHCs damage and IHC normal functioning, cells in the dorsal cochlear nuclei (DCNs) show increased spontaneous activity because of IHC normal input but decrease OHCs input. This spontaneous activity is perceived as tinnitus. The OHCs normally recover within a few days, but this can be delayed for up to a few months. Therefore, it is
hypothesized that tinnitus represents a consequence of a central gain adaptation mechanism when the auditory system is confronted with a hearing loss (Two types of tinnitus have been identified; tonal and complex. Tonal tinnitus results from discordant dysfunction of OHCs and IHCs manifesting in a single area, whereas complex tinnitus results from multiple areas of discordance (Jastreboff, 2004). However, when patients clearly have the central type of tinnitus, such as after transaction of the auditory nerve, the OHC concept is not applicable and alternative mechanisms need to be considered.
Auditory and Vestibular Nerve pathophysiology: studied 75 patients age ranging from 20 to 45 years, to evaluate electrophysiologically the auditory nerve and the auditory brainstem function of patients with tinnitus and normal-hearing thresholds using the auditory brainstem response (ABR). Abnormal results were found in 43% in at least 1 of the 8 parameters evaluated. Although within normal limits, the tinnitus group presented with a significant prolongation of the latencies of wave I, III, and V. The interpeak I-III, and I-V values were within the normal limits, whereas the interpeak III-V value was significantly (p=.003) enlarged in the experimental group. The V/I amplitude was within normal limits; however, a significant (p=.004) difference was present between the two groups. The interaural latency difference of wave V did not show significant differences. The study suggests even though most parameters were within normal limits that there are changes in the central pathways in the tinnitus group. Schaette and McAlpine (2011) reported tinnitus is triggered by cochlear damage, however, many tinnitus patient’s audiogram are normal. They explain, in tinnitus patients with normal audiogram, ABR show significantly reduced amplitude of the wave I potential but normal amplitudes of the more centrally generated wave V. This is direct evidence of “hidden hearing loss” that manifests as reduced neural output from the cochlea and consequent renormalization of neuronal response magnitude within the brainstem included 43 unilateral tinnitus patients (19 males, 24 females) and 18 control participants with normal hearing thresholds in their study. The amplitudes of wave I and V were measured at 90 dB nHL and UCLs at 500 Hz and 3000 Hz pure tones in each TE and NTE were assessed. The within-patients comparison between TEs and NTEs showed no significant differences in wave I and wave V amplitude but individual data revealed increased V/I amplitude ratios (mean± 2 SD) in 3 TEs in the experimental ears. No significant differences in UCL at 500 Hz and 3000 Hz between the TEs and NTEs were found but were lower than that of control group. ABR results do not represent meaningful evidence, however, reduced sound level tolerance in both TEs and NTEs might replicate increased central gain following hidden synaptopathy that was subsequently balanced between the ears by lateral olivocochlear efferents. Conducted a study with twenty six participants the aim to evaluate existence of any association between tinnitus loudness/onset duration and audiological profile to explain differences in prognosis. The significant differences were found in extended high frequency for pure-tone audiometry hearing thresholds and tinnitus loudness/onset duration into tinnitus and nontinnitus ears. These are associated with an increase in tinnitus loudness and its onset duration. Nerve compression may cause artificial synapses to be formed between nerve fibres of the cochlear and vestibular nerve (crosstalk), this may occur when auditory nerve fibers are intact and some other cranial nerves are damaged. This may result in the phase-locking of the spontaneous activity of groups of auditory neurons. The breakdown of the myelin insulation of the nerve fibers may further enhance coupling. In the absence of any external sounds, it creates a neural pattern that resembles the patterns evoked by actual sounds. These cranial nerves are sensitive to compression at the root entry zone, where they are covered by myelin. This notion is applied to the cochlear-vestibular nerve, which is myelinated, and is vulnerable to compression from blood vessels or tumors impinging upon the nerve (e.g., vestibular schwannoma). This might lead to tinnitus if synchronization of the stochastic firing in the cochlear nerve is perceived as sound, as well as to cross talk synapses and tinnitus development, a process seen with vestibular schwannoma or neurovascular conflict.
Dorsal Cochlear Nucleus pathophysiology: The Dorsal Cochlear Nucleus (DCN) has been implicated as a possible site for the generation of tinnitus-related signals owing to its tendency to become hyperactive following exposure to tinnitus-inducing agents such as intense sound and cisplatin. However, OHC damage triggers plastic readjustments in the DCN, resulting in DCN fusiform cells become spontaneously hyperactive a reduction in auditory nerve input leads to elevated DCN fusiform output that act as a trigger for tinnitus-related neural activity rostral to the cochlear nucleus and an increase in spontaneous activity in the central auditory system, which manifests as tinnitus. This mechanism could explain the temporary ringing sensation that can follow exposure to loud sound the plastic readjustments in the dorsal cochlear nucleus are slow and lead to tinnitus with a delayed onset. IHC damage prevents hyperactivity in the DCN, damage to the cochlea enhances neural activity in the central auditory pathway. Auditory plasticity emerges as a consequence of the aberrant pathway, and tinnitus might be considered as an auditory system analog to phantom limb sensations in amputees studied auditory nerve and brainstem function in response to sound assessed via auditory brainstem responses (ABR) in humans with and without tinnitus. Tinnitus participants showed reduced wave I amplitude but enhanced wave V suggestive of reduced auditory nerve activity and elevated input to the inferior follicular, compared with non-tinnitus participants. It was concluded that the elevated III/I and V/I amplitude ratios in tinnitus participants reflect disproportionately high activity in the spherical bushy cells (SBC) pathway for a given amount of peripheral input. The results imply a role for the VCN in tinnitus and suggest the SBC pathway as a target for tinnitus treatment.
Auditory cortex pathophysiology: it reported that tinnitus might be generated in the temporal lobe in the auditory association cortex and inferior colliculus on low-frequency fluctuations of fMRI confirms that chronic tinnitus patients have aberrant significant increased spontaneous neuronal activity within the right middle temporal gyrus (MTG), right superior frontal gyrus (SFG) and right angular gyrus. On the other hand, decreased spontaneous neuronal activity was detected in the right middle occipital gyrus and bilateral thalamus. The tinnitus duration (longer vs shorter) correlated positively with higher spontaneous neuronal activity values in right superior frontal gyrus (SFG). Using magnetic source imaging test (MRI), a marked shift of the auditory cortical representation of the tinnitus frequency into an area adjacent to the expected tonotopic location was observed. Also a strong positive correlation was found between the subjective loudness of the tinnitus and the amount of cortical reorganization (r = 0.82, p < 0.01). Another reported that frequency region corresponding to the tinnitus pitch is known to be abnormally represented in auditory cortex. This appears to be correlated with the perceived loudness of tinnitus but not with the amount of hearing loss, which is the primary determinant of changes in tonotopic maps. These results strongly demonstrate that tinnitus is related to plastic alterations in auditory cortex.
Somatosensory pathophysiology: the activity of the Dorsal Cochlear Nucleus (DCN) is also influenced by stimulation of nonauditory structures. However, the somatosensory system is the only nonauditory sensory system that appears to be related to tinnitus (e.g., temporomandibular-joint). Somatic tinnitus can develop from activation of underlying oto-somatic interaction (Levin, 2004) and is caused by disinhibition of the ipsilateral DCN; this is mediated by nerve fibers whose cell bodies lie in the ipsilateral medullary somatosensory nuclei. These neurons receive inputs from the nearby trigeminal tract, the facial, vagal, and glossopharyngeal nerve fibers innervating the middle and external ears. Thus activity of the DCN is influenced by stimulation of somatosensory system (nonauditory structures) that has been also implicated in development of tinnitus. There is a correlation with limbic structures that has been clearly documented with anxiety, depression and negative psychological state and increased tinnitus. The ability of some individuals to modulate tinnitus by performing voluntary somatosensory or motor actions (forceful head and neck contractions) is probably attributable to somato-sensory-auditory interactions within the central nervous system of these patients which account for somatic modulation of tinnitus. It appears that etiology of tinnitus is linked to some factors including hearing loss, noise, age, psychological status, ototoxicity, medical problem such as vestibular schwannoma, multiple sclerosis, somatosensory damage. The pathophysiology of tinnitus is associated with cochlea, auditory and vestibular nerve, dorsal cochlear nucleus, auditory cortex, pathology in limbic and autonomous nervous system. This systematic review is an attempt to find evidence to support these causes and pathophysiology for further tinnitus management. The studies reported differ based on sample size, study design, selection of control and experimental group, methodology, and instruments used. The authors explore the differences in studies to explain how evidence is stronger to accept the different causes and pathology underlying tinnitus generation and that influence it. According to, systematic reviews formulate research questions that are broad in scope, and identify and synthesize studies that directly relate to the systematic review question. It uses PRISMA guidelines to collect secondary data, and combine findings qualitatively or quantitatively. Systematic review studies can be rejected based on research methodology and adequate sample size is important in an effective studies otherwise, small sample sizes may limit generalization of the findings. In this systematic review, sample size was wide ranging. Twelve studies had between 9 to 50 participants, three studies had 50 to 100 participants, three studies had 150 to 450, and one study had 14178 participants while only one study had 75764 participants. Randomized control trial (RCT), meta-analysis, cohort study, case control study, case series are important for evaluation of clinical findings. There were only three randomized
DISCUSSION
It appears that etiology of tinnitus is linked to several factors including hearing loss, noise, age, psychological status, ototoxicity, medical problem such as vestibular schwannoma, multiple sclerosis, somatosensory damage. The pathophysiology of tinnitus is associated with cochlea, auditory and vestibular nerve, dorsal cochlear nucleus, auditory cortex, pathology in limbic and autonomous nervous system. This systematic review is an attempt to find evidence to support these causes and pathophysiology for further tinnitus management. The studies reported differ based on sample size, study design, selection of control and experimental group, methodology, and instruments used. The authors explore the differences in studies to explain how evidence is stronger to accept the different causes and pathology underlying tinnitus generation and that influence it. According to, systematic reviews formulate research questions that are broad in scope, and identify and synthesize studies that directly relate to the systematic review question. It uses PRISMA guidelines to collect secondary data, and combine findings qualitatively or quantitatively. Systematic review studies can be rejected based on research methodology and adequate sample size is important in an effective studies otherwise, small sample sizes may limit generalization of the findings. In this systematic review, sample size was wide ranging. Twelve studies had between 9 to 50 participants, three studies had 50 to 100 participants, three studies had 150 to 450, and one study had 14178 participants while only one study had 75764 participants. Randomized control trial (RCT), meta-analysis, cohort study, case control study, case series are important for evaluation of clinical findings. There were only three randomized
control trials, two experimental studies, twelve cross section analytic trials and five case control studies. In this review, the most widely used tests for finding aetiology and pathophysiology were PTA, IA, SRT, SDS, TEOAE, DPOAE, ABR, MLR, TEN, fMRI, MRI, PET, 40Hz ASSR, MEG, 148 channel magnetometer (4D- neuroimaging) rTMS, PET and tinnitus pitch and loudness measurement. Therefore, based on these findings it can be drawn that a reduction in cochlear activity due to aetiology such as noise, age and hearing loss, leads to changes in the neural activity across several areas of the central auditory system and non-auditory canters including limbic system and the cerebellum associated with the perception of tinnitus. Thus for assessment of tinnitus PTA, IA, OAE, ABR, fMRI and PET tests are required. In a systematic review, level of evidence is an important parameter to explore strength of the study. It was possible to rate level of evidence for 30 studies in this systematic review. To interpret the level of evidence, guidelines developed by Ackley, Swan, Ladwig, and Tucker (2008) were used. It is clear that only three studies have level I- meta analysis of randomised control trial, 25 studies have level II- well-designed controlled trials and three studies have level IV- well-designed case-control or cohort studies to find out the aetiology and pathophysiology of tinnitus. Also these findings suggest that multidisciplinary professionals such as ENT, audiologist, radiologist and psychologist are required for a combined approach for better tinnitus assessment in future. It is possible to bring out standard normative clinical practice guidelines for tinnitus assessment based on systematic reviews of more RCTs of good quality that have similar results. This systematic review explores the many etiologic causes of tinnitus that have been proposed; but for the majority of patients with tinnitus, the etiology remains idiopathic. An extensive range of disorders can cause damage to the auditory system, potentially leading to the development of tinnitus. The major causative factors are: 1) viral infection; 2) vascular impairment; 3) immune-mediated mechanisms; 4) inner ear abnormality; and (5) CNS abnormalities, including tumors, trauma, hemorrhage, infarction, and other pathologies. The most common suspected etiologies of tinnitus in adult patients is idiopathic (71.0%), infections (12.8%), otologic disease (4.7%), trauma (4.2%), vascular or hematologic (2.8%), neoplastic (2.3%), and other causes (2.2%) as potential suspected etiologies. Multiple pathophysiology were identified, including OHC and IHC pathology, auditory nerve synchronisation central nervous system anomalies and limbic and autonomous nervous system (problems). Establishment of a direct causal link between tinnitus and these etiologies remains elusive. Diagnostic imaging (fMRI, PET) is a useful method for identification of temporal bone or intracranial pathology that can present with tinnitus as a primary symptom.
CONCLUSION
A detailed case history and physical examination of patients with tinnitus may identify some causes, for example, trauma, cerebrovascular accident and ear surgery. MRI scanning of the internal auditory canal and cerebellopontine angle is required in patients suffering from tinnitus where the cause cannot be identified by case history or physical examination in order to rule out cases of vestibular schwannoma. Evidence suggests that tinnitus involves neuroplastic changes in central auditory structures which takes place when the brain is deprived of its normal input due to cochlear pathology. Cochlear pathology may not be identified in audiograms in some cases but can be detected with the help of OAE or ABR measures. Neural changes may occur at the level of synapses between inner hair cells and the auditory nerve or at in any level of the central auditory pathway. Presence of tinnitus for a very long time/duration usually is the results of functioning of complex network structure including central auditory and non-auditory system.
CONFLICT OF INTEREST
The authors declares no conflict of interest
ACKNOWLEDGEMENT
The authors sincerely acknowledge the contribution of Dr. Geetha Mukundan and Prof Geeta Gore, for their immense support and guidance.
REFERENCES
1. Armstrong R, Hall BJ, Doyle J, Waters E. “Cochrane Update. ‘Scoping the scope’ of a cochrane review”. Journal of Public Health. 2011;33:147-50.
2. Bhatt JM, Lin HW, Bhattacharyya N. Prevalence, severity, exposures, and treatment patterns of tinnitus in the United States. JAMA Otolaryngology–Head & Neck Surgery. 2016;142: 959-965.
3. Bauer CA, Brozoski TJ, Myers K. Primary afferent dendrite degeneration as a cause of tinnitus. Journal of Neuroscience. 2009;85:1489-98.
4. Baizer JS, Manohar S, Paolone NA, Weinstock N, Salvi RJ. Understanding tinnitus: The dorsal cochlear nucleus, organization and plasticity. Brain Research. 2012;1485:40-53.
5. Baldo P, Doree C, Molin P. Antidepressants for patients with tinnitus. Cochrane Database Syst Rev. 2009;1:1-6.
6. Bankstahl US, Elkin EP, Gebauer A, Görtelmeyer R. Validation of the THI-12 questionnaire for international use in assessing tinnitus: A multi-centre, prospective, observational study. International Journal of Audiology. 2009;51:671-677.
7. Bauch CD, Lynn SG, Williams DEMellon, MW, Weaver, AL. Tinnitus impact: Three different measurement tools. Journal of the American Academy of Audiology. 2003;14: 584-92.
8. Bogo R, Farah A, Karlsson KK, Pedersen NL, Svartengren M, Skjönsberg, Å. Prevalence, incidence proportion, and heritability for tinnitus: a longitudinal twin study. Ear and Hearing. 2003; 38: 292-300.
9. Bellini JL, Rumrill Jr PD. Research in rehabilitation counseling: A guide to design, methodology, and utilization. Charles C Thomas Publisher. Springfield. 2009;1:6-9.
10. Chang, H, Chen K, Kaltenbach JA, Zhang J, Godfrey, DA. Effects of acoustic trauma on dorsal cochlear nucleus neuron activity in slices. Hearing Research. 2004;164:59-68.
11. Chen YC, Zhang J, Li XW, Xia W, Feng X, Gao B. Aberrant spontaneous brain activity in chronic tinnitus patients revealed by resting-state functional MRI. NeuroImage: Clinical. 2004;6:222-8.
12. Chen GD, Fechter LD. The relationship between noise-induced hearing loss and hair cell loss in rats. Hear Res. 2003;1:81-90.
13. Chrbalka P, Paluch Z, Aluşik Š. Current perspectives of tinnitus and its therapeutic options. European Geriatric Medicine. 2005;6:170-4.
14. Daniell WE, Swan SS, McDaniel MM, Stebbins JG, Seixas NS, Morgan MS. Noise exposure and hearing conservation practices in an industry with high incidence of workers’ compensation claims for hearing loss. American Journal of Industrial Medicine. 2009; 42:309-317.
15. Dehmel S, Cui YL, Shore SE. Cross-modal interactions of auditory and somatic inputs in the brainstem and midbrain and their imbalance in tinnitus and deafness. American Journal of Audiology. 2009;1:1-6.
16. Dong, S, Rodger J, Mulders WH, Robertson D. Tonotopic changes in GABA receptor expression in guinea pig inferior colliculus after partial unilateral hearing loss. Brain Research. 2010;1432:24-32.
17. Emmerich E, Richter F, Reinhold U, Linss V, Linss W. Effects of industrial noise exposure on distortion product otoacoustic emissions (DPOAEs) and hair cell loss of the cochlea long term experiments in awake guinea pigs. Hear Res. 2010;148:9-17.
18. Flor H, Elbert T, Knecht S, Wienbruch C, Pantev C, Birbaumer N, Taube E. Phantom-limb pain as a perceptual correlate of cortical reorganization following arm amputation. Nature. 2010;375:4-82.
19. Gay LR, Mills GE, Airasian PW. Educational research: Competencies for analysis and application. Boston: Pearson Education. 2011;1:1-8.
20. Geocze L, Mucci S, Abranches D, de Marco, MA, de Oliveira Penido N. Systematic review on the evidences of an association between tinnitus and depression. Brazilian Journal of otorhinolaryngology. 2011; 79:106-111.
21. Han BI, Lee HW, Kim TY, Lim JS, Shin KS. Tinnitus: Characteristics, causes, mechanisms, and treatments. Journal of Clinical Neurology. 2009;5:11-19.
22. Henry JA. Audiologic Assessment. In: Snow J. Tinnitus: Theory and management. Ontario: BC Decker Inc. 2004;1:220-36.
23. Henry JA, Dennis KC, Schechter MA. General review of tinnitus. Journal of Speech, Language, and Hearing research. 2006;1:1-6.
24. Henry JA, Roberts LE, Caspary DM, Theodoroff SM, Salvi RJ. Underlying mechanisms of tinnitus: Review and clinical implications. Journal of the American Academy of Audiology. 2014; 25: 5-22.
25. Hobson J, Chisholm E, El Refaie, A. Sound therapy (masking) in the management of tinnitus in adults. Cochrane Database of Systematic Reviews. 2011;1:1-11.
26. Holt AG, Bissig D, Mirza N, Rajah G, Berkowitz B. Evidence of key tinnitus-related brain regions documented by a unique combination of manganese-enhanced MRI and acoustic startle reflex testing. PloS one. 2010;5:14-260.
27. Jastreboff PJ, Jastreboff MM. Tinnitus retraining therapy. In: Snow J. Tinnitus: Theory and management. Ontario: BC Decker Inc. 2014;1: 295-309.
28. Jastreboff PJ, Hazell JW. Tinnitus Retraining Therapy. New York: Cambridge University Press. 2004.
29. Jastreboff PJ. The neurophysiological model of tinnitus. In: Snow J. Tinnitus: Theory and management. Ontario: BC Decker Inc. 2004;1:96-107.
30. Job A, Raynal M, Kossowski M. Susceptibility to tinnitus revealed at 2 kHz range by bilateral lower DPOAEs in normal hearing subjects with noise exposure. Audiology and Neurotology. 2003;12:137-144.
31. Kaltenbach JA. Tinnitus: models and mechanisms. Hearing research. 2002;6: 52-60.
32. Kaltenbach JA, Zhang J, Finlayson P. Tinnitus as a plastic phenomenon and its possible neural underpinnings in the dorsal cochlear nucleus. Hear Res. 2002;6:200-26.
33. Kaltenbach JA, Zhang J, Afman CE. Plasticity of spontaneous neural activity in the dorsal cochlear nucleus after intense sound exposure. Hear Res. 2003;147:282-92.
34. Koehler SD, Pradhan S, Manis PB, Shore SE. Somatosensory inputs modify auditory spike timing in dorsal cochlear nucleus principal cells. European Journal of Neuroscience. 2011;33:409-20.
35. Koester M, Storm C, Zorowka P. Tinnitus-classification, causes, diagnosis, treatment and prognosis. MMW Fortschr. Medizin. 2014;146:23-4.
36. Kraus KS, Canlon B. Neuronal connectivity and interactions between the auditory and limbic systems: Effects of noise and tinnitus. Hearing Research. 2014;88:34-46.
37. Kujawa SG, Liberman MC. Adding insult to injury: Cochlear nerve degeneration after “temporary” noise-induced hearing loss. Journal of Neuroscience. 2019; 29:14077-85.
38. Lanting CP, De Kleine E, Van Dijk, P. Neural activity underlying tinnitus generation: Results from PET and fMRI. Hearing Research. 2009;255:1-13.
39. Leaver AM, Renier L, Chevillet MA, Morgan S, Kim HJ, Rausschecker JP. Dysregulation of limbic and auditory networks in tinnitus. Neuron. 2009;69:33-43.
40. Levine RA, Abel M, Cheng H. CNS somatosensory-auditory interactions elicit or modulate tinnitus. Experimental Brain Research. 2009;153:643-48.
41. Levin RA. Tinnitus: Theory and management. Somatic tinnitus. In: Snow, J. B. editor. Ontario: BC Decker Inc. 2008;1:108–124.
42. Lockwood AH, Salvi RJ, Burkard RF, Galantowicz PJ, Coad ML, Wack DS. Neuroanatomy of tinnitus. Scand Audiol Suppl.1990;51:47-52.
43. Lockwood AH, Salvi RJ, Burkard RF. Tinnitus. N Engl J Med. 2002;347:904-10.
44. López-González, MA, Cambil E, Abrante A, López-Fernández R, Barea E, Esteban F. Tinnitus measurement with conventional audiometer versus high-frequency audiometer. Acta Otorrinolaringol Lager. (English Edition). 2003;63: 102-105.
45. McCormack A, Edmondson-Jones M, Somerset S, Hall D. A systematic review of the reporting of tinnitus prevalence and severity. Hearing research. 2010;337:70-9.
46. Mirz F, Gjedde A, Sdkilde-Jrgensen H, Pedersen C. Functional brain imaging of tinnitus-like perception induced by aversive auditory stimuli. Neuroreport. 2010;11:633-7.
47. Møller AR, Langguth B, DeRidder D, Kleinjung T. Textbook of tinnitus. Springer Science & Business Media. 2010.
48. Mühlnickel W, Elbert T, Taub E, Flor H. Reorganization of auditory cortex in tinnitus. Proceedings of the National Academy of Sciences. 2010; 95:10340-43.
49. Nuttal AL, Meikle MB, Trune DR. Peripheral process involved in tinnitus. In: Snow JB. Tinnitus: Theory and management. Ontario: BC Decker Inc. 2004; 52-68.
50. Parra,LC, Pearlmutter BA. Illusory percepts from auditory adaptation. J Acoust Soc Am. 2007;121:1632-41.
51. Reyes SA, Salvi RJ, Burkard RF, Coad ML, Wack DS. Galantowicz, PJ, et al. Brain imaging of the effects of lidocaine on tinnitus. Hear Res. 2002;171:43-50.
52. Rauschecker JP, Leaver AM, Mühlau M. Tuning out the noise: limbic-auditory interactions in tinnitus. Neuron. 2007;66:819-26.
53. Salvi RJ, Wang J, Ding D. Auditory plasticity and hyperactivity following cochlear damage. Hear Res. 2007;147:261-74.
54. Samarei R, Fatholahi N. Causes of tinnitus in patients referred to ENT clinic of Imam Khomeini hospital in Urmia, 2012-2013. Global Journal of Health Science. 2004;6:136-8.
55. Schreiner CE, Cheung SW. Cortical plasticity and tinnitus. In: Snow JB. Tinnitus: Theory and management. Ontario: BC Decker Inc. 2004;1:189-204.
56. Shargorodsky J, Curhan GC, Farwell WR. Prevalence and characteristics of tinnitus among US adults. The American Journal of Medicine. 2004; 123: 711-18.
57. Shore SE. Plasticity of somatosensory inputs to the cochlear nucleus–implications for tinnitus. Hearing Research. 2011; 281:38-46.
58. Shore S, Zhou J, Koehler S. Neural mechanisms underlying somatic tinnitus. Prog Brain Res. 2007,166:107-23.
59. Tang J, Jii B, Liu L. Study of hearing loss in 200 patients with subjective tinnitus. Lin-Bed Journal of Otorhinolaryngology Head and Neck Surgery. 2005; 25: 726-29.
60. Wienbruch C, Paul I, Weisz N, Elbert T, Roberts LE. Frequency organization of the 40-Hz auditory steady-state response in normal hearing and in tinnitus. Neuroimage. 2006;33:180-94.
61. Yildirim, G, Berkiten G, Kuzdere M, Ugras H. High frequency audiometry in patients presenting with tinnitus. The Journal of International Advanced Otology. 2006;6: 401-6. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2660,
1
],
[
2660,
8946,
2
],
[
8946,
12482,
3
],
[
12482,
16075,
4
],
[
16075,
20342,
5
],
[
20342,
26525,
6
],
[
26525,
32971,
7
],
[
32971,
39365,
8
],
[
39365,
45401,
9
],
[
45401,
51244,
10
],
[
51244,
53644,
11
]
]
} |
1f820b4b0b61afd0d620bed5d19eab24f1d96c8e | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 11,
"total-fallback-pages": 0,
"total-input-tokens": 36749,
"total-output-tokens": 13222,
"fieldofstudy": [
"Medicine"
]
} | Original Article
Palliative Lung Radiotherapy: Higher Dose Leads to Improved Survival?
T.S. Lewis *, J.A. Kennedy *, G.J. Price *, T. Mee *, D.K. Woolf *, N.A. Bayman *, C. Chan *, J.H. Coote *, C. Faivre-Finn *, M.A. Harris *, A.M. Hudson *, L.S. Pemberton *, A. Salem *, H.Y. Sheikh *, H.B. Mistry *†, D.C.P. Cobben *‡
*Department of Radiotherapy Related Research, The Christie NHS Foundation Trust, Manchester, UK
†Manchester Academic Health Science Centre, University of Manchester, Manchester, UK
‡Division of Cancer Sciences, School of Medical Sciences, Faculty of Biology, Medicine and Health, University of Manchester, Manchester, UK
§Division of Pharmacy and Optometry, School of Health Sciences, Faculty of Biology, Medicine and Health, University of Manchester, Manchester, UK
Received 14 January 2020; received in revised form 16 April 2020; accepted 6 May 2020
Abstract
Aims: Choosing the optimal palliative lung radiotherapy regimen is challenging. Guidance from The Royal College of Radiologists recommends treatment stratification based on performance status, but evidence suggests that higher radiotherapy doses may be associated with survival benefits. The aim of this study was to investigate the effects of fractionation regimen and additional factors on the survival of palliative lung cancer radiotherapy patients.
Materials and methods: A retrospective univariable (n = 925) and multivariable (n = 422) survival analysis of the prognostic significance of baseline patient characteristics and treatment prescription was carried out on patients with non-small cell and small cell lung cancer treated with palliative lung radiotherapy. The covariates investigated included: gender, age, performance status, histology, comorbidities, stage, tumour location, tumour side, smoking status, pack year characteristics and treatment prescription was carried out on patients with non-small cell and small cell lung cancer treated with palliative lung radiotherapy.
Methods: This retrospective single-centre analysis of palliative lung radiotherapy, increased total dose (up to and including 30 Gy/10 fractions) was associated with better survival regardless of performance status.
Results: Univariable analysis revealed that performance status (P < 0.001), fractionation scheme (P < 0.001), comorbidities (P = 0.02), small cell histology (P = 0.02), ‘lifelong never’ smoking status (P = 0.01) and gender (P = 0.06) were associated with survival. Upon multivariable analysis, only better performance status (P = 0.01) and increased dose/fractionationregimens of up to 30 Gy/10 fractions (P < 0.001) were associated with increased survival. Eighty-five (9.2%) and 316 patients (34%) died within 30 and 90 days of treatment, respectively.
Conclusion: In this retrospective single-centre analysis of palliative lung radiotherapy, increased total dose (up to and including 30 Gy/10 fractions) was associated with better survival regardless of performance status.
Key words: Lung cancer; outcomes research; palliative care; radiotherapy
Introduction
Lung cancer is the malignancy with the highest incidence worldwide and the leading cause of cancer death [1]. In the UK, lung cancer accounts for 13% of new cancer cases and 22% of cancer deaths. Eighty-five per cent of lung cancer cases are non-small cell lung cancer (NSCLC), with most of the remainder being small cell lung cancer (SCLC) (13%). One-year survival in England and Wales ranges from 82% for patients with stage I NSCLC to 16% for patients with stage IV NSCLC [2]. At presentation, 57% of patients are not candidates for curative therapy due to tumour volume, presence of metastases, patient fitness and/or comorbidities [3]. An increasing number of patients are receiving immunotherapy (sometimes in combination with chemotherapy) or tyrosine kinase inhibitors, which have been shown to improve survival [4,5]. Across hospitals in England in the
The primary treatment of patients with advanced lung cancer is systemic therapy (including chemotherapy, immunotherapy and targeted agents). However, palliative radiotherapy still has a role for those who are unresponsive to systemic therapy, those who relapse and those who have contraindications to, or are not fit for, systemic therapy [7]. Palliative radiotherapy is also often used to manage local symptoms [8,9]. These symptoms are often linked to local tumour effects, such as haemoptysis, chest pain, dyspnoea, cough, dysphagia and superior vena cava compression [10]. Palliative radiotherapy is intended to alleviate the aforementioned symptoms and improve quality of life. In a 2008 systematic review of palliative radiotherapy for lung cancer, improvement in total symptom score was reported in 65.4–77.1% of patients depending on the dose of radiotherapy administered [11].
The dose-fractionation schedule is selected when palliative radiotherapy is recommended to a patient. A balance between successful palliation of the symptoms, fitness of the patient, toxicity and convenience is sought in collaboration with the patient [10]. Toxicities of palliative radiotherapy may include: fatigue, dysphagia, odynophagia, dyspnoea, cough, skin erythema and, rarely, radiation myelopathy [10].
The choice of radiotherapy dose and fractionation scheme in the palliative setting is challenging because there is conflicting evidence regarding the optimal fractionation scheme in order to achieve palliation of symptoms and possibly improve survival. A 2015 meta-analysis found that when the patients were stratified by performance status no significant difference was found in 1-year overall survival [10]. More recently, two studies reported that higher fractionation schemes were associated with increased survival [12,13]. Fractionation schemes utilised varied from 10 Gy/one fraction up to doses more typically associated with the curative intent setting, such as 60 Gy/30 fractions [10].
Current Royal College of Radiologists (RCR) and American Society for Radiation Oncology (ASTRO) guidance suggest the use of palliative regimens with doses up to 39 Gy/13 fractions and 42 Gy/14 fractions, respectively, for patients with NSCLC [14,15]. Longer fractionation schemes can inconvenience patients with multiple hospital visits towards the end of their lives and also have healthcare resource implications.
The time taken for palliative radiotherapy to reach effect has been shown to occur at 5–7 weeks, with palliation occurring 2 weeks earlier in the 16 Gy/two fraction arm compared with the 30 Gy/three fraction arm [16]. Peak palliation occurs at 8–9 weeks. Frank et al. [17] defined radiotherapy as futile if the patient dies less than 30 days after treatment, as the patient has not yet benefitted fully from the treatment but has still been exposed to the risks of radiotherapy-related acute toxicity [17]. This issue has been debated at RCR forums and a consensus agreed that there should be a target of under 20% of patients that die within 30 days of palliative radiotherapy [18]. Patients with an acute presentation of symptoms, such as superior vena cava obstruction, often also have a short life expectancy and are incorporated into this figure.
There are predictive factors that have been investigated to guide the treatment decisions and give prognostic information in the context of palliative radiotherapy. From the literature, the following factors have been found to be significantly correlated with survival during multivariable analysis: T and N status, extrathoracic disease status, lactate dehydrogenase levels, completion of planned treatment, leukocyte count and C-reactive protein levels [10,12,13,17,19,20]. These factors have not been consistently examined through the literature and when included they are not always reproducibly significant and as such they are not incorporated in commonly used guidelines [14,15,21].
The aim of our study was to retrospectively analyse predictive factors for survival in palliative radiotherapy in lung cancer.
Materials and Methods
Cohort Selection
Patients treated for lung tumours with palliative radiotherapy between 1 January 2013 and 8 May 2018 were identified from the UK Computer-Aided Theragnostics (UKCAT) database. The ukCAT database contains the anonymised electronic patient records from a single large cancer centre and was established to model clinical outcomes. Consent is on an opt-out basis (REC reference 17/NW/0060). For this study, consent for patient data access was granted by the ukCAT database management committee (reference: 2017–008). Data from this study were part of a clinical audit (reference: SE18/2221). All research was carried out in accordance with the Declaration of Helsinki. Further details are given in the Supplementary Material.
Detailed patient and tumour characteristics are collected prospectively at the time of the first appointment in our institution. All patients included within the analysis had confirmed histology consistent with lung cancer (see Figure 1). The data specification included the following items: treatment intent and lung or mediastinal cancer. Local guidance recommends the following fractionation schemes: 30 Gy/10 fractions, 20 Gy/five fractions or 10 Gy/one fraction. Therefore, 30 Gy/10 fractions was the highest fractionation scheme included in this study.
Patients were staged with the IASLC seventh edition for TNM staging [22]. The comorbidity score was an overall score calculated with the Adult Comorbidity Evaluation 27 tool [23].
Primary Technique for Radiotherapy Delivery
The primary techniques for radiotherapy delivery were grouped into four larger groups that were deemed to be sufficiently similar. Parallel pair, two field or tangential pair (n = 890); single field (n = 28); three or more fields (n = 5); and all the intensity-modulated radiotherapy techniques
were grouped together \((n = 2)\). Brachytherapy was excluded.
**Statistical Analysis**
A combined NSCLC and SCLC patient cohort \((n = 925)\) and NSCLC-only patient cohort \((n = 664)\) were analysed.
Overall survival was measured from the date of the first fraction. Patients who had not died by 8 May 2018 were considered to be right-censored and were excluded from the analysis. The percentages of patients who died within 30 and 90 days of receiving radiotherapy were calculated.
A univariable and multivariable survival analysis was conducted using the Cox proportional hazards model. The multivariable model was built using a complete case (no missing variable data) analysis \((n = 422)\). P-values and hazard ratios with 95% confidence intervals were reported. Survival curves were plotted using the Kaplan–Meier method and differences between survival curves were assessed using the Log-rank test. A SCLC patient cohort was not analysed separately, as the number of complete cases was deemed to be insufficient \((n = 95)\).
Due to there being a low number of patients with performance status 0 and 4, performance status was grouped as follows: good (0–1), mid (2) and poor (3–4), for the survival analysis. Due to there being a low number of patients treated with 8 Gy/one fraction, they were grouped with the patients treated with 10 Gy/one fraction for the survival analysis. The software used for statistical analysis was R® Version 3.5.1.
**Results**
**Patient Characteristics**
In total, 925 patients with NSCLC and SCLC remained in the cohort for analysis after filtering the originally extracted patient data. Figure 1 shows how the initial patient data downloaded from the electronic patient record system was refined in order to provide a more complete and comparable dataset. Any outlying data were checked manually. There were 816 events within the cohort; 109 patients were censored. The median overall survival was 129 days (95% confidence interval 120–138).
Table 1 summarises the main patient, tumour and treatment characteristics. The gender distribution of the patients was 55.45 male to female. The most common performance status was 2 (35%). In total, 545 of 925 (76%) patients had stage IV disease; 261 of 925 (28%) patients were treated for SCLC and 664 of 925 (72%) patients were treated for NSCLC. Of the patients with NSCLC, most (97%) had either squamous cell carcinoma or adenocarcinoma. As expected, the patients with a high performance status had a high comorbidity score.
The most frequently used fractionation scheme was 30 Gy/10 fractions, with 551/925 (60%) patients being prescribed this regimen.
**Death Within 30 and 90 Days of Treatment**
Eighty-five patients (9%) in the combined NSCLC and SCLC cohort died within 30 days of treatment. Three hundred and sixteen patients (34%) of the combined NSCLC and SCLC cohort died within 90 days of treatment. Seventy-two patients (11%) with NSCLC died within 30 days of treatment. Two hundred and forty-five patients (37%) with NSCLC died within 90 days of treatment.
**Univariable Survival Analysis**
The univariable analysis, see Table 2, highlighted six covariates: performance status, fractionation scheme, comorbidities, small cell histology, gender and ‘lifelong never’ smoking status that were associated with patient survival.
**Univariable Subset Survival Analysis**
When the patients were subdivided into good, mid and poor performance status the fractionation scheme was still found to be a predictor of patient survival. The 30 Gy/10 fractions scheme showed a clear survival advantage in each performance status subset (see Figure 2). This correlation of increased survival with increased fractionation persisted when SCLC patients were removed from the dataset for patients with good, mid and poor performance status (see Tables 2 and 3).
**Multivariable Survival Analysis**
The multivariable analysis highlighted that only fractionation scheme and performance status were predictors of patient survival (see Table 2).
The lack of interaction between fractionation scheme and performance status via Kaplan–Meier survival plots discussed above was further assessed in a multivariable analysis including only these two variables \((n = 925)\). No interaction was found (interaction hazard ratio = 0.97, confidence interval = 0.85–1.10, \(P = 0.60\)) and both performance status and fractionation scheme retained independent effects on overall survival (performance status hazard ratio = 1.20; confidence interval = 1.07–1.35; \(P = 0.002\) and fractionation scheme hazard ratio = 1.70; confidence interval = 1.42–2.02; \(P < 0.001\)).
**Discussion**
In this single-centre retrospective analysis of palliative lung radiotherapy, performance status and fractionation scheme were the only covariates shown to have a significant correlation with patient survival on multivariable analysis. Performance status was correlated with overall survival in a predictable manner: those with a good performance status out-survived those with a mid performance status, who out-survived those with a poor performance status.
In this cohort, when examining both NSCLC and SCLC together, every increase in fractionation regimen through
all performance status strata resulted in an increased median overall survival. The difference in median overall survival between receiving 10 Gy/one fraction and 30 Gy/10 fractions in patients with a good, mid and poor performance status was 126, 80.5 and 77 days, respectively. The cohort of NSCLC patients was also examined in isolation to ensure that there was not a confounding effect from the SCLC patients. The results were similar, with every increase in fractionation regimen through all performance status strata resulting in an increased median overall survival. It should be noted that to date all published prospective studies comparing different palliative thoracic radiotherapy fractionation schemes have been carried out in the NSCLC setting.
Performance Status
The finding that performance status is significantly correlated with survival is in concordance with other survival analyses [12,15,17]. There has been less clarity as to the optimal fractionation scheme in order to increase survival in palliative lung radiotherapy.
Radiotherapy Fractionation and Overall Survival
Janssen et al. [13] reported in a retrospective analysis of 125 patients that increasing equivalent dose in 2 Gy fractions (EQD2) led to significantly better survival outcomes. These patients had stage III and IV lung cancer, including both NSCLC and SCLC. EQD2 of 31–40, 41–46 and 47–52 Gy led to 6-month overall survival of 30, 38 and 57%, respectively, and 1-year overall survival of 11, 26 and 36%, respectively [13]. On multivariable analysis, EQD2 was significant, although the confidence intervals were wide (n = 125, relative risk = 1.43, confidence interval = 1.06–1.94, P = 0.018) [13]. The doses of radiotherapy were higher compared with those used in this study. It should be noted that 29% of the patients could
not complete their full fractionation course due to acute toxicity [13].
Nieder et al. [12] found that lower dose/fractionation regimens (17 Gy/two fractions and 20–24 Gy/five to six fractions) were significantly associated with lower overall survival on multivariable analysis when compared with regimens with an EQD2 of 45Gy. Unlike our study, when Nieder et al. [12] carried out a subset analysis and excluded those with performance status 3–4 this survival advantage was no longer significant.
More recently, Nieder et al. [19] compared the following palliative fractionation regimens of 17 Gy/two fractions versus 30 Gy/10 fractions versus regimens with an EQD2 of 34–50 Gy for those aged 80 years and over. They found a median overall survival difference of 2.4, 2.6 and 11.8 months, respectively, with significant differences in survival for doses /C20 30 Gy and doses >30 Gy. This could be an analysis of a subset of the patients of those included in the previously mentioned study, although this is not explicitly mentioned (the cohort is selected from the same hospital and the time period) [12,19].
In the 2015 Cochrane analysis, a meta-analysis incorporating 14 trials, the authors were unable to obtain enough original individual patient data in order to conduct a time-to-event analysis [16,24–26]. Therefore, the authors were only able to perform a meta-analysis of 1-year overall survival. This meta-analysis of 1-year overall survival for all patients regardless of performance status showed that receiving more fractions and a higher dose was favourable for survival, depending on which model was used (fixed effects versus random effects model) [10]. Due to large heterogeneity in the data for good performance status patients, the authors did not present these data in a summary statistic [10]. Although the data for poor performance status patients showed low heterogeneity and no 1-year overall survival advantage in using a more fractionated regimen, the evidence was rated as moderate [10]. In addition, Frank et al. [17] investigated 159 patients with NSCLC and
| Table 1 (continued) |
|---------------------|------------------|
| Covariate | Number of patients (% proportion of patients with data present) |
| Mean | 42 |
| N/A | 234 |
| Fractionation scheme| |
| 8 Gy/1 fraction | 10 (1%) |
| 10 Gy/1 fraction | 97 (10%) |
| 20 Gy/5 fractions | 267 (29%) |
| 30 Gy/10 fractions | 551 (60%) |
| 8 Gy + 10 Gy/1 fraction | 107 (12%) |
| Primary radiotherapy technique | |
| Intensity-modulated radiotherapy | 2 (0%) |
| Parallel pair and two field | 890 (96%) |
| Single field | 28 (3%) |
| Three+ field | 5 (1%) |
| Table 1 |
|---------|
| Baseline patient characteristics of this large cancer centre’s cohort |
| Covariate | Number of patients (% proportion of patients with data present) |
| Age | |
| Range | 36–93 |
| Interquartile range | 14 |
| Mean | 69 |
| ECOG performance status| |
| 0 | 65 (7%) |
| 1 | 315 (34%) |
| 2 | 325 (35%) |
| 3 | 213 (23%) |
| 4 | 7 (1%) |
| Good (0–1) | 380 (41%) |
| Mid (2) | 325 (35%) |
| Poor (3–4) | 220 (24%) |
| Histology | |
| Small cell | 261 (28%) |
| lung cancer | |
| Non-small cell | 664 (72%) |
| lung cancer | |
| Non-small cell lung cancer subgroups | |
| Adenocarcinoma | 323 (35%) |
| Adenosquamous cell carcinoma | 7 (1%) |
| Large cell carcinoma | 14 (1%) |
| Squamous cell carcinoma| 320 (35%) |
| Comorbidities | |
| 0 | 221 (28%) |
| 1 | 257 (33%) |
| 2 | 202 (26%) |
| 3 | 96 (12%) |
| N/A | 149 |
| TNM stage | |
| 2a | 9 (1%) |
| 2b | 11 (2%) |
| 3a | 61 (8%) |
| 3b | 95 (13%) |
| 4 | 545 (76%) |
| N/A | 204 |
| Tumour location | |
| Lung, upper lobe | 533 (65%) |
| Lung, middle lobe | 60 (7%) |
| Lung, lower lobe | 230 (28%) |
| Lung, not otherwise specified (N/A) | 102 |
| Tumour side | |
| Left | 301 (40%) |
| Right | 462 (60%) |
| N/A | 162 |
| Smoking | |
| Lifelong never | 27 (3%) |
| Light former | 7 (1%) |
| Ex-smoker | 457 (60%) |
| Current smoker | 272 (36%) |
| N/A | 159 |
| Pack years | |
| Range | 0–150 |
| Interquartile range | 20 |
Table 2
Univariable and multivariable survival analysis
| Covariate | Univariable survival analysis NSCLC + SCLC cohort | Multivariable analysis NSCLC + SCLC cohort | Multivariable analysis NSCLC-only cohort |
|----------------------------------|--------------------------------------------------|------------------------------------------|----------------------------------------|
| | $N (E)$ | Hazard ratio (95% confidence interval) | P-value | Hazard ratio (95% confidence interval) | P-value | Hazard ratio (95% confidence interval) | P-value |
| Sex | | | | | | | |
| Female (reference) | 925 (816) | 1 | 1 | 1 | | | |
| versus male | 1.14 (0.99–1.31) | 0.06 | 1.03 (0.83–1.27) | 0.82 | 0.95 (0.74–1.22) | 0.67 |
| Age | 925 (816) | 1.00 (1.00–1.01) | 0.35 | 1.00 (0.99–1.01) | 0.85 | 0.99 (0.98–1.01) | 0.48 |
| Performance status | | | | | | | |
| Good (0–1) versus mid (2) | 925 (816) | 1.32 (1.21–1.45) | <0.001 | 1.22 (1.05–1.42) | 0.01 | 1.25 (1.05–1.49) | 0.01 |
| Comorbidities | | | | | | | |
| 0 versus 1 versus 2 versus 3 | 776 (673) | 1.09 (1.02–1.18) | 0.02 | 1.06 (0.94–1.18) | 0.34 | 1.08 (0.94–1.23) | 0.27 |
| Combined TNM stage | | | | | | | |
| IV (reference) | 721 (642) | 1 | 1 | 1 | | | |
| versus III | 0.89 (0.74–1.08) | 0.23 | 0.98 (0.75–1.28) | 0.87 | 1.05 (0.76–1.43) | 0.78 |
| versus I-II | 1.03 (0.63–1.67) | 0.92 | 0.74 (0.38–1.44) | 0.38 | 0.70 (0.33–1.49) | 0.36 |
| Tumour location | | | | | | | |
| Lower lobe (reference) | 823 (720) | 1 | 1 | 1 | | | |
| versus middle lobe | 0.87 (0.64–1.17) | 0.35 | 1.83 (0.79–4.23) | 0.16 | 1.37 (0.49–3.79) | 0.55 |
| versus upper lobe | 0.93 (0.80–1.11) | 0.46 | 0.90 (0.71–1.14) | 0.36 | 0.87 (0.67–1.14) | 0.32 |
| Tumour side | | | | | | | |
| Left (reference) | 763 (669) | 1 | 1 | 1 | | | |
| versus right | 0.99 (0.85–1.15) | 0.88 | 1.04 (0.83–1.29) | 0.75 | 0.98 (0.76–1.25) | 0.85 |
| Smoking status | | | | | | | |
| Current smoker (reference) | 766 (663) | 1 | 1 | 1 | | | |
| versus light former | 0.88 (0.36–2.14) | 0.78 | 0.42 (0.13–1.39) | 0.16 | 0.43 (0.13–1.43) | 0.17 |
| versus ex-smoker | 0.92 (0.78–1.08) | 0.31 | 0.95 (0.75–1.21) | 0.69 | 0.98 (0.74–1.29) | 0.88 |
| versus lifelong never | 0.57 (0.37–0.89) | 0.01 | NA* | NA* | NA* | NA* | NA* |
| Pack years | 691 (598) | 1.00 (1.00–1.01) | 0.4 | 1.00 (0.99–1.00) | 0.59 | 1.00 (0.99–1.00) | 0.30 |
| Fractionation scheme | | | | | | | |
| 30 Gy/10F versus 20 Gy/5F | 925 (816) | 1.73 (1.56–1.91) | <0.001 | 1.48 (1.23–1.77) | <0.001 | 1.54 (1.25–1.89) | <0.001 |
| Gy/1F | | | | | | | |
| Primary radiotherapy technique | | | | | | | |
| IMRT (reference) | 925 (816) | 1 | 1 | 1 | | | |
| versus parallel pair and two field | 2.03 (0.29–14.41) | 0.48 | 2.03 (0.28–14.86) | 0.49 | 2.30 (0.31–17.05) | 0.42 |
| versus single field | 2.13 (0.29–15.74) | 0.46 | 2.03 (0.25–16.50) | 0.51 | 2.23 (0.27–18.30) | 0.46 |
| versus 3+ field | 1.64 (0.19–14.06) | 0.65 | 1.54 (0.16–15.15) | 0.71 | 1.83 (0.18–18.15) | 0.61 |
IMRT, intensity-modulated radiotherapy; $N$, number of patients; $E$, number of events.
* There were no lifelong never smokers when performing a complete case analysis.
compared 30 Gy/10 fractions, 25 Gy/five fractions, 15 Gy/three fractions and 10 Gy/one fraction, finding no statistically significant correlation between overall survival and radiotherapy regimen.
It is difficult to directly compare this study with others finding a positive correlation between increased fractionation and overall survival, as the fractionation schemes utilised in each study are variable with a large range in EQD2. The maximum palliative lung fractionation scheme used in this cohort was 30 Gy/10 fractions (EQD2 32.5), whereas Nieder et al. [12] and Janssen et al. [13] reported much higher doses used, with maximum radiotherapy doses of EQD2 45 and 47–52 Gy, respectively.
Fig 2. Univariable subset analysis Kaplan–Meier survival curves examining varying fractionation schemes and the correlating overall survival when all patients were divided by performance status strata in the combined non-small cell lung cancer/small cell lung cancer patient cohort.
There were several potential biases in our study that need to be explored. This was a retrospective, single-centre study. The conclusions that can be drawn are therefore limited due to both known and unknown confounding factors. It is likely that the patients receiving a larger number of fractions are also the patients receiving systemic treatment. It would have been beneficial to include previous systemic therapies (chemotherapy, tyrosine kinase inhibitors and immunotherapies) and previous radiotherapy in our analysis, as these treatments could have a major impact on survival. It is likely that patients with SCLC would have only received higher fractionation schemes if they had shown a good response to systemic therapies, as recommended in RCR guidance (see Table 4). This could have been a confounding factor, as these patients showing a good response to systemic therapy would probably go on to have longer survival. Unfortunately, due to some patients receiving systemic treatment at multiple hospitals and a lack of integrated databases there was not sufficient access to information on these previous treatments. Improvements in electronic registration and database integration mean that in future analysis we expect to take the role of systemic treatments in palliative lung radiotherapy into account.
It would have also been interesting to see if mutation status (such as epidermal growth factor receptor and anaplastic lymphoma kinase rearrangement) was correlated to survival, but unfortunately these data were unavailable. There was a large proportion of missing data in several of the covariates examined (see Supplementary Material) and this reduced the power of the multivariable analysis. It would be informative to carry out regular audits on which data are not being fully recorded and why.
Treatment field size was another important prognostic indicator that would have been valuable to include, as tumour volume is known to be a factor associated with poorer survival, but this was also unavailable [27]. There was little variance in the radiotherapy technique, with the vast majority of patients treated with parallel pair/two field (96%). Therefore, a meaningful comparison of radiotherapy technique effect on overall survival did not take place. Granton et al. [28] have recently reported a decreased incidence of dysphagia following oesophageal-sparing intensity-modulated radiotherapy as opposed to parallel pair beams. If this technique becomes standard of care, it would be worthwhile investigating its effect on overall survival, toxicity and patient-reported symptoms.
Performance status was shown in this study to be a prognostic indicator of overall survival and is used in multiple guidelines to determine treatment. Yet, performance status is not entirely objective, with the clinician’s judgement playing a large role in determining the patient’s score.
| Table 3 |
|------------------------------------------------------------------------|
| Univariable subset survival analysis results examining varying fractionation schemes and the correlating overall survival when all patients were divided by performance status strata in both the combined non-small cell lung cancer/small cell lung cancer (NSCLC/SCLC) patient cohort and the NSCLC patient-only cohort |
| Fractionation scheme 8 ÷ 10 Gy/1 fraction | Fractionation scheme 20 Gy/5 fractions | Fractionation scheme 30 Gy/10 fractions |
| Good performance status (0–1) | Median overall survival 67 days (n = 19, 95% confidence interval 58–108) | Median overall survival 112 days (n = 77, 95% confidence interval 95–158) |
| Combined NSCLC and SCLC cohorts | Median overall survival 71.5 days (n = 32, 95% confidence interval 57–131) | Median overall survival 88 days (n = 107, 95% confidence interval 78–109) |
| Mid performance status (2) | Poor performance status (3–4) Median overall survival 72 days (n = 56, 95% confidence interval 40–90) | Median overall survival 80 days (n = 83, 95% confidence interval 66–107) |
| Good performance status (0–1) | Median overall survival 67 days (n = 17, 95% confidence interval 58–129) | Median overall survival 106 days (n = 57, 95% confidence interval 92–137) |
| NSCLC cohort only | Median overall survival 67 days (n = 29, 95% confidence interval 55–131) | Median overall survival 84 days (n = 88, 95% confidence interval 67–102) |
| Mid performance status (2) | Poor performance status (3–4) Median overall survival 64 days (n = 42, 95% confidence interval 38–103) | Median overall survival 75 days (n = 58, 95% confidence interval 56–108) |
| Good performance status (0–1) | Median overall survival 67 days (n = 17, 95% confidence interval 58–129) | Median overall survival 106 days (n = 57, 95% confidence interval 92–137) |
| NSCLC cohort only | Median overall survival 67 days (n = 29, 95% confidence interval 55–131) | Median overall survival 84 days (n = 88, 95% confidence interval 67–102) |
| Mid performance status (2) | Poor performance status (3–4) Median overall survival 64 days (n = 42, 95% confidence interval 38–103) | Median overall survival 75 days (n = 58, 95% confidence interval 56–108) |
There was a large proportion of missing data in several of the covariates examined (see Supplementary Material) and this reduced the power of the multivariable analysis. It would be informative to carry out regular audits on which data are not being fully recorded and why.
Discrepancies between clinician- and patient-reported performance status have been documented and been shown to be associated with poorer survival [29]. As performance status determines treatment regimen and is sometimes an entry criteria to clinical trials, the treating clinician may be assigning performance status to fit the treatment rather than vice versa [30].
Previous studies have established that fractionation does not have a bearing on symptom control but can increase acute toxicity [10]. Unfortunately, in this cohort it was not possible to carry out an analysis on symptomatic improvement or toxicity due to its retrospective nature. The question remains: does a higher fractionation scheme not only lead to a longer overall survival, but also to a better quality of life?
**RCR and ASTRO palliative lung radiotherapy fractionation guidance** is summarised in Table 4. Neither ASTRO nor RCR guidance defines good, mid or poor performance status [14,15]. This makes it difficult to determine if clinicians are following the guidance explicitly. In the 2015 Cochrane review on palliative lung radiotherapy, good performance status was defined as a score of 0–1 and poor as 2–4 [10]. Other studies have classified a score of 0–1 as good, 2 as moderate and 3–4 as poor [31]. A number of patients in this study were prescribed an alternative dose of radiotherapy than that recommended by the RCR or ASTRO. According to RCR guidance, in this cohort, only 17 patients (2.6%) with a good performance status were undertreated (with 8 Gy/one fraction or 10 Gy/one fraction) and 110 patients (16.6%) with a poor performance status were over-treated (with 20 Gy/five fractions or 30 Gy/10 fractions). According to ASTRO guidance, 74 patients (11.1%) with a good performance status were undertreated as they were given <30 Gy/10 fractions and 54 patients (8.1%) with a poor performance status were over-treated as they were given >20 Gy/five fractions. Moderate performance status patients are not defined within either guidance.
Nine per cent of patients within this large centre’s cohort died within 30 days of receiving palliative radiotherapy. This is consistent with other published data, including Spencer et al. [32,33], who examined the 30-day mortality of 3628 patients who received palliative lung radiotherapy, resulting in a 30-day mortality rate of 14%. It is also within the RCR forum suggested limit of 20% [18].
**Future**
Although this study has limitations, it adds to the justification for the need of a prospective multicentred randomised controlled trial to examine the effects of varying fractionation schemes and radiotherapy techniques on survival in today’s era of modern systemic therapies. This future study should include both doctor- and patient-reported outcomes. The TOURIST (Thoracic Umbrella Radiotherapy Study in Stage IV) trial, a UK-based trial, is currently under development and is aiming to answer these questions (Woolf D, Lee C, Shah R, Ahmed M, Fraser I, Billingham L et al., unpublished data).
An area of unmet need are studies evaluating dose/fractionation regimens in the SCLC setting. To our knowledge there is only one prospective trial, currently recruiting, that is looking into a dose—effect relationship in patients with extensive stage SCLC. This trial compares 30 Gy/10 fractions versus 45 Gy/15 fractions in patients who have shown a response to standard of care chemotherapy (Clinicaltrials.gov identifier: NCT02675088).
The data coming from ongoing and future studies should be used to create decision tools to help patients with lung cancer considered for palliative treatment to balance survival gain and quality of life.
**Conclusions**
In this retrospective, single-centre analysis of palliative lung radiotherapy, although limited by a lack of data on systemic anticancer treatments, toxicity and quality of life, we found that increased fractionation regimens (up to and
including 30 Gy (10 fractions) were associated with better survival regardless of performance status.
Conflict of interest
C. FAivre-Finn has declared research grants from Astrazeneca, MSD and Elekta and sits on the advisory boards of Astrazeneca and Pfizer. D. Woolf has declared travel grants, consultancy and speaker fees from Astrazeneca and travel grants from Roche. G. Price acknowledges the support of Cancer Research UK via the funding to Cancer Research Manchester Centre (C147/A18083) and (C147/A25254).
Acknowledgements
The key results presented in this paper have been published previously as part of a poster presentation at the 2019 British Thoracic Oncology Group and European Society for Radiotherapy and Oncology conferences. The authors are grateful for the assistance of Dr Kate Wicks in preparing the manuscript. This work was supported by Cancer Research UK via the funding to Cancer Research UK Manchester Centre: (C147/A18083) and (C147/A25254). Cancer Research UK did not have any input into the study design, data analysis or interpretation, the writing of the report, or the decision to submit the article for publication. C. FAivre-Finn was supported by the National Institute for Health Research: Manchester Biomedical Research Centre.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.org/10.1016/j.clon.2020.05.003.
References
[1] Ferlay J, Soerjomataram I, Dikshit R, Eser S, Mathers C, Rebelo M, et al. Cancer incidence and mortality worldwide: sources, methods and major patterns in GLOBOCAN 2012. Int J Cancer 2015;136:E359–E386.
[2] Royal College of Physicians. National lung cancer audit annual report 2017. 1–29 2018. Available at: https://www.rcplondon.ac.uk/projects/outputs/nica-annual-report-2017. [Accessed 4 October 2019].
[3] Royal College of Physicians. National lung cancer audit annual report 2018. 1–22 2019. Available at: https://www.rcplondon.ac.uk/projects/outputs/nica-annual-report-2018. [Accessed 8 March 2020].
[4] WOO H, MHaskar R, Kumar A, Miladinovic B, Djulbegovic B. Survival of patients with non-small cell lung cancer without treatment: a systematic review and meta-analysis. Syst Rev 2013;2:10.
[5] MOya-Horno I, Viteri S, Karachaliou N, Rosell R. Combination of immunotherapy with targeted therapies in advanced non-small cell lung cancer (NSCLC). Therapeut Adv Med Oncol 2018; 10:1–12.
[6] The National Cancer Registration and Analysis Service. The radiotherapy dataset 2019. Available at: https://www.cancerdata.nhs.uk/radiotherapy. [Accessed 4 October 2019].
[7] Stewart LA. Chemotherapy in non-small cell lung cancer: a meta-analysis using updated data on individual patients from 52 randomised clinical trials. BMJ 1995;311:899–909.
[8] National Institute for Health and Care Excellence. Lung cancer: diagnosis and management 2019. Available at: https://pathways.nice.org.uk/pathways/lung-cancer#path=view%3Apathways/lung-cancer/supportive-and-palliative-care-for-lung-cancer.xml&content=view-node%3A3Anodes-palliative-radiotherapy. [Accessed 17 February 2020].
[9] Rodrigues G, Videtic GM, SUR R, Bezjak A, Bradey J, Hahn CA, et al. Palliative thoracic radiotherapy in lung cancer: an American Society for Radiation Oncology evidence-based clinical practice guideline. Pract Radiat Oncol 2011;1:60–71.
[10] Stevens R, Macbeth F, Toy E, Colles B, Lester JF. Palliative radiotherapy regimens for patients with thoracic symptoms from non-small cell lung cancer. Cochrane Database Syst Rev 2015;1:CD002143.
[11] Fairchild A, Harris K, Barnes E, Wong R, Lutz S, Bezjak A, et al. Palliative thoracic radiotherapy for lung cancer: a systematic review. J Clin Oncol 2008;26:4001–4011.
[12] Nieder C, Tollali T, Yobuta R, Reigstad A, Flatoy LR, Pawinski A. Palliative thoracic radiotherapy for lung cancer: what is the impact of total radiation dose on survival? J Clin Med Res 2017;9:482–487.
[13] Janssen S, Kaesmann L, Schild SE, Rades D. Impact of the radiation dose and completion of palliative radiotherapy on survival in patients treated for locally advanced lung cancer. Anticancer Res 2016;36:1825–1828.
[14] The Royal College of Radiologists. Radiotherapy dose fractionation, Second Edition 2019. p. 49–57. Available at: https://www.rcr.ac.uk/system/files/publication/field_publication_files/bfc0193_radiotherapy_dose_fractionation_third-edition-lung-cancer_0.pdf. [Accessed 17 February 2020].
[15] Moeller B, Balagamwala EH, Chen A, Creach KM, Giaconne G, Koshy M, et al. Palliative thoracic radiation therapy for non-small cell lung cancer: 2018 Update of an American Society for Radiation Oncology (ASTRO) Evidence-Based Guideline. Pract Radiat Oncol 2018;8:245–250.
[16] Kramer GW, Wanders SL, Noordijk EM, Vonk EJ, van Houwelingen HC, van den Hout WB, et al. Results of the Dutch national study of the palliative effect of irradiation using two different treatment schemes for non-small-cell lung cancer, J Clin Oncol 2005;23:2962–2970.
[17] Frank MS, Nørøxe DS, Nygaard L, Persson FG. Fractionated palliative thoracic radiotherapy in non-small cell lung cancer — futile or worth-while? BMC Palliat Care 2018:17:15.
[18] Lees K. In: Audit of 30 day mortality following palliative radiotherapy, vols. 1–3. Available at: https://www.rcr.ac.uk/audit/audit-30-day-mortality-following-palliative-radiotherapy. [Accessed 1 May 2018].
[19] Nieder C, Yobuta R, Mannsaker B, Dalhaug A. How should palliative thoracic radiotherapy be fractionated for octogenarians with lung cancer? In Vivo 2018;32:331–336.
[20] Lotayef M, Elkader YA, Amin A, Taher A, Ehab EK, Abdelall M. A prospective study of the effect of different palliative radiotherapy fractionation schedules on tumor response and toxicity in advanced non-small cell lung cancer (NSCLC) patients. J Cancer Ther 2016;7:924–938.
[21] Planchar D, Popat S, Kerr K, Novello S, Smitt EF, Faivre-Finn C, et al. Metastatic non-small cell lung cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann Oncol 2018;29:iv192–iv237.
[22] Mirsadraee S. The 7th lung cancer TNM classification and staging system: Review of the changes and implications. World J Radiol 2012;4:128–134.
[23] Binder PS, Peipert JF, Kallogjeri D, Brooks RA, Massad LS, Mutch DG, et al. Adult Comorbidity Evaluation 27 score as a predictor of survival in endometrial cancer patients. *Am J Obstet Gynecol* 2016;215:766.e1–766.e9.
[24] Bezjak A, Dixon P, Brundage M, Tu D, Palmer MJ, Blood P, et al. Randomized phase III trial of single versus fractionated thoracic radiation in the palliation of patients with lung cancer. *Int J Radiat Oncol Biol Phys* 2002;54:719–728.
[25] Reinfuss M, Glinski B, Kowalska T, Kulpa J, Zawila K, Reinfuss K, et al. Radiotherapy in stage III, unresectable, asymptomatic non-small cell lung cancer. Final results of a prospective randomized study (240 patients). *Cancer Radiother* 1999;3:475–479.
[26] Macbeth FR, Bolger JJ, Hopwood P, Bleehen NM, Cartmell J, Girling DJ, et al. Randomized trial of palliative two-fraction versus more intensive 13-fraction radiotherapy for patients with inoperable non-small cell lung cancer and good performance status. *Clin Oncol* 1996;8:167–175.
[27] Plataniotis G, Theofanopoulou MA, Sotiriadou K, Theodorou K, Mavroidis P, Kyrgias G. Palliative hypofractionated radiotherapy for non-small-cell lung cancer (NSCLC) patients previously treated by induction chemotherapy. *J Thorac Dis* 2009;1:5–10.
[28] Granton PV, Palma DA, Louie AV. Intentional avoidance of the esophagus using intensity modulated radiation therapy to reduce dysphagia after palliative thoracic radiation. *Radiat Oncol* 2017;12:1–5.
[29] Schnadig ID, Fromme EK, Loprinzi CL, Sloan JA, Mori M, Li H, et al. Patient-physician disagreement regarding performance status is associated with worse survivorship in patients with advanced cancer. *Cancer* 2008;113:2205–2214.
[30] Blagden SP, Charman SC, Sharples LD, Magee LR, Gilligan D. Performance status score: Do patients and their oncologists agree? *Br J Cancer* 2003;89:1022–1027.
[31] Prigerson HG, Bao Y, Shah MA, Paulk ME, LeBlanc TW, Schneider BJ, et al. Chemotherapy use, performance status, and quality of life at the end of life. *JAMA Oncol* 2015;1:778–784.
[32] Park KR, Lee CG, Tseng YD, Liao JJ, Reddy S, Bruera E, et al. Palliative radiation therapy in the last 30 days of life: a systematic review. *Radiother Oncol* 2017;125:193–199.
[33] Spencer K, Morris E, Dugdale E, Newsham A, Sebag-Montefiore D, Turner R, et al. 30 day mortality in adult palliative radiotherapy - a retrospective population based study of 14,972 treatment episodes. *Radiother Oncol* 2015;115:264–271. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3931,
1
],
[
3931,
9863,
2
],
[
9863,
15093,
3
],
[
15093,
16915,
4
],
[
16915,
22410,
5
],
[
22410,
32421,
6
],
[
32421,
32704,
7
],
[
32704,
38612,
8
],
[
38612,
42556,
9
],
[
42556,
48738,
10
],
[
48738,
51223,
11
]
]
} |
47f07a00b79d8b71363dcac45c3c98dcf02bcb52 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 12,
"total-fallback-pages": 0,
"total-input-tokens": 35576,
"total-output-tokens": 14567,
"fieldofstudy": [
"Education"
]
} | Secondary Education in Uganda: Resource Mobilization and Efficiency
Godwin Tindyebwa Muhangi
Mbarara University of Science and Technology, P. O. Box 1410 Mbarara, Uganda
Abstract
Over the years, scholars and researchers alike have evaluated the basic components of quality secondary education. This has been the case in Uganda and other countries. Success on such an undertaking requires that countries will have to confront the resource requirements for maintaining growth and expansion trends in secondary education. Uganda as a developing country is faced with limited educational resources and a concerted effort from all stake holders must be put in place to mobilize the required resources to improve efficiency of this sector. Such would allow schools operate with minimum difficulty and realise the objectives of secondary education. Adequate resources have been not mobilized to allow secondary schools in Uganda fully realize the governments’ existing plans for secondary education. The limited available resources are competed for with other demands in the public sector; making it difficult for the government to mobilize additional resources to make secondary education efficient. Additionally, challenges related to increased access to education, financing, increased refugee population, teacher supply and resource allocation all limit efficiency of secondary schools in Uganda. Also, the secondary education curriculum has been said to be outdated, irrelevant, or poorly implemented and this reduces efficiency of secondary education. Without practical steps to improve efficiency in terms of quality and relevance, secondary education is likely to consume vast amounts of resources without leading to the hoped benefits of improved social and economic development. This paper illumines secondary school resources in Uganda, how such have been put to use and how this has not realised the desired levels of efficiency. This could be attributed to unsatisfactory resource mobilization strategies. The stakeholders need to step up their resource mobilization strategies. Alternative sources of funding such as engaging in income generating activities will provide desired finances and certainly improve the efficiency of the sector.
Keywords: Secondary, education, Resource, mobilization, efficiency
1. Introduction
Uganda recognizes and upholds the right to education as indicated by various policy documents (Right to Education Project, 2012). As a nation, Uganda appreciates education as a “tool” for national stability and development (UNESCO, 2014) and an agent for nation building and social cohesion (Zebun, 2016). While issues associated to primary and higher education have been diagnosed for long; secondary education and its peculiar attributes are still scanty in the education literature (Jacob & Lehner, 2012). It is important to critique issues surrounding secondary education in Uganda. According to Chapman, Burton and Werner (2010) for example, wider access to good secondary education is a critical element for the achievement of the goals of human development, political stability, and economic competitiveness. This can only be possible with enough resources for efficient operations within the sector.
Secondary education as an intermediary level between primary and higher education prepares the youth for the world of work and equips a largely adolescent population with the skills, aptitudes, and social values for a productive and healthy adult life (Jacob & Lehner, 2012). It is therefore fundamental that resources for secondary education be mobilized to allow schools take care of the educational needs of such youths. It has thus been suggested that a well-planned and adequately resourced secondary education has potentials for equipping students with skills and knowledge required for better economic and social participation in a peaceful, democratic society. Also, secondary education contributes towards the development of the knowledge necessary to avoid risky behavior for better and healthy lives (UNESCO, 2010).
While the above depicts the desired goals for secondary education, such have not been fully achieved (Sentongo, 2018). This has been attributed to insufficient resources in terms of finances, teachers or technical support to the secondary education sector (Chapman, Burton & Werner, 2010). Many factors relating to secondary education resources have limited the achievement of secondary education goals. For example, a shortage of schools, as well as the inability to pay for education, have limited achievement of secondary education goals such that the quality of secondary in Uganda is often poor (Uganda Government, 2018). Achieving the goals of secondary education requires increasing resource supply including scholastic materials, teachers, finances and infrastructure (Obadara & Alaka, 2010).
Several challenges limit the achievement of secondary education in Uganda. These challenges are largely related to resource mobilization. If this is not achieved, the challenge will become insurmountable making the
goals of secondary education unachievable. Issues of financing, curriculum, and access together with their consequent effects on efficiency in secondary schools have been cited but these have not been well knitted together in the context of Uganda. Additionally, mobilization of funds and other resources for secondary schools in Ugandan is something that requires analysis. This paper makes suggestions on how Uganda can address the multiple challenges related to resource mobilization faced in secondary education and then grow the sector efficiently.
1.1 The secondary education context in Uganda
There is a great international heterogeneity regarding secondary education provision, its content, length, and the age cohort engaged in it (OECD, 2018). Consequently, while many people have an idea of what secondary education is or what it should be, there exist differences in national interpretations of secondary education (Jacob & Lehner, 2012). Many countries define secondary education differently (Eubanks & Eubanks, 2011). In the Ugandan context, secondary education is a six-year cycle from senior one to senior six having learners with average age between 14 years and 19 years (National Curriculum Development Centre, 2018). It is one of the different options for educational progress within Uganda’s education system (Uganda National Commission for UNESCO, 2010).
The secondary cycle of education in Uganda lasts for six normal progress years including four years ordinary secondary level (lower secondary) and two for Advanced secondary level (higher secondary) (JICA & IDCI, 2012). On successful completion of the ordinary level of secondary education, students are awarded the Uganda Certificate of Education i.e. UCE (Nuffic, 2016). Successful senior four leavers have four possible paths through which they can attain further education: they can either proceed to advanced level of education; join two-year advanced crafts courses in technical institutes; join a two-year grade III primary teaching programme; or join any of the government’s departmental programmes such as agriculture, health, veterinary, and cooperatives (Uganda Investment Authority, 2010). On successful completion of upper secondary, students are awarded Uganda Advanced Certificate of Education (UACE) (Uganda Government, 2012). Successful upper secondary school leavers proceed to university or join a two-year courses leading to ordinary diploma in teacher education, technical education; business studies or join departmental programmes (Uganda Investment Authority, 2010).
1.2 Provision of secondary education in Uganda
Secondary education occurs in distinct secondary schools whose ownership is variable. According to Uganda Bureau of Statistics (2017), provision of secondary education is through a three types of secondary schools i.e. government, private sector and community owned secondary schools. Secondary schools owned by the government may either be government aided schools or grant aided schools previously community owned but have been taken over by government (Education Act, 2008). On the other hand, private secondary schools are founded and owned by private individuals, religious bodies or NGOs and they currently constitute 66% of all secondary schools in Uganda (Kazuya & Chikako, 2016). International schools that deliver foreign curricula also exist though they are only a few of them (Sentongo, 2018).
Irrespective of their ownership, secondary schools are managed and supervised by Ministry of Education and Sports (MoES) whose mission is to provide, support, guide, coordinate, regulate, and promote quality education and sports for national integration, individual and national development of Uganda (MoES, 2013). Thus, secondary education in Uganda is provided by both the government and the private sector (Education Statistical Abstract, 2016). Some private schools receive some funding from the government (hence called government grant-aided schools) under the Public Private Partnership (PPP) (Namusobya, Aubry & McKernan, 2015). Also, the Ministry of Education ensures that trained teachers are deployed; salaries and allowances are paid to teachers; educational materials and other capital development inputs provided; national selection. Besides admission guidelines are followed for all students enrolled in both government and government-aided schools (Makaaru, Cunningham, Kisaame, Nansozi & Bogere, 2015). In private secondary schools, the government mainly ensures that private schools conform to rules and regulations governing provision of education services (Huylebroeck & Kristof, 2015). The government is also responsible for deployment of teachers and head teachers, placing students and organizing national examinations.
The Government of Uganda introduced Universal Secondary Education (USE) in 2007 to increase access to secondary education for economically vulnerable families and communities (MoES, 2013). This is carried out in some government and private schools termed as USE and PPP Schools (Bo-Jo, 2011). In such schools, students are automatically promoted but to proceed from O-level to A-level and graduating from A’ level, one needs to pass national exams administered by the Uganda National Examination Board or UNEB (Huylebroeck & Kristof, 2015).
The available literature on secondary education does not provide insights into efficiency and resource mobilization in secondary schools in Uganda. Much of the literature for example focuses on the impact of USE
on secondary school enrolments (Kazuya & Chikako, 2016), the impact of USE on teachers and moonlighting activities and the importance of involving head teachers in policies such as USE (Chapman, Burton, Werner, 2010). Other studies have investigated incidences of poor service delivery in secondary schools (Urwick & Kisa, 2013) characterised by higher teacher absenteeism (Molyneaux, 2011). There have been no attempts to diagnose efficiency and resource mobilization in secondary schools in Uganda. This paper provides an insight on mobilization and resource efficiency in secondary education in Uganda; and serves as a catalyst for studies in this area.
As suggested in this paper, variation in the nature of secondary schools in Uganda has implications for efficiency and resource mobilization strategies for such schools. According to the Education Statistical Abstract (2016), there are about 3,070 secondary schools in Uganda including 1,058 government secondary schools and 2012 private secondary schools. 690 of the private secondary school are under public private partnership and thus receive funding from the government together with 902 USE school. 372 secondary schools (12%) are registered while 497 (16%) are licensed and 2,201 secondary schools (72%) are registered. It is argued in this paper, that the different secondary school types have different sources of resources. The different sources influence the resource mobilization strategy and certainly impact the efficiency. In a country with one of the worlds’ highest youthful populations (Uganda Bureau of Statistics and ICF, 2017), a variety of educational providers, and an influx of refugees, there is need for a critical consideration of resource provision to the secondary education sector in Uganda for purposes of improving efficiency (Uganda National Commission for UNESCO, 2010).
1.3. Role of secondary education in Uganda
Development of human capital through education significantly contributes towards national economic growth, development and improvement in well-being among citizens (Anikina, Ivankina, Tumanova, 2015). In recognition of this, there has been intense policy discussion about education, society and national development (African Network Campaign on Education for All, 2012). Concomitantly, secondary education has become a policy concern of developing countries, particularly among those that have progress in terms of universal primary education (UPE), and those in which demographic transition has shifted towards adolescents (Uganda technology and Management University (2016). This is well represented in academic literature and is reflected in political debates for developed and developing countries (Kulid, 2014). This makes reference to Uganda; a developing country that has progressed in terms of UPE provision.
Secondary education being intermediate between primary and higher education plays specific roles (Damon, Glewwe, Wisniewski & Sun, 2016). Because of the different education policy issues that have been associated with education in Uganda, the context has been changing over time (Uganda National Commission for UNESCO, 2010). This has significantly influenced the roles of secondary education. Policy makers in Uganda have had a thorough consideration of these roles. According to the Curriculum Development Centre (2018), the roles of secondary education in Uganda include the following:
- Secondary education instills and promotes national unity and an understanding of social and civic responsibilities; love and care for others and respect for public property, as well as an appreciation of international relations and beneficial international co-operation.
- It promotes appreciation and understanding of the Uganda cultural heritage and its languages.
- Through secondary education, sense of self-discipline, ethical and spiritual values, personal and collective responsibility and imitativeness are imparted and promoted.
- Secondary education enables students to acquire, develop and appreciate emerging societal and economic needs.
- It provides updated and comprehensive knowledge in theoretical and practical aspects of innovation, modern management methods in commerce and industry and their application in socio-economic development.
- It allows development of basic scientific, technological, technical, agricultural and commercial skills required for self-employment.
- It enables development of personal skills in problem-solving, information gathering and interpretation, independent reading and writing and physical or social self-improvement.
- It provides a foundation for further education.
- Allows individuals to apply acquired skills in solving problems of the community and to develop a strong sense of constructive and beneficial belonging to that community.
- Instills positive attitudes towards productive work and strong respect for the dignity of labour and those who engage in productive labour activities.
2. Resource mobilization for secondary education in Uganda
The 2012 Rio de Janeiro United Nations Conference gave rise to Sustainable Development Goals (SDGs). These are universal goals to meet world environmental, political and economic challenges (United Nations, 2012) that replaced the Millennium Development Goals (MDGs). MDGs set a platform for human development and expanded primary education to all children, among other development priorities (United Nations, 2017). Also, MDGs facilitated growth, reduced poverty, improved access to water and sanitation; reduced child mortality and improved maternal health (United Nations, 2015). Most importantly, MDGs led to the adoption of UPE which increased the number of candidates for secondary education and certainly presented quality related challenges (Jacob & Lehner, 2012).
Secondary education needs in Uganda have been on an increase due to the introduction of UPE in 1997 (Asankha & Takashi, 2014). However, it is also recognised that quality secondary education generates opportunities and benefits for social and economic development as envisaged by the SDGs (Patrinos, 2016). This notwithstanding, secondary enrolment grew from 954,000 in 2007 to 1.5 million by 2016 (World Bank, 2018). This challenges provision of quality secondary education and requires more resources towards secondary education (Ministry of Finance, Planning and Economic Development, 2018). Considering the SDGs, goal No. 4 emphasizes the need for ‘quality education’ by ensuring that education is inclusive and promotes lifelong learning (United Nations, 2012). For this goal to be achieved, governments and donors need to commit resources to mitigate factors that limit attainment of quality education.
Secondary education resources are obtainable from the government, parents and donors (Baghdady & Zaki, 2019). Thus, depending on the nature of the school, resources are mobilized from any or a combination of the mentioned sources. In government non USE secondary schools, government funds are used to pay salaries - a fraction of the total expenses on secondary education (Sentongo, 2018). Consequently, secondary education in non-government aided secondary schools relies on resource mobilization from sources such as households mainly through user charges (Omoeva & Gale, 2016). Thus, schools should initiate alternative sources of income and try to make up for the budgetary deficits. This requires a thorough consideration of the alternative funding mechanisms, the budgeting process, expenditure items, and the design of the fiscal system (Gongera & Okoth, 2013). Funding sources themselves influence education efficiency and quality (Gogo, 2012). This paper discusses different funding sources with the basic argument that Uganda should consider rearranging or reforming the financing mechanisms so as to ensure sustainable resource mobilization and efficiency of secondary schools.
Better resource mobilization for secondary schools in Uganda entails expanding public funding, encouraging contributions from the private sector, or the international community (UNESCO-UIS, 2011) besides encouraging income generating activities at school (Omuoba, Simatwa & Ayodo, 2011). Additionally, financing programs need to mix different funding sources in innovative ways (Education Task Force of innovative financing to fund development leading group, 2012). Without such engagements, the quality of secondary education will be unattainable and certainly bring about inefficiency.
2.1. Sources of funds for secondary education in Uganda
According to World Bank (2018b), Uganda is on track towards the achievement of universal primary education (UPE). This requires Uganda to determine strategies and financing options for the expanded secondary education that are consistent with national human capital development goals. This is because fiscal constraints cannot allow the country to solely rely on government financing for the expanded secondary education (Glennerster, Kremer, Mbiti & Takvarasha, 2011). With the eminent difficulties, adoption of policies to reduce the burden of secondary education funding has been inevitable. The strategies include (1) charging tuition fees as a contribution towards the cost of public secondary education; and/or (2) privatization policies to encourage development of private schools and shoulder part of surging numbers (UNESCO-IIEP, Pole de Dakar, UNESCO-UIS, 2016a). This informed four sources of funding to secondary education in Uganda and these are the state, households, local communities (including donations from), and Multinational development partners (Gongera & Okoth, 2013).
2.1.1. The state (Public sector)
Uganda government has the responsibility for education provision; it plays a central role in ensuring efficient, equitable and effective management and financing of public education (Makaaru, et al; 2015). This depends on the school status: whether government aided or not; for example out of the 3,149 secondary schools in Uganda, only 39.4 % are government funded (MoES, 2016). By being government aided (including government non USE, USE and PPP), schools receive funding from the government as a proportion of the national GDP. In 2013, government expenditure on secondary education was reported to be 0.6426 % of GDP of Uganda (Trading economics, 2013) and this is a very low value which certainly limits efficiency the sector. National governments worldwide spend copious amounts of public resources to fund education services, yet access to those services is not always equitable for all intended beneficiaries (Mugendawala, 2012). This limits efficiency since not all the intended users access it (OECD, 2012). Access to secondary education is skewed towards specific segments such
as educated classes, schools, regions and wealth quantiles; only those with access to public schools benefit from public education resources (UNESCO-IIEP, et al; 2016a).
In attempt to improve access and efficiency in Ugandan secondary education, the government introduced USE programme (Bo-Joe, 2011). Through this programme, the government offers eligible students free places in some government aided secondary schools (Asankha & Takashi, 2012). The government aided schools were supplemented by some private schools through the PPP since there was a limited number of government aided secondary schools when the programme was introduced (Bo-Joe, 2011). Like the government aided schools, PPP schools receive specific support from the government including the provision of textbooks and other teaching.
While the introduction of USE is appreciated for it solved some challenges related to access secondary education, the programme has been said to be inefficient (Huylbroeck & Kristof, 2015). It is thus imperative that the government revisits the USE policy so as to marry efficiency with the purposes of USE.
2.1.2 Households
This relates to the funds paid by families for their children to attain secondary education. Such expenses relate to tuition fees paid directly to the school and other school related payments made both inside and outside of schools, such as catering costs, transportation costs, uniforms, and textbooks (Ministry of Education & Sports, 2013). It has been reported that households fund more than one-half of education expenditure in Uganda (UNESCO-IIEP, et al; 2016a). For private secondary education however, households are the main sources of funding. Additionally, households fund a considerable portion of secondary education spending in government funded schools. Such spending may be formal or informal; direct or indirect contributions from households (Omoeva & Gale, 2016).
For private secondary schools, fees and other charges payable to schools constitute the most significant items paid for by the students’ household (Smith & Baker, 2017). In addition to such payments, households pay for other charges such as registration and examination fees, auxiliary fees, contributions to parent-teacher associations or school management committee fees (World Bank, 2018b). In Uganda, other payments represent 38% of household expenditure on education in public schools but when taken together, payments made outside of schools for items such as uniforms, teaching materials, private classes and other expenses represent more than half of what households spend on education, especially in public schools (UNESCO-IIEP, 2016a). It has also been reported that households fund at least 50% of the cost of teaching materials in Uganda (ibid). This represents a heavy burden of students’ secondary education costs borne by households. If not considered carefully, they exclude some students from secondary education.
From the above, it is shown that household contributions constitute an important source of secondary education financing and, with good policy design; they could provide sustainable subsidies for the poor and help improve efficiency and accountability by increasing the schools’ responsibility to parents (Crawfurd, 2016). Furthermore, household contributions can increase reliability and resilience of institutional service delivery and thus help to ensure that the education provided meets the demand (Zheng & Libertus, 2018). However, excessive reliance on household contributions can depress enrollments, even when there is excess capacity (Alecke, Burgard, Mitze, 2013). User charges and fees have a substantial impact on parents’ decisions about their children’s schooling (Kiage, Simatwa & Ayodo, 2014). The government should come in to regulate fees charged especially in private secondary schools.
As already indicated, teaching materials are largely funded by households and this affects the quality of learning; poorer households experience greater difficulty because they are expected to fund teaching materials (UNESCO-IIEP, 2016a). Children from poorer households have less access to adequate materials, which may hamper their learning. Thus, when the burden on households to pay for their education is too heavy, issues of equity and accessibility may arise (OECD, 2018). This is an area that policy makers need to consider critically.
2.1.3 Funding from multinational development partners
Multinational development partners play a small but significant role towards secondary education funding to a tune of about 7% of total secondary education funding (UNESCO-IIEP, 2016a). They include the European Union, the World Bank, African Development Bank, and Master Card Foundation among others (UNECO, 2014). Their funds come in through the government either as loan or grants (UNESCO-IIEP, 2016a). For effective policy formulation and analysis, it is important to identify whether the support is given as a loan or grant (MoES, 2017). Loans have to be reimbursed and could be understood as part of national government funding yet this has long term impact on national economy. Grants on the other hand are sources of external funding and contribute towards educational development with no ‘strings attached’ (Gabriela, 2013).
An Education Funding Agency Group (EFAG) has been established to bring together all donors that provide financial and technical support to the education sector to improve coordination among these donors (Sentongo, 2018). While this is done, Uganda needs to evaluate the impact of educational loans from development partners on education. Continuous borrowing without evaluation of the impact may not support the goal of attaining middle income status in the near future. The country could certainly do better by generating revenues locally to finance her secondary education.
2.1.4 Donor funds from NGOs and religious bodies or individuals
Secondary education to particular students in Uganda is funded through donor funds generated from local NGOs and religious bodies or even philanthropic individuals (Hedger, Williamson, Muzoora & Stroh, 2010). Some faith-based organizations (FBOs) are important actors in either managing or financing (or both) secondary education activities (Ministry of gender labour and social development, 2011). Others have fully or partially been financing students’ secondary education (Uganda secondary education improvement project, 2018). Also, community-based organizations (CBOs) which are non-profit entities working at a local level to improve the lives of residents in social fields and have often been involved in providing and/or funding educational services at local level (Ministry of gender labour and social development, 2011).
These entities fund secondary education through school fees payment and /or other charges especially in form of scholarships or bursaries (Alcott, Rose & Sabates, 2016). In some cases, the educational support is in form of cash transfers to students or their families who then use the funds to pay for tuition and/ or other education expenses (Gordon & Rose, 2018). While the end education service is attained by students, the source of the funding is the donor agency. In some other cases, the scholarship is transferred through the school (the producing unit) before reaching the student (UNESCO-IIEP et al; 2016a).
Thus, the local community (such as local NGO and the church) fund secondary education by supporting poor students or their household to enable students who would have otherwise not been able, access secondary education. This is an avenue that can further be developed especially with the government coming in to support more learners to access education. The government can for example commit funds in form of scholarship directed to schools which can in turn use the funds to exempt, partially or totally, the targeted students from payment of tuition or user fees (UNESCO-IIEP et al; 2016a). A similar case would be a government subsidizing students attending private schools by channeling its support directly through the student with a scholarship (Sentongo, 2019). Alternatively, the government could give the institution a specific amount for each student it subsidizes. In all cases, access and efficient operation of secondary education can be achieved.
3.0. Efficiency of secondary education in Uganda
Efficiency of a sector depends on the extent and nature of funding extended to the sector in consideration (Nunnenkamp & Öhler, 2010). With this in consideration, public spending for education in Uganda, as a share of GDP compared with countries with similar GDP per capita, is well below expectations for secondary education (World Bank, 2018b). This affects efficiency of the secondary education sector. National spending as a share of the national budget on secondary education has been on a decrease i.e. from 15% to 11% over the last few years despite the introduction of the Universal Secondary Education (USE) policy in 2007 (Bashir, Lockheed, Ninan Dulvy, Tan, 2017). Due to the decrease in public funding to the sector, the burden of financing has been shifting to households in both public and non-government schools (World Bank, 2018b). Since the economic muscle of some households is weak, some students miss a chance of attaining education. Thus, the government needs to step in and iron out this anomaly.
Access to secondary education in Uganda has been on the increase largely due to the existence of private secondary schools and presence of the USE programme (World Bank, 2018b). This has however not been experienced in the world of work; the average level of education of the work force remains low and does not meet labor market requirements (World Bank, 2018a). This notwithstanding estimates from the National Household Survey (2016) show that only entrants with postsecondary education can escape informal sector work. This can be attributed to an inefficient secondary education sector. Thus, to increase employability and productivity of the expanding workforce, supply of quality education, especially for low-income, rural households and girls, is critical. Only one in five people (20%) aged 15 and above have completed secondary education (National Household Survey, 2016). This has led to a large number of youth entering the job market without foundational skills of basic literacy and numeracy, as well as generic skills essential for life and work (World Bank, 2018a).
The above shows that Uganda faces a challenge of expanding secondary school enrolment (Bashir, Lockheed, Ninan Dulvy & Tan, 2017). Though enrollments in secondary education have increased since the introduction of USE at an average rate of 6% per year, growing from a total of 954,000 students enrolled in 2007 to over 1.5 million in 2016, the pace of increase in enrollment remains low in comparison to Uganda’s neighbors (Bashir, et al; 2018). Such an increase is attributed to the many private secondary schools in the country which stand at about 59% of secondary schools enrolling slightly above half of enrollment (World Development, 2018). This has nevertheless not improved the secondary enrollment rate which is significantly below that of regional comparators (World Bank, 2018b). In 2016 the average annual enrollment increase stood at 25 % in Kenya and 16 % in Rwanda (World Development Report, 2018). Furthermore, Gross Enrollment Rates (GER) have stagnated since 2007 with GER at only 28 % in 2017, much below enrollment rates in neighboring countries. GER in Kenya, Rwanda and Ethiopia was 58% (in 2009), 37% (in 2016) and 38 % (in 2012) respectively (UNESCO Institute of Statistics, 2018). Very low enrollment rates in secondary education and the lack of
progress indicate inefficiency of the system.
There is also notable enrollment disparity between boys and girls all over Uganda with the disparity being skewed positively towards boys (Education and Sports Sector Annual Performance Report, 2016). Great disparities are in urban centers of Soroti, Mbale and Kampala which all have high GERs of 47, 60 and 53% respectively (ibid). The high GERs have not even improved Gender Parity Indices (GPIs) implying that a high GER is not indicative of a high GPI. This suggests that increasing supply itself does not necessarily address gender imbalance in schools. Efforts to address the challenge of gender imbalance in secondary schools should be devised.
The combination of issues discussed above leads to an overall low performance of the education system including poor learning outcomes and thus an inefficient secondary education sector. This has resulted in a declining trend of student learning especially science subjects and math (Education and Sports Sector Annual Performance Report, 2016). Additionally, there are differentials across gender especially in sciences and math and across urban and rural schools (ibid). The persistent decline in performance points to an urgent need to address the quality of secondary education. An analysis of UCE results in 2015 and in 2016 showed that over 52% and 58% respectively, of the candidates failed to pass in at least division 3 and therefore have marginal prospects for further education and employment (World Bank, 2018b). Students perform worst in science subjects with approximately half of those who sit not passing mathematics, physics, chemistry and biology (ibid). Similar experience is observed at advanced level though performance in UACE examinations is only slightly better given that only the best performing O-level students proceed to A-level. This weakens the foundation for further learning besides undermining the country’s potential to tackle poverty.
3.1. Restraints to efficiency in secondary education in Uganda
Section 3.0 is an overview of efficiency related issues in secondary schools in Uganda. Insufficient resources coupled with improper allocation attracts inefficiencies in the sector and consequently undermines performance and productivity of the entire sector. While inefficiency is the observed factor, several other challenges augment the status quo which inter alia includes: poorly distributed and few secondary schools in the country which limit efficiency of the sector (Statistical Abstract, 2016). This reduces the capacity to address the existing and growing demand of secondary education resulting from high population growth rate, the rise in primary school completion, and the recent influx of refugees.
Low survival rate throughout the education cycles also limits internal efficiency (World Bank, 2019). In 2017, primary school survival rate was 56%, which is considerably below the primary survival rate in Kenya, which is close to 100%, Ethiopia at 72%, or Rwanda at 68% (World Bank, 2018b). This implies that it takes almost twice as many years of schooling than normal to ‘produce’ a graduate in primary and secondary education. The inefficiencies persist through secondary school, largely as a consequence of low transition rates to lower secondary (Sentongo, 2018). This increases the cost of education such that, the cost of service provision at the secondary level in 2013 was 2.3 times higher than what it should have been (Wodon, 2016).
Improper use of resources characterised by inefficient deployment of teachers across secondary schools also limits efficiency and this has resulted in several schools ending up with too many or too few teachers (UNESCO 2014). It has been reported that 40% of teachers in secondary schools in Uganda are placed in schools based on factors other than the class time required by students (ibid). This has been attributed to an outdated curriculum which further complicates teacher allocation across schools by imposing too many subjects that require specialized teachers (Uganda technology and Management University (2016).
The influx of refugees into Uganda further exacerbates inefficiency especially in terms of access (Uganda CPF, 2016). Uganda is Africa’s largest refugee hosting country and one of the five largest refugee hosting countries in the world (UNESCO-IIEP, et al, 2016a). According to the UN High Commission for Refugees (UNHCR), Uganda hosts over 1.4 million refugees. Since refugees share all social services with host communities, the refugee population strains the already limited resources, including education (UNHCR, 2019). Majority of the refugees are hosted in the Northern districts of Uganda yet these are the least developed in the country with much lower levels of human capital and enrollment in education (World Bank, 2018b).
Refugees currently make up over half of the total populations in some districts and in the 12 refugee hosting districts, the secondary school-aged population (13-17 years of age) including both refugee and host communities is estimated at 310,121 (with the refugee population estimated at 147,020 and the host community at 163,192) (Education Response Plan for Refugees and Host Communities in Uganda 2018). Secondary school provision is limited for refugees and host communities in the refugee hosting districts and this limits efficiency-an issue that should be looked into. Only 11% of refugees in eight of the refugee hosting districts have accessed secondary education, with only 33% of these being girls (Education Response Plan for Refugees and Host Communities in Uganda 2018). In the same twelve districts, only 18% of the host community secondary school aged children are enrolled, which is considerably below the national average (ibid). These additional pressures further exacerbate the crisis-like situation.
Worth noting also is that girls’ experience of secondary education is characterized by lower access, high dropout and low transition rates compared to boys (Jones, 2011). In 2016 the enrollment rate for boys was 29% compared to 25% for girls (i.e. 4% lower than that for boys) and the GPI was 86%. In the same year, completion rates at ordinary level for boys were 40%, compared to 36% for girls. Such disparities widened at the transition point to Senior 5 such that 34% of boys and 24% of girls transitioned to the next level (Education and Sports Sector Annual Performance Report, 2016). Also, learning outcomes tend to be lower for girls in certain subjects. For instance, in 2016 only 33% of girls in Senior 2 were proficient in mathematics in comparison with 49% of boys (ibid). Thus, a lot needs to be done to bring about efficiency.
According to Ahikire & Madanda (2011), the main reasons for girls dropping out of secondary school are pregnancy (40%), marriage (28%) and the high cost for secondary education (7.3%). Uganda’s levels of child marriage are above expectations (Walakira, Muhangti, Munyuvinyi, Matovu, Awich, 2016) leading many women (36.5%) that marry before 18 years (Demographic and Health Survey, 2011). The probability of completing secondary education for women aged 25-34 who married after 18 is 12.9 points higher than for women who married earlier (World Bank, 2018b). This suggests that continued schooling delays marriage and appropriate policies should be implemented by schools to prevent early marriage and pregnancy. Related to the preceding, distance to lower secondary schools for young adolescents, especially girls from poor families tends to raise opportunity costs and physical risks (ibid).
Despite the relative importance of the PPP arrangement, there are several challenges in its design and implementation which undermine its potential effectiveness (Matsiko, 2017). There is lack of accountability to government and poor learning outcomes in some private schools (Barrera-Osorio, de Galbert, Habyarimana & Sabarwal, 2015). In addition, private provision tends to focus on densely populated areas with strong demand, often ignoring the poorest and most underserved parts of Uganda (Day Ashley, Mcloughlin, Aslam, et al; 2014). It is for this reason that policy makers have been advocating to end the PPP arrangement and shift its funding to the construction of new government secondary schools though this choice of action is potentially detrimental to the system (Barungi, 2017).
Inefficiency can also be attributed to low levels of access and equity that have been exacerbated by an outdated curriculum and low quality teachers (Uganda secondary education improvement project, 2018). The secondary education curriculum has been under revision with the aim of improving the quality and relevance of education and training leading to better learning outcomes. The curriculum is ‘overloaded’ with too many subjects and does not fully meet labor market competencies (Nsbuga & Okwakol, 2014). This should be reformed to reduce the number of subjects and refocus pedagogy making it more student-centered and competency-based in order to produce graduates with skills relevant to the changing labor market (World Bank, 2018a).
A low quality teaching force, often lacking the necessary skills, leads to poor learning outcomes hence an efficient education sector (Ministry of Education & Sports, 2013). The current teacher policies are inefficient in terms of attracting talent, supporting strong head teachers, and providing support to teachers to improve instruction (World Bank, 2018b). There is also a lack of subject proficiency and all these impact on student learning. Though 90% of secondary school teachers in Uganda have the required formal qualifications, research has shown that secondary teachers do not have the content knowledge and also lack sufficient pedagogical skills to teach (UNESCO, 2014). For example, only 66%, 70% and 17% of teachers are proficient in English, mathematics and biology respectively. This has a great impact on the efficiency of secondary education.
3.2. Improving efficiency of secondary education in Uganda
It is no doubt that the private sector plays a significant role in increasing access to secondary education in Uganda (Namusobya, Aubry & McKernan, 2015). Most secondary schools in Uganda are private schools and provide education to the majority of students enrolled in secondary education (ibid). There are 1.6 million students in secondary education, out of which 41% are in government managed schools and 59% in independently-managed schools. It is therefore worthwhile for government to step in to support private secondary schools to allow them overcome efficiency bottlenecks.
To further improve efficiency, secondary education should adopt new funding sources to generate additional revenues. Additional incomes from alternative sources make school management easier (Gongera, & Okoth, 2013), enable schools to pay workers on time and even hire extra labor whenever it was required (Omukoba, Simatwa, & Ayodo, 2011). The extra funds can also be used to hire more teachers, take students for benchmarking and motivate the students and staff and infrastructure improvement (Gogo, 2012).
Secondary schools should also be more grounded in their immediate communities so as to improve their efficiency. The relationship between the school and immediate community should be strengthened to facilitate community participation in school activities (Sa’ad & Sadiq, 2014). The parent-teacher associations should be made more active, school management committees formed and, in certain instances, local-based committees created. These interventions are instrumental in making community participation an effective channel: providing
a legal framework for local participation, creating structures for participation and training and orientation programmes to developing local capacity (Bakwai, 2013).
There is a need to move from a control-based and visitor oriented inspection system to a mechanism that provides innovative supervision and support to schools where needed (Directorate Education Standards, 2014). The idea that supervisors need to be more involved with teacher training and demonstration classes, to assist schools to evolve development plans etc. is important. A substantial part of what is now the job of the supervisor can be replaced by strengthening internal management of schools, i.e. internal and local monitoring systems. Unfortunately, developing countries like Uganda invest too little in supervision activities to make them effective which limit the attainment of the desired end.
4. The final thoughts
4.1. Conclusions
The paper examined alternative sources of financing as a precursor to resource mobilization in secondary education in Uganda and how this has influenced the observed efficiency of secondary education in Uganda. The following conclusions are worth noting;
In addition to secondary school government funding, household sponsorship, and donors that may be local or international were also identified. These sources are however insufficient regarding to the current needs of the existing secondary sector. The funding modes have failed to improve gender disparity and a greater burden is currently being borne by households. This has consequently affected the efficiency of the secondary schools especially with regard to meeting the development needs of the country. Secondary school graduates are of poor quality because the secondary education sector is inefficient. Resource mobilization should be addressed so as to improve secondary education efficiency.
4.2. Recommendations
In addition to the existing funding sources, secondary schools should explore other alternative sources (income generating projects) that may be agricultural based, commercial based or service based. Agricultural based income generating projects include livestock and crop farming. Commercial based activities may include school canteen, confectionery shop and bakery. Service based income generating activities may include school bus hire, hire of furniture and school fields, school halls and equipment. Other than income generating projects, other funding options may include government bursaries, and Fundraising activities.
It is also recommended that there should be interventions aimed at increasing access to secondary schools and reducing cost of education for poor households and providing incentives for girls to stay in school. Such will positively affect education access and attainment. The government through the Ministry of education should formulate policies that will enable schools to have alternative sources of financing secondary education to break the overdependence on free day secondary funding that does not seem to be sustainable from the findings of the study.
The government through the ministry of education should develop structures within secondary schools and also promote in-service training courses to train school managers on effective management of projects. This will allow schools to generate funds internally and hence be able to subsidize the fees charged and therefore affordable to most households. This will increase access, improve retention by preventing drop-out and facilitate achievement of SDG no. 4.
The secondary school curriculum should be revised to give more emphasis to subjects that encourage development of both cognitive and psychomotor skills. This will lead to realization of quality education for self-reliance, in line with the national goals of secondary education. Additionally, teacher re-training and continuous professional development of all teachers should be initiated and maintained to refresh teachers with desirable skills, knowledge and attitudes for the realisation of goals of secondary education.
5. References
African Network Campaign on Education For All, (2012). Civil Society Advocacy Paper for the Second Decade for Education in Africa (2006-15).
Ahikire, J & Madanda, A. (2011). A survey on re-entry of pregnant girls in primary and secondary schools in Uganda. FAWE Uganda
Alcott, B. Rose, P. & Sabates, R. (2016) ‘Targeted Multidimensional Approaches to Overcome Inequalities in Secondary Education: Case Study of Camfed in Tanzania,’ Background Paper, The Learning Generation, The Education Commission
Alecke, B., Burgard,C., Mitze, T. (2013). The Effect of Tuition Fees on Student Enrollment and Location Choice – Interregional Migration, Border Effects and Gender Differences. Ruhr Economic Paper No. 404
Anikina, E., Ivankina, L., Tumanova, I. (2015). Human well-being and educational investment efficiency. International Conference on Research Paradigms Transformation in Social Sciences 2014, Procedia - Social
and Behavioral Sciences 166, 48 – 52
Asankha, P & Takashi, P. (2014). Impacts of Universal Secondary Education Policy on Secondary School
Enrollments in Uganda. National Graduate Institute for Policy Studies (GRIPS), Tokyo, Japan.
Asankha, P., Takashi, Y., (2011). “Impacts of Universal Secondary Education policy on secondary school
enrolments in Uganda”, Journal of Accounting, Finance and Economics, 1(1), 16-30.
Baghdady, A & Zaki, O. (2019). Secondary Education Governance in Sub-Saharan Africa. World Innovation
Summit for Education, Qatar.
Bakwai, B. (2013). Assessment of the role of school community relationship in the development of secondary
education in Zamfara State. Sokoto Educational Review Journal, 14(1): 3543.
Barrera-Osorio, F., de Galbert, P., Habyarimana, J., & Sabarwal, S. (2015). The Impact of Public-Private
Partnerships on Private School Performance: Evidence from a Randomized Controlled Trial in Uganda.
Policy Research Working Paper; No. 7905. World Bank, Washington, DC.
Barungi, M., (2017) “Does government have to abandon aiding private USE schools?” Commentary, Daily
Monitor 31 August, 2017.
Bashir S., Lockheed M., Ninan Dulvy E., Tan J.P. (2017). Facing Forward: Schooling with Learning in Africa.
World Bank, Washington DC.
Bo-Joe, B. (2011). Public Private Partnerships in Education: Analyzing PPPs as a policy tool for Universal
Secondary Education in Uganda. Universiteit van Amsterdam
Chapman, D., Burton, L., Werner, J. (2010). Universal secondary education in Uganda: the head teachers’
dilemma, International Journal of Educational Development, 30(1), 77-82.
Crawford, L. (2016). School Management and Public-Private Partnerships. Center for Global Development &
University of Sussex.
Damon, A., Glewwe, P., Wisniewski, S. &Sun, B (2016). Education in developing countries – what policies and
programmes affect learning and time in school? Elanders Sverige AB Stockholm.
Day Ashley L, Mcloughlin C, Aslam M, Engel J, Wales J, Rawal S, Batley R, Kingdon G,Nicolai S, Rose P
(2014) The role and impact of private schools in developing countries: a rigorous review of the evidence.
Final report. Education Rigorous Literature Review. Department for International Development.
Dejaeghere, J.G., Williams, R., Kyejune, R., Ugandan secondary school head teachers’ efficacy: What kind of
training for whom? International Journal of Educational Development, 29, 312–320.
Directorate Education Standards (2014). Enhancing the quality of school management, teaching and Learning
processes in education institutions: challenges, best practices and proposed interventions. Education and
Sports Sector Review, Kampala, author.
Education Task Force of innovative financing to fund development leading group (2012). Innovative financing
for Education: Moving Forward. International Expert Report. Paris, France.
Education and Sports Sector (2016). Annual Performance Report, FY16-17, pp. 200 – 201. Uganda.
Education System: A Review of the Progress, Challenges and Potential Solutions. The Abdul Latif Poverty Action
Lab at MIT (J-PAL) and Innovations for Poverty Action (IPA).
Eubanks, D., Eubanks, L.T. (2011). Quality of human resources: Education. The Importance of Secondary
Education, Vol 2.
Encyclopedia Britannica (2019). Secondary education. Standard Edition. Chicago:
Gabriela, S.M. (2013). The role of international organizations in the global economic governance – an assessment.
Romanian Economic and Business Review – Special issue, 308-316
Glennerster, R., Kremer, M., Mbiti, I. & Takvarasha, K. (2011). Access and Quality in the Kenya. Nairobi.
Gongera, E & Okoth, N.O. (2013).Alternative sources of financing secondary schools in rural Kenya: A case of
Kisii County, Kenya. Journal of Education and Practice, 4(4) 104-108
Gogo, J.G.O. (2012). A study on cost effective measures to reduce operational costs of secondary education, a
case study of Nyando District. Daystar University Centre for Research, Publications and Consultancy.
Working Paper Series, Number DU/2012/003.
Gordon, R. & Rose, P. (2018). Targeting scholarships and cash transfers for secondary education in Sub-Saharan
Africa. Research for Equitable Access and Learning (REAL) Centre, University of Cambridge
Hedger, E., Williamson, T. Muzoora, T. & Stroh, J. (2010). Sector Budget Support in Practice Case Study
Education Sector in Uganda. Overseas Development Institute, UK.
Huylebroeck, L. & Kristof, T. (2015) ‘Universal Secondary Education (USE) in Uganda: blessing or curse? The
impact of USE on educational attainment and performance’. In: Reyntjens, F., Vandegistne, S. and M.
Verpoorten (eds.) L’Afrique des Grands Lacs: Annuaire 2014-2015. Antwerp: University Press Antwerp,
349-372.
Jacob, W. J. & Lehner, S. (2012). EQUIP2 State-of-the-Art Knowledge in Education; Secondary Education: A
Guide to Education Project Design Based on a Comprehensive Literature and Project Review. EQUIP2
State-of-the-Art Knowledge Series.
JICA & IDCJ (2012). *Education Sector Analysis Report-Uganda*. JICA
Jones, S. (2011). Girls' secondary education in Uganda: assessing policy within the women's empowerment framework. *Gender and education*, 23(4), 385-413
Kazuya, M. & Chikako, Y. (2016). *The Effects of Universal Fee-Free Secondary Education Program on Students’ Access, Composition, and Achievement: Evidence from Uganda*. Hitotsubashi University.
Kiage J.M, Simatwa E.M.W, Ayodo T.M.O. (2014). Effect of school fees and levies on enrollment of girls in Boarding Public Secondary Schools in Transmara Sub County, Kenya: an analytical study. *Educ.Res*. 5(5):152-165
Kulild, V. (2014). *Role of education in ending extreme poverty -Taking a global lead*. Carita seminar on the United Nations Sustainable Development Goals and Post-2015 Development Agenda, Oslo Norway.
Makaaru, A. J., Cunningham, K., Kisaame, K.E., Nansozi, S.M & Bogere, G. (2015). *Public Expenditure Governance in Uganda’s Education Sector; Application of an Innovative Assessment Framework*. *ACODE Policy Research Series* No.67, 2015
Matsiko G. M, (2017). Embracing Public Partnerships to Improve Learning Outcomes in Education. *CEPA Policy Series Papers* NO.15 of 2017. Kampala
Ministry of Education and Sports, (2013). *Teacher issues in Uganda: a diagnosis for a shared vision on issues and the designing of a feasible, indigenous and effective teachers’ policy*. Author Kampala
Ministry of Finance, Planning and Economic Development (2018). Provision of Inclusive Education in Uganda: What are the challenges? *BMAU briefing paper* (13/18), Kampala, Author
Ministry Of Education and Sports (2016). *Education Abstract 2016*. Kampala, Education Planning and Policy Analysis Department.
Ministry of gender labour and social development (2011). *Resource Mobilization Strategy for Implementation of the National Response to Vulnerable Children*. Kampala, Author
Ministry Of Education and Sports, (2013). *A Comprehensive Report on Universal Post Primary Education & Training (UPPET/USE) & Universal Post O-Level Education & Training (UPOLET) National Headcount Exercise*. Education Planning and Policy Analysis Department, Kampala.
MoES (2016). *Education and sports sector strategic plan 2017/18 - 2019/20*. Author, Kampala.
Mugendawala, H. (2012). *Inequity in educational attainment in Uganda: implications for government policy*. *International Journal of Global Education*. 1(1), 93-101
National Curriculum Development Centre (2018). *Secondary Curriculum*. https://www.ncdc.go.ug/
Nunnenkamp, P & Öhler, H (2010). Funding, Competition and the Efficiency of NGOs: An Empirical Analysis of Non-charitable Expenditure of US NGOs Engaged in Foreign Aid. *Kiel Institute for the World Economy*, Working Paper No.1640. Hindenburgufer 66, 24105 Kiel, Germany.
Obadar, O & Alaka, A (2010) "Influence of Resources Allocation in Education on Secondary School Students’ Outcome in Nigeria," *Academic Leadership: The Online Journal*, 8(4), Article 38.
OECD (2012), *Equity and Quality in Education: Supporting Disadvantaged Students and Schools*, OECD Publishing. http://dx.doi.org/10.1787/9789264130852-en
Omoeva, C. and Gale, C., (2016). Universal, but not free: Household schooling costs and equityeffects of Uganda’s Universal Secondary Education policy. *International Journal of Educational Development*, 50, pp.41-50.
Omukoba, H.O, Simatwa, E.M.& Ayodo, T.M. (2011). Contribution of income generating activities to financing secondary school education in Kenya: A case study of Eldoret municipality. *Education Research*, 2(2), 884-897.
Patrinos, H.A. (2016). *Why education matters for economic development*. Education for Global development. World Bank Group
Revealed through National Education Accounts. *UNESCO Institute for Statistics*, Montreal.http://uis.unesco.org/sites/default/files/documents/who-pays-for-what-in-education-nationalrevealed-through-accounts-2016-en_0.pdf
Right to Education Project (2012). *The right to education Country Fact Sheet; Uganda: saving the future generation*. www.right-to-education.org/sites/right-to-education.org/files/resource-attachments/RTE_Country_Factsheet_Uganda_2012.pdf
The Education (pre-primary, primary and post-primary) Act, (2008). Acts Supplement No.8, *The Uganda Gazette*, 91(44), 13.
Sa’ad, T.U & Sadiq, A.M. (2014). The Relevance of School Community Relationship on the Development of Primary Education in Azare Metropolis of Bauchi State, Nigeria. *OSR Journal of Research & Method in Education (IOSR-JRME)*. 4(6), 23-29
Sentongo, J. (2018). The Efficiency of Secondary Education in Sub-Saharan Africa: EESSA Project; The Case
of Uganda. Mastercard Foundation report, *Secondary Education in Africa: Preparing Youth for the Future of Work.*
Smith, W. C. & Baker, T. (2017). *From free to fee: Are for-profit, fee-charging private schools The solution for the world’s poor?* RESULTS Educational Fund.
Trading Economics (2019). *Uganda - Government expenditure on secondary education as % of GDP.* https://tradingeconomics.com/uganda/government-expenditure-on-secondary-education-as-%-of-gdp-%-wb-data.html
Uganda secondary education improvement project (2018). *Social assessment report.* Kampala, Author
Uganda technology and Management University (2016). *Curriculum review report on a framework for the harmonization of lower secondary teacher education curriculum with the revised lower secondary school curriculum, assessment and examination reform in Uganda.* UTAM, Kampala.
Uganda Investment Authority (2010). *Education Sector Profile.* Uganda Investment Authority.
Uganda Bureau of Statistics and ICF (2017), *Uganda Demographic and Health Survey, 2016: Key Indicators Report.* Kampala, UBOS and Rockville, Maryland, USA: UBOS and ICF.
Uganda National Commission for UNESCO, (2010). *Uganda: education for sustainable development implementation strategy.* Kampala, Author
UNHCR (2019). *Uganda country refugee response plan. The integrated response plan for refugees from South Sudan, Burundi and the Democratic Republic of the Congo.* UNHCR
UNESCO (2014), *Teacher Issues in Uganda: A shared vision for an effective teachers‘ policy.* UNESCO
UNESCO. (2010). *Sustainable development begins with education: how education can contribute to the proposed post-2015 goals.* UNESCO Digital Library
UNESCO. (2010). *Out-of-School Adolescents.* Montreal, Quebec: UNESCO Institute for Statistics.
UNESCO Institute of Statistics (2018). *Education Response Plan for Refugees and Host Communities in Uganda.* UNESCO.
UNESCO (2014) *Teacher Issues in Uganda: A shared vision for an effective teachers‘ policy.* UNESCO.
UNESCO-UIS (2011). *Financing Education in Sub-Saharan Africa. Meeting the Challenge of Expansion, Equity and Quality.* UNESCO/UIS, Montreal.
UNESCO-IIEP, Pole de Dakar, UNESCO-UIS (2016a). *Who Pays for What in Education? The Real Costs.* UNESCO
UNESCO-IIEP, Pole de Dakar, UNESCO-UIS (2016b). Methodology of national education accounts. *UNESCO Institute for Statistics, Montreal*
UNESCO (2014). *Sub-Saharan Africa 2013 EFA Report.* Global Education for All Meeting, Muscat, Sultanat of Oman.
United Nations, (2012). *The future we want: Outcome document of the United Nations Conference on Sustainable Development.* Rio de Janeiro, Brazil.
United Nations, (2017). *From Millennium Development Goals to Sustainable Development Goals: Laying the base for 2030.* Viantiane, Lao PDR.
United Nations, (2015). *The Millennium Development Goals Report 2015.* United Nations, New York
Urwick, J., Kisa, S., (2013). Science teacher shortage and the moonlighting culture: The pathology of the teacher labour market in Uganda, *International Journal of Educational Development.*
Walakira, E.J., D. Muhangi, S. Munyuwiny, F. Matovu, E. Awich, I. Ddumba Nyanzi, J. Kayiwa, J. Akellot, P. Mubiri, J. Majugo, A. Mutebi, M. Ruiz-Rodriguez, (2016). *The State of the Ugandan Child – An Analytical Overview.* Kampala/ Washington ton DC: USAI D/QED
World Development Report, (2018). *World Development, Author*
Wodon, K. (2016), *Uganda Note: Brief Summary for a first set of notes* World Bank, (2018), *Uganda job diagnostics/strategy draft.* World Bank
World Bank (2018a), Uganda job diagnostics/strategy draft.
World Bank (2018b). *Uganda Secondary Education Expansion Project (P166570): Project Information Document/ Integrated Safeguards Data Sheet (PID/ISDS).* Report No: PIDISDSC24831. World Bank.
Zebun, N.K. (2016). *Role of Education in Building Social Cohesion.* *International Journal of Secondary Education.* 4(2) 23-26
Zheng, P & Libertus, M. (2018). *The Role of Parental Education, Household Income, and Race on Parents’ Academic Beliefs and the Provision of Home Learning Opportunities for 4- to 8-Year-Old Children.* *Journal of Educational and Developmental Psychology,* 8(1), 118-132. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
5088,
1
],
[
5088,
10613,
2
],
[
10613,
15578,
3
],
[
15578,
21375,
4
],
[
21375,
27242,
5
],
[
27242,
33144,
6
],
[
33144,
39022,
7
],
[
39022,
44848,
8
],
[
44848,
49851,
9
],
[
49851,
54798,
10
],
[
54798,
59456,
11
],
[
59456,
63662,
12
]
]
} |
cb1f3ba5c71344af0b36eeace630644e450c261a | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 9,
"total-fallback-pages": 0,
"total-input-tokens": 30985,
"total-output-tokens": 8718,
"fieldofstudy": [
"Biology"
]
} | Analysis of the genetic diversity of the nematode parasite *Baylisascaris schroederi* from wild giant pandas in different mountain ranges in China
Xuan Zhou1, Yue Xie1, Zhi-he Zhang2, Cheng-dong Wang2, Yun Sun1, Xiao-bin Gu1, Shu-xian Wang1, Xue-rong Peng3 and Guang-you Yang1*
**Abstract**
**Background:** *Baylisascaris schroederi* is one of the most common nematodes of the giant panda, and can cause severe baylisascarosis in both wild and captive giant pandas. Previous studies of the giant pandas indicated that this population is genetically distinct, implying the presence of a new subspecies. Based on the co-evolution between the parasite and the host, the aim of this study was to investigate the genetic differentiation in the *B. schroederi* population collected from giant pandas inhabiting different mountain ranges, and further to identify whether the evolution of this parasite correlates with the evolution of giant pandas.
**Methods:** In this study, 48 *B. schroederi* were collected from 28 wild giant pandas inhabiting the Qinling, Minshan and Qionglai mountain ranges in China. The complete sequence of the mitochondrial cytochrome b (mtCytb) gene was amplified by PCR, and the corresponding population genetic diversity of the three mountain populations was determined. In addition, we discussed the evolutionary relationship between *B. schroederi* and its host giant panda.
**Results:** For the DNA dataset, insignificant *Fst* values and a significant, high level of gene flow were detected among the three mountain populations of *B. schroederi*, and high genetic variation within populations and a low genetic distance were observed. Both phylogenetic analyses and network mapping of the 16 haplotypes revealed a dispersed pattern and an absence of branches strictly corresponding to the three mountain range sampling sites. Neutrality tests and mismatch analysis indicated that *B. schroederi* experienced a population expansion in the past.
**Conclusions:** Taken together, the dispersed haplotype map, extremely high gene flow among the three populations of *B. schroederi*, low genetic structure and rapid evolutionary rate suggest that the *B. schroederi* populations did not follow a pattern of isolation by distance, indicating the existence of physical connections before these populations became geographically separated.
**Keywords:** Giant panda, *Baylisascaris schroederi*, Mountain ranges, Genetic diversity, Genetic structure, Phylogeography
**Background**
The giant panda, one of the world’s most iconic and threatened species, is considered a precious natural resource [1-3]. Giant pandas are currently restricted to the Qinling, Minshan, Qionglai, Daxiangling, Xiaoxiangling and Liangshan Mountains on the Eastern edge of the Tibetan plateau, and number about 1,600 individuals, of which 17.23% are in Qinling, 44.36% in Minshan and 27.38% in Qionglai [4,5]. *Baylisascaris schroederi* is one of the most common nematodes of giant panda, and can cause severe baylisascarosis in both wild and captive giant pandas [1,2,6]. The parasite mainly lives in the small intestine of the giant panda. However, the migrating *B. schroederi* larvae are sometimes also recovered post-mortem from the liver, lungs, heart or brain, which is associated with an increased clinical severity and pathologic manifestations, making *B. schroederi* one of the most serious parasites of giant panda [2].
Understanding organismal biodiversity based on the analysis of genetic variation has important implications for studying the evolutionary history and genetic structure of populations, and may therefore provide basic data for disease control. In general, genetic diversity, population hierarchical structure, population mutation rate, rate of gene flow and selective neutrality are the quantifiable components of genetic structure [7]. Parasitic nematodes are unexceptional to this universal biological rule [7]. During recent decades, a significant amount of genetic data has been generated on nematode parasite populations, in an attempt to explain micro-evolutionary processes [8]. However, unlike most free-living organisms, not only could the parasites own reproductive and transmission patterns but also their host genetics and behavior (e.g. migration) influence the genetic variations in parasites. Thus, genetic markers can be used to understand and depict the population genetic structure and diversity of a number of animal nematodes [8].
Previous studies of the giant pandas in the Qinling Mountain range indicated that this population is distinct from other wild giant panda populations, implying the presence of a new subspecies [9]. However, recent investigations based on whole genome data analysis have indicated that the Minshan and Qionglai-Daxiangling-Xiaoxiangling-Liangshan population were also genetically distinct [10]. In general, the genetic correlations between a population of hosts and their parasites are usually consistent, due to their similar demographic history and local adaptations. In this study, we investigated whether there is geographical genetic differentiation within these populations, and further identified whether the evolution of this parasite correlates with the evolution of the giant panda. The complete mtCytb gene (1107 bp) as a genetic marker in 48 isolates of *B. Schroederi* parasitizing 28 pandas inhabiting three different mountain ranges (Qinling, Minshan and Qionglai) was sequenced. Based on these samples, the genetic diversity and geographical relationships of the *B. Schroederi* populations were examined in the three relatively isolated mountain ranges, which the giant panda mainly inhabits.
**Methods**
**Ethics statement**
The present study was performed in strict accordance with the Guidelines and Recommendations for the Care and Use of Laboratory Animals of the Ministry of Health of the People’s Republic of China, and all protocols were reviewed and approved by the Research Ethics Committee of Sichuan Agricultural University (Ya’an, China).
**Sample collection, identification and DNA extraction**
Adult *B. Schroederi* (*n* = 48) were collected from 28 dead, injured or rescued giant pandas from three mountain ranges (Qinling, Minshan and Qionglai) during the period between May 2000 and May 2012 (details in Additional file 1; Figure 1). Individual worms were washed in physiological saline, identified morphologically as *B. Schroederi* [6], and stored at −20°C before DNA extraction. Total nematode genomic DNA was extracted from each specimen by standard proteinase K treatment and phenol/chloroform extraction [11], eluted into 30 μL TE buffer (pH 8.0) and stored at −20°C until analysis.
**Polymerase chain reaction (PCR) amplification, cloning and DNA sequencing**
Using the published mtDNA sequence of *B. Schroederi* in GenBank (HQ671081) [1], the primers Cytb-1 (GGT GCTATGCTCGGTTCAGG) and Cytb-2 (CCACTAAG CCCTCCAATT) were designed to amplify the whole mtCytb gene (~1500 bp) from the 48 nematode specimens. The optimal cycling conditions for specific and efficient amplification were obtained by testing varying annealing temperatures. PCR reactions (25 μL) containing 1 μL of template genomic DNA, 1 μL of each primer (10 pmol each), 12.5 μL of 2× Taq MasterMix (Beijing ComWin Biotech, Beijing, China), 9.5 μL of ddH2O were performed in a S1000 Thermal Cycler (Bio-Rad, USA) using the following conditions: initial denaturation at 94°C for 5 min; 35 cycles of 94°C for 1 min (denaturation), 50°C for 1 min (annealing), 72°C for 1 min (extension); followed by a final extension at 72°C for 10 min. The amplified PCR products were visualized on ethidium bromide-stained 1.0% agarose-TAE gels under UV light, excised and purified using spin columns (Wizard PCR Prep, Promega, USA). The purified products were cloned into the vector pMD19-T (TaKaRa, Dalian, China) using standard molecular procedures. Each clone was sequenced three times on an automatic DNA sequencer (ABI Applied Biosystems Model 3730) by Invitrogen (Shanghai, China).
**Sequence alignment and phylogenetic analyses**
All sequences were initially aligned using ClustalX 1.81 [12] with the following parameters: gap opening penalty = 10.0, gap extension penalty = 5.0, and then the alignments were modified by eye. A consensus phylogeny of the mtDNA haplotypes were estimated by two methods: maximum-parsimony (MP) tree using MEGA 5.10 [13] with the significance of each node estimated using 10000 bootstrap replicates of the data set, and the Bayesian inference (BI) using MrBayes 3.1 [14]. Four independent Markov chains were simultaneously run for the 1,630,000 replicates by sampling one tree per 1000 replicates with the Bayesian procedure. The first 250 trees were discarded as part of a burn-in procedure, and the remaining samples were used to generate a 50% majority rule consensus tree.
*Baylisascaris transfuga* (accession number: NC_015924.1)
for the mtCyb gene), a closely related species [1], was used as the outgroup in these phylogenies.
Population differentiation
The program Dna SP 5.0 [15] was used to calculate the nucleotide diversity within populations, number of informative sites, neutrality test (Fu’s Fs and Tajima’s D value), average number of nucleotide differences (K), nucleotide divergence (Dxy) and the net genetic distance (Da) between populations; the Fst statistic and value were computed from the haplotype frequencies using the program Arlequin Version 3.5 [16]. To evaluate whether genetic differentiation between populations was associated with geographical isolation of the mountains, analysis of molecular variance (AMOVA) [16] was used to examine genetic structuring among the sequences of the populations from the three mountain ranges: (i) Qinling Mountain range; (ii) Minshan Mountain range; (iii) Qionglai Mountain range. Gene flow was also estimated using Dna SP 5.0 [14].
The pairwise genetic distances between individual haplotypes were calculated following the Kimura 2-parameter model, using MEGA 5.0 [17]; phylogenetic relationships were estimated using Arlequin Version 3.5 [16] and then implemented in Network 4.6 [18] to illustrate the genetic relationships among the Cyb gene haplotypes in *B. Schroederi* parasitizing giant pandas in three different mountain ranges.
Population expansions
For population expansion analysis, the Arlequin Version 3.5 [16] coupled with 1000 simulations was used to obtain the distribution of Fu’s Fs value of neutrality and P value for each population, to assess whether population expansion occurred in *B. Schroederi*. To calculate the distribution of the number of pairwise differences between the 48 sample sequences, a
mismatch analysis was conducted using Dna SP 5.0 [15] over 1000 simulations.
Estimation of time differentiation
Due to the lack of fossil records for *B. schroederi*, calibration of the absolute rate of evolution of these parasites is generally problematic. Therefore, the evolutionary rate of the mtCytb gene from the parasite nematode *Heligmosomoides polygyrus* (which belongs to the Nematoda, the same as *B. schroederi* [3.5-3.7% K2P distance per million years] [19] in this study), was used to estimate the mean Kimura 2-parameter (K2P) distance between the 48 parasite mtCytb sequences in order to date the isolation time of *B. schroederi*.
Results
Haplotype and haplotype distribution
The complete sequence (1107 bp) of the mtCytb in 48 *B. schroederi* isolates was determined and deposited in GenBank under the accession numbers KC796955-KC797002. The sequence analysis showed 31 variable sites on the sequences (2.80%), including 24 singleton variable sites, defining 16 haplotypes (Additional file 2). Nucleotide diversity in the 48 samples was 0.201%, and nucleotide diversity among the populations isolated in the different mountain ranges ranged from 0.1489% (Minshan) to 0.222% (Qionglai). The Minshan population had the lowest diversity (0.1489 ± 0.1046%). The number of haplotypes, haplotype diversity and nucleotide diversity of each population are presented in Table 1.
Phylogenetic relationships
To clarify the phylogenetic relationship among the haplotypes of the three populations of *B. schroederi*, the sequence divergence values to construct hypothetical phylogenetic trees were calculated by the MP method and BI procedure. The two methods of phylogenetic analysis using the mtCytb marker led to very similar trees with shallow branches (Figure 2). The topology of the resultant trees was supported by the bootstrap values. Interestingly, as shown in Figure 2, haplotype H7, which was specific to the Minshan population, closely clustered with haplotypes H9 and H11 in the Qionglai population; while the Minshan population-specific haplotype H3 was close to the Qinling population-specific haplotype H8; three haplotypes in the Qinling population (H6, H12 and H15) clustered together; the other haplotypes formed isolated clusters. Several relationships among these haplotypes were also confirmed using the Network method (Figure 3). On the basis of the 16 haplotypes detected for the mtCytb gene, the network map revealed a star-like pattern around haplotype H1 (16 individuals, 33.3% of total haplotypes). Seven of the haplotypes were found in the Minshan population (H1-H7), three in the Qinling population (H1, H2, H8) and twelve in the Qionglai population (H1-H3, H6, H9-H16). While H1 and H2 were present in the populations from all three mountain ranges, haplotype H1, which had a cumulative frequency of 33.33%, was more frequent than H2 (18.8%). The Qionglai and Minshan populations shared haplotypes H3 and H6 (Figure 3). Furthermore, the haplotypes present in each mountain population were highly dispersed, and no obvious correlation of sampled clusters was detected.
Population genetic structure
To analyze the population genetic structure, several AMOVA analyses at different hierarchical levels were performed. Based on these AMOVA analyses, 98.03% of the variation occurred within the individual populations, and only 1.97% occurred across the three populations. As shown in Table 2, the highest *Fst* value was recorded for the Qinling and Qionglai populations (*Fst = 0.02875, P > 0.05*), while the lowest value was recorded for the Minshan and Qionglai populations (*Fst = 0.01911, P > 0.05*). The pairwise genetic distances between haplotypes are presented in Table 3; the pairwise genetic distances between haplotypes were generally low, except for H16 compared with H8 (1.7%), with a mean overall haplotype distance of 0.4% K2P. Nucleotide divergence (Dxy) among the three populations was low, ranging from 0.182% to 0.221%. Taken together, the three mountain range populations were not significantly differentiated.
Table 1 Summary of the genetic diversity of the three populations of *B. schroederi* collected from giant pandas inhabiting different mountain ranges
| Populations | No. individual | No. of haplotypes | No. of variable sites | Haplotype diversity (±SD) | Nucleotide diversity (±SD) | Tajima’s D | Fu’s Fs | Average no. of nucleotide differences (k) |
|-------------|----------------|------------------|----------------------|--------------------------|---------------------------|-----------|--------|----------------------------------------|
| Minshan | 14 | 7 | 10 | 0.7582 ± 0.1158 | 0.001489 ± 0.0001046 | −1.85262**| −2.69374* | 1.648 |
| Qinling | 8 | 3 | 8 | 0.6786 ± 0.1220 | 0.002065 ± 0.001438 | −1.25337 | 1.87372 | 2.286 |
| Qionglai | 26 | 12 | 21 | 0.8985 ± 0.0327 | 0.002221 ± 0.001382 | −1.99313***| −4.91076*** | 2.458 |
| Total | 48 | 16 | 31 | 0.8440 ± 0.0378 | 0.002009 ± 0.001252 | −1.69971* | −7.72261** | 2.224 |
*P < 0.05; **P < 0.01; ***P < 0.001.
SD standard deviation.
Population expansion
The complete data set of 48 B. schroederi sequences had a significant, large negative Fu’s Fs value (−7.72261, P < 0.005). To further explore the demographic history of the B. schroederi population, the Fs statistic values for Cytb for the three populations were also estimated. Overall, the Fs values were −2.69374 (P < 0.05), 1.87372 (P > 0.05) and −4.91076 (P < 0.005) for the Minshan, Qinling and Qionglai populations, respectively. Mismatch distribution analysis of the complete datasets revealed the presence of a major peak, which suggests that at least one expansion event occurred in the population demographic history of the B. schroederi population (Figure 4). It is possible that demographic expansions of the parasite occurred after introducing particular individuals into the endemic areas by anthropogenic movements of the giant panda.
Discussion
In recent years, successive studies employing molecular tools have attempted to understand the genetic diversity, population differentiation, and evolutionary or taxonomic relationships between closely related species. There is no doubt that there has been relatively wide interest in the genetic diversity of parasites. MtDNA markers have higher Fst values than nuclear sequences, and the mtDNA of nematodes evolves more quickly than the mtDNA of other parasites [8,20,21]. Recently, some mt genes (e.g., mtCytb) have been widely applied for analysis of the phylogenetic relationships between nematodes at the subspecies, species, genus and order levels [9,22,23]. Here, the complete Cytb gene was sequenced in 48 B. schroederi isolates sampled from 28 giant pandas inhabiting three different mountain ranges, and the corresponding population genetic diversity of the nematode parasites was subsequently determined.
In general, standard indices of genetic diversity are represented by the number of distinct haplotypes, haplotype diversity (h) and nucleotide diversity (π) [24]. Our results indicated that the three populations of B. schroederi from the different mountain ranges had a high haplotype diversity and low nucleotide diversity. This phenomenon is frequently observed in a number of other invertebrate animals with large standing population sizes and an extremely high fecundity [25,26], and may reflect the high matrilineal effective population sizes in B. schroederi, or indicate the occurrence of expansion after a period of low effective population size, as rapid population growth enhances the retention of new mutations.
Fst is appropriate for assessing the levels of differentiation within populations, as genetic drift is assumed to be the major factor that leads to genetic differentiation among closely related populations or over short-term evolution [27]. Our assessment of population genetic structure using the Fst index revealed an absence of...
significant differentiation among the three *B. schroederi* populations (including Qinling mountain), consistent with previous reports based on the ribosome DNA (ITS-1, ITS-2 and 5.8S) [9,28]. Furthermore, the high Nm value indicated the occurrence of strong gene flow among the *B. schroederi* populations over time. The most likely major contributor to gene flow is the movement of the giant panda. Generally, network approaches are better methods for representing genealogical relationships at a population level than traditional phylogenetic methods, and these approaches take account of several features associated with intraspecific gene evolution, including the persistence of ancestral haplotypes, the existence of multiple descendant haplotypes and often low levels of sequence variation [24]. Based on the network and phylogenetic tree, there was strong support in this study for the existence of a low structure among the *B. schroederi* populations from different mountain ranges, in contrast with that reported in giant pandas [10], suggesting absence of significant evolution correlation between *B. schroederi* and its host. Additionally, efforts were also made to determine whether a correlation existed between genetic differentiation and geographical distance. The calculated “within population” similarity indices showed that the populations did not follow a pattern of differentiation by distance. Collectively, these results clearly indicate the absence of significant geographical structuring among *B. schroederi*. Genetic diversity within populations is affected by their effective population sizes, immigration from other populations and mutation rates, whereas gene flow among populations and the time of divergence of a population from a common ancestral population affect the extent of genetic diversity [29].
In this study, we accounted for differences in sampling effort and the different numbers of individuals sequenced per population. However, it was difficult to obtain sufficient numbers of (sequenced) samples, especially from Qinling Mountains. Although the limit of three populations used for analysis might not be sufficient, strong correlations emerged when the phylogenetic relationship among these haplotypes, even for individual haplotypes (H16) contradicted the general trend (Figure 3). Some *B. schroederi* samples from different mountain ranges shared the same haplotypes, indicating that some of the mountain systems have a physical connection, perhaps due to human activity or host movement. Additionally, the haplotype network analysis revealed that the H1 haplotype is the most ancient haplotype (Figure 3), as previously reported ancient haplotypes often have a high frequency and display a trend towards a widespread geographic distribution [30].
Even though the present study was carried out on a small number of populations, genetic diversity was still observable for the parasite. As indicated by the phylogenetic trees (MP and BI) and network, the H8, H14 and H16 haplotypes clearly have a wide distribution, especially the H16 haplotype which could be derivatized by two paths (H1→H5→H16; H1→H14→H16). This phenomenon of genetic diversity in *B. schroederi* populations is likely to be influenced by a variety of factors. On the one hand, drainage systems and isolation of the mountain ranges, in particular geographical isolation, have been identified as factors leading to evolutionarily distinct clusters. The Min River valley and the existence of six relatively isolated mountain ranges that giant
Table 2 Pairwise comparisons based on parameters in the three *B. schroederi* populations
| Population 1 | Population 2 | Nm | Fst | Dxy | Da |
|--------------|--------------|------|--------|-------|-------|
| Minshan | Qinling | 9.68 | 0.02519| 0.00182| 0.00005|
| Minshan | Qionglai | 12.83| 0.01911| 0.00189| 0.00004|
| Qionglai | Qinling | 8.45 | 0.02875| 0.00221| 0.00006|
Nm, pairwise comparisons based on gene flow; Fst, genetic variance contained in a subpopulation relative to the total genetic variance; Dxy, nucleotide divergence; Da, net genetic distance for the three populations of *B. schroederi* collected from giant pandas inhabiting different mountain ranges; a, no significance.
panda inhabit, three of which were sampled in this study, might result in a distinct population structure among the populations. Second, unlike most free-living organisms, not only the parasites own reproductive and transmission patterns but also the genetics and behavior of the host could influence their genomic variation [31]. Host movement is an important determinant of population genetic structure in parasitic nematodes, and the frequent gene flow and weak population subdivisions will result from mobile vertebrate hosts [32].
The giant panda population has experienced two population expansions, two bottlenecks and two divergences. Evidence indicates that global changes in climate were the primary drivers of population fluctuation for millions of years, and human activities have negatively affected giant pandas for approximately 3,000 years [10]. Previous studies indicated a distinct Qinling Mountain giant panda population [33]. Subsequently, Zhao et al. (2012) carried out whole-genome re-sequencing of 34 wild giant pandas to continuously outline the history of the
| Table 3 Pairwise comparison of genetic distance and the percentage of haplotypes for the *B. Schroederi* population |
|----|----|----|----|----|----|----|----|----|----|----|----|
| H1 | H2 | H3 | H4 | H5 | H6 | H7 | H8 | H9 | H10 | H11 | H12 |
|----|----|----|----|----|----|----|----|----|----|----|----|
| 0.0001 | 0.0001 | 0.0001 | 0.0001 | 0.0001 | 0.0001 | 0.0001 | 0.0001 | 0.0001 | 0.0001 | 0.0001 | 0.0001 |
*Note: pairwise comparison of genetic distance is below diagonal and the percentage of haplotypes is above the diagonal.*
Figure 4 Mismatch distribution to test the expansion of 48 *B. Schroederi* isolates. The number of nucleotide differences between pairs of sequences is indicated along the x axis, and their frequency along the y axis.
giant panda population, and revealed that the Minshan and Qionglai-Daxiangling-Liangshan populations were also genetically distinct [10]. Prior to the events leading to the divergence of these populations, opportunities for B. Schroederi genetic exchange due to giant panda and, especially, human movements have may independently led to the presence of a large number of haplotypes in the B. Schroederi populations in each area. Indeed, the network map clearly revealed each B. Schroederi population has its own unique haplotypes (Figure 3).
Placing a time scale on molecular data remains a difficult problem, since very little information is available on the Cytb gene evolution rate, particularly in parasitic nematodes including ascaridoid species. In the present study, the nematode H. polygyrus was chosen as a comparison for exploring the evolutionary rate of Cytb gene in B. Schroederi due to its known Cytb evolution, taxonomic and phylogenetic relationships with B. Schroederi. Based on the molecular evolution rate of H. polygyrus Cytb gene (3.5-3.7% K2P distance per million years) [19], the B. Schroederi divergence time is estimated to be 0.108-0.114 million years (Myr). In addition, it is well-known that sequence diversity can result from an accelerated rate of nucleotide substitution [22]. In our study, the rate of evolution of the Cytb gene is approximately 7-fold higher in B. Schroederi than in the giant panda [34]. This result is consistent with phylogenetic studies highlighting the faster molecular evolutionary rate for mtDNA in nematodes, compared with other taxa [34]. A large number of similar studies have analyzed these genes in parasite taxa relative to their hosts [34,35]. In the present study, the 48 complete Cytb genes were sequenced and clearly demonstrated the genetic relationships and gene diversity of B. Schroederi populations from different mountain ranges.
Conclusion
This first investigation of the variability of the complete Cytb gene in 48 B. Schroederi isolates sampled from the three main giant panda habitats revealed a lack of population structure (Fst) and a relatively high gene flow for B. Schroederi, and indicates that the low genetic diversity might be a result of either a low DNA mutation rate, frequent movement of the hosts, or a combination of both of these factors. We also presume that B. Schroederi has experienced lineage re-arrangement, as indicated by the high rate of gene flow and lack of a sufficient differentiation period. In addition, expansion phenomena have occurred among the B. Schroederi populations, as most of the genetic variation identified in this study was found within populations, suggesting that a more intensive sampling strategy could possibly uncover the geographic structure of B. Schroederi in more detail.
Additional files
Additional file 1: Origins and numbers of B. Schroederi isolates and their hosts. a, GenBank accession numbers of mitochondrial Cytb sequences of these B. Schroederi samples.
Additional file 2: Polymorphic sites in complete mtCytb gene. The nucleotides present at each variable site among the 48 B. Schroederi isolates.
Abbreviations
mtCytb: Mitochondrial cytochrome b; ITS1: Internal transcribed spacer region 1; ITS2: The second internal transcribed spacer; can: Cytochrome c oxidase subunit 1; NAD1: NADH dehydrogenase subunit 1; MP: Maximum-parsimony; BI: Bayesian inference; K2P: Kimura 2-parameter; mtDNA: Mitochondrial DNA; Myr: Million years.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
YGY conceived of the project. ZZH, WCD, OXB, WXY, PXR and YGY collected samples. ZXY and SY performed the lab work. ZXY and XXY carried out the statistical analyses. ZXY wrote the initial draft of the manuscript. YGY and XY revised the manuscript. All authors read and approved the final manuscript.
Acknowledgments
This work was supported by grants from the Science & Technology Ministry, China (No. 200910188) and the Research Fund for the Chengdu Giant Panda Breeding (No. CPF-2012-13). We would like to thank Yu Wang and Ran He (College of Veterinary Medicine, Sichuan Agricultural University, China) for their technical assistance; Cai-Wu Li and his staff (Ya’an Bifengxia Research Base of Giant Panda Breeding, China) for materials.
Author details
1Department of Parasitology, College of Veterinary Medicine, Sichuan Agricultural University, Ya’an 625014, China. 2Chengdu Research Base of Giant Panda Breeding, Chengdu 610081, China. 3Department of Chemistry, College of Life and Basic Science, Sichuan Agricultural University, Ya’an 625014, China.
Received: 15 April 2013 Accepted: 6 August 2013
Published: 8 August 2013
References
1. Xie Y, Zhang Z, Wang C, Lan J, Li Y, Chen Z, Fu Y, Nie H, Yan N, Gu X, Wang S, Peng X, Yang G: Complete mitochondrial genomes of Baylisascaris Schroederi, Baylisascaris ailuri and Baylisascaris Transfuga from giant panda, red panda and polar bear. Gene 2011, 482:59–67.
2. Yang GY, Zhang ZH: Parasitic diseases of wildlife. Beijing: Science Press; 2013:458–465.
3. Zhao GH, Xu MJ, Zhu XQ: Identification and characterization of microRNAs in Baylisascaris Schroederi of the giant panda. Parasit Vectors 2013, 6:216.
4. Zhang ZH, Wei FW: Giant panda: ex-situ conservation theory and practice. Beijing: Science Press; 2006.
5. State Forestry Administration: The third national survey report on giant panda in China. Beijing: Chinese Science Press House; 2006:278–280. in Chinese.
6. Yang GY: Advance on parasites and parasitosis of giant panda. Chin J Vet Sci 1998, 18:158–208. in Chinese.
7. Hartl DL, Clark AG: Principles of Population Genetics. 4th edition. Sunderland MA: Sinauer; 2007.
8. Anderson TJ, Blouin MS, Beech RN: Population biology of parasitic nematodes: applications of genetic markers. Adv Parasitol 1998, 41:219–283.
9. Zhao GH, Li HM, Ryan UM, Cong MM, Hu B, Gao M, Ren WX, Wang XY, Zhang SP, Lin Q, Zhu XQ, Yu SK: Phylogenetic study of Baylisascaris Schroederi isolated from Qinling subspecies of giant panda in China based on combined nuclear 5.8S and the second internal transcribed spacer (ITS-2) ribosomal DNA sequences. Parasitol Int 2012, 61:497–500.
Toward defining the course of evolution: minimum change for a specific tree topology. *Syst Zool* 1971, 20:406–416.
Excoffier L, Laval G, Schneider S. DnaSP version 3: an integrated program for molecular population genetics and molecular evolution analysis. *Bioinformatics* 1999, 15:174–175.
Lavery S, Moritz C, Fielder DR. Shallow population histories in deep evolutionary networks. *Mol Ecol* 2013, 22:456–471.
Liu GH, Wang Y, Song HQ, Li MW, Ai L, Yu XL, Zhu XQ. Characterization of the complete mitochondrial genome of *Spirocerca lupi*: sequence, gene organization and phylogenetic implications. *Parasit Vectors* 2013, 6:45.
Lavery S, Montz C, Fielder DR. Indo-Pacific population structure and evolutionary history of the coconut crab *Birgus latro*. *Mol Ecol* 1996, 5:557–570.
Grant WAS, Bowen BW. Shallow population histories in deep evolutionary lineages of marine fishes: insights from sardines and anchovies and lessons for conservation. *J Hered* 1998, 89:415–426.
Weir BS, Cockernham CC. Estimating F-Statistics for the analysis of population structure. *Evolution* 1984, 38:1358–1370.
Lin Q, Li HM, Gao M, Wang XY, Ren WX, Cong MM, Tan XC, Chen CX, Yu SK, Zhao GH. Characterization of *Baylisascaris Schroederi* from Qinling subspecies of giant panda in China by the first internal transcribed spacer (ITS-1) inferred from mitochondrial COI gene. *Ver Parasitol* 2009, 164:248–256.
Submit your next manuscript to BioMed Central and take full advantage of:
- Convenient online submission
- Thorough peer review
- No space constraints or color figure charges
- Immediate publication on acceptance
- Inclusion in PubMed, CAS, Scopus and Google Scholar
- Research which is freely available for redistribution
Submit your manuscript at www.biomedcentral.com/submit | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3436,
1
],
[
3436,
8958,
2
],
[
8958,
10718,
3
],
[
10718,
16180,
4
],
[
16180,
19042,
5
],
[
19042,
23343,
6
],
[
23343,
25191,
7
],
[
25191,
31393,
8
],
[
31393,
33178,
9
]
]
} |
2f52e48fbef56aaef719085683fca11db6bab682 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 9,
"total-fallback-pages": 0,
"total-input-tokens": 44864,
"total-output-tokens": 12568,
"fieldofstudy": [
"Medicine"
]
} | Since January 2020 Elsevier has created a COVID-19 resource centre with free information in English and Mandarin on the novel coronavirus COVID-19. The COVID-19 resource centre is hosted on Elsevier Connect, the company's public news and information website.
Elsevier hereby grants permission to make all its COVID-19-related research that is available on the COVID-19 resource centre - including this research content - immediately available in PubMed Central and other publicly funded repositories, such as the WHO COVID database with rights for unrestricted research re-use and analyses in any form or by any means with acknowledgement of the original source. These permissions are granted for free by Elsevier for as long as the COVID-19 resource centre remains active.
Mortality in children with positive SARS-CoV-2 polymerase chain reaction test: Lessons learned from a tertiary referral hospital in Indonesia
Rismala Dewi¹, Nastiti Kaswandani¹, Mulya Rahma Karyanti, Darmawan Budi Setyanto, Antonius Hocky Pudjiadi, Aryono Hendarto, Mulyadi M. Djer, Ari Prayitno, Irene Yuniar, Wahyuni Indawati, Yogi Prawira, Setyo Handryastuti, Hikari Ambara Sjakti, Eka Laksmi Hidayati, Dina Muktiarti, Amanda Soebadi, Niken Wahyu Puspaningtyas, Riski Muhamin, Anisa Rahmadhany, Gilbert Sterling Octavius, Henny Adriani Puspitasari, Madeleine Ramdhani Jasin, Tartila Tartila, Nina Dwi Putri*
Department of Paediatrics, Dr. Cipto Mangunkusumo National Central Hospital, Faculty of Medicine, Universitas Indonesia, Jakarta, Indonesia
A R T I C L E I N F O
Article history:
Received 3 December 2020
Received in revised form 5 April 2021
Accepted 7 April 2021
Keywords:
COVID-19
SARS-CoV-2
Outcome
Children
Indonesia
A B S T R A C T
Background: The incidence of coronavirus disease 2019 (COVID-19) is still increasing rapidly, but little is known about the prevalence and characteristics of fatal cases in children in Indonesia. This study aimed to describe the characteristics of children with COVID-19 with fatal outcomes in a tertiary referral hospital in Indonesia.
Methods: This cross-sectional study used data collected from the medical records of patients with COVID-19 admitted to Dr. Cipto Mangunkusumo Hospital, Jakarta, Indonesia from March to October 2020.
Results: During the study period, 490 patients were admitted and diagnosed with suspected and probable COVID-19. Of these patients, 50 (10.2%) were confirmed to have COVID-19, and 20 (40%) had a fatal outcome. The fatality rate was higher in patients aged ≥10 years, categorized with severe disease upon admission, PaO₂/FiO₂ ratio <300 mmHg and chronic underlying diseases. The most common clinical manifestations were generalized symptoms, while acute respiratory distress syndrome (8/20) and septic shock (7/20) were the two most common causes of death. Increased procalcitonin, D-dimer, lactate dehydrogenase and presepsin levels were found in all fatal cases. One patient met the criteria of multisystem inflammatory syndrome in children.
Conclusion: Our work highlights the high mortality rate in paediatric patients with positive SARS-CoV-2 polymerase chain reaction test. These findings might be related to or co-incident with COVID-19 infection. Further studies are needed to improve understanding of the role of severe acute respiratory syndrome coronavirus-2 in elaborating the mechanisms leading to death in children with comorbidities.
© 2021 The Author(s). Published by Elsevier Ltd on behalf of International Society for Infectious Diseases. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Introduction
Severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) was first reported in Wuhan, Hubei Province, China in December 2019. The disease caused by this virus later became known as coronavirus disease 2019 (COVID-19) (World Health Organization, 2020a). Most reports have indicated that children and adolescents comprise a small proportion of confirmed cases, and that these populations are less likely to be severely affected than adults (Castagnoli et al., 2020; Dong et al., 2020a; Ludvigsson, 2020; Rodriguez-Morales et al., 2020). Furthermore, studies have reported a good health status in children with underlying chronic conditions and those on immunosuppressive treatment (Nicastro et al., 2020; Di Giorgio et al., 2021). One study reported 80 deaths in children aged 0–14 years in a population of 137,047,945, resulting in a mortality rate of 0.06 per 100,000 population (Bhopal et al., 2020).
However, in early May 2020, an increasing amount of evidence emerged from the UK, the USA and Europe of a different manifestation of COVID-19 in paediatric patients, namely...
hyperinflammatory shock with multi-organ involvement (Riphagen et al., 2020). This condition is interchangeably referred to as paediatric inflammatory multisystem syndrome temporally associated with SARS-CoV-2 (PIMS-TS) or multisystem inflammatory syndrome in children (MIS-C) associated with COVID-19 (Centers for Disease Control and Prevention, 2020; Royal College of Paediatrics and Child Health, 2020; World Health Organization, 2020b). The clinical manifestations of MIS-C and PIMS-TS are both distinct from and similar to other inflammatory syndromes in children, such as Kawasaki disease, Kawasaki disease shock syndrome and toxic shock syndrome (World Health Organization, 2020).
Systematic reviews have shown that among 662 patients who fulfilled the MIS-C criteria, only 11 deaths (1.7%) were reported (Ahmed et al., 2020; Jiang et al., 2020). However, as of 20 July 2020, the Indonesian Pediatrician Society had reported 2,712 confirmed paediatric cases of COVID-19 with 51 deaths (1.9%) (Pulungan, 2020). There are limited data on the clinical characteristics of paediatric cases with COVID-19. More reliable data are needed to determine the disease burden to create better screening and intervention strategies for the Indonesian paediatric population. For these reasons, this study aimed to describe the characteristics of paediatric patients with fatal outcomes with positive COVID-19 and/or MIS-C tests admitted to a tertiary referral hospital in Indonesia.
### Materials and methods
**Patients, clinical data and sample collection**
This is a cross sectional study with data collected from the medical records of suspected and confirmed paediatric cases of COVID-19 at the study site.
| Table 1 | Demographic data of the confirmed paediatric cases of coronavirus disease 2019 (n = 20) at Dr. Cipto Mangunkusumo National Central Hospital, Indonesia, 2020. |
|---------|--------------------------------------------------------------------------------------------------------|
| **Parameter** | **Results** |
| Sex (n = 20) | |
| Male | 10 |
| Female | 10 |
| BMI, median (range) (n = 15) | |
| Severely underweight and underweight | 3 |
| Normal weight | 9 |
| Overweight and obese | 3 |
| Age in years, median (range) (n = 20) | |
| <1 | 0 |
| 1–5 | 5 |
| 5–10 | 5 |
| >10 | 10 |
| Rapid antibody tests (n = 20) | |
| IgM positive | 0 |
| IgG positive | 2 |
| Not tested | 7 |
| Source of RT-PCR samples (n = 20) | |
| Naso-oropharyngeal | 19 |
| Sputum | 1 |
| Ct values (N gene Cq), median (range) | |
| First sample (n = 9) | 33.2 (21.41–36.26) |
| Second sample (n = 6) | 33.8 (22.78–38.1) |
| Third sample (n = 3) | 31.3 (22.4–32.47) |
| Fourth sample (n = 2) | 30.0 (20.05–35.88) |
| Clinical manifestations | |
| Generalized symptoms | 12 |
| Respiratory symptoms | 9 |
| Gastrointestinal symptoms | 8 |
| Neurologic symptoms | 3 |
| Comorbidities | |
| Kidney diseases | 8 |
| Cardiovascular diseases | 6 |
| Malignancy | 6 |
| Neurological diseases | 4 |
| Overweight and obesity | 3 |
| Underweight | 3 |
| Burn injury | 2 |
| Systemic lupus erythematosus | 2 |
| Deep vein thrombosis | 1 |
| Acute appendicitis with generalized peritonitis | 1 |
| Biliary atresia | 1 |
| Intestinal tuberculosis | 1 |
| Number of comorbidities | |
| Single | 4 |
| Multiple (two or more) | 16 |
| Exposure to healthcare facilities or professionals (n = 20) | |
| History of contact with suspected or confirmed cases (n = 20) | |
| Shock (n = 20) | |
| Septic | 9 |
| Not in shock | 7 |
| Hypovolaemic | 4 |
BMI, body mass index; RT-PCR, reverse transcriptase polymerase chain reaction; Ct, cycle threshold.
COVID-19 admitted to Dr. Cipto Mangunkusumo National Central Hospital, Jakarta, a tertiary referral hospital in Indonesia, from March 2020 (when the first Indonesian case of COVID-19 was announced) to October 2020 (World Health Organization, 2020c). Before the pandemic, Dr. Cipto Mangunkusumo National Central Hospital was a general hospital that serves pediatric and adult patients with 1001 beds capacity. During the pandemic, the new pediatric unit was converted into a COVID-19 isolation unit that serves 237 beds (13 beds for family-centered wards, eight beds for children only, eight beds for paediatric intensive care unit (PICU), and eight beds for neonatal intensive care unit (NICU) isolation room). The total bed capacity was reduced to 888 beds due to a lack of personnel. In 2020, 31,075 patients across all ages visited the emergency department, with 1373 (4.41%) patients confirmed as positive for COVID-19. Demographic data (age, sex, weight and height), COVID-19 status [rapid antibody test results, reverse transcriptase polymerase chain reaction (RT-PCR) results, cycle threshold (Ct) values], signs and symptoms (such as fever and lethargy), respiratory symptoms, gastrointestinal symptoms, and neurological symptoms, comorbidities, PICU status, cause of death and laboratory data were obtained. We also obtained the length of stay in PICU and the total length of stay from admission to discharge or death.
This study included all paediatric patients (0–18 years old) who had tested positive for COVID-19 infection using RT-PCR from any sample involving a swab or other specimen and had a fatal outcome. Probable cases were excluded from this study. We used WHO's guideline to define probable cases as (1) a suspect case for whom testing for the COVID-19 virus is inconclusive (inconclusive being the result of the test reported by the laboratory); or (2) a suspect case for whom testing could not be performed for any reasons (WHO, 2020d).
Patients were classified according to their clinical presentation upon admission as: (1) asymptomatic (absence of signs and symptoms associated with COVID-19, normal clinical imaging, but positive ribonucleic acid SARS-CoV-2 test); (2) mild (presence of symptoms limited to upper respiratory tract (including fever, fatigue, myalgia, cough, sore throat, runny nose or nasal congestion) or gastrointestinal symptoms (including nausea, vomiting and abdominal pain, with normal lung auscultation); (3) moderate (presence of symptoms mentioned in the mild category together with clinical signs and symptoms of pneumonia but without hypoxaemia); (4) severe (presence of signs and symptoms mentioned above together with dyspnoea, central cyanosis and oxygen saturation of <92%); and (5) critical (presence of acute respiratory distress syndrome, respiratory failure, encephalopathy, myocardial injury, coagulation dysfunction and acute kidney injury) (Dong et al., 2020b).
The Centers for Disease Control and Prevention (CDC) criteria for MIS-C associated with COVID-19 was used in this study. The
| Parameter | Results |
|-----------|---------|
| Days to worsening clinical manifestations since the first day of admission, median (range) (n = 20) | 3 (0–50) |
| Days to intubation since the first day of admission, median (range) (n = 10) | 5.5 (0–55) |
| Days to ICU admission since the first day of admission, median (range) (n = 16) | 2.5 (0–50) |
| Days spent in ICU, median (range) (n = 16) | 2.5 (0–50) |
| Days to death since the first day of admission, median (range) (n = 20) | 1.5 (0–11) |
| Days to death since the onset of first clinical manifestations, median (range) (n = 20) | 7 (0–51) |
| Dialysis (n = 20) | 8 (1–65) |
| Fluid resuscitation (n = 20) | No |
| Medications (n = 20) | Yes |
| Vasopressors | Yes |
| Antibiotics | Yes |
| Steroids | Yes |
| IVIG | Yes |
| Enoxaparin | Yes |
| Lopinavir + ritonavir | Yes |
| Ventilator use (n = 20) | No |
| PaO₂/FIO₂ ratio (n = 16) | >300 |
| Clinical condition upon admission (n = 20) | Mild |
| Acute respiratory distress syndrome | Yes |
| Septic shock | Yes |
| Hypovolaemic shock | Yes |
| Encephalopathy sepsis | Yes |
| Medical and surgical bleeding | Yes |
| Pulmonary thrombosis | Yes |
| Multi-organ dysfunction syndrome | Yes |
| MIS-C (n = 20) | Present |
| Absent | Absent |
ICU, intensive care unit; IVIG, intravenous immunoglobulin; MIS-C, multisystem inflammatory syndrome in children.
Organization oropharyngeal normality samples within values were involvement (cardiac, renal, respiratory, haematological, gastrointestinal, dermatological or neurological); other plausible alternative diagnoses had been excluded; and (5) they had a positive RT-PCR, serology or antigen test or COVID-19 exposure within 4 weeks preceding the onset of symptoms (Centers for Disease Control and Prevention, 2020). The CDC criteria were chosen because they were released earlier than the World Health Organization criteria. However, later on, our national guideline adopted the World Health Organization criteria of MIS-C.
Detection of COVID-19 infection
All naso-oropharyngeal and sputum/endotracheal tube aspirate samples were tested for the presence of SARS-CoV-2, and the N gene Cq was used as the parameter for the RT-PCR target. The standard protocol for obtaining the samples was via naso-oropharyngeal swabs with a minimum of two samples within a 1-day interval. If the samples were positive, subsequent samples were obtained every 5–7 days until conversion was achieved. Ct values >40 (detection limit) were reported as negative, while Ct values <37 were considered positive (Ile et al., 2020). A medium load (Ct value 37–40) requires confirmation via at least one repeat sample in the study institute.
Statistical analysis was done using IBM SPSS 22.0 (Statistical Package for the Social Sciences, IBM Corp., Armonk, NY, USA). The normality test was carried out using the Kolmogorov-Smirnov method, and if the p-value is greater than 0.05, the data were considered to have a normal distribution. The data would be presented in mean and standard deviation if the distribution was normal, while median and range would be used if it was not normal.
Results
In total, 490 paediatric cases were categorized as suspected or probable COVID-19. Of these, 50 (10.2%) cases were confirmed by RT-PCR. Among the confirmed cases, 20 patients (40%) died and were included in this analysis. The mortality rate in confirmed cases of COVID-19 in children was 40%.
There was no difference in mortality between males and females in patients positive for COVID-19 (Table 1). Most (18/20) of the patients had previous exposure to healthcare facilities or professionals. The highest mortality rate was seen in patients aged >10 years and those placed in the severe category upon admission. Most (12/20) patients presented with generalized or systemic symptoms such as fever, malaise, myalgia and fatigue. Kidney diseases, such as nephritic lupus with secondary hypertension, acute kidney failure and chronic kidney disease, were the most common comorbidities, found in eight patients, with four patients requiring dialysis. Three of these patients were previously on chronic dialysis, while one of them was receiving continuous renal replacement therapy regularly before their COVID-19 diagnosis.
Septic shock was the most common type of shock (9/20) seen among subjects, and acute respiratory distress syndrome was the most common cause of death (8/20). On average, it took 3 days from admission and 5.5 days from the onset of the first clinical manifestations of COVID-19 for their conditions to worsen (Table 2). Of the 20 patients who died, 16 required admission to the intensive care unit (ICU), with the median number of days from hospital admission to ICU admission being 2.5 days (range 0–50). Vasopressors (19/20) and antibiotics (18/20) were the two most common medications used during hospitalization, while mechanical ventilation was needed in half of the patients. According to national and hospital guidelines, patients were given antibiotics and antivirals according to the clinical severity of their symptoms. Most notably, one of the patients met the MIS-C criteria.
SARS-CoV-2 RNA was detected in a sputum sample/endotracheal tube aspirate in one patient and naso-oropharyngeal swab specimens in the remaining patients. The median first-sample Ct value was 33.2, with a range of 21.41–36.26.
All patients were admitted with increased procalcitonin, D-dimer, lactate dehydrogenase and presepsin levels. White blood cell (WBC) count, platelet count, lactic acid, prothrombin time and creatinine levels were normal in most patients on Days 1 and 3 (Table 3).
Detailed clinical characteristics of each subject are described in Table 4. Two cases were asymptomatic (Cases 1 and 9), as they were initially admitted for severe burns affecting body surface areas of 59% and 45%, respectively, and were later diagnosed with COVID-19 by RT-PCR testing. Notably, sixteen out of twenty patients had more than one comorbidity. Echocardiography was performed on one patient (Case 5) and showed pericardial effusion with an ejection fraction of 77%. Two cases presented with moderate illness (Cases 3 and 10) and passed away due to surgical complications related to bleeding. The preliminary findings showed that increments in creatinine levels between Day 1 and
### Table 3
| Parameter | Normal value | Median (range) | Elevated, n | Normal, n | Decreased, n |
|----------------------------|--------------|----------------|-------------|------------|--------------|
| Haemoglobin (g/dL) | 12.0–15.0 | 10.2 (3.7–17.6) | 2 | 5 | 13 |
| White blood cells (10³/μL) | 4.0–10.0 | 12.7 (1.22–24.6) | 8 | 9 | 3 |
| Platelets (10³/μL) | 150–410 | 274 (1–818) | 1 | 12 | 7 |
| CRP (mg/L) | <5.0 | 25.6 (5–472.5) | 16 | 1 | N/A |
| Procalcitonin (ng/mL) | <0.05 | 3.1 (0.1–48.1) | 18 | 0 | N/A |
| Fibrinogen (mg/dL) | 200–400 | 4475 (44–1118) | 5 | 2 | 3 |
| D-dimer (μg/L) | <440 | 5490 (1120–6820) | 9 | 0 | N/A |
| Ferritin (ng/mL) | 20–200 | 1411.4 (1–28,740) | 5 | 1 | 1 |
| SGOT (U/L) | 10–40 | 69 (14–3173) | 10 | 6 | 1 |
| SGPT (U/L) | 5.9–37 | 41 (10–1095) | 9 | 8 | 0 |
| Lactic acid (mmol/L) | 0.7–2.5 | 1.6 (0.6–4.6) | 2 | 7 | 1 |
| Prothrombin time (s) | 9.8–12.6 | 13.2 (10.1–120) | 9 | 9 | 0 |
| Activated partial prothrombin time (s) | 31.0–47.0 | 50.4 (26.9–180) | 10 | 7 | 1 |
| Lactate dehydrogenase (B) (U/L) | 125–220 | 713.5 (364–1063) | 2 | 0 | 0 |
| Presepsin (pg/mL) | <300 | 2274.5 (1257–3292) | 2 | 0 | N/A |
| Creatinine (Day 1) (mg/dL) | 0.22–0.59 | 0.6 (0.2–5) | 9 | 10 | 0 |
| Creatinine (Day 3) (mg/dL) | 0.22–0.59 | 0.7 (0.1–4.9) | 4 | 5 | 0 |
**Note:** CRP, C-reactive protein.
| No | PCR-positive sample | Age (years) | Mean Ct values | Rapid test | Clinical manifestations associated with COVID-19 | Complications | Shock | Mechanical ventilation | Fluid resuscitation | Medications | Laboratory findings | Clinical condition upon admission | Days in PICU | Days to intubation | Days to death | Cause of death |
|----|---------------------|------------|----------------|------------|------------------------------------------------|--------------|-------|------------------------|-------------------|-------------|-------------------|-----------------------------|-----------|------------------|--------------|----------------|
| | Nasopharyngeal | | | | | | | | | | | | | | | |
| | Specimen/ETT | | | | | | | | | | | | | | | |
| 1 | - | 78.3 | NA | IgG | Asymptomatic | Superficial to full-thickness burn with 5% of body surface area affected; multi-organ dysfunction syndrome; sepsis; NK/CD8+ T cells | Hypovolemic | Yes | Yes | 17% WBC, neutrophil count, PCT, CRP, SCr, ALT | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 2 | + | 6.7 | 35.7 | IgG | Fever, fatigue, mumps | Neuroinflamma | No | No | 17% WBC, neutrophil count, PCT, CRP, SCr, ALT | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 3 | - | 14.7 | 34.5 | Neg | Fever, malaise, nausea, vomiting | Acute lymphopenic leukaemia, anaemia, neutropenia, moderate protein-energy malnutrition, anaemia, thrombocytopenia, cardiomyopathy, gastrointestinal, toxic liver changes | No | No | Yes | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 4 | - | -14.8 | NA | Not | Fever | Suspected deep vein thrombosis and unexplained severe protein-energy malnutrition | No | No | No | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 5 | + | 15.1 | 30.025 | Not | Fever, dyspnoea, rash, mucosal changes | ARDS, septic lung grade II | No | Yes | Yes | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 6 | - | 11.1 | 38.1 | Not | Diarrhoea, altered mental status | Hepatic failure, intrahepatic cholestasis, CMV infection, prolonged diarrhea coagulation defects, metabolic encephalopathy | Hypovolemic | Yes | Yes | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 7 | - | 13.3 | 33.16 | Not | Fever, cough, rhinorrhea | Chronic kidney failure, essential hypertension, anemia, ulceration, obesity | No | Yes | Yes | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 8 | + | 3.3 | 54.245 | Not | Diarrhoea, vomiting | Haemorrhagic due to rupture of oesophagale varices, biliary atresia, acute diarrhoea with mild-moderate dehydration, hyperaemia, marasmus, haemorrhagic, hypotension, anaemia, severe underweight | No | Yes | Yes | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 9 | + | 1.8 | 36.26 | Neg | Asymptomatic | Superficial to mild-throat burn with 45% of body surface area affected, haemorrhage due to stress ulcers | No | No | Yes | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 10 | + | 17.8 | 21.225 | Neg | Cough, dyspnoea | Polypneusus tuberculosis, ac тре, obstruction of bile duct, protein-energy malnutrition | Hypovolemic | Yes | Yes | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 11 | + | 13.7 | 2718 | Neg | Cough (haemoptysis), nasal congestion, dyspnoea | ARDS, SIL, chronic kidney failure, suppurative cutaneous pleurisy, septic encephalopathy | No | Yes | Yes | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 12 | - | 1.7 | NA | Not | Fever, diaphoresis, altered mental status, seizure | Septic encephalopathy, underweight | Hypovolemic | Yes | Yes | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 13 | - | 9.1 | NA | Not | Fever, abdominal pain | Acute myeloid leukaemia, febrile neutropenia, respiratory failure | Septic | No | No | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 14 | - | 14.6 | NA | Not | Cough, dyspnoea, lymphopenopathy | Leukaemia, toxic liver disease, cardiomegaly | Septic | No | No | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 15 | - | 9.3 | NA | Not | Cough, dyspnoea | SLE | Septic | No | No | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 16 | - | 17.9 | NA | Not | Fever, cough, nasal congestion, dyspnoea | Rhabdomyosarcoma, hypoplasia, acute kidney failure, encephalopathy, sepsis, respiratory failure | Septic | No | No | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
| 17 | + | 1 | NA | Not | Fever, cough, nasal congestion, dyspnoea | Encephalopathy, gastrointestinal | Septic | No | No | Neutrophil count, metabolic acidosis | Critical | 0 | 0 | MODS |
Day 3 led to prolonged hospitalization, except for one case (Case 13). This patient showed a decrease in creatinine level, although no association or significance can be inferred.
**Discussion**
The clinical and laboratory characteristics of paediatric patients with COVID-19 with fatal outcomes were studied. The proportion of COVID-19-associated deaths in this study is higher than the COVID-19 case fatality rate in 42 states in the USA, which reported mortality rates of 0–0.23% as of 22 October 2020 (American Academy of Pediatrics, 2020). It is also higher than the 1.9% nationwide case fatality rate reported in Indonesia (Pulungan, 2020). As the study centre is a national tertiary referral hospital, patients often present with one or more pre-existing underlying chronic diseases that will affect their prognoses and mortality. Nearly all of the patients in this study had at least one comorbidity, with the most common being kidney disease (8/20 cases), followed by malignancy and cardiovascular disease (6/20 cases each). Chronic kidney disease is associated with a poorer prognosis due to disturbances in the innate and adaptive immune responses, rendering such patients more susceptible to all infections (Gagliardi et al., 2020). The present findings differ from those of another study which reported that 86% of patients had at least one comorbidity, with the most prevalent pre-existing conditions being medically complex conditions (40%), immunosuppression or malignancy (23%), and obesity (15%). There were two deaths reported in this study, and both of the patients who died had comorbidities (Shekerdemian et al., 2020).
Obesity and overweight are the two comorbidities frequently mentioned as risk factors for mortality in COVID-19 or MIS-C in children (Ahmed et al., 2020; Jiang et al., 2020). However, underweight is a comorbidity that has not been discussed in detail to date, especially in children. Studies performed in adult populations show conflicting results; one study found that underweight individuals tended to trend towards increased risk of contracting COVID-19, but this trend was not significant (Jung et al., 2020). Another retrospective cohort study of 2466 hospitalized adults found that underweight individuals had a borderline significant association with increased risk of death or intubation (Anderson et al., 2020).
Another reason accounting for the high mortality rate seen in this study is the severity of clinical manifestations upon presentation. One review found that non-mild disease, defined as pneumonia or need for hospitalization, accounted for 33.3% of cases. In contrast, more severe illness accounted for 9.1% of cases, which contrasts with the 55% and 35% rates, respectively, that were observed in the present study (Anderson et al., 2020). Among 58 patients in three studies, 35 required invasive mechanical ventilation (60.3%) (Belhadjer et al., 2020; Escosa-Garcia et al., 2020; Toubiana et al., 2020). While this number is slightly lower in the present study (10/20), it is lower because six parents signed ‘do not resuscitate’ forms, making treatment suboptimal for these patients. Other reasons that could explain the high mortality rate in the present study are overcrowding in the hospital wards due to the sudden surge of new cases of COVID-19, delayed presentation of chronic patients to the hospital and coupled with the lack of human resources to combat the pandemic initially.
The median first-sample Ct value in this study was 33.2, similar to the results from a study in China that examined 10 paediatric patients (median Ct value of 33.5) (He et al., 2020). Low SARS-CoV-2 Ct values were associated with the increased likelihood of progression to more severe disease, increased mortality, and the presence of biochemical and hematological markers (Rao et al., 2020). According to one study, the median Ct value of the present
study is classified as a low viral load (30–39.9) (Karahan et al., 2020).
In the present study, mortality also tended to be higher in patients with PaO2/FiO2 ratios ≤300 mmHg, in line with other studies performed in Europe (Wendel et al., 2020). The present study produced results similar to those of other meta-analyses (Elshazli et al., 2020; Henry et al., 2020) and one review study (Leticia et al., 2020), which reported that increased D-dimer, fibrinogen, procalcitonin, CRP and ferritin levels, as well as low haemoglobin levels were associated with severe disease and mortality. Although increased WBC counts were consistently cited as one of the significant predictors for severe disease (Elshazli et al., 2020; Henry et al., 2020; Leticia et al., 2020), nine of the patients in the present study had normal WBC counts, which might be explained by the inclusion of six patients with haematological malignancies with the potential to impair WBC count.
One patient (Case 18) met the CDC criteria for MIS-C, meaning that a positive RT-PCR, serology or antigen test or COVID-19 exposure within 4 weeks preceding the onset of symptoms was required (Ahmed et al., 2020; Jiang et al., 2020). Although two patients presented as IgG-positive on serologic testing, they did not meet the other criteria, as one presented with severe burns and no COVID-19-related symptoms (Case 1). In contrast, the other case (Case 2) presented with a high Ct value, indicating recent infection. There was limited knowledge concerning MIS-C early in the pandemic; therefore, limited data were available on its physical manifestations, such as Kawasaki-like symptoms, and diagnostic SARS-CoV-2 serology, cardiac markers and echocardiography, which may have led to the underdiagnosis of MIS-C in the study patients.
Patients were managed conservatively, as almost all paediatric guidelines recommend mainly supportive treatment. Most patients present with presumed sepsis and/or pneumonia as evidenced by clinical manifestations and elevated inflammatory markers. Hence, empirical antibiotics were given until PCR or culture and sensitivity results came back. The practice of prescribing empirical antibiotics follows the national and hospital guidelines which recommend administering antibiotics according to clinical severity (Kementerian Kesehatan Republik Indonesia, 2020). Antivirals were also given to some patients due to underlying comorbidities resulting in immunocompromised conditions. As the knowledge of COVID-19 is always evolving, knowledge about the use of intravenous immunoglobulin, steroids and low-molecular-weight heparin for prophylaxis of thrombosis was not widespread early in the pandemic. It hence reflected the lack of specific treatments for COVID-19. This study also shows that most patients were exposed to healthcare facilities. This highlights the urgent need for infection prevention education protocols, especially for children with chronic medical conditions necessitating multiple hospital visits.
This study has several limitations. First, as Dr. Cipto Mangunkusumo Hospital is a referral hospital for managing patients with COVID-19, especially those with comorbidities, the mortality rate for paediatric cases of COVID-19 reported in this study cannot be extrapolated to other hospitals, cities or regions in the country. Secondly, the authors could not establish significant associations between several of the variables mentioned above and mortality as well as MIS-C. Thirdly, the authors could not determine whether the cause of death was attributable to COVID-19 or underlying comorbidities. Finally, several laboratory panels, such as interleukins and other cytokines, were not checked to measure the severity of COVID-19. Nevertheless, despite these limitations, this study revealed a high mortality rate in paediatric patients with COVID-19. To the authors’ knowledge, this study is the first to describe the clinical characteristics of an Indonesian population.
Conclusion
This study described cases of mortality in paediatric patients with positive tests for COVID-19. A higher proportion of deaths was observed in patients aged >10 years with severe manifestations upon admission, and with PaO2/FiO2 ratios ≤300 mmHg. This is the first study in Indonesia to highlight the mortality-related or coincidental to SARS-CoV-2. However, further multicentre studies and better intervention and management studies are required to optimize public health measures, especially for paediatric patients with severe and critical COVID-19. Further studies are also needed to improve understanding of the role of SARS-CoV-2 in supporting the mechanisms leading to mortality in children with associated comorbidities.
Author contributions
Conceptualization and study design: NDP, RD, NK, TT, MRJ, HAP, AH.
Data curation and management: NDP, GSO, TT, MRJ, HAP.
Data analysis: NDP, TT, GSO, MRJ, HAP.
Funding acquisition: NDP, MMD, AH.
Clinical data collection: NDP, TT, MRJ, HAP, RM, HAP, RD, NK, MRK, DBS, AHP, MMD, AP, WI, YP, HAS, ELH, DM, NWP, RM, AR, IY, AS, SH.
Supervision: NDP, RD, NK, MRK, DBS, AHP, MMD, AP, AH.
Writing (original draft preparation): NDP, HAP, GSO, RD, NK, TT, MRJ.
Writing (review and editing): NDP, HAP, TT, MRJ, GSO, RD, NK, MRK, DBS, AHP, MMD, AP, WI, YP, HAS, ELH, NWP, DM, RM, AR, IY, SH, AS, AH.
Ethical approval
The Ethics Committee of the Faculty of Medicine, Universitas Indonesia approved this study (Ref. 596/UN2.F1/ETIK/PPM.00.02/2020).
Funding
This study was funded by a research grant from Cipto Mangunkusumo Hospital.
Conflict of interests
None declared.
Acknowledgements
The authors wish to thank the Director of Cipto Mangunkusumo Hospital, all patients and their families, paediatric residents of the University of Indonesia, and the Kiara Ultimate medical team for their help and support.
References
Ahmed M, Advani S, Moreira A, Zoretic S, Martinez J, Chorath K, et al. Multisystem inflammatory syndrome in children: a systematic review. EclincialMedicine 2020;26:100527.
American Academy of Pediatrics. Children and COVID-19: state-level data report. Washington, DC: AAP. 2020 Available at: https://services.aap.org/en/pages/2019-novel-coronavirus-covid-19-infections/children-and-covid-19-state-level-data-report/ [Accessed 20 October 2020].
Andersson M, Gelers J, Anderson D, Zucker J, Nobel Y, Freedberg D, et al. Body mass index and risk for intubation or death in SARS-CoV-2 infection. Ann Intern Med 2020;173:782–90.
Belhadjer Z, Mfot M, Bajolle F, Khraiche D, Legendre A, Abaika S, et al. Acute heart failure in multisystem inflammatory syndrome in children in the context of global SARS-CoV-2 pandemic. Circulation 2020;142:429–36.
Bhopal S, Bagaria J, Olabi B, Bhopal R. COVID-19 deaths in children: comparison with all- and other causes and trends in incidence of mortality. Public Health 2020;188:32–4.
Cadagnoli R, Votto M, Liciari A, Brambilla I, Bruno R, Perlini S, et al. Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection in children and adolescents. JAMA Pediatr 2020;174:382–90.
Centers for Disease Control and Prevention. Multisystem inflammatory syndrome in children (MIS-C). Atlanta, GA: CDC; 2020 Available at: https://www.cdc.gov/mis-c/hcp/ [Last accessed 20 October 2020].
Di Giorgio A, Nicastro E, Arnaboldi S, Montoni O, Di Stasio F, D’Antiga L, et al. Health status of children with chronic liver disease during the SARS-CoV-2 outbreak: results from a multicentre study. Clin Res Hepatol Gastroenterol 2021;45:101610.
Dong Y, Mo X, Hu Y, Qi X, Jiang F, Jiang Z, et al. Epidemiological characteristics of 2143 pediatric patients with 2019 coronavirus disease in China. Pediatrics 2020a;146:e20200702.
Dong Y, Mo X, Hu Y, Qi X, Jiang F, Jiang Z, et al. Epidemiology of COVID-19 among children in China. Pediatrics 2020b;145:e20200702.
Elihazli R, Toralh E, Elgamli A, El-Mowafy M, El-Mesery M. Amin M. Diagnostic and prognostic value of hematological and immunological markers in COVID-19 infection: a meta-analysis of 6320 patients. PLoS One 2020;15:e0238160.
Escosa-Garcia L, Aguilera-Alonso D, Calvo C, Mellado M, Baquero-Artigao F. Ten key points about COVID-19 in children: the shadows on the wall. Pediatr Pulmonol 2020;55:2576–86.
Gagliardi I, Patella G, Michael A, Serra R, Provenzano M, Andreucci M. COVID-19 and the kidney: from epidemiology to clinical practice. J Clin Med 2020;9:3506–35.
He X, Lau E, Wu P, Deng X, Wang J, Hao X, et al. Temporal dynamics in viral shedding and transmissibility of COVID-19. Nat Med 2020;26:672–5.
Henry B, de Oliveira M, Benoit S, Plebani M, Lippi G. Hematological, biochemical and immune biomarker abnormalities associated with severe illness and mortality in coronavirus disease 2019 (COVID-19): a meta-analysis. Clin Chem Lab Med 2020;58:1021–8.
Jiang L, Tang X, Levin M, Irfan O, Morris S, Wilson K, et al. COVID-19 and multisystem inflammatory syndrome in children and adolescents. Lancet Infect Dis 2020;20: e276–88.
Jung C, Park H, Kim D, Lim H, Chang J, Choi Y, et al. Association between body mass index and risk of coronavirus disease 2019 (COVID-19): a nationwide case-control study in South Korea. Clin Infect Dis 2020;ciaa1257.
Karahan A, Sarinoglu R, Bilgin H, Yanilmaz O, Sayin E, Deniz G, et al. Relationship of the cycle threshold values of SARS-CoV-2 polymerase chain reaction and total severity score of computerized tomography in patients with COVID 19. Int J Infect Dis 2020;101:160–6.
Kementerian Kesehatan Republik Indonesia. Pedoman Pencegahan Pengedalian Coronavirus Disease (COVID-19). Kementerian Kesehatan Republik Indonesia; 2020 Available at: https://covid19.go.id/storage/app/media/Protocol/2020/juli/REV-05_Pedoman_F2_COVID-19_13_juli_2020.pdf [Accessed 20 October 2020].
Leticia S, Sousa L, das Graças Carvalho M, Romana D, de Barros Pinheiro M. COVID-19: review and hematologic impact. Clin Chim Acta 2020;510:170–6.
Ludvigsson J. Systematic review of COVID-19 in children shows milder cases and a better prognosis than adults. Acta Paediatr 2020;109:1088–95.
Nicastro E, Di Giorgio A, Zambelli M, Ginanni M, Bravi M, Stroppa P, et al. Impact of the severe acute respiratory syndrome coronavirus 2 outbreak on pediatric liver transplant recipients in Lombardy, Northern Italy. LTIX 2020;26:1359–62.
Pulungan AB. Indonesia set to have world’s highest rate of child deaths from COVID-19. APPA Bulletin; 2020 Available at: http://a-p-p-a.org/pdf/1202-APPA-Bulletin-79-69-may-aug-2020.pdf [Accessed 20 October 2020].
Rao S, Manserro D, Steele V, Fareja JA. Narrative systematic review of the clinical utility of cycle threshold values in the context of COVID-19. Infect Dis Ther 2020;9:573–86.
Royal College of Paediatrics and Child Health. Paediatric multisystem inflammatory syndrome temporally associated with COVID-19 (PIMS) – guidance for clinicians. London: RCPCH; 2020 Available at: https://www.rcpch.ac.uk/resources/paediatric-multisystem-inflammatory-syndrome-temporally-associated-covid-19-pims-guidance?dm_i=1Z51z%2C6UTD%2C6ZU%2CR1H1%3C2C1 [Accessed 20 October 2020].
Rophagen S, Gomez X, Gonzalez-Martinez C, Wilkinson N, Theoharis P. Hyper-inflamatory shock in children during COVID-19 pandemic. Lancet 2020;395:1907–8.
Rodriguez-Morales A, Cardosa-Ospina J, Gutiérrez-Ocampo E, Villamizar-Peña R, Holguin-Rivera Y, Escalera-Antezana J, et al. Clinical, laboratory and imaging features of COVID-19: a systematic review and meta-analysis. Travel Med Infect Dis 2020;34:101623.
Shekerdemian L, Mahmood N, Wolfe K, Riggs B, Ross C, McKiernan C, et al. Characteristics and outcomes of children with coronavirus disease 2019 (COVID-19) infection admitted to US and Canadian pediatric intensive care units. JAMA Pediatr 2020;174:868–73.
Toubiana J, Poirault C, Corsia A, Bajolle F, Fourgeaud J, Angoulvant F, et al. Kawasaki-like multisystem inflammatory syndrome in children during the COVID-19 pandemic in Paris, France: prospective observational study. BMJ 2020;369:m2084.
Wendel P, Fumeaux T, Guerci P, Heuberger D, Montomoli J, Roche-Campo F, et al. Prognostic factors associated with mortality risk and disease progression in 639 critically ill patients with COVID-19 in Europe: initial report of the international RISC-19-ICU prospective observational cohort. EClinicalMedicine 2020;25:100449.
World Health Organization. Coronavirus disease (COVID-19) dashboard. Geneva: WHO; 2020 Available at: https://covid19.who.int [Last accessed 20 October 2020].
World Health Organization. Multisystem inflammatory syndrome in children and adolescents temporally related to COVID-19. Geneva: WHO; 2020 Available at: https://www.who.int/news-room/commentaries/detail/multisystem-inflammatory-syndrome-in-children-and-adolescents-with-covid-19 [Accessed 20 October 2020].
World Health Organization. Coronavirus disease 2019 (COVID-19) situation report – 1. Geneva: WHO; 2020 Available at: https://www.who.int/docs/default-source/serie/indonesia/covid19/who-indonesia-situation-report-1.pdf?sfvrsn=54eb3b39_0 [Accessed 20 October 2020].
WHO. COVID-19 case definition; 2020. COVID-19 case definition. [Accessed 20 October 2020]. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
775,
1
],
[
775,
4740,
2
],
[
4740,
8364,
3
],
[
8364,
12793,
4
],
[
12793,
19831,
5
],
[
19831,
27110,
6
],
[
27110,
31019,
7
],
[
31019,
37744,
8
],
[
37744,
44181,
9
]
]
} |
3fafea4f9c99ce3107f4a1801fbd4b125890a191 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 11,
"total-fallback-pages": 0,
"total-input-tokens": 36259,
"total-output-tokens": 10491,
"fieldofstudy": [
"Engineering"
]
} | Vector Diagram: How to Get More Useful Information from Hydraulic Unit Monitoring
Evgeniia Georgievskaja
Center of Design and Technological Innovation LLC, 197376 Saint-Petersburg, Russia; [email protected]; Tel.: +7-921-971-6443
Abstract: The development of an effective hydraulic unit (HU) diagnostic system, which recognizes critical fault at an early stage of their development, is a relevant and practically important objective for many hydroelectric power plants (HPP). The main trends in recent years in this area are the expansion of the hardware base and the introduction of digital technologies to handle the increasing flow of data monitoring. At the same time, much less attention is paid to the improvement of algorithms for potential defect recognition and the creation of new diagnostic rules; much of the monitoring data are used only for the trend construction, and do not participate in the diagnosis. The method of vector diagrams is proposed for estimating actual forces applied towards the support joints of an HU: each mode of operation is represented on a phase plane in the form of a vector proportional to the vibration displacement. The vector difference determines the actual forces of mechanical, electromagnetic and hydraulic origin in each operating mode, taking into account the individual characteristics of the HU. The example of vector diagram analysis shows how to obtain more useful information about the technical status of the unit from the same amount of basic vibration monitoring data, to predict possible deterioration in the operation of the equipment well before the generation of a warning or an alarm signal by the diagnostic system.
Keywords: hydraulic unit; shaft dynamics; relative vibration; vector diagram; vibration monitoring; diagnostics; lifetime; guide bearing
1. Introduction
Modern hydraulic unit (HU) diagnostic systems are equipped with multiple sensors (from 50 up to 300 pcs) which regularly analyze equipment operation data. As with any rotating unit, vibration monitoring plays a significant role in the diagnosis. Unlike other types of non-destructive control, vibration monitoring allows the rotating unit state monitoring directly during operation and not only during equipment shutdowns. The vibration state determines the reliability of the HU operation to a large extent. Increased vibration results in accelerated wear of the critical components and premature failure of the HU [1,2]. Vibration, as an integral response of the system to an external impact, together with a number of technological parameters, is an objective indicator of the presence of faults or discrepancies in the operation of the equipment. Therefore, HU vibration monitoring is now an integral part of the safe operation of most HPPs [3,4].
Vibration monitoring of the HU provides continuous or periodic measurement of the vibration state, its conversion, processing, recording, archiving, comparison of measured values with threshold values, tracking and qualitative analysis of trends in the changing of vibration parameters, splitting vibration signals into frequency components, generation of warning and alarm signals. Controlled vibration parameters are considered to be:
- the absolute vibrations of the stationary parts of the unit (thrust and guide bearing housings, the hydro turbine cover, the spider, the hydro generator stator’s steel structures);
- the relative vibrations of rotating elements (usually shaft vibrations near guide bearings).
The measured value can be vibration displacement, vibration velocity, or vibration acceleration. Changes in the status of the hydroelectric unit can be seen from changes in the vibration signals. Vibrations are quantified by comparing with thresholds defined by regulatory documentation [5–9], operation experience and manufacturers’ recommendations.
Some fault recognition algorithms based on vibration monitoring data have now been successfully implemented in many hydropower plants operation practice [10–15], allowing the detection of a series of dangerous faults:
• deviations in the operation of the turbine regulator;
• weakening of support structure fasteners;
• gap geometry distortion;
• shaft line defects;
• weakening of the flange joints;
• deterioration of kinematics;
• modification of the imbalances of mechanical, hydraulic and electromagnetic origin;
• excitation of eigenfrequencies, as well as approaching resonant zones, etc.
The data collected by the vibration monitoring system serve as initial information for the diagnostic system, whose main objective is making a diagnosis (HU technical condition assessment). The diagnostic system aims at providing maintenance according to the actual technical condition of the HU, taking into account changes in the energy sector, prevention of industrial accidents and downtimes, lifetime increase, minimization of negative factors affecting equipment operating modes, as well as optimization of operating costs. The diagnostic system objectives (in terms of vibration parameter processing) also include:
• interpretation of measurements from the vibration monitoring system;
• identification of defects (discrepancies, faults);
• identification of the correlation between controlled vibration characteristics and process variables taking daily, seasonal, annual changes into account;
• determining the causes of changes in vibrations;
The diagnostic system should also allow trend analysis (short-term and long-term) for each monitored vibration parameter. In case of significant changes, even if the parameter remains in the acceptable value field, the reasons for such changes should be investigated, because unexplained changes may be associated with the beginning critical HU components’ destruction. However, at present, in existing diagnostic systems this function is not generally automated, and finding the causes of changes implies contacting an expert.
Usually, only a small part of the monitoring system data collected is used to diagnose the technical condition of the hydraulic unit, mainly information on the component’s temperature status, change in the gaps between the rotating and stationary parts of the HU, and integral characteristics of absolute and relative vibrations in support structures [5–9].
Trends in recent years in the development of monitoring and diagnostic systems for hydraulic units are the expansion of the hardware base and the introduction of digital technologies (for example, Big Data, Machine Learning, Artificial Neural Networks algorithms) for processing the increasing flow of monitoring data [16–19], when much less attention is paid to creating new diagnostic rules (new defect recognition algorithms). The systems currently operating at HPPs are essentially monitoring systems with separate diagnostic functions. They have been successful in protecting the unit against critical faults, in generating warning and alarm signals, and in detecting some faults at the qualitative level. However, it is too early to fully diagnose the unit based on the data collected by the monitoring system, for the following main reasons:
• the diagnosis is mainly qualitative: presence or absence of imbalance, increased vibration in a specific zone, unfavourable cavitation conditions etc.;
• no quantitative force impact effects on equipment, which may in the future make it possible to assess when a defect appears, its danger and development pace, and to predict possible faults in the operation of the hydraulic unit or the time it takes to be in need of repairs long before a warning or alarm signal is generated;
insufficient efficiency in finding faults and defects: HU emergency stops occur [20] or
defects are found during equipment repairs (not predicted by the diagnostic system);
this is due to the difficulty in extracting important informative components from the
total vibration signal;
• inability to identify a number of defects at an early stage of development: often the
vibration monitoring system detects a defect only when there are significant changes
in the vibration state of the unit, and the size of the defect is already almost critical
(immediate stop and long and expensive repairs required);
• lack of understanding of the impact of operation on components of equipment (insuf-
icient analysis of the influence of the operation conditions on the values of the control
parameters and their change trends);
• insufficient consideration of individual characteristics of the hydraulic unit (vibration
thresholds are often assumed to be the same for different units or even for different
types of units);
• lack of information analysis on transient operating modes that play a significant role
in the performance and lifetime of an HU.
Relying solely on vibration characteristics, it may be noted that the volume of data
from the vibration monitoring system is much larger than what is required to assess the
vibration state of the unit in accordance with regulations and currently established criteria.
The rest of the pieces of the vibration state monitoring data remain without application
in the archives, being often unstructured and unavailable for further analysis. While this
article does not purport to cover all factors and all features of monitoring and diagnostic
systems, it reflects on some aspects. It offers additional possibilities in the transition from
monitoring to full HU diagnostic, referring to examples of analysis of the dynamics of the
HU shaft by means of vector diagrams.
2. Object of Research
The object of research is a vertical three-point hydraulic unit, installed at an HPP
in Russia. The HPP’s head is about 30 m. The unit is equipped with a Kaplan turbine
with a nominal capacity of about 80 MW. Kaplan turbine power control is achieved by
rotating guide vanes as well as rotating hydraulic runner (HR) blades about its axis. This
allows the efficient use of the watercourse, maintaining high efficiency in almost the whole
operating range. The HU shaft speed is about 83 rpm. A hydraulic runner has seven blades,
approximately 7 m in diameter. The maximum HR blade rotation angle is 35°. The unit’s
shaft line consists of a turbine shaft and a generator shaft, firmly connected by a flange joint.
The diameter of the turbine shaft along the bearing is 1.23 m. The guide vanes’ quantity
is 32. This unit is selected as an example, all conclusions of this article may be extended
to other powerful vertical HUs with HR diameters above 5 m, for which the influence of
individual characteristics is most pronounced.
Figure 1 shows the HU structural design. The structural design reads: $C_{TB}$—thrust
bearing (TB) rigidity, $C_{UGB}$, $C_{LGB}$, $C_{TGB}$ are the rigidities of the upper generator guide
bearing (UGB), the lower generator guide bearing (LGB) and the turbine (guide) bearing
(TGB), respectively. Each bearing’s rigidity consists of bearing housing rigidity and bearing
liner rigidity, the latter being of crucial importance. For further calculations, the following
bearing liner rigidity values were adopted, previously calculated experimentally for the
reviewed unit: UGB—$0.3 \cdot 10^6$ N/m, LGB—$0.5 \cdot 10^6$ N/m, TGB—$0.3 \cdot 10^6$ N/m.
3. Vector Diagram Method
The aim is to obtain information on actual operating loads towards the unit shaft and support structures, taking into account existing geometry parameter imperfections (shape distortion of rotor, stator and generator air gap, hydraulic runner, runner chamber and chamber-to-blade gap). For this purpose, the analysis of the amplitude spectrum of the relative vibrations of the generator and turbine shafts is carried out near the guide bearings (UGB, LGB and TGB) specifying the amplitude and the 1st harmonic’s phase. The relative shaft vibration sensors are installed near each bearing in two mutually perpendicular directions: upstream–downstream (US–DS) and left bank–right bank (LB–RB). In accordance with Russian regulations [8,9], the normalized relative vibration parameter is the range of vibration displacement (doubled amplitude) averaged over several rotations.
The 1st harmonic usually contributes the most to the total vibration state. Therefore, much determines the force impact on the unit. Most vibration monitoring systems, including old ones, allow the identification of a 1st harmonic with oscillation phase separation. Oscillation phase is important for further analysis because it allows moving on from scalar values (total level of relative vibration) to vector values (amplitudes/phase couple on the phase plane).
The analysis of the spectrum of the relative vibration 1st harmonic is performed on the characteristic operating modes of the unit:
1. run-out from the speed-no-load mode without excitation (free HU rotation before stopping);
2. speed-no-load with no excitation (SNLNE) at nominal rotating speed $N_{nom}$;
3. speed-no-load with excitation (SNLWE) at nominal shaft rotation speed;
4. low power parallel to mains operation mode—10–15% of the nominal value $N_{nom}$ (here taken as 10% of $N_{nom}$);
5. medium power parallel to mains operation mode—40–60% of the nominal value $N_{nom}$ (here taken as 50% of $N_{nom}$);
6. high power generator’s parallel to mains operation mode—75–90% of the nominal value $N_{nom}$ (here taken as 85% of $N_{nom}$);
7. full power generator’s parallel to mains operation mode—100% of the nominal value $N_{nom}$.
In run-out mode, after distributor closure, the unit is freely rotating at insignificant speed (less than 20 rpm) and not subject to any external effects of hydraulic, mechanical or electromagnetic origin. During run-out, no external forces are applied to the unit, the vibration state is determined only by the geometry of the shaft and the individual characteristics of the HU. This mode is taken as a ‘zero position’ and serves as a reference point for the analysis of all other modes. The force impact on the components of the unit in the remaining modes reveals itself by changing the value and phase of the 1st harmonic relative to the corresponding values in the run-out mode. From the increments of the vibration displacement vectors, the corresponding force impacts on the shaft of the unit can be determined during the transition from one mode to another, and subsequently the
Figure 1. Vertical three-point HU structural design (image turned for ease of presentation).
load on the HU guide bearings can be determined, their performance and the repairs time needed can also be assessed.
In speed-no-load with no excitation (SNL_{NE}) mode, only centrifugal forces, determined by available mechanical imbalance and speed, are applied towards the HU. Hydraulic forces are not significant and may not be taken into account at this stage, and no electromagnetic force effects are present. The mechanical imbalance is created by the unevenness of the mass distribution in the circumference and manifests itself in the massive parts of the HU (generator rotor, hydraulic runner) in the form of the periodic result force of the rotational frequency. The resulting mechanical force on the hydraulic runner and on the generator rotor, together with the values of bearing rigidities, determines the vibration displacement of the shaft near guide bearings. This displacement is recorded by the vibration monitoring system for two mutually perpendicular directions US–DS and LB–RB.
During the unit’s speed-no-load with excitation (SNL_{WE}) in addition to mechanical forces, electromagnetic fields begin their influence, determined by the electromagnetic imbalance, the eccentricity of the rotor position relative to the generator stator, rotor shape distortion, as well as deformation of the stator and the generator air gap. Consequently, there is an additional resulting force on the generator rotor that has an electromagnetic origin. This force also manifests itself in a periodic impact on HU elements. The cumulative effect of mechanical and electromagnetic imbalance is reflected in the recorded vibration displacement of the shaft near the guide bearings in SNL_{WE} mode.
In parallel to mains operation modes at different power, the hydraulic load from the flow-through path is supplemented. The additional resulting hydraulic force arises as a result of hydraulic imbalance of the hydraulic runner: technological deviations in the grid of the bladed system, imperfections in the fixed part of the flow-through path, misalignment between the axis of the hydraulic runner and the axis of the stator of the hydraulic turbine, distortions of the chamber-to-blade gap geometry, and so on. The hydraulic force, as well as the mechanical and electromagnetic force, creates periodic effects on the supporting structures of the HU. Vibration sensors fix this contribution to the total vibration response in the given modes.
The method of the vector diagrams consists of constructing a broken-line graph on a phase plane, reflecting the dynamic displacement of the shaft in different operating modes according to the data received from the vibration monitoring system. In this case, the vibration displacements (ranges—doubled amplitude) near the three guide bearings (UGB, LGB и TGB) are considered in two mutually perpendicular directions, US–DS and LB–RB. Each point in the diagram is the end of the vector corresponding to the mode with that number. The start of the vector is at the origin. The length of the vector corresponds to the range of the 1st harmonic of the vibration displacement, the rotation of the vector is determined by the phase of oscillation of the 1st harmonic. The vector represents the position of a point on the surface of the shaft close to a specified guide bearing for a given mode in a specified direction. The change in mode leads to a change in the point position and a vector change. The difference vector between the final and initial position of the vectors on the diagram during the change in mode (vector increment) is proportional to the force impact generated by a change in modes. The proportionality coefficient is the value of the liner rigidity of the bearing, near which the vibration displacements are measured. The values of rigidities are presented in Section 2. The modulus and phase of the vector increment are calculated by vector algebra methods.
The points are sequentially placed on the diagram, which are corresponding to all HU operating modes considered. The lines connect the points and show the variation in the vector when changing from one mode to another. The result is a visual representation of the dynamic displacement of the shaft during operation. Phase plane boundaries (maximum value among the axles) correspond to the threshold values recommended by the manufacturer for the unit: 500 μm for UGB, 600 μm for LGB, 400 μm for TGB.
As opposed to the normative approach [5–9], which regulates the assessment of only the total level of relative vibrations and some frequency constituents in order to determine
the permissibility or impermissibility of prolonged operation of the HU in a given mode, the vector diagram method provides much more useful information for technical diagnosis.
4. Results
Table 1 gives an example of HU vibration monitoring system data recorded for a specific time: amplitudes and phases of vibration displacement of the 1st harmonic for each of the two mutually perpendicular directions, US–DS and LB–RB, for each of the three guide bearings, UGB, LGB and TGB.
Table 1. Initial vibration monitoring data (1st harmonic).
| Mode Number | Directions | UGB Amplitude, µm | Phase, ° | LGB Amplitude, µm | Phase, ° | TGB Amplitude, µm | Phase, ° |
|-------------|------------|-------------------|---------|-------------------|---------|-------------------|---------|
| (1) | US-DS | 78 | 249 | 72 | 169 | 173 | 113 |
| | LB-RB | 132 | 138 | 70 | 115 | 180 | 20 |
| | US-DS | 454 | 227 | 321 | 175 | 275 | 147 |
| (2) | LB-RB | 502 | 103 | 344 | 84 | 266 | 62 |
| | US-DS | 489 | 257 | 291 | 231 | 195 | 151 |
| (3) | LB-RB | 544 | 128 | 389 | 123 | 202 | 72 |
| | US-DS | 472 | 256 | 335 | 217 | 214 | 147 |
| (4) | LB-RB | 527 | 127 | 415 | 116 | 230 | 69 |
| | US-DS | 472 | 259 | 366 | 219 | 206 | 149 |
| (5) | LB-RB | 521 | 131 | 423 | 120 | 227 | 74 |
| | US-DS | 347 | 294 | 275 | 282 | 146 | 132 |
| (6) | LB-RB | 478 | 159 | 399 | 166 | 140 | 84 |
| | US-DS | 241 | 286 | 105 | 294 | 185 | 119 |
| (7) | LB-RB | 435 | 151 | 310 | 162 | 141 | 68 |
Figure 2 shows vector diagrams of dynamic shaft displacements near guide bearings UGB, LGB and TGB, charted from Table 1 data for each of the two directions US-DS and LB-RB (source data of the vibration monitoring). The point number in the diagram corresponds to the mode number.
Figure 2. Vector diagrams of the dynamic shaft displacements near guide bearings (µm): (a) UGB—upper generator guide bearing; (b) LGB—lower generator guide bearing; (c) TGB—turbine (guide) bearing (the numbers (1) ÷ (7) correspond to mode numbers in Table 1; description of the modes is given in Section 3).
Table 2, Table 3 and Figure 3, show the results of the vibration experiment data processing, presented in Table 1: the vector increment of the 1st harmonic in a given mode relative to run-out, and the increment from external forces of mechanical, hydraulic and
electromagnetic origin (force increments). The point numbers in the diagrams are represented by the difference between the initial mode and the run-out mode. Force increments are the vector differences between the respective modes: mechanical—between the SNL_{NE} mode and run-out, electromagnetic—between the SNL_{WE} and SNL_{NE}, hydraulic—between operating conditions at the appropriate power mode and SNL_{WE}. Separation of force factors by nature of origin makes it easier to determine the cause of increased vibrations than total vibration displacement recording.
Table 2. Vector increments of the 1st harmonic on different operating modes.
| Mode Number | Directions | UGB | LGB | TGB |
|-------------|------------|-----|-----|-----|
| | | Amplitude, µm | Phase, ° | Amplitude, µm | Phase, ° | Amplitude, µm | Phase, ° |
| (2-1) | US-DS | 383 | 313 | 250 | 267 | 163 | 273 |
| | LB-RB | 401 | 182 | 286 | 167 | 179 | 194 |
| (3-1) | US-DS | 412 | 349 | 265 | 335 | 122 | 302 |
| | LB-RB | 415 | 215 | 320 | 215 | 169 | 219 |
| (4-1) | US-DS | 395 | 347 | 292 | 318 | 120 | 291 |
| | LB-RB | 398 | 213 | 345 | 206 | 176 | 210 |
| (5-1) | US-DS | 395 | 351 | 324 | 319 | 121 | 296 |
| | LB-RB | 390 | 219 | 353 | 211 | 189 | 214 |
| (6-1) | US-DS | 471 | 31 | 344 | 15 | 271 | 260 |
| | LB-RB | 358 | 257 | 359 | 265 | 173 | 243 |
| (7-1) | US-DS | 185 | 31 | 158 | 46 | 22 | 263 |
| | LB-RB | 308 | 247 | 267 | 263 | 135 | 239 |
Table 3. Force increments of the 1st harmonic.
| Force | Directions | UGB | LGB | TGB |
|-------------|------------|-----|-----|-----|
| | | Amplitude, µm | Phase, ° | Amplitude, µm | Phase, ° | Amplitude, µm | Phase, ° |
| Mechanical | US-DS | 383 | 313 | 250 | 267 | 163 | 273 |
| Fm | LB-RB | 401 | 182 | 286 | 167 | 179 | 194 |
| Electromagnetic | US-DS | 246 | 54 | 289 | 28 | 82 | 47 |
| Fe | LB-RB | 230 | 285 | 248 | 274 | 76 | 304 |
| Hydraulic | US-DS | 19 | 193 | 88 | 254 | 24 | 202 |
| Fh (10% N_{nom}) | LB-RB | 19 | 66 | 56 | 147 | 30 | 139 |
| Hydraulic | US-DS | 24 | 123 | 101 | 272 | 13 | 208 |
| Fh (50% N_{nom}) | LB-RB | 36 | 349 | 40 | 179 | 26 | 180 |
| Hydraulic | US-DS | 321 | 90 | 220 | 65 | 198 | 236 |
| Fh (85% N_{nom}) | LB-RB | 280 | 337 | 289 | 323 | 71 | 318 |
| Hydraulic | US-DS | 302 | 144 | 261 | 120 | 105 | 130 |
| Fh (100% N_{nom}) | LB-RB | 222 | 348 | 245 | 340 | 62 | 351 |
Table 4 and Figure 4 show the forces of mechanical, electromagnetic and hydraulic origin calculation results; these forces being applied to both directions on the supporting structures of the unit near the guide bearings in different operating modes. The load values are the product of the calculated force increments and the corresponding rigidities of the bearing liners.
a regular phase jump –
**Figure 3.** Vector diagrams of the dynamic shaft displacement 1st harmonic’s increment near guide bearings (μm): (a) UGB—upper generator guide bearing; (b) LGB—lower generator guide bearing; (c) TGB—turbine (guide) bearing (the numbers (1) ÷ (7) correspond to mode numbers in Table 1; description of the modes is given in Section 3).
**Table 3.** Force increments of the 1st harmonic.
| Force (t) | Directions | UGB | LGB | TGB |
|-----------|------------|------|------|------|
| Mechanical | US-DS | 5.7 | 7.8 | 3.7 |
| Fm | LB-RB | 6.0 | 4.6 | 4.3 |
| Electromagnetic | US-DS | 3.7 | 1.4 | 4.3 |
| Fe | LB-RB | 3.5 | 7.1 | 3.7 |
| Hydraulic | US-DS | 0.3 | 4.8 | 1.3 |
| Fh (10% $N_{nom}$) | LB-RB | 0.3 | 1.7 | 0.8 |
| Hydraulic | US-DS | 0.4 | 3.1 | 1.5 |
| Fh (50% $N_{nom}$) | LB-RB | 0.5 | 8.7 | 0.6 |
| Hydraulic | US-DS | 4.8 | 2.2 | 3.3 |
| Fh (85% $N_{nom}$) | LB-RB | 4.2 | 8.4 | 4.3 |
| Hydraulic | US-DS | 4.5 | 3.6 | 3.9 |
| Fh (100% $N_{nom}$) | LB-RB | 3.3 | 8.7 | 3.7 |
**Figure 4.** Force impacts on the shaft of the unit (t): (a) upstream-downstream direction (US-DS); (b) left bank-right bank direction (LB-RB); Fm—mechanical force, Fe—electromagnetic force, Fh—hydraulic force with indication of relative power.
5. Discussion
Vector diagrams illustrate the dynamic behaviour of the shaft when changing operating modes. Switching from run-out mode (1) to SNLNE mode (2) shall occur strictly by the radius, because the level of relative vibration of the shaft is driven by mechanical imbalance. In Figure 2, this condition is well met only for LGB in the US–DS direction. Deviation of the vector chart trajectory from the radius (phase shift) indicates system instability, for example, tension weakening in the contact area or rotating parts fastening or loosening, or other imbalance mass offset.
Inclusion into mains, switching mode (2) to mode (3), causes a regular phase jump and a corresponding change in vector direction. The variation in the vector indicates the magnitude of the emerging electromagnetic forces. In Figure 2, the vector module for LGB and UGB changes insignificantly, as the phase changes approximately the same for both directions: US–DS and LB–RB. The TGB vector module changes significantly for both directions, as the phase changes insignificantly, the electromagnetic forces partially balance the available mechanical imbalance.
Switching from run-out mode (3) to operation on (4) ÷ (7) power modes with a gradual increase in the flow of water through the unit coincides with the appearance and growth of hydraulic effects. Modes (4) and (5) of low and medium power (10% and 50% of Nnom) are associated with flux instability in the flow part of the unit, expressed as small shifts in amplitude, as well as in phase without pronounced movement trend. Further power increase (transition to high-power mode (6)—85% of Nnom) leads to stabilization of processes and an increase in hydraulic force on the hydraulic runner. The vector rotates, indicating a change in the distribution of hydraulic forces in the flow section. In Figure 2, for all values except LGB in the LB–RB direction, the vector module is reduced by the partial balancing of mechanical and electromagnetic forces.
In the last phase of the transition from high-power mode (6) to full power mode (7), a vector jump occurs again (phase changes on Figure 2), which seems to be linked to changes in flow conditions at maximum power modes. Furthermore, for all bearings, the end of the vector is shifted to the center, therefore, at maximum power, the hydraulic load balances the mechanical and electromagnetic imbalances to a maximum extent. It should be noted that this jump is not observed for all HUs and is more an individual characteristic of the units of this HPP than a characteristic of all units of any type.
The presented method of vector diagrams makes it possible to analyze the ratio of forces acting on a hydraulic unit in addition to the nature of shaft displacements. The magnitude of the force can be used to indicate internal stresses in the hydroelectric units and, using the known mathematical tools, to proceed to assess stress–strain state and remaining lifetime, that is, to plan properly the time needed for necessary repairs. This possibility can be automated and implemented in diagnostic systems in the form of a periodic report on the actual load of the unit’s support joints.
The results presented in Table 3, Table 4 and Figure 4 show that, for the unit in question, the forces of mechanic (Fm) and electromagnetic (Fe) origin are quite different for the directions US–DS and LB–RB. Especially visible is the difference for LGB. The main reason for this is probably the large deviations in the generator rotor and generator stator geometry: the presence of eccentricity and ovality, as well as the asymmetry of the generator air gap. Different sensitivities of LGB and UGB to uneven loads along the directions can be explained by differences in the rigidity of their liners.
Figures 2 and 3 clearly show that for TGB in both directions, the mechanical imbalance is largely balanced by the hydraulic imbalance: final vector amplitude in mode (7) is close to the initial vector amplitude at point (1), although the vector itself is rotated. In practice, this means that, near the nominal operating mode, the low level of relative shaft vibrations is not an indication of a good technical state of the unit or of the absence of significant force impacts on its joints, but of the presence of a couple of forces counterbalancing each other. As a result, the impact of these forces will lead to premature defects, which cannot be predicted on the basis of a normative analysis of relative shaft vibrations alone. The
influence of this pair of forces can also be seen on the generator bearings in the direction US–DS, where the spin up line in speed-no-load mode (from mode (1) to mode (2)) and power build-up mode at greater energy consumption mode (from mode (5) to mode (6)) are almost parallel, as the transition to mode (7) moves the end of the vector even closer to the center, but in a different direction.
Figure 4 presents the histograms of actual force values of different origin without their direction. It can be seen that LGB is under increased loads in the direction of LB–RB in almost all operating modes, including at high-power modes. At the same time, the total level of relative vibration on the LGB in all modes is not more than 400 µm, which is much lower than the manufacturer’s recommended limits (600 µm).
A similar analysis by vibration monitoring can be carried out to assess the unit’s history of operation, the degree of loading of its components according to operating modes, and to compare the pre- and post-repair state of the HU. Based on the analysis described, new diagnostic rules can be developed that take into account not only the total level of relative vibrations but also the actual loads level.
The automation of the proposed algorithm for calculating actual loads on HU key components opens up wide possibilities on express-assessment of the actual technical state of the unit without stopping it, based on vector diagrams identifying problem areas and forecasts of the dates and size of repairs required. If information on phases for higher harmonics is available in the monitoring system, using the technology mentioned above, the influence of factors can be estimated, which manifest themselves on the second, the third, etc., rotating frequencies. The diagnosis can also be clarified based on the analysis of the 1st harmonic.
6. Conclusions
As opposed to the normative approach, which regulates the assessment of only the total level of relative vibrations and some frequency constituents in order to determine the permissibility or impermissibility of a given operating mode, the vector diagram method provides much more useful information for technical diagnosis while installation of additional sensors or accessories is not required. The monitoring system has already collected all the information necessary for analysis.
An example of data on the relative vibrations of the unit shaft shows the possibilities of qualitative and quantitative analysis by the vector diagram method showing the dynamic displacement of the shaft at transitions from one operating mode to another. The vector algebra methods help calculate the increments of vectors and the corresponding force increments proportional to the loads involved when operating modes change. Furthermore, the components formed by each type of load are distinguished from the total vibration signal, i.e., the total force impact is resolved into the mechanical, electromagnetic and hydraulic load according to the direction of action thereof. The example of a powerful three-point HU shows that vibration-safe modes do not always indicate that the unit is not subject to significant loads. It is possible that there are couples of forces whose actions partially counterbalance each other, but at the same time adversely affect the overall load of the HU components and result in an accelerated lifetime exhaustion.
The proposed approach makes it possible to estimate the stability/instability of the system on the basis of changes in the vector, identify trends of deterioration or balancing problems before the monitored parameters exceed the thresholds. The calculation of force impacts using the vector diagram method allows further evaluation of the stress-deformation state of the HU elements and remaining lifetime forecast by actual data and actual operating modes, as well as allowing the prediction of possible deterioration in equipment operation long before a warning or alarm signal is generated by the diagnostic system.
In the future, it is possible to formulate new diagnostic rules based on vector diagrams, to be introduced into diagnostic systems to improve their performance without expanding the hardware base.
**Funding:** This research was funded by the Foundation for Assistance to Small Innovative Enterprises in Science and Technology (FASIE), Russia, Grant No. 4373Г/2/55639, 1 December 2021, and the Skolkovo Innovation center, Grant No. 40113/07002/10329-2020, 22 April 2020.
**Institutional Review Board Statement:** Not applicable.
**Informed Consent Statement:** Not applicable.
**Data Availability Statement:** Not applicable.
**Conflicts of Interest:** The authors declare no conflict of interest.
**References**
1. Mohanta, R.K.; Chelliah, T.R.; Allamsetty, S.; Akula, A.; Ghosh, R. Sources of vibration and their treatment in hydro power stations—A review. *Eng. Sci. Technol.* **2017**, *20*, 637–648. [CrossRef]
2. Nässelqvist, M.; Gustavsson, R.; Aidanpää, J.-O. A methodology for protective vibration monitoring of hydropower units based on the mechanical properties. *J. Dyn. Syst. Meas. Control.* **2013**, *135*, 410071–410078. [CrossRef] [PubMed]
3. Valero, C.; Egusquiza, E.; Pressas, A.; Valentín, D.; Egusquiza, M.; Bossio, M. Condition monitoring of a prototype turbine. Description of the system and main results. *J. Phys. Conf. Ser.* **2017**, *813*, 012041. [CrossRef]
4. Alagöz, İ.; Bulut, M.; Geylani, V.; Yıldırım, A. Importance of real-time hydro power plant condition monitoring systems and contribution to electricity production. *Turk. J. Electr. Power Energy Syst.* **2021**, *1*, 1–11. [CrossRef]
5. ISO 13373-7:2017; Condition monitoring and diagnostics of machines—Vibration condition monitoring—Part 7: Diagnostic techniques for machine sets in hydraulic power generating and pump-storage plants. ISO copyright office: Geneva, Switzerland, 2017.
6. ISO 19283:2020; Condition monitoring and diagnostics of machines—Hydroelectric generating units. ISO copyright office: Geneva, Switzerland, 2020.
7. ISO 20816-5:2018; Mechanical vibration—Measurement and evaluation of machine vibration—Part 5: Machine sets in hydraulic power generating and pump-storage plants. ISO copyright office: Geneva, Switzerland, 2018.
8. GOST R 55260.3.3-2013; Hydro power plants. Part 3-3. Hydraulic turbines. Technical requirements for operation monitoring systems. UNIDO: Viena, Austria, 2019.
9. GOST R 55260.3.2-2013; Hydro Power Plants. Part 3-2. Estimation procedures of hydroturbine operating conditions. Standartinform: Moscow, Russia, 2013.
10. Ivanchenko, I.; Prokopenko, A. Analiz sistem monitoringa i diagnostiki tekhnicheskogo sostoyaniya gidroturbin. *Gidrotechnika* **2011**, *2*, 22. (In Russian)
11. DIA TECH Monitoring and Diagnosis System. Andritz Hydro. Available online: https://pdfs.semanticscholar.org/7259/027a40481ad0dcfd5a5be433cd84c8b80.pdf?_ga=2.253670409.1010097152.1584874762-1584874762.1584874762 (accessed on 5 February 2020).
12. Betti, A.; Crisostomi, E.; Paolinelli, G.; Piazzì, A.; Ruffini, F.; Tucci, M. Condition monitoring and early diagnostics methodologies for hydropower plants. *arXiv* 2019. arXiv:1911.06242.
13. Voith. Monitoring, Analysis and Diagnosis. OnCare. Health Hydro. Available online: https://voith.com/corp-en/VH_OnCare-HealthHydro_19_vvk_VH3401_en.pdf (accessed on 15 December 2021).
14. Emerson Hydropower Solutions. Available online: https://www.emerson.com/documents/automation/hydropower-dcs-scada-mechanical-solutions-en-67608.pdf (accessed on 20 March 2020).
15. Bently Nevada. Condition Monitoring and Protection Solutions Brief. Available online: https://dam.bakerhughes.com/m/m/4ae70971bd1624b5/original/GEA31903D-BN_Product_Brochure_R3-1-pdf.pdf (accessed on 25 December 2021).
16. Zhao, W.; Egusquiza, M.; Valero, C.; Valentín, D.; Pressas, A.; Egusquiza, E. On the use of artificial neural networks for condition monitoring of pump-turbines with extended operation. *Measurement* **2020**, *163*, 107952. [CrossRef]
17. Zhao, R.; Yan, R.; Chen, Z.; Mao, K.; Wang, P.; Gao, R.X. Deep learning and its applications to machine health monitoring. *Mech. Syst. Signal Process.* **2019**, *115*, 213–237. [CrossRef]
18. Selak, L.; Butala, P.; Sluga, A. Condition monitoring and fault diagnostics for hydropower plants. *Comput. Ind.* **2014**, *65*, 924–936. [CrossRef]
19. Betti, A.; Crisostomi, E.; Paolinelli, G.; Piazzì, A.; Ruffini, F.; Tucci, M. Condition monitoring and predictive maintenance methodologies for hydropower plants equipment. *Renew. Energy* **2021**, *171*, 246–253. [CrossRef]
20. Georgievskaya, E. Destruction of the hydraulic unit shaft: Why it is possible? *Forces Mech.* **2021**, *4*, 100026. [CrossRef] | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3509,
1
],
[
3509,
7634,
2
],
[
7634,
11242,
3
],
[
11242,
14464,
4
],
[
14464,
19065,
5
],
[
19065,
22366,
6
],
[
22366,
25420,
7
],
[
25420,
26771,
8
],
[
26771,
31293,
9
],
[
31293,
35521,
10
],
[
35521,
40029,
11
]
]
} |
572a38faca6c83f7419001308ec75ac05ba67568 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 15,
"total-fallback-pages": 0,
"total-input-tokens": 45974,
"total-output-tokens": 13416,
"fieldofstudy": [
"Biology"
]
} | RESEARCH ARTICLE
Methanesulfonate (MSA) Catabolic Genes from Marine and Estuarine Bacteria
Ana C. Henriques, Paolo De Marco*
Instituto de Investigação e Formação Avançada em Ciências e Tecnologias da Saúde, CESPU, Rua Central de Gandra 1317, 4585–116 Paredes, Portugal
* [email protected]
Abstract
Quantitatively, methanesulfonate (MSA) is a very relevant compound in the global biogeochemical sulfur cycle. Its utilization by bacteria as a source of carbon and energy has been described and a specific enzyme, methanesulfonate monooxygenase (MSAMO), has been found to perform the first catabolic step of its oxidation. Other proteins seemingly involved in the import of MSA into bacterial cells have been reported. In this study, we obtained novel sequences of genes msmA and msmE from marine, estuary and soil MSA-degraders (encoding the large subunit of the MSAMO enzyme and the periplasmic component of the import system, respectively). We also obtained whole-genome sequences of two novel marine Filomicrobium strains, Y and W, and annotated two full msm operons in these genomes. Furthermore, msmA and msmE sequences were amplified from North Atlantic seawater and analyzed. Good conservation of the MsmA deduced protein sequence was observed in both cultured strains and metagenomic clones. A long spacer sequence in the Rieske-type [2Fe-2S] cluster-binding motif within MsmA was found to be conserved in all instances, supporting the hypothesis that this feature is specific to the large (α) subunit of the MSAMO enzyme. The msmE gene was more difficult to amplify, from both cultivated isolates and marine metagenomic DNA. However, 3 novel msmE sequences were obtained from isolated strains and one directly from seawater. With both genes, our results combined with previous metagenomic analyses seem to imply that moderate to high-GC strains are somehow favored during enrichment and isolation of MSA-utilizing bacteria, while the majority of msm genes obtained by cultivation-independent methods have low levels of GC%, which is a clear example of the misrepresentation of natural populations that culturing, more often than not, entails. Nevertheless, the data obtained in this work show that MSA-degrading bacteria are abundant in surface seawater, which suggests ecological relevance for this metabolic group of bacteria.
Introduction
The ocean constitutes a large reservoir of sulfur and hence the transfer of volatile sulfur compounds from the sea to the atmosphere represents a key process in the sulfur cycle [1].
Dimethylsulfide (DMS) is the main component of marine emissions of volatile sulfur, contributing an estimated 98% of atmospheric DMS [2]. DMS is produced in the marine environment by degradation of its precursor dimethylsulfoniopropionate (DMSP) [3], which in turn is synthesized by microalgae and cyanobacteria starting from seawater sulfate. Only a small portion of the DMS produced in the oceans escapes to the atmosphere. Here it can be oxidized by the action of radical species such as OH, BrO, Cl and/or NO₃, through several intermediate reactions, forming a variety of products such as sulfur dioxide (SO₂), methanesulfonic acid (MSA) and dimethylsulfoxide (DMSO) [1,4,5].
MSA (formula CH₃SO₃H, the smallest organic sulfonic acid) is one of the main products of DMS oxidation since it is estimated that 25–70% of the flux of dimethylsulfide is oxidized to methanesulfonate (approximately 10¹⁰ Kg/year) [2,6–10]. Due to its hygroscopic nature, MSA takes part in the formation of cloud condensation nuclei (CCN), contributes to the regulation of cloud formation and thus has a significant impact on albedo regulation [4,11,12]. MSA falls onto lands and oceans in wet and dry precipitation [2,10,13] and, despite its high chemical stability, has only been found to accumulate in detectable levels in the frozen layers of snow of Antarctica and Greenland [14–16]. MSA can be used as a sulfur source by some aerobic bacteria [10]. On the other hand, several methylotrophic bacteria isolated from different environments have shown the ability to grow using MSA as the sole source of carbon and energy. The soil bacterium Methylosulfonomonas (Me.) methyllovora strain M2 was the first such isolate to be described [17–19], followed by the marine Marinosulfonomonas (Ma.) methylotropha strains TR3 and PSCH4 [20,21]. All of these isolates contained an inducible multicomponent cytoplasmic enzyme, MSA monooxygenase (MSAMO). This enzyme is responsible for splitting the C-S bond, catalyzing the first oxidative step of MSA to the central methylotrophic intermediate formaldehyde with the release of sulfite, which is subsequently oxidized to sulfate. Formaldehyde was assimilated through the serine cycle or fully oxidized to CO₂ and H₂O, in order to yield reducing power and energy. MSAMO was purified from Me. methyllovora str. M2 and its four components, all necessary for enzyme activity, were identified: the large (or α) subunit of the hydroxylase (MsmA—48 kDa), the small (or β) subunit of the hydroxylase (MsmB—20 kDa), a 16 kDa ferredoxin (MsmC), and a 38 kDa reductase component (MsmD) [10,18,22,23]. Subsequently, the genes encoding MSAMO (msmA, msmB, msmC and msmD) were cloned and sequenced from Me. methyllovora str. M2 and Ma. methyllovora str. TR3. In the case of Me. methyllovora str. M2, an msmABCD operon was found [18]. The msm genes from Ma. methylotropha str. TR3 were located in two separate but complementary operons, msmABC and msmABD [24], which showed general synteny and high similarity levels with the corresponding operon from Me. methyllovora str. M2 [15,24].
Several other bacterial strains capable of growing on MSA as the sole carbon and energy source have since been isolated from soil (Hyphomicrobium, Methylobacterium [24,25]), river sediments (Methylobacterium, Flavobacterium, Rhodococcus, Afipia felis [18,26]), seawater (Pedomicrobium [24]) and Antarctic lakes (Afipia felis [26]. Most of them have been tested for the presence of the msm genes either by Southern blotting [25] or by PCR with a set of primers targeting the msmA gene [24,26].
In addition to the results obtained with cultured isolated strains, msmA sequences were also amplified by Baxter et al. [24] from DNA extracted directly from a soil sample and from soil and marine enrichments. A search for msm gene homologues in the Sargasso Sea Metagenome (SSM) database [27] was performed by Leitão et al. [15] who retrieved two scaffolds bearing genes with high identity to the msmABCD cluster and several singletons with high identity to shorter segments of the msm clusters or individual msm genes.
The predicted gene product of gene msmA revealed a unique sequence in the region associated to the Rieske-type [2Fe–2S] cluster, with a longer-than-usual 26-amino acid spacer.
between the two highly conserved cysteine—histidine groups in the CXH-Xn-CXXH conserved motif [15,18,24,26]. However, not all the sequences annotated as MsmA in the databases include this motif. The genome of the first cultured representative of the marine SAR116 clade, Candidatus Puniceispirillum marinum str. IMCC1322, was sequenced [28] and two hypothetical proteins were annotated as hydroxylase alpha subunit of MSAMO (MsmA), but only one of these (NCBI Reference Sequence: YP_003552429) contains the CXH-Xn-CXXH Rieske-associated motif. The absence of this element is observed in many predicted MsmA proteins in the Global Ocean Sampling (GOS) metagenomic data [29] and marine virome sequences, including the one found in bacteriophage HMO-2011, which is known to infect C. Puniceispirillum marinum str. IMCC1322 [30]. On the other hand, the predicted proteomes of methylotrophic strains Methylibium petroleiphilum str. PM1 [31] and Methyloversatilis universalis str. FAM5 [32] contain polypeptides annotated as MsmA with a shorter spacer (n = 17) in the Rieske-associated motif. However, all the organisms so far isolated for which growth on MSA has been positively proved revealed the presence of a Rieske motif with a longer spacer [18,24].
The genes involved in MSA transport have also been investigated. De Marco et al. [18] and Jamshad et al. [33] sequenced and analyzed the msmEFGH operon from Me. methyllovora str. M2, adjacent to operon msmABCD but transcribed in the opposite direction. The protein components encoded by these genes were proposed to constitute an MSA/sulfonate transport system belonging to the ABC-type superfamily of transporters. msmE and msmF genes would encode, respectively, a putative periplasmic substrate-binding protein and a putative membrane-associated protein, while the products of msmG and msmH were proposed to be, respectively, an ATP-binding protein and an outer membrane-associated permease. The search for msm genes in the Sargasso Sea Metagenome by Leitão et al. [15] also yielded a scaffold bearing genes highly identical to the msmEFGH operon.
More recent evidence also suggests that the msm genes are functionally very active in the oceanic environment. In a metatranscriptome study [34] it was shown that in North Atlantic coastal seawater collected at Sapelo Island (Georgia, USA) msm genes similar to those found in C. Puniceispirillum marinum str. IMCC1322 were highly expressed.
In this work, we looked for msmA and msmE gene sequences in previously isolated MSA-degrading strains, as well as in the genomes of two novel marine Filomicrobium isolates. We also amplified and analyzed msmA and msmE gene sequences from coastal ocean surface water metagenomic DNA in order to extend our knowledge on these ecofunctional genetic markers of MSA degradation.
Material and Methods
Enrichment and isolation of marine strains Y and W
A 10 L ocean surface water sample was collected roughly 5 Km off the coast of Matosinhos, Portugal (approximate coordinates 41.1822, -8.7587), and biomass was obtained by successive filtration through 8 μm, 0.45 μm and 0.2 μm filters. Since sampling was performed in publicly accessible ocean waters and involved no environmental risk or damage and no commercial exploitation, no specific permissions were required. This study did not involve endangered or protected species. This biomass was then resuspended in the final 200 mL of the ocean water sample and incubated aerobically at room temperature (15 to 25°C) in the dark, continuously mixed by a magnetic stirrer. Two milliliters of alkaline (pH 8) sodium methanesulfonate 1 M were added to the suspension (10 mM final concentration) and similar amounts were used to spike the enrichment every time a drop to pH 6 or lower was observed. This schedule was maintained over 2 months. The enrichment began to become turbid with cells and a salt precipitate. Small samples were periodically removed to inoculate solid minimal medium MinE.
amended with 3% NaCl and sodium methanesulfonate 10 mM. Several types of bacterial colonies grew on the agar plates, but most did not survive replication. Two strains grew well in these conditions, one producing white colonies (strain W) and one presenting a yellow pigmentation (strain Y). Amplification and sequencing of the SSU rRNA gene from the two strains revealed that both belonged to the genus *Filomicrobium* within the *Hyphomicrobiaceae* (*Alphaproteobacteria*) and had almost identical 16S rRNA gene sequences (99.7% identity).
**DNA extraction from isolated MSA-degrading bacteria**
Previously described strains used in this work were: *Methylobacterium* sp. str. P1 and *Hyphomicrobium* sp. str. P2 [25], *Methylobacterium* sp. str. RD4.1 [36], and *Marinosulfonomonas methylotropha* str. TR3 [20] (kind gift of Prof. J. Colin Murrell, University of East Anglia, UK). The extraction of genomic DNA from bacterial strains, previously described or new, was performed with a Maxwell 16 Cell DNA Purification Kit and the Maxwell 16 robot (Promega Corporation), accordingly to the manufacturer’s instructions.
**Seawater sample collection and metagenomic DNA isolation**
Atlantic Ocean surface water was collected along the coast of Leça da Palmeira, Portugal (coordinates: 41.226956, -8.720528). Briefly, approximately 8 liters of seawater were collected off the rocky shore at high tide into clean bottles, which were immediately transported to the lab in an isothermal bag with ice packs. Five liters of water were successively filtered through 8 μm, 0.45 μm and 0.2 μm filters. Metagenomic DNA was extracted from the three filters with the PowerWater DNA Isolation Kit (MO BIO Laboratories, Inc.) according to the manufacturer’s instructions.
**Amplification and sequencing of the *msmA* and *msmE* genes**
MsmA or MsmE homologs available in the databases were aligned using ClustalW [37] and the corresponding gene sequences were aligned using the output from ClustalW as a scaffold in RevTrans 1.4 [38]. Since the *msmA* and *msmE* genes of *Me. methyllovora* str. M2 and *Ma. methylotropha* str. TR3 are fairly divergent from their Sargasso Sea Metagenome homologs (GenBank accession numbers: EF103447 and EF103448), it proved impossible to design PCR primer pairs common to both types. As such, two primer sets (synthesized by Stabvida Lda., Caparica, Portugal) were employed for each gene: one based on the *Me. methyllovora* str. M2 sequence and the other based on the SSM sequences. The primer pairs used are listed in S1 Table. The DNA extracted from the coastal seawater sample microbes trapped on 0.2, 0.45 and 8 μm filters was tested both with the *msmA* - and *msmE*-directed primers. Different brands of DNA polymerase were used and several reaction parameters had to be adjusted in order to optimize amplification including the usage of additives betaine and DMSO (S2 Table). We initially used Taq Plus DNA polymerase (Citomed, Portugal), commonly employed in our lab. With the amplification of *msmA* from *Hyphomicrobium* sp. str. P2, however, we only obtained results employing the iProof High-Fidelity DNA Polymerase (Bio-Rad). We used GoTaq G2 Flexi DNA polymerase (Promega Corporation) for the rest of our PCR reactions. In the amplification of *msmE* from seawater metagenomic DNA, despite all the attempts, only tenuous bands were obtained, so a nested PCR approach with primer sets SarE133fwd/SarE1119rev and internal primers SarE322fwd/SarE828rev was carried out. Negative controls received PCR water instead of DNA. Positive controls contained DNA from *Me. methyllovora* str. M2 or from SSM clone EF103447, accordingly. The sizes of the resulting PCR products were confirmed by gel electrophoresis. Products were then purified from agarose (GRS PCR & Gel Band Purification Kit, GRISP, Portugal) and cloned into *Escherichia coli* str. DH5α competent cells using the pGEM-T Easy vector System (Promega Corporation) followed by sequencing with BigDye.
Terminator 3 (Applied Biosystems) in an ABI 3730 XL sequencer (Stabvida Lda., Caparica, Portugal) and vector-based primers M13fwd and M13rev. An outline of results and conditions is shown in S2 Table and a full list of the sequences used in this work is presented in S3 Table.
**Genome sequencing of the Filomicrobium isolates**
The genomes of the two Filomicrobium sp. Y and W isolates were sequenced using the MiSeq Illumina sequencing platform by Molecular Research LP (Shallowater, Texas, USA). Coverage was 381x and 294x, respectively. Sequence reads were assembled using NGEN assembler (DNASTAR, Inc.).
**Bioinformatic analysis of the genomic sequencing data**
Through local tblastn (http://www.ncbi.nlm.nih.gov/BLAST/) [39] searches, the genes encoding the putative Msm proteins were found in both str. Y and str. W genome sequences. As the 20,300 bp genome region containing the proposed msm genes in the two strains was 100% identical, we proceeded with the analysis of the segment from just strain Y (deposited in GenBank under accession number KM879220). For open reading frame (ORF) discovery and annotation we used the Glimmer gene prediction software v3.02 (http://www.ncbi.nlm.nih.gov/genomes/MICROBES/glimmer_3.cgi) [40] as well as the ORF Finder (www.ncbi.nlm.nih.gov/projects/gorf/) [41] program both available on the NCBI platform. Blastp searches were performed in order to support the results.
MsmA and MsmE trees were obtained by subjecting sequence alignments to tree inference by PhyML (Maximum Likelihood method) with 100 bootstrap iterations at the Mobyle site (mobyle.pasteur.fr) [42].
**Results**
**Novel msm gene sequences obtained from isolated strains and seawater sample DNA.**
**Amplification of msmA sequences.** PCR using primers aimed at the msmA sequence from *Me. methyllovora* str. M2 (M2A136fwd and M2A1044rev—S2 Table) was successful with Filomicrobium sp. strs. Y and W, Methylobacterium sp. strs. P1 and RD4.1, and Hyphomicrobi um sp. str. P2 and resulted in products around the expected size of 908 bp. With metagenomic seawater DNA, the amplification of a product of the right size (approximately 929 bp) was successfully achieved only with the primer set aimed at the msmA genes found in the SSM (Sar-A124fwd/SarA1053rev), while no amplification was obtained with the primers aimed at soil strain M2. The conditions for successful amplification in each case are summarized in S2 Table.
The sequences of msmA from Filomicrobium sp. strs. Y and W were 100% identical to each other (these data were later confirmed by the whole genome sequencing of the two strains). Also the sequence from soil Methylobacterium sp. str. P1 shared identity and similarity values higher than 99% with that of *Me. methyllovora* str. M2. In general, the msmA gene sequences (and their deduced protein sequences) from the MSA-isolates showed much higher identity to one another (78.7 to 99.3% at the nucleotide level; 84.4 to 99.3% at the amino acid level) than to the SSM sequences (59.4 to 63.4% at the nucleotide level; 73.1 to 75.1% at the amino acid level).
On the other hand, all the msmA sequences obtained from seawater metagenomic DNA, here designated as SCA1 to SCA10, revealed higher identity values relatively to the SSM sequences (77.5 to 99.0% at the nucleotide level; 88.2 to 100% at the amino acid level) than to those from the isolated strains (58.9 to 63.3% at the nucleotide level; 70.0 to 75.1% at the amino acid level). This finding is not unexpected, since these latter amplicons were obtained
employing primers aimed at the SSM sequences and is in line with the differences in GC% observed: indeed, the \textit{msmA} genes of all strains isolated on MSA have moderately high GC% (55 to 62%) while the SSM \textit{msmA} and the SCA sequences have much lower GC% (36 to 38%) (\textit{S3 Table}). Sequences SCA1 and SCA3 seemed to form a subset within the seawater group with reciprocal identity values higher than 99%.
\textbf{Amplification of \textit{msmE} sequences.} In the case of \textit{msmE}, primers M2E76fwd and M2E736rev were successfully employed with isolates \textit{Methylobacterium} sp. str. P1 and \textit{Ma. methylotropha} str. TR3, resulting in products with approximately 697 bp. The predicted peptide sequences from these two amplicons showed very high similarity values between each other and with \textit{Me. methyllovora} str. M2 (98.5 to 99.5% amino acid identity and 99.2 to 99.6% identity at the nucleotide level). On the other hand, these sequences were significantly less similar to their SSM homolog (61.9 to 62.8% amino acid identity; 57.8 to 58.3% identity at the nucleotide level). Amplification of the \textit{msmE} gene from the other isolates (\textit{Methylobacterium} sp. str. RD4.1, \textit{Hyphomicrobium} sp. str. P2 and \textit{Filomicrobium} str. Y and W) failed with both the primer sets aimed at \textit{Me. methyllovora} str. M2 or aimed at the SSM sequences.
In the case of seawater DNA, amplification with primer set SarE133fwd/SarE1119rev yielded just a tenuous band of the expected size (986bp) and several nonspecific bands: therefore a nested-PCR was employed with a second amplification round using internal primers SarE322fwd/SarE828rev: a band of the expected size (around 500 bp) was obtained, cloned, and 30 clones were inspected by insert analysis. Eighteen clones with inserts of roughly the expected size were sent for sequencing. The 18 sequences obtained were all different and only one corresponded to \textit{msmE} (SCE2). The predicted protein sequence from clone SCE2 was 100% similar (99.33% identity; 95.11% identity at the nucleotide level) to \textit{MsmE} from SSM clone EF103447 and much more distant from the homologous sequence of \textit{Me. methyllovora} str. M2, \textit{Methylobacterium} sp. str. P1 or \textit{Ma. methylotropha} str. TR3 (72.08 to 72.73% similarity).
In line with what was observed with gene \textit{msmA}, the \textit{msmE} sequences obtained from cultivated strains showed higher GC% contents (49 to 66.3%), while lower values were observed in the metagenomic sequences (39 to 41.5%) (\textit{S3 Table}).
\textbf{Analysis of the genomic region containing \textit{msm} genes from \textit{Filomicrobium} strain Y}
Since we were not successful in amplifying the \textit{msmE} gene from \textit{Filomicrobium} sp. strains Y and W, their genomes were sequenced and partially annotated. The two genomes are different but share large regions of identical sequence. A 20,300 bp segment (identical in the two strains; deposited in GenBank under accession number KM879220) was examined with gene prediction bioinformatic tools. The search confirmed the presence of two full \textit{msm} operons, \textit{msmABCD} and \textit{msmEFGH}, and uncovered nine extra open reading frames (see Table 1). The order and arrangement of the \textit{msm} genes within the two operons were very similar to those found in \textit{Me. methyllovora} str. M2 (see Fig 1).
The two \textit{msm} operons in \textit{Filomicrobium} sp. strains Y and W are divergently transcribed, like in \textit{Me. methyllovora} str. M2. However, the highest-scoring blast hits for the \textit{msm} genes found on this genomic fragment were obtained with their homologs from marine strain \textit{C. Puniceispirillum marinum} str. IMCC1322.
The additional open reading frames found in this genomic fragment downstream of ORF \textit{msmH} (not shown in Fig 1) encode hypothetical proteins associated with ABC-transport systems, with sulfur compounds metabolism (a sulfite exporter and a putative SoxD (cytochrome c)-SoxC (sulfite dehydrogenase) pair) and with reactions linked to methylotrophic pathways (glycine cleavage system protein T or 3-methyl-2-oxobutanoate hydroxymethyltransferase) (see Table 1).
Table 1. Open reading frames annotated on the 20,300 bp genome segment from *Filomicrobium* sp. str. Y.
| Hypothetical protein encoded | ORF location | Function |
|------------------------------------------------------------------|--------------|---------------------------------|
| LysR family transcriptional regulator a | 1417–335 | Gene regulation |
| Membrane protein sulfite exporter TauE/SafE a | 1688–2449 | Sulfite export |
| glycine cleavage system protein T a | 2462–3457 | Methylotrophic metabolism |
| 3-methyl-2-oxobutanoate hydroxymethyltransferase a | 3640–4548 | C1-moiety transfer |
| molydbdopterin-binding protein sulfite dehydrogenase (SoxC) a | 4903–6315 | Sulfite oxidation |
| cytochrome c (SoxD) a | 6389–6967 | Sulfite oxidation |
| ABC transporter a | 9711–7087 | ABC-transport system |
| ABC transporter a | 10447–9716 | ABC-transport system |
| nitrate/sulfonate/bicarbonate ABC transporter ATPase a | 11624–10737 | ABC-transport system |
| Putative ABC MSA transporter membrane-associated permease component (MsmH) | 12540–11674 | MSA transport |
| Putative ABC MSA transporter ATP-binding component (MsmG) | 13449–12568 | MSA transport |
| Putative ABC MSA transporter membrane-associated permease component (MsmF) | 14372–13485 | MSA transport |
| Putative ABC MSA transporter periplasmic protein (MsmE) | 15705–14569 | MSA transport |
| MSA monooxygenase, hydroxylase alpha subunit (MsmA) | 16483–17754 | MSA metabolism |
| MSA monooxygenase, beta subunit (MsmB) | 17941–18432 | MSA metabolism |
| MSA monooxygenase, ferredoxin (MsmC) | 18546–18923 | MSA metabolism |
| MSA monooxygenase, reductase (MsmD) | 19010–20056 | MSA metabolism |
* Genes not shown in map of Fig 1.
doi:10.1371/journal.pone.0125735.t001
---
Fig 1. Graphic alignment of the *msm* operons from *Me. methylovora* strain M2, *Ma. methylotropha* strain TR3, Sargasso Sea clone EF103447 and *Filomicrobium* strain Y. The arrows represent *msm* genes: A, *msmA*; B, *msmB*; C, *msmC*; D, *msmD*; E, *msmE*; F, *msmF*; G, *msmG*; H, *msmH*. Blue arrows represent genes encoding MSA-monooxygenase and light brown ones correspond to MSA transport genes. *orfX* and *orfY* encode putative regulators of the *msm* operons [24]. A complete description of all genes on the *Filomicrobium* fragment is provided in Table 1.
doi:10.1371/journal.pone.0125735.g001
Conservation of the Rieske-associated motif in the MsmA sequences
All the msmA sequences amplified in this work both from MSA-degrading isolates and seawater metagenomic DNA encode a Rieske-associated motif CXXH-Xn-CXXH with a conserved 26-amino acid spacer between the two cysteine—histidine groups. This amino acid spacer sequence is much shorter (16 to 18 residues) in non-MSAMO monooxygenases. The longer-than-usual spacer found in MsmA proteins appears to be a constant characteristic in spite of the disparate origins of all the MSA-strains and environmental samples analyzed so far.
Phylogenetic trees of Msm sequences
The MsmA sequences obtained in this work were aligned with those previously obtained from cultivated strains, metagenomic SSM sequences EF103447 and EF103448 [15] and similar sequences from the GOS project. The cladogram obtained by PhyML analysis (Fig 2) clearly shows a split into two major groups, one including the Alphaproteobacteria and metagenomic data and another corresponding to Beta and Gammaproteobacteria. Within the Alphaproteobacteria branch two subgroups clearly emerge, one comprising all sequences of the cultivated isolates, and the other made up of just seawater metagenomic sequences. Within the group containing Beta and Gammaproteobacteria, the sequence from Betaproteobacteria Burkholderia cepacia GG4 and Burkholderia sp. RPE67 curiously show a higher proximity to that of Gammaproteobacterium Pseudomonas xanthomarina than to the MsmA from Ralstonia PBA (a Betaproteobacterium), which suggests a possible recent event of inter-class horizontal transfer of the msmA gene between these strains.
The four novel MsmE sequences were aligned with that from Me. methylovora str. M2 as well as with the highest scoring blastp/tblastn hits of Me. methylovora str. M2, including SSM EF103448, C. Puniceispirillum marinum IMCC1322, and GOS sequences. Through the observation of the cladogram (Fig 3) it is easy to recognize two groups of Alpha and Betaproteobacteria. Within the Alphaproteobacteria, a subgroup containing the proteins from low-GC% sequences SCE2 and SSM EF103448 clearly separates from the remaining cluster, which in turn splits in two subgroups, one containing only our novel sequences from cultivated isolates and another including the sequences from GOS and C. Puniceispirillum marinum IMCC1322.
Discussion
The quantitative importance of MSA degradation by bacteria in the general biogeochemical sulfur cycle prompts the investigation of the mechanisms underlying MSA catabolism present in these microorganisms.
Methanesulfonate monooxygenase (MSAMO), an enzyme responsible for the first oxidative step of MSA to the central methylotrophic intermediate formaldehyde, was first discovered in soil bacterium Methylosulphonomonas methylovora strain M2 and sea strains Marinoulomonas methylotropha TR3 and PSCH4. After that, the presence of MSA monooxygenase hydroxylase alpha subunit gene (msmA) was detected in novel strains able to use MSA as the sole source of carbon and energy.
In the present work PCR was successfully used to amplify msmA and msmE genes from cultured bacterial strains isolated from different environmental sources (marine surface water, estuarine sediments and soil) and belonging to diverse genera within the Alphaproteobacteria class. The amplification of msmA from seawater biomass was successful only using a primer pair aimed at sequences previously discovered in the Sargasso Sea Metagenome. Indeed, metagenomic seawater msmA and msmE sequences have low GC contents and, in general, are phylogenetically close to one another, while the homologs from the cultivated strains showed moderately high GC% and clustered separately. It is well accepted that the culturing of most...
Fig 2. Phylogenetic tree of the MsmA sequences. Novel sequences are in boldface. Marine metagenomic sequences are in blue. Only sequences containing the 26 amino acid spacer in the conserved Rieske-associated motif (CXX-X_{26}-CXXH) were considered for the analysis. A maximum likelihood method (PhyML) was used for tree inference. Bootstrap values at nodes are for 100 iterations; only values > 50 are shown.
doi:10.1371/journal.pone.0125735.g002
Fig 3. Phylogenetic tree of the MsmE sequences. Novel sequences are in boldface. Marine metagenomic sequences are in blue. A maximum likelihood method (PhyML) was used for tree inference. Bootstrap values at nodes are for 100 iterations; only values > 50 are shown.
doi:10.1371/journal.pone.0125735.g003
microorganisms present in environmental samples is not feasible, so it is not surprising that the use of culture-independent methods should yield dissimilar results from traditional laboratory isolation [43]. Indeed, in our results we observe a clear cleavage between culturable MSA-degrading bacteria with moderate to high genomic GC content and most of the uncultured types with low GC%.
All the msmA sequences obtained in this work from both isolates and seawater metagenomic DNA were predicted to encode a Rieske-associated motif \( \text{CXH-X}_{26}-\text{CXXH} \) with a conserved 26-amino acid internal spacer, confirming previous findings [15,18,24,26]. Several predicted proteins annotated as MsmA but completely devoid of the four cysteine/histidine residues (needed to bind the Rieske iron—sulfur cluster) exist in databases (GOS metagenome and virome sequences) [30]. Several other polypeptides with a short spacer \((n = 17)\) have also been described as “MSA monooxygenase large subunits” in methylotrophic strains such as \( \text{Methylibium petroleiphilum} \) str. PM1 [31], \( \text{Methyloversatilis universalis} \) str. FAM5 [32], \( \text{Rhodocyclus} \) str. RZ94 (WP_019918795; unpublished), and \( \text{Thiobacillus} \) spp. (WP_018508129; unpublished). However, we are not aware of any hard data showing that any of these strains do actually employ these polypeptides in the oxidation of MSA and the designation as “MSA monooxygenase” in these cases appears to be merely the consequence of blastp-based automatic annotation. Our results combined with previous findings strongly suggest that the longer-than-usual spacer found in MsmA polypeptides may be a characteristic signature of MSAMOs in spite of the disparate origins of the MSA-utilizing strains and environmental samples analyzed so far. A preliminary structure of the two-component hydroxylase of MSAMO has been obtained [44], but specific structure-function studies have not been performed yet so the potential function of such a peculiarly long spacer remains unexplained.
Regarding the MSA/sulfonate transport set of genes (\textit{msmEFGH}), tests were performed with \textit{msmE}, coding for a putative periplasmic substrate binding protein involved in MSA translocation across the membrane [33]. Our attempts were successful only with two of the 6 isolates tested, soil strain \( \text{Methylobacterium} \) P1 and marine strain \( \text{Ma. methylotropha} \) TR3. In the case of seawater metagenomic DNA, within the 30 clones analyzed in this work from that amplicon, only one actually contained \textit{msmE} sequence. This greater difficulty we met in amplifying \textit{msmE} relatively to \textit{msmA} is likely due to the lower levels of conservation of the \textit{msmE} gene sequence. This conjecture is also supported by the results one obtains when searching databases using the MsmE sequence as query: much lower numbers of significant hits and lower identity/similarity levels and scores. Consistently, however, the discrepancy in GC% content witnessed with \textit{msmA} sequences was also observed with \textit{msmE}, with higher values in genes obtained from cultivated isolates than directly from seawater DNA (S3 Table).
The \textit{msmE} gene was found to be part of operon \textit{msmEFGH} in \( \text{Me. methyllovora} \) str. M2 [15,18] and marker exchange mutagenesis data suggested a coordinated expression of this gene with the MSAMO enzyme [33]. This same organization has also been revealed in the whole-genome sequence of \( \text{C. Puniceispirillum marimarinum} \) str. IMCC1322 while in SSM clone EF103447 the two operons are oriented in the same direction. The \textit{Filomicrobium} marine strains Y and W isolated from marine water in this study were shown to contain two full \textit{msm} operons, \textit{msmABCD} and \textit{msmEFGH} adjacent to each other and divergently transcribed (Fig 1). Fittingly, the highest-scoring blast hits for the \textit{msm} genes from these marine isolates were with their homologs from marine strain \( \text{C. Puniceispirillum marimarinum} \) str. IMCC1322. Despite the differences at sequence level, general synteny was demonstrated between the \textit{msm} regions of \( \text{Me. methyllovora} \) str. M2, \( \text{C. Puniceispirillum marimarinum} \) str. IMCC1322 and \textit{Filomicrobium} strains Y and W.
The annotation of the 20,300 bp segment from the genome of \textit{Filomicrobium} sp. strain Y also yielded nine extra open reading frames localized downstream of ORF \textit{msmH}. Three of
these ORFs encoded hypothetical proteins related with sulfite metabolism, namely a sulfite ex-
porter and a SoxD (cytochrome c)-SoxC (sulfite dehydrogenase) pair. According to the litera-
ture, periplasmic sulfite dehydrogenases SoxCD (commonly termed “oxidases” in databases)
involve a catalytic unit, SoxC, bound to a molybdenum cofactor (Moco), which feeds electrons
into the electron transport chain through a cytochrome c (SoxD) [45,46]. These sulfite dehy-
drogenases are associated with energy conservation during the oxidation of reduced inorganic
sulfur species [46]. Indeed, sulfite dehydrogenases are also present in bacteria degrading lon-
ger-chain alkylsulfonates [47,48]. The oxidation of MSA into formaldehyde performed by
MSAMO releases sulfite, which can be further oxidized to sulfate either enzymatically or by re-
action with oxidizing small molecules. Although this is a process that would allow limited pro-
ton-motive force gain, it is theoretically possible to derive energy from the oxidation of the
sulfite released from methanesulfonate [10].
In conclusion, this study delivered a palette of 14 novel *msmA* gene sequences (4 from culti-
vated species and 10 metagenomic) and 4 *msmE* gene sequences (3 from cultivated species and
1 metagenomic). Clearly, the *msmA* genes (and derived proteins) show less sequence variability
than *msmE*. This added to the apparently stable conservation of a peculiarly long Rieske-associ-
ated motif reinforce the value of gene *msmA* as ecofunctional indicator for methanesulfonate
cycling by bacterial natural communities. The data obtained in this work corroborate the suspi-
cion of a strong bias in favor of higher genomic GC-content species when culturing MSA utiliz-
ers, especially from marine water. This notion has to be kept in due account in further studies
on the degradation of MSA when assessing the representativeness of the results vis-à-vis real
natural communities.
**Supporting Information**
**S1 Table.** Primers successfully employed in the amplification of *msmA* and *msmE* genes
from MSA-degrading isolates and seawater metagenomic DNA.
(DOCX)
**S2 Table.** Successful PCR conditions for the amplification of the *msmA* and *msmE* genes
from MSA-degrading isolates and seawater metagenomic DNA. In all cases, PCR was per-
formed in 25 μL volume using the manufacturer’s buffer associated with the Taq polymerase
employed, 1.5 mM MgSO₄ and 200 μM of each dNTP.
(DOCX)
**S3 Table.** Listing of amplification and sequencing results ordered by decreasing GC% con-
tent.
(DOCX)
**Acknowledgments**
We are grateful to the Portuguese Science Fund (FCT) and the Competitiveness Factors Opera-
tional Program (COMPETE) for funding this work through projects PTDC/BIA-MIC/3623/
2012 and FCOMP-01-0124-FEDER-028330. We are also indebted with Prof. J. Colin Murrell
(U. of East Anglia, UK) for the kind gift of *Marinosulfonomonas methylotropha* str. TR3 bio-
mass and advice. We also wish to acknowledge here the valuable help received by lab colleagues
and especially lab managers Rita Reis and Patricia Duarte. Our thanks also to the two anony-
mous reviewers of this paper for their useful observations.
Author Contributions
Conceived and designed the experiments: ACH PDM. Performed the experiments: ACH PDM. Analyzed the data: ACH PDM. Wrote the paper: ACH PDM. Designed the initial project and secured funds to support the study: PDM.
References
1. Vogt M, Liss PS. Dimethylsulfide and climate. In: Quéré C. L., Saltzman ES, editors. Surface Ocean-Lower Atmosphere Processes. 2009. p. 197–232.
2. Gondwe M, Krol M, Gieskes W, Klaassen W, de Baar H. The contribution of ocean-leaving DMS to the global atmospheric burdens of DMS, MSA, SO2, and NSS SO4. Global Biogeochem Cycles. 2003 Jun 30; 17(2).
3. Curson ARJ, Todd JD, Sullivan MJ, Johnston AWB. Catabolism of dimethylsulphoniopropionate: microorganisms, enzymes and genes. Nat Rev Microbiol. 2011; 9(October):849–59.
4. Glasow R von, Crutzen PJ. Model study of multiphase DMS oxidation with a focus on halogens. Atmos Chem Phys. 2004 Apr 14; 4(3):589–608.
5. Barnes I, Hjorth J, Mihalopoulos N. Dimethyl sulfide and Dimethyl sulfoxide and their oxidation in the atmosphere. Chem Rev. 2006; 106(3):940–75. PMID:16522014
6. Andreae MO. The ocean as a source of atmospheric sulfur compounds. In: Patrick Buat-Ménard, editor. The role of air-sea exchange in geochemical cycling. 1986. p. 331–62.
7. Hynes AJ, Wine PH, Semmes DH. kinetics and mechanism of OH reactions with organic sulfides. J Phys Chem. 1986; 90(23):4148–56.
8. Koga S, Tanaka H. numerical study of the oxidation process of dimethylsulfide in the marine atmosphere. J Atmos Chem. 1993; 17(3):201–28.
9. Mihalopoulos N, Nguyen BC, Boissard C, Putaud JP, Belviso S. Field study of dimethylsulfide oxidation in the boundary layer: variations of dimethylsulfide, methanesulfonic acid, sulfur dioxide, non-sea-salt sulfate and alken nuclei at a coastal site. J Atmos Chem. 1992; 14(1–4):459–77.
10. Kelly DP, Murrell JC. Microbial metabolism of methanesulfonic acid. Arch Microbiol. 1999; 5(6):341–8.
11. Chen X, Minofar B, Jungwirth P, Allen HC. Interfacial molecular organization at aqueous solution surfaces of atmospherically relevant dimethyl sulfoxide and methanesulfonic acid using sum frequency spectroscopy and molecular dynamics simulation. J Phyl Chem. 2010; 114(47):15546–53. doi:10.1021/jp1078339 PMID: 21062065
12. Charlson RJ, Lovelock JE, Andreae MO, Warren SG. Oceanic phytoplankton, atmospheric sulphur, cloud albedo and climate. Nature. 1987; 326:655–61.
13. Langner J, Rodhe H. A global three-dimensional model of the tropospheric sulfur cycle. J Atmos Chem. 1991; 13:225–63.
14. Whung PY, Saltzman ES, Spencer MJ, Mayewski PA. Two-hundred-year record of biogenic sulfur in a south greenland ice core (20D). J Geophys Res. 1994; 99(D1):1147–56.
15. Leitão E, Moradas-Ferreira P, De Marco P. Evidence of methanesulfonate utilizers in the Sargasso Sea metagenome. J Basic Microbiol. 2009 Sep; 49(Suppl 1):S24–30. doi: 10.1002/jobm.200800223 PMID: 19322831
16. Legrand M, Feniet-Saigne C. Methanesulfonic acid in south polar snow layers: a record of strong El Nino? Geophys Res. 1991; 18(2):187–90.
17. Kelly DP, Baker SC. The organosulphur cycle: aerobic and anaerobic processes leading to turnover of C1-Sulphur compounds. FEMS Microbiol. 1990; 87(3–4):241–6.
18. De Marco P, Moradas-Ferreira P, Higgins TP, McDonald I, Kenna EM, Murrell JC. Molecular analysis of a novel methanesulfonic acid monoxygenase from the methyloph Methylsulfonomonas methyllovora. J Bacteriol. 1999 Apr; 181(7):2244–51. PMID: 10094704
19. Baker SC, Kelly DP, Murrell JC. Microbial degradation of methanesulphonic acid: a missing link in the biochemical sulphur cycle. Nature. 1991; 350:627–8.
20. Thompson AS, Owens NJP, Murrell JC. Isolation and characterisation of methanesulfonic Acid-degrading bacteria from the marine environment. Appl Environ Microbiol. 1995 Jun; 61(6):2388–93. PMID: 16535055
21. Holmes AJ, Kelly DP, Baker SC, Thompson AS, De Marco P, Kenna EM, et al. and Marinosulfonomonas methylotropha gen. nov., sp. nov.: novel methylotrophs able to grow on methanesulfonic acid. Arch Microbiol. 1997; 167:46–53. PMID: 9000341
22. Higgins TP, De Marco P, Murrell JC. Purification and molecular characterization of the electron transfer protein of methanesulfonic acid monoxygenase. J Bacteriol. 1997 Mar; 179(6):1974–9. PMID: 9068643
23. Murrell JC, Higgins T, Kelly DP. Bacterial metabolism of methanesulfonic acid. Microbiol Atmos Trace Gases. 1996; 39:243–53.
24. Baxter NJ, Scanlan J, De Marco P, Wood AP, Murrell JC. Duplicate copies of genes encoding methanesulfonate monooxygenase in Marinisosulfonomonas methylotropha strain TR3 and detection of methanesulfonate utilizers in the environment. Appl Environ Microbiol. 2002 Jan; 68(1):289–96. PMID: 11772638
25. De Marco P, Murrell JC, Bordalo AA, Moradas-Ferreira P. Isolation and characterization of two new methanesulfonic acid-degrading bacterial isolates from a Portuguese soil sample. Arch Microbiol. 2000 Feb; 173(2):146–53. PMID: 15643932
26. De Marco P, Murrell JC, Bordalo AA, Moradas-Ferreira P. Isolation and characterization of two new methanesulfonic acid-degrading bacterial isolates from a Portuguese soil sample. Arch Microbiol. 2000 Feb; 173(2):146–53. PMID: 10795686
27. Moosvi SA, Pacheco CC, McDonald IR, De Marco P, Pearce DA, Kelly DP, et al. Isolation and properties of methanesulfonate-degrading Afipia felis from Antarctica and comparison with other strains of A. felis. Environ Microbiol. 2005 Jan; 7(1):22–33. PMID: 16391054
28. Gifford SM, Sharma S, Booth M, Moran MA. Expression patterns reveal niche diversification in a marine microbial assemblage. ISME J. Nature Publishing Group; 2013 Feb; 7(2):281–98.
29. Kelly DP, Baker SC, Trickett J, Davey M, Murrell JC. Methanesulphonate utilization by a novel methylotrophic bacterium involves an unusual monooxygenase. Microbiology. 1994 Jun 1; 140(6):1419–26.
30. Delcher AL, Bratke KA, Powers EC, Salzberg SL. Identifying bacterial genes and endosymbiont DNA with Glimmer. Bioinformatics. 2007; 23(21):2947–8. PMID: 17846036
31. McGinnis S, Madden TL. BLAST: At the core of a powerful and diverse set of sequence analysis tools. Nucleic Acids Res. 2004; 32:20–5.
45. Friedrich CG, Quentmeier A, Bardschewsky F, Rother D, Orawski G, Hellwig P, et al. Redox control of chemotrophic sulfur oxidation of Paracoccus pantotrophus. In: Dahl C, Friedrich CG, editors. Microbial Sulfur Metabolism. Springer Berlin Heidelberg; 2008. p. 139–50.
46. Denger K, Weinitschke S, Smits THM, Schleheck D, Cook AM. Bacterial sulfite dehydrogenases in organotrophic metabolism: separation and identification in Cupriavidus necator H16 and in Delftia acidovorans SPH-1. Microbiology. 2008 Jan; 154(Pt 1):256–63. doi: 10.1099/mic.0.2007/011650-0 PMID: 18174144
47. Reichenbecher W, Murrell JC. Linear alkanesulfonates as carbon and energy sources for gram-positive and gram-negative bacteria. Arch Microbiol. 1999; 171(6):430–8. PMID: 10369899
48. Reichenbecher W, Kelly DP, Murrell JC. Desulfonation of propanesulfonic acid by Comamonas acido- vorans strain P53: evidence for an alkanesulfonate sulfonatase and an atypical sulfite dehydrogenase. Arch Microbiol. 1999; 172(6):387–92. PMID: 10591848 | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2549,
1
],
[
2549,
6832,
2
],
[
6832,
10816,
3
],
[
10816,
14812,
4
],
[
14812,
18361,
5
],
[
18361,
22630,
6
],
[
22630,
25680,
7
],
[
25680,
29453,
8
],
[
29453,
29902,
9
],
[
29902,
30207,
10
],
[
30207,
34779,
11
],
[
34779,
37974,
12
],
[
37974,
42229,
13
],
[
42229,
44100,
14
],
[
44100,
45116,
15
]
]
} |
1a677c5f7eadbc60a3f2462a084fdda7bff0a61d | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 6,
"total-fallback-pages": 0,
"total-input-tokens": 14877,
"total-output-tokens": 2830,
"fieldofstudy": [
"Physics",
"Computer Science"
]
} | Entropy of 2+1 dimensional de Sitter space in terms of brick wall method
Won Tae Kim∗
Department of Physics and Basic Science Research Institute,
Sogang University, C.P.O. Box 1142, Seoul 100-611, Korea
(August 1998)
Abstract
We calculate the statistical entropy of a scalar field on the background of three-dimensional De Sitter space in terms of the brick wall method and finally derive the perimeter law of the entropy.
∗Electronic address: [email protected]
Entropy of black holes has an universal area law \cite{1} and the entropy of the well-known
Schwarzschild black hole satisfies the area law by means of thermal radiation based on the
quantum field theory \cite{2}. On the other hand, 't Hooft has argued that when one calculates
the black hole entropy, the modes of a quantum field in the vicinity of a black hole horizon
should be cut off due to gravitational effects rather than infinitely piling up by suitably
choosing a brick-wall cutoff just beyond the horizon \cite{3}. Most of brick-wall calculations have
been done for the asymptotically flat cases.
For 2+1 dimensional anti-de Sitter space, Bañados, Teitelboim, and Zanelli (BTZ) have
obtained a black hole solution which is asymptotically anti-de Sitter spacetime rather than
asymptotically flat \cite{4}. The thermodynamic properties has been extensively studied in
this black hole \cite{5,6}. Recently the statistical entropy of the de Sitter(DS) space is studied
in terms of Chern-Simons formulation \cite{7}. The DS space has a cosmological horizon and
asymptotically non-flat spacetime, furthermore the spacetime is bounded by the horizon as
the two-dimensional cavity.
In this Brief Report, we shall calculate the entropy of a scalar field on the DS space
background by using the brick wall method. As a result, the divergent entropy is obtained
in the vicinity of horizon and by properly choosing the brick-wall cutoff we finally obtain
the expected perimeter law.
Let us start with the following action
\[
I = \frac{1}{2\pi} \int d^3 x \sqrt{-g} \left[ R - \frac{2}{l^2} \right],
\]
where \(\Lambda = \frac{1}{l^2}\) is a cosmological constant. Then the classical equation of motion yields the
DS metric as
\[
ds^2 = -g(r) dt^2 + \frac{1}{g(r)} dr^2 + r^2 d\theta^2,
\]
\[
g(r) = \left(1 - \frac{r^2}{l^2}\right).
\]
The horizon is located at \(r = l\) and our spacetime is defined within \(0 \leq r \leq l\). The inverse
of Hawking temperature is given by
\[ \beta_H = 2\pi l. \]
(5)
Let us now introduce a Klein-Gordon field equation on the DS background,
\[ \Box \Phi = 0, \]
(6)
where we consider the massless case for simplicity. The Eq. (6) can be solved through the separation of variables and we can write the wave function as
\[ \Phi(r, \phi, t) = e^{-iEt} e^{i m \phi} R_E^m(r), \]
(7)
where \( m \) is an azimuthal quantum number of the scalar field. Then the radial wave equation is written as
\[ \frac{1}{rg(r)} \partial_r [rg(r)\partial_r R_E^m(r)] + k^2(r, m, E) R_E^m(r) = 0, \]
(8)
where the radial wave number is given by
\[ k^2(r, m, E) = \frac{1}{g^2(r)} \left[ E^2 - \frac{m^2g(r)}{r^2} \right] \]
(9)
in the WKB approximation [3]. According to the semi-classical quantization rule, the radial wave number is quantized as
\[ \pi n_r(m, E) = \int_L^{r_H-\epsilon} dr k(r, m, E), \]
(10)
under the brick wall boundary conditions: \( \Phi = 0 \) at \( r = L, \ r = r_H - \epsilon \). Note that \( n_r \) is assumed to be a nonnegative integer, and \( \epsilon \) and \( L \) are ultraviolet and infrared regulators, respectively where \( \epsilon > 0 \) and \( 0 \leq L \leq r_H - \epsilon \). In this range, the energy \( E \) is always positive and the wave number \( k \) is real.
The free energy at inverse temperature \( \beta \) is given by
\[ e^{-\beta F} = \prod_k \left[ 1 - e^{-\beta E_k} \right]^{-1}, \]
(11)
where $K$ represents the set of quantum numbers. By using Eq. (10), the free energy can be rewritten as
$$
F = \frac{1}{\beta} \sum_K \ln \left[ 1 - e^{-\beta E_K} \right] \approx \frac{1}{\beta} \int dn_r \int dm \ln \left[ 1 - e^{-\beta E} \right]
$$
$$\approx -\frac{1}{\pi} \int dm \int dE \frac{n_r}{e^{\beta E} - 1} \int dr k(r, m, E), \quad (12)
$$
where we have taken the continuum limit in the first line and integrated by parts in the second line in Eq. (12). The explicit form of the free energy is given by
$$
F = -\frac{1}{2} \int_0^\infty dE \frac{1}{E^2} \int dE E \frac{1}{g^3(r)} \int dr k(r, m, E).
$$
$$\approx -\frac{1}{\pi} \int dm \int dE \frac{1}{e^{\beta E} - 1} \int dr k(r, m, E), \quad (13)
$$
Note that the integration with respect to angular variable $m$ is taken over values for which the square root is real. Performing the remaining integrations, the free energy is written by
$$
F = -\frac{\zeta(3) l^3}{2 \beta^3} \left( \frac{1}{\sqrt{l^2 - (r_H - \epsilon)^2}} - \frac{1}{\sqrt{l^2 - L^2}} \right).
$$
Let us now evaluate the entropy for the massless field, which can be obtained from the free energy (14) at the Hawking temperature, then the entropy is
$$
S = \beta^2 \frac{\partial F}{\partial \beta} \bigg|_{\beta = \beta_H}
\approx 4\pi a \left( \frac{l}{\sqrt{l^2 - (r_H - \epsilon)^2}} - \frac{l}{\sqrt{l^2 - L^2}} \right), \quad (15)
$$
where the constant is defined by $a \equiv \frac{3\zeta(3)}{2\pi^3}$. This result shows that the entropy behaves as $1/\sqrt{\epsilon}$ at $\epsilon \to 0$ which corresponds to the ultraviolet divergence of the entropy. On the other hand, the distance of the brick wall from the horizon is related to the ultraviolet cutoff as
$$
\tilde{\epsilon} = \int_{r_H - \epsilon}^{r_H} \frac{dr}{\sqrt{g(r)}}
\approx l \left( \frac{\pi}{2} - \sin^{-1} \frac{l - \epsilon}{l} \right).
$$
4
Then the entropy (15) is neatly represented in terms of the invariant cutoff (16) as follows,
\[ S = 4\pi a \left( \frac{1}{\sin \tilde{\epsilon}} - 1 \right) \]
(17)
where we simply fix the infrared cutoff as \( L = 0 \) without loss of generality since there does not exist any infrared divergence even though we consider massless scalar field. Note that the entropy is always positive in Eq. (17). If we choose the cutoff as
\[ \tilde{\epsilon} = l \sin^{-1} \left( \frac{a}{a + l} \right) \]
(18)
then the entropy is written by the perimeter law
\[ S = 2 \cdot 2\pi r_+ . \]
(19)
Note that for \( l \gg a \), the invariant cutoff is simply written as \( \tilde{\epsilon} \approx a \) and it gives the entropy \( S = 4\pi l \).
As a comment, at first sight the infrared regulator \( L \) seems to be necessary in our massless field, however, it does not play an important role in DS space since our spacetime is bounded by the horizon and in some sense spacetime is surrounded by the black hole which is similar to the confined field in the cavity. Therefore, the finite volume of DS spacetime evaluated as \( V(r) = 2\pi l^2 \left[ 1 - \sqrt{1 - r^2/l^2} \right] \) removes infrared divergence in contrast to the asymptotically flat spacetime whose volume is infinite.
ACKNOWLEDGMENTS
This work was supported by the Korea Research Foundation (1997).
REFERENCES
[1] J. D. Bekenstein, Lett. Nuovo Cimento 4, 737 (1972); Phys. Rev. D7, 2333 (1973); D9, 3292 (1974).
[2] S. W. Hawking, Commun. Math. Phys. 43, 199 (1975).
[3] G. ’t Hooft, Nucl. Phys. B256, 727 (1985).
[4] M. Bañados, C. Teitelboim, and J. Zanelli, Phys. Rev. Lett. 69, 1849 (1992).
[5] S. Carlip and C. Teitelboim, Phys. Rev. D51, 622 (1995).
[6] S. Carlip, Phys. Rev. D51, 632 (1995).
[7] J. Maldacena and A. Strominger, gr-qc/9801096. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
472,
1
],
[
472,
2420,
2
],
[
2420,
3872,
3
],
[
3872,
5744,
4
],
[
5744,
7116,
5
],
[
7116,
7573,
6
]
]
} |
762b9abb1d3264d7342594a8468c484e4942a056 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 15,
"total-fallback-pages": 0,
"total-input-tokens": 63175,
"total-output-tokens": 12920,
"fieldofstudy": [
"Biology"
]
} | The Effect of the Pore Entrance on Particle Motion in Slit Pores: Implications for Ultrathin Membranes
Armin Delavari and Ruth Baltus *
Department of Chemical & Biomolecular Engineering, Clarkson University, Potsdam, NY 13699-5705, USA; [email protected]
* Correspondence: [email protected]; Tel.: +1-315-268-2368
Received: 30 June 2017; Accepted: 31 July 2017; Published: 10 August 2017
Abstract: Membrane rejection models generally neglect the effect of the pore entrance on intrapore particle transport. However, entrance effects are expected to be particularly important with ultrathin membranes, where membrane thickness is typically comparable to pore size. In this work, a 2D model was developed to simulate particle motion for spherical particles moving at small Re and infinite Pe from the reservoir outside the pore into a slit pore. Using a finite element method, particles were tracked as they accelerated across the pore entrance until they reached a steady velocity in the pore. The axial position in the pore where particle motion becomes steady is defined as the particle entrance length (PEL). PELs were found to be comparable to the fluid entrance length, larger than the pore size and larger than the thickness typical of many ultrathin membranes. Results also show that, in the absence of particle diffusion, hydrodynamic particle–membrane interactions at the pore mouth result in particle “funneling” in the pore, yielding cross-pore particle concentration profiles focused at the pore centerline. The implications of these phenomena on rejection from ultrathin membranes are examined.
Keywords: membrane transport models; ultrathin membranes; ultrafiltration; microfiltration; pore entrance
1. Introduction
With the development of many new fabrication techniques, there has been an increasing interest in using ultrathin membranes in water purification, for biological, pharmaceutical and other separations as well as for sensing devices [1–13]. The thickness of these membranes is generally tens to several hundred nanometers, orders of magnitude smaller than the thickness of conventional membranes made using phase inversion and track-etch techniques. Ultrathin membranes allow for higher permeabilities with lower applied pressure compared to conventional membranes.
The degree of separation in porous membranes is typically characterized using the rejection coefficient, $\sigma_f$, defined by [14]:
$$\sigma_f = 1 - \left[ \frac{N}{\bar{C}_\infty} \right]_{\Delta C_\infty=0}$$
where $N$ is the particle flux, $f_0$ is the solvent velocity across the membrane and $C_\infty$ is the particle concentration in the feed. A general expression for rigid particles is developed by averaging the particle velocity over the pore cross-section:
$$\sigma_f = 1 - \frac{\left\langle \frac{\pi}{4h} \int G \cdot \exp\left(\frac{-E}{kT}\right) d\theta \right\rangle}{\left\langle u \right\rangle}$$
Membranes 2017, 7, 42; doi:10.3390/membranes7030042 www.mdpi.com/journal/membranes
where $\vec{u}$ is the fluid velocity in the absence of particles in the pore, $\hat{e}$ is the orientation of the particle in the pore, $G$ is the hydrodynamic lag coefficient and $E$ is the interaction energy between the particle and the pore wall. The brackets $< >$ are defined as an average over the pore cross-sectional area.
Equation (2) is focused on particle transport within a pore and is based on the assumption that thermodynamic equilibrium is established at the pore mouth. This means that a Boltzmann expression can be used to give the time-averaged probability of finding a particle at a given position and orientation in the pore. When only steric interactions are considered, the probability of finding a particle at a stERICALLY allowed location in the pore is 1 ($E = 0$) and is 0 ($E = \infty$) for disallowed positions and orientations [14]. A Boltzmann distribution in the pore is valid when the particles have sufficient time to diffuse radially inside the pore and sample all allowed positions and orientations as they are carried through the pore. For small particles and long pores, this assumption is easily satisfied. However, its validity for ultrathin membranes is questionable.
Equation (2) is also based on the assumption that fluid travels at a steady velocity in the pore. At the pore entrance, fluid is funneled from a reservoir where it has a relatively low velocity and accelerates as it enters the pore, reaching a steady velocity after it travels a distance we define as the ‘fluid entrance length’ (FEL). Fluid entrance effects for flow into cylindrical tubes and channels has been an extensively studied problem where the distance required for flow to transition from plug flow (flat profile) to parabolic flow has been determined using both theoretical and experimental approaches. For flow into parallel plate channels, the FEL can be estimated using [15,16].
$$\frac{FEL}{h/2} = 1.25 + 0.088Re_h$$ \hspace{1cm} (3)
where $h$ is the width of the channel and $Re_h$ is defined with the channel width as the characteristic length scale. Equation (3) shows that when $Re_h$ is small, the FEL is of order of the pore $\frac{1}{2}$ width. When the pore length (i.e., membrane thickness) and pore size are of comparable magnitude (as is the case with ultrathin membranes [4,6,8,10,17]), Equation (3) predicts that the fluid profile may not be fully developed within the membrane, making Equation (2) of questionable validity for estimating particle rejection.
Previous studies of pore entrance effects have generally focused on particle motion in the reservoir outside the pore, examining the balance of forces controlling particle motion as it approaches a pore. The impact of different pore entrance shapes was studied by Bowen and Sharif [18] who used a finite element approach to relate the hydrodynamic and electrostatic forces on a spherical particle to the critical velocity of the system. Interactions of the particle with the pore walls at the entrance were studied for different pore mouth geometries; results showed that a rounded pore entrance has the highest critical velocity and is therefore the optimal choice.
Kao et al. [19] studied the impact of the pore entrance on trajectories of microscale particles by considering the influence of inertial and molecular forces for different Stokes number, showing that for low Stokes number, particle trajectories are affected largely by hydrodynamic forces. Other studies of pore entrance effects investigated particle–particle interactions in the vicinity of the pore entrance [20,21], the impact of electrostatic, Brownian and hydrodynamic forces on particle trajectories [22,23] and the effect of pore entrance size on the mass transfer of gas molecules [24,25].
As noted, previous studies have generally focused on particle transport within a pore or on the motion of a particle as it approaches a pore. In this study, we couple these to examine the impact of the pore entrance on intrapore particle transport. We report results from a model describing spherical particle motion from the reservoir outside the membrane into a slit pore. Solution of the coupled equations for fluid and particle motion yields the particle entrance length (PEL). Analogous to the FEL, the PEL is defined as the axial distance traversed by the particle within the pore before it achieves a steady velocity. The relationship between the PEL and the FEL, relative particle size, pore entrance geometry and $Re_h$ are examined and discussed in the context of ultrathin membranes where pore entrance effects are expected to be important.
2. Theoretical Model
2.1. Modeling Domain
Particle motion from a reservoir into a slit pore was modeled in a 2D domain, with the system shown in Figure 1.
2.2. Immersed Boundary Method
The immersed boundary method was used to simulate particle motion from the reservoir into the pore [26]. In this approach, the particle was considered in a Lagrangian frame; an Eulerian perspective was used for the surrounding fluid. Using the Arbitrary Lagrangian–Eulerian concept, an appropriate mesh map was generated in the domain. To compute the forces imposed on the particle surface, a finite element method was employed to solve the Navier–Stokes equation coupled with Newton’s equation of motion throughout the 2D domain. In this approach, the particle boundaries consist of multiple nodes, with the number of nodes dependent on the mesh resolution. This approach enables one to appropriately include steric exclusion and the coupled motion of fluid and particle (i.e., the fluid impacts particle trajectory and particle motion impacts the fluid).
2.3. Model Equations
The particle trajectory was determined by solving Newton’s equation of motion in the absence of Brownian motion.
\[ m_{\text{particle}} \frac{d\vec{u}_{\text{particle}}}{dt} = \vec{F}_{\text{hydrodynamic}} \]
where \( m_{\text{particle}} \) is the particle mass and \( \vec{F}_{\text{hydrodynamic}} \) is the hydrodynamic force on the particle surface. The Navier–Stokes and continuity equations were used to simulate fluid flow:
\[ \rho_{\text{fluid}} \left( \frac{\partial \vec{u}}{\partial t} + \vec{u} \cdot \nabla \vec{u} \right) = \nabla \cdot \left[ -p \vec{1} + \mu \left( \nabla \vec{u} + \left( \nabla \vec{u} \right)^T \right) \right] \]
\[ \rho_{\text{fluid}} \nabla \cdot \vec{u} = 0 \]
The hydrodynamic force on the particle surface was determined by integration of the normal and shear stresses over the particle surface:
$$\vec{F}_{\text{Hydrodynamic}} = \int \left[ -P\vec{I} + \mu \left( \nabla \vec{u} + \left( \nabla \vec{u} \right)^T \right) \right] \cdot \vec{n} \, dS \quad (7)$$
where $\vec{n}$ is the normal vector on the particle surface.
The particle diameter, $d_{\text{particle}}$, and the fluid velocity at the pore centerline, $u_o$, were used as scaling parameters to non-dimensionalize these equations:
$$\overline{u} = \frac{u}{u_o}, \quad \nabla = \frac{d_{\text{particle}}}{u_o}, \quad \nabla t = \frac{t}{u_o d_{\text{particle}}}, \quad P = \frac{P_{\text{particle}}}{u_o^2}, \quad \mu = \frac{\mu_{\text{particle}}}{u_o d_{\text{particle}}}, \quad \text{Re}_{\text{particle}} = \frac{\rho_{\text{fluid}} u_o d_{\text{particle}}}{\mu}, \quad S = \frac{s}{d_{\text{particle}}^2} \quad (8)$$
Note that the characteristic length in $\text{Re}_{\text{particle}}$ is the particle diameter, different than $\text{Re}_h$ where the channel width is the characteristic length (Equation (3)). The dimensionless forms of equations of Equations (4)–(6) are
$$\overline{m}_{\text{particle}} \frac{d \overline{u}_{\text{particle}}}{dt} = \frac{\vec{F}_{\text{hydrodynamic}}}{\mu d_{\text{particle}} u_o} \quad (9)$$
$$\text{Re}_{\text{particle}} \left( \frac{d \vec{u}}{dt} + \vec{u} \cdot \nabla \vec{u} \right) = \nabla \cdot \left[ -\overline{P} \vec{I} + \left( \nabla \vec{u} + \left( \nabla \vec{u} \right)^T \right) \right] \quad (10)$$
$$\nabla \cdot \vec{u} = 0 \quad (11)$$
and the dimensionless hydrodynamic stress is:
$$\frac{\vec{F}_{\text{hydrodynamic}}}{\mu d_{\text{particle}} u_o} = \int \left[ -\overline{P} \vec{I} + \left( \nabla \vec{u} + \left( \nabla \vec{u} \right)^T \right) \right] \cdot \vec{n} \, dS \quad (12)$$
Particle diffusion in both cross-pore and axial directions is neglected in this analysis. Particle–particle interactions are also neglected. While the particle diameter was used as the length scale to non-dimensionalize the governing equations, results are presented with the pore width as the length scale to non-dimensionalize positions in the pore and in the reservoir outside the pore.
For all calculations, a parabolic velocity profile was defined at the reservoir inlet ($X = -10$) and the pressure at the pore outlet was set to zero. While the reservoir in the image in Figure 1 spans from $Y = -10$ to $Y = +10$, the reservoir in our simulations spanned from $Y = -24$ to $Y = +24$. A no-slip condition was applied on the particle surface, on the pore wall and on the reservoir boundaries. With the reservoir width 24 times larger than the pore width, the reservoir boundaries have minimal effect on the flow into the pore. At time $t = 0$, a spherical particle was placed at position $X_{\text{initial}}, Y_{\text{initial}}$ in the reservoir outside the pore.
2.4. Model Parameter Values
For most simulations, the average fluid velocity at the reservoir inlet ($X = -10$) was $4.08 \times 10^{-4}$ m/s, which was based on a volumetric flow rate of 0.5 mL/min, typical of filtration measurements with track-etched membranes [27]. This leads to an average velocity in the pore of 0.0098 m/s and $\text{Re}_{\text{particle}}$ is of $O(10^{-4})$. Our results will show that the particle trajectories and the PEL are independent of velocity when the fluid velocity is <0.1 m/s. The pore length was set to 20 times the pore width, which was found to be sufficient for the particle to reach a steady velocity in the pore. The radius of curvature of the pore mouth, $r_c$, varied from 10 to 500% of the pore width, with most calculations performed with $r_c = h/2$. Calculations were limited to spherical particles, with dimensionless size $\lambda = 0.1$ to 0.9, where $\lambda$ is defined as $d_{\text{particle}}/(h/2)$. Particle trajectories
were determined for particles with specific gravity = 1.1, considering these to be representative of
viruses and other bioparticles or macromolecules. Some simulations involved particles with specific
gravity = 11, considering these to be representative of metallic nanoparticles. Simulations were
repeated with different initial particle positions in the reservoir, $X_{\text{initial}} = -1$ to $-9$ and $Y_{\text{initial}} = 0$ to 5.
The largest $Y_{\text{initial}}$ was chosen to be 5 because the pore-to-pore distance in track-etched membranes is
estimated to be ~10 times the pore radius.
2.5. Solution of Model Equations
Equations (9)–(12) were solved simultaneously using COMSOL Multiphysics (COMSOL, Inc,
Burlington, MA, USA), yielding particle position and velocity as the particle moves from the reservoir
into the pore. Mesh refinement was performed to prevent distorted elements.
The particle accelerates as it moves from the reservoir into the pore and eventually reaches a
steady velocity. The total force on the particle reaches its maximum when the particle passes the
entrance of the pore ($X = 0$) and gradually declines as the particle attains its steady velocity. The particle
entrance length (PEL) is defined as the axial particle position ($X$) where the particle has achieved a
velocity 99.9% of its steady velocity.
The cross-sectional particle concentration profile at the PEL was determined by repeating
simulations with 102 particles uniformly distributed in the reservoir.
3. Results and Discussion
3.1. Particle Trajectories
The trajectories of four particles are illustrated in Figure 2 where the particles were initially placed
at $X_{\text{initial}} = -5$ and $Y_{\text{initial}} = 0, 0.6, 1.4$ and 2.2. Note that the computational domain is bounded at
$X = -10$ (see Figure 1), with a parabolic velocity profile prescribed at this boundary. This surface
has width $= 48$ dimensionless length units, whereas the pore width is 2 dimensionless length units.
The reservoir is sufficiently large such that the velocity profile is approximately flat until it is influenced
by the pore entrance (at $X \approx -2$).
**Figure 2.** Velocity profile of fluid and particle trajectories for particles with $\lambda = 0.1$ and specific
gravity = 1.1, initially placed in the reservoir at $X_{\text{initial}} = -5$ and $Y_{\text{initial}} = 0, 0.6, 1.4$ and 2.2. For these
simulations, $r_c = h/2$.
3.2. Particle Velocities
The effect of the initial particle placement in the reservoir on particle velocity is presented in
Figure 3a for particles with $\lambda = 0.1$ and in Figure 3b for particles with $\lambda = 0.8$. Here, particles were
initially placed in the reservoir at $Y_{\text{initial}} = 1$ and $X_{\text{initial}} = -1, -3, -5, -7$ and $-9$. These results
illustrate the acceleration experienced by the particles as they approach and enter the pore. Results show that particle trajectories are indistinguishable when \(-9 < X_{\text{initial}} < -5\) because particles initially placed more than five pore widths from the membrane surface get carried into the pore along the same streamlines. Results also show that the steady intrapore particle velocity increases as |\(X_{\text{initial}}|\) increases because particles initially placed farther from the membrane surface are carried towards the pore centerline.
The larger particles (\(\lambda = 0.8\), Figure 3b) show a smaller and tighter range of intrapore particle velocities when compared to the smaller particles (\(\lambda = 0.1\), Figure 3a). This results from larger hydrodynamic resistances experienced by the larger particles and from steric constraints which restrict the larger particles to centerline positions (\(-0.6 < Y < 0.6\)).
Figure 3. Particle velocity as a function of \(X\) for particles moving from the reservoir into the pore for (a) \(\lambda = 0.1\) and (b) \(\lambda = 0.8\); particle specific gravity = 1.1, \(Y_{\text{initial}} = 1\) and \(X_{\text{initial}} = -1, -3, -5, -7, -9\).
### 3.3. Fluid Entrance Length
Simulations were also performed with a particle-free system and the fluid velocity profile in the pore was examined. The FEL was defined as the axial (\(X\)) position in the pore where the axial velocity was 99.9% of its fully developed value (where velocity is independent of axial position). Results for FEL/(\(h/2\)) as a function of \(Y\) are shown in Figure S1 in the Supplementary Information. These results show that FEL/(\(h/2\)) values are constant across the pore (FEL/(\(h/2\)) = 1.65 to 1.75), except at \(Y = 0.3\)–0.4 (minimum FEL/(\(h/2\)) = 1.26 at \(Y = 0.4\)). To transition from a relatively flat velocity profile at \(X = 0\) to a steady parabolic profile at \(X = \text{PEL}\), the fluid velocity at the centerline increases and the fluid velocity near the pore wall decreases. Between these two regions, the fluid velocity does not change significantly; hence the minimum FEL at this point.
To be consistent with previous reports of entrance lengths [15,16], we define FEL based on the velocity at the pore centerline, where FEL/(\(h/2\)) = 1.69. This is larger than the value predicted from Equation (3) (FEL/(\(h/2\)) = 1.25). As in other previous studies of entrance lengths, the FEL reported in Equation (3) was based on a centerline velocity equal to 99% of the value at fully developed flow, whereas our results were determined with a tighter criterion (99.9%). When the criterion is set to 99%, our simulations yield a centerline FEL/(\(h/2\)) = 1.28, in close agreement with Equation (3). In addition, Equation (3) was derived by considering boundary layer flow near the channel wall when flow transitions from a flat profile to a parabolic flow [15], a situation which is similar, but not the same as the problem considered here. In this paper, we use the centerline value FEL/(\(h/2\)) = 1.69 and this value is included in Figure 3.
3.4. Particle Entrance Lengths
The particle entrance length (PEL) and the cross-pore particle position (Y) at the PEL were determined by examining the results presented in Figure 3; results are presented in Figure 4. For both $\lambda = 0.1$ and $\lambda = 0.8$, the PEL is generally independent of $X_{\text{initial}}$ when $X_{\text{initial}} = -5, -7$ and $-9$ (Figure 4a); these particles also reach approximately the same cross-pore position in the pore (Figure 4b). These results arise because particles that are initially placed 5 or more pore $\frac{1}{2}$ widths from the membrane surface follow approximately the same trajectories into the pore.
Particles initially placed closer to the membrane surface ($X_{\text{initial}} = -1$ and $-3$) have smaller PEL and are carried closer to the pore wall when compared to those initially placed farther from the membrane. Particles that are carried closer to the pore wall achieve a smaller steady velocity (Figure 3). With a smaller change in particle velocity between $X = 0$ and $X = \text{PEL}$, these particles achieve a steady velocity in less time (and therefore in a shorter distance) when compared to particles initially placed farther from the membrane face.
The results in Figure 4a show that PEL for the smaller particles is greater than the PEL for the larger particles. This can again be explained by considering the smaller change in velocity experienced by the larger particle when it travels from the pore entrance to the PEL. Particles with $\lambda = 0.1$ and placed initially at $X_{\text{initial}} = -5$ undergo a 35% increase in velocity from $X = 0$ to $X = \text{PEL}$ whereas the velocity of the larger particles increases by only 12% from $X = 0$ to $X = \text{PEL}$.
The difference between the PEL and FEL for $\lambda = 0.1$ particles that are placed at $X_{\text{initial}} = -1$ can be explained by recognizing that these particles achieve their steady velocity at $Y = 0.45$ (Figure 4b). As shown in Figure S1 (Supplementary Information), this is the pore region where the FEL is considerably smaller than the centerline value ($\text{FEL}/(h/2) = 1.26$ at $Y = 0.4$). The difference between the FEL and the PEL is relatively large because the FEL value presented in Figure 4a is based on the centerline velocity. The relatively small value for the PEL for the $\lambda = 0.8$ particles that are placed at $X_{\text{initial}} = -1$ arises from the fact that the steady velocity reached by these particles is relatively small because they reach that steady velocity at $Y = 0.13$, closer to the pore wall than for the same-sized particles initially placed farther from the membrane surface. Again, a smaller steady velocity means less time and a shorter distance is required for the particle to reach the PEL.
The results presented in Figure 4 demonstrate that results for particles that approach the membrane from five or more $\frac{1}{2}$ pore widths from the surface are essentially the same. Therefore,
in subsequent simulations, particles were placed at $X_{initial} = -5$. This should be a reasonable representation of particle trajectories to a membrane pore from a reservoir.
The dependence of the PEL on $Y_{initial}$ was examined for particles with relative size $\lambda = 0.1, 0.3, 0.5$ and $0.8$ and $X_{initial} = -5$; results are shown in Figure 5a. The cross-sectional particle position ($Y$) at the PEL as a function of $Y_{initial}$ from these simulations is shown in Figure 5b. Particles that are initially placed at $Y_{initial} = 0$ maintain this cross-pore position as they are carried into and through the pore because particle diffusion is neglected and there is a symmetric flow field around the particles in the $Y$ direction. In general, the smaller particles ($\lambda = 0.1, 0.3$ and $0.5$) have similar PEL values, which are generally independent of $Y_{initial}$ for particles initially placed at $Y_{initial} = 0, 1$ and $2$. The PEL for these particles decreases and the cross-pore position at the PEL increases (Figure 5b) when particles are initially placed in the reservoir farther from the pore centerline ($Y_{initial} = 3, 4$ and $5$). These trends can again be explained by recognizing that these particles are carried into the pore at $Y$ positions close to the pore wall (larger $Y$ at the PEL) and therefore achieve smaller particle velocities compared to particles initially placed at smaller $Y_{initial}$. The relationship between PEL, $Y_{initial}$ and $\lambda$ appears to be complex when $Y_{initial} = 3, 4$ and $5$, particularly for the smaller particles ($\lambda = 0.1, 0.3$ and $0.5$). These results arise because these particles reach their PEL between $Y = 0.25$ and $0.47$. This is the area in the pore where the PEL shows a minimum (Figure S1 in Supplementary Information). Therefore, the confluence of different steric constraints and hydrodynamic interactions influencing the different-sized particles and the fluid response to the pore entrance (FEL($Y$)) leads to complex relationships between PEL and particle size and $Y_{initial}$. The PELs for the larger particles ($\lambda = 0.8$) are less sensitive to $Y_{initial}$ than the smaller particles. Since steric constraints limit available positions in the pore for these particles, $Y$ at the PEL is also less sensitive to $Y_{initial}$, leading to particle velocities, and therefore PELs, that vary little with $Y_{initial}$.
**Figure 5.** (a) PEL as a function of $Y_{initial}$ and (b) $Y$ at PEL as a function of $Y_{initial}$ for particles with $\lambda = 0.1, 0.3, 0.5$ and $0.8$. Particles were initially placed at $X_{initial} = -5$ and particle specific gravity $= 1.1$. The pore centerline is at $Y = 0$ (Figure 1). The radius of curvature of the pore entrance, $r_c/(h/2) = 1$.
A general conclusion that can be drawn from the results presented in Figures 4 and 5 is that as $Y$ at the PEL increases, the PEL decreases. This trend can be attributed to the smaller velocities achieved by particles when they are carried towards the pore wall.
Finally, the results in Figures 4b and 5b show that hydrodynamic forces leave a significant portion of the pore free of particles. For example, equilibrium steric restrictions allow particles with $\lambda = 0.1$ to sample cross-sectional positions from $Y = -0.95$ to $Y = 0.95$, but our results show that, without diffusion, these particles will be found within $-0.5 < Y < 0.5$. This ‘funneling’ phenomenon is addressed in further detail in the subsequent discussion of particle concentration profiles.
3.5. Particle Concentration Profiles
Simulations were repeated with 102 particles that were initially placed at \( X_{\text{initial}} = -5 \) and were uniformly distributed in the reservoir between \( Y_{\text{initial}} = -5 \) and \( Y_{\text{initial}} = 5 \). The trajectories of these particles were tracked as they reached a steady velocity in the pore channel. These simulations were repeated with collections of particles with different size (\( \lambda = 0.1, 0.3, 0.5 \) and 0.8). The number of particles in each downstream cross-sectional pore segment (at the PEL) were counted, yielding particle concentration profiles shown in Figure 6. These concentration profiles are compared to the uniform profile predicted for a Boltzmann distribution. These results show that hydrodynamic particle–pore entrance interactions ‘funnel’ the particles towards the pore center. The difference between the profiles predicted from our pore entrance model and the uniform Boltzmann profile is most pronounced for the smallest particles (\( \lambda = 0.1 \)). Equilibrium steric constraints restrict these particles to \(-0.95 < Y < 0.95\), but results show that the hydrodynamic forces place these particles only within the window \(-0.5 < Y < 0.5\). The particles are most concentrated at the outer edge of this window because all particles initially placed at larger \( Y \) in the reservoir are carried into these regions in the pore (as shown in Figure 5b). For larger particles, the difference between the two profiles is reduced, with only small differences for particles with \( \lambda = 0.8 \), because these particles are restricted by steric constraints to a narrow window near the pore centerline.
**Figure 6.** Relative particle concentration profile at the PEL predicted from our pore entrance model compared to a Boltzmann probability for different particle sizes: (a) \( \lambda = 0.10 \); (b) \( \lambda = 0.30 \); (c) \( \lambda = 0.50 \); (d) \( \lambda = 0.80 \). Simulations were repeated with 102 particles for each particle size, initially placed at \( X_{\text{initial}} = -5 \) and evenly distributed between \(-5 < Y_{\text{initial}} < +5\). The radius of curvature of the pore entrance, \( r_c/(h/2) = 1 \).
The results in Figure 6 show that, in the absence of particle diffusion, particles at the PEL are concentrated closer to the pore centerline than is predicted from a Boltzmann probability. For thick membranes and small particles, sufficient residence time in the pore is expected to enable particles to diffuse across the pore and a Boltzmann profile will develop within a relatively short distance, enabling one to use Equation (2) to predict rejections. However, that may not be the case for ultrathin membranes with thicknesses that are less than several hundred nm; asymmetric membranes have selective layers which are ~1 µm thick. For these systems, the results in Figure 6 and our calculations of particle transit and cross-pore diffusion times indicate that particle residence time in the pore is likely to be less than the time required for a Boltzmann concentration profile to develop. The ‘non-equilibrium’ concentration profiles shown in Figure 6 should be considered when developing particle rejection models for these systems.
3.6. Effect of the Radius of Curvature of the Pore Entrance
Simulations were performed for membranes with different radius of curvature at the pore mouth for particles with λ = 0.1 that were initially placed in the reservoir at \( Y_{\text{initial}} = 1 \) and \( X_{\text{initial}} = -5 \). The relationship between PEL and FEL and the radius of curvature are shown in Figure 7a. These results show a strong correlation between the PEL and the FEL and the radius of curvature, with a linear relationship for the PEL and a nearly linear relationship for the FEL. The estimation for FEL in Equation (3) does not include the radius of curvature of the channel entrance because, as noted earlier, the analysis leading to this expression considers the transition from a flat velocity profile to a steady parabolic profile within a channel; it does not consider flow into the channel.
**Figure 7.** Effect of the radius of curvature at the pore mouth on (a) PEL and fluid entrance length (FEL) and (b) \( Y \) at the PEL. Results are shown for a system with \( \lambda = 0.1 \), particle specific gravity = 1.1, \( Y_{\text{initial}} = 1 \) and \( X_{\text{initial}} = -5 \).
When $r_c/(h/2) < 3$, there is close agreement between the PEL and the FEL, with both increasing by a factor of ~4 when $r_c/(h/2)$ increases from 0.1 (essentially a square corner) to 3. For larger (and perhaps unrealistic) $r_c$, FEL > PEL. These results illustrate the importance of the geometry of the membrane face on effective entrance lengths. This geometry will be influenced by the process used to fabricate the membrane.
The cross-sectional position of the particle at the PEL for different $r_c$ is presented in Figure 7b. These results show a generally weak dependence on the final particle position and $r_c$. We expect a stronger dependence for larger particles where steric restrictions will be more important. In our discussion of the results in Figures 4 and 5, we noted the relationship between PEL and $Y$ at PEL, with the PEL decreasing as particles get carried closer to the pore wall. The results in Figure 7 indicate that the relationship between PEL and $Y$ at the PEL is perhaps more complicated. These results show a large increase in PEL when $r_c/(h/2)$ increases from 0.1 to 5 with only a small change in the particle position at the PEL.
3.7. Effect of $Re_h$
The results presented in Figures 3–7 were obtained for a specific reservoir velocity ($4.08 \times 10^{-4}$ m/s, $Re_h = 2 \times 10^{-5}$). Simulations were also performed with higher fluid velocities; PEL and FEL versus $Re_h$ are shown in Figure 8. The particle Stokes number ($\left( \frac{\rho_{particle}}{\rho_{fluid}} \right) \left( \frac{d_{particle}}{h} \right)$ $Re_{dparticle}$) is also included in this figure. These results show that, for $Re_h < 2$ (Stokes number < 0.03), the PEL and the FEL are both independent of $Re_h$, with close agreement between the PEL and the FEL. Both the PEL and the FEL increase for larger $Re_h$, with a much larger increase for the PEL compared to the FEL because of particle inertia at these conditions. Note that this increase in PEL occurs when the Stokes number is less than 1. While UF and MF separations are typically performed at low $Re_h$ conditions ($Re_h << 1$), where inertial effects should not be important, the fact that PEL/FEL increases from ~1 at low $Re_h$ to ~2 when $Re_h > 10$ is an interesting observation. The higher $Re_h$ results may also be important in gas filtration processes where viscous effects are generally not significant.
. Effect of $Re_h$ on PEL and FEL. Results are presented for a system with $\lambda = 0.1$, particle specific gravity = 1.1, $Y_{initial} = 1$, $X_{initial} = -5$ and $r_c/(h/2) = 1$.
**Figure 8.** Effect of $Re_h$ on PEL and FEL. Results are presented for a system with $\lambda = 0.1$, particle specific gravity = 1.1, $Y_{initial} = 1$, $X_{initial} = -5$ and $r_c/(h/2) = 1$.
3.8. Effect of Particle Specific Gravity
The results presented in Figures 3–8 were obtained for particles with specific gravity = 1.1, characteristic of viruses and other bioparticles or macromolecules. Results for these particles are compared to results from simulations performed with particles with specific gravity = 11 in Figure 9, with PEL versus Re\textsubscript{particle} shown in Figure 9a and \(Y\) at PEL versus Re\textsubscript{particle} shown in Figure 9b. At Re\textsubscript{particle} < 0.04, results for the two particles are essentially the same because transport at these conditions is governed by viscous forces and inertia is insignificant. These results illustrate that one can predict behavior for bioparticles (S.G. = 1.1) from measurements with metallic nanoparticles (S.G. = 11) or vice versa. The particle systems differ only when Re\textsubscript{particle} > 0.1, conditions that are generally not experienced in membrane filtration with liquids, but may be important in gas separations.
Similarly, the results in Figure 9b show no difference between the steady cross-pore positions achieved by the two different particles when Re\textsubscript{particle} is small. At higher Re\textsubscript{particle}, inertia is important, yielding differences in \(Y\) at the PEL for the two particles. An interesting observation can be made for the results for the denser particle at Re\textsubscript{particle} = 4. Here, the particle follows a trajectory that carries it to \(Y = -0.16\); inertia has carried the particle across the pore centerline.
**Figure 9.** (a) PEL as a function of Re\textsubscript{particle} for particles with specific gravity = 1.1 and 11; (b) \(Y\) at PEL as a function of Re\textsubscript{particle} for these same particles. Results are presented for a system with \(\lambda = 0.1\), \(Y\textsubscript{initial} = 1\), \(X\textsubscript{initial} = -5\) and \(r\textsubscript{c}/(h/2) = 1\).
3.9. Pore Geometry
Our model neglects the impact of neighboring pores on particle motion into a pore, an assumption that may be questionable for high porosity membranes. To model higher porosity membranes, initial particle placement in the reservoir would be limited to smaller \(Y\textsubscript{initial}\) values where PEL is larger and particles will be carried close to the pore centerline (Figures 3 and 5). Here, one would expect more pronounced particle ‘funneling’ than demonstrated with the current results. Therefore, the impact of the pore entrance is expected to be even more important for higher porosity membranes.
3.10. Particle Concentration/Feed Mixtures
The results presented here were obtained with a model in which particle–particle interactions were neglected, i.e., low feed concentrations are assumed. To simulate systems with higher particle concentrations requires that one add particle–particle interactions to the particle–membrane...
interactions that are included in the current model. This would add considerable complexity, and the concomitant computational time, to the simulations.
The results in Figure 4a show that the PEL increases as particle size decreases. Therefore, in order to put an upper bound on the PEL for a low concentration mixture containing particles of various sizes, one should consider the smallest particles in the mixture.
3.11. Implications for Rejection Models
As noted earlier, the general expression for membrane rejection (Equation (2)) is based on the assumptions that equilibrium is established at the pore entrance and that particle and fluid velocities are independent of axial position in the pore. The results reported here put into question the validity of these assumptions for ultrathin membranes.
The Boltzmann expression in Equation (2) involves potential particle–pore wall interaction energies. To correct the Boltzmann term for the non-steady particle and fluid velocities in the pore entrance region, particle kinetic energies can be added to the potential energy. We are currently developing a rejection model that incorporates axially dependent fluid and particle velocities and the non-uniform cross-pore concentration profiles that were determined in this study. Results will be forthcoming.
4. Conclusions
A 2D particle transport model was developed to examine spherical particle motion from a reservoir into a single-slit pore. Results show that particles accelerate as they travel into the pore and achieve a steady velocity a distance into the pore we define as the particle entrance length, PEL. The effects of relative particle size, pore entrance geometry and initial particle placement in the reservoir on the PEL were examined. The centerline fluid entrance length, FEL, was also determined for a particle-free system, with the FEL found to be 69% larger than the pore width when the radius of curvature of the pore mouth was equal to the pore width. This value is larger than values reported in previous studies of fluid entrance lengths because we have placed a tighter criterion (99.9%) on agreement between centerline fluid velocity and the velocity with fully developed flow. The different system considered in this study (flow from a reservoir into a channel) compared to previous studies may also play a role in this difference.
Results show that, in general, the PEL is comparable to the FEL, with the PEL decreasing when particles are carried to intrapore positions closer to the pore wall. The relationship between PEL, particle size and initial particle placement in reservoir is complex because of the confluence of steric constraints, hydrodynamic particle–membrane resistances and the response of the fluid as it enters the pore. The PEL was found to depend strongly on the radius of curvature of the pore entrance, with the PEL increasing by a factor of ~6 when the radius of curvature increased from 10 to 500% of the pore width.
In the absence of particle diffusion, hydrodynamic interactions between the particles and the pore entrance lead to ‘funneling’ of the particles towards the pore centerline, yielding particle cross-pore concentration profiles that can differ significantly from profiles predicted from a Boltzmann probability.
Ultrathin membranes can have thickness comparable to the pore size; therefore, pore entrance effects are expected to be particularly important for these systems. Results from this study indicate that particles may exit pores in these membranes before they achieve a steady velocity and before Boltzmann concentration profiles can develop by cross-pore diffusion. Therefore, conventional models may not accurately describe particle transport and rejections for these systems. We are currently developing an alternative approach to model such systems.
Supplementary Materials: The following are available online at http://www.mdpi.com/2077-0375/7/3/41/s1, Figure S1: Dimensionless FEL as a function of cross-pore position (Y).
Acknowledgments: The authors gratefully acknowledge the support from the National Science Foundation through grants CBET-0966934 and CBET-1604715.
Author Contributions: Armin Delavari performed all of the simulations. Ruth Baltus and Armin Delavari analyzed and interpreted the results and wrote the paper.
Conflicts of Interest: The authors declare no conflict of interest. The National Science Foundation had no role in the design of the study; in the generation, analyses, or interpretation of data; in the writing of the manuscript, and in the decision to publish the results.
References
1. Striemer, C.C.; Gaborski, T.R.; Mcgrath, J.L.; Fauchet, P.M. Charge-and size-based separation of macromolecules using ultrathin silicon membranes. *Nature* 2007, 445, 749–753. [CrossRef] [PubMed]
2. Akbari, A.; Sheath, P.; Martin, S.T.; Shinde, D.B.; Shaibani, M.; Banerjee, P.C.; Tkacz, R.; Bhattacharyya, D.; Majumder, M. Large-area graphene-based nanofiltration membranes by shear alignment of discotic nematic liquid crystals of graphene oxide. *Nat. Commun.* 2016. [CrossRef] [PubMed]
3. Ling, S.; Jin, K.; Kaplan, D.L.; Buehler, M.J. Ultrathin free-standing bombyx mori silk nanofibril membranes. *Nano Lett.* 2016, 16, 3795–3800. [CrossRef] [PubMed]
4. Lee, Y.M.; Jung, B.; Kim, Y.H.; Park, A.R.; Han, S.; Choe, W.S.; Yoo, P.J. Nanomesh-structured ultrathin membranes harnessing the unidirectional alignment of viruses on a graphene-oxide film. *Adv. Mater.* 2014, 26, 3899–3904. [CrossRef] [PubMed]
5. Zhu, Y.; Gao, S.; Hu, L.; Jin, J. Thermoresponsive Ultrathin Membranes with Precisely Tuned Nanopores for High-Flux Separation. *ACS Appl. Mater. Interfaces* 2016, 8, 13607–13614. [CrossRef] [PubMed]
6. Popa, A.-M.; Niedermann, P.; Heinzelmann, H.; Hubbell, J.A.; Fugin, R. Fabrication of nanopore arrays and ultrathin silicon nitride membranes by block-copolymer-assisted lithography. *Adv. Mater.* 2009, 20, 485303. [CrossRef] [PubMed]
7. Liu, G.; Jiang, Z.; Cheng, X.; Chen, C.; Yang, H.; Wu, H.; Pan, F.; Zhang, P.; Cao, X. Elevating the selectivity of layer-by-layer membranes by in situ bioinspired mineralization. *J. Membr. Sci.* 2016, 520, 364–373. [CrossRef] [PubMed]
8. Fang, D.Z.; Striemer, C.C.; Gaborski, T.R.; McGrath, J.L.; Fauchet, P.M. Methods for Controlling Pore Properties of Ultra-thin Nanocrystalline Silicon Membranes. *J. Phys. Condens. Matter* 2010, 22, 454134. [CrossRef] [PubMed]
9. Feinberg, B.J.; Hsiao, J.C.; Park, J.; Zydney, A.L.; Fissell, W.H.; Roy, S. Silicon nanoporous membranes as a rigorous platform for validation of biomolecular transport models. *J. Membr. Sci.* 2017, 536, 44–51. [CrossRef]
10. Han, Y.; Xu, Z.; Gao, C. Ultrathin Graphene Nanofiltration Membrane for Water Purification. *Adv. Funct. Mater.* 2013, 23, 3693–3700. [CrossRef]
11. Mireles, M.; Gaborski, T.R. Fabrication techniques enabling ultrathin nanostructured membranes for separations. *Electrophoresis* 2017. [CrossRef] [PubMed]
12. Wong, H.C.; Zhang, Y.; Viasnoff, V.; Low, H.Y. Predictive Design, Etch-Free Fabrication of Through-Hole Membrane with Ordered Pores and Hierarchical Layer Structure. *Adv. Mater. Technol.* 2017, 2, 1600169. [CrossRef] [PubMed]
13. Warkiani, M.E.; Bhagat, A.A.S.; Khoo, B.L.; Han, J.; Lim, C.T.; Gong, H.Q.; Fane, A.G. Isoporous micro/nanoengineered membranes. *ACS Nano* 2013, 7, 1882–1904. [CrossRef] [PubMed]
14. Anderson, J.L. Configurational effect on the reflection coefficient for rigid solutes in capillary pores. *J. Theor. Biol.* 1981, 90, 405–426. [CrossRef]
15. Deen, W.M. *Analysis of Transport Phenomena*; Oxford University Press: New York, NY, USA, 1998.
16. Atkinson, B.; Brocklebank, M.P.; Card, C.C.H.; Smith, J.M. Low Reynolds Number Developing Flows. *AIChE J.* 1969, 15, 548–553. [CrossRef]
17. Burgin, T.; Johnson, D.; Chung, H.; Clark, A.; McGrath, J. Analytical and finite element modeling of nanomembranes for miniaturized, continuous hemodialysis. *Membranes (Basel)* 2015, 6, 6. [CrossRef] [PubMed]
18. Bowen, W.R.; Sharif, A.O. Prediction of optimum membrane design: Pore entrance shape and surface potential. *Colloids Surf. A: Physicochem. Eng. Asp.* 2002, 201, 207–217. [CrossRef]
19. Kao, J.-N.; Wang, Y.; Pfeffer, R.; Weinbaum, S. A theoretical model for nucleopore filters including hydrodynamic and molecular wall interaction effects. *J. Colloid Interface Sci.* 1988, 121, 543–557. [CrossRef]
20. Agbangla, G.C.; Bacchin, P.; Climent, E. Collective dynamics of flowing colloids during pore clogging. *Soft Matter* 2014, 10, 6303–6315. [CrossRef] [PubMed]
21. Ramachandran, V.; Venkatesan, R.; Tryggvason, G.; Scott, F.H. Low Reynolds Number Interactions between Colloidal Particles near the Entrance to a Cylindrical Pore. *J. Colloid Interface Sci.* 2000, 229, 311–322. [CrossRef] [PubMed]
22. Kim, M.; Zydney, A.L. Effect of electrostatic, hydrodynamic, and Brownian forces on particle trajectories and sieving in normal flow filtration. *J. Colloid Interface Sci.* 2004, 269, 425–431. [CrossRef] [PubMed]
23. Bowen, R.W.; Filippov, A.N.; Sharif, A.O.; Starov, V.M. A model of the interaction between a charged particle and a pore in a charged membrane surface. *Adv. Colloid Interface Sci.* 1999. [CrossRef]
24. Ford, D.M.; Glandt, E.D. Steric hindrance at the entrances to small pores. *J. Memb. Sci.* 1995, 107, 47–57. [CrossRef]
25. Lim, Y.-I.; Bhatia, S.K. Simulation of methane permeability in carbon slit pores. *J. Memb. Sci.* 2011, 369, 319–328. [CrossRef]
26. Peskin, C.S. The immersed boundary method. *Acta Numer.* 2002, 11, 479–517. [CrossRef]
27. Agasanapura, B.; Baltus, R.E.; Tanneru, C.; Chellam, S. Membrane rejection of nonspherical particles: Modeling and experiment. *AIChE J.* 2013, 59, 3863–3873. [CrossRef] | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3014,
1
],
[
3014,
7637,
2
],
[
7637,
9713,
3
],
[
9713,
13637,
4
],
[
13637,
16460,
5
],
[
16460,
19589,
6
],
[
19589,
22949,
7
],
[
22949,
26542,
8
],
[
26542,
28796,
9
],
[
28796,
31034,
10
],
[
31034,
33839,
11
],
[
33839,
36760,
12
],
[
36760,
40926,
13
],
[
40926,
45323,
14
],
[
45323,
46506,
15
]
]
} |
26f7fe4f8dbcd71ec0588471fc4d446d62be863c | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 17,
"total-fallback-pages": 0,
"total-input-tokens": 57979,
"total-output-tokens": 21710,
"fieldofstudy": [
"Environmental Science"
]
} | Factors influencing productivity of eastern wild turkeys in northeastern South Dakota
Reina M. Tyl¹,³ | Christopher T. Rota¹ | Chadwick P. Lehman²
¹Division of Forestry and Natural Resources, West Virginia University, Morgantown, WV, USA
²South Dakota Game, Fish and Parks, Custer, SD, USA
³Missouri Department of Conservation, Central Regional Office and Conservation Research Center, Columbia, MO, USA
Correspondence
Reina M. Tyl, Missouri Department of Conservation, Central Regional Office and Conservation Research Center, 3500 East Gans Road, Columbia, MO 65201, USA. Email: [email protected]
Present address
Missouri Department of Conservation, Central Regional Office and Conservation Research Center, Columbia, MO, USA
Funding information
South Dakota Department of Game, Fish and Parks; National Wild Turkey Federation; West Virginia University; USDA National Institute of Food and Agriculture, McIntire Stennis, Grant/Award Number: WVA00124
Abstract
Population growth is highly sensitive to changes in reproductive rates for many avian species. Understanding how reproductive rates are related to environmental conditions can give managers insight into factors contributing to population change. Harvest trends of eastern wild turkey in northeastern South Dakota suggest a decline in abundance. We investigated factors influencing reproductive success of this important game bird to identify potential factors contributing to the decline. We monitored nesting rate, nest survival, renesting rate, clutch size, hatchability, and poult survival of 116 eastern wild turkey hens using VHF radio transmitters during the springs and summers of 2017 and 2018. Heavier hens were more likely to attempt to nest than lighter hens, and adult hens were more likely to renest than yearling hens. Nest survival probability was lowest in agricultural fields relative to all other cover types and positively related to horizontal visual obstruction and distance to the nearest road. Daily nest survival probability demonstrated an interaction between temperature and precipitation, such that nest survival probability was lower on warm, wet days, but lowest on dry days. Egg predation was the leading cause of nest failure, followed by haying of the nest bowl and death of the incubating hen. Poults reared by adult hens had a greater probability of survival than poults reared by yearling hens. Our estimate of survival probability of poults raised by yearling hens was low relative to other studies, which may be contributing to the apparent regional population decline. However, there is little managers can do to influence poult survival in yearling hens. Alternatively, we found nest survival probability was lowest for nests initiated in agricultural fields. Wildlife-friendly harvesting practices such as delayed haying or installation of flushing bars could help increase productivity of eastern wild turkey in northeastern South Dakota.
KEYWORDS
clutch size, hatchability, nest survival, nesting rate, poult survival, renesting rate
1 | INTRODUCTION
Population dynamics in closed systems are governed by survival and reproduction (Caswell, 2001). For many avian species, population growth is highly sensitive to changes in reproductive rates (Sæther & Bakke, 2000). Many factors contribute to variation in reproductive rates, and knowledge of how reproductive rates are related to environmental variation can give insight into drivers of population dynamics through time. This can guide management activities for species of conservation concern by identifying environmental variables that are associated with reproductive rates and how incremental changes in these variables are likely to influence reproduction and ultimately population growth (Mills, 2007).
Eastern wild turkey (Meleagris gallopavo silvestris; hereafter turkey) are an important game species across North America. This species tends to be relatively short-lived with high reproductive output (McRoberts, Wallace, & Eaton, 2014), making population growth rates highly sensitive to reproductive rates (Sæther & Bakke, 2000). Studies in New York (Roberts, Coffey, & Porter, 1995) and Wisconsin (Pollentier, Hull, & Lutz, 2014; Rolley, Kubisiak, Paisley, & Wright, 1998) have demonstrated population growth can be highly sensitive to changes in reproductive rates. Understanding factors influencing reproduction in turkey is therefore necessary for effective management of this important game species.
Both intrinsic (e.g., body condition, age) and extrinsic environmental variables influence reproductive rates for turkey. Hen body condition is an important determinant of reproductive success. For example, hen weight can be positively associated with nesting probability and nest success (Porter, Nelson, & Mattson, 1983; Vander Haegen, Dodge, & Sayre, 1988). Similarly, adult hens are more likely to nest and renest than yearling hens (Lehman, Flake, Leif, & Shields, 2001; Paisley, Wright, Kubisiak, & Rolley, 1998; Pollentier, Lutz, & Hull, 2014; Porter et al., 1983; Shields & Flake, 2006; Vander Haegen et al., 1988). Adult hens may also enhance survival probability of their poults relative to yearling hens (Porter et al., 1983). Extrinsic environmental variables can also influence turkey reproductive rates. For example, precipitation can be negatively associated with daily nest survival and survival of poults <2 weeks old (Healy, 1992; Healy & Nenno, 1985; Lehman, Flake, Rumble, & Thompson, 2008; Roberts & Porter, 1998a; Vangilder & Kurzejeski, 1995). In northern populations, cold weather during the brood-rearing season can be detrimental to poult survival and overall reproductive success (Healy, 1992; Healy & Nenno, 1985).
Although intrinsic and extrinsic environmental variables can be important determinants of turkey reproductive success, managers have little or no ability to influence these variables. In contrast, managers often have some ability to manipulate habitat-related environmental variables. One habitat-related environmental variable that can be important to turkey reproductive success is cover type. For example, Clawson and Rotella (1998) found that artificial nests located in Conservation Reserve Program (CRP) fields had greater success relative to nests located in non-CRP cover types. In contrast, hens nesting in agricultural cover types such as alfalfa fields may be subject to increase risk of nest failure and hen mortality (Paisley et al., 1998; Shields & Flake, 2006; Vangilder & Kurzejeski, 1995). South Dakota has seen large-scale landscape changes as grasslands have been converted into row crop or other agricultural cover types, with the greatest losses having occurred in the northeastern region (16.9% between 2006 and 2012) (Reitsma et al., 2014). This loss of grassland cover types parallels declines in the amount of land enrolled in CRP (Hellerstein, 2017), which has been particularly steep in northeastern South Dakota (USDA, 2016). Microhabitat conditions can also be strongly associated with reproductive rates. For example, increased visual obstruction can be positively associated with nest success (Badyaev, 1995; Lutz & Crawford, 1987), likely because of reduced detectability by predators.
For this study, we evaluated factors influencing reproductive success of turkeys in northeastern South Dakota. Harvest of turkeys in northeastern South Dakota during the spring prairie turkey season declined more than 50% between 2010 and 2016 (Huxoll, 2016), prompting managers to study potential causes of this apparent decline. Since productivity has a strong influence on population growth of this species (Pollentier, Hull, et al., 2014; Roberts et al., 1995; Rolley et al., 1998), understanding the factors that influence reproduction in this population is necessary to identify and potentially reverse the causes of this apparent decline. The objectives of this study are to (a) obtain baseline estimates of nesting rate, nest survival, renesting rate, clutch size, and hatchability; (b) obtain estimates of poult survival over the 28-day posthatch interval; and (c) determine the effects of intrinsic and environmental variables on nest and poult survival for turkey hens in northeastern South Dakota. The results of this study will improve understanding of the factors influencing turkey productivity in open, agriculturally dominated landscapes and inform management of turkeys in northeastern South Dakota.
2 | MATERIALS AND METHODS
2.1 | Study area
The study was conducted in Codington, Deuel, Grant, and Roberts counties in northeastern South Dakota. The study area was split between the Minnesota River-Red River Lowland in the eastern half of the study area and Coteau des Prairies physiographic region in the western half (Flint, 1955; Johnson, Higgins, & Hubbard, 1995). The Coteau begins in the northwest and extends in a southeasterly direction across the study area (Miller, Kempf, & Koopman, 1979). On top of the Coteau, the relief is gently undulating to hilly, while down in the Lowlands the land is nearly level (Flint, 1955; Miller et al., 1979). Elevations ranged from over 600 m above mean sea level on top of the Coteau to about 300 m above sea level in the Lowland (Miller et al., 1979). Most of the study area consisted of privately owned lands with some state-owned (e.g., Game Production Areas) and federally owned (e.g., Waterfowl Production Areas) lands scattered throughout.
Agriculture dominates land use in the study area, with most land being used for either cropland, rangeland, or to grow alfalfa or hay for livestock feed (Miller et al., 1979) (Figure 1). Most of the grain farming (i.e., corn and soybeans) occurs in the Lowland (Miller et al., 1979). The highlands of the Coteau support native tallgrass prairie which is used primarily for rangeland; however, scattered fields of hay and alfalfa are in the highlands as well (Miller et al., 1979). Common grasses include warm-season grasses such as big bluestem (Andropogon gerardii), little bluestem (Schizachyrium scoparium), Indiangrass (Sorghastrum nutans), switchgrass (Panicum virgatum), and sideoats grama (Bouteloua curtipendula) (Johnson & Larson, 2007). Common cool-season grasses include smooth brome (Bromus inermis), Kentucky bluegrass (Poa pratensis), western wheatgrass (Pascopyrum smithii), and green needlegrass (Stipa viridula) (Johnson & Larson, 2007). Numerous forbs and patches of western snowberry (Symphoricarpos occidentalis) are scattered throughout the pasture lands (Johnson & Larson, 2007). Forested areas along the east-facing breaks where the Coteau descends into the Lowlands are dominated by bur oak (Quercus macrocarpa) on the drier slopes (Leatherberry, Piva, & Josten, 2000). More mesic areas are dominated by elm-ash (Fraxinus spp.; Ulmus spp.) forests (Leatherberry et al., 2000) that are intermixed with trembling aspen (Populus tremuloides), box elder (Acer negundo), eastern cottonwood (Populus deltoides), and sugar maple (Acer saccharum) (Knupp Moore & Flake, 1994).
Northeastern South Dakota is in a humid continental climate region, with mean annual precipitation of 57 cm and mean annual temperature of 6.5°C across the study area (Menne et al., 2012). Early spring snowfall is possible, with about one-quarter (26%) of the total annual snowfall occurring from March through May (Menne et al., 2012). About 60% of the total annual precipitation occurs during the nesting and brood-rearing seasons (April through August; Menne et al., 2012). Northeastern South Dakota received below average precipitation (i.e., rainfall) and approximately average temperatures during spring seasons (1 April–30
**FIGURE 1** Map of land cover types (adapted from the National Land Cover Database 2016 land cover raster layer; Yang et al. 2018) in Codington, Deuel, Grant, and Roberts Counties in northeastern South Dakota, USA
June) over the course of this study (Figure 2). Spring and summer temperatures can be highly variable, with average minimum temperatures near 0°C in early spring to average maximum temperatures near 28°C during summer; however, normal mean temperatures for the spring and summer months range from 13 to 19°C (Menne et al., 2012).
2.2 | Capture and radio telemetry
We monitored reproduction of turkeys by fitting female turkeys with radio transmitters. We captured turkeys by first locating flocks of turkeys during the winter (1 January–31 March) and then baiting turkeys into capture sites. We captured turkeys using rocket nets (Thompson & Delong, 1967; Wunz, 1984). Following capture, we aged female turkeys as adult or yearling based on the presence or absence of barring on the 9th and 10th primary feathers (Williams, 1961) and weighed each bird. We secured 80-g very high-frequency (VHF) radio telemetry transmitters (Advanced Telemetry Systems) using a shock cord harness and backpack mount. Transmitters were <3% of the hens’ body weight to reduce the risk of the transmitter interfering with survival and reproduction (Fair, Paul, & Jones, 2010). Transmitters were equipped with an activity signal, a nonmoving (loafing) signal that is activated instantaneously whenever the hen is not in motion, and mortality signal set to activate after 8 hr of inactivity. We monitored turkeys 6 days per week during the spring and summer (1 April–31 July) by locating each transmitter signal and listening to the nature of the signal; however, turkeys that were incubating nests were monitored daily. A moving signal indicated the hen was alive but not incubating a nest, a nonmoving signal indicated the hen was alive and incubating a nest, and a mortality signal indicated a hen was no longer alive. All handling, marking, and monitoring procedures were approved by the West Virginia University Institutional Animal Care and Use Committee (Permit No. 1606003205; South Dakota State Permit 37).
2.3 | Nest marking and monitoring
We monitored nesting activity of hens daily from 1 April to 6 August, 2017–2018. We first determined onset of incubation by listening for nonmoving signals from VHF transmitters. Once a nonmoving signal was obtained, we located nesting hens via homing and marked the nest. We marked the nest area with ~4 flags at distances of 20–40 m from the nest bowl depending on cover height and density of vegetation while attempting to minimize disturbance. If a nonmoving signal was observed on a subsequent day, we assumed that the hen was still tending the nest. If a moving signal was observed, we visually inspected the nest bowl to determine whether the hen was temporarily away (i.e., eggs and nest bowl still active and not disturbed), or whether it was lost due to predation (i.e., smashed or removed eggs). If a mortality signal was observed, we assumed that the hen died while tending the nest and we located the transmitter and assessed the cause of death. We classified nests as successful by the presence of hatched eggshells, or as failed if nest contents were depredated, destroyed, or abandoned (Lehman et al., 2001). If the nest was successful, we determined the number of eggs that hatched from the total clutch size by counting eggshell fragments and membranes (Lehman et al., 2001). We counted the number of eggs in failed nests to determine clutch size if the eggs were relatively intact and undisturbed (Lehman et al., 2001). If the clutch size of failed nests could not be accurately determined, we did not include that nest in the analysis of clutch size.
2.4 | Poult monitoring
We determined the initial number of poults that hatched from each successful nest based on egg shell and membrane remains (Lehman et al., 2001). The number of poults in each brood was counted at 1, 2, and 4 weeks posthatch by observing broods feeding in open areas (Lehman, Flake, et al., 2008); however, if dense vegetation interfered with observations, broods were flushed to count poults. Broods
**FIGURE 2** Total precipitation accumulation (i.e., rainfall) (cm) and mean air temperature (°C) during the springs (1 April to 30 June) of 2017 and 2018 in northeastern South Dakota, USA. The 30-year average (1989–2018) for total precipitation accumulation (24.0-cm) and mean air temperature (13.1°C) during spring in Milbank, South Dakota, USA, are indicated by the horizontal dashed lines (Menne et al., 2012)
often formed crèches (multiple hens with a group of commingled poults) after poults were 2 weeks old, and crèches were common when broods were 4 weeks old, making it difficult to differentiate individual broods during the day. If we could not determine the number of poults in a brood during the day due to the formation of a crèche, we performed another poult count for that brood again at night. During night brood counts, we observed the brood while in the roost with the hen, being careful to not flush the group from the roost, to obtain an accurate count of poults (Lehman, Flake, et al., 2008).
2.5 Environmental and spatial covariate estimation
We sought to determine how nest-site characteristics influenced nest success. Therefore, we quantified nest-site characteristics on the hatch date for successful nests and on the projected hatch date for failed nests (Gibson, Blomberg, & Sedinger, 2016; McConnell, Monroe, Burger, & Martin, 2017; Smith et al., 2018). We measured understory visual obstruction readings (VOR) of vegetation by placing a Robel pole with 2.54 cm increments in the nest bowl and at 1 m from the nest in the four cardinal directions (Benkobi, Uresk, Schenbeck, & King, 2000; Robel, Briggs, Dayton, & Hulbert, 1970). We recorded the lowest visible increment on the pole from a distance of 4 m while kneeling to a height of 1 m (Lehman, Rumble, Flake, & Thompson, 2008; Robel et al., 1970). We measured VOR from the four cardinal directions at the nest bowl; however, at the peripheral 1 m from the nest measurements, we estimated VOR from only three cardinal directions, ignoring the 4th direction back across the nest bowl so as not to duplicate visual obstruction readings across the nest bowl (Lehman, Rumble, et al., 2008). We measured the height (in centimeters) of living vegetation at the nest bowl and at 1 m from the nest in the 4 cardinal directions (Lehman, Rumble, et al., 2008). We estimated total cover of grass, forbs, shrubs, and other cover using a 0.1-m² quadrat at the nest bowl, and at 5, 1-m intervals in the cardinal directions (Daubenmire, 1959). We qualitatively categorized the dominant land cover type within the area surrounding the nest bowl as either grassland, pasture, agriculture, or forest. We classified the land cover as pasture if grazing was currently occurring or had occurred that year. Alfalfa hayfields and row crop fields were classified as agriculture. CRP grasslands, old fields, and other land cover where the dominant vegetation was grass and forbs and where grazing did not occur were classified as grasslands. If a nest was located within a road ditch, we classified the land cover according to the dominant land use adjacent to the road ditch (e.g., a nest placed in a road ditch next to a corn field would be classified as agriculture). We used ArcMap version 10.6.1 (Environmental Systems Research Institute) to calculate the distance from each nest to the nearest road (i.e., interstate, federal highway, state highway, local paved road, local unpaved road), obtained from the South Dakota Department of Transportation (SDDOT, 2017).
We placed 10 precipitation and temperature monitoring stations throughout the study area before the onset of nesting and retrieved the monitoring stations after all broods were >4 weeks old. Monitoring stations consisted of a rain gauge and a HOBO Pendant Temperature Data Logger (Onset Computer Corporation) that recorded 4 or 6 temperature readings at evenly spaced intervals each day. Monitoring stations were placed systematically throughout the study area to cover the extent of all radio-marked hen locations. Rain gauges were checked after every precipitation event, and we calculated daily precipitation accumulation (mm) (hereafter precipitation) and daily mean temperature (°C) (hereafter temperature) for each monitoring station for each day of the study. We assigned precipitation and temperature covariates to each individual nest and each individual brood (for nest and poult survival analyses, respectively) by assuming precipitation and temperature conditions at the location of each nest and at the location of each brood was equal to the precipitation and temperature conditions observed at the closest monitoring station.
2.6 Modeling reproductive parameters
2.6.1 Nesting rate
We modeled nesting rate as the probability an individual hen that was alive on 1 April would attempt to nest that year using Bayesian logistic regression. We modeled nesting rate as a function of the age-class of each hen (adult or yearling), year of the study (2017 or 2018), and weight of each hen (kg). We used informative prior distributions for the intercept (log odds a juvenile initiates a nest when all other coefficients fixed at 0) and the slope coefficient describing the difference in log odds of nesting (i.e., log odds ratio [LOR]) between adult and juveniles. Drawing upon the studies in Table 2, we used a Gaussian (mean = 0.9; SD = 0.2) prior distribution for the intercept coefficient and a Gaussian (mean = 1.6; SD = 0.8) prior distribution for the adult LOR coefficient. Details on how we derived informative prior distributions are in Appendix 2. We assumed logistic (location = 0; scale = 1) prior distributions for all other slope coefficients.
2.6.2 Nest survival
We assumed survival of nest i during day t was a Bernoulli random variable:
\[ y_{it} \sim \text{Bernoulli}(y_{it-1}, p_i) \]
where \( y_{it} = 1 \) if nest i survived day t, \( y_{it} = 0 \) if nest i failed during day t, and \( p_i \) represents daily survival probability (Royle & Dorazio, 2008). We further assumed a logit-linear model for daily survival probability which we model as a function of age-class of the nesting hen (adult or yearling), precipitation, temperature, land cover type (agriculture, forest, grassland, or pasture), mean VOR, mean total cover, and distance to the nearest road (m). We included an interactive effect of precipitation on temperature, since the effect of precipitation may vary
depending on the temperature. Fifteen failed nests were missing VOR and total cover observations because the vegetation surrounding the nest was removed via haying before it could be measured. Rather than discard those nests, we imputed missing predictor variables, accounting for uncertainty in unmeasured variables (Gelman et al., 2013).
We accounted for repeated observations on individual nests by fitting a random coefficients model (Gelman & Hill, 2007). We assumed each coefficient $\beta_{ji}$ was a Gaussian random variable:
$$\beta_{ji} \sim \text{Gaussian}(\mu_j, \tau_j)$$
where $\mu_j$ and $\tau_j$ represent the population-level mean and precision, respectively, of slope coefficient $j$. We used informative prior distributions for the intercept and adult LOR population-level mean parameters. Drawing upon the studies in Table 2, we used a Gaussian (mean = 3.2; $SD = 0.3$) prior distribution for the intercept population-level mean parameter and a Gaussian (mean = 0.3; $SD = 0.4$) prior distribution for the adult LOR population-level mean parameter. Details on how we derived informative prior distributions are in Appendix 2. We selected logistic (location = 0; scale = 1) prior distributions for all other population-level mean parameters and gamma (shape = 1; rate = 1) prior distributions for precision parameters $\tau_p$ which provided little prior information.
### 2.6.3 Renesting rate
We modeled renesting rate as the probability an individual hen would attempt a second nest, conditional upon failure of the first nest, using Bayesian logistic regression. We considered a hen unavailable for renesting if she was killed while tending the first nest. We modeled renesting rate as a function of the age-class of each hen (adult or yearling), year of the study (2017 or 2018), ordinal date of failure for the previous nest attempt, length of the incubation period, and weight of the nesting hen (kg). We used informative prior distributions for the intercept and adult LOR coefficients. Drawing upon the studies in Table 2, we used a Gaussian (mean = 1.3; $SD = 0.9$) prior distribution for the intercept coefficient and a Gaussian (mean = −0.2; $SD = 1.1$) prior distribution for the adult LOR coefficient. Details on how we derived informative prior distributions are in Appendix 2. We selected logistic (location = 0; scale = 1) prior distributions for all other slope coefficients.
### 2.6.4 Clutch size
We modeled mean clutch size using Bayesian Poisson regression based on the number of eggs laid in each individual nest. Nests were excluded from the analysis if an accurate count of eggs could not be obtained (i.e., the nest was depredated and only some egg fragments remained). We modeled clutch size as a function of age-class of the nesting hen (adult or yearling), year of the study (2017 or 2018), weight of the nesting hen (kg), and nest attempt (first or second). We used informative prior distributions for the intercept (log expected count when all other slope coefficients equal 0) and the coefficient describing the difference in log expected count between adults and juveniles. Drawing upon the studies in Table 2, we used a Gaussian (mean = 2.4; $SD = 0.4$) prior distribution for the intercept coefficient and a Gaussian (mean = 0.0; $SD = 0.6$) prior distribution for the adult coefficient. Details on how we derived informative prior distributions are in Appendix 2. We selected Gaussian (mean = 0; $SD = 1$) prior distributions for all other slope coefficients.
### 2.6.5 Hatchability
We modeled hatchability as the proportion of eggs that hatched from each individual nest based on the total number of eggs laid in each individual nest using Bayesian logistic regression. We modeled hatchability as a function of age-class of the nesting hen (adult or yearling), year of the study (2017 or 2018), and weight of the nesting hen (kg). We used informative prior distributions for the intercept and adult LOR coefficients. Drawing upon the studies in Table 2, we used a Gaussian (mean = 1.3; $SD = 0.9$) prior distribution for the intercept coefficient and a Gaussian (mean = −0.2; $SD = 1.1$) prior distribution for the adult LOR coefficient. Details on how we derived informative prior distributions are in Appendix 2. We selected logistic (location = 0; scale = 1) prior distributions for all other slope coefficients.
### 2.6.6 Poults survival
We assumed the number of poults alive in each brood $i$ at each day posthatch $t$ was a Binomial random variable:
$$N_{it} \sim \text{Binomial}(\phi_{it}, N_{it-1})$$
where $N_{it}$ was equal to the initial number of poults in each brood $i$ and $\phi_{it}$ represents daily survival probability of each poult in brood $i$ between time $t = 1$ and time $t$. If a brood-rearing hen died during the 28-day posthatch interval, we assumed $N_{it} = 0$ for all subsequent poult counts. We treated the number of poults counted during each of 3 poult monitoring events at 7, 14, and 28 days posthatch ($t = 8, 15, 29$) as fixed and known, and treated $N_{it}$ as a latent random variable during all other time steps.
We further assumed a logit-linear model for daily survival probability which we model as a function of brood age (1–28 days posthatch), age of the brood-rearing hen (adult or yearling), year of the study (2017 or 2018), precipitation, and temperature. We additionally modeled the interaction between precipitation and temperature. We used an informative prior distribution for the intercept coefficient. Note that we did not include an informative prior distribution for the difference in log odds of poult survival between those raised by adults and juveniles because previous studies (Table 2) did not make this distinction. Drawing upon the studies in Table 2,
we used a Gaussian (mean = 3.4; SD = 0.1) prior distribution for the intercept coefficient. Details on how we derived informative prior distributions are in Appendix 2. We selected logistic (location = 0; scale = 1) prior distributions for all other slope coefficients.
We fit all models using Bayesian methods to maintain a consistent analytical approach. We fit each model with JAGS version 4.3.0 (Plummer, 2003) via the jagsUI version 1.4.9 interface (Kellner, 2018) in program R version 3.5.1 (R Core Team, 2018). We ran three chains for each model using trace plots to determine an adequate burn-in period and subsequently ran models until we achieved reasonable convergence (R < 1.1; Gelman et al., 2013). We concluded that slope coefficients were different from 0 if 95% credible intervals (CI) did not overlap 0.
3 | RESULTS
We captured 42 adult and 34 yearling turkey hens during the winter of 2017, and we captured an additional 40 yearling turkey hens during the winter of 2018. Sixteen yearling hens captured during the first year of the study transitioned to the adult age-class for the second year of the study; twenty-three adult hens captured during the first year of the study remained in the adult age-class for the second year of the study. Ultimately, we estimated factors influencing reproductive rates across the 2-year study from 116 individual turkey hens.
3.1 | Nesting rate
Nesting rate probabilities were estimated from a total of 155 nesting opportunities during 2017 and 2018 (76 hens were available to nest in 2017 and 79 hens were available to nest in 2018; note that 39 hens were available to nest during both years). Our estimate of nesting rate is likely biased low due to our inability to detect nests that were lost during the laying period. Although adult hens had a slightly greater estimated nesting rate (0.82, 95% CI = [0.72, 0.89]) than yearling hens (0.71, 95% CI = [0.65, 0.77]), 95% credible intervals of the age slope coefficient overlapped 0, indicating no strong effect of age. Hen body weight at the time of capture had a positive effect on nesting rate (Figure 3, Table A1). Nesting rate did not differ between years of the study (Table A2).
3.2 | Nest survival
We observed a total of 147 nest attempts during this study. In 2017, 48 hens attempted one nest, seven hens attempted two nests, and one hen attempted three nests (65 nests total). In 2018, 42 hens attempted one nest, 17 hens attempted two nests, and two hens attempted three nests (82 nests total). We recorded at least one nest attempt in both years from 28 hens. Across both years of the study, five nests were censored due to investigator interference causing the hen to abandon the nest. Five additional nests were censored because we were unable to visit the nest site due to a lack of landowner permissions. Therefore, nest survival probabilities were estimated from a total of 137 nests, across 2,412 days where an individual nest was at risk of failure.
Fifty-six of 137 nest attempts were successful. Predation of eggs was the leading cause of nest failure, accounting for over half of all failed nest attempts (Table 1). Haying of vegetation surrounding the nest and death of the incubating hen were also major sources of nest failure, accounting for 16% and 12%, respectively, of all failed nest attempts (Table 1). Over a 28-day average incubation period, estimated survival probability of nests laid by adult hens (0.44, 95% CI = [0.25, 0.62]) was greater than survival probability of nests laid by yearling hens (0.40, 95% CI = [0.20, 0.55]).
We found a strong effect of cover type on nest success probability. Nests located in areas classified as agriculture had a much...
lower success probability relative to nest located in any other cover type (Figure 4, Table A2). All nests that were laid in alfalfa fields failed because fields were hayed \( (n = 10) \) or depredated \( (n = 2) \) before eggs hatched. We also found nests with greater visual obstruction (Figure 5, Table A2) and that were placed further from roads (Figure 6, Table A2) had higher daily nest survival probability. Finally, precipitation and temperature had an interactive effect on daily nest survival probability. On relatively cool days, we found a positive effect of precipitation on daily nest survival probability. However, as temperature increased, the effect of precipitation on daily survival diminished. Thus, for a fixed amount of precipitation, predicted daily nest survival tended to be lower on warmer days (Figure 7, Table A2). Daily nest survival probability was not strongly affected by mean total cover (i.e., 95% CI of the population-level mean overlapped 0; Table A2).
### 3.3 Renesting rate
Renesting rate probabilities were estimated from a total of 58 hens that were available to renest after a failed first nest attempt (25 in 2017 and 33 in 2018). Adult hens were more likely to renest \( (0.59, 95\% \text{ CI} = [0.41, 0.76]) \) than yearling hens \( (0.26, 95\% \text{ CI} = [0.14, 0.40]) \). Probability of renesting was lower when the date of nest failure for the previous nest attempt was later in the season (Figure 8, Table A3). Renesting rate did not differ between years of the study and was not affected by hen weight at the time of capture or duration of previous nesting attempt (Table A3).
### 3.4 Clutch size
Clutch size was estimated from a sample size of 105 nests (48 nests in 2017 and 57 nests in 2018). We were unable to determine
**Figure 4** Daily nest survival probability across cover types of adult eastern wild turkey (Meleagris gallopavo silvestris) hens nesting in northeastern South Dakota, USA, in 2017 and 2018
**Figure 5** Daily nest survival probability as a function of visual obstruction of adult eastern wild turkey (Meleagris gallopavo silvestris) hens nesting in northeastern South Dakota, USA, in 2017 and 2018
**Figure 6** Daily nest survival probability as a function of distance to road of adult eastern wild turkey (Meleagris gallopavo silvestris) hens nesting in northeastern South Dakota, USA, in 2017 and 2018
accurate clutch counts for 40 out of 91 failed nest attempts and two out of 56 successful nest attempts over the course of this study, and therefore, omitted these nests from the clutch size analysis. During the first nest attempt, the mean clutch size laid by adult hens was 10.6 (95% CI = [9.8, 11.5]), the mean clutch size laid by yearling hens was 10.0 (95% CI = [8.9, 11.2]), and clutch size did not vary by age-class. Clutch size did not differ between nesting attempts or years of the study and was not affected by hen body weight at the time of capture (Table A4).
3.5 | Hatchability
Hatchability was estimated from a sample size of 54 successful nests (25 nests in 2017 and 29 nests in 2018; seven hens successfully hatched a clutch during both years of the study). Hatchability of clutches laid by adult hens (0.88, 95% CI = [0.85, 0.92]) was not different from clutches laid by yearling hens (0.87, 95% CI = [0.82, 0.92]). Hatchability did not differ between years of the study and was not affected by hen body weight (kg) at the time of capture (Table A5).
## Table 2: Comparison of estimates of eastern wild turkey (*Meleagris gallopavo silvestris*) reproductive rates obtained from northeastern South Dakota in 2017 and 2018 with previously published estimates
| Parameter | This Study | Published | Reference |
|---------------------|------------|-----------|------------------------------------------------|
| | Adult | Yearling | Combined | |
| Nesting rate | 0.82 | 0.71 | 0.90 (0.03) 0.34 (0.09) | Pollentier, Lutz, et al., 2014 |
| | 0.94 (0.03) 0.91 (0.06) | Porter et al., 1983 |
| | 0.98 (0.01) 0.79 (0.06) | Paisley et al., 1998 |
| | 0.96 | 0.88 | - | Shields & Flake, 2006 |
| Nest survival | 0.44 | 0.40 | 0.16 (0.03) 0.05 (0.03) | Pollentier, Lutz, et al., 2014 |
| | 0.23 (0.06) 0.12 (0.13) | Pollentier, Lutz, et al., 2014 |
| | 0.35 (0.09) 0.20 (0.04) | Pollentier, Lutz, et al., 2014 |
| | - | - | 0.31 | Vangilder & Kurzejeski, 1995 |
| | 0.64 | 0.61 | - | Porter et al., 1983 |
| | 0.68 | 0.33 | - | Vangilder et al., 1987 |
| Renesting rate | 0.59 | 0.26 | 0.51 (0.05) 0.47 (0.11) | Shields & Flake, 2006 |
| | 0.39 (0.08) 0.46 (0.08) | Vangilder & Kurzejeski, 1995 |
| | 0.41 (0.09) 0.00 (0.00) | Pollentier, Lutz, et al., 2014 |
| | - | - | 0.5 | Vangilder et al., 1987 |
| | - | - | 0.55 | Vangilder et al., 1987 |
| | 0.60 (0.05) 0.42 (0.08) | Pollentier, Lutz, et al., 2014 |
| | - | - | 0.65 | Porter et al., 1983 |
| Clutch size | 10.6 | 10.0 | 10.03 (0.24) | Vangilder et al., 1987 |
| | - | - | 10.38 (0.21) | Vangilder & Kurzejeski, 1995 |
| | 11.2 (0.36) 10.3 (0.48) | Pollentier, Lutz, et al., 2014 |
| | 11.7 | 12.8 | - | Vangilder et al., 1987 |
| | 12.8 (1.9) 11.1 (1.9) | Porter et al., 1983 |
| Hatchability | 0.88 | 0.87 | 0.76 (0.28) 0.83 (0.21) | Porter et al., 1983 |
| | 0.83 (0.07) 0.86 (0.06) | Vangilder et al., 1988 |
| | - | - | 0.87 (0.04) | Paisley et al., 1998 |
| Poult survival | 0.33 | 0.16 | 0.36 (0.05) | Shields & Flake, 2006 |
| | - | - | 0.37 (0.06) | Pollentier, Lutz, et al., 2014 |
| | - | - | 0.381 | Vangilder et al., 1987 |
| | - | - | 0.40 (0.15) | Hubbard et al., 1999 |
| | - | - | 0.41 | Roberts & Porter, 1998 |
| | - | - | 0.45 | Vangilder & Kurzejeski, 1995 |
| | - | - | 0.47 (0.12) | Paisley et al., 1998 |
Note: Published estimates are distinguished by age (adult or yearling) when possible or reported as “combined” if the authors did not distinguish estimates by age. Standard errors of published point estimates are presented in parentheses when reported by the authors.
Poult survival was estimated from a total of 55 broods (26 broods in 2017 and 29 broods in 2018; seven hens had broods in both years). Poult survival probability was well within the range of published estimates, we identified important sources of variation in reproductive rates and identify factors potentially contributing to the apparent decline. Many of our estimates of reproductive rates—nest survival, renesting rate, clutch size, and hatchability—were well within the range of previously reported estimates for these reproductive rates (Table 2). However, adult hen nesting rate and the probability of poult survival probability increased as the number of days posthatch increased (i.e., as poult age increases; Figure 9, Table A6). Daily poult survival probability was not affected by precipitation or temperature and did not differ between years of the study (Table A6).
4 | DISCUSSION
Our investigation of a turkey population exhibiting an apparent decline allowed us to identify important sources of variation in reproductive rates and identify factors potentially contributing to the apparent decline. Many of our estimates of reproductive rates—nest survival, renesting rate, clutch size, and hatchability—were well within the range of previously reported estimates for these reproductive rates (Table 2). However, adult hen nesting rate and the probability of poult survival probability increased as the number of days posthatch increased (i.e., as poult age increases; Figure 9, Table A6). Daily poult survival probability was not affected by precipitation or temperature and did not differ between years of the study (Table A6).
Although we found nest survival probability was well within the range of published estimates, we identified important sources of variation in this reproductive rate. In particular, we found that nest survival was substantially lower in agricultural fields relative to other cover types. This was largely driven by failure of all nests placed in alfalfa fields, which were mostly lost due to haying operations. Haying can be an important cause of mortality and nest failure for turkey and other avian species within agricultural landscapes (Bollinger, Bollinger, & Gavin, 1990; Shields & Flake, 2006; Wright, Paisley, & Kubisiak, 1996). Standard recommendations for wildlife-friendly farming include delaying haying until mid-July to allow nests in agricultural fields to fledge, haying from the center of the field and working out to allow nesting birds to escape, and installing flushing bars to minimize mortality (NRCS, 1999). Incentive programs designed to restrict haying during grassland bird nesting seasons have been demonstrated to improve avian reproductive success (Perlut, Strong, & Alexander, 2011). However, delaying haying operations can come at high economic costs to farmers and such wildlife-friendly practices are unlikely to succeed without established incentive programs.
Like the agricultural activities described above, predation is another important source of nest failure. Nest-site selection is strongly driven by avoidance of predation (Gill, 2007). Turkeys are known to select nests sites that are visually obstructed to predators (Isabelle, Conway, Comer, Calkins, & Hardin, 2016; Wood, Cohen, Conner, Collier, & Chamberlain, 2019), and our study demonstrated that selection for nest sites with high visual obstruction may reduce predation risk, as the probability of nest success was greater at sites with relatively high visual obstruction (Yarnall, Litt, & Lehman, 2019). Turkeys may also select nests located farther away from roads as a predator avoidance strategy. Beasley, DeVault, and Rhodes (2007) noted high raccoon prevalence near roads, and coyotes are known to use roads as travel corridors and forage along roads (Tigas, Van Vuren, & Sauvajot, 2002; Gosselink, Van Deelen, Warner & Joselyn, 2003; Hinton, van Manen, & Chamberlain, 2015). Our study found that nests located close to roads had a lower probability of survival, likely due to increased predation risk near roads. This result is similar to Badayev (1995), who found that successful turkey nests were located farther from roads, on average, than unsuccessful nests.
While many predators hunt visually, many also hunt by olfaction. Roberts and Porter (1998b) and Lehman, Rumble, et al. (2008) found a negative association between precipitation and nest survival probability, presumably because of increased detection of incubating hens by predators. Our finding of a positive association with daily nest survival probability and amount of precipitation at cold temperatures seemingly contradicts these findings. However, we found that for a given amount of precipitation, survival probability decreased as temperature increased, which is consistent with Roberts and Porter’s (1998b) suggestion that olfactory cues to locate nests may be enhanced during warm, wet periods. Alternatively, the nesting seasons captured during this study were, on average, drier than a typical spring in our study area (Figure 2). It is possible that the effects of precipitation on nest survival may extend beyond individual rainfall events and be cumulative in nature, such that rainfall events in relatively dry years may influence nest survival differently than rainfall events in relatively wet years.
In addition to external factors such as cover type, temperature, and precipitation, we found intrinsic factors such as experience and body condition were important sources of variation in reproductive rates. For example, nesting rate increased as hen body weight increased, likely because heavier birds are better able to complete the energetically expensive task of incubating a clutch (Porter et al., 1983). Following nest failure, adult hens were more likely to renest than yearling hens, which is consistent with previous research (Lehman et al., 2001; Pollentier, Lutz, et al., 2014; Shields & Flake, 2006). Similarly, poult survival probability increased with the number of days posthatch, which is also consistent with many other studies (Hubbard et al., 1999; Lehman et al., 2001; Paisley et al., 1998; Pollentier, Lutz, et al., 2014; Porter et al., 1983; Shields & Flake, 2006; Switzer & Tucker, 2009; Vander Haegen et al., 1988; Vangilder & Kurzjeski, 1995). Older poult are better able to escape ground predators (Spears et al., 2007) and thermoregulate (Healy &
Nenno, 1985; Schmidt-Nielsen, 1997), increasing daily survival probability as they age.
An important source of uncertainty in the apparent regional decline in turkey abundance is the role of loss of land enrolled in the Conservation Reserve Program (CRP). The amount of land enrolled in CRP has been steadily declining for more than a decade (Hellerstein, 2017), with some of the steepest declines occurring in northeastern South Dakota (USDA, 2016). It is possible that conversion of former CRP lands to agricultural production could translate to increased probability of nesting in agricultural fields, leading to decreases in reproductive output and consequently declining population growth rates. Future research should determine how nest-site selection varies as a consequence of loss of CRP habitat, and whether loss of CRP habitat increases the probability of nesting in low productivity habitats such as agricultural fields.
Our research highlighted reproductive parameters that may be contributing to an apparent population decline in turkey in northeastern South Dakota. Survival of poult raised by yearlings is low compared to other published studies. However, since uncontrollable intrinsic (e.g., hen age) and extrinsic (e.g., precipitation) factors appeared to have the greatest effects on poult survival, there is little managers can do to increase poult survival, particularly without knowledge of factors contributing to low survival. Of all the reproductive parameters we evaluated, nest survival may be the most amenable to changes through management action. Although our estimates of nest survival were not low compared to other studies, neither were the estimates close to upper limits observed in other studies and there may be room to increase nest survival and reproductive output. In particular, we found low daily survival probability in agricultural fields relative to other cover types. To increase nest survival in these cover types, managers could promote wildlife-friendly practices such as delayed haying (NRCS, 1999). Alternatively, increasing the availability of suitable nesting cover types (e.g., CRP fields) may lead to lower probabilities of nesting in agricultural fields. Improvements to productivity and recruitment could potentially stabilize, or lead to growth of, the population of turkeys in northeastern South Dakota.
ACKNOWLEDGMENTS
This project was funded by the South Dakota Department of Game, Fish and Parks (Grant W-75-R, Federal Aid Study 7564), the National Wild Turkey Federation (NWTF Project No.: 2.1.2017), and West Virginia University. CTR was supported in part by the USDA National Institute of Food and Agriculture, McIntire Stennis project WVA00124. We thank research technicians M. Ahern, K. Ostrander, N. Vruno, and J. Wolf for their help with field data collection, and N. Rossman and N. Markl for assistance with wild turkey capture events, aerial monitoring of radio-marked wild turkeys, and other field operations.
CONFLICT OF INTEREST
The authors have declared that no competing interests exist.
AUTHOR CONTRIBUTIONS
Reina M. Tyl: Data curation (lead); formal analysis (equal); investigation (lead); methodology (equal); visualization (equal); writing – original draft (lead); writing – review and editing (equal). Christopher T. Rota: Conceptualization (equal); formal analysis (equal); funding acquisition (equal); methodology (equal); project administration (equal); resources (equal); software (equal); supervision (equal); visualization (equal); writing – original draft (equal); writing – review and editing (equal). Chadwick P. Lehman: Conceptualization (equal); funding acquisition (equal); methodology (equal); project administration (equal); resources (equal); supervision (equal); writing – review and editing (equal).
OPEN RESEARCH BADGES
This article has earned an Open Data Badge for making publicly available the digitally-shareable data necessary to reproduce the reported results. The data is available at https://doi.org/10.5061/dryad.5dv41ns3d.
DATA AVAILABILITY STATEMENT
Data are archived in the Dryad Data Repository. Link to data during review process: https://datadryad.org/stash/share/0SbTuPgkKGU6HOTBud3yLDsHeGb5ooCZrtcFwTZ3s and DOI upon data publication: https://doi.org/10.5061/dryad.5dv41ns3d.
ORCID
Reina M. Tyl https://orcid.org/0000-0003-4540-778X
REFERENCES
Badyaev, A. V. (1995). Nesting habitat and nesting success of eastern wild turkeys in the Arkansas Ozark highlands. The Condor, 97, 221–232. https://doi.org/10.2307/1368998
Beasley, J. C., Devault, T. L., & Rhodes, O. E. Jr (2007). Home-range attributes of raccoons in a fragmented agricultural region of northern Indiana. The Journal of wildlife management, 71(3), 844–850.
Benkobi, L., Uresk, D. W., Schenbeck, G., & King, R. M. (2000). Protocol for monitoring standing crop in grasslands using visual obstruction. Journal of Range Management, 53, 627–633. https://doi.org/10.2307/4003158
Bollinger, E. K., Bollinger, P. B., & Gavin, T. A. (1990). Effects of hay-cropping on eastern populations of the bobolink. Wildlife Society Bulletin, 111, 142–150.
Caswell, H. (2001). Matrix population models. Sunderland, MA: Sinauer Associates Inc.
Clawson, M. R., & Rotella, J. J. (1998). Success of artificial nests in CRP fields, native vegetation, and field borders in southwestern Montana. Journal of Field Ornithology, 69, 180–191.
Daubenmire, R. F. (1959). Canopy coverage method of vegetation analysis, Northwest Science, 33, 43–64.
Fair, J. M., Paul, E., & Jones, J. (2010). Guidelines to the use of wild birds in research. Washington, DC: Ornithological Council.
Flint, R. F. (1955). Pleistocene Geology of Eastern South Dakota. Washington, DC: Government Printing Office.
Gelman, A., & Hill, J. (2007). Data analysis using regression and multilevel/hierarchical models. New York, NY: Cambridge University Press.
Gelman, A., Stern, H. S., Carlin, J. B., Dunson, D. B., Vehtari, A., & Rubin, D. B. (2013). *Bayesian data analysis* (3rd ed.). Boca Raton, FL: Chapman and Hall/CRC Press.
Gibson, D., Blomberg, E. J., & Sedinger, J. S. (2016). Evaluating vegetation effects on animal demographics: The role of plant phenology and sampling bias. *Ecology and Evolution*, 6, 3621–3631. https://doi.org/10.1002/ece3.2148
Gill, F. B. (2007). *Ornithology* (3rd ed.). New York, NY: W.H. Freeman & Company.
Gosselink, T. E., Van Deelen, T. R., Warner, R. E., & Joselyn, M. G. (2003). Temporal habitat partitioning and spatial use of coyotes and red foxes in east-central Illinois. *The Journal of Wildlife Management*, 67, 90–103.
Healy, W. M. (1992). Population influences: Environment. In J. G. Dickson (Ed.), *The wild turkey: Biology and management* (pp. 129–143). Harrisburg, PA: Stackpole Books.
Healy, W. M., & Nenno, E. S. (1985). Effect of weather on wild turkey poult survival. *Proceedings of the National Wild Turkey Symposium*, 5, 91–101.
Hellerstein, D. M. (2017). The US Conservation Reserve Program: The evolution of an enrollment mechanism. *Land Use Policy*, 63, 601–610. https://doi.org/10.1016/j.landusepol.2015.07.017
Hinton, J. W., van Manen, F. T., & Chamberlain, M. J. (2015). Space use and habitat selection by resident and transient coyotes (Canis latrans). *PLoS One*, 10(7), e0132203.
Hubbard, M. W., Garner, D. L., & Klaas, E. E. (1999). Wild turkey poult survival in southcentral Iowa. *The Journal of Wildlife Management*, 63, 199–203. https://doi.org/10.2307/3802501
Huxoll, C. (2016). *Annual report big game harvest projections. South Dakota game report No. 2017-06*. Pierre, SD: South Dakota Department of Game, Fish, and Parks.
Isabelle, J. L., Conway, W. C., Comer, C. E., Calkins, G. E., & Hardin, J. B. (2016). Reproductive ecology and nest-site selection of eastern wild turkeys translocated to east Texas. *Wildlife Society Bulletin*, 40, 88–96. https://doi.org/10.1002/wsb.632
Johnson, J. R., & Larson, G. E. (2007). *Grassland plants of South Dakota and the northern great plains*. Brookings, SD: South Dakota State University College of Agriculture & Biological Sciences and South Dakota Agricultural Experiment Station.
Johnson, R. R., Higgins, K. F., & Hubbard, D. E. (1995). Using soils to delineate South Dakota physiographic regions. *Great Plains Research*, 5, 309–322.
Kellner, K. (2018). *jagsUI*: A wrapper around RJAGS to streamline JAGS analyses. R package version 1.4.9. Retrieved from https://CRAN.R-project.org/package=jagsUI
Knupp Moore, P. M., & Flake, L. D. (1994). Forest characteristics in eastern and central South Dakota. *Proceedings of the South Dakota Academy of Science*, 73, 163–174.
Leatherberry, E. C., Piva, R. J., & Josten, G. J. (2000). South Dakota’s forest resources outside the Black Hills National Forest, 1996. St. Paul, MN: North Central Research Station, USDA Forest Service.
Lehman, C. P., Flake, L. D., Leif, A. P., & Shields, R. D. (2001). Comparative survival and reproduction of sympatric eastern and Rio Grande wild turkey females in northeastern South Dakota. *Proceedings of the National Wild Turkey Symposium*, 8, 123–125.
Lehman, C. P., Flake, L. D., Rumble, M. A., & Thompson, D. J. (2008). Merriam’s turkey poult survival in the Black Hills, South Dakota. *Intermountain Journal of Sciences*, 14, 78–88.
Lehman, C. P., Rumble, M. A., Flake, L. D., & Thompson, D. J. (2008). Merriam’s turkey nest survival and factors affecting nest predation by mammals. *The Journal of Wildlife Management*, 72, 1765–1774. https://doi.org/10.2193/2007-519
Lutz, R. S., & Crawford, J. A. (1987). Reproductive success and nesting habitat of Merriam’s wild turkeys in Oregon. *The Journal of Wildlife Management*, 51, 783–787. https://doi.org/10.2307/3801740
McConnell, M. D., Monroe, A. P., Burger, L. W. Jr., & Martin, J. A. (2017). Timing of nest vegetation measurement may obscure adaptive significance of nest-site characteristics: A simulation study. *Ecology and Evolution*, 7, 1259–1270. https://doi.org/10.1002/ece3.2767
McRoberts, T. T., Wallace, M. C., & Eaton, S. W. (2014). Wild Turkey (*Meleagris gallopavo*), version 2.0. In A. F. Poole (Ed.), *The birds of North America*. Ithaca, NY: Cornell Lab of Ornithology. https://doi.org/10.2173/bna.22
Menne, M. J., Durre, I., Korzeniewski, B., McNeal, S., Thomas, K., Yin, X., ... Houston, T. G. (2012). *Global historical climatology network – daily (GHCN-Daily)*. Version 3.26. Asheville, NC: NOAA National Climatic Data Center. https://doi.org/10.7289/V5D21VHZ
Miller, K. F., Kempf, L. S., & Koopman, V. F. (1979). *Soil survey of grant county, South Dakota*. Washington, DC: USDA Soil Conservation Service.
Mills, L. S. (2007). *Conservation of wild turkey populations*: Demography, genetics, and management. Malden, MA: Blackwell Publishing.
Natural Resources Conservation Service [NRCs] (1999). *Grassland birds. Fish and wildlife habitat management leaflet number 8*. Madison, MS: Natural Resources Conservation Service.
Paisley, R. N., Wright, R. G., Kubisja, J. F., & Rolley, R. E. (1998). Reproductive ecology of eastern wild turkeys in southwestern Wisconsin. *The Journal of Wildlife Management*, 62, 911–916. https://doi.org/10.2307/3802542
Perlut, N. G., Strong, A. M., & Alexander, T. J. (2011). A model for integrating wildlife science and agri-environmental policy in the conservation of declining species. *Journal of Wildlife Management*, 75, 1657–1663. https://doi.org/10.1002/jwmg.199
Plummer, M. (2003). *JAGS: A program for analysis of Bayesian graphical models using Gibbs sampling*. Proceedings of the 3rd international workshop on distributed statistical computing, Vienna, Austria.
Pollentier, C. D., Hull, S. D., & Lutz, R. S. (2014). Eastern wild turkey demography: sensitivity of vital rates between landscapes. *The Journal of Wildlife Management*, 78, 1372–1382. https://doi.org/10.1002/jwmg.787
Pollentier, C. D., Lutz, R. S., & Hull, S. D. (2014). Survival and productivity of eastern wild turkey females in contrasting landscapes in Wisconsin. *The Journal of Wildlife Management*, 78, 985–996. https://doi.org/10.1002/jwmg.749
Porter, W. F., Nelson, G. C., & Mattson, K. (1983). Effects of winter conditions on reproduction in a northern wild turkey population. *The Journal of Wildlife Management*, 47, 281–290. https://doi.org/10.2307/3808500
R Core Team (2018). *R: A Language and Environment for Statistical Computing*. Vienna, Austria: R Foundation for Statistical Computing. Retrieved from https://www.R-project.org/
Reitsma, K. D., Clay, D. E., Carlson, C. G., Dunn, B. H., Smart, A. J., Wright, D. L., & Clay, S. A. (2014). Estimated South Dakota land use change from 2006 to 2012. *Agronomy, Horticulture and Plant Science Faculty Publications*, 18. Retrieved from https://openprairie.sdstate.edu/plant_faculty_pubs/18
Robel, R. J., Briggs, J. N., Dayton, A. D., & Hulbert, L. C. (1970). Relationships between visual obstruction measurements and weight of grassland vegetation. *Journal of Range Management*, 23, 295–297. https://doi.org/10.2307/3896225
Roberts, S. D., Coffey, J. M., & Porter, W. F. (1995). Survival and reproduction of female wild turkeys in New York. *The Journal of Wildlife Management*, 59, 437–447. https://doi.org/10.2307/3802449
Roberts, S. D., & Porter, W. F. (1998a). Influence of temperature and precipitation on survival of wild turkey poult. *The Journal of Wildlife Management*, 62, 1499–1505. https://doi.org/10.2307/3802016
Roberts, S. D., & Porter, W. F. (1998b). Relation between weather and survival of wild turkey nests. *The Journal of Wildlife Management*, 62, 1492–1498. https://doi.org/10.2307/3802015
Rolley, R. E., Kubisiak, J. F., Paisley, R. N., & Wright, R. G. (1998). Wild turkey population dynamics in southwestern Wisconsin. The Journal of Wildlife Management, 62, 917–924. https://doi.org/10.2307/3802543
Royle, J. A., & Dorazio, R. M. (2008). Hierarchical modeling and inference in ecology: The analysis of data from populations, metapopulations, and communities. San Diego, CA: Academic Press.
Saether, B.-E., & Bakke, Ø. (2000). Avian life history variation and contributions of demographic traits to the population growth rate. Ecology, 81, 642–653. https://doi.org/10.1890/0012-9658(2000)081[0642:ALHVA]2.0.CO;2
Schmidt-Nielsen, K. (1997). Animal physiology: Adaptation and environment. Cambridge, UK: Cambridge University Press.
Shields, R. D., & Flack, L. D. (2006). Survival and reproduction of translocated eastern wild turkeys in a sparsetly wooded landscape in north eastern South Dakota. Western North American Naturalist, 66, 298–309. https://doi.org/10.3398/1527-0904(2006)66[298:SAROT]2.0.CO;2
Smith, J. T., Tack, J. D., Doherty, K. E., Allred, B. W., Maestas, J. D., Tigas, L. A., Van Vuren, D. H., & Sauvajot, R. M. (2002). Behavioral responses of bobcats and coyotes to habitat fragmentation and corridors in an urban environment. Biological Conservation, 108(3), 299–306.
USDA (2016). Change in CRP enrollment 2007-2016. Washington, DC: USDA. Retrieved from https://www.fsa.usda.gov/Assets/USDA-FSA-Public/usdafiles/Conservation/PDF/ChangeInCRPAcreagefrom2007_2016.pdf
Vander Haegen, W. M., Dodge, W. E., & Sayre, M. W. (1988). Factors affecting productivity in a northern wild turkey population. The Journal of Wildlife Management, 51, 127–133. https://doi.org/10.2307/3801072
Vangilder, L. D., & Kurzejeski, E. W. (1995). Population ecology of the eastern wild turkey in northern Missouri. Wildlife Monographs, 130, 1–50.
Vangilder, L. D., Kurzejeski, E. W., Kimmel-Truitt, V. L., & Lewis, J. B. (1997). Reproductive parameters of wild turkey hens in north Missouri. The Journal of Wildlife Management, 51, 535–540. https://doi.org/10.2307/3801265
Williams, L. E. Jr (1961). Notes on wing molt in the yearling wild turkey. The Journal of Wildlife Management, 25, 439–440. https://doi.org/10.2307/3798838
Wood, J. D., Cohen, B. S., Conner, L. M., Collier, B. A., & Chamberlain, M. J. (2019). Nest and brood site selection of eastern wild turkeys. Journal of Wildlife Management, 83, 192–204. https://doi.org/10.1002/jwmg.21562
Wright, R. G., Paisley, R. N., & Kubisiak, J. F. (1996). Survival of wild turkey hens in southwestern Wisconsin. Journal of Wildlife Management, 60, 313–320. https://doi.org/10.2307/3802230
Wunz, G. A. (1984). Rocket-net innovations for capturing wild turkeys and waterfowl. Transactions of the Northeast Section of the Wildlife Society, 41, 219.
Yang, L., Jin, S., Danielson, P., Homer, C., Gass, L., Bender, S. M., … Funk, M. (2018). A new generation of the United States National Land Cover Database: Requirements, research priorities, design, and implementation strategies. ISPRS Journal of Photogrammetry and Remote Sensing, 146, 108–123.
Yarnall, M. J., Litt, A. R., & Lehman, C. P. (2019). Timing of vegetation sampling does not influence associations between visual obstruction and turkey nest survival in a montane forest. Ecology and Evolution, 9, 11791–11798. https://doi.org/10.1002/ece3.5681
How to cite this article: Tyl RM, Rota CT, Lehman CP. Factors influencing productivity of eastern wild turkeys in northeastern South Dakota. Ecol Evol. 2020;10:8838–8854. https://doi.org/10.1002/ece3.6583
## APPENDIX 1
### TABLE A1 Log odds ratio (LOR), lower 95% credible interval (CI) level, and upper 95% CI level for each slope coefficient obtained from Bayesian nesting rate model fit to data collected in 2017 and 2018 from eastern wild turkeys \textit{(Meleagris gallopavo silvestris)} in northeastern South Dakota, USA
| Covariate | LOR | Lower CI level | Upper CI level |
|---------------------|------|----------------|----------------|
| (Intercept) | 0.92 | 0.62 | 1.25 |
| Hen age: adult | 0.59 | −0.03 | 1.24 |
| Study year: 2018* | 0.16 | −0.23 | 0.53 |
| Hen weight | 0.45 | 0.00 | 0.90 |
*Note, we used sum-to-zero constraints for study year. Therefore, the coefficient for study year 2017 = −1 × the coefficient reported above.
### TABLE A2 Population-level mean log odds ratio (LOR), lower 95% credible interval (CI) level, and upper 95% CI level for each slope coefficient obtained from Bayesian nest survival model fit to data collected in 2017 and 2018 from eastern wild turkeys \textit{(Meleagris gallopavo silvestris)} in northeastern South Dakota, USA
| Covariate | LOR | Lower CI level | Upper CI level |
|---------------------|------|----------------|----------------|
| (Intercept) | 3.49 | 3.07 | 3.91 |
| Hen age: adult | 0.60 | 0.08 | 1.15 |
| Precipitation | 1.07 | 0.43 | 1.97 |
| Temperature | 0.05 | −0.32 | 0.41 |
| Precipitation × temperature | −0.85 | −1.59 | −0.21 |
| Cover type: agriculture | −1.35 | −2.44 | −0.28 |
| Cover type: forest | 0.30 | −0.71 | 1.36 |
| Cover type: pasture | 0.85 | −0.19 | 1.96 |
| Mean total cover | −0.32| −0.89 | 0.25 |
| Mean VOR | 0.89 | 0.32 | 1.45 |
| Distance to road | 0.48 | 0.03 | 0.95 |
### TABLE A3 Log odds ratio (LOR), lower 95% credible interval (CI) level, and upper 95% CI level for each slope coefficient obtained from Bayesian renesting rate model fit to data collected in 2017 and 2018 from eastern wild turkeys \textit{(Meleagris gallopavo silvestris)} in northeastern South Dakota, USA
| Covariate | LOR | Lower CI level | Upper CI level |
|---------------------|------|----------------|----------------|
| (Intercept) | −1.10| −1.76 | −0.41 |
| Hen age: adult | 1.49 | 0.57 | 2.38 |
| Study year: 2018* | 0.61 | −0.00 | 1.28 |
| Hen weight | 0.03 | −0.65 | 0.74 |
| Previous nest duration | −0.02 | −0.75 | 0.71 |
| Previous fail date | −0.96| −1.81 | −0.20 |
*Note, we used sum-to-zero constraints for study year. Therefore, the coefficient for study year 2017 = −1 × the coefficient reported above.
### TABLE A4 Log proportional change, lower 95% credible interval (CI) level, and upper 95% CI level for each slope coefficient obtained from Bayesian clutch size model fit to data collected in 2017 and 2018 from eastern wild turkeys \textit{(Meleagris gallopavo silvestris)} in northeastern South Dakota, USA
| Covariate | LOR | Lower CI level | Upper CI level |
|---------------------|------|----------------|----------------|
| (Intercept) | 2.31 | 2.18 | 2.42 |
| Hen age: adult | 0.05 | −0.08 | 0.19 |
| Study year: 2018* | −0.02| −0.08 | 0.04 |
| Hen weight | 0.03 | −0.04 | 0.09 |
| Nest attempt: renest* | 0.00 | −0.07 | 0.07 |
*Note, we used sum-to-zero constraints. Therefore, the coefficient for the second category (year 2017 or nest attempt 1) = −1 × the coefficient reported above.
### TABLE A5 Log odds ratio (LOR), lower 95% credible interval (CI) level, and upper 95% CI level for each slope coefficient obtained from Bayesian hatchability model fit to data collected in 2017 and 2018 from eastern wild turkeys \textit{(Meleagris gallopavo silvestris)} in northeastern South Dakota, USA
| Covariate | LOR | Lower CI level | Upper CI level |
|---------------------|------|----------------|----------------|
| (Intercept) | 1.97 | 1.52 | 2.46 |
| Hen age: adult | 0.07 | −0.53 | 0.66 |
| Study year: 2018* | −0.09| −0.36 | 0.17 |
| Hen weight | −0.01| −0.30 | 0.30 |
*Note, we used sum-to-zero constraints for study year. Therefore, the coefficient for study year 2017 = −1 × the coefficient reported above.
### TABLE A6 Log odds ratio (LOR), lower 95% credible interval (CI) level, and upper 95% CI level for each slope coefficient obtained from Bayesian poult survival model fit to data collected in 2017 and 2018 from eastern wild turkeys \textit{(Meleagris gallopavo silvestris)} in northeastern South Dakota, USA
| Covariate | LOR | Lower CI level | Upper CI level |
|---------------------|------|----------------|----------------|
| (Intercept) | 3.53 | 3.38 | 3.69 |
| Poult age | 1.13 | 0.97 | 1.29 |
| Hen age: adult | 0.25 | 0.13 | 0.36 |
| Study year: 2018* | −0.10| −0.32 | 0.11 |
| Precipitation | 0.34 | −0.11 | 1.09 |
| Temperature | −0.13| −0.40 | 0.09 |
| Precipitation × temperature | −0.03 | −1.11 | 0.62 |
APPENDIX 2
Details regarding how informative prior distributions were obtained.
NESTING RATE, RENESTING RATE, AND HATCHABILITY
We derived informative prior distributions for the intercept and adult log odds ratio (LOR) coefficients by drawing on those studies in Table 2 that distinguish between adult and juvenile nesting rate, renesting rate, and hatchability. For each reproductive parameter, we integrated previous studies into a beta-binomial model:
\[ y_i \sim \text{Binomial}(p_i, N_i) \]
\[ p_i \sim \text{Beta}(a, b) \]
where \( y_i \) is the number of “successes” from study \( i \) (defined as hens nesting, hens renesting, or eggs hatching), \( N_i \) is the total number of “trials” included in study \( i \) (hens available to nest, hens available to renest, or eggs laid), \( p_i \) is the probability of success, and \( a \) and \( b \) are shape parameters governing the beta distribution. We assumed gamma (shape \( = 0.1 \); rate \( = 0.1 \)) prior distributions for both \( a \) and \( b \). For each reproductive parameter, we separately estimated posterior distributions of \( a \) and \( b \) for both adults and juveniles.
After obtaining posterior distributions of \( a \) and \( b \) for both adults and juveniles, we estimated the posterior distribution of the probability of success as:
\[ r = \frac{a}{a + b} \]
which is the expected value of a beta random variable. Since we modeled the probability of success for each reproductive parameter on the logit-linear scale, we transformed the quantity \( r \) (which is bound between 0 and 1) to the log odds scale:
\[ s = \log(r) - \log(1 - r), \]
which is the well-known logit transformation. To approximate the adult LOR coefficient, we calculated the difference in \( s \) obtained from juvenile and adult log odds of success:
\[ t = s_a - s_j \]
where \( s_a \) and \( s_j \) are posterior distributions of \( s \) obtained from adults and juveniles, respectively.
Distributions of both \( s \) and \( t \) were very well approximated with a Gaussian distribution. We therefore used Gaussian distributions as prior distributions for the intercept and adult LOR coefficients. We used a Gaussian prior distribution with mean and SD calculated from posterior samples of \( s_j \) for the intercept coefficient of our logit-linear models of nesting probability, renesting probability, and hatchability. We used a Gaussian prior distribution with mean and SD calculated from posterior samples of \( t \) for the adult LOR slope coefficient of our logit-linear models of nesting probability, renesting probability, and hatchability.
NEST SURVIVAL
We derived informative prior distributions for the intercept and adult log odds ratio (LOR) coefficients by drawing on those studies in Table 2 that distinguish between adult and juvenile nest success. We assumed the probability that a hen successfully hatched at least 1 poult reported from study \( i \) was a beta random variable:
\[ p_i \sim \text{Beta}(a, b) \]
where \( a \) and \( b \) are shape parameters governing the beta distribution. We assumed gamma (shape \( = 0.1 \); rate \( = 0.1 \)) prior distributions for both \( a \) and \( b \). We separately estimated posterior distributions of \( a \) and \( b \) for both adults and juveniles.
After obtaining posterior distributions of \( a \) and \( b \) for both adults and juveniles, we estimated the posterior distribution of a hen successfully hatching at least 1 poult as:
\[ r = \frac{a}{a + b} \]
which is the expected value of a beta random variable. Since we modeled the daily probability of nest survival on the logit-linear scale, we transformed the quantity \( r \) (which is bound between 0 and 1) to the log odds scale:
\[ s = \log(r^{1/28}) - \log(1 - r^{1/28}), \]
which is the well-known logit transformation. Note that we also transformed the probability of hatching at least 1 poult to daily nest survival probability by taking the 28th root of \( r \), to reflect a 28-day incubation period. To approximate the adult LOR coefficient, we calculated the difference in \( s \) obtained from juvenile and adult log odds of daily nest survival:
\[ t = s_a - s_j \]
where \( s_a \) and \( s_j \) are posterior distributions of \( s \) obtained from adults and juveniles, respectively.
Distributions of both \( s \) and \( t \) were very well approximated with a Gaussian distribution. We therefore used Gaussian distributions as prior distributions for the intercept and adult LOR coefficients. We used a Gaussian prior distribution with mean and SD calculated from posterior samples of \( s_j \) for the intercept coefficient of our logit-linear models of daily nest survival. We used a Gaussian prior distribution with mean and SD calculated from posterior samples of \( t \) for the adult LOR slope coefficient of our logit-linear models of daily nest survival.
POULT SURVIVAL
We derived informative prior distributions for the intercept coefficient by drawing on those studies in Table 2 that report the probability of poult survival to 28 days. We assumed the probability that a poult survived to 28 days reported from study \( i \) was a beta random variable:
where \(a\) and \(b\) are shape parameters governing the beta distribution. We assumed gamma (shape = 0.1; rate = 0.1) prior distributions for both \(a\) and \(b\). We did not distinguish between poults raised by adults or yearlings, as this distinction was not made in studies listed in Table 2.
After obtaining posterior distributions of \(a\) and \(b\), we estimated the posterior distribution of a poult surviving to 28 days as:
\[ r = \frac{a}{a+b} \]
which is the expected value of a beta random variable. Since we modeled the daily probability of poult survival on the logit-linear scale, we transformed the quantity \(r\) (which is bound between 0 and 1) to the log odds scale:
\[ s = \log \left( \frac{r^{1/28}}{1 - r^{1/28}} \right), \]
which is the well-known logit transformation. Note that we also transformed the probability of a poult surviving to 28 days to daily poult survival probability by taking the 28th root of \(r\). The distribution of \(s\) was very well approximated with a Gaussian distribution. We therefore used a Gaussian prior distribution with mean and SD calculated from posterior samples of \(s\) for the intercept coefficient of our logit-linear model of daily poult survival.
**CLUTCH SIZE**
We derived informative prior distributions for the intercept and difference in log expected clutch size between adults and juveniles (hereafter simply “adult”) coefficients by drawing on those studies in Table 2 that distinguish between adult and juvenile clutch size. We assume mean clutch size reported from study \(i\) was a gamma random variable:
\[ p_i \sim \text{Gamma}(a, b) \]
where \(a\) and \(b\) are shape and rate parameters, respectively, governing the gamma distribution. We assumed gamma (shape = 1, rate = 1) prior distributions for both \(a\) and \(b\). We separately estimated posterior distributions of \(a\) and \(b\) for both adults and juveniles.
After obtaining posterior distributions of \(a\) and \(b\) for both adults and juveniles, we estimated the posterior distribution of expected clutch size as:
\[ r = \frac{a}{b} \]
which is the expected value of a gamma random variable. Since we modeled expected clutch size on the log scale, we transformed the quantity \(r\) as:
\[ s = \log(r). \]
To approximate the adult coefficient, we calculated the difference in \(s\) obtained from juvenile and adult clutch size models:
\[ t = s_a - s_j \]
where \(s_a\) and \(s_j\) are posterior distributions of \(s\) obtained from adults and juveniles, respectively.
Distributions of both \(s\) and \(t\) were very well approximated with a Gaussian distribution. We therefore used Gaussian distributions as prior distributions for the intercept and adult coefficients. We used a Gaussian prior distribution with mean and SD calculated from posterior samples of \(s_j\) for the intercept coefficient of our log-linear model of clutch size. We used a Gaussian prior distribution with mean and SD calculated from posterior samples of \(t\) for the adult coefficient of our log-linear model of clutch size. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3047,
1
],
[
3047,
9460,
2
],
[
9460,
11899,
3
],
[
11899,
16361,
4
],
[
16361,
22391,
5
],
[
22391,
28174,
6
],
[
28174,
31884,
7
],
[
31884,
34350,
8
],
[
34350,
35421,
9
],
[
35421,
39245,
10
],
[
39245,
45660,
11
],
[
45660,
51561,
12
],
[
51561,
59297,
13
],
[
59297,
62877,
14
],
[
62877,
68588,
15
],
[
68588,
73774,
16
],
[
73774,
76835,
17
]
]
} |
cd237a53a46044df499c8b2768bc1b051c427c5d | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 14,
"total-fallback-pages": 0,
"total-input-tokens": 36908,
"total-output-tokens": 11745,
"fieldofstudy": [
"Biology"
]
} | Activation of Chymotrypsin-Like Activity of the Proteasome during Ischemia Induces Myocardial Dysfunction and Death
Gina Sanchez1,2*, Daniela Berrios1, Ivonne Olmedo1, Javier Pezoa3, Jaime A. Riquelme4, Luis Montecinos3, Zully Pedrozo3,4, Paulina Donoso2,3*
1 Programa de Fisiopatología, Instituto de Ciencias Biomédicas, Facultad de Medicina, Universidad de Chile, Santiago, Chile, 2 Centro de Estudios Moleculares de la Célula, Facultad de Medicina, Universidad de Chile, Santiago, Chile, 3 Programa de Fisiología y Biofísica, Instituto de Ciencias Biomédicas, Facultad de Medicina, Universidad de Chile, Santiago, Chile, 4 Advanced Center for Chronic Diseases, Facultad de Ciencias Químicas y Farmacéuticas, Universidad de Chile, Santiago, Chile
* [email protected] (GS); [email protected] (PD)
Abstract
Inhibitors of the ubiquitin-proteasome system improve hemodynamic parameters and decrease the infarct size after ischemia reperfusion. The molecular basis of this protection is not fully understood since most available data report inhibition of the 26 proteasome after ischemia reperfusion. The decrease in cellular ATP levels during ischemia leads to the dissociation of the 26S proteasome into the 19S regulatory complex and the 20S catalytic core, which results in protein degradation independently of ubiquitination. There is scarce information on the activity of the 20S proteasome during cardiac ischemia. Accordingly, the aim of this work was to determine the effects of 30 minutes of ischemia, or 30 min of ischemia followed by 60 minutes of reperfusion on the three main peptidase activities of the 20S proteasome in Langendorff perfused rat hearts. We found that 30 min of ischemia produced a significant increase in the chymotrypsin-like activity of the proteasome, without changes in its caspase-like or trypsin-like activities. In contrast, all three activities were decreased upon reperfusion. Ixazomib, perfused before ischemia at a concentration that reduced the chymotrypsin-like activity to 50% of the control values, without affecting the other proteasomal activities, improved the hemodynamic parameters upon reperfusion and decreased the infarct size. Ixazomib also prevented the 50% reduction in RyR2 content observed after ischemia. The protection was lost, however, when simultaneous inhibition of chymotrypsin-like and caspase-like activities of the proteasome was achieved at higher concentration of ixazomib. Our results suggest that selective inhibition of chymotrypsin-like activity of the proteasome during ischemia preserves key proteins for cardiomyocyte function and exerts a positive impact on cardiac performance after reperfusion.
Introduction
Myocardial ischemia represents a severe cellular stress that triggers dramatic biochemical and metabolic changes in the heart. The generation of reactive oxygen species (ROS) during ischemia [1–3] initiates the oxidation and modification of cellular proteins by lipid hydroperoxides [4] that eventually produce irreversible cell damage and death. The length of the ischemic period is a critical determinant of cell survival or death. Reperfusion is essential to keep cells alive, but the burst of ROS generation and calcium overload that takes place upon reperfusion further increases cellular damage [5,6].
Protein degradation during ischemia provides amino acids to be used as substrates for mitochondrial energy production and avoids the accumulation of toxic aggregates. Proteasomes are proteolytic complexes responsible for the degradation of over 90% of cellular proteins. The 26S proteasome, composed by the 20S catalytic core plus the 19S regulatory complex, mediates the ATP-dependent degradation of ubiquitinated proteins while the 20S proteasome, that contains the catalytic subunits, degrades oxidized proteins independent of ubiquitination. Both, the 20S proteasome and the 26S proteasome coexist in the heart [7,8]. Inhibition of the proteasome as a pharmacological strategy to prevent cell damage in ischemia reperfusion has produced conflicting results. Several studies report that the activity of the 26S proteasome decreases after ischemia reperfusion [9–11] and that further pharmacological inhibition produces more damage [12]. Conversely, an increasing number of studies have shown that proteasome inhibitors protect the heart from IR damage [13–16].
The decrease in cellular ATP content that occurs during ischemia promotes the dissociation of the 20S catalytic core from its associated regulatory particles in the 26S proteasome [17]. There is scarce information on the activity of the 20S proteasome during ischemia. Accordingly, the aim of this work was to determine this activity in isolated rat hearts and to evaluate the effect of ixazomib, the first oral proteasome inhibitor approved by the FDA for the treatment of multiple myeloma, on ischemia reperfusion injury.
Methods
Male Sprague-Dawley (SD) rats (220–240 g) were obtained from the animal facility of the School of Medicine, University of Chile. Rats were kept at a temperature of 22 ± 3°C and 12 h light-dark cycle, with free access to standard food and water. All procedures in this study conform to the Guide for the Care and Use of Laboratory Animals, published by the U.S. National Institutes of Health (NIH, Publication No. 85–23, revised in 1996), and were approved by the Institutional Ethics Committee of the School of Medicine, Universidad de Chile (Protocol CBA#0399FMUCH).
Experimental protocol
SD male rats were anesthetized with pentobarbital (80 mg/kg, intraperitoneal) and heparin 100 U/kg was injected into the right atria. The heart was rapidly excised, mounted in a temperature regulated heart chamber and perfused at 37°C via the ascending aorta using a peristaltic infusion pump at a constant flow of 10–14 mL/min. The Krebs Henseleit solution contained (in mmol/L): 128.3 NaCl, 4.7 KCl, 1.35 CaCl₂, 1.1 MgSO₄, 20.2 NaHCO₃, 0.4 NaH₂PO₄, pH 7.4, and 11.1 glucose, equilibrated with a gas mixture of 95% O₂/5% CO₂. Left ventricular hemodynamic parameters were measured with a latex balloon inserted into the left ventricle and connected to a pressure transducer. After 20 min of stabilization, the hearts were subjected to 30 min of global ischemia at 37°C. Hearts were either frozen in liquid N₂ immediately after ischemia or perfused with Krebs Henseleit oxygenated solution for 60 min before freezing. The proteasome inhibitor ixazomib (MLN 9708, Selleckchem, Houston, TX, USA) was perfused during...
10 min before ischemia at concentrations of 0.1 or 1 μmol/L. To compare with published data, MG132 (Merck Millipore, Billerica, MA, USA) was perfused as above at concentrations of 0.5 or 6 μmol/L. Control hearts, not subjected to ischemia, were perfused with or without proteasome inhibitors for 50 (control for ischemia) or 110 min (control for ischemia–reperfusion). At the end of the reperfusion period, hearts were snap frozen in liquid N₂ or perfused with triphenyl tetrazolium (TTC) to measure the infarct size.
**Preparation of whole ventricle homogenates**
Frozen ventricles were reduced to powder under liquid N₂ and homogenized under slightly different conditions according to the ensuing biochemical determination. To measure the activity of the proteasome, frozen tissue powder was homogenized in 5 volumes of a solution containing (in mmol/L) 50 NaCl, 1 Na₂EDTA, 10 HEPES-NaOH, pH 8.0, 250 Sucrose, 0.2% Triton X-100, as described [18]. This fraction was prepared without reducing agents or protease inhibitors just before the determination of proteasome activity.
To measure [³H]-ryanodine binding, the frozen powder was homogenized in 4 volumes of a solution containing (in mmol/L) 300 sucrose, 20 MOPS-Tris buffer, pH 7.0, with protease inhibitors (1 mmol/L PMSF, 1 mmol/L benzamidine, 2 μg/mL leupeptine, 1 μg/mL pepstatin). Unbroken cells and debris were eliminated by centrifugation at 600 x g for 10 min, at 4°C. The supernatant (whole homogenate) was fractioned into small aliquots, frozen in liquid N₂ and kept at -80°C. For western blot analysis, the frozen tissue was homogenized as above, but the buffer contained in addition 2 mmol/L EDTA, 2 mmol/L EGTA and the detergents NP-40 (1%) and SDS (1%). The supernatant was recovered by centrifugation at 1000 x g for 20 min at 4°C. Aliquots were frozen in liquid N₂ and kept at -80°C as above. Protein concentration was determined by the method of Hartree [19].
**Proteasome activity**
The three main peptidase activities of the proteasome were determined in the absence of reducing agents as described before [18]. Briefly, heart homogenates (0.3 mg protein/mL) were incubated with fluorogenic proteasome substrates in a solution containing (mmol/L) 50 NaCl, 1 EDTA, 250 sucrose, 10 HEPES-NaOH, pH 8.0. The substrates used were Suc-LLVY-amc (21 μmol/L) for chymotrypsin-like activity, Z-LLE-amc (105 μmol/L) for caspase-like activity and Boc-LSTR-amc (34 μmol/L) for trypsin-like activity. Fluorescence was measured at 30°C in a plate reader at 380 nm excitation / 440 nm emission wavelengths. Non-specific proteolysis was determined in the presence of MG132 (30 μmol/L). All proteasome substrates were obtained from Sigma-Aldrich (St Louis, MI).
**Western blots**
Proteins were separated by electrophoresis in 3%–8% Tris-Acetate gels (Criterion XT, Bio-Rad, Hercules, CA) under reducing conditions. After transfer to polyvinylidene difluoride membranes, proteins were probed with anti-RyR2 antibody (Thermo Scientific; Rockford, IL) or anti-caspase-3 antibody (Cell Signaling Technology Inc. Danvers, MA). Anti-GAPDH antibody (Sigma-Aldrich, St Louis, MI) was used as loading control. The bands were quantified by densitometry and the results were normalized with respect to controls run in the same gel.
**[3H]-Ryanodine binding**
[3H]-ryanodine binding was measured in whole homogenates from frozen hearts (0.5 mg protein/mL) following 90 min incubation at 37°C with 10 nM [3H]-ryanodine (Perkin Elmer, Boston, MA) in 150 mol/L KCl, 0.5 mmol/L AMPPNP at pCa 5, as previously described [20].
**Infarct size**
Infarct size was assessed by the triphenyltetrazolium chloride (TTC, Sigma-Aldrich, St Louis, MI) technique as described [21].
**Statistical analysis**
Data are expressed as mean ± S.E.M. Statistical data were analyzed by ANOVA followed by Tukey post test. Differences were considered significant at p<0.05.
**Results**
**Dose dependent effects of proteasome inhibitors on hemodynamic parameters and infarct size after ischemia reperfusion**
The divergent effects of proteasome inhibitors on the cardiac parameters after IR reported in the literature may arise from different models that employ different types and concentrations of inhibitors. We therefore compared the effect of ixazomib with the widely used classical proteasome inhibitor MG132. In the absence of inhibitors, isolated rat hearts subjected to 30 minutes of non-flow global ischemia at 37°C displayed a severe contractile dysfunction upon reperfusion. After 60 minutes of reperfusion, the left ventricular developed pressure (LVDP, Fig 1A) and the maximum rate of left ventricular pressure rise (+dP/dt, Fig 1B) decreased by 85 and 90%, respectively, relative to the pre ischemic values. Perfusion of hearts with ixazomib for 10 minutes before ischemia produced a biphasic effect: at low concentration (0.1 μmol/L), ixazomib significantly improved LVDP (Fig 1A) and +dP/dt after IR (Fig 1B). At high concentration (1 μmol/L), however, this protective effect was lost and both, LVDP and +dP/dt, were not different to the values found after IR in the absence of inhibitor (Fig 1A and 1B). The same effect, protection at low concentration (0.5 μmol/L) and loss of protection at high concentration (6 μmol/L), was observed when MG132 was administered before ischemia (Fig 1A and 1B). At the concentrations used in this work, neither ixazomib nor MG132, at 0.5 μmol/L, produce significant changes in hemodynamic parameters in control hearts, not subjected to ischemia (S1 Fig). To corroborate the protective effect of a low concentration of proteasome inhibitors, we measured the infarct size after IR. In the absence of inhibitors, the infarct size was 39±8% of the cardiac volume (Fig 2). MG132 and ixazomib significantly reduced the infarct size to 15±8% and 24±6% of the heart volume, respectively, when administered at a low concentration, but did not protect at higher concentration (Fig 2). Therefore, ixazomib and MG132 exert a similar dose dependent cardioprotective effect. For this reason we used only ixazomib in most of the following experiments. No measurable infarction was detected in control hearts, perfused with Krebs Henseleit solution with or without proteasome inhibitors.
**Determination of proteasome activity**
The results shown above suggest that proteasomes are key players in the damage induced by IR and the biphasic effect of the inhibitors may be produced by different inhibition of proteasome peptidases at the different concentrations of ixazomib or MG132 used. Reports in the literature consistently show decreased proteasome activity after reperfusion (reviewed in [22]), but there...
Fig 1. Effect of proteasome inhibition on hemodynamic parameters after ischemia reperfusion. (A) Left ventricular developed pressure (LVDP) and (B) maximal rates of contraction (+ dP/dt) measured in control hearts or after IR with or without the indicated proteasome inhibitor. Bar graphs show the mean ± S.E.M of values measured at 60 minutes of reperfusion of the number of hearts shown in each bar. * p<0.05 vs IR without inhibitor.
doi:10.1371/journal.pone.0161068.g001
Fig 2. Effect of proteasome inhibition on the infarct size. Representative heart slices stained with TTC after IR (left) with or without the indicated proteasome inhibitor. The bar graph shows the mean ± S.E.M of the infarct size as % of total heart volume calculated in hearts like those shown at the left. Number of hearts analyzed in each condition is shown in each bar. * p<0.05 vs IR without inhibitor.
doi:10.1371/journal.pone.0161068.g002
is less information on the effect of ischemia on proteasome activity. Therefore we determined the effect of ixazomib on the activities of the 20S proteasome at the end of ischemia or at the end of the reperfusion period. We made the novel observation that after 30 minutes of ischemia in the absence of inhibitor, and before reperfusion, the chymotrypsin like (CT–like) activity was significantly increased by 50% compared to the value observed in control hearts (Fig 3A).
Fig 3. Proteasome activities after I or IR. Chymotrypsin-like (panel A), caspase-like (panel B) and trypsin-like (panel C) activities of the proteasome are shown in controls (C, white bars), after 30 min of ischemia (I, black bars) or after 30 min of ischemia followed by 60 min of reperfusion (IR, grey bars) in hearts perfused with the indicated concentration of ixazomib. Activities were normalized with respect to controls. Bars show the mean ± S.E.M. of different hearts as indicated in each bar. *p< 0.05 vs C without inhibitor.
doi:10.1371/journal.pone.0161068.g003
Ischemia did not produce changes in the caspase-like (C-like) or the trypsin-like (T-like) activities of the proteasome (Fig 3B and 3C). In contrast, after reperfusion all three main activities of the proteasome were significantly inhibited. On average, we observed reductions of 53%, 30% and 23% for CT-like, C-like and T-like activities respectively (Fig 3A–3C). In hearts perfused with ixazomib, 0.1 μmol/L, the CT-like activity was reduced approximately by half of the basal value both, in control and after 30 min of ischemia. Ixazomib did not cause further reduction of this activity upon reperfusion (Fig 3A). The C-like and the T-like activities of the proteasome were not inhibited by ixazomib 0.1 μmol/L, compared to the same condition without the inhibitor (Fig 3B and 3C). Increasing ixazomib to 1 μmol/L produced a further reduction in CT-like activity to about 25% of the basal value (Fig 3A) and also inhibited the C-like activity to less than 10% of the activity in the absence of inhibitor (Fig 3B). In contrast, perfusion of hearts with ixazomib 1 μmol/L increased the T-like activity of the proteasome in control hearts by 47% (Fig 3C). After 30 min of ischemia, the T-like activity remained significantly elevated but decreased after reperfusion to the value observed in the absence of inhibitor (Fig 3C). In summary, a low concentration of ixazomib (0.1 μmol/L), prevents the increase in CT-like activity induced by ischemia without inhibiting the C-like or T-like activities, while a higher concentration of ixazomib (1 μmol/L) significantly reduces both, the CT-like and C-like activities and increases the T-like activity of the proteasome under basal conditions and this activity remained elevated after 30 min of ischemia. Besides, ixazomib increased the content of ubiquitinated proteins in whole ventricle homogenates of control hearts in a dose dependent manner (S2 Fig), indicating that proteasomes were effectively inhibited by the concentrations used in this work. These results suggest that partial inhibition of CT-like activity during ischemia protects from IR damage, but simultaneous inhibition of CT-like and C-like activities is detrimental. The simultaneous increase in T-like activity may also contribute to the loss of the protective effect of ixazomib.
Effect of proteasome inhibition on RyR2 protein content
RyR2, the calcium release channel of the sarcoplasmic reticulum is reduced by half after myocardial ischemia in adult hearts [23,24] or after simulated IR in neonatal cardiomyocytes [25]. MG132 preserves the content of RyR2 in neonatal cardiomyocytes, suggesting that proteasomes are involved in the reduction of this protein during ischemia [25]. Therefore, we used RyR2 as a representative protein to corroborate the protective effect of ixazomib during ischemia. We measured RyR2 activity and protein content after ischemia or reperfusion by measuring the binding of [3H]-ryanodine and by performing western blots analysis of whole heart ventricles. We found that 30 minutes of global ischemia caused 50% decrease in [3H]-ryanodine binding density. No further decrease was observed upon reperfusion (Fig 4A). MG132, 0.5 μM, prevented the decrease in [3H]-ryanodine binding when perfused before ischemia but not when added from the very start of reperfusion (Fig 4A, red bars). Ixazomib, which did not modify [3H]-ryanodine binding in control hearts, prevented the decrease in [3H]-ryanodine binding when perfused before ischemia at 0.1 μmol/L but not at 1 μmol/L (Fig 4B). Since ryanodine binds preferentially to the open conformation of RyR2, its binding depends on the functional state of the channels. To test if the decrease in [3H]-ryanodine binding represents a true decrease in RyR2 protein content and does not result from RyR2 modifications that occurred during ischemia and affected the opening of the channel, we also quantified RyR2 content in western blots. As shown in Fig 5, RyR2 protein content decreased by 50% after ischemia and remained at this level after reperfusion; the presence of ixazomib during ischemia prevented the decrease in RyR2 at 0.1 but not at 1 μmol/L. These results strongly suggest that
preventing the increase in CT-like activity of the proteasome with ixazomib avoids the loss of RyR2 and other cellular proteins producing the observed beneficial effects.
**Activation of caspase 3 by ixazomib**
Inhibition of the proteasome induces apoptosis in different cells [26–28] and the loss of protection by stronger inhibition of the proteasome during myocardial ischemia suggests that a cell death program was activated in this condition. We explored this possibility by measuring caspase-3, an executioner caspase that mediates cellular apoptosis. As shown in Fig 6, a low concentration of ixazomib did not change the cleaved caspase-3 (Fig 6A and 6B) or pro-caspase-3 content (Fig 6A and 6C) in controls, after ischemia or after ischemia reperfusion; in contrast, ixazomib at 1 μM, increased both, the active caspase-3 content (Fig 6A and 6B) and the total pro-caspase-3 (Fig 6A and 6C) in hearts subjected to ischemia and in control hearts as well. The increase in active caspase-3 suggests that apoptosis was activated in these hearts with the consequent increase in protein degradation and cell death. The apparent normality of control hearts...
perfused with high concentration of ixazomib, in spite of the increased active caspase-3, is dis-
cussed below.
**Discussion**
Our results show that thirty minutes of warm ischemia in isolated rat hearts causes the death of
almost 40% of the heart and a poor recovery of hemodynamic parameters upon reperfusion.
We found that the CT-like activity of the 20S proteasome increases significantly after 30 min of
ischemia, and decreases below control values after reperfusion. The C-like and T-like activities,
which did not change during ischemia, decreased as well after reperfusion. These results sug-
gest that the CT-like activity of the proteasome is a privileged target of ischemia-induced
injury. The partial inhibition of CT-like activity during ischemia (to 50% of the control value)
produced by ixazomib reduced the infarct size by half and produced a significant recovery of
hemodynamic parameters, suggesting that during ischemia the 20S proteasome degrades key
cellular proteins, such as RyR2, and promotes cell dysfunction and death. Protection was
completely lost when the simultaneous inhibition of CT-like and C-like activity was achieved
by a higher concentration of ixazomib, which caused a simultaneous activation of T-like
activity.
Ixazomib belongs to the second-generation of proteasome inhibitors and it has been
recently approved by the FDA for the treatment of multiple myeloma. It inhibits preferentially
---
**Fig 5. Effect of ixazomib on RyR2 protein content.** Representative western blots show RyR2 content in
whole ventricle homogenates obtained from controls hearts (C), after ischemia (I), after ischemia-reperfusion
(IR) with the indicated concentration of ixazomib (ixa). Graph bar show the mean ± S.E.M of values obtained
in different hearts as indicated in each bar. Results were normalized by the content of GAPDH in the same
membrane. *: p < 0.05 vs control.
doi:10.1371/journal.pone.0161068.g005
the CT-like activity of the proteasome over the caspase-like activity (IC50 values of 3.4 and 31 nmol/L, respectively). A higher concentration is required for the inhibition of the T-like activity (IC50 3.5 μmol/L) [29]. The inhibition of the proteasome activities observed in this study is in agreement with these reported inhibitory constants. Further kinetic analysis would be necessary to clarify the molecular basis of the increase in T-like activity observed at the higher concentration of ixazomib. Nevertheless, a simultaneous inhibition of CT-like activity and enhancement of T-like activity is also produced by the inhibitor of the human immunodeficiency virus, type I (HIV-I) protease, ritonavir [30]. This effect would be produced by the binding of ritanovir to a non-catalytic modifier site of the proteasome [30]. A similar mechanism could explain the increase in T-like activity produced by ixazomib.
Previous studies report beneficial [13,14,31], or deleterious [9,10,12] effects of proteasome inhibitors on cardiac IR. The activity of the proteasome was not measured in all these studies and differences in the degree of proteasome inhibition, may explain the discrepancies. As shown here, the same inhibitor can improve or worsen cardiac function after an episode of IR, depending on the degree of inhibition of the different peptidases. Those studies that measure proteasome activity consistently found that ischemia-reperfusion decreases the activity of the proteasome [9,11,12]. We confirmed this inhibition in the present work. In contrast, there is
**Fig 6. Effect of ixazomib on the activation of caspase3.** Representative western blots (A) show cleaved caspase 3 and procaspase 3 content in whole ventricle homogenates obtained from controls hearts (C), after ischemia (I), and after ischemia-reperfusion (IR) with the indicated concentration of ixazomib (ixa). Graph bars show the mean ± S.E.M of cleaved caspase 3 (B) or procaspase 3 (C) obtained in different hearts as indicated in each bar. Results were normalized by the content of GAPDH in the same membrane. *: p < 0.05 vs control without inhibitor.
doi:10.1371/journal.pone.0161068.g006
not enough information about the activity of the proteasome after ischemia. Some studies report decreased 26S proteasome activity during ischemia (reviewed in [22]) and only one study reports that the 20S proteasome is also inhibited after ischemia [12]. At difference with that work, we did not use reducing agents in the preparation of heart extracts or in the assay medium. The presence of reducing agents, such as dithiothreitol, may have removed reversible redox modifications like S-glutathionylation that increases the activity of the 20S proteasome [32]. In rat hearts kept in Wisconsin solution at 4°C (cold ischemia), the activity of the 26S proteasome increases more than two fold as ATP content decreases during ischemia [33]. The inhibition of the proteasome at this stage preserves the ultra structural integrity of the hearts and reduces ischemia-reperfusion injury [33] prolonging the viability of the organ for transplant [34]. Cold ischemia did not produce changes in the 20S proteasome [33]. Warm ischemia (37°C), such as that induced here, is a totally different setting because diverse signaling pathways are activated at the beginning of ischemia before the intracellular media changes extensively. In addition, as the ATP concentration decreases, the disociation of the 26S proteasome may increase the abundance of the 20S proteasome, which actively degrades oxidized proteins independently of ATP [35,36].
A detailed elucidation of the molecular mechanisms responsible for the increased proteasome activity produced by ischemia is beyond the scope of the present work. Proteasomes are modulated by a number of post translational modifications [37] and proteomic studies are needed to further clarify this regulation.
Proteasome inhibitors were deliberately designed to produce apoptosis in cancer cells [38,39]. Ixazomib induces apoptosis in multiple myeloma cells through activation of caspases and other proapoptotic proteins such as p53, PUMA and Noxa and proteins involved in the endoplasmic reticulum stress pathway [40]. The increase in the executioner caspase-3 by ixazomib 1 μM, suggests that apoptosis increases when proteasomes are strongly inhibited during ischemia. The lack of visible deleterious effects in control hearts treated with this concentration of ixazomib suggests that apoptosis takes a longer time to complete in normal cells. In fact, proteasome inhibition-induced apoptosis is potentiated by the generation of reactive oxygen species and by increased calcium concentration [27], two conditions that have been amply demonstrated to occur in cardiac ischemia.
In addition, inhibition of the proteasome could have simultaneously activated other cell death pathways such as chaperon mediated autophagy and macroautophagy [41] which may have also contributed to the degradation of RyR2 [42] and to increase the infarct size upon reperfusion [43].
Supporting Information
S1 Fig. Effect of proteasome inhibition on hemodynamic parameters in control hearts. (A) Left ventricular developed pressure (LVDP) and (B) maximal rates of contraction (+ dP/dt) measured in control hearts with the indicated proteasome inhibitor. Inhibitors were perfused at the indicated concentration in oxygenated Krebs Henseleit solution at 37°C and at 10 mL/min during 10 min. Perfusion continued without inhibitor for 90 min, time at which values were measured. Bars show the mean ± S.E.M of 4 different hearts.
S2 Fig. Effect of ixazomib on the accumulation of ubiquitinated proteins. Whole ventricle homogenates were analyzed in Western blots for the presence of ubiquitinated proteins after perfusion with the indicated concentration of ixazomib. Total anti-ubiquitin immunoreactivity of each lane was normalized by the content of GAPDH in the same lane. Anti-ubiquitin antibody was obtained from Cell signaling (Danvers, MA). Bars show Mean ± S.E.M values
obtained in different hearts as indicated in the bars. *: p < 0.05.
Acknowledgments
The authors thank Dr. Cecilia Hidalgo for her critical reading of the manuscript. The technical assistance of Guillermo Arce and Rodrigo Durán is gratefully acknowledged.
Author Contributions
Conceived and designed the experiments: GS ZP PD.
Performed the experiments: GS DB IO JP JAR LM.
Analyzed the data: GS DB IO JP JAR LM ZP PD.
Contributed reagents/materials/analysis tools: GS ZP PD.
Wrote the paper: GS ZP IO JAR PD.
References
1. Vanden Hoek TL, Becker LB, Shao Z, Li C, Schumacker PT. Reactive oxygen species released from mitochondria during brief hypoxia induce preconditioning in cardiomyocytes. J Biol Chem. 1998; 273: 18092–18098. PMID: 9660766
2. Robin E, Guzy RD, Loor G, Iwase H, Waypa GB, Marks JD, et al. Oxidant stress during simulated ischemia primes cardiomyocytes for cell death during reperfusion. J Biol Chem. 2007; 282: 19133–19143. PMID: 17488710
3. Zhu X, Zuo L. Characterization of oxygen radical formation mechanism at early cardiac ischemia. Cell Death Dis. 2013; 4: e678. doi: 10.1038/cddis.2013.313 PMID: 24008731
4. Eaton P, Hearse DJ, Shattock MJ. Lipid hydroperoxide modification of proteins during myocardial ischemia. Cardiovasc Res. 2001; 51: 294–303. PMID: 11470469
5. Garcia-Dorado D, Ruiz-Meana M, Inserne J, Rodriguez-Sinovas A, Piper HM. Calcium-mediated cell death during myocardial reperfusion. Cardiovasc Res. 2012; 94: 168–180. doi: 10.1093/cvr/cvs116 PMID: 22499772
6. Hageneslo DJ, Yellon DM. Myocardial ischemia-reperfusion injury: a neglected therapeutic target. J Clin Invest. 2013; 123: 99–100. doi: 10.1172/JCI62874 PMID: 23281415
7. Gomes AV, Young GW, Wang Y, Zong C, Eghbali M, Drews O, et al. Contrasting proteome biology and functional heterogeneity of the 20 S proteasome complexes in mammalian tissues. Mol Cell Proteomics. 2008; 8: 302–315. doi: 10.1074/mcp.M800056-MCP200 PMID: 18931397
8. Gomes AV, Zong C, Edmondson RD, Li X, Stefani E, Zhang J, et al. Mapping the murine cardiac 26S proteasome complexes. Circ Res. 2006; 99: 362–371. PMID: 16857966
9. Bulteau AL, Lundberg KC, Humphries KM, Sadek HA, Szweda PA, Friguet B, et al. Oxidative modification and inactivation of the proteasome during coronary occlusion/reperfusion. J Biol Chem. 2001; 276: 30057–30063. PMID: 11375979
10. Divall A, Kivity S, Wang P, Hochhauser E, Roberts B, Teichberg S, et al. Myocardial ischemic preconditioning preserves postischemic function of the 26S proteasome through diminished oxidative damage to 19S regulatory particle subunits. Circ Res. 2010; 106: 1829–1838. doi: 10.1161/CIRCRESAHA.110.219485 PMID: 20431057
11. Powell SR, Davies KJ, Divall A. Optimal determination of heart tissue 26S-proteasome activity requires maximal stimulating ATP concentrations. J Mol Cell Cardiol. 2007; 42: 265–269. PMID: 17140599
12. Powell SR, Wang P, Katzef H, Shringarpure R, Teoh C, Khaliulin I, et al. Oxidized and ubiquitinated proteins may predict recovery of postischemic cardiac function: essential role of the proteasome. Antioxid Redox Signal. 2005; 7: 536–546. PMID: 15899999
13. Campbell B, Adams J, Shin YK, Lefter AM. Cardioprotective effects of a novel proteasome inhibitor following ischemia and reperfusion in the isolated perfused rat heart. J Mol Cell Cardiol. 1999; 31: 467–476. PMID: 10093058
14. Pye J, Ardeshirpour F, McCain A, Bellinger DA, Merricks E, Adams J, et al. Proteasome inhibition ablates activation of NF-kappa B in myocardial reperfusion and reduces reperfusion injury. Am J Physiol Heart Circ Physiol. 2003; 284: H919–926. PMID:12424098
15. Stansfield WE, Moss NC, Willis MS, Tang R, Selzman CH. Proteasome inhibition attenuates infarct size and preserves cardiac function in a murine model of myocardial ischemia-reperfusion injury. Ann Thorac Surg. 2007; 84: 120–125. PMID:17588397
16. Adams B, Mapanga RF, Essop MF. Partial inhibition of the ubiquitin-proteasome system ameliorates cardiac dysfunction following ischemia-reperfusion in the presence of high glucose. Cardiovasc Diabetol. 2015; 14: 94 doi: 10.1186/s12933-015-0258-4 PMID: 26216448
17. Asai A, Tanahashi N, Qiu JH, Saito N, Chi S, Kawahara N, et al. Selective proteasomal dysfunction in the hippocampal CA1 region after transient forebrain ischemia. J Cereb Blood Flow Metab. 2002; 22: 705–710. PMID: 12045669
18. Lima CF, Rattan SI. Determination of proteasomal activities. Methods Mol Biol 2010; 648: 183–192. doi: 10.1007/978-1-60761-756-3_12 PMID: 20700713
19. Hartree EF. Determination of protein: a modification of the Lowry method that gives a linear photometric response. Anal Biochem. 1972; 48: 422–427. PMID:4115981
20. Bull R, Marengo JJ, Suarez-Isla BA, Donoso P, Sutko JL, Hidalgo C, et al. Activation of calcium channels in sarcoplasmic reticulum from frog muscle by nanomolar concentrations of ryanodine. Biophys J. 1989; 56: 749–756. PMID:2554991
21. Vila-Petroff M, Salas MA, Said M, Valverde CA, Sapia L, Portiansky E, et al. CaMKII inhibition protects against necrosis and apoptosis in irreversible ischemia-reperfusion injury. Cardiovasc Res. 2007; 73: 689–698. PMID: 17217936
22. Majetschak M. Regulation of the proteasome by ATP: implications for ischemic myocardial injury and donor heart preservation. Am J Physiol Heart Circ Physiol. 2013; 305: H267–278. doi: 10.1152/ajpheart.00206.2012 PMID: 23709597
23. Temsah RM, Netticadan T, Chapman D, Takeda S, Mochizuki S, Dhalla NS, et al. Alterations in sarcoplasmic reticulum function and gene expression in ischemic-reperfused rat heart. Am J Physiol. 1999; 277: H584–594. PMID: 1044483
24. Domenech RJ, Sanchez G, Donoso P, Parra V, Macho P. Effect of tachycardia on myocardial sarcoplasmic reticulum and Ca2+ dynamics: a mechanism for preconditioning? J Mol Cell Cardiol. 2003; 35: 1429–1437. PMID: 14654369
25. Pedrozo Z, Sanchez G, Torrealba N, Valenzuela R, Fernandez C, Hidalgo C, et al. Calpains and proteasomes mediate degradation of ryanodine receptors in a model of cardiac ischemic reperfusion. Biochim Biophys Acta. 2010; 1802: 356–362. doi: 10.1016/j.bbadis.2009.12.005 PMID: 20026269
26. Muz B, Ghazarian RN, Ou M, Luderer MJ, Kusdono HD, Azab AK, et al. Spotlight on ixazomib: potential in the treatment of multiple myeloma. Drug Des Devel Ther. 2016; 10: 217–226. doi: 10.2147/DDDT.S93602 PMID: 26811670
27. Kupperman E, Lee EC, Cao Y, Bannerman B, Fitzgerald M, Berger A, et al. Evaluation of the proteasome inhibitor MLN9708 in preclinical models of human cancer. Cancer Res. 2010; 70: 1970–1980. doi: 10.1158/0008-5472.CAN-09-2766 PMID: 20160034
28. Schmidtke G, Emch S, Groettrup M, Holzhueter HG. Evidence for the existence of a non-catalytic modifier site of peptide hydrolysis by the 20 S proteasome. J Biol Chem. 2000; 275: 22056–22063. PMID: 10806206
29. Yu X, Kem DC. Proteasome inhibition during myocardial infarction. Cardiovasc Res. 2010; 85: 312–320. doi: 10.1093/cvr/cvp309 PMID: 19744947
30. Silva GM, Netto LE, Simoes V, Santos LF, Gozzo FC, Demasi MA, et al. Redox control of 20S proteasome gating. Antioxid Redox Signal. 2012; 16: 1183–1194. doi: 10.1089/ars.2011.4210 PMID: 22229461
31. Geng Q, Romero J, Saini V, Baker TA, Picken MM, Gamelli RL, et al. A subset of 26S proteasomes is activated at critically low ATP concentrations and contributes to myocardial injury during cold ischemia. Biochem Biophys Res Commun. 2009; 390: 1136–1141. doi: 10.1016/j.bbrc.2009.10.067 PMID: 19944202
34. Baker TA, Geng Q, Romero J, Picken MM, Gamelli RL, Majetschak M, et al. Prolongation of myocardial viability by proteasome inhibition during hypothermic organ preservation. Biochem Biophys Res Commun. 2010; 401: 548–553. doi:10.1016/j.bbrc.2010.09.093 PMID: 20875792
35. Hohn TJ, Grune T. The proteasome and the degradation of oxidized proteins: Part III-Redox regulation of the proteasomal system. Redox Biol. 2014; 2: 388–394. doi:10.1016/j.redox.2013.12.029 PMID: 24563857
36. Davies KJ. Degradation of oxidized proteins by the 20S proteasome. Biochimie. 2001; 83: 301–310. PMID: 11295490
37. Cui Z, Scruggs SB, Gilda JE, Ping P, Gomes AV. Regulation of cardiac proteasomes by ubiquitination, SUMOylation, and beyond. J Mol Cell Cardiol. 2014; 71: 32–42. doi:10.1016/j.yjmcc.2013.10.008 PMID: 24140722
38. Chauhan D, Hideshima T, Anderson KC. Proteasome inhibition in multiple myeloma: therapeutic implication. Annu Rev Pharmacol Toxicol. 2005; 45: 465–476. PMID: 15822185
39. Henderson CJ, Aleo E, Fontanini A, Maestro R, Paroni G, Brancolini C. Caspase activation and apoptosis in response to proteasome inhibitors. Cell Death Differ. 2005; 12: 1240–1254. PMID: 16094404
40. Chauhan D, Tian Z, Zhou B, Kuhn D, Orlowski R, Raje N, et al. In vitro and in vivo selective antitumor activity of a novel orally bioavailable proteasome inhibitor MLN9708 against multiple myeloma cells. Clin Cancer Res. 2011; 17: 5311–5321. doi:10.1158/1078-0432.CCR-11-0476 PMID: 21724551
41. Wong E, Cuervo AM. Integration of clearance mechanisms: the proteasome and autophagy. Cold Spring Harb Perspect Biol. 2010; 2: a006734. doi:10.1101/cshperspect.a006734 PMID: 21068151
42. Pedrozo Z, Torrealba N, Fernandez C, Gatica D, Toro B, Quiroga C, et al. Cardiomyocyte ryanodine receptor degradation by chaperone-mediated autophagy. Cardiovasc Res. 2013; 98: 277–285. doi:10.1093/cvr/cvt029 PMID: 23404999
43. Whelan RS, Kaplinskiy V, Kitisis RN. Cell death in the pathogenesis of heart disease: mechanisms and significance. Annu Rev Physiol. 2010; 72: 19–44. doi:10.1146/annurev.physiol.010908.163111 PMID: 20148665 | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2689,
1
],
[
2689,
6519,
2
],
[
6519,
9804,
3
],
[
9804,
13202,
4
],
[
13202,
14125,
5
],
[
14125,
15173,
6
],
[
15173,
19355,
7
],
[
19355,
20517,
8
],
[
20517,
22457,
9
],
[
22457,
24672,
10
],
[
24672,
28561,
11
],
[
28561,
31903,
12
],
[
31903,
36006,
13
],
[
36006,
38115,
14
]
]
} |
e838204f59ed9495545726d5345fbf1a04d67e4f | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 6,
"total-fallback-pages": 0,
"total-input-tokens": 16683,
"total-output-tokens": 4071,
"fieldofstudy": [
"Engineering"
]
} | Impact of undissolved gas on the performance characteristics of hydraulic fluids and characteristic of hydraulic system
To cite this article: A S Lunev et al 2020 J. Phys.: Conf. Ser. 1515 022013
View the article online for updates and enhancements.
Impact of undissolved gas on the performance characteristics of hydraulic fluids and characteristic of hydraulic system
A S Lunev, A A Nikitin, V A Ionova, A V Novik and V A Kramarenko
Siberian Federal University, 79, Svobodny Avenu, Krasnoyarsk, 660041, Russia
E-mail: [email protected]
Abstract. Almost in all cases, the hydraulic fluid contains not only the liquid phase of the used oil but also by the gaseous phase, which is often air. Air and oil form a while tanking and pressure drop in particular areas of fluid flow. It can be during dynamic processes due to different rates of gas dissolution and evolution from the fluid. The value of gas compressibility is much bigger than the fluid one, it is necessary to investigate the impact of the gas content of hydraulic fluids on the operation of hydraulic systems. This paper considers analyzing the impact of the gas content of oils on characteristics of hydraulic devices and the properties of hydraulic fluids and the degree of their variation with the amount of undissolved air.
1. Introduction
One of the properties of hydraulic fluids used as energy carriers in hydraulic systems is the solubility gas. It means that gas molecules are in the intermolecular fluid space. Such the gas is dissolved, unlike the entrained gas that has a form of bubbles in a fluid.
The impact of dissolved gas on the operation of hydraulic devices is pervasive in apparatus having significant volumes of the hydraulic fluid with the non-permanent pressure and temperature. Such apparatus are actuators and connecting pipelines in electro-hydraulic mechanisms.
The value of $c_1$ is the air solubility of the pressure equal to 1 atm. In this case, the air solubility for mineral oils and hydraulic fluid of 1 atm depends on the specific gravity of the oil:
$$c_1 \cdot 10^5 = 45,11 - 41,64 \cdot \gamma,$$
where $\gamma$ – the specific gravity of the oil.
This equation above permits to approximately find the air solubility for mineral oils if their specific gravity is foregone. It represents the impact of density that is one of the main parameters of mineral oil on the air solubility. The density increasing, air solubility of mineral oil decreases.
Hydraulic devices sometimes have areas where the hydraulic fluid is tangent to air and is entrained into the hydraulic system, for example, in an intensifier with a jet pipe. Entrained air under pressure begins to dissolve in the hydraulic fluid. That is what increases the content of dissolved air in the fluid. However, if there are no areas of direct contact of the hydraulic fluid and air during the operation of the hydraulic system, the hydraulic fluid contains the dissolved air even before filling the system under ordinary conditions and atmospheric pressure.
The content of undissolved gas adversely affects the reliability of the operational properties of the hydraulic fluids and the parameters of hydraulic equipment. Thereby the performance of the hydraulic
system reduces. Figure 1 shows the classification of the impact of the gas factor on the properties of mineral oils and the parameters of hydraulic systems.
.
**Figure 1.** The classification of the impact of the gas factor on the properties of mineral oils and hydraulic system parameters.
2. **The impact of the gas factor on the characteristics of hydraulic fluids**
Investigative and field researches [1] show that the process of fluid degassing is very slowly. The content of undissolved air in the hydraulic fluid significantly increases the surface used to oxygen diffusion. The result of oxidation is the increase of the oil viscosity [2, 3, 4], buildup acidic products of deterioration, such as hydroxy acids, resins, asphaltenes that decays its anti-corrosive ability. The lubrication performance and demulsibility get worse. Papers [2, 5, 6] tell that the resinous oxidation product precipitates. These cause the increased wear of the friction couple, jamming of the spool-type
couple, siltation of throttling trench, and a thrashing of filter elements [7]. The presence of air bubbles causes the dangerous of the probability of a “diesel effect” [2, 8, 9, 10] in the medium under isentropic compression.
The fluid having the p undissolved gas in dispersed form, its viscosity increases. For example, mineral oil containing 10% air has a viscosity of 15% higher than the pure one [11]. In some cases, it impacts on the value of the hydrodynamic load-carrying capacity of the oil film. Air bubbles disrupt oil film that leads to an extension of the space of metal-to-metal contact and increased wear [6]. The presence of a dispersed gas-air factor in the hydraulic fluid and especially the accumulation of it in the form of lumped volumes in the dead ends of the hydraulic systems adversely affect the heat exchange process, so the heat air conduction is much lower than the oil one [6]. The presence of air bubbles in the fluid reduces the bulk modulus of the compressible medium. Thus, one percent of undissolved air decays the bulk modulus of the mixture by 40% [12].
The following equation gives bulk modulus of the mixture
\[ B_{\text{mix}} = -V_{\text{mix}} \frac{dp}{dV_{\text{mix}}} \]
(2)
where \( B_{\text{mix}} \) – the bulk elastic modulus of the mixture under unrestricted pressure \( p \); \( V_{\text{mix}} \) – the volume of the mixture under unrestricted pressure \( p \); \( dp \) – an infinitesimal increment of pressure; \( dV_{\text{mix}} \) – an infinitesimal increment of the volume of a mixture.
The maximal change in the bulk mineral oil modulus due to the content of free gas occurs under the pressure of less than 5 MPa [13].
An increase in the compressibility of the medium causes a deterioration in its function of the link transferring pressure energy. It is an undesirable event for hydraulic self-control systems requiring high accuracy and operation speed. Thus, researches [14] of the dependence of the response time and excessive correction on a unit disturbance at the inlet of the hydraulic booster with and without air in cavities show that the presence of undissolved air in the hydraulic fluid increases the excessive correction to 20% and the response time to 30%. The undissolved air can also lead to a decrease in the stability of the hydraulic system, in particular, the follower system.
The formula [12] determines the volume content of gas under atmospheric pressure \( p_0 \):
\[ \alpha_g = V_{g,0} / V_{\text{mix},0} \]
(3)
where \( \alpha_g \) – the volume gas content; \( V_{\text{mix},0} \) – the volume of the mixture under atmospheric pressure \( p_0 \); \( V_{g,0} \) – the volume of gas under atmospheric pressure \( p_0 \);
\[ V_{\text{mix},0} = V_{l,0} + V_{g,0} \]
(4)
where \( V_{l,0} \) – the volume of liquid under atmospheric pressure \( p_0 \).
Then the following formula characterizes the ratio of the volume of the liquid component to the volume of the mixture under atmospheric pressure \( p_0 \):
\[ \frac{V_{l,0}}{V_{\text{mix},0}} = 1 - \alpha_g. \]
(5)
3. The impact of gas content on the operation of hydraulic devices
The presence of an undissolved gas-air component in the hydraulic fluid reduces the performance of hydraulic equipment. For example, the compressing time of the hydraulic press arrangement increases sharply when the elastic modulus of the fluid decreases due to an enhance in gas content [15]. The compressibility of the fluid filling the work volume of the hydraulic actuator requires additional energy consumption for its compression. The working volume of the hydraulic motor and the speed of its operation increasing, the power loss increases. For example, while using a hydraulic fluid with a bulk modulus of 1406 MPa in a drive with a response frequency of 100 Hz and a hydraulic cylinder
with a volume of 164 cm³, the power loss will be about 5 kW for every 5.8 cm² of piston area under a system pressure of 21 MPa. The hydraulic system operating under pressure equal to 21 MPa with a flow rate of 15 lpm and a power of 5 kW cannot move the load with a frequency of 100 Hz because it needs to take all power for compressing of the fluid [16].
The undissolved gas in the hydraulic fluid enhances the high-frequency pressure vibration in the duct of the hydraulic equipment and pipelines. That leads to an increase in the noise of the hydraulic system. According to the data given in [17–18], the presence of 2–3% of undissolved air can increase the noise by up to 10 dBA. The tests conducting with a pumping unit based on a pump with a valve distributor show a decrease in the noise level by 3 dBA after fluid degassing.
4. Gas evolution from the hydraulic fluid
It is experimentally proved [19] that the presence of a dissolved gas component does not change the elastic properties of liquids. However, the dissolved phase may cause such an unpleasant phenomenon as cavitation.
The evolution of a gas-air component in the increasing space of the pumps (gas cavitation) leads to the underfilling of the operating chambers and a decrease in displacement to its starvation [20, 21]. It is estimated that when the content of the undissolved gas factor is 5%, the delivery coefficient under a pressure of 20 MPa decreases by about 10% [16]. The gas amount released from the hydraulic fluid depends on three things: the fluid type, the rate among the fluid flow pressure and the bubble point pressure of the dissolved gas, and temperature. So intensive air release in the AMG-10 oil (aviation hydraulic fluid) leading to pump starvation starts at a temperature of 18 °C when the pressure in the stream decreases to a value equal to half the pressure of saturation of the liquid with air. The temperature of the fluid increased, the critical pressure in the flow increases. It is already 75% of the saturation pressure at 60 °C [20].
The phenomena of gas cavitation in the throttling elements change the flow structure and pressure distribution in the wetted part. It can significantly affect the capacity of the throttle gap and the forces acting on the shut-off-and-regulating element of the hydraulic unit. The phenomena lead to instability of the flow characteristics and the blockage effect of the throttling element [22-23]. The blockage effect is that the fluid flow through the throttling element stops to increase together with enhancing pressure drop under constant pressure at the inlet of the throttling element. The minimum value of the pressure drop is a critical one. It is because the fluid flow velocity cannot exceed the propagation velocity of disturbances in it (local sound velocity). The propagation velocity can be lower than the sound one in the air for certain combinations of the minimum pressure in the throttling zone and the volume content of undissolved gas [24, 25].
It is found that the process of gas evolution from the fluid (“capillary” gas) can accompany the flow of hydraulics fluids through capillary channels under certain conditions. This process causes passages to plug. This phenomenon called the filtration effect can account for a decrease in the efficiency of filter made of foam materials [26].
5. Conclusions
Therefore, one of the significant factors, which negatively affect the performance of the hydraulic system elements, its lifetime, and durability, is the presence of a gas-air factor in the hydraulic fluid in the dissolved and, especially, undissolved phase.
References
[1] Gallant H 1962 Untersuchungen von kavitations blasen Ostereichische Ingenieur-Zeitschrift 5(3) 74-83
[2] Backe W 1976 Influence of dispersed air on the pressure medium Conference on Contamination in Fluid Power Systems 77-84
[3] Cibula G 1966 Die oxydation bei schmierolen Mineraloltechnik 11 8-9
[4] Vorberg K 1969 Druckflussigkeiten in hydraulikanlagen Maschinenmarka 75(40) 834-8
[5] Thoenes H W 1976 Zum einflug von luft und wasser auf die leitungsfahigkeit von druckubertragungsmedium und von hydraulikanlagen Industrie Anzeiger 98(51) 888-91
[6] Kazanskij V N 1974 Steam Turbine Lubrication Systems (Moscow: Energy)
[7] Avrunin G A 1982 Improving the reliability and durability of hydraulic equipment and mineral oils when using hydrodynamic dispersants in hydraulic systems (Moscow: Research institute Mash)
[8] Lohrentz H J 1968 Die entwicklung extrem hoher temperatur in hydraulik systemen und die einflusse dieser temperatur auf die bauteil und ihre funktion Mineraloltechnik 13 14
[9] Lipphardt P 1976 Kompression von dispergieter luft in hydrauliksystemen und deren auswirkungen auf das druckubertragungsmedium Industrie Anzeiger 98(51) 883-7
[10] Beyer R 1981 Druckflussigkeiten Q+P 25(21) 37-9
[11] Bashta T M 1971 Engineering hydraulics (Moscow: Machine building)
[12] Popov D N 1977 Dynamics and Regulation of Hydraulic and Pneumatic Systems (Moscow: Machine building)
[13] Hohlov V A, Prokof'ev V N and Borisova N A 1971 Electro-Hydraulic Tracking Systems (Moscow: Machine building)
[14] Kuznecov V N and Golovko Y S 1970 The effect of undissolved air on the transient hydraulic amplifier Technology and organization of production 4 74-5
[15] Belopuhov A K 1971 Injection Molding. Prepress Problems (Moscow: Machine building)
[16] Magorien V J 1969 What is bulk modulus and when is it important? H+P 98-100
[17] Maas J 1969 Neue wege in der entwicklung hydrostatischer antriebe zu wirksamer larmbekampfung und hohenen wirkungsgrad dur verminderung der hysteresisverluste 0+P vol 13(11) 531
[18] Turovskij Z G 1973 Cavitation Research in Piston Pumps (Kiev: Book)
[19] Prokof'ev V N 1968 An experimental study of the elastic properties of two-phase working fluids of volumetric hydraulic actuators Machine building 2 87-93
[20] Glazkov M M 1987 Cavitation in aircraft liquid systems (Kiev: Book)
[21] Bashkirov V S, Dudkov Y N and Fedin V I 1977 Methods of experimental study of gas generation during unsteady fluid motion in the lines of volumetric hydraulic drives Hydraulic Drive and Control Systems 137-42
[22] Abarinova I A, Efimceva N F and Piskunov Y A 1980 Features of the throttling elements of hydraulic systems on a two-phase flow of a working fluid 198th All-Union Conf. Scientific and Tech. Progress in Mech. Engineering and Instrument Engineering 55-7
[23] Backe W and Benning P 1962 Uber kavitation - serscheinungen in querschnittsverengungen von sopr hydraulischen systemen Industrie Anzeiger 63 29-36
[24] Martin K S and Padmanabhan M 1979 Pressure pulse propagation in a two-component slug flow Theoretical Foundations of Engineering Calculations 1 16-71
[25] Popov D N 1982 Non-Stationary Hydromechanical Processes (Moscow: Machine building)
[26] Timirkheev R G and Sapozhnikov V M 1986 Industrial Cleanliness and Fine Filtration of Aircraft Operating Fluids (Moscow: Machine building) | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
252,
1
],
[
252,
3230,
2
],
[
3230,
4241,
3
],
[
4241,
8074,
4
],
[
8074,
12101,
5
],
[
12101,
15059,
6
]
]
} |
b7fbcb350d05be229c3ca4bf4a43ceb48fdf0d8e | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 12,
"total-fallback-pages": 0,
"total-input-tokens": 41243,
"total-output-tokens": 14876,
"fieldofstudy": [
"Biology"
]
} | Lysosomal Storage Disorders Shed Light on Lysosomal Dysfunction in Parkinson’s Disease
Shani Blumenreich 1, Or B. Barav 1, Bethan J. Jenkins 1,2 and Anthony H. Futerman 1,* †
1 Department of Biomolecular Sciences, Weizmann Institute of Science, Rehovot 76100, Israel; [email protected] (S.B.); [email protected] (O.B.B.); [email protected] (B.J.J.)
2 Department of Neurobiology, Max Planck Institute of Neurobiology, 82152 Planegg, Germany
* Correspondence: [email protected]; Tel.: +972-8-9342704; Fax: +972-8-9344112
† The Joseph Meyerhoff Professor of Biochemistry at the Weizmann Institute of Science.
Abstract: The lysosome is a central player in the cell, acting as a clearing house for macromolecular degradation, but also plays a critical role in a variety of additional metabolic and regulatory processes. The lysosome has recently attracted the attention of neurobiologists and neurologists since a number of neurological diseases involve a lysosomal component. Among these is Parkinson’s disease (PD). While heterozygous and homozygous mutations in GBA1 are the highest genetic risk factor for PD, studies performed over the past decade have suggested that lysosomal loss of function is likely involved in PD pathology, since a significant percent of PD patients have a mutation in one or more genes that cause a lysosomal storage disease (LSD). Although the mechanistic connection between the lysosome and PD remains somewhat enigmatic, significant evidence is accumulating that lysosomal dysfunction plays a central role in PD pathophysiology. Thus, lysosomal dysfunction, resulting from mutations in lysosomal genes, may enhance the accumulation of α-synuclein in the brain, which may result in the earlier development of PD.
Keywords: lysosomal storage diseases; Parkinson’s disease; Gaucher disease; α-synuclein
1. Introduction
The lysosome is a membrane-bound organelle first described by Christian de Duve in the 1950s [1,2]. In addition to its well-characterized role in macromolecular degradation [3,4], the lysosome plays pivotal roles in other aspects of cell homeostasis and is crucial to numerous physiological processes. Subsequent to the discovery of the lysosome, a family of diseases was shown to be associated with the defective activity of lysosomal proteins. These diseases, known as lysosomal storage diseases (LSDs), are typically classified according to the substrate that accumulates [5]. To date, about 70 genetically distinct conditions causing different LSDs have been described, although more are likely based on the number of known lysosomal proteins.
Mutations in one of these lysosomal genes, namely GBA1, result in Gaucher disease (GD) [6], and mutations in GBA1 are now recognized as the highest genetic risk factor for Parkinson’s disease (PD) [7]. However, the mechanistic association between GBA1 mutations and PD is unclear, with some favoring the “gain of function” hypothesis (i.e., a mutated lysosomal protein gains a new function when mutated) and others favoring the “loss of function” hypothesis (i.e., one or another lysosomal function is compromised [8]). We are of the opinion that most of the evidence [9] is consistent with the loss-of-function hypothesis, which is supported by a recent whole exome sequencing study on PD patients [10].
In this sequencing study [10], 54 genes associated with LSDs were analyzed (Table 1). Remarkably, the majority of PD patients (about 56%) displayed at least one mutation in a gene
that causes an LSD, and 21% had two or more, suggesting that a combination of mutations in genes encoding proteins that cause an LSD could contribute to lysosomal dysfunction, and thereby increase PD susceptibility. The prevalence of LSD mutations in PD patients strengthens the idea that lysosomal dysfunction is a key player in PD pathogenesis and identifies a burden of LSD variants associated with PD. This study also revealed new susceptibility loci. While these results appear to be supportive of an association between lysosomal genes and PD, considerable additional work is needed to cement this relationship. For instance, the Human Lysosome Gene Database contains about 400 lysosomal proteins (http://lysosome.unipg.it) [11] and a recent proteomics study suggested 343 unique lysosomal proteins [12]. These numbers are significantly more than the roughly 50 proteins currently known to be associated with an LSD, and that were analyzed in the whole exome sequencing study [10]. Might some of these other lysosomal genes also be associated with LSDs, even though not all of the approximately 400 proteins in this database are likely to be bonafide lysosomal proteins?
Table 1. Genes that cause lysosomal storage diseases (LSDs) and the percent of mutated variants compared to the number of tested variants. Data were obtained by whole exome sequencing of PD patients. The analysis in this study targeted several variants from each LSD gene, looking for an association with PD. Variants that were significantly associated with PD are given as the percentage of total variants examined for each gene. The study was performed under strict conditions, only using patients with a common European ancestry and quality-controlled samples. Moreover, subjects were excluded with mutations in well-established Mendelian genes known to cause PD. Adapted from [10].
| LSD | Gene | Mutated Variants (%) |
|------------------------------------------|-------|----------------------|
| Action mycolonus-renal failure syndrome | SCARB2| 70.0 |
| Alpha-mannosidosis | MAN2B1| 91.7 |
| Aspartylglucosaminuria | AGA | 33.3 |
| Beta-mannosidosis | MANBA | 83.3 |
| Cystinosis | CTNS | 92.3 |
| Danon disease | LAMP2 | 77.8 |
| Fabry disease | GLA | 77.8 |
| Faber Lipogranulomatosis | ASAH1 | 85.0 |
| Fucosidosis | FUCA1 | 80.0 |
| Galactosialidosis | CTSA | 78.6 |
| Gaucher disease | GBA | 82.1 |
| GM1-gangliosidosis/Morquio B | GLB1 | 50.0 |
| GM2-gangliosidosis | GM2A | 100.0 |
| Hunter syndrome | IDS | 88.9 |
| Hurler syndrome | IDUA | 50.0 |
| I-Cell disease | GNPTAB| 79.5 |
| Krabbe disease | GALT | 83.3 |
| Kufer-Rakeb syndrome | ATP13A2| 75.0 |
| Maroteaux–Lamy disease | ARSB | 90.9 |
| Metachromatic leukodystrophy | ARSA | 100.0 |
| Morquio A disease | GALNS | 63.6 |
| Mucolipidosis type IV | MCOLN1| 73.7 |
Just under ten years ago [9], we documented the prevalence of PD features in LSD patients and in cellular and animal models, and came to the somewhat surprising conclusion, at least at that time, that there was indeed an association between LSDs and PD rather than just between GD and PD. We suggested that additional genetic, epidemiological, and clinical studies should be performed to check the precise incidence of mutations in genes encoding lysosomal proteins in patients displaying PD symptoms. In the current review, we update studies performed in the last decade or so that lend further support to the association between the lysosome and PD, and then briefly discuss progress on understanding the mechanistic relationship between the lysosome and PD.
2. Associations Between LSDs and PD
LSDs are monogenic diseases that are mainly inherited in a recessive manner. Individually, LSDs are rare, but collectively are estimated to occur in as many as 1:5000 live births [13]. However, due to their low individual prevalence, early age of onset and death in many cases, limited pathophysiological
information as well as limited information on natural history, especially in late-onset forms, is available for many of the diseases. Thus, in some cases, it is difficult to make a compelling argument for an association with PD. Having said that, our earlier review [9] documented that PD was detected in LSD carriers and patients, as well as in relatives of LSD patients. Additionally, in some LSD animal models, PD features such as substantia nigra (SN) pathology, Lewy body formation or α-synuclein aggregation, ubiquitinated protein aggregates, and/or down-regulation of UCH-L1 were observed [9]. We will now discuss the significant progress that has been made in the last few years that supports a clear association between three of the most common LSDs and PD.
2.1. Gaucher Disease
Mutations in GBA1, which encodes acid-β-glucosidase (GCase), are the highest genetic risk factor for PD [7]. A large multi-center study showed that by the age of 80, GD patients have a 9.1% risk of developing PD, and GBA1 carriers have a risk of 7.7% [14], consistent with an earlier study [7]. Furthermore, the severity of GBA1 mutations, determined by their strong association with the presence or absence of a GD neuronopathic phenotype [15], correlates with the risk and severity of PD [16–19]. PD patients with a GBA1 mutation leading to a severe neurological form of GD displayed worse motor and non-motor symptoms than GBA1 mutations that lead to a milder form of GD in both heterozygotes and homozygotes [19]. However, results have not been consistent, with some suggesting that PD patients with or without a GBA1 mutation are clinically and cognitively heterogeneous [20], though some studies were limited by the number of available patients and the kind of mutations examined. Numerous other studies have attempted to address both the genetic and mechanistic relationship between GBA1 and PD (see, for instance, [21–26]).
2.2. Niemann–Pick Disease
Significant advances in delineating the relationship between mutations in the SMPD1 gene, which causes Niemann–Pick disease types A and B [27], and PD have been reported in the past few years, with SMPD1 repeatedly identified as a genetic risk factor for PD [28,29]. Thus, 3.1% of Jewish Ashkenazi PD patients were SMPD1 carriers [30], and an association between SMPD1 and sporadic PD was also reported in Chinese patients [31]. Similar to GD, different SMPD1 mutations may influence the risk and the course of PD; thus, patients with the L302P mutation have a greater chance of developing PD than those with the R496L mutation [29]. Although there are no documented cases of PD in Niemann–Pick disease, parkinsonian phenotypes were reported in a 9-month-old Niemann–Pick type A patient with tremors on one side of her body [32]. Niemann–Pick type C disease was originally classified as a similar disease to Niemann–Pick types A and B, although it is now known to be caused by mutations in two completely different genes, namely NPC1 and NPC2 [33]. Two adult heterozygous carriers with mutations in NPC1 have been reported with PD and a further heterozygous individual has been identified with Parkinsonism [34].
2.3. Fabry Disease
Fabry disease is an X-linked disorder caused by mutations in GLA, which encodes α-galactosidase A (α-Gal). Several studies suggesting an association between Fabry disease and PD have recently been published. In the first, α-Gal activity was about 10% lower in blood spots from PD patients [35]. This reduction was seen in all types of PD patients, including idiopathic PD and patients with LRRK2 or GBA1 mutations, suggesting that the reduced activity of α-Gal in PD patients is not affected by genetic risk factors. A second study revealed that α-Gal levels and activity were reduced in the temporal cortex in late-stage PD, which correlated with elevated α-synuclein levels [36]. A third study demonstrated reduced α-Gal activity along with the reduced activity of multiple lysosomal hydrolases in the SN of PD patients [37]. A further study revealed that 8.3% of Fabry patients over the age of 60 were diagnosed with PD and 7.4% of the families of these patients had a close relative that matched the criteria for a PD diagnosis [38]; however, the latter was an online survey with no
direct clinical examination of PD. In addition, α-Gal activity was significantly lower in the serum of PD patients compared to parkinsonian syndrome [39]. Finally, reduced nigral volume (suggesting neurodegeneration in this region) correlating with increased susceptibility of this region was observed in Fabry disease patients [40].
2.4. Other LSDs
A number of other LSD genes were identified in a genome-wide association study (GWAS) of PD patients, including GALC [41] (Krabbe disease) and IDUA [42] (mucopolysaccharidoses I, also known as Hurler syndrome). Moreover, focused candidate gene studies were performed to determine the association between PD and SCARB2 (myoclonus–renal failure) with polymorphisms in the SCARB2 locus associated with the risk of developing PD [43,44], although another study suggested that SCARB2 does not confer a significant risk for PD [45]. A retrospective study demonstrated that clinical symptoms in three families with ARSA haploinsufficiency [46] (metachromatic leukodystrophy, or MLD) were similar to those described in PD patients with GBA1 mutations. Mutations in one of the genes causing neuronal ceroid lipofuscinoses, ATP13A2 (also known as PARKA), were also shown to cause a form of early-onset Parkinsonism with pyramidal degeneration and dementia [43,47].
In summary, most of the data available in 2011 [9] was based on clinical and pathological observations, while genetic associations between LSD genes and PD were limited (except for GBA1). Since then, a significant amount of genetic, clinical, and pathological data has accrued, indicating that the connection between LSDs and PD is significantly tighter than in 2011, strengthening the argument that a detailed understanding of lysosomal (dys)function in PD is of critical importance to delineating the mechanistic link between the lysosome and PD.
3. Lysosomal Dysfunction and PD
A unified hypothesis explaining the mechanistic association between mutations in lysosomal genes and PD is currently lacking. While we favor the loss-of-function paradigm, the appearance of endoplasmic reticulum stress and of the unfolded protein response in some LSDs [48–52] and in PD [53–56] lends some credence to the gain-of-function hypothesis (Figure 1, pathway B). In reality, both loss and gain of function probably contribute to the association between the lysosome and PD.
Lysosomal dysfunction, although a rather loose term used mainly to describe the inappropriate execution of lysosomal function, occurs due to mutations in lysosomal proteins or to alterations in lysosomal acidification (Figure 1, pathway A). Since most LSDs are recessive and carriers do not display overt LSD symptoms, it is difficult to explain why mutations in one allele encoding for a lysosomal protein increase the risk of PD [7]. A consensus is emerging that general impairment of lysosomal function could occur over the long lifespan of an individual and may be exacerbated in the case of carriers. This is supported by data showing that GCase activity decreases with age in normal mouse brains, and that glucosylsphingosine (GlcSph) levels increase with age in normal mice [57]. Similar changes are seen in the SN and hippocampus of sporadic PD patients without GBA1 mutations [58], along with a reduction in the activity of GCase, α-mannosidase, β-mannosidase, and β-hexosaminidase in the cerebrospinal fluid of PD patients [59]. A recent study also demonstrated a reduction in GCase activity in the SN of PD patients, in addition to substrate accumulation, suggesting it is not the result of neuronal death [37].
There is only very limited information examining lipid accumulation in LSD carriers. In one study [60], the accumulation of ceramides and sphingolipids was observed in Lewy body dementia patients carrying a GBA1 mutation compared with controls, although other studies [61,62] suggested no lipid accumulation in LSD carriers. Clearly, further systematic studies on LSD carriers are required to evaluate whether substrate levels are altered. Since carriers do not present disease symptoms, it is close to impossible to obtain human brain tissue, although such tissues can be easily obtained from animal models of the relevant diseases.
Figure 1. Possible pathophysiological pathways for exacerbation of Parkinson’s disease (PD) symptoms due to mutations in LSD-causing genes. Lysosomal dysfunction may be a key player in PD pathogenesis and could be triggered or exacerbated due to LSD gene mutations. With the lysosome being involved in numerous cellular processes, lysosomal dysfunction could explain some of the symptoms observed in PD. LSDs; lysosomal storage diseases, ER; endoplasmic reticulum, UPR; unfolded protein response, ROS; reactive oxygen species. See text for more details.
Irrespective of the extent of substrate accumulation in LSD carriers, changes in lysosomal function could lead to inappropriate clearance of proteins, including α-synuclein (Figure 1, pathway A1), and a crosstalk between α-synuclein and lysosomal enzyme levels has been described [63–65]. Presumably, α-synuclein accumulation with age due to a reduction in lysosomal function may be exacerbated in LSD carriers or in individuals with multiple haploid mutations [10] due to accumulative damage, perhaps explaining the earlier onset of PD.
Evidence for lysosomal dysfunction in PD is also accumulating [66]. Thus, a reduction in some lysosomal markers in the SN was observed in some early studies in PD patients [67,68], as was the number of lysosomes [68]. Also, the selective reduction of LAMP2 and GCase in regions accumulating α-synuclein in the early stages of PD was observed, suggesting altered lysosomal function, including alterations in chaperone-mediated autophagy pathways [69]. Several familial PD-related genes are strongly linked to endo-lysosomal and autophagic pathways [70–72]. Treatment of neuroblastoma cells with the neurotoxin 1-methyl-4-phenylpyridinium (MPP+) resulted in reduced lysosomal markers and accumulation of autophagosomes [68]. Impairment of lysosomal degradation might also explain why genomic variants elevate α-synuclein levels, such as mutations causing SNCA multiplication or mutations causing the enhancement of its promoter [73,74]. There is also some evidence that lysosomal integrity is altered in PD and may cause the seeding of α-synuclein aggregates [67,68,75]. Clearly, loss of lysosomal function, either in PD or as the result of an LSD, will impact a variety of up- and downstream pathways, such as autophagy, which is indeed impaired in 14 different LSDs [76], and mitophagy [77] (Figure 1, pathway A1a), which would exacerbate mitochondrial dysfunction. The mitochondrial dysfunction, observed in a number of LSDs [78], might be due to the strong reciprocal relationship
between the lysosome and mitochondria [79]. Finally, different neuronal populations might be more or less sensitive to lysosomal dysfunction; thus, selective loss of GCase activity was observed in the SN and caudate, but not in the frontal cortex, hippocampus, cerebellum, or putamen of PD patients [80]. The basal activity of lysosomal enzymes (in both the control and patients) was generally higher in the SN and hippocampus than in six other brain regions [80].
As a result of lysosomal dysfunction, cellular lipid composition may also be altered. A variety of lipids, including sphingolipids, cholesterol, and fatty acids interact directly or indirectly with α-synuclein [81] (Figure 1, pathway A3a). In PD, α-synuclein is converted from a soluble protein into insoluble amyloid-like fibrils [82], which might be mediated by interactions with these lipids. Thus, when human midbrain neurons were treated with conduritol B-epoxide (CBE, a chemical inhibitor of GCase), a reversible conformational change in α-synuclein was observed [83]. Liposomes containing gangliosides GM1 and GM2 and glucosylceramide (GlcCer) induce a catalytic environment for nucleation of α-synuclein aggregation [84], and some of these gangliosides accumulate as secondary metabolites in some LSDs [85]. The effect of lipid composition is not limited to interactions with α-synuclein, since neurotransmission is also affected by membrane lipid composition (Figure 1, pathway A3b); as a result, lipid accumulation in LSD patients and carriers may alter neurotransmission and affect both motor and non-motor symptoms, such as the reduction in exocytosis observed in mucopolysaccharidosis IIIA mice [86]. Moreover, some neurotransmitters use membrane-dependent mechanisms for their uptake and thus might be dependent on membrane composition for their normal function [87,88].
Another critical pathway affected in both LSDs and PD is calcium homeostasis (Figure 1, pathway A4). Induced pluripotent stem cell (iPSC)-derived neurons from PD patients carrying a GBA1 mutation (PD-GBA) and GD patients show impaired calcium homeostasis upon stress stimulation, in addition to impaired autophagy [89]. Moreover, a reduction in lysosomal calcium store content in PD and PD-GBA fibroblasts, as well as disturbances in lysosomal morphology [90], have been observed. Altered calcium homeostasis is observed in a number of LSDs (reviewed in [91]). Since calcium is an important player in many neuronal events, it may be a critical player in the relationship between PD and LSDs.
In summary, it is likely that the lysosome acts as a central protective hub against α-synuclein toxicity, autophagy impairment, altered neurotransmission, and alterations in calcium homeostasis. Therefore, any impairment of this pathway, such as that which occurs over the long lifetime of an LSD carrier, could increase the risk of developing PD.
4. Concluding Remarks
The notion that lysosomal processes play a central role in PD pathophysiology is now gaining momentum. The lysosome is crucial for various cellular processes, and, upon disruption, lysosomal dysfunction is likely to lead to α-synuclein accumulation (Figure 1). This, together with the discovery that mutations in GBA1 are the highest genetic risk factor for PD [7], instigated the search for the mechanistic connection between PD and various LSDs [9].
The majority of clinical studies on the relationship between PD and LSDs were performed on carriers rather than LSD patients, due to the low individual prevalence of LSDs and their early age of death. In LSD carriers, lysosomal dysfunction could be exacerbated over the years which, along with a reduction in lysosomal activity [57,58], could culminate in α-synuclein accumulation (Figure 1). Assuming that α-synuclein causes PD, LSD carriers and patients are more likely to develop early onset PD, depending on whether they accumulate α-synuclein or not (and other genetic and environmental factors). We further suggest that lysosomal dysfunction could explain both motor and non-motor observations in LSD-related PD (Figure 1). Indeed, we recently documented evidence for co-existence of early PD-related non-motor symptoms in LSDs (Blumenreich et al., in press). We thus speculate that lysosomal dysfunction due to LSD mutations may enhance the aggregation and spreading of α-synuclein, and therefore trigger the PD cascade.
Author Contributions: Writing—original draft preparation, S.B., O.B.B., B.J.J. and A.H.F.; writing—review and editing, S.B. and A.H.F.; visualization, S.B.; supervision, A.H.F. All authors have read and agreed to the published version of the manuscript.
Funding: Work in the Futerman laboratory on Parkinson’s disease is supported by the Legacy Heritage Biomedical Science Partnership Program of the Israel Science Foundation (grant # 2240/17), by the Rolf Wiklund and Alice Wiklund Parkinson’s Disease Research Fund, the Children’s Gaucher Research Fund, and the Michael J. Fox Foundation.
Conflicts of Interest: The authors declare no conflicts of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
Abbreviations
GCase Acid-β-glucosidase
GD Gaucher disease
LSDs Lysosomal storage diseases
PD Parkinson’s disease
PD-GBA PD patient with a GBA1 mutation
SN Substantia nigra
References
1. de Duve, C. Exploring cells with a centrifuge. Science 1975, 189, 186–194. [CrossRef]
2. de Duve, C. The lysosome turns fifty. Nat. Cell Biol. 2005, 7, 847–849. [CrossRef]
3. Saftig, P.; Klumperman, J. Lysosome biogenesis and lysosomal membrane proteins: Trafficking meets function. Nat. Rev. Mol. Cell Biol. 2009, 10, 623–635. [CrossRef]
4. Settembre, C.; Fraldi, A.; Medina, D.L.; Ballabio, A. Signals from the lysosome: A control centre for cellular clearance and energy metabolism. Nat. Rev. Mol. Cell Biol. 2013, 14, 283–296. [CrossRef] [PubMed]
5. Futerman, A.H.; van Meer, G. The cell biology of lysosomal storage disorders. Nat. Rev. Mol. Cell Biol. 2004, 5, 554–565. [CrossRef] [PubMed]
6. Futerman, A.H.; Zimran, A. Gaucher Disease, 1st ed.; Futerman, A., Zimran, A., Eds.; CRC Press: Boca Raton, FL, USA, 2006.
7. Sidransky, E.; Nalls, M.A.; Aasly, J.O.; Aharon-Peretz, J.; Annesi, G.; Barbosa, E.R.; Bar-Shira, A.; Berg, D.; Bras, J.; Brice, A.; et al. Multicenter analysis of glucocerebrosidase mutations in Parkinson’s disease. N. Engl. J. Med. 2009, 361, 1651–1661. [CrossRef]
8. Futerman, A.H.; Hardy, J. Perspective: Finding common ground. Nature 2016, 537, 160–161. [CrossRef] [PubMed]
9. Shachar, T.; Bianco, L.C.; Recchia, A.; Wiessner, C.; Raas-Rothschild, A.; Futerman, A.H. Lysosomal storage disorders and Parkinson’s disease: Gaucher disease and beyond. Mov. Disord. 2011, 26, 1593–1604. [CrossRef] [PubMed]
10. Robak, L.A.; Jansen, I.E.; van Rooij, J.; Uitterlinden, A.G.; Kraaij, R.; Jankovic, J.; International Parkinson’s Disease Genomics Consortium (IPDGC); Heutink, P.; Shulman, J.M. Excessive burden of lysosomal storage disorder gene variants in Parkinson’s disease. Brain 2017, 140, 3191–3203. [CrossRef]
11. Brozzi, A.; Urbanelli, L.; Germain, P.L.; Magini, A.; Emiliani, C. hLGDB: A database of human lysosomal genes and their regulation. Database (Oxford) 2013, 2013, bat024. [CrossRef]
12. Wyant, G.A.; Abu-Remaileh, M.; Frenkel, E.M.; Laqtom, N.N.; Dharamdasani, V.; Lewis, C.A.; Chan, S.H.; Heinze, I.; Ori, A.; Sabatini, D.M. NUFIP1 is a ribosome receptor for starvation-induced ribophagy. Science 2018, 360, 751–758. [CrossRef] [PubMed]
13. Meikle, P.J.; Brooks, D.A.; Ravenscroft, E.M.; Yan, M.; Williams, R.E.; Jaunzemis, A.E.; Chataway, T.K.; Karageorgos, L.E.; Davey, R.C.; Boulter, C.D.; et al. Diagnosis of lysosomal storage disorders: Evaluation of lysosome-associated membrane protein LAMP-1 as a diagnostic marker. Clin. Chem. 1997, 43, 1325–1335. [CrossRef] [PubMed]
14. Alcalay, R.N.; Dinur, T.; Quinn, T.; Sakanaka, K.; Levy, O.; Waters, C.; Fahn, S.; Dorovski, T.; Chung, W.K.; Pauciulo, M.; et al. Comparison of Parkinson risk in Ashkenazi Jewish patients with Gaucher disease and GBA heterozygotes. JAMA Neurol. 2014, 71, 752–757. [CrossRef] [PubMed]
15. Migdalska-Richards, A.; Schapira, A.H.V. The relationship between glucocerebrosidase mutations and Parkinson disease. *J. Neurochem.* 2016, 139, 77–90. [CrossRef] [PubMed]
16. Gan-Or, Z.; Amshalom, I.; Kilarski, L.L.; Bar-Shira, A.; Gana-Weisz, M.; Mirelman, A.; Marder, K.; Bressman, S.; Giladi, N.; Orr-Urtreger, A. Differential effects of severe vs mild GBA mutations on Parkinson disease. *Neurology* 2015, 84, 880–887. [CrossRef] [PubMed]
17. Cilia, R.; Tunesi, S.; Marotta, G.; Cereda, E.; Siri, C.; Tesei, S.; Zecchinelli, A.L.; Canesi, M.; Mariani, C.B.; Meucci, N.; et al. Survival and dementia in GBA-associated Parkinson’s disease: The mutation matters. *Ann. Neurol.* 2016, 80, 662–673. [CrossRef]
18. Liu, G.; Boot, B.; Locascio, J.J.; Jansen, I.E.; Rhodes, S.W.; Eberly, S.; Elbaz, A.; Brice, A.; Ravina, B.; van Hilten, J.J.; et al. Specifically neuropathic Gaucher’s mutations accelerate cognitive decline in Parkinson’s. *Ann. Neurol.* 2016, 80, 674–685. [CrossRef]
19. Thaler, A.; Bregman, N.; Gurevich, T.; Shiner, T.; Dror, Y.; Zmira, O.; Gan-Or, Z.; Bar-Shira, A.; Gana-Weisz, M.; Orr-Urtreger, A.; et al. Parkinson’s disease phenotype is influenced by the severity of the mutations in the GBA gene. *Parkinsonism Relat. Disord.* 2018, 55, 45–49. [CrossRef]
20. Collins, L.M.; Williams-Gray, C.H.; Morris, E.; Deegan, P.; Cox, T.M.; Barker, R.A. The motor and cognitive features of Parkinson’s disease in patients with concurrent Gaucher disease over 2 years: A case series. *J. Neurol.* 2018, 61, 1789–1794. [CrossRef]
21. Tayebi, N.; Parisiadou, L.; Berhe, B.; Gonzalez, A.N.; Serra-Vinardell, J.; Tamargo, R.J.; Maniwang, E.; Thaler, A.; Bregman, N.; Gurevich, T.; Shiner, T.; Dror, Y.; Zmira, O.; Gan-Or, Z.; Bar-Shira, A.; Gana-Weisz, M.; Mirelman, A.; Marder, K.; Bressman, S.; Chandra, S.S. Glucosylsphingosine Promotes α-Synuclein Pathology in Mutant GBA-Associated Parkinson’s Disease. *J. Neurosci.* 2017, 37, 9617–9631. [PubMed]
22. Ruskey, J.A.; Zhou, S.; Santiago, R.; Franche, L.-A.; Alam, A.; Ronci, J.; Keutzer, J.; Mistry, P.K.; Chandra, S.S. Glucocerebrosidase deficiency promotes protein aggregation through dysregulation of extracellular vesicles. *PLoS Genet.* 2018, 14, e1007694. [CrossRef] [PubMed]
23. Li, H.; Ham, A.; Ma, T.C.; Kuo, S.-H.; Kanter, E.; Kim, D.; Ko, H.S.; Quan, Y.; Sardi, S.P.; Li, A.; et al. Mitochondrial dysfunction and mitophagy defect triggered by heterozygous GBA mutations. *Autophagy* 2018, 15, 113–130. [CrossRef]
24. Arkadir, D.; Dinur, T.; Cohen, M.B.; Vilk, S.R.; Tiomkin, M.; Brüggemann, N.; Cozma, C.; Rolsf, A.; Zimran, A. Prodromal substantia nigra nigrostriatal dysfunction underlines suggested association between substrate accumulation and the risk for GBA-related Parkinson’s disease. *Eur. J. Neurosci.* 2019, 26, 1013–1018. [CrossRef]
25. Schuchman, E.H.; Wasserstein, M.P. Types A and B Niemann-Pick Disease. *Pediatr. Endocrinol. Rev.* 2016, 13, 674–681.
26. Gan-Or, Z.; Ozelius, L.J.; Bar-Shira, A.; Saunders-Pullman, R.; Mirelman, A.; Kornreich, R.; Gana-Weisz, M.; Raymond, D.; Rozenkrantz, L.; Deik, A.; et al. The p.L302P mutation in the lysosomal enzyme gene SMPD1 is a risk factor for Parkinson disease. *Neurology* 2013, 80, 1606–1610. [CrossRef]
27. Gan-Or, Z.; Orr-Urtreger, A.; Alcalay, R.N.; Bressman, S.; Giladi, N.; Rouleau, G.A. The emerging role of SMPD1 mutations in Parkinson’s disease: Implications for future studies. *Parkinsonism Relat. Disord.* 2015, 21, 1294–1295. [CrossRef]
28. Dagan, E.; Schlesinger, I.; Ayoub, M.; Mory, A.; Nassar, M.; Kurolap, A.; Peretz-Aharon, J.; Gershoni-Baruch, R. The contribution of Niemann-Pick SMPD1 mutations to Parkinson disease in Ashkenazi Jews. *Parkinsonism Relat. Disord.* 2015, 21, 1067–1071. [CrossRef]
29. Mao, C.-Y.; Yang, J.; Wang, H.; Zhang, S.-Y.; Yang, Z.-H.; Luo, H.-Y.; Li, F.; Shi, M.; Liu, Y.-T.; Zhuang, Z.-P.; et al. SMPD1 variants in Chinese Han patients with sporadic Parkinson’s disease. *Parkinsonism Relat. Disord.* 2017, 34, 59–61. [CrossRef]
30. Vykuntaraju, K.N.; Lokanatha, H. Shivananda Niemann-Pick disease type A presenting as unilateral tremors. *Indian Pediatr.* 2012, 49, 919–920. [CrossRef] [PubMed]
31. Evans, E.L.; Platt, F.M. Lipids on Trial: The Search for the Offending Metabolite in Niemann-Pick type C Disease. *Traffic* 2010, 11, 419–428. [CrossRef] [PubMed]
34. Kluenemann, H.H.; Nett, J.G.; Davis, M.Y.; Bird, T.D. Parkinsonism syndrome in heterozygotes for Niemann-Pick C1. *J. Neurol. Sci.* 2013, 335, 219–220. [CrossRef] [PubMed]
35. Alcalay, R.N.; Wolf, P.; Levy, O.A.; Kang, U.J.; Waters, C.; Fahn, S.; Ford, B.; Kuo, S.H.; Vanegas, N.; Shah, H.; et al. Alpha galactosidase A activity in Parkinson’s disease. *Neurobiol. Dis.* 2018, 112, 85–90. [CrossRef]
36. Nelson, M.P.; Boutin, M.; Tse, T.E.; Lu, H.; Haley, E.D.; Ouyang, X.; Zhang, J.; Auray-Blais, C.; Shacka, J.J. The lysosomal enzyme alpha-Galactosidase A is deficient in Parkinson’s disease brain in association with the pathologic accumulation of alpha-synuclein. *Neurobiol. Dis.* 2018, 110, 68–81. [CrossRef]
37. Huebecker, M.; Moloney, E.B.; van der Spoel, A.C.; Priestman, D.A.; Isacson, O.; Hallett, P.J.; Platt, F.M. Reduced sphingolipid hydrolase activities, substrate accumulation and ganglioside decline in Parkinson’s disease. *Mol. Neurodegener.* 2019, 14, 40. [CrossRef]
38. Wise, A.H.; Yang, A.; Naik, H.; Stauffer, C.; Zeitd, N.; Liong, C.; Balwani, M.; Desnick, R.J.; Alcalay, R.N. Parkinson’s disease prevalence in Fabry disease: A survey study. *Mol. Genet. Metab Rep.* 2018, 14, 27–30. [CrossRef]
39. Niimi, Y.; Ito, S.; Mizutani, Y.; Murate, K.; Shima, S.; Ueda, A.; Satake, W.; Hattori, N.; Toda, T.; Mutoh, T. Altered regulation of serum lysosomal acid hydrolase activities in Parkinson’s disease: A potential peripheral biomarker? *Parkinsonism Relat. Disord.* 2019, 61, 132–137. [CrossRef]
40. Russo, C.; Pontillo, G.; Pisani, A.; Saccà, F.; Riccio, E.; Maera, A.; Rusconi, G.; Stanzione, A.; Borrelli, P.; Brescia Morra, V.; et al. Striationagal involvement in Fabry Disease: A quantitative and volumetric Magnetic Resonance Imaging study. *Parkinsonism Relat. Disord.* 2018, 57, 27–32. [CrossRef]
41. Chang, D.; Nalls, M.A.; Hallgrímsdóttir, I.B.; Hunkapiller, J.; van der Brug, M.; Cai, F.; International Parkinson’s Disease Genomics Consortium; 23andMe Research Team; Kerchner, G.A.; Ayalon, G.; et al. A meta-analysis of genome-wide association studies identifies 17 new Parkinson’s disease risk loci. *Nat. Genet.* 2017, 49, 1511–1516. [CrossRef]
42. Pankratz, N.; Wilk, J.B.; Latourelle, J.C.; DeStefano, A.L.; Halter, C.; Pugh, E.W.; Doheny, K.F.; Gusella, J.F.; Nichols, W.C.; Foroud, T.; et al. Genomewide association study for susceptibility genes contributing to familial Parkinson disease. *Hum. Genet.* 2008, 124, 593–605. [CrossRef]
43. Do, C.B.; Tung, J.Y.; Dorfman, E.; Kiefer, A.K.; Drabant, E.M.; Francke, U.; Mountain, J.L.; Goldman, S.M.; Tanner, C.M.; Langston, J.W.; et al. Web-Based Genome-Wide Association Study Identifies Two Novel Loci and a Substantial Genetic Component for Parkinson’s Disease. *PLoS Genet.* 2011, 7, e1002141. [CrossRef] [PubMed]
44. Michelakakis, H.; Xiomerisiousi, G.; Dardiotis, E.; Bozi, M.; Vassilatis, D.; Kountra, P.; Patramani, G.; Moraitou, M.; Papadimitriou, D.; Stamboulis, E.; et al. Evidence of an association between the scavenger receptor class B member 2 gene and Parkinson’s disease. *Mov. Disord.* 2012, 27, 400–405. [CrossRef] [PubMed]
45. Chen, Y.; Yuan, X.; Cao, B.; Wei, Q.; Ou, R.; Yang, J.; Chen, X.; Zhao, B.; Song, W.; Wu, Y.; et al. No association of FAM47E rs6812193, SCARB2 rs6825004 and STX1B rs4889603 polymorphisms with Parkinson’s disease in a Chinese Han population. *Mov. Disord.* 2012, 27, 400–405. [CrossRef] [PubMed]
46. Chen, Y.; Yuan, X.; Cao, B.; Wei, Q.; Ou, R.; Yang, J.; Chen, X.; Zhao, B.; Song, W.; Wu, Y.; et al. No association of FAM47E rs6812193, SCARB2 rs6825004 and STX1B rs4889603 polymorphisms with Parkinson’s disease in a Chinese Han population. *Mov. Disord.* 2012, 27, 400–405. [CrossRef] [PubMed]
47. Najim al-Din, A.S.; Wriekat, A.; Mubaidin, A.; Dasouki, M.; Hiari, M. Pallido-pyramidal degeneration, supranuclear upgaze paresis and dementia: Kufor-Rakeb syndrome. *Acta Neurol. Scand.* 1994, 89, 347–352. [CrossRef] [PubMed]
48. Moraitou, M.; Papadimitriou, D.; Stamboulis, E.; et al. Evidence of an association between the scavenger receptor class B member 2 gene and Parkinson’s disease. *Mov. Disord.* 2012, 27, 400–405. [CrossRef] [PubMed]
49. Chang, D.; Nalls, M.A.; Hallgrímsdóttir, I.B.; Hunkapiller, J.; van der Brug, M.; Cai, F.; International Parkinson’s Disease Genomics Consortium; 23andMe Research Team; Kerchner, G.A.; Ayalon, G.; et al. A meta-analysis of genome-wide association studies identifies 17 new Parkinson’s disease risk loci. *Nat. Genet.* 2017, 49, 1511–1516. [CrossRef]
50. Michelakakis, H.; Xiomerisiousi, G.; Dardiotis, E.; Bozi, M.; Vassilatis, D.; Kountra, P.; Patramani, G.; Moraitou, M.; Papadimitriou, D.; Stamboulis, E.; et al. Evidence of an association between the scavenger receptor class B member 2 gene and Parkinson’s disease. *Mov. Disord.* 2012, 27, 400–405. [CrossRef] [PubMed]
51. Chen, Y.; Yuan, X.; Cao, B.; Wei, Q.; Ou, R.; Yang, J.; Chen, X.; Zhao, B.; Song, W.; Wu, Y.; et al. No association of FAM47E rs6812193, SCARB2 rs6825004 and STX1B rs4889603 polymorphisms with Parkinson’s disease in a Chinese Han population. *Mov. Disord.* 2012, 27, 400–405. [CrossRef] [PubMed]
52. Chen, Y.; Yuan, X.; Cao, B.; Wei, Q.; Ou, R.; Yang, J.; Chen, X.; Zhao, B.; Song, W.; Wu, Y.; et al. No association of FAM47E rs6812193, SCARB2 rs6825004 and STX1B rs4889603 polymorphisms with Parkinson’s disease in a Chinese Han population. *Mov. Disord.* 2012, 27, 400–405. [CrossRef] [PubMed]
53. Chen, Y.; Yuan, X.; Cao, B.; Wei, Q.; Ou, R.; Yang, J.; Chen, X.; Zhao, B.; Song, W.; Wu, Y.; et al. No association of FAM47E rs6812193, SCARB2 rs6825004 and STX1B rs4889603 polymorphisms with Parkinson’s disease in a Chinese Han population. *Mov. Disord.* 2012, 27, 400–405. [CrossRef] [PubMed]
54. Chen, Y.; Yuan, X.; Cao, B.; Wei, Q.; Ou, R.; Yang, J.; Chen, X.; Zhao, B.; Song, W.; Wu, Y.; et al. No association of FAM47E rs6812193, SCARB2 rs6825004 and STX1B rs4889603 polymorphisms with Parkinson’s disease in a Chinese Han population. *Mov. Disord.* 2012, 27, 400–405. [CrossRef] [PubMed]
55. Chen, Y.; Yuan, X.; Cao, B.; Wei, Q.; Ou, R.; Yang, J.; Chen, X.; Zhao, B.; Song, W.; Wu, Y.; et al. No association of FAM47E rs6812193, SCARB2 rs6825004 and STX1B rs4889603 polymorphisms with Parkinson’s disease in a Chinese Han population. *Mov. Disord.* 2012, 27, 400–405. [CrossRef] [PubMed]
56. Chen, Y.; Yuan, X.; Cao, B.; Wei, Q.; Ou, R.; Yang, J.; Chen, X.; Zhao, B.; Song, W.; Wu, Y.; et al. No association of FAM47E rs6812193, SCARB2 rs6825004 and STX1B rs4889603 polymorphisms with Parkinson’s disease in a Chinese Han population. *Mov. Disord.* 2012, 27, 400–405. [CrossRef] [PubMed]
57. Chen, Y.; Yuan, X.; Cao, B.; Wei, Q.; Ou, R.; Yang, J.; Chen, X.; Zhao, B.; Song, W.; Wu, Y.; et al. No association of FAM47E rs6812193, SCARB2 rs6825004 and STX1B rs4889603 polymorphisms with Parkinson’s disease in a Chinese Han population. *Mov. Disord.* 2012, 27, 400–405. [CrossRef] [PubMed]
52. Irahara-Miyana, K.; Otomo, T.; Kondo, H.; Hossain, M.A.; Ozono, K.; Sakai, N. Unfolded protein response is activated in Krabbe disease in a manner dependent on the mutation type. J. Hum. Genet. 2018, 63, 699–706. [CrossRef]
53. Mercado, G.; Castillo, V.; Soto, P.; Sidhu, A. ER stress and Parkinson’s disease: Pathological inputs that converge into the secretory pathway. Brain Res. 2016, 1648, 626–632. [CrossRef] [PubMed]
54. Hoozemans, J.J.M.; van Haastert, E.S.; Eikelenboom, P.; de Vos, R.A.I.; Rozemuller, J.M.; Schepers, W. Activation of the unfolded protein response in Parkinson’s disease. Biochem. Biophys. Res. Commun. 2007, 354, 707–711. [CrossRef] [PubMed]
55. Heman-Ackah, S.M.; Manzano, R.; Hoozemans, J.J.M.; Schepers, W.; Flynn, R.; Haerty, W.; Cowley, S.A.; Boudewyn, L.C.; Walkley, S.U. Current concepts in the neuropathogenesis of mucolipidosis type IV. Int. J. Mol. Sci. 2020, 21, 4966.
56. Valdés, P.; Mercado, G.; Vidal, R.L.; Molina, C.; Parsons, G.; Court, F.A.; Martinez, A.; Galleguillos, D.; Armentano, D.; Schneider, B.L.; et al. Control of dopaminergic neuron survival by the unfolded protein response transcription factor XBP1. Proc. Natl. Acad. Sci. USA 2014, 111, 6804–6809. [CrossRef] [PubMed]
57. Hallett, P.J.; Huebecker, M.; Brekk, O.R.; Moloney, E.B.; Rocha, E.M.; Priestman, D.A.; Platt, F.M.; Isacson, O. Glycosphingolipid levels and glucocerebrosidase activity are altered in normal aging of the mouse brain. Neurobiol. Aging 2018, 67, 189–200. [CrossRef]
58. Rocha, E.M.; Smith, G.A.; Park, E.; Cao, H.; Brown, E.; Hallett, P.; Isacson, O. Progressive decline of glucocerebrosidase in aging and Parkinson’s disease. Ann. Clin. Transl. Neurol. 2015, 2, 433–438. [CrossRef]
59. Balducci, C.; Pierguidi, L.; Persichetti, E.; Parnetti, L.; Sbaragli, M.; Tassi, C.; Orlacchio, A.; Calabresi, P.; Beccari, T.; Rossi, A. Lysosomal hydrolases in cerebrospinal fluid from subjects with Parkinson’s disease. Mov. Disord. 2007, 22, 1481–1484. [CrossRef]
60. Clark, L.N.; Chan, R.; Cheng, R.; Liu, X.; Park, N.; Parmalee, N.; Kisselev, S.; Cortes, E.; Torres, P.A.; Pastores, G.M.; et al. Gene-Wise Association of Variants in Four Lysosomal Storage Disorder Genes in Neuropathologically Confirmed Lewy Body Disease. PLoS ONE 2015, 10, e0125204. [CrossRef]
61. Gegg, M.E.; Sweet, L.; Wang, B.H.; Shihabuddin, L.S.; Sardi, S.P.; Schapira, A.H.V. No evidence for substrate accumulation in Parkinson brains with GBA mutations. Mov. Disord. 2015, 30, 1085–1089. [CrossRef]
62. Boudewyn, L.C.; Walkley, S.U. Current concepts in the neuropathogenesis of mucolipidosis type IV. J. Neurochem. 2018, 148, 669–689. [CrossRef] [PubMed]
63. Rothaug, M.; Zunke, F.; Mazzulli, J.R.; Schweizer, M.; Altmepen, H.; Lüllmann-Rauch, R.; Kallemeijn, W.W.; Gaspar, P.; Aerts, J.M.; Glatzel, M.; et al. LIMP-2 expression is critical for β-glucocerebrosidase activity and α-synuclein clearance. Proc. Natl. Acad. Sci. USA 2014, 111, 15573–15578. [CrossRef] [PubMed]
64. Sardi, S.P.; Clarke, J.; Kinnekom, C.; Tamsset, T.J.; Li, L.; Stanek, L.M.; Passini, M.A.; Grabowski, G.A.; Schlossmacher, M.G.; Sidman, R.L.; et al. CNS expression of glucocerebrosidase corrects alpha-synuclein pathology and memory in a mouse model of Gaucher-related synucleinopathy. Proc. Natl. Acad. Sci. USA 2011, 108, 12101–12106. [CrossRef]
65. Mazzulli, J.R.; Zunke, F.; Tsunemi, T.; Toker, N.J.; Jeon, S.; Burbulla, L.F.; Patnaik, S.; Sidransky, E.; Marugan, J.J.; Sue, C.M.; et al. Activation of β-Glucocerebrosidase Reduces Pathological α-Synuclein and Restores Lysosomal Function in Parkinson’s Patient Midbrain Neurons. J. Neurosci. 2016, 36, 7693–7706. [CrossRef]
66. Klein, A.D.; Mazzulli, J.R. Is Parkinson’s disease a lysosomal disorder? Brain 2018, 141, 2255–2262. [CrossRef]
67. Chu, Y.; Dodia, H.; Aeberscher, P.; Olanow, C.W.; Kordower, J.H. Alterations in lysosomal and proteasomal markers in Parkinson’s disease: Relationship to alpha-synuclein inclusions. Neurobiol. Dis. 2009, 35, 385–398. [CrossRef] [PubMed]
68. Dehay, B.; Bové, J.; Rodríguez-Muela, N.; Perier, C.; Recasens, A.; Boya, P.; Vila, M. Pathogenic Lysosomal Depletion in Parkinson’s Disease. J. Neurosci. 2010, 30, 12535–12544. [CrossRef] [PubMed]
69. Murphy, K.E.; Gysbers, A.M.; Abbott, S.K.; Tayebi, N.; Kim, W.S.; Sidransky, E.; Cooper, A.; Garner, B.; Halliday, G.M. Reduced glucocerebrosidase is associated with increased α-synuclein in sporadic Parkinson’s disease. Brain 2014, 137, 834–848. [CrossRef]
70. Kett, L.R.; Dauer, W.T. Endolysosomal dysfunction in Parkinson’s disease: Recent developments and future challenges. Mov. Disord. 2016, 31, 1433–1443. [CrossRef]
71. Balestrino, R.; Schapira, A.H.V. Glucocerebrosidase and Parkinson Disease: Molecular, Clinical, and Therapeutic Implications. Neuroscientist 2018, 20, 1–20. [CrossRef]
72. Kilpatrick, B.S.; Magalhaes, J.; Beavan, M.S.; McNeil, A.; Gegg, M.E.; Cleeter, M.W.J.; Bloor-Young, D.; Gan-Or, Z.; Dion, P.A.; Rouleau, G.A. Genetic perspective on the role of the autophagy-lysosome pathway in Parkinson disease. *Autophagy* 2015, 11, 1443–1457. [CrossRef] [PubMed]
73. Singleton, A.B.; Farrr, M.; Johnson, J.; Singleton, A.; Hague, S.; Kachergus, J.; Hulihan, M.; Peuralinna, T.; Dutra, A.; Nussbaum, R.; et al. alpha-Synuclein locus triplication causes Parkinson’s disease. *Science* 2003, 302, 841. [CrossRef] [PubMed]
74. Soldner, F.; Stelzer, Y.; Shivalilova, C.S.; Abraham, B.J.; Barrasa, M.I.; Goldmann, J.; Myers, R.H.; Young, R.A.; Jaenisch, R. Parkinson-associated risk variant in distal enhancer of alpha-synuclein modulates target gene expression. *Nature Medicine* 2016, 533, 95–99. [CrossRef]
75. Jiang, P.; Gan, M.; Yen, S.-H.; McLean, P.J.; Dickson, D.W. Impaired endo-lysosomal membrane integrity accelerates the seeding progression of alpha-synuclein aggregates. *Sci. Rep.* 2017, 7, 7690. [CrossRef]
76. Lieberman, A.P.; Puertollano, R.; Raben, N.; Slaugenhaupt, S.; Walkley, S.U. Autophagy in lysosomal storage disorders. *Autophagy* 2012, 8, 719–730. [CrossRef] [PubMed]
77. Plotegher, N.; Duchen, M.R. Mitochondrial Dysfunction and Neurodegeneration in Lysosomal Storage Disorders. *Trends Mol. Med.* 2017, 23, 116–134. [CrossRef]
78. Oselame, L.D.; Duchen, M.R. Quality control gone wrong: Mitochondria, lysosomal storage disorders and neurodegeneration. *Br. J. Pharmacol.* 2014, 171, 1958–1972. [CrossRef]
79. Plotegher, N.; Duchen, M.R. Crosstalk between Lysosomes and Mitochondria in Parkinson’s Disease. *Front. Cell Dev. Biol.* 2017, 5, 1–8. [CrossRef]
80. Chiasserini, D.; Paciotti, S.; Eusebi, P.; Persichetti, E.; Tasegian, A.; Kurzawa-Akanbi, M.; Chinnery, P.F.; Morris, C.M.; Calabresi, P.; Parnetti, L.; et al. Selective loss of glucocerebrosidase activity in sporadic Parkinson’s disease and dementia with Lewy bodies. *Mol. Neurodegener.* 2015, 10, 1–6. [CrossRef]
81. Galvagnion, C. The Role of Lipids Interacting with alpha-Synuclein in the Pathogenesis of Parkinson’s Disease. *J. Parkinsoms Dis.* 2017, 7, 433–450. [CrossRef]
82. Goedert, M.; Spillantini, M.G.; Del Tredici, K.; Braak, H. 100 years of Lewy pathology. *Nat. Rev. Neurol.* 2013, 9, 13–24. [CrossRef]
83. Suzuki, M.; Sango, K.; Wada, K.; Nagai, Y. Pathological role of lipid interaction with alpha-synuclein in Parkinson’s disease. *Neurochem. Int.* 2018, 119, 97–106. [CrossRef] [PubMed]
84. Walkley, S.U. Secondary accumulation of gangliosides in lysosomal storage disorders. *Cell Calcium* 2010, 59, 12–20. [CrossRef]
85. Vitting, E.B.; Platt, F.M.; Futereman, A.H. Common and uncommon pathogenic cascades in lysosomal storage diseases. *J. Biol. Chem.* 2010, 285, 20423–20427. [CrossRef] | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3519,
1
],
[
3519,
7208,
2
],
[
7208,
8312,
3
],
[
8312,
12583,
4
],
[
12583,
16817,
5
],
[
16817,
19394,
6
],
[
19394,
23793,
7
],
[
23793,
27637,
8
],
[
27637,
32003,
9
],
[
32003,
38988,
10
],
[
38988,
43805,
11
],
[
43805,
46636,
12
]
]
} |
a5edea167382276e0c1b989a3a8abf25ab32b432 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 11,
"total-fallback-pages": 0,
"total-input-tokens": 39219,
"total-output-tokens": 12070,
"fieldofstudy": [
"Biology"
]
} | During prolonged low-level contractions, synergist muscles are activated in an alternating pattern of activity and silence called as alternate muscle activity. Resting muscle stiffness is considered to increase due to muscle fatigue. Thus, we investigated whether the difference in the extent of fatigue of each plantar flexor synergist corresponded to the difference in the frequency of alternate muscle activity between the synergists using muscle shear modulus as an index of muscle stiffness. Nineteen young men voluntarily participated in this study. The shear moduli of the resting medial and lateral gastrocnemius muscles (MG and LG) and soleus muscle (SOL) were measured using shear wave ultrasound elastography before and after a 1-h sustained contraction at 10% peak torque during maximal voluntary contraction of isometric plantar flexion. One subject did not accomplish the task and the alternate muscle activity for MG was not found in 2 subjects; therefore, data for 16 subjects were used for further analyses. The magnitude of muscle activation during the fatiguing task was similar in MG and SOL. The percent change in shear modulus before and after the fatiguing task (MG: 16.7 ± 12.0%, SOL: −4.1 ± 13.9%; mean ± standard deviation) and the alternate muscle activity during the fatiguing task (MG: 33 [20–51] times, SOL: 30 [17–36] times; median [25th–75th percentile]) were significantly higher in MG than in SOL. The contraction-induced change in shear modulus (7.4 ± 20.3%) and the alternate muscle activity (37 [20–45] times) of LG with the lowest magnitude of muscle activation during the fatiguing task among the plantar flexors were not significantly different from those of the other muscles. These results suggest that the degree of increase in muscle shear modulus induced by prolonged contraction corresponds to the frequency of alternate muscle activity between MG and SOL during prolonged contraction. Thus, it is likely that, compared with SOL, the alternate muscle activity of MG occurs more frequently during prolonged contraction due to the greater increase in fatigue of MG induced by the progression of a fatiguing task.
**Keywords:** muscle fatigue, joint torque, shear wave ultrasound elastography, evoked torque, electromyography
INTRODUCTION
Human joint movement is produced by multiple muscles that act as synergists. During prolonged low-level contractions, the synergist muscles are activated in an alternating pattern of activity and silence (Hellsing and Lindström, 1983; Sjøgaard et al., 1986; Tamaki et al., 1998, 2011; Semmler et al., 1999, 2000; Kouzaki et al., 2002; Akima et al., 2012), referred to as alternate muscle activity (Kouzaki et al., 2002). The system of alternate activity among the synergists provides muscles with time to recover from the fatigue that develops during prolonged contraction (Tamaki et al., 2011). During prolonged (more than 1 h) low-level contraction, the alternate muscle activity was found to occur more frequently in the latter half than in the former half of the prolonged contraction to maintain the target level of joint torque in previous studies (Tamaki et al., 1998; Kouzaki et al., 2002; Akima et al., 2012). This finding is considered to be affected by the muscle fatigue (Tamaki et al., 1998) because it increases as the fatiguing task progresses. In addition, some previous studies (Enoka and Stuart, 1992; Gandevia, 2001; Kouzaki and Shinohara, 2006) have suggested that the alternate muscle activity between the synergist muscles is effective for minimizing or attenuating muscle fatigue. Therefore, there is a possibility that the difference in the fatigue of each synergist corresponds to the difference in the frequency of alternate muscle activity among the synergists.
Resting muscles become stiffer under several conditions, including those involving cramps and damage (Fischer, 1987; Murayama et al., 2000). Similarly, muscle fatigue has been considered as one of possible causes for the increment of muscle stiffness (Morisada et al., 2006; Descarreaux et al., 2010). It is known that a resting tension develops when muscles fail to fully relax during fatigue (Gong et al., 2000, 2003). Given that muscle tension increases muscle stiffness (Dresner et al., 2001), it is reasonable to assume that resting muscle stiffness increases due to muscle fatigue. Recently, resting muscle shear modulus was calculated by shear wave propagation speed within a muscle, which is simply and non-invasively determined by shear wave ultrasound elastography, and has been used as an index of muscle stiffness (Eby et al., 2015; Dieterich et al., 2017). The combination of this new technique and traditional methods, such as surface electromyography (EMG) and evoked torque(force), can be useful to examine the connection between the extent of fatigue of each synergist and the frequencies of alternate muscle activity among the muscles.
In the plantar flexors, alternate muscle activity occurs frequently during prolonged low-level contraction (more than 1 h; Tamaki et al., 1998, 2011; Kishibuchi and Kouzaki, 2013). It has been demonstrated that the medial gastrocnemius and soleus muscles (MG and SOL) play a key role in producing target plantar flexion torque during prolonged contraction compared to the lateral gastrocnemius muscle (LG) (Tamaki et al., 1998, 2011); possibly producing the greater fatigue of MG and SOL compared with LG after the prolonged contraction. To prevent the fatigue of MG and SOL, the alternate muscle activity of MG and SOL may occur more frequently compared with that of LG during the prolonged contraction. In this study, we determined the shear moduli of the plantar flexor synergists before and after 1-h sustained low-level contraction using shear wave ultrasound elastography and tested the following two hypotheses: (1) The contraction-induced increases in the shear moduli of MG and SOL are more prominent than that of LG and (2) of MG and SOL, the degree of increase in muscle shear modulus induced by prolonged contraction is larger in the muscle in which the alternate muscle activity occurs more frequently than in the other muscle.
MATERIALS AND METHODS
Subjects
After providing written informed consent, 19 healthy young men voluntarily participated in this study (age: 22 ± 1 year, height: 170.7 ± 4.5 cm, body mass: 62.6 ± 7.3 kg; mean ± standard deviation [SD]). We confirmed existence or non-existence of the subjects’ cardiovascular diseases and physical activity levels orally before starting the experiment. They were free of cardiovascular diseases. Also, they were sedentary and did not take any exercise. All measurements were performed with the subjects’ right legs. This study was approved by the Ethics Committee of the Shibaura Institute of Technology and conducted according to the Declaration of Helsinki.
Experimental Procedures
Figure 1 shows the experimental procedures. After determining the sites of surface EMG signals and tibial nerve stimulation, subjects performed several submaximal plantar flexion contractions as a warm-up and then rested for 3 min. Subsequently, the measurements of resting evoked singlet and triplet responses were performed every 10 s. After a 30-s rest period, the subjects performed maximal voluntary contraction (MVC) of isometric plantar flexion for 3 s two or three times. The measurements of the shear moduli of resting MG, LG, and SOL were performed after the MVC task. The subjects took a 10-min rest and started low-level (10%MVC) sustained contraction for 1 h as the fatiguing task. Immediately after finishing this task, the measurements of resting evoked singlet and triplet responses and the MVC task was repeated as described above. Last, the shear moduli of each muscle were measured.
Shear Modulus Measurement
Both before and after performing the fatiguing task, the shear wave propagation speed within the resting MG, LG, and SOL was measured three times (Figure 2). The measurement sites of shear wave propagation speed within each muscle were determined as follows. For MG, at the proximal 40% of the lower leg length from the popliteal crease to lateral malleolus, the boundaries between MG and LG or the tibia were determined using ultrasonography. The elastographic image including MG was obtained at 40% of the girth from the boundary between MG and the tibia to that between MG and LG medially. For LG and SOL, the corresponding boundaries at the proximal 30% of the lower leg length were similarly determined. The elastographic image including both LG and SOL was obtained at 20% of the girth from the boundary between MG and the tibia to that between
Akagi et al. Fatigue-Induced Change in Muscle Stiffness
**FIGURE 1** | Experimental procedures. MVC, maximal voluntary contraction of isometric plantar flexion.
**FIGURE 2** | Typical elastographic images of the medial and lateral gastrocnemius muscles (MG and LG) and the soleus muscle (SOL) before (upper) and after (lower) the fatiguing task. Shear wave ultrasound elastography generated color-coded images with a scale from blue (soft) to red (hard) depending on the magnitude of shear wave propagation speed.
MG and LG laterally. At each measurement site, a 45-mm electronic linear array probe (9L4 Transducer, 4–9 MHz, Siemens Medical Solutions, USA) attached to a B-mode ultrasound apparatus (ACUSON S2000, Siemens Medical Solutions, USA) was longitudinally placed and its direction was adjusted to match the orientation of muscle fascicle before the first measurement of each shear wave propagation speed. Then, the measurement site was marked with a pen. Special care was taken to match the measurement site each time. The electronic linear array probe was placed at the measurement site with water-soluble transmission gel and without depression of the tissues. Three obtained elastographic images of sufficient quality were copied to a personal computer. The shear wave propagation speeds within MG, LG, and SOL were determined using image analysis software (MSI Analyzer version 2.0AqI, Institute of Rehabilitation Science, Tokuyukai Medical Corporation, Japan). A quadrangular region of interest (ROI) was set on the muscle to be as large as possible within the color-coded area of the elastographic image, and the mean value of the shear wave propagation speed within the ROI was automatically obtained at 0.01 m/s. The analyses of each image were conducted once. The mean values of the three measurements were used for further analyses. The coefficients of variance and intraclass correlation coefficients type 1, 3 for these measurements were 2.0 ± 1.1% and 0.955 (P < 0.001) for MG, 2.4 ± 2.1% and 0.966 (P < 0.001) for LG and 3.7 ± 1.9% and 0.912 (P < 0.001) for SOL.
The shear modulus of a muscle is calculated as the product of muscle density and shear wave velocity squared (Nordez and Hug, 2010; Akagi et al., 2015). In this study, the muscle density was assumed to be 1,084 kg/m³, which was the mean of the two values (1,112 and 1,055 kg/m³) obtained from different methods in a previous study (Ward and Lieber, 2005).
**EMG and Joint Torque Measurements**
Surface EMG signals were acquired from MG, LG, and SOL. The muscle belly and fascicle longitudinal directions were confirmed using a B-mode ultrasound apparatus (ACUSON S2000, Siemens Medical Solutions, USA) with the 45-mm electronic linear array probe (9L4 Transducer, 4–9 MHz, Siemens Medical Solutions, USA) perpendicular to the skin for identifying the fascicles. After skin shaving, rubbing with sandpaper and cleaning with alcohol, bipolar Ag/AgCl surface electrodes (F-150M, size adjustment by cutting to 10 × 20 mm, Nihon Koden, Tokyo, Japan; 20 mm inter-electrode distance) with high-pass filtering at 5 Hz using a bioamplifier system (MEG-6108, Nihon Koden, Tokyo, Japan) were placed at the proximal 40% (MG) and 30% (LG) of the lower leg length. The electrodes of MG and LG were set at 1 cm laterally from each site where the ultrasound probe was placed to determine the shear wave propagation speed within MG and LG before and after the fatiguing task. For SOL, electrode placement was ∼5 cm distal from the proximal 40% of the lower leg length, but there were inter-individual variations due to the variability of the superficial region of SOL.
Muscle fatigue is generally defined as any exercise-induced reduction in the ability of a muscle to generate force or power (Gandevia, 2001). In this study, MVC torque (TQ_{MVC}) and evoked peak triplet torque (TQ_{TRI}) were determined because it is considered that the exercise-induced decrease in TQ_{MVC} depends on both peripheral and central factors (Fernandez-del-Olmo et al., 2013) whereas that in TQ_{TRI} reflects peripheral fatigue (Miyamoto et al., 2011).
To obtain TQ_{TRI} and the peak-to-peak compound muscle action potential amplitude (Mmax), the bioamplifier system
Before and after the fatiguing task, Mmax, TQ_{TRI}, and TQ_{MVC} were measured. Subjects sat on a seat of a dynamometer (CON-TREX MJ, Physiomed, Germany) with the hip at 80° of flexion, the knee fully extended and the ankle at 20° of plantar flexion. This ankle joint angle was set based on a previous finding that the appropriate ankle position for detecting frequent alternations of activity in plantar flexor synergists was 20° of plantar flexion during static contractions at 10%MVC (Tamaki et al., 2011). The subject’s pelvis, torso and ankle were secured on the reclining seat and dynamometer with non-elastic straps and/or seat belt. Care was taken to adjust centers of the rotation of the ankle joint and the dynamometer. Firstly, stimulus intensity was increased until a plateau in the twitch torque and Mmax were reached at 20° of plantar flexion and supramaximal stimulus intensity was set (66 ± 25 mA; mean ± SD). Then, two Mmax and two TQ_{TRI} (100 Hz) were obtained every 10 s, and averaged across the two contractions, respectively. Thereafter, the peak torque during MVC was measured two or three times with a 1 min interval. If the difference between the first two values of peak torque was >10% of the higher one, the peak torque was measured one more time. Before the fatiguing task, the highest value of two or three peak torque measurements was adopted as TQ_{MVC}. After the fatiguing task, TQ_{MVC} measurement was performed only once. The EMG and joint torque signals were recorded at a sampling frequency of 2000 Hz and stored in a personal computer using LabChart software (v8.1.5, ADInstruments, Australia) after A/D conversion (PowerLab16/35, ADInstruments, Australia).
During the fatiguing task, the torque data were displayed as waveforms on a monitor of a personal computer using LabChart software in real time. The horizontal target line (i.e., 10%MVC) was also displayed on the monitor using the software, and the monitor was placed in front of the subjects to provide visual feedback. As described above, the subject’s pelvis, torso and ankle were secured on the reclining seat and dynamometer with non-elastic straps and/or seat belt during the fatiguing task, but the fixation of the subject’s ankle was unfixed slightly to avoid the numbness in the right leg or foot. The subjects tried to match their torque level and the target line, and were instructed not to alter joint angles during the fatiguing task. The ankle joint angle was determined with an electronic goniometer (SG110/A, Biometrics, UK).
EMG Data Analysis
EMG signals during the MVC tasks were full-wave rectified and averaged 0.5 s in the period around the peak torque to calculate the average EMG signals (AEMG) in each MVC task. Each AEMG in the selected MVC tasks before and after the fatiguing task was used. Moreover, the AEMG was normalized by Mmax in each muscle both before and after the fatiguing task to minimize peripheral contamination and thus to assess the level of central motor output (Place et al., 2010).
Alternate muscle activity was observed through EMG activity between the plantar flexor synergists during the fatiguing task. The alternate muscle activity was defined and measured according to previously established methods (Kouzaki et al., 2002; Kouzaki and Shinohara, 2006). The EMG signals were full-wave rectified and averaged over 15 s to yield the AEMG every 15 s (Figure 3). The calculated AEMG of each muscle was smoothed by a five-point moving average and differentiated (dAEMG/dt). Eight sample points of dAEMG/dt immediately after the onset of exercise were extracted and 3 SDs of dAEMG/dt during this period were determined as a normal fluctuation. The criterion for an outlier was defined as dAEMG/dt throughout the fatiguing task that exceeds ±3 SDs, including both upper and lower limits as a normal fluctuation (Figure 4). The extracted outliers were classified into positive (+) and negative (−) outliers per muscle. The alternate muscle activity between the plantar flexor synergists was defined as the case in which the positive and the negative outliers were simultaneously observed between the muscles (Figure 4). The overlap in time for extraction of alternate muscle activity was accepted for a 15-s period. The alternate muscle activity between MG and LG (1), MG and SOL (2), and LG and SOL (3) was counted. In Figure 4, for example, (1) = numbers of a + b = 14, (2) = numbers of c + d = 16, and (3) = numbers of e + f = 6. In addition, the frequency of alternate muscle activity (N_{0−60}) for each muscle was calculated as follows: MG = (1) + (2), LG = (1) + (3), SOL = (2) + (3). These frequencies were also counted every 30 min (N_{30−60} and N_{60−90}). In each muscle, the AEMG every 15 s during the fatiguing task was normalized using the AEMG in the selected MVC task before the fatiguing task, and the mean value during the fatiguing task was calculated as %AEMG_{0−60}.
Statistical Analyses
Of the 19 subjects, 1 was unable to complete the fatiguing task and alternate muscle activity for MG was not found in 2 participants because of the great variability in the 8 sample points of dAEMG/dt immediately after the onset of exercise. Consequently, the data for 16 subjects were used for further analyses.
A two-way analysis of variance (ANOVA) with two within-group factors (time [before and after the fatiguing task] and muscle [MG, LG, and SOL]) was used to evaluate changes induced by the fatiguing task in muscle shear moduli, Mmax, and AEMG during the MVC task normalized by Mmax. When a significant interaction was detected, an additional ANOVA with the Bonferroni multiple-comparison test was performed. The percentage changes in the shear moduli and normalized AEMG during the MVC task of each muscle from before to after the fatiguing task were also calculated and their differences were examined using a one-way ANOVA followed by the Bonferroni multiple-comparison test. Differences in TQ_{MVC} and TQ_{TRI} before and after the fatiguing task were examined using a paired t-test.
The one-way ANOVA followed by Bonferroni multiple-comparison test was used to investigate differences in
%AEMG of MG, LG, and SOL were tested using the Friedman test followed by the Wilcoxon post-hoc test. In addition, relationships of N0−60, N0−30, or N30−60 and the percentage change in the muscle shear modulus were tested in each muscle using Spearman’s rank correlation coefficients.
To test the effect of the individual variability of joint torque level during the fatiguing task on the fatigue-induced decline in joint torque, the data of joint torque during the fatiguing task were firstly normalized by TQ MVC as the joint torque level during the fatiguing task. Then, the percentage changes in TQ MVC and TQ TRI from before to after the fatiguing task (%ΔTQ MVC and %ΔTQ TRI) were computed, and Pearson’s product-moment correlation coefficients were calculated between the joint torque level and %ΔTQ MVC or %ΔTQ TRI.
Parametric data are presented as means ± SDs. Statistical significance was set at P < 0.05. Statistical analyses were performed using statistical analysis software (SPSS 24.0, IBM, USA).
RESULTS
Shear Moduli and Joint Torque
Figure 5 shows the muscle shear moduli before and after the fatiguing task. A significant time × muscle interaction (P = 0.001) was found without significant main effects of muscle (P = 0.057) or time (P = 0.277). The shear modulus of MG significantly increased after the fatiguing task (P < 0.001), but there were no significant differences in the shear modulus before and after the fatiguing task in LG (P = 0.193) and SOL (P = 0.258). In addition, the shear modulus of MG was significantly higher than that of SOL only after the fatiguing task (P = 0.049). Correspondingly, for the percentage change in the muscle shear moduli from before to after the fatiguing task, a main effect of muscle was significant (P = 0.028), and the percentage change in MG (16.7 ± 12.0%) was significantly higher than that in SOL (−4.1 ± 13.9%; P < 0.001), but not than that in LG (7.4 ± 20.3%; P = 0.306). TQ MVC and TQ TRI were significantly decreased after the fatiguing task (both P < 0.001; Figure 6).
EMG
For Mmax, there was neither a significant main effect of time (P = 0.823) nor a significant time × muscle interaction (P = 0.825). In respect to AEMG during the MVC task normalized by Mmax, there was a significant main effect of time (P = 0.009) without a significant time × muscle interaction (P = 0.280) (Figure 7A). There was no significant main effect of muscle on the percent changes in normalized AEMG during the MVC task (MG: −20.6 ± 29.9%; LG: −23.8 ± 20.2%; SOL: −21.1 ± 26.6%; P = 0.880). Regarding %AEMG0−60, a significant main effect of muscle was found (P < 0.001), and %AEMG0−60 of LG (10.3 ± 3.6%) was significantly lower than those of MG (20.3 ± 7.1%; P < 0.001) and SOL (19.9 ± 6.9%; P = 0.001; Figure 7B).
Frequency of Alternate Muscle Activity
For N0−60 and N30−60, a main effect of muscle was significant (N0−60: P = 0.022; N30−60: P = 0.006). Wilcoxon post-hoc test revealed that N0−60 and N30−60 were significantly higher in MG than in SOL (N0−60: P = 0.040; N30−60: P = 0.006; Figure 8).
FIGURE 4 | Measuring alternate muscle activity during the 1-h fatiguing task. The 15-s averaged electromyography signals of the medial and lateral gastrocnemius muscles (MG and LG) and the soleus muscle (SOL) were smoothed by five-point moving average and differentiated (dAEMG/dt). The left figures show the time-course changes in dAEMG/dt of each muscle. Eight sample points of dAEMG/dt immediately after the onset of exercise were extracted and 3 standard deviations (3 SDs) of dAEMG/dt during this period were determined as the normal fluctuation. The dashed line indicates the ±3 SDs. The criterion for an outlier was defined as dAEMG/dt throughout the fatiguing task that exceeded the dashed lines. As shown on the right, the extracted outliers were classified into positive (+) and negative (−) outliers per muscle. The alternate muscle activity of LG → MG, MG → LG, SOL → MG, MG → SOL, SOL → LG, and LG → SOL are indicated using a, b, c, d, e, and f, respectively.
Regarding N₀−₃₀, a significant main effect of muscle was not found (P = 0.074; Figure 8). When examining relationships between the frequency of alternate muscle activity and the percentage change in the muscle shear modulus, none of the correlations were significant (MG: rₕ = 0.284, P = 0.286 [N₀−₆₀]; rₜ = 0.225, P = 0.402 [N₀−₃₀]; rₖ = 0.066, P = 0.807 [N₃₀−₆₀]; LG: rₕ = 0.397, P = 0.128 [N₀−₆₀]; rₜ = 0.489, P = 0.054 [N₀−₃₀]; rₖ = 0.440, P = 0.088 [N₃₀−₆₀]; SOL: rₕ = 0.068, P = 0.803 [N₀−₆₀]; rₜ = −0.040, P = 0.883 [N₀−₃₀]; rₖ = −0.317, P = 0.231 [N₃₀−₆₀]).
Variability in Ankle Joint Angle and Joint Torque during the Fatiguing Task
The SD of change in ankle joint angle during the fatiguing task was 1.6 ± 1.8° (min: 0.2°, max: 7.3°). The mean values and SD of joint torque level during the fatiguing task were 9.5 ± 0.6% (min: 8.7%, max: 11.1%) and 1.1 ± 0.5% (min: 0.4%, max: 1.9%). There were no significant correlations between the joint torque level during the fatiguing task and ΔTQ/MVC (r = −0.272, P = 0.309) or ΔTQ/TRI (r = −0.391, P = 0.134).
**DISCUSSION**
After sustaining 10% MVC for 1 h, both TQ_{MVC} and TQ_{TRI} were significantly decreased (Figure 6). It is as that the extent of the prolonged contraction-induced decrease in TQ_{TRI} reflects peripheral fatigue (Miyamoto et al., 2011) whereas that in TQ_{MVC} depends on both peripheral and central factors (Fernandez-del-Olmo et al., 2013). Therefore, both peripheral and central fatigue are suggested to result in the decline in joint torque after the fatiguing task. N_{0−60} and N_{30−60} were significantly higher in MG than in SOL; however, there was no significant difference in N_{0−30} among the plantar flexors (Figure 8). These results suggest that the difference in the frequency of alternate muscle activity among the plantar flexors is prominent in the latter half of the prolonged contraction because muscle fatigue increases as the fatiguing task progresses. Based on these suggestions, we would like to refer to the fatigue-induced changes in the shear moduli of the plantar flexor synergist.
In the current study, the %AEMG_{0−60} was significantly higher in MG and SOL than in LG (Figure 7B), which is in agreement with previous findings (Tamaki et al., 1998, 2011). Considering the present results and the previous findings that the ratio of LG volume to total muscle volume of MG and SOL was <20% (Fukunaga et al., 1992), it is reasonable to suggest that MG and SOL contributed more greatly to sustaining 10% MVC during the fatiguing task compared with LG. The hypotheses of the current study were that (1) the increases in the shear moduli of MG and SOL are more prominent than that of LG after the prolonged contraction and (2) of MG and SOL, the alternate
muscle activity occurs more frequently in the muscle with the larger degree of increase in muscle shear modulus induced by the prolonged contraction. In the present study, the significant increase in shear modulus after the fatiguing task was found only in MG (Figure 5) but the percent change in shear modulus of MG was not significantly higher than that of LG. Furthermore, there was no significant difference in the percent change in shear modulus between LG and SOL. Thus, the first hypothesis was rejected. Meanwhile, N₀₋₆₀ and N₃₀₋₆₀ were significantly higher in MG than in SOL (Figure 8) and the increase in shear modulus induced by the fatiguing task was found in MG but not in SOL (Figure 5). Additionally, the percent change in shear modulus of MG was significantly higher than that of SOL. These results indicate that the fatigue-induced increase in muscle shear modulus occurred in MG but not in SOL, supporting the second hypothesis.
To clarify the reasons for the rejection of the first hypothesis, we would like to discuss the results of MG and LG and those of LG and SOL separately. In respect to MG and LG, the significant increase in shear modulus after the fatiguing task was found in MG but not in LG (Figure 5). As described above, MG is suggested to contribute more greatly to sustaining 10%MVC during the fatiguing task compared with LG. Therefore, despite no significant differences in the frequency of alternate muscle activity between MG and LG (Figure 8), the difference in the contribution to sustaining 10%MVC between them appears to influence the current results of their shear moduli (Figure 5). Regarding LG and SOL, the contribution of LG to sustaining 10%MVC during the fatiguing task seemed to be lower than that of SOL whereas the fatigability is greater for LG than for SOL (Ochs et al., 1977; Supert and Munson, 1981). Hence, fatigue of LG can be regulated by the lower contribution of LG to sustaining 10%MVC and the greater fatigability for LG compared with SOL, likely affecting the current results that there were no significant differences in the shear moduli before and after the fatiguing task (Figure 5) and in N₀₋₆₀, N₀₋₃₀, and N₃₀₋₆₀ between LG and SOL (Figure 8).
As a reason for supporting the second hypothesis, the effect of the difference in fatigability between MG and SOL on the difference in frequency of alternate muscle activity during the prolonged low-level contraction between them is cited. Physical fluctuations of the muscle of ~8–12 Hz, which are termed physiological tremors and arise from Iα afferent activity and neural oscillations from central commands to motoneuron pool, have been observed in fatigued muscles (McAuley and Marsden, 2000). In accordance with a previous study (Kouzaki et al., 2004) investigating the physiological tremor of knee extension force fluctuations during 1-h knee extension at 2.5% of MVC, its change has been suggested to be caused by the unique muscle activity of the rectus femoris muscle (RF) during the alternate muscle activity due to the more fatigable RF than the other synergists. Another study (Kouzaki and Shinohara, 2006) has also pointed out that, for the alternate muscle activity among the knee extensors, fatigue-related feedback information is likely to be transmitted from the afferents of RF to α-motoneurons in each knee extensor muscle, most likely via interneurons, leading to the alternate muscle activity. Thus, the more fatigable RF compared with the other synergists is expected to be a key muscle for the alternate muscle activity among the knee extensors during prolonged contractions in low-level. In fact, the alternative muscle activity appears to occur more frequently in RF than in the other synergists during such contractions as shown previously (Kouzaki et al., 2002; Akima et al., 2012). In a previous study investigating the plantar flexors (Kishibuchi and Kouzaki, 2013), alternate muscle activity during a 2-h prolonged contraction in 10%MVC was associated with changes in physiological tremor of ankle angular acceleration when the AEMG of MG dramatically decreased with the increase in the AEMG of LG and/or SOL. In particular, the activity of MG, but not the activity of the other synergists, was accompanied by physiological tremor. Hence, Kishibuchi and Kouzaki (2013) have demonstrated that these findings would correspond to the findings of RF described above (Kouzaki et al., 2004), and that...
MG is a key muscle for the alternate muscle activity among the plantar flexors during the prolonged low-level contraction. MG has been suggested to have the greater fatigability compared with SOL based on the previous findings of their contraction-induced fatigue and/or fiber type composition (Ochs et al., 1977; Sybert and Munson, 1981). In addition, the higher N₀⁻₆₀ and N₃₀⁻₆₀ of MG than of SOL were found in the current study (Figure 8). Therefore, considering the similarity between MG and RF from the perspective of fatigability, it is possible that the increase in the muscle shear modulus after the fatiguing task was found in the more fatigable MG but not in the less fatigible SOL in the present study (Figure 5).
On the other hand, there were no significant correlations between N₀⁻₆₀, N₀⁻₃₀, or N₃₀⁻₆₀ and the percentage change in the muscle shear modulus in any plantar flexors. As described in the Section Introduction, the system of alternate activity among the synergists provides muscles with time to recover from the fatigue that develops during prolonged contraction (Tamaki et al., 2011). However, it is not clear whether the degree of recovery from the fatigue induced by the alternate muscle activity is consistent from person to person. If this is inconsistent among individuals, the relationships between the frequency of alternate muscle activity and the percentage change in the muscle shear modulus should be weakened. In addition, based on previous findings (Tamaki et al., 1998, 2011), Kishibuchi and Kouzaki (2013) indicated that alternate muscle activity of plantar flexors emerges without regularity, and that overlapped activities between the muscles are observed in the plantar flexors. Thus, the alternations of plantar flexor during prolonged contraction are considered to be complicated. This complexity may also weaken the aforementioned relationships. Taken together, the present results suggest the difficulty of comparing the effect of the frequency of alternate muscle activity on the muscle shear modulus among individuals. In other words, the second hypothesis is likely to be supported based on the results of group means, regardless of the results of the correlations between muscle shear modulus and the frequency of alternate muscle activity.
We discuss implications of the present findings in this part. During the 1-h fatiguing task, the magnitudes of muscle activations of MG and SOL were higher than that of LG. Based on this result and the previous findings that the ratio of LG volume to total muscle volume of MG and SOL was <20% (Fukunaga et al., 1992), MG and SOL can be considered to contribute more greatly to perform plantar flexion compared with LG as described in the earlier part. On the other hand, of MG and SOL, the task-induced increase in muscle fatigue evaluated by shear wave ultrasound elastography was found only in MG. In other words, the muscle fatigue of SOL during the task was likely to be attenuated by the alternate muscle activity. Thus, the difference in fatigability was prominent between MG and SOL during performing plantar flexion. Therefore, when conducting a fatigue-training of the plantar flexor synergists, for instance, the differences in the contributions to plantar flexion strength and the fatigue profiles of the three muscles could imply inter-muscle differences in the training modality. Furthermore, it has been reported that the gastrocnemius muscle is more susceptible to injury than SOL, with by far the majority of the strain injuries involving MG as opposed to LG (Koulouris et al., 2007). Hence, when performing exercises as sports or recreational activities which could result in the muscle fatigue of the plantar flexors, special care may be needed for the muscle fatigue of MG.
The present study had three limitations. First, there was variability in the ankle joint angles and joint torque level during the 1-h fatiguing task. Tamaki et al. (2011) found that there were no significant differences in the averaged EMG activities of MG, LG, and SOL during MVC between the ankle joint angles of 20° and 10° or 30° of plantar flexion. Because the ankle joint angle of 20° of plantar flexion was used in the current study, it is unlikely that the interpretation of the current study was greatly influenced by the changes in the ankle joint angle observed here (SD of change in ankle joint angle during the fatiguing task: 1.6±1.8° [0.2–7.3°]). The current results suggest that the target level of joint torque was roughly maintained during the fatiguing task. Moreover, judging from the result that there were no significant correlations between the joint torque level during the fatiguing task and %ΔTQ_{MVC} or %ΔTQ_{TRI}, the effect of the individual variability of joint torque level during the fatiguing task on the fatigue-induced decline in joint torque is likely to be small. These results indicate that the effects of variability in the ankle joint angles and joint torque level during the fatiguing task are ignorable in this study. Second, changes in muscle water content would affect the values of muscle shear moduli because the subject’s posture was nearly unchanged through the experiment (i.e., for over 1 h). This fact may influence the differences in muscle shear moduli before and after the fatiguing task in each muscle. However, it is hard to think that there was a large difference in the change in muscle water content among the plantar flexors. In other words, it is very difficult to explain the reason why the shear modulus significantly increased after the fatiguing task only in MG from the standpoint of change in muscle water content. Therefore, the effect of the changes in muscle water content on the interpretation of the present results seems to be small. Third, changes in muscle blood volume was not considered in the present study. Kouzaki et al. (2003) investigated the relationship between local blood circulation and alternate muscle activity in RF and vastus lateralis muscle during 1-h knee extension at 2.5%MVC. In the previous study, local blood circulation was modulated by the alternate muscle activity of knee extensor synergists, and a negative correlation between the muscle activity and blood volume sequences was found in the more fatigable RF but not in vastus lateralis muscle. This phenomenon can also be observed among the plantar flexor synergists during the prolonged contraction. If so, the data of muscle shear modulus in the present study might be affected by the muscle blood volume. This effect needs to be investigated in the future to strengthen the present findings.
In summary, the magnitude of muscle activation during the 1-h fatiguing task was similar in MG and SOL. The muscle shear modulus increased after the 1-h fatiguing task only in MG and the percent change in shear modulus of MG from before to after the fatiguing task was higher than that of SOL. Furthermore, N₀⁻₆₀ and N₃₀⁻₆₀ were higher in MG than in...
SOL. The contraction-induced change in shear modulus and alternate muscle activity during the prolonged contraction of LG, which had the lowest magnitude of muscle activation during the fatiguing task among the plantar flexors were not significantly different from those of the other muscles. These results suggest the correspondence of the degree of increase in muscle shear modulus induced by prolonged contraction to the frequency of alternate muscle activity especially in the latter half of the prolonged contraction between MG and SOL. Thus, based on the results of muscle shear modulus, it can be concluded that, compared with SOL, the alternate muscle activity of MG occurs more frequently—especially in the latter half of the prolonged contraction due to the greater increase in fatigue of MG induced by the progression of the fatiguing task.
REFERENCES
Akagi, R., Yamashita, Y., and Ueyasu, Y. (2015). Age-related differences in muscle shear moduli in the lower extremity. *Ultrasound Med. Biol.* 41, 2906–2012. doi: 10.1016/j.ultrasmedbio.2015.07.011
Akima, H., Saito, A., Watanabe, K., and Kouzaki, M. (2012). Alternate muscle activity patterns among synergists of the quadriceps femoris including the vasti intermedius during low-level sustained contraction in men. *Muscle Nerve* 46, 86–95. doi: 10.1002/mus.23268
Descarreaux, M., Lafond, D., and Cantin, V. (2010). Changes in the flexion-relaxation response induced by hip extensor and erector spine muscle fatigue. *BMC Musculoskelet. Disord.* 11:112. doi: 10.1186/1471-2474-11-112
Dietrich, A. V., Andrade, R. J., Le Sant, G., Falla, D., Petzke, F., Hug, F., et al. (2017). Shear wave elastography reveals different degrees of passive and active stiffness of the neck extensor muscles. *Eur. J. Appl. Physiol.* 117, 171–178. doi: 10.1007/s00421-016-3509-5
Dresner, M. A., Rose, G. H., Rossman, P. J., Muthupillai, R., Eby, S. F., Cloud, B. A., Brandenburg, J. E., Giambini, H., Song, P., Chen, S., Fernandez-del-Olmo, M., Rodriguez, F. A., Marquez, G., Iglesias, X., Marina, M., Fukunaga, T., Roy, R. R., Shellock, F. G., Hodgson, J. A., Day, M. K., McDougal, A., and Ehman, R. L. (2001). Magnetic resonance elastography of skeletal muscle. *J. Magn. Reson. Imaging* 13, 269–276. doi: 10.1002/1522-2586(20010213)13:2<269::AID-JMRI1039>3.0.CO;2-1
Eby, S. F., Cloud, B. A., Brandenburg, J. E., Giambini, H., Song, P., Chen, S., et al. (2015). Shear wave elastography of passive skeletal muscle stiffness: influences of sex and age throughout adulthood. *Clin. Biomech.* 30, 22–27. doi: 10.1016/j.clinbiomech.2014.11.011
Enoka, R. M., and Stuart, S. G. (1992). Neurobiology of muscle fatigue. *J. Appl. Physiol.* 72, 1631–1648.
Fernandez-del-Olmo, M., Rodriguez, F. A., Marquez, G., Iglesias, X., Marina, M., Benitez, A., et al. (2013). Isometric knee extensor fatigue following a Wingate test: peripheral and central mechanisms. *Scand. J. Med. Sci. Sports* 23, 57–65. doi: 10.1111/j.1600-0838.2011.01355.x
Fischer, A. A. (1987). Clinical use of tissue compliance meter for documentation of soft tissue pathology. *Clin. J. Pain* 3, 23–30. doi: 10.1097/00002508-19870310-00005
Fukunaga, T., Roy, R. R., Shellyock, F. G., Hodgson, J. A., Day, M. K., Lee, P. L., et al. (1992). Physiological cross-sectional area of human leg muscles based on magnetic resonance imaging. *J. Orthop. Res.* 10, 928–934. doi: 10.1002/jor.110010623
Gandevia, S. C. (2001). Spinal and supraspinal factors in human muscle fatigue. *Physiol. Rev.* 81, 1725–1789. Available online at: http://physrev.physiology.org/content/81/4/1725
Gong, B., Legault, D., Miki, T., Seino, S., and Renaud, J. M. (2003). K_{ATP} channels depress force by reducing action potential amplitude in mouse EDL and soleus muscle. *Am. J. Physiol. Cell Physiol.* 285, C1464–C1474. doi: 10.1152/ajpcell.00728.2003
Gong, B., Miki, T., Seino, S., and Renaud, J. M. (2000). A K_{ATP} channel deficiency affects resting tension, not contractile force, during fatigue in skeletal muscle.
**AUTHOR CONTRIBUTIONS**
Conceived and designed the experiments: RA and RE. Performed experiment: TF, MK, and MN. Analyzed data: RA and TF. Drafted manuscript and prepared tables and figures: RA. All authors interpreted results of research, edited, critically revised, and approved the final version of manuscript, and have agreed to be accountable for all aspects of the work related to its accuracy and integrity.
**FUNDING**
This study was partly supported by JSPS KAKENHI Grant Number JP16H05918 [Grant-in-Aid for Young Scientists (A)].
Semmler, J. G., Kutzscher, D. V., and Enoka, R. M. (1999). Gender differences in the fatigability of human skeletal muscle. *J. Neurophysiol.* 82, 3590–3593.
Semmler, J. G., Kutzscher, D. V., and Enoka, R. M. (2000). Limb immobilization alters muscle activation pattern during a fatiguing isometric contraction. *Muscle Nerve* 23, 1381–1392. doi: 10.1002/1097-4598(200009)23:9<1381::AID-MUS9>3.0.CO;2-5
Sjøgaard, G., Kiens, B., Jørgensen, K., and Saltin, B. (1986). Intramuscular pressure, EMG and blood flow during low-level prolonged static contraction in man. *Acta Physiol. Scand.* 128, 475–484. doi: 10.1111/j.1748-1716.1986.tb08002.x
Sypert, G. W., and Munson, J. B. (1981). Basis of segmental motor control: motoneuron size or motor unit type? *Neurosurgery* 8, 608–621. doi: 10.1227/00006123-198105000-00020
Tamaki, H., Kirimoto, H., Yotani, K., and Takekura, H. (2011). Frequent alternate muscle activity of plantar flexor synergists and muscle endurance during low-level static contractions as a function of ankle position. *J. Physiol. Sci.* 61, 411–419. doi: 10.1007/s12576-011-0157-8
Tamaki, H., Kutada, K., Akamine, T., Murata, F., Sakou, T., and Kurata, H. (1998). Alternate activity in the synergistic muscles during prolonged low-level contractions. *J. Appl. Physiol.* 84, 1943–1951.
Ward, S. R., and Lieber, R. L. (2005). Density and hydration of fresh and fixed human skeletal muscle. *J. Biomech.* 38, 2317–2320. doi: 10.1016/j.jbiomech.2004.10.001
**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Copyright © 2017 Akagi, Fukui, Kubota, Nakamura and Ema. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2270,
1
],
[
2270,
8671,
2
],
[
8671,
12900,
3
],
[
12900,
19065,
4
],
[
19065,
22124,
5
],
[
22124,
24163,
6
],
[
24163,
25862,
7
],
[
25862,
30305,
8
],
[
30305,
37298,
9
],
[
37298,
41852,
10
],
[
41852,
44029,
11
]
]
} |
a685ef295ac6f08c272a06f436c62272027f57f6 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 6,
"total-fallback-pages": 0,
"total-input-tokens": 21094,
"total-output-tokens": 8184,
"fieldofstudy": [
"Medicine",
"Biology"
]
} | Expression and Clinical Significance of ANXA1 and DICER1 in Myelodysplastic Syndromes
Wenqi Wu, Chengyao Wan, Qiongni Xie, Xiaolin Liang, Jing Wen and Zhenfang Liu
Department of Hematology, The First Affiliated Hospital of Guangxi Medical University, Nanning, Guangxi, China
ABSTRACT
Objective: To explore the expression status of ANXA1 and DICER1 and their clinical significance in myelodysplastic syndromes (MDS) patients.
Study Design: A case control study.
Place and Duration of Study: Department of Hematology, The First Affiliated Hospital of Guangxi Medical University, Nanning, Guangxi, China, from January 2011 to June 2020.
Methodology: Quantitative real-time fluorescence PCR (qRT-PCR) was carried out to detect ANXA1 and DICER1 expression levels in the bone marrow from 49 MDS patients and 12 healthy volunteers as control. The correlation of the clinical parameters and ANXA1 or DICER1 expression was then analysed in MDS.
Results: Compared with normal controls, the expression of bone marrow ANXA1 and DICER1 in MDS patients was significantly decreased, especially in patients with secondary acute myeloid leukemia (s-AML). Moreover, down-expression of ANXA1 had a great effect on differentiating s-AML subjects from MDS cases, and DICER1 expression showed good performance to screen MDS subjects from normal controls. ANXA1 expression was negatively correlated with clinical features of poor prognosis such as the percentage of bone marrow blasts, IPSS and WHO subtypes.
Conclusions: Bone marrow ANXA1 may be a potential biomarker for the risk prediction of leukemia transformation in MDS. DICER1 may have diagnostic significance to MDS.
Key Words: Myelodysplastic syndrome, Secondary acute myeloid leukemia, ANXA1, DICER1.
How to cite this article: Wu W, Wan C, Xie Q, Liang X, Wen J, Liu Z. Expression and Clinical Significance of ANXA1 and DICER1 in Myelodysplastic Syndromes. J Coll Physicians Surg Pak 2020; 30(12):1291-1296
INTRODUCTION
Myelodysplastic syndromes (MDS), which originates from hematopoietic stem cell, is a clonal malignant disorder characterised by dysplastic hyperplasia in one or more cell lineages, ineffective hematopoiesis, refractory hemocytopenia and a variable inclination to develop into acute myeloid leukemia (AML). The incidence of MDS is approximately 4/100000 population/year, but it is principally a disease of the elderly (over 70 years old) with an incidence of more than 0.03%/year. The clinical manifestations and prognosis of patients are considerably different. Correct and early diagnosis and treatment of MDS are often extremely difficult.
malignancies, which regulate suppressors or oncogenes to facilitate cancer biologies such as tumor growth, invasion, angiogenesis, and immune evasion. Recently, it is reported that part miRNAs were global down expressed in bone marrow mesenchymal stromal cells (MSCs) of MDS, suggesting that the abnormal expression of miRNA may be involved in the pathogenesis of MDS. Authors’ preliminary work discovered that the expression of BM miR-196b in MDS patients increased significantly and rose with the disease risk. Bioinformatics techniques were used to predict potential target genes of miR-196b. We verified the reliability and validity of some genes by quantitative real-time fluorescence PCR (qRT-PCR) and found that human Annexin A1 (ANXA1) and DICER1 were differentially expressed in bone marrow from MDS patients.
However, the patterns of bone marrow ANXA1 and DICER1 expression and their clinical significance in MDS remain unknown. To address this problem, the aim of the current study was to explore the expression and clinical value of ANXA1 and DICER1 in MDS patients.
**METHODOLOGY**
The study was approved by the Human Ethics Committee of The First Affiliated Hospital of Guangxi Medical University. A total of 61 bone marrow samples were obtained from 38 MDS patients before any chemotherapy, 11 MDS patients who subsequently developed into secondary acute myeloid leukemia (s-AML) and 12 healthy volunteers from 2011 to 2014 in The First Affiliated Hospital of Guangxi Medical University. All MDS patients received regular follow-up and the follow-up period was up to June 2020 (84 months). The patients who had other hematological diseases or malignant tumors were excluded. Twelve healthy volunteers had no obvious abnormalities in all examination indexes. Written informed consents were obtained from all the participants according to the Declaration of Helsinki prior to bone marrow collection. MDS patients were classified according to the World Health Organization (WHO, 2016) Criteria. Refer to the International Prognostic Scoring System (IPSS), MDS patients were divided into three groups: relatively low-risk group (low risk + intermediate risk I, n=23), relatively high-risk group (intermediate risk II + high risk, n=15), and s-AML group (n=11). Detailed clinical features of 49 MDS patients were provided in Table I.
Up to 2ml BM sample was extracted from each participant and BM mononuclear cells was seperated by density gradient centrifugation. Total RNA was isolated from BM cells using TRIzol reagent (Invitgen, USA) according to the manufacturer’s instructions. RNA was then reverse transcribed to cDNA. Reverse transcription was performed with SuperScriptTM III Reverse Transcriptase (Invitrogen) on Gene Amp PCR System 9700 (Applied Biosystems). QRT-PCR was processed on ViIA 7 Real-time PCR System (Applied Biosystems) using 2X PCR master mix Kit (Arraystar). Conditions of qRT-PCR were as follows: 95°C for 10 min, followed by 40 PCR cycles (95°C for 10s) and 60°C for 1 min. The primers used in the present study were as follows: GAPDH forward: 5’GGGAAAATCTGTGCGTGA3’ and reverse: 5’GAGTGGGTTGTCGCTGTGTA3’; ANXA1 forward: 5’ACCTCAATCCATCCTCAG3’ and reverse: 5’TGACTCTGGACCTCTGTC3’; DICER1 forward: 5’TGCAATGTGAGACCGAATG3’ and reverse: 5’CATAGTAGGACTGGCGGGAAG3’. The relative ANXA1 and DICER1 expression was calculated with the comparative 2−ΔΔCt method using housekeeping gene GAPDH as the endogenous normaliser.
As the data were not subjected to normal distribution, the relative ANXA1 and DICER1 expression among different groups was analysed using the Mann-Whitney U-test or Kruskal-Wallis test as appropriate. Quantitative variables were expressed as median (IQR). With the median expression of ANXA1 and DICER1 as the cut-off value, patients with MDS were divided into high expression group and low expression group. Qualitative variables were expressed as frequencies and percentages. Pearson Chi-square test or Fisher’s exact probabilities was employed to detect the relationship between the expression of ANXA1 or DICER1 and clinical characteristics. And r value was used to evaluate the degree of linear correlation between two variables. Receiver operating characteristic (ROC) curve and the area under the curve (AUC), sensitivity and specificity were used to assess the diagnosis value of bone marrow ANXA1 and DICER1 expression. Kaplan-Meier survival analysis was carried out to explore the relationship between the expression of ANXA1/DICER1 and patient clinical outcome. The statistical analyses were processed with SPSS 25.0 (SPSS Inc, Chicago, IL, USA) and GraphPad Prism 7 (GraphPad Software Inc., La Jolla, CA, USA) software. Results were considered statistically significant when p-value <0.05.
**RESULTS**
QRT-PCR results revealed that bone marrow ANXA1 expression in each subgroup of MDS was significantly lower than normal controls (p<0.05, Figure 1A). And s-AML patients had dramatically lower ANXA1 expression than patients with relatively low-risk MDS and patients with relatively high-risk MDS (p<0.05, Figure 1B). More importantly, ROC curve analysis showed that ANXA1 was a potential indicator for distinguishing s-AML patients from MDS patients with AUC of 0.799, and the sensitivity and specificity were 81.82% and 84.21%, respectively (p=0.003, Figure 2).
On the other hand, the expression of DICER1 from each MDS subgroup was obviously more reduced compared to normal controls, especially in the s-AML group (all p <0.01, Figure 1C-D). ROC curve analysis showed that compared to healthy controls, low-expression of DICER1 may have the diagnostic reference value for MDS as the AUC value was 0.821 and its sensitivity and specificity were 66.67% and 79.59%, respectively (p=0.001, Figure 3). The relative expressions of ANXA1 and DICER1 were shown in Table II.
The median values of absolute neutrophil counts (ANC), hemoglobin and platelets were used as the cut-off values to divide the MDS patients into high group and low group. As shown in Table I, ANXA1 expression was negatively correlated to the percentage of bone marrow (BM) blasts (r=−0.315, p=0.027, WHO classification (r=−0.259, p=0.009), IPSS (r=−0.365, p=0.011) in MDS patients.
Expression and clinical significance of ANXA1 and DICER1 in myelodysplastic syndromes
Table I: Association between different clinical features and ANXA1 or DICER1 expression in MDS patients.
| Clinical variables | ANXA1 Expression | DICER1 Expression |
|--------------------|------------------|-------------------|
| | Low(n=25) | High(n=24) | P | Low(n=25) | High(n=24) | P |
| Gender | | | 0.196 | | | 0.913 |
| Male | 18(72%) | 13(54.2%) | | 16(64%) | 15(62.5%) | |
| Female | 7(28%) | 11(45.8%) | 0.027 | 9(36%) | 9(37.5%) | 0.296 |
| BM blasts (%) | | | 0.027 | | | 0.321 |
| <5 | 11(44%) | 18(75%) | | 13(52%) | 16(66.7%) | |
| >5 | 14(56%) | 6(25%) | | 12(48%) | 8(33.3%) | |
| PLT(×109/L) | | | 0.195 | | | 0.879 |
| <50 | 14(56%) | 9(37.5%) | | 10(40%) | 13(54.2%) | |
| >50 | 11(44%) | 15(62.5%) | | 15(60%) | 11(45.8%) | |
| Hb(g/L) | | | 0.674 | | | 0.879 |
| <70 | 14(56%) | 12(50%) | | 13(52%) | 13(54.2%) | |
| >70 | 11(44%) | 12(50%) | | 12(48%) | 11(45.8%) | |
| ANC (×109/L) | | | 0.199 | | | 0.889 |
| =5 | 15(60%) | 10(41.7%) | | 13(52%) | 12(50%) | |
| >5 | 10(40%) | 14(58.3%) | | 12(48%) | 12(50%) | |
| WHO subtype | | | 0.009 | | | 0.036 |
| MDS-SLD | 2(8%) | 1(4.2%) | | 1(4%) | 2(8.3%) | |
| MDS-MLD | 9(36%) | 13(54.2%) | | 11(44%) | 11(45.9%) | |
| MDS-EB | 4(16%) | 9(37.5%) | | 5(20%) | 8(33.3%) | |
| s-AML | 10(40%) | 1(4.2%) | | 8(32%) | 3(12.5%) | |
| IPSS | | | 0.011 | | | 0.258 |
| low+INT1 | 9(36%) | 14(58.3%) | | 10(40%) | 13(54.2%) | |
| high+INT2 | 6(24%) | 9(37.5%) | | 7(28%) | 8(33.3%) | |
| s-AML | 10(40%) | 1(4.2%) | | 8(32%) | 3(12.5%) | |
| Cytogenetics | | | 0.312 | | | >0.999|
| Favorable | 15(60%) | 13(54.2%) | | 14(56%) | 14(58.3%) | |
| Intermediate | 0(0%) | 3(12.5%) | | 1(4%) | 2(8.3%) | |
| Unfavorable | 6(24%) | 8(33.3%) | | 7(28%) | 7(29.2%) | |
There was no correlation between ANXA1 expression and other clinical parameters, including gender, cytogenetics, hemoglobin levels, platelet counts and ANC nevertheless. In like manner, however, there was no relevance between DICER1 expression and these clinical parameters. During the 84-month follow-up, there was no statistically significant difference in overall survival or patient outcome between high expression group and low expression group of ANXA1 or DICER1 (p>0.05).
Table II: The relative expression of ANXA1 and DICER1 in subgroups of MDS and normal controls.
| Group | Patients (N=61) | ANXA1 Expression | DICER1 Expression |
|------------------|----------------|------------------|------------------|
| MDS groups | n=49 | 0.534(0.589) | 0.822(1.419) |
| Relatively low-risk group | n=23 | 0.610(0.492) | 0.952(1.612) |
| Relatively high-risk group | n=15 | 0.624(0.596) | 0.920(1.326) |
| s-AML group | n=11 | 0.252(0.057) | 0.322(1.425) |
| Normal controls | n=12 | 1.150(0.976) | 2.342(3.209) |
DISCUSSION
MDS is a highly heterogeneous hematopoietic malignancy with a high risk of transformation to AML. At present, the occurrence and progression of MDS are complex multi-gene, multi-stage pathological processes, and its pathogenesis remains elusive. DICER1, belongs to RNase III family, is extensively expressed in various tissues and plays a crucial part in the maturation of miRNA. Accumulated evidence has indicated that DICER1 expression is different among tumors,
and varies according to the development stages in the same tumor. For instance, DICER1 was down-regulated in liver cancer, cervical cancer, ovarian cancer, and breast cancer, while over-expressed in cutaneous melanoma, prostate cancer and rectal cancer.
In the current study, bone marrow DICER1 expression in MDS patients was significantly down-regulated compared to normal controls. These results were consistent with other studies regarding MDS. Hakan et al. found that DICER1 expression in multipotent mesenchymal stem cells from MDS and AML patients was lower than healthy controls, and gradually decreased from healthy controls to AML. Some differentially expressed miRNAs suggested that DICER1 may be involved in the pathogenesis of MDS and AML, which may become a new target gene for therapy. In addition, Raaij-makes et al. discovered that the specific deletion of DICER1 in mouse osteoprogenitor cells can give rise to hemocytopenia, myelodysplasia and induce into MDS and s-AML. Subsequently, studies confirmed that DICER1 gene expression in MDS patients was lower than healthy controls.
Figure 1A: The relative expression of ANXA1 in each MDS subgroup and normal controls. low+INT-1, low risk+ intermediate risk I. INT-2+high: intermediate risk II+ high risk.
Figure 1B: The relative expression of ANXA1 in relatively low + high risk MDS groups and s-AML group.
In the current study, bone marrow DICER1 expression in MDS patients was significantly down-regulated compared to normal controls. These results were consistent with other studies regarding MDS. Hakan et al. found that DICER1 expression in multipotent mesenchymal stem cells from MDS and AML patients was lower than healthy controls, and gradually decreased from healthy controls to AML. Some differentially expressed miRNAs suggested that DICER1 may be involved in the pathogenesis of MDS and AML, which may become a new target gene for therapy. In addition, Raaij-makes et al. discovered that the specific deletion of DICER1 in mouse osteoprogenitor cells can give rise to hemocytopenia, myelodysplasia and induce into MDS and s-AML. Subsequently, studies confirmed that DICER1 gene expression in MDS patients was lower than healthy controls.
Figure 1C: The relative expression of DICER1 in each MDS subgroup and normal controls.
Figure 1D: The relative expression of DICER1 in three MDS groups and normal controls.
Figure 2: ROC analysis using ANXA1 for screening AML cases from MDS patients.
In this study bone marrow DICER1 expression in MDS and s-AML patients was greatly lower than healthy controls, especially in s-AML. In MDS subjects, the expression level of DICER1 was significantly down-regulated. ROC analysis showed that DICER1 low-expression could distinguish MDS cases from normal controls. Thus, we believe that DICER1 may be a potential biomarker for accurate diagnosis for MDS.
ANXA1, a calcium-dependent phospholipid-binding protein, is the first member of the Annexin family. Since its discovery, ANXA1 was considered to involve in anti-inflammatory response, cell differentiation, proliferation and apoptosis, signal regulation and other cellular biological activities. Recent studies have shown that ANXA1 may play multiple roles in the occurrence and development of malignancies at different levels (from cancer initiation to metastasis).
ANXA1 expression in various types of tumors was contradictory. Gao. et al. showed that ANXA1 was down-expressed in gastric and bile duct cancers. Forced-expressed ANXA1 in gastric cancer cells strongly restrained cell growth and modulated COX-2 expression. Similar studies were observed in other tumors. However, ANXA1 was up-expressed in some cancer types such as hepatocellular carcinoma, non-small cell lung cancer, melanoma and esophageal cancer.
As for hematologic malignancies, Sabran et al. found that ANXA1 expression in acute lymphoblastic leukemia (Jurkat), AML (U937) and chronic myeloid leukemia (K562) cell lines were visibly higher compared to peripheral blood mononuclear cells (PBMCs) from normal controls. Moreover, ANXA1 expression in U937 cells was lower than in K562 and Jurkat cells, indicating that ANXA1 can be used as a biomarker to distinguish AML from healthy people. The expression of ANXA1 was up-regulated in hairy cell leukemia (HCL). Brunangelo. et al did immunostaining in 500 B-cell tumors with a specific anti-ANXA1 monoclonal antibody and demonstrated that ANXA1 expression was unique to HCL.
From these results, it is speculated that the expression of ANXA1 is abnormal in MDS. However, studies on ANXA1 in bone marrow of MDS patients are scant. This study showed that compared with relatively low-risk MDS group, ANXA1 expression was down-regulated in relatively high-risk MDS patients. Furthermore, in s-AML patients, the expression was more decreased. As a result, ANXA1 expression in MDS patients is negatively correlated with the risk of leukemia transformation. At present, the IPSS risk score of MDS is mainly determined by the percentage of BM blasts, cytogenetics and peripheral blood. Hence, this study analysed the correlation between the relative expression of ANXA1 and these clinical parameters. Results indicated that ANXA1 expression in MDS patients was negatively correlated with the percentage of BM blasts, IPSS and WHO stratification, but did not correlate with cytogenetics and peripheral blood. As the percentage of BM blasts, IPSS and WHO stratification are important indices for MDS disease classification, risk classification, disease progression and prognosis, low expression of ANXA1 may contribute to the disease progression and poor prognosis in MDS, suggesting that ANXA1 may be conducive to early diagnosis and management of MDS. In addition, the ROC curve suggested that ANXA1 down-expression may lead to the occurrence of AML, which can be used to distinguish s-AML patients from MDS patients. In conclusion, we speculate that ANXA1 may play a crucial role in the transformation from MDS to AML.
Unfortunately, the univariate survival analysis showed that the expression of ANXA1 or DICER1 was not correlated with prognosis. The reasons may be sufficient sample size and the therapy of MDS patients is not unified, thus the survival time cannot be compared. Next, the authors will expand the sample size, strictly screen and follow up the eligible patients and then analyze the correlation between the disease prognosis and ANXA1 or DICER1.
**CONCLUSION**
Taken together, this data provide a convincing evidence that bone marrow ANXA1 and DICER1 expressions in MDS patients are significantly decreased compared to healthy controls, and their abnormal expressions may play an important role in MDS pathogenesis. ANXA1 can be used as a potential biomarker to predict the leukemia transformation risk of MDS, and DICER1 may contribute to the diagnosis of MDS.
**FUNDING:**
This work was supported by the National Natural Science Foundation of China (No. 81560028 and No.81160072).
**CONFLICT OF INTEREST:**
The authors have no conflicts of interest to declare.
**ETHICAL APPROVAL:**
This study was approved by the Human Ethics Committee of The First Affiliated Hospital of Guangxi Medical University, China. (2020, KY-E-101).
AUTHORS’ CONTRIBUTION:
WW, ZL: Conception or design of the work.
WW, QX, XL, JW: Acquisition and analysis of data.
WW, CW, QX, XL, JW: Drafting the work.
CW, ZL: Final approval of the version to be published.
REFERENCES
1. Killick SB, Bown N, Cavenagh J, Dokal I, Foukaneli T, Hill A, et al. Guidelines for the diagnosis and management of adult aplastic anaemia. Br J Haematol 2016; 172(2):187-207. doi:10.1111/bjh.13853.
2. Wang J, Chen J, Sen S. MicroRNA as Biomarkers and Diagnostics. J Cell Physiol 2016; 231(1):25-30. doi:10.1002/ jcp.25056.
3. Wu SG, Chang TH, Liu YN, Shih JY. MicroRNA in Lung Cancer Metastasis. Cancers (Basel) 2019; 11(2):265. doi:10.3390/ cancers11020265.
4. Santamaría C, Muntiño S, Rosón B, Blanco B, López-Villar O, Carrancio S, et al. Impaired expression of Dicer, Drosha, Sbds and some microRNAs in mesenchymal stromal cells from myelodysplastic syndrome patients. Haematologica 2012; 97(8):1218-24. doi:10.3324/haematologica.2011.054437.
5. Zhu Y, Cai L, Guo J, Chen N, Yi X, Zhao Y, et al. Depletion of Dicer promotes epithelial ovarian cancer progression by elevating Pdia3 expression. Tumour Biol 2016; 37(10):14009-23. doi:10.1007/s13277-016-5218-4.
6. Poursadegh Zonouzi AA, Nejatizadeh A, Rahmani-Yamchi M, Fardmanesh H, Shakerizadeh S, Poursadegh Zonouzi A. Dysregulated expression of Dicer in invasive ductal breast carcinoma. Med Oncol 2015; 32(7):203. doi:10.1007/s12032-015-0643-6.
7. Hoffend NC, Magner WJ, Tomasi TB. The modulation of Dicer regulates tumor immunogenicity in melanoma. Oncotarget 2016; 7(30):47663-73. doi:10.18632/oncotarget.10273.
8. Bian XJ, Zhang GM, Gu CY, Cai Y, Wang CF, Shen YJ, et al. Down-regulation of dicer and ago2 is associated with cell proliferation and apoptosis in human prostate cancer. J Cell Physiol 2014; 231(7):852-7. doi:10.1002/jcp.25056.
9. Ozdogan H, Gur Dedeoglu B, Oztemur Islakoglu Y, Aydos A, Atalay A, Kose S, Atalay A, et al. Dicer1 gene and miRNA dysregulation in mesenchymal stem cells of patients with myelodysplastic syndrome and acute myeloblastic leukemia. Leuk Res 2017; 63:62-71. doi:10.1016/j.leukres.2017.10.006.
10. Raajmakers MH, Mukherjee S, Guo S, Zhang S, Kobayashi T, Schoonmaker JA, et al. Bone progenitor dysfunction induces myelodysplasia and secondary leukemia. Nature 2010; 464(7290):852-7. doi:10.1038/nature08851.
11. Geyh S, Oz S, Cadeddu RP, Fröbel J, Brückner B, Kündgen A, et al. Insufficient stromal support in MDS results from molecular and functional deficits of mesenchymal stromal cells. Leukemia 2013; 27(9):1841-51. doi:10.1038/ leu.2013.193.
12. Tebar F, Gelabert-Baldrich M, Hoque M, Cairns R, Rentero C, Pol A, et al. Annexins and endosomal signaling. Methods Enzymol 2014; 535:55-74. doi:10.1016/B978-0-12-397925-4.00004-3.
13. Sheikh MH, Solito E. Annexin A1: Uncovering the many talents of an old protein. Int J Mol Sci 2018; 19(4):1045. doi:10.3390/ijms19041045.
14. Boudhraa Z, Bouchon B, Viaillard C, D’Incan M, Degoul F. Annexin A1 localization and its relevance to cancer. Clin Sci (Lond) 2016; 130(4):205-20. doi:10.1042/CS20150415.
15. Gao Y, Chen Y, Xu D, Wang J, Yu G. Differential expression of ANXA1 in benign human gastrointestinal tissues and cancers. BMC Cancer 2014; 14:520. doi:10.1186/ 1471-2407-14-520.
16. Zhu DW, Yang X, Yang CZ, Ma J, Liu Y, Yan M, et al. Annexin A1 down-regulation in oral squamous cell carcinoma correlates to pathological differentiation grade. Oral Oncol 2013; 49(6):542-50. doi:10.1016/j.oraloncology.2013.01.002.
17. Mu D, Gao Z, Guo H, Zhou G, Sun B. Sodium butyrate induces growth inhibition and apoptosis in human prostate cancer DU145 cells by up-regulation of the expression of annexin A1. PLoS One 2013; 8(9):e74922. doi:10.1371/journal. pone.0074922.
18. Álvarez-Teijeiro S, Menéndez ST, Villaronga MA, Pena-Alonso E, Rodrigo JP, Morgan RO, et al. Annexin A1 down-regulation in head and neck squamous cell carcinoma is mediated via transcriptional control with direct involvement of miR-196a/b. Sci Rep 2017; 7(1):6790. doi:10.1038/ s41598-017-07169-w.
19. Lin Y, Lin G, Fang W, Zhu H, Chu K. Increased expression of annexin A1 predicts poor prognosis in human hepatocellular carcinoma and enhances cell malignant phenotype. Med Oncol 2014; 31(12):327. doi:10.1007/s12032-014-0327-7.
20. Fang Y, Guan X, Cai T, Long J, Wang H, Xie X, et al. Knockdown of ANXA1 suppresses the biological behavior of human NSCLC cells in vitro. Mol Med Rep 2016; 13(5):3858-66. doi:10.3892/mmr.2016.5022.
21. Boudhraa Z, Rondepierre F, Ouchchane L, Kintossou R, Trzeciakiewicz A, Franck F, et al. Annexin A1 in primary tumors promotes melanoma dissemination. Clin Exp Metastasis 2014; 31(7):749-60. doi:10.1007/s10585-014-9665-2.
22. Han G, Lu K, Huang J, Ye J, Dai S, Ye Y, et al. Effect of Annexin A1 gene on the proliferation and invasion of esophageal squamous cell carcinoma cells and its regulatory mechanisms. Int J Mol Med 2017; 39(2):357-63. doi:10.3892/ijmm.2016.2840.
23. Sabran A, Kumolosasi E, Jantan I. Effects of annexin A1 on apoptosis and cell cycle arrest in human leukaemic cell lines. Acta Pharm 2019; 69(1):75-86. doi:10.2478/acph-2019-0005.
24. M., Tiacci E, Liso A, Basso K, Sabattini E, Pacini R, et al. Simple diagnostic assay for hairy cell leukaemia by immunocytochemical detection of annexin A1 (ANXA1). Lancet 2004; 363(9424):1869-70. doi:10.1016/S0140- 6736(04) 16356-3. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2613,
1
],
[
2613,
8834,
2
],
[
8834,
13200,
3
],
[
13200,
15675,
4
],
[
15675,
20444,
5
],
[
20444,
25839,
6
]
]
} |
bb3ab43ec57eec26b04d55b77640232cd1ad1069 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 5,
"total-fallback-pages": 0,
"total-input-tokens": 15722,
"total-output-tokens": 4680,
"fieldofstudy": [
"Computer Science"
]
} | Typical Cases of Annotators’ Disagreement in Discourse Annotations in Prague Dependency Treebank
Šárka Zikánová, Lucie Mladová, Jiří Mírovský, Pavlína Jínová
Charles University in Prague, Faculty of Mathematics and Physics
Institute of Formal and Applied Linguistics
Malostranské nám. 25, 118 00 Prague 1, Czech Republic
E-mail: {zikanova, mladova, mirovsky, jinova}@ufal.mff.cuni.cz
Abstract
In this paper, we present the first results of the parallel Czech discourse annotation in the Prague Dependency Treebank 2.0. Having established an annotation scenario for capturing semantic relations crossing the sentence boundary in a discourse, and having annotated the first sections of the treebank according to these guidelines, we report now on the results of the first evaluation of these manual annotations. We give an overview of the process of the annotation itself, which we believe is to a large degree language-independent and therefore accessible to any discourse researcher. Next, we describe the inter-annotator agreement measurement, and, most importantly, we classify and analyze the most common types of annotators’ disagreement and propose solutions for the next phase of the annotation. The annotation is carried out on dependency trees (on the tectogrammatical layer), this approach is quite novel and it brings us some advantages when interpreting the syntactic structure of the discourse units.
1. Introduction
Current discourse annotation of Czech texts in the Prague Dependency Treebank (PDT) brought out some general questions which are common for annotation of higher and more complicated units in texts. Contrary to e.g. morphological tagging on the word layer, annotation of discourse, coreference, direct-speech-attribution etc. encounters two basic problems:
1) In classical linguistics, the system of “higher” language levels is usually less described than morphology and syntax. Therefore, there is a large amount of unclear cases at the beginning of annotation of these structures, where the very existence of a discourse/coreference/etc. relation in a single case is put under question.
2) The extent of the discourse units (discourse arguments) is formally almost non-predictable and it can depend on the annotators’ understanding of the text.
Both these problems influence the measurement of inter-annotator agreement. In this paper, we present examples how we solve typical problematic structures in discourse annotation of Czech. Since our solutions should lead to a clear and reusable system, they are generally based on two principles: first, the nature of a “discourse relation” is getting more strict and gets into an opposition to other types of text relations (especially to coreference). Second, the syntactic structure of discourse arguments is taken into account.
The discourse annotation in PDT is linked to the previous layers of annotations, such as morphological analysis, syntactico-semantic analysis (so called tectogrammatics) including topic-focus articulation and some coreference types (Hajič et al., 2006). The aim of the discourse annotation is to indicate semantic relations crossing the sentence boundary (see Mladová et al., 2008). A discourse relation, be it expressed explicitly by means of a discourse connective, or implicitly, connects two “discourse arguments” (abstract objects, i.e. independent events, expressed mainly by independent clauses, cf. Asher, 1993). Every single discourse argument has a certain partial semantic feature, building together with its counterpart argument the whole of a discourse relation, e.g. the relation of reason links the Argument1 expressing the reason itself to the Argument2 expressing the fact. In the first phase, the annotation is limited to the relations that are expressed by explicit discourse connectives (coordinating and subordinating conjunctions, particles, adverbs etc., cf. Prasad et al. 2007, 2008). This temporary formal restriction helps us understand better the character of discourse relations, set the annotation scenario clearly and train annotators. Implicit discourse relations will be annotated in further phases.
2. The process of discourse annotation in the Prague Dependency Treebank
PDT contains journalistic texts of all kinds, including e.g. sport results and television programs. For discourse annotation training, larger narrative texts (30 sentences and more) were selected, in which a higher occurrence of discourse relations can be assumed. Annotators have at their disposal both plain text and the tectogrammatical analysis (tree structures). Annotation is carried out on the tectogrammatical trees (since we do not want to lose connection with the analyses of previous levels); however, its representation for annotators is very close to the plain text.
Annotators first search in the plain texts for possible discourse connectives and arguments of the connectives. Then they mark the assumed extent of discourse arguments on the tectogrammatical layer, link them with a discourse relation, and choose from the list of possible semantic types of the discourse relation. The appropriate discourse connective is also marked.
3. Evaluation of parallel annotations
In order to evaluate the inter-annotator agreement on selected text annotated by two or more annotators, we use F$_1$-measure for the agreement on arrows, types and connectives (their various combinations), and Cohen’s $\kappa$ (Cohen, 1960) for the agreement on types of arrows. By the agreement on arrows we mean agreement on the start and target nodes. Cohen’s $\kappa$ is used for measuring the agreement on types of those arrows where the annotators agreed on the start and target nodes. Tables 1 and 2 show results of three subsequent measurements (each performed on different data). The measurement #1 shows the average inter-annotator agreement between each of three annotators and an exemplar annotation. The measurements #2 and #3 show the agreement between two selected annotators.
![Image]
Table 1: The inter-annotator agreement on arrows, on arrows and types, and on arrows and connectives
| Measurement | F$_1$ on arrows | F$_1$ on arrows and types | F$_1$ on arrows and connectives |
|-------------------|----------------|--------------------------|--------------------------------|
| Measurement #1 | 0.43 | 0.3 | 0.35 |
| (74 sentences) | | | |
| Measurement #2 | 0.44 | 0.39 | 0.44 |
| (71 sentences) | | | |
| Measurement #3 | 0.55 | 0.39 | 0.5 |
| (68 sentences) | | | |
Table 2: The inter-annotator agreement on arrows, types and connectives, and on types
| Measurement | F$_1$ on arrows, types and connectives | Cohen’s $\kappa$ on types |
|-------------------|----------------------------------------|--------------------------|
| Measurement #1 | 0.22 | 0.63 |
| (74 sentences) | | |
| Measurement #2 | 0.39 | 0.81 |
| (71 sentences) | | |
| Measurement #3 | 0.33 | 0.59 |
| (68 sentences) | | |
This first attempt at inter-annotator agreement confirms the general feelings that the taste of discourse annotation is more difficult than an annotation of lower levels in that it relies to a greater extent on individual annotators’ interpretation of a broader context. If some of the restrictions are relaxed, the figures demonstrate a certain improvement, see below Sect. 4.3.
4. Cases of typical disagreement
The first evaluation of parallel annotations of selected texts brought up some interesting observations. Reflecting the results, we were able to distinguish several repeatedly occurring problematic issues in the annotations. The nature of these disagreements corresponds to the general problem of a formal description on such a high level of language, namely – the texts sometimes allow for different, equally relevant interpretations. So, as for the annotators, two general issues appeared to be difficult to decide: where the connective indeed connects two discourse-relevant text units, and, second, what is the exact extent of these units (arguments of the relation). These issues are closely analyzed in the sections 4.1 to 4.3, with real-data examples.
4.1 Semantic types of discourse relations
Contrary to our assumptions, a disagreement in the semantic type of the assigned relation is not so frequent. In other words, when annotators recognize presence of a discourse connective and determine the discourse arguments, all of them usually mark up the same type of the semantic relation.
4.2 Discourse and non-discourse relations: NPs and elided verbs
However, there was a relatively high degree of disagreement in the very recognition of a discourse relation in some typical cases. (The further examples represent the most common questions of the annotators.)
A trivial example is the fact that expressions acting as discourse connectives can be used in non-discourse contexts. Cf. He took his hat and went home. (discourse-relevant coordination) mother and father (discourse-irrelevant NP coordination) The disagreement occurs when it is not clear whether the potential discourse connective refers to the whole sentence as an independent abstract object (discourse argument), or just to its complement, typically an NP. This ambiguity is common in sentences including verbs with a vague, general meaning (cf. 1; discourse connective is in bold).
(1) [Arg1: The case has several problematic points.] [Arg2: The first is the fact that although a female candidate succeeded in the entrance test better than male candidates (there were 17 accepted with worse results), she has not been accepted to study precisely and only because she is a woman.] (specification)
According to one of the possible interpretations, the second sentence of the example (1) is a specification of the content of the first sentence. In this case, the relation is considered a discourse-relevant relation. In another interpretation, the second sentence characterizes solely the NP several problematic points. Then the relation is not a matter of discourse analysis but
rather a relation of (one type of) coreference. Example (2) points to a similar situation.
(2) [Arg1: Při prohlídce střech Šternberského paláce si lze všimnout drobného, avšak charakteristického rozdílu mezi přístupem památkářů koncem 80. let a nyní:COLON] [Arg2: zatímco komínů staré sněmovny byly zbouřány jako zbytečné a zůstala jen holá střecha, dělníci KDM mají přikázáno komín všech čtyř objektů nejen ponechat, ale dokonce mírně přizdobit, aby tradiční kolorit malostranských střech čásem nezmizel.] (specification)
[Arg1: When observing the roofs of the Sternberg Palace it is possible to note a small, but distinctive difference between the approaches of preservationists of late 80's and now:COLON] [Arg2: while chimneys of the old Parliament were demolished as functionless and only a clear roof was retained, the KDM workers are ordered not only to maintain chimneys of all the four objects, but even to decorate them slightly, so that the traditional local atmosphere of Lesser Town roofs does not eventually disappear.] (specification)
In this case, the first argument involves either the whole clause, or just the NP a small, but distinctive difference between the approaches of preservationists of late 80's and now.
For a unification of annotation, we decided to consider the relations in these cases (1-2) as being coreferential rather than discourse-relevant.
In a similar vein, the existence of a discourse argument is often doubtful in structures with an elided verb in which a potential discourse connective occurs, cf. (3).
(3) Tato fakta svědčí i tom, že [Arg1: státní úředníci nemají dostatečný respekt,] [Arg2: možná snad ani představu o požadavcích Listiny základních práv a svobod]. (gradation)
These facts also suggest that [Arg1: state officials do not have enough respect,] [Arg2: perhaps not even an idea of the requirements of the Charter of Fundamental Rights and Freedoms]. (gradation)
A question arises whether in (3) the connective connects independent abstract objects (they have no respect and they have no idea, cf. Figure 1 (3a)), or just parts dependent on the verb that are not discourse arguments (they have no respect and no idea, cf. Figure 2 (3b)).
Semantic difference between the two possible interpretations of structures like (1-3) may often not be crucial. Nevertheless, it is crucial to catch the same linguistic phenomena in the same way and to set clear borders of discourse annotation, in order to provide systematic and coherent linguistic data.
4.3 Extent (scope) of a discourse argument: Verbs of thinking and speaking
In some cases, there are no doubts about the existence of a discourse relation, but the extent (scope) of the discourse argument is arguable. Typically, there is annotators’ disagreement in structures with governing verbs of thinking or speaking. Often it is not clear whether the discourse argument contains the governing verb or just the content of thought or speech (dictum), cf. (4).
(4)
[Arg1: Na tom, aby ve Šternberku ani v paláci Smiřických nevznikaly žádné příčky, trvají památkáři.] [Arg2: Poslancům tudiž nebude dopřáno žádné velké soukromí.] (reason)
[Arg1: Preservationists insist that no partition walls will be built up neither in the Sternberg Palace nor in the Smiřický Palace.] [Arg2: Therefore, MP’s will not enjoy great privacy.] (reason)
In one of the annotators’ interpretation of the discourse structure of (4), the governing verb is not included into the discourse argument:
[Arg1: No partition walls will be built up in the buildings.] [Arg2: Therefore, MP’s will have no privacy.]
In another solution the first argument is larger:
[Arg1: Preservationists insist that no partition walls will be built up in the buildings.] [Arg2: Therefore, MP’s will have no privacy.]
To ensure agreement, we recommended in these cases to take into consideration whether the meaning of the governing clause is substantial, i.e. whether it reflects an important operation to be carried out on the idea of the dependent clause. In (4) the governing verb is unambiguously a part of the discourse argument: it is necessary to know whether the preservationists insist on the idea, or for example, they forbid it.
To address this issue in the evaluation of the inter-annotator agreement, we have performed the same tests as before, this time allowing the annotators to disagree slightly either in the start or the target node of the arrows. By “slightly” we mean a difference of one level in the tree. For example, if node A is a parent of node B, then we consider arrows A→C and B→C to be in agreement, as well as arrows D→A and D→B. Tables 3 and 4 show results of the three measurements of the inter-annotator agreement, this time allowing for skipping one level at either the start or the target node.
| Measurement | F₁ on arrows | F₁ on arrows and types | F₁ on arrows and connectives |
|--------------|--------------|------------------------|-------------------------------|
| Measurement #1 (74 sentences) | 0.53 | 0.33 | 0.41 |
| Measurement #2 (71 sentences) | 0.5 | 0.44 | 0.5 |
| Measurement #3 (68 sentences) | 0.67 | 0.44 | 0.61 |
Table 3: The inter-annotator agreement on arrows, on arrows and types, and on arrows and connectives, with allowed skipping of one level either at the start or the target node
| Measurement | F₁ on arrows, types and connectives | Cohen’s κ on types |
|--------------|-----------------------------------|-------------------|
| Measurement #1 (74 sentences) | 0.24 | 0.56 |
| Measurement #2 (71 sentences) | 0.44 | 0.84 |
| Measurement #3 (68 sentences) | 0.39 | 0.53 |
Table 4: The inter-annotator agreement on arrows, types and connectives, and on types, with allowed skipping of one level either at the start or the target node
The numbers show improvement in agreement on arrows (about 10%) and on the combination with agreement on types and/or connectives (less than 5%), while Cohen’s κ – measured solely on types – has either slightly improved or worsened, depending on the measurement (i.e. on the material tested).
5. Conclusion
As demonstrated by the results of parallel annotations, it is crucial at this moment to distinguish discourse relations from other types of relations within the sentence and in the text. At this stage, the research of discourse semantic relations and unification of discourse annotation is closely linked to syntactic analysis. Setting of annotation scenario can be only done consistently with regard to the syntactic construction. Likewise, it is necessary to determine the extent (scope) of discourse arguments in definable cases on the basis of syntactic structure. In these tasks, the connection to the syntactico-semantic analysis of the tectogrammatical layer in the Prague Dependency Treebank appears as a rather convenient tool. It makes it possible to work with already established (and coherent) solutions of typical syntactic constructions, such as ellipses, coordinations etc.
6. Acknowledgements
The research reported in this contribution has been carried out under the grant projects of the Grant Agency of Czech Republic (GACR 405/09/0729, GACR 201/09/H057), the Center for Computational Linguistics CKL (LC536), the Czech Ministry of Education (MSM-0021620838, ME-10018), and the Grant Agency of Charles University in Prague (GAUK 103609).
7. References
Asher, N. (1993). *Reference to Abstract Objects in Discourse*. Dordrecht: Kluwer Academic Publishers.
Cohen, J. (1960). A coefficient of agreement for nominal scales. *Educational and Psychological Measurement*, 20(1), pp. 37-46.
Hajič, J. et al. (2006). *Prague Dependency Treebank 2.0*. Philadelphia: Linguistic Data Consortium.
Mladová, L., Zikánová, Š., Hajičová, E. (2008). From Sentence to Discourse: Building an Annotation Scheme for Discourse Based on Prague Dependency Treebank. In *Proceedings of the 6th International Conference on Language Resources and Evaluation (LREC 2008)*. Marrakech, Morocco: European Language Resources Association, ISBN 2-9517408-4-0, pp. 1-7.
Prasad, R. et al. (2007). The Penn Discourse Treebank 2.0 Annotation Manual.
Prasad, R., Dinish N., Lee A., Miltsakaki, E., Robaldo, L., Joshi, A., Webber, B. (2008). The Penn Discourse Treebank 2.0. In *Proceedings of the 6th International Conference on Language Resources and Evaluation (LREC 2008)*.
http://www.seas.upenn.edu/~pdtb/PDTBAPI/pdtb-annotation-manual.pdf | 2025-03-06T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
5171,
1
],
[
5171,
10793,
2
],
[
10793,
12997,
3
],
[
12997,
16863,
4
],
[
16863,
19209,
5
]
]
} |
bca9e62dbe3e70175b6a6a6d6c0adf240feee676 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 11,
"total-fallback-pages": 0,
"total-input-tokens": 23034,
"total-output-tokens": 3725,
"fieldofstudy": [
"Chemistry"
]
} | On the Promotion of Catalytic Reactions by Surface Acoustic Waves
Bernhard von Boehn, Michael Foerster, Moritz von Boehn, Jordi Prat, Ferran Macià, Blai Casals, Muhammad Waqas Khaliq, Alberto Hernández-Minguez, Lucia Aballe, and Ronald Imbihl*
Content:
1. Pt electrical characterization
2. Film morphology characterization
3. Conversion of image intensity into work function change
4. SAW amplitude as function of distance
5. Strain calculation
6. Calculation of expected work function change
7. Irreversible work function changes due to SAW treatment
1. Pt electrical characterization
The Pt thin film structures used in this study were characterized electrically by resistivity measurements in a two point probe method, using tips applied to two samples: A control sample with a pristine Pt film (nominal thickness: 20 nm) and the very same sample as used to obtain the work function oscillations in Fig. 2 of the main manuscript, after extracting it from the UHV system. The in situ surface preparation of the second sample included three cycles of Ar$^+$ ion sputtering (10 min, 500 V, 5×10$^{-6}$ mbar, 3 µA, incidence 45°) and oxygen treatment (2×10$^{-5}$ mbar, 600 K, 10 min), which are expected to reduce the Pt thickness. Measured resistance data along 100 µm wide stripes, defined by optical lithography, are shown in Fig. S1, together with the results of linear fits. Similar constant offsets correspond to the contact resistance of the measurement setup, while the slopes of the linear fit indicate the Pt resistivity. From the data of the pristine film, a resistivity of 13.8×10$^{-8}$ Ωm is calculated, in reasonable agreement with the bulk value for Pt 10.8×10$^{-8}$ Ωm at 300 K (K. H. Hellwege, Ed., Landolt-Börnstein Numerical Data and Functional Relationships in Science and Technology, Group III, Vol. 15, Subvol. a, Springer-Verlag, Heidelberg, 1982.), indicating good metallic behavior of the deposited Pt film. The sputter cleaned sample shows an increased sheet resistance, which we ascribe predominantly to a reduction of thickness during the in situ surface cleaning by Ar$^+$ sputtering. Assuming the resistivity of both Pt samples is the same, it translates into a reduced thickness of around 9 nm.
These results rule out the possibility of an incomplete electric screening of the LiNbO$_3$ surface electric potential through the Pt layer to be the origin of the observed work function oscillation in Fig. 2. Note that the ratio between the potential oscillations in LiNbO$_3$ (1.04 V peak to peak) and the measured work function oscillation (0.910 mV peak to peak) correspond numerically to only 8 - 9 screening lengths.
Figure S1: The resistance of a pristine 20 nm thick Pt film and a sputter-cleaned 20 nm thick Pt film as measured using a two-point probe method.
2. Film morphology characterization
The Pt films were prepared by magnetron sputtering from a Pt target at room temperature. In this way polycrystalline Pt film are obtained, with the roughness mainly determined by the surface morphology of the LiNbO₃ substrate single crystal. The low annealing / oxygen treatment temperature of 600 K is not sufficient to cause an atomically flat surface with large corn grains. Due to the small size of the Pt grains in the film, a LEED characterization is not possible in this case. However, we expect a broad distribution of surface orientations with a large portion of the grains representing low index planes of (111) or (100) orientation, as a result of the prolonged high temperature / oxygen treatment. An FTIR characterization of a similar system, a Cu film on LiNbO₃, showed that 3 h annealing at 573 K suffices to generate a surface consisting predominantly of Cu(111) (Inoue, Y., Surf. Sci. Rep., 2007, 62: 305). A dominant (111) orientation is also what is expected from thermodynamics. In order to obtain an estimate of the quality of the Pt film used for the SAW experiments on a macroscopic scale, white light interferometry 3D microscopy (WLI) was used. Fig. S2 shows three WLI images recorded on different areas of the Pt/LiNbO₃ sample. The first image in Fig. S2 gives an overview of the Pt microstructure. Two horizontal and three vertical Pt lines of ~9 nm height are seen on the LiNbO₃ substrate. One notices several bright features, which are particles probably caused by scratching the Pt film during the electrical characterization. The microstructured Pt film represents a cross pattern, which would have allowed several parallel current paths. In order to achieve a well-defined single current path, some intersections were removed carefully with a knife.
The bottom left image shows a magnification of a horizontal Pt stripe and the LiNbO₃ substrate, the bottom right image shows a magnified portion of a Pt stripe. The average $R_q$ roughness, calculated as:
$$ R_q = \sqrt{L^{-2} \sum (h(\vec{r}) - \bar{h})^2}, $$
with $\vec{r}$ being the position vector, $L$ the length of the platinum film and $\bar{h}$ its mean height
$$ \bar{h} = L^{-2} \sum h(\vec{r}), $$
measured over entire field of view of the top image is 96 nm.
The nanoscale roughness was characterized by atomic force microscopy (AFM). A representative $5 \times 5 \mu m^2$ AFM image of the Pt sample used in the SAW experiments is shown in Fig. S3A, together with an AFM image of an unprepared reference sample of the same batch. Both samples have been stored in air for $\sim 8$ months between film preparation (and SAW experiment) and AFM characterization. Fig. S3B shows three-dimensional representations of the data in A. Fig. 3C shows two line profiles taken along the white lines depicted in Fig. 5A. The AFM measurements demonstrate a strong effect of the sample preparation (repeated cycles of $Ar^+$ ion sputtering and oxygen treatment at 600 K) on the film quality. On the Pt/LiNbO$_3$ sample used in the SAW experiments a $R_q$ roughness averaged over the imaged area of 0.64 nm is determined, whereas the corresponding value of the unprepared reference sample is 2.24 nm. The roughness of both films is comparable with values reported in the literature (Dolatshahi-Pirouz, A., et al. Phys. Rev. B, 2008, 77: 115427).
Figure S3: AFM images of the Pt film used in the SAW experiments (left) and of an unprepared reference sample (right). The AFM characterization was performed eight months after film growth and SAW experiment. During this period, the sample was stored in air. In case of the SAW sample, the Pt film was additionally touched with needle contacts for the electrical characterization. A) 2D AFM images of the Pt film used in the SAW experiments (left) and of a pristine reference sample (right). B) 3D representation of the data in A. C) Representative line profiles taken along the white lines shown in A.
3. **Conversion of image intensity into work function change**
In order to reduce noise and to amplify the signal, SAW difference images have been calculated in the following way:
\[
\frac{I_0 - I_\pi}{I_0 + I_\pi}
\]
with \(I_0\) and \(I_\pi\) being the images recorded with the SAW phase 0 and \(\pi\). Due to the subtraction of the images, the intensity variation caused by the phase shifted SAW double in the numerator, whereas the their influence cancels out in the denominator. The denominator is used to normalize the image intensity of the difference image and removes most of the background intensity.
To translate the image intensity into work function change \(\Delta w_f\) between the opposite SAW phases, we use the linear approximation
\[
\frac{I_0 - I_\pi}{I_0 + I_\pi} = \frac{\Delta I}{2I} \approx \frac{1}{2I} \frac{dI}{dV} \Delta w_f,
\]
thus, the whole image has to be multiplied by two times the intensity of the Pt surface at \(-0.5\) V bias voltage \((I(-0.5\) V)), the voltage at which the difference image was recorded, and divided by the slope of the secondary electron emission onset \((d'(-0.5\) V)), these values are shown in Fig. 2c:
\[
\Delta w_f = \frac{I_0 - I_\pi}{I_0 + I_\pi} \cdot \frac{2I(-0.5 \text{ V})}{I'(-0.5 \text{ V})}.
\]
In our case, the average conversion factor for the two phases is 0.50 eV, calculated as:
\[
\frac{1}{2} \cdot \left( \frac{2 \cdot I_0(-0.5 \text{ V})}{I_0'(-0.5 \text{ V})} + \frac{2 \cdot I_\pi(-0.5 \text{ V})}{I_\pi'(-0.5 \text{ V})} \right) \cdot e = \frac{1}{2} \cdot \left( \frac{2 \cdot 60.84}{231.8 \text{ V}^{-1}} + \frac{2 \cdot 67.46}{281.4 \text{ V}^{-1}} \right) \cdot e.
\]
4. SAW amplitude as function of distance
Figure S4: WF changes in the Pt thin film under SAW as determined from stroboscopic XPEEM images (Fig. 2). Shown is the intensity of the SAW induced brightness variations as a function of distance from the Pt/LiNbO$_3$ interface. Left: color coded differential image, showing the Pt surface (top) and the uncovered LiNbO$_3$ substrate (bottom). Right: seven line profiles taken along the horizontal white lines in the image on the left side. The line profiles 1 and 2 were taken in the vicinity of the interface Pt/LiNbO$_3$ / (phase jump). Line profiles 3 to 7 were taken in steps of roughly 9 μm from the interface towards the center of the Pt stripe. The intensity of the line profiles 3 to 7 is multiplied by 500, in order to allow for comparison with the line profiles 1 and 2. The area used for quantification of the SAW induced work function change in Fig. 2 of the main manuscript is the area enclosed by two horizontal black lines in the left frame, far from the stripe edge.
5. **Strain calculation**
**Figure S5**: Calculated strain ($S_{xx}$ and $S_{zz}$) for a SAW of 8 µm wavelength propagating along the $x$ direction of a LiNbO$_3$ substrate partially covered by a 10 nm Pt film. The dashed lines are the strain components at the surface of the Pt-free LiNbO$_3$, while the solid lines are the strain calculated at the Pt film. The strain tensor components are calculated by numerically solving the coupled differential equations of the mechanical and electric displacements and setting a linear power density of 29.5 W/m to obtain the same piezoelectric voltage amplitude at the surface of the Pt-free LiNbO$_3$ as measured experimentally (1.3 V). The Pt film screens the piezoelectric field at the sample surface and modifies the amplitude of the strain fields.
6. Determination of surface potential change over LiNbO₃
**Figure S6:** Determination of the SAW piezoelectric voltage amplitude. Left: XPEEM image recorded at a bias voltage of -0.65 eV at a photon energy of 300 eV. Right: Intensity vs. bias voltage curves showing the onset of secondary photoelectron emission. The intensity is integrated over the two white rectangles shown in the right frame. The amplitude of the SAW is evaluated by numerically fitting the intensity onset with sigmoid functions (not shown). A shift of 2.6 eV between the two curves indicates a SAW piezoelectric voltage amplitude of 1.3 V.
With the SAW piezoelectric voltage amplitude of 1.3 V we calculate the amplitude of the strain components $S_{xx} = +2.03 \times 10^{-4}$ and $S_{yy} = -1.23 \times 10^{-4}$, resulting in a total volume change of:
$$\alpha = (1 + 4.06 \times 10^{-4}) \times (1 - 2.46 \times 10^{-4}) - 1 = 1.6 \times 10^{-4}.$$
Taking the experimentally measured work function difference of 910 µeV we arrive at 28.4 meV/% volume change, taking into account, that the difference image includes an effective factor 2. For comparison, a value of approximately 13.4 meV/% volume change is reported in the literature (Wang, X. F., et al. *Appl. Phys. Lett.*, 2013, 102: 223504).
7. Work function changes due to SAW treatment
Figure S7: Measurement of the onset of secondary photoelectron emission. Changes of the local work function are reflected by an energy shift to lower energy (lower work function) or to higher energy (higher work function). The black squares represent a measurement recorded before the sample was exposed to SAW for 11 h. Data acquired after 11 h of SAW at different positions on the Pt surface (X-ray beam irradiated area and non-irradiated area) exposure are represented by filled circles. For comparison, the corresponding measurement outside the acoustic path on the Pt surface (no SAW exposure) is represented by blue triangles. The application of SAW for 11 h causes a work function decrease of 30 – 75 meV, compared to the situation before SAW excitation. On the Pt surface not exposed to SAW, a work function increase of 200 meV is measured, probably due to adsorption from the residual gas. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
245,
1
],
[
245,
554,
2
],
[
554,
2802,
3
],
[
2802,
5098,
4
],
[
5098,
6168,
5
],
[
6168,
6771,
6
],
[
6771,
8436,
7
],
[
8436,
9463,
8
],
[
9463,
10303,
9
],
[
10303,
11581,
10
],
[
11581,
12526,
11
]
]
} |
dcdd81c07514738865abc6db2d7d98325db6b5e0 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 18,
"total-fallback-pages": 0,
"total-input-tokens": 53431,
"total-output-tokens": 16274,
"fieldofstudy": [
"Biology",
"Medicine"
]
} | Signatures of oral microbiome in HIV-infected individuals with oral Kaposi's sarcoma and cell-associated KSHV DNA
Marion Gruffaz1, Tinghe Zhang2, Vickie Marshall3, Priscila Gonçalves4✉, Ramya Ramaswami4, Nazzarena Labo3, Denise Whitby3, Thomas S. Uldrick4,5, Robert Yarchoan4, Yufei Huang2,6, Shou-Jiang Gao1,7✉
1 Department of Molecular Microbiology and Immunology, Keck School of Medicine, University of Southern California, Los Angeles, United States of America, 2 Department of Electrical and Computer Engineering, University of Texas at San Antonio, San Antonio, Texas, United States of America, 3 Viral Oncology Section, AIDS and Cancer Virus Program, Leidos Biomedical Research Inc., Frederick National Laboratory for Cancer Research, Frederick, Maryland, United States of America, 4 HIV and AIDS Malignancy Branch, Center for Cancer Research, National Cancer Institute, Bethesda, Maryland, United States of America, 5 Fred Hutchinson Cancer Research Center, Seattle, Washington, United States of America, 6 Department of Epidemiology and Biostatistics, The University of Texas Health San Antonio, San Antonio, Texas, United States of America, 7 UPMC Hillman Cancer Center, Department of Microbiology and Molecular Genetics, University of Pittsburgh, Pittsburgh, Pennsylvania, United States of America
✉ Current address: Center for Novel Cancer Therapies, Monter Cancer Center, Lake Success, New York, United States of America
✉ [email protected]
Abstract
Infection by Kaposi's sarcoma-associated herpesvirus (KSHV) is necessary for the development of Kaposi's sarcoma (KS), which most often develops in HIV-infected individuals. KS frequently has oral manifestations and KSHV DNA can be detected in oral cells. Numerous types of cancer are associated with the alteration of microbiome including bacteria and virus. We hypothesize that oral bacterial microbiota affects or is affected by oral KS and the presence of oral cell-associated KSHV DNA. In this study, oral and blood specimens were collected from a cohort of HIV/KSHV-coinfected individuals all previously diagnosed with KS, and were classified as having oral KS with any oral cell-associated KSHV DNA status (O-KS, n = 9), no oral KS but with oral cell-associated KSHV DNA (O-KSHV, n = 10), or with neither oral KS nor oral cell-associated KSHV DNA (No KSHV, n = 10). We sequenced the hypervariable V1-V2 region of the 16S rRNA gene present in oral cell-associated DNA by next generation sequencing. The diversity, richness, relative abundance of operational taxonomic units (OTUs) and taxonomic composition of oral microbiota were analyzed and compared across the 3 studied groups. We found impoverishment of oral microbial diversity and enrichment of specific microbiota in O-KS individuals compared to O-KSHV or No KSHV individuals. These results suggest that HIV/KSHV coinfection and oral microbiota might impact one another and influence the development of oral KS.
Kaposi’s sarcoma (KS) is the most common cancer occurring in HIV-infected individuals worldwide, and often involves the mouth. While infection by Kaposi’s sarcoma-associated herpesvirus (KSHV) is necessary for the development of KS, other cofactors remain unclear. In this study, we evaluated the impact of oral bacterial microbiota on the development of oral KS and the presence of oral cell-associated KSHV DNA by studying a cohort of HIV/KSHV-coinfected individuals all previously diagnosed with KS, classified as having oral KS with any oral cell-associated KSHV DNA status (O-KS), no oral KS but with oral cell-associated KSHV DNA (O-KSHV), or with neither oral KS nor oral cell-associated KSHV DNA (No KSHV). We observed impoverishment of oral microbial diversity and enrichment of specific types of microbes in O-KS individuals compared to O-KSHV or No KSHV individuals. Hence, HIV/KSHV coinfection and oral microbiota might impact one another and influence the development of oral KS.
Introduction
Infection by Kaposi’s sarcoma-associated virus (KSHV), also called human herpesvirus-8 (HHV-8), is associated with several human malignancies or hyperinflammatory conditions including Kaposi’s sarcoma (KS), primary effusion lymphoma (PEL), multicentric Castleman’s disease (MCD) and KSHV inflammatory cytokine syndrome (KICS)[1, 2]. KSHV is a gamma-herpesvirus having latent and lytic replication phases[3]. KSHV DNA can often be detected in oral cells of individuals with asymptomatic KSHV infection[4]. In KSHV-associated malignancies, KSHV latent infection and latent genes are essential for the proliferation, survival and immune evasion of tumor cells[1, 5]. However, KSHV lytic replication, detected in a small proportion of cells within tumors, also contributes to disease pathogenesis[1, 5].
KS is the most common KSHV-associated malignancy worldwide. It is characterized by the proliferation of vascular spindle tumors cells, and extensive inflammatory infiltration and angiogenesis[1]. AIDS-associated KS (AIDS-KS) may affect the skin, the oral cavity, lymph nodes and internal organs including gut, stomach, liver and lung. Oral KS is the first manifestation in 20% of AIDS-KS individuals[6]. Before the introduction of effective antiretroviral therapy, up to 70% of AIDS-KS individuals in the US eventually developed oral, visceral or cutaneous KS[6]. Antiretroviral therapy effectively decreases KS incidence[7] and the presence of KSHV in oral cells[8]. However, even in the era of combination antiretroviral therapy (cART), KS remains one of the most common AIDS-related cancers in the United States and throughout sub-Saharan Africa[9]. Upon initiation of cART, KS can also progress, possibly triggered by, or exacerbated by a KS immune reconstitution inflammatory syndrome (IRIS) [10]. AIDS-KS continues to have a high mortality rate in sub-Saharan Africa[11] and is still associated with morbidity and mortality in the US among HIV-infected patients[12].
The development of next generation sequencing (NGS) has enabled the identification and quantification of the microbiome including bacteria, viruses and fungi in healthy individuals, or in a context of disease[13]. Indeed, numerous metagenomics studies of the microbiome have highlighted microbial pattern modifications in various types of cancer and viral infections. Particularly, over the course of HIV infection, microbial diversity is altered as a result of host-microbiota interactions[14]. Moreover, in cancer, specific microbial DNA signatures have been identified[15]. For example, two bacteria commonly associated with gastric and colorectal cancers in humans are Helicobacter pylori and Fusobacterium nucleatum, respectively[16, 17]. Other studies also showed a positive association between Propionibacterium...
acnes and the development of prostate cancer\[18\]. Some enterotoxigenic Escherichia coli (E. coli) strains are associated with tumorigenesis in mouse models of colorectal cancers\[19\]. We have recently shown that E. coli C25 strain could promote KSHV-induced tumorigenesis in a KS-like mouse model\[20\].
Numerous pathogenic mechanisms, including the production of genotoxic inducers, the activation of TLRs and pro-inflammatory pathways, the inhibition of the immune response, and the increase of the cellular turnover have been proposed to explain the oncogenic properties of some types of bacteria\[15\]. Bacterial byproducts such as short-chain fatty acids (SCFA) that are highly abundant in individuals suffering from periodontal disease can induce KSHV reactivation\[21, 22\]. Interestingly, HIV-infected individuals display a higher rate of periodontal disease\[23\], which has been proposed to promote oral KS development by inducing pro-inflammatory cytokines or releasing SCFA. We previously showed that E. coli infection or bacterial ligands such as LPS can promote KSHV-induced tumorigenesis in mice through the activation of TLR4 and pro-inflammatory pathways\[20\].
In this study, we examined the oral bacterial microbiota in a cohort consisting of HIV/KSHV-coinfected individuals all previously diagnosed with KS, with or without oral KS whose status of oral cell-associated KSHV DNA was assessed. We showed a diminution of the oral microbial diversity and enrichment of specific bacteria in HIV/KSHV-coinfected individuals with oral KS, which was independent of the presence of oral cell-associated KSHV DNA. Hence, the composition of the oral microbiota may contribute to the development of oral KS.
Results
Cohort characteristics and clinical status
We recruited 29 HIV/KSHV-coinfected individuals with a history of pathology-confirmed KS by physician physical examination, including 9 individuals who had oral KS involvement with any oral cell-associated KSHV DNA status (O-KS), 10 individuals who had detectable intermittent oral cell-associated KSHV DNA without oral KS (O-KSHV), and 10 individuals who had neither oral cell-associated KSHV DNA nor oral KS (No KSHV) (Table 1). While all individuals had KS when admitted into the cohort, 5 of these subjects no longer had detectable KS at the time of sampling but had either KSHV-MCD or PEL (Table 1). Intermittent oral cell-associated KSHV DNA was defined based on the detection of KSHV DNA in oral cells in longitudinal follow-ups in at least 3 visits. Oral microbiota, oral cell-associated KSHV DNA, KSHV blood DNA detected in peripheral blood mononuclear cells (PBMC), and HIV load were examined at the time of sampling while CD4+ T cell count and CD8+ T cell count were obtained from medical records. Of note, while all individuals were infected with HIV when they were admitted into the cohort, only nine had detectable (>50 copies/mL) HIV RNA at the time of sampling, two of whom had >400 copies/mL. Two individuals in the O-KSHV group were negative for oral cell-associated KSHV DNA at the time of sampling; however, they had a significant history of detectable oral cell-associated KSHV DNA with one individual positive for 17 of 20 time points (85%) tested over a 43 months period and the second individual positive for 8 of 15 time points (53%) tested over a 55 months period. All individuals were men and had a median age of 45 years.
Few individuals had detectable KSHV in PBMC and no significant difference between the studied groups was observed (Fig 1A). Likewise, no significant difference in HIV load (Fig 1B) was noted between groups. There were 5 of 9 and 8 of 10 individuals who had detectable oral cell-associated KSHV DNA in the O-KS and O-KSHV groups, respectively, but no significant difference was observed between the O-KSHV and O-KS groups at the time of collection (Fig 1C).
| Subject | KS status at the time of sampling | HIV antiretroviral therapy (ART) | HIV ART duration (month) | Time from HIV diagnosis (month) | CD4 (cells/μl) | CD8 (cells/μl) | HIV in blood (copies/ml) | PBMC KSHV DNA (copies/10^6 cells) | Oral cell-associated KSHV DNA (copies/10^6 cells) |
|---------|---------------------------------|---------------------------------|--------------------------|------------------------------|---------------|---------------|-------------------------|---------------------------------|---------------------------------|
| **GROUP 1: O-KS Oral KS** | | | | | | | | | |
| 1 | O-KS | Darunavir, Tenofovir, Emtricitabin, Ritonavir | 3.2 | 3.9 | 46 | 475 | 1150 | QP | <3 |
| 2 | O-KS | Tenofovir, Emtricitabin, Efavirenz | 8.4 | 8.4 | 47 | 1099 | <50 | QP | <3 |
| 3 | O-KS | Tenofovir, Emtricitabin, Ritonavir, Atazanavir | 15.9 | 24.1 | 106 | 423 | 74 | ND | 58650 |
| 4 | O-KS | Abacavir, Etravirine, Raltegravir | 6.4 | 7.3 | 49 | 113 | <20 | QP | 25 |
| 5 | O-KS |Efavirenz, Cobicistat, Emtricitabin, Tenofovir | 110.6 | 110.6 | 753 | 424 | <20 | QP | 1365000 |
| 6 | O-KS | Abacavir, Dolutegravin, Lamivudine | 260.0 | 261.0 | 39 | 402 | <20 | QP | <3 |
| 7 | O-KS | Abacavir, Dolutegravin, Lamivudine | 6.8 | 9.4 | 151 | 1890 | <20 | <3 | <3 |
| 8 | O-KS | Emtricitabin, Tenofovir, Rilpivirine | 0 | 127.7 | 6 | 389 | 156292 | <20 | <3 |
| 9 | O-KS | Abacavir, Lamivudine, Dolutegravin | 2.4 | 30.8 | 16 | 500 | <20 | QP | 3400 |
| **GROUP 2: O-KSHV No oral KS, intermittent detection of oral cell-associated KSHV DNA** | | | | | | | | | |
| 10 | No KS |Efavirenz, Emtricitabin, Tenofovir | 219.2 | 295.2 | 571 | 1288 | <20 | 24500 | 1640 |
| 11 | KS | Atazanavir, Lamivudine, Zidovudine | 45.8 | 91.9 | 637 | 845 | <50 | <3 | 157000 |
| 12 | KS | Lamivudine, Zidovudine, Lopinavir, Ritonavir | 33.3 | 40.8 | 119 | 937 | <20 | QP | 185 |
| 13 | KS | Efavirenz, Tenofovir, Emtricitabin | 6.0 | 131.8 | 285 | 463 | <50 | <3 | 38000 |
| 14 | KS | Tenofovir, Emtricitabin, Dolutegravin | 241.2 | 241.2 | 180 | 382 | <20 | <3 | 3500 |
| 15 | KS | Abacavir, Dolutegravin, Lamivudine | 187.6 | 189.6 | 449 | 555 | <20 | <3 | 36000 |
| 16 | KS | Ritonavir, Tenofovir, Emtricitabin, Darunavir | 5.2 | 26.2 | 307 | 1112 | 75 | 2000 | 4400 |
| 17 | No KS | Tenofovir, Emtricitabin, Ritonavir, Reyataz | 187.5 | 190.5 | 584 | 1148 | <50 | 570 | <3 |
| 18 | No KS | Raltegravir, Darunavir, Tenofovir, Emtricitabin | 279.1 | 315.2 | 594 | 1700 | <20 | 42000 | 1300 |
| 19 | KS | Efavirenz, Emtricitabin, Tenofovir | 17.7 | 53.8 | 287 | 718 | <50 | 270500 | 2200 |
| **GROUP 3: No KSHV No oral KS, no detection of oral cell-associated KSHV DNA** | | | | | | | | | |
| 20 | KS |Efavirenz, Emtricitabin, Tenofovir | 201.5 | 225.6 | 215 | 479 | <20 | <3 | <3 |
| 21 | No KS |Efavirenz, Emtricitabin, Tenofovir | 49.0 | 55.0 | 386 | 625 | <20 | QP | <3 |
| 22 | KS | Tenofovir, Emtricitabin, Raltegravir | 39.4 | 147.6 | 396 | 667 | <50 | <3 | <3 |
| 23 | KS | Ritonavir, Lamivudine, Abacavir, Darunavir | 68.3 | 69.5 | 465 | 712 | <20 | <3 | <3 |
| 24 | KS | Tenofovir, Emtricitabin, Raltegravir | 3.6 | 67.4 | 185 | 842 | 27 | QP | <3 |
| 25 | No KS | Tenofovir, Emtricitabin, Efavirenz | 29.5 | 286.5 | 378 | 1493 | 336 | <3 | <3 |
| 26 | KS | Abacavir, Lamivudine, Zidovudine, Atazanavir | 58.7 | 185.8 | 482 | 385 | <50 | <3 | <3 |
| 27 | KS | Norvir, Tenofovir, Emtricitabin, Ritonavir | 81.7 | 81.7 | 437 | 1308 | 58 | 240 | <3 |
| 28 | KS | Raltegravir, Darunavir, Etravirine, Ritonavir | 189.3 | 249.4 | 97 | 518 | 109 | 500 | <3 |
| 29 | KS | Lamivudine, Tenofovir, Nelfinavir | 183.0 | 183.0 | 445 | 950 | 447 | 60 | <3 |
ND: Not done.
QP: Qualitative positive, i.e. detectable KSHV DNA <3 copies/10^6 cells (assay cut-off).
*: No detectable oral cell-associated KSHV DNA over 3 or more time points.
https://doi.org/10.1371/journal.ppat.1008114.t001
Fig 1. Viral and immunological status in three studied groups of HIV/KSHV-coinfected individuals. (A–C) Quantification of PBMC KSHV DNA (A), HIV load (B), and oral cell-associated KSHV DNA (C). (D–E) Quantification of levels of CD4+ T cell count (D) and CD8+ T cell count (E). (F) Correlation of CD4+ T cell count with oral cell-associated KSHV DNA. (G) Correlation of CD4+ T cell count with PBMC KSHV DNA. (H) Correlation of HIV load and PBMC KSHV DNA. (I) Correlation of CD4+ T cell count with HIV load. P-value ≤0.05 (*) was considered as significant. NS indicates not significant (P-value ≥0.05).
https://doi.org/10.1371/journal.ppat.1008114.g001
The CD4+ T cells was significantly lower in the O-KS group than in the other two groups (Fig 1D), and 8 of 9 individuals in the O-KS group had advanced immunodeficiency with a CD4+ T cell count <200 cells/μL. However, no difference of CD8+ T cell count was observed between the O-KS and either of the other two groups (Fig 1E). There was a significant positive correlation between oral cell-associated KSHV DNA and CD4+ T cell count (Fig 1F) as previously described[8, 24]. However, there was no association between CD4+ T cell count and PBMC KSHV DNA (Fig 1G). There was a weak positive correlation between HIV load and PBMC KSHV DNA (Fig 1H) as previously described[25]. No correlation between CD4+ T cell count and HIV load was observed (Fig 1I).
**Microbiota richness and diversity across samples**
We sequenced the hypervariable V1-V2 region of 16S rRNA gene from oral specimens of the 29 individuals using an Illumina MiSeq system. A total of 15,261,199 raw sequences were generated. After quality control and filtering, 13,098,094 high-quality sequences with an average length of 263 bp were recovered for further analysis, with an average of 451,658 reads per
After alignment with QIIME database, unique representative sequences were classified into 1,886 operational taxonomic units (OTUs), from which 16 phyla, 28 classes, 41 orders, 79 families, 125 genera and 148 species were identified. Shannon diversity index was used to evaluate the sequencing depth. All the rarefaction curves reached plateau indicating that there was sufficient sequencing coverage depth (Fig 2A). Using Venn diagram, we observed that all 3 groups shared 1,242 OTUs, whereas 62, 129 and 97 OTUs were specific to the O-KS, O-KSHV and No KSHV groups, respectively (Fig 2B).
We calculated the species richness (using the Ace and Chao1 nonparametric methods for estimating the number of species in a community) (Fig 2C) and the α-diversity of the observed species (using the Shannon index measuring how evenly OTUs are distributed in a sample) among all groups (Fig 2D). We observed a strong diminution of the α-diversity and richness in individuals in the O-KS group compared to the other two groups. Using Wilcoxon signed-rank test, we tested the statistical difference between the richness and α-diversity of each pair of groups (Fig 2E). We observed a significant dissimilarity between the O-KS and O-KSHV groups (P-value 2.2e-16), and between the O-KS and No KSHV groups (P-value 2.35e-13), whereas the O-KSHV and No KSHV groups were similar (P-value 0.117). Hence, individuals of the O-KS group tend to cluster together and were more distant from individuals of the O-KSHV or No KSHV groups, suggesting an association between the oral microbiota and the presence of oral KS in HIV/KSHV-positive individuals.
Relative abundances of OTUs and taxonomic compositions of bacterial populations at phylum and species levels across the 3 different groups of HIV/KSHV-coinfected individuals
The relative abundances of OTUs, as well as the taxonomic compositions of bacterial populations in individuals of the 3 studied groups were analyzed at different taxonomic levels. The top 5 most abundant identifiable phyla in the oral microbiota were Firmicutes, Bacteroidetes, Actinobacteria, Proteobacteria and Fusobacteria (Fig 3A), which were consistent with previous reports[26]. At the species level, Streptococcus, Prevotella, Lactobacillus, Dispar and Selenomonas were the top 5 most abundant species for all 3 groups (Fig 3B). To investigate the taxonomic composition of the bacterial populations in all 29 subjects, heatmaps were constructed by clustering each individual OTU compositions (S1A and S1B Fig), as well as each mean of relative OTU abundances for the 3 studied groups (Fig 3C and 3D). The heatmaps constructed at phylum and species levels demonstrated that the O-KSHV and No KSHV groups clustered together, whereas the O-KS group was phylogenetically more distant from other groups (Fig 3C and 3D), indicating that distinct oral microbiome might influence the development of oral KS in HIV/KSHV-coinfected individuals or that the oral microbiome is perturbed by the development of oral KS.
Alterations of the oral microbiome in HIV/KSHV-coinfected individuals with oral cell-associated KSHV DNA or oral KS
We further examined the differences of bacterial distributions across all 3 studied groups at all phylogenetic levels. We observed a significant diminution of Pasteurellales and Burkholderiales at order level in the O-KS group compared to the O-KSHV and No KSHV groups (Fig 4A). At family level, Bacillaceae was enriched in the O-KS group, whereas the abundance of Burkholderiaceae decreased (Fig 4B). At genus level, the abundances of Aggregibacter and Lautropia were decreased in the O-KS group, whereas those of Corynebacterium and Shuttleworthia were increased (Fig 4C). At the species level, the abundances of Dialister and Satelles were increased.
Fig 3. Relative abundances of OTUs and taxonomic compositions of bacterial communities at phylum and species levels in three studied groups of HIV/KSHV-coinfected individuals. (A) Mean of relative OTU abundances at phylum level for the three studied groups. (B) Mean of relative OTU abundances of the 30 most abundant species for the three studied groups. (C) Mean of the relative OTU abundances at phylum level clustered by the three studied groups. (D) Mean of the relative OTU abundances of the 30 most abundant species clustered by the three studied groups.
https://doi.org/10.1371/journal.ppat.1008114.g003
Fig 4. Alterations of oral microbiota in HIV/KSHV-coinfected individuals with oral cell-associated KSHV DNA or oral KS. (A-D) Distinct signatures of O-KS group shown by relative abundances of OTUs at order level (A), family level (B), genus level (C) and species level (D). (E-G) Distinct signatures of No KSHV group shown by relative abundances of OTUs at order level (E), family level (F) and species level (G). Statistical analysis was performed using Student’s t-test with the GraphPad Prism software. Only significant results are represented.
https://doi.org/10.1371/journal.ppat.1008114.g004
whereas those of Lautropia and Porphyromonas were decreased in the O-KS group (Fig 4D).
No significant differences were observed among all 3 groups at phylum and class levels. The increase in OTUs in the O-KS group clustered in the Firmicutes and Actinobacteria phylum, whereas the decrease in OTUs in the O-KS group centered in the Bacteroides and Proteobacteria phylum (Fig 5A).
When individual bacterial orders, families, or species groups were examined, there were some significant differences between individuals with oral cell-associated KSHV DNA (O-KSHV) as compared to those without oral cell-associated KSHV DNA (No KSHV). Some of these differences were independent of oral KS status as they were also observed between the O-KS group and the No KSHV group, most likely due to the fact that 5 of 9 individuals in the O-KS group had detectable oral cell-associated KSHV DNA. For example, the abundance of Bacillales order increased in both O-KSHV and O-KS groups (Fig 4E), whereas those of Gemellaceae family and the Gemella species decreased (Fig 4F and 4G and Fig 5B).
Altogether, these results indicated that alterations of the oral microbiota might influence the detection of oral cell-associated KSHV DNA and development of oral KS in HIV/KSHV-coinfected individuals, or that the presence of oral cell-associated KSHV DNA and development of oral KS may influence the composition of the oral microbiome.
Discussion
Altered oral microbiota has been observed in several diseases including diabetes, bacteremia, endocarditis, cancer and autoimmune disease, and in some cases can influence disease progression[27] or tumor response to immunotherapy[28]. Furthermore, we previously demonstrated that TLR4 stimulation with LPS from either E. coli strain K12 or C25 promoted KSHV-induced cellular transformation and tumorigenesis in a KS-like animal model[20]. Therefore, the purpose of this study was to investigate whether oral KS was associated with an alteration of the bacterial microbiota in the oral cavity. Our results revealed for the first time significant changes in diversity in oral microbiota in HIV/KSHV-coinfected individuals with oral KS compared to those without oral KS. These observations support the close interactions of microbiome, viral infections and cancer development[29, 30], and to our knowledge, it is the first study in the context of a KSHV-associated cancer in HIV-infected individuals.
We observed a strong diminution of microbial α-diversity and richness in individuals developing oral KS compared to the other two groups without oral KS. Imbalance in microbial flora composition is correlated with impaired immune cell activity and the decrease of oral microbial diversity could affect immune responses[31]. We indeed have observed lower CD4 + T cell counts in the O-KS group than the other two groups (Fig 1D). The impoverishment in oral microbiota may have implications for the immune reconstitution during cART in HIV-infected patients developing oral KS. In melanoma patients undergoing anti-PD-1 immunotherapy, significant differences were observed in the diversity and composition of the gut microbiome of responders versus non-responders. Higher gut microbiome diversity was associated with improved response to anti-PD-1 immunotherapy in patients with metastatic melanoma[28]. Nevertheless, advanced immunodeficiency can also affect microbial diversity. Further investigations are required to elucidate the role of decreased microbial diversity and alterations of specific microbiota in the development oral KS.
In healthy individuals, the oral microbiota is usually composed of the phyla Firmicutes, Proteobacteria, Bacteroidetes, Actinobacteria, and Fusobacteria, with a predominance in the genus Streptococcus followed by Prevotella, Veillonella, Neisseria, and Haemophilus[26]. As observed in the oral cavity of healthy individuals[26], the top 5 phyla in all 3 groups of HIV/KSHV-coinfected individuals consisted of the same 5 phyla (Fig 3A). At species level, because of the high variability within each individual, no strict consensus has been identified so far regarding the bacterial relative abundance in the oral cavity in healthy individuals[26, 30]. However, we identified Streptococcus, Prevotella, Lactobacillus, Dispar sp. and Selenomonas as the top 5 species in all 3 groups of HIV/KSHV-coinfected individuals (Fig 3B). Although there was no difference in the distribution of the top 5 phyla or species across all 3 groups (Fig 3B), the cluster analysis showed that the phylum and species in the O-KS group were distinct from the O-KSHV and No KSHV groups (Fig 3C and 3D), highlighting the modification of the microbial pattern in individuals developing cancer as previously reported[29].
Oral squamous cell cancer (OSCC) is the oral cancer with the highest incidence, and ranked the 15th place in frequency among all types of cancer in 2012[32]. OSCC has been associated with increases of oral Capnocytophaga gingivalis and Prevotella melaninigenica belonging to the Bacteroidetes phylum, and Streptococcus mitis belonging to the Firmicutes phylum, as well as decreases of Citrobacter and Neisseraceae belonging to the Proteobacteria phylum compared to the cancer-free control groups[33–35]. In our study, we observed a diminution of the Burkholderiales and Pasteurellales order belonging to the Proteobacteria phylum, and a decrease in species of Porphyromonas belonging to the Bacteroidetes phylum in individuals who developed oral KS compared to control groups without oral KS. In parallel, there were increases in OTUs belonging to the Firmicutes phylum such as Satelles species, Dialister order and Bacillaceae
family in the O-KS group (Fig 4 and Fig 5). Hence, the bacterial signature of oral KS is different from that of OSCC despite both types of cancer having an increase in OTUs of Firmicutes phylum and a decrease in OTUs of Proteobacteria phylum. However, the diminution of OTUs belonging to the Bacteroidetes phylum seems to be specific to individuals who developed oral KS and has not been described in other types of oral cancer. Moreover, as observed in the O-KS group, an increase of Firmicutes and decreases of Bacteroidetes and Proteobacteria phyla-totypes have previously been reported in other types of malignancies such as colorectal cancer (CRC)[36].
Various factors such as diet, oral hygiene, tobacco and alcohol consumption, stress, hormonal imbalance, diabetes, and gingival inflammation can perturb the oral bacterial community[37]. Studies have demonstrated that viral infections such as HIV, CMV, EBV and HSV-1 can also impact the composition of oral microbiome[38–40]. Other reports highlighted the negative effect of HIV load on microbiome. For examples, there were increases of Porphyromonas sp. and Corynebacterium order in HIV-infected individuals compared to HIV-negative individuals[41, 42].
We observed significant differences of some microbiota between individuals with neither oral KS nor detectable oral cell-associated KSHV DNA (No KSHV) and those with oral KS (O-KS) or detectable oral cell-associated KSHV DNA (O-KSHV) (Fig 4 and Fig 5). These results suggested that these differences were independent of KS status despite 5 of 9 individuals in the O-KS group had detectable oral cell-associated KSHV DNA (Table 1). Particularly, there was an increase in Bacillales OTU of the Firmicutes phylum in individuals with oral cell-associated KSHV DNA in the O-KSHV and O-KS groups. Firmicutes lineage has been associated with inflammation and cancer development in other reports. Indeed, Clostridia from Firmicutes lineage promoted carcinogenesis by inducing pro-inflammatory Th1 and Th17 immune responses in mice[43]. Moreover, anti-inflammatory responses can be induced by the generation of regulatory T cells through the production of short-chain fatty acid (SCFA) by bacteria belonging to the clostridia cluster[44–46]. Also, and perhaps even more important, SCFA, such as sodium butyrate and valproic acid, that act as histone deacetylase (HDAC) inhibitors, can reactivate KSHV[21, 22], and therefore increased butyrate production, might promote KS tumorigenesis through lytic activation of KSHV[47]. Interestingly, HIV-infected individuals with severe periodontal disease display a higher level of SCFA in the saliva compared to healthy individuals[48]. Firmicutes abundance has also been linked to TNF-α serum concentration in young obese patients[49]. However, since our study was cross-sectional, it is also possible that changes in the oral microbiome were a consequence of detectable oral cell-associated KSHV DNA or oral KS development, which is closely linked to the immune status of the subjects.
This study of the oral bacterial signature in AIDS-KS highlights for the first time the link between the oral microbiota and oral KS. A broader evaluation of the microbiome in the context of the development of visceral and systemic KS as well as resolution with immune reconstitution following antiretroviral therapy is warranted and may lead to novel biomarkers or probiotic approaches to treating KS.
Materials and methods
Study design and individual selection
We conducted a cross-sectional analysis of the oral microbiome in twenty-nine individuals with pathology-confirmed KS who were serologically positive for KSHV and HIV, seen from December 2004 to February 2015 as part of clinical research protocols in the HIV and AIDS Malignancy Branch of the National Cancer Institute (NCI), which allowed the evaluation of
oral cell-associated KSHV DNA. Individuals were selected from cohorts of AIDS-KS patients who were participating in clinical studies. Participants with at least 3 oral cell-associated KSHV DNA measurements were included. KS diagnosis was initially made by the attending physicians and confirmed by pathological examination by NCI pathologists with a biopsy by immunohistochemistry and sometimes by PCR for the presence of KSHV DNA. The KS stage of all patients was determined using the AIDS Clinical Trials Group Network (ACTG) criteria [50]. Presence or absence of macroscopically visible KS lesions in the oral mucosa was obtained from records of research physicians who performed physical examinations and was supported by photography. Individuals were classified into three groups based on oral cell-associated KSHV DNA patterns and clinical documentation: oral KS with any oral cell-associated KSHV DNA status (O-KS, n = 9); no oral KS but with detectable oral cell-associated KSHV DNA (O-KSHV, n = 10); and with neither oral KS nor detectable oral cell-associated KSHV DNA over 3 or more time points (No KSHV, n = 10). Intermittent oral cell-associated KSHV DNA was defined based on the detection of KSHV DNA in oral cells in longitudinal follow-ups in at least 3 visits. Oral microbiota, oral cell-associated KSHV DNA, KSHV blood DNA detected in PBMC, and HIV load were examined at the time of sampling while CD4+ T cell count, CD8+ T cell count, the type of HIV therapy and duration, and years since HIV diagnosis were obtained from medical records. All participants provided written informed consent.
Sampling and DNA extraction
Approximately 5 ml of Scope mouthwash was used to collect oral fluids and cellular materials. Samples were centrifuged at 8,000 g for 5 min to pellet cellular materials. The supernatant was transferred to another tube without disturbing the pellet. Pellets were stored at -80 °C prior to extraction. PBMC materials were matched with oral cells (same draw date) in every incidence except one intermittent shedder who had PBMC draw next day and for two never shedders who had PBMC materials within a month of the collection of oral cells. PBMC were isolated from blood collected in acid citrate dextrose (ACD) tubes by Ficoll (GE Healthcare) centrifugation with Leucosep tubes (Greiner bio-one). Red blood cells were lysed with ammonium-chloride-potassium (ACK) buffer following manufacturer’s protocol (Thermo Scientific). The purified PBMC were counted using a Nexcellom Cellometer Vision instrument. Pellets of approximately 1–2 million PBMC were obtained for each individual and stored at -80 °C prior to DNA extraction.
Genomic DNA was extracted from oral cells and PBMC pellets using QIAamp Blood Mini Kits according to the manufacturer’s instructions (Qiagen). The quality and quantity of the DNA extracted were measured using a UV spectrophotometer at 260 and 280 nm (Nanodrop 2000). All DNA samples were stored at -80 °C.
Measurements of KSHV DNA and antibodies
KSHV DNA in oral cell pellet and PBMC collected at the time of oral sampling were measured as previously described[51]. Briefly, KSHV DNA was measured in a real-time PCR assay based on the ORF-K6 gene. Triplicate assays were performed, and the averages of the triplicate values were used to determine the viral copies. The cell-associated KSHV DNA values were converted to copies per million cells using a cell quantitation assay based on the ERV-3 gene[52] with an assay sensitivity of 3 copies/10⁶ cells. The assays are Clinical Laboratory Improvement Amendments (CLIA) certified and conducted using stringent procedures to prevent contamination.
KSHV serological status was determined using recombinant protein ELISA for ORF73 and K8.1. Participants were considered KSHV-seropositive if they had antibodies to either antigen.
Next-generation sequencing
The first PCR amplification of the bacterial 16S rRNA gene hypervariable V1-V2 region was performed with DNA isolated from oral cell pellet using the forward primer 5'-TCGTCGGCAGCTCTACGATGTATAGACAG-3' and the reverse primer 5'-GTCTCGTGGCCTCATATAGACAG-3', both of which contain overhang adapters. The PCR reaction was carried out by 2 min initial denaturation at 95 °C, 25 cycles of 30 sec denaturation at 95 °C, 30 sec elongation at 72 °C, and a 5 min one final extension at 72 °C. Illumina sequencing adapters and indexes were attached using Nextera XT Index kit (Cat. FC-131-1001, Illumina). The amplicon library was built and sequenced with the Illumina MiSeq V3 600 Cycle Kit according to the manufacturer's instructions (Illumina) using the universal 16S primers 27F (5'-AGAGTTTGATCMTGGCTCAG-3') and 355R (5'-GCTGCCTCCTCCTTAG-3')[53].
Bioinformatics and statistical analysis
Raw pyrosequencing results from the hypervariable region V1·V2 of the 16S rRNA gene were filtered and unique reads were assigned to the corresponding samples after mapping with barcode and primer sequences. FastQC was used to evaluate the sample reads for sequencing quality[54]. Sequences were analyzed using quantitative insights into microbial ecology (QIIME) software[55]. Sequences were assigned to operational taxonomic units (OTUs) at 97% similarity at different taxonomic levels (from phylum to species) to the Human Oral Microbiome Database with a 97% cutoff value. Bacterial diversity was determined by performing a sampling-based OTU analysis and was displayed as rarefaction curves (Shannon index curves). Bacterial richness and diversity across samples were analyzed using the following α indexes: Shannon, Chao, Ace and Observed species. Comparison of OTUs abundance across groups was performed using Student's t-test with the GraphPad Prism software[56]. For all statistic tests, P-value \( \leq 0.05 \) (\(*\)), \( \leq 0.005 \) (\(**\)) or \( \leq 0.0005 \) (\(***)\) were considered significant.
Ethics statement
All patients were enrolled on National Institutes of Health Clinical Center Protocol 01-C-0038 (registered in clinicaltrials.gov as NCT00006518), which was approved by the National Cancer Institute Institutional Review Board, and the samples were obtained and studied as part of this protocol. All individuals gave written informed consent to clinical examination, sample acquisition, and testing on clinical samples. However, all the data were analyzed anonymously.
Supporting information
S1 Fig. Taxonomic compositions of bacterial communities at phylum and species levels for each individual HIV/KSHV-coinfected subject. (A) Relative OTU abundances at phylum level clustered by each individual subject. (B) Relative OTU abundances of the 30 most abundant species clustered by each individual subject. (TIF)
Acknowledgments
We thank members of Dr. Gao’s laboratory for technical assistances and helpful discussions. We thank Kathleen Wyvill, Karen Aleman, Anna Widell and Matthew Lindsley at the NIH Center for Cancer Research for collection of oral samples and care of the patients.
Author Contributions
Conceptualization: Shou-Jiang Gao.
Data curation: Marion Gruffaz, Vickie Marshall, Priscila Gonçalves, Ramya Ramaswami, Nazarena Labo.
Formal analysis: Marion Gruffaz, Tinghe Zhang.
Funding acquisition: Denise Whitby, Robert Yarchoan, Yufei Huang, Shou-Jiang Gao.
Investigation: Marion Gruffaz, Tinghe Zhang, Vickie Marshall, Priscila Gonçalves, Ramya Ramaswami, Nazarena Labo, Denise Whitby, Thomas S. Uldrick, Yufei Huang.
Methodology: Marion Gruffaz, Tinghe Zhang, Vickie Marshall, Priscila Gonçalves, Ramya Ramaswami, Nazarena Labo, Yufei Huang.
Project administration: Denise Whitby, Robert Yarchoan, Yufei Huang, Shou-Jiang Gao.
Resources: Denise Whitby, Robert Yarchoan, Yufei Huang, Shou-Jiang Gao.
Software: Tinghe Zhang, Yufei Huang.
Supervision: Denise Whitby, Thomas S. Uldrick, Robert Yarchoan, Yufei Huang, Shou-Jiang Gao.
Validation: Marion Gruffaz, Tinghe Zhang, Denise Whitby, Robert Yarchoan, Yufei Huang, Shou-Jiang Gao.
Visualization: Marion Gruffaz, Tinghe Zhang.
Writing – original draft: Marion Gruffaz, Shou-Jiang Gao.
Writing – review & editing: Marion Gruffaz, Tinghe Zhang, Vickie Marshall, Priscila Gonçalves, Ramya Ramaswami, Nazarena Labo, Denise Whitby, Thomas S. Uldrick, Robert Yarchoan, Yufei Huang, Shou-Jiang Gao.
References
1. Greene W, Kuhne K, Ye F, Chen J, Zhou F, Lei X, et al. Molecular biology of KSHV in relation to AIDS-associated oncogenesis. Cancer Treat Res. 2007; 133:69–127. Epub 2007/08/04. https://doi.org/10.1007/978-0-387-46816-7_3 PMID: 17672038; PubMed Central PMCID: PMC2798888.
2. Uldrick TS, Wang V, O'Mahony D, Aleman K, Wyvill KM, Marshall V, et al. An interleukin-6-related systemic inflammatory syndrome in patients co-infected with Kaposi sarcoma-associated herpesvirus and HIV but without Multicentric Castleman disease. Clin Infect Dis. 2010; 51(3):350–8. Epub 2010/06/30. https://doi.org/10.1086/654798 PMID: 20583924; PubMed Central PMCID: PMC2946207.
3. Ye F, Lei X, Gao SJ. Mechanisms of Kaposi’s sarcoma-associated herpesvirus latency and reactivation. Adv Virol. 2011;2011. Epub 2011/06/01. https://doi.org/10.1155/2011/193860 PMID: 21625290; PubMed Central PMCID: PMC3103228.
4. Newton R, Labo N, Wakeham K, Marshall V, Roshan R, Nalwoga A, et al. Determinants of gammaherpesvirus shedding in saliva among Ugandan children and their mothers. J Infect Dis. 2018; 218(6):892–900. https://doi.org/10.1093/infdis/jiy262 PMID: 29762709; PubMed Central PMCID: PMC6093317.
5. Ganem D. KSHV and the pathogenesis of Kaposi sarcoma: listening to human biology and medicine. J Clin Invest. 2010; 120(4):939–49. https://doi.org/10.1172/JCI40567 PMID: 20364091; PubMed Central PMCID: PMC2947423.
6. Pantanowitz L, Khammissa RA, Lemmer J, Feller L. Oral HIV-associated Kaposi sarcoma. J Oral Pathol Med. 2013; 42(3):201–7. Epub 2012/06/08. https://doi.org/10.1111/j.1600-0714.2012.01180.x PMID: 22672182.
7. Shieh MS, Islam JY, Rosenberg PS, Hall HI, Jacobson E, Engels EA. Projected cancer incidence rates and burden of incident cancer cases in HIV-infected adults in the United States through 2030. Ann Intern Med. 2018; 168(12):866–73. https://doi.org/10.7326/M17-2499 PMID: 29801099.
Gruffaz M, Vasan K, Tan B, Ramos da Silva S, Gao SJ. TLR4-mediated inflammation promotes KSHV-20.
Gorres KL, Daigle D, Mohanram S, Miller G. Activation and repression of Epstein-Barr virus and Kapo-22.
Fais T, Delmas J, Cougnoux A, Dalmaso G, Bonnet R. Targeting colorectal cancer-associated bacte-19.
Watari J, Chen N, Amenta PS, Fukui H, Oshima T, Tomita T, et al. Helicobacter pylori associated 17.
Yu X, Shahir AM, Sha J, Feng Z, Eapen B, Nithianantham S, et al. Short-chain fatty acids from peri-21.
Mataftsi M, Skoura L, Sakellaridou D. HIV infection and periodontal diseases: an overview of the post-23.
Bullman S, Pedamallu CS, Sicinska E, Clancy TE, Zhang X, Cai D, et al. Analysis of Fusobacterium per-16.
Guadalupe M, Pollock BH, Westbrook S, Redding S, Bullock D, Anstead G, et al. Risk factors influenc-25.
under antiretroviral therapy. J Acquir Immune Defic Syndr. 2011; 56(1):83–90. https://doi.org/10.1097/QAI.OB013e3181fcd928 PMID: 21084997; PubMed Central PMCID: PMC3051282.
26. Bik EM, Long CD, Armitage GC, Loomer P, Emerson J, Mongodin EF, et al. Bacterial diversity in the oral cavity of 10 healthy individuals. ISME J. 2010; 4(8):962–74. Epub 2010/03/26. https://doi.org/10.1038/ismej.2010.30 PMID: 20336157; PubMed Central PMCID: PMC2941673.
27. Verma D, Garg PK, Dubey AK. Insights into the human oral microbiome. Arch Microbiol. 2018; 200(4):525–40. Epub 2018/03/25. https://doi.org/10.1007/s00203-018-1505-3 PMID: 29572583.
28. Gopalakrishnan V, Spencer CN, Nezi L, Reuben A, Andrews MC, Karpinets TV, et al. Gut microbiome modulates response to anti-PD-1 immunotherapy in melanoma patients. Science. 2018; 359(6371):97–103. https://doi.org/10.1126/science.aan4236 PMID: 29097493; PubMed Central PMCID: PMCS829766.
29. Goodman B, Gardner H. The microbiome and cancer. J Pathol. 2018; 244(5):667–76. Epub 2018/01/30. https://doi.org/10.1002/path.5047 PMID: 29377130.
30. Aviles-Jimenez F, Yu G, Torres-Poveda K, Madrid-Marina V, Torres J. On the Search to Elucidate the Role of Microbiota in the Genesis of Cancer: The Cases of Gastrointestinal and Cervical Cancer. Arch Med Res. 2017; 48(8):754–65. Epub 2017/12/06. https://doi.org/10.1016/j.arcmed.2017.11.008 PMID: 29203054.
31. Routy B, Le Chatelier E, Derosa L, Duong CPM, Alou MT, Daillere R, et al. Gut microbiome influences efficacy of PD-1-based immunotherapy against epithelial tumors. Science. 2018; 359(6371):91–7. https://doi.org/10.1126/science.aan3706 PMID: 29097494.
32. Thomson PJ. Perspectives on oral squamous cell carcinoma prevention-proliferation, position, progression and prediction. J Oral Pathol Med. 2018. Epub 2018/05/13. https://doi.org/10.1111/jop.12733 PMID: 29752860.
33. Shiga K, Takeda M, Sajo S, Hori T, Sato I, Tateno H, et al. Presence of Streptococcus infection in extraoralpharyngeal head and neck squamous cell carcinoma and its implication in carcinogenesis. Oncol Rep. 2001; 8(2):245–8. Epub 2001/02/22. PMID: 11182034.
34. Mager DL, Haftfaje AD, Devlin PM, Norris CM, Posner MR, Goodson JM. The salivary microbiota as a diagnostic indicator of oral cancer: a descriptive, non-randomized study of cancer-free and oral squamous cell carcinoma subjects. J Transl Med. 2005; 3:27. Epub 2005/07/01. https://doi.org/10.1186/1479-5876-3-27 PMID: 15987522; PubMed Central PMCID: PMC1226180.
35. Guerrero-Preston R, Godoy-Vitorino F, Jedlicka A, Rodriguez-Hilario A, Gonzalez H, Bondy J, et al. 16S rRNA amplicon sequencing identifies microbiota associated with oral cancer, human papilloma virus infection and surgical treatment. Oncotarget. 2016; 7(32):51320–34. Epub 2016/06/05. https://doi.org/10.18632/oncotarget.9710 PMID: 27259999; PubMed Central PMCID: PMC5239478.
36. Chen W, Liu F, Ling Z, Tong X, Xiang C. Human intestinal lumen and mucosa-associated microbiota in patients with colorectal cancer. PLoS One. 2012; 7(6):e39743. Epub 2012/07/05. https://doi.org/10.1371/journal.pone.0039743 PMID: 22761885; PubMed Central PMCID: PMC3886193.
37. Kilian M, Chapple IL, Hannig M, Marsh PD, Meuric V, Pedersen AM, et al. The oral microbiome—an update for oral healthcare professionals. Br Dent J. 2016; 221(10):657–66. Epub 2016/11/20. https://doi.org/10.1038/sj.bdj.2016.855 PMID: 27857087.
38. Passariello C, Gigola P, Testarelli L, Puttini M, Schippa S, Petti S. Evaluation of microbiota associated with Herpesviruses in active sites of generalized aggressive periodontitis. Ann Stomatol (Roma). 2017; 8(2):59–70. Epub 2018/01/05. https://doi.org/10.11138/ads/2017.8.2.071 PMID: 29299190; PubMed Central PMCID: PMC5749375.
39. Lewy T, Hong BY, Weiser B, Burger H, Tremain A, Weinstock G, et al. Oral microbiome in HIV-infected women: Shifts in the abundance of pathogenic and beneficial bacteria are associated with aging, HIV load, CD4 count, and ART, AIDS Res Hum Retroviruses. 2018. Epub 2018/05/29. https://doi.org/10.1089/AID.2017.0200 PMID: 29808701.
40. Nogueira-Julian M, Guillem Y, Petersen J, Reznik D, Harris EV, Joseph SJ, et al. Oral microbiome in HIV-associated periodontitis. Medicine (Baltimore). 2017; 96(12):e5821. Epub 2017/03/23. https://doi.org/10.1097/MD.0000000000005821 PMID: 28328799; PubMed Central PMCID: PMC5371436.
41. Starr JR, Huang Y, Lee KH, Murphy CM, Moscicki AB, Shiboski CH, et al. Oral microbiota in youth with perinatally acquired HIV infection. Microbiome. 2018; 6(1):100. Epub 2018/06/02. https://doi.org/10.1186/s40168-018-0484-6 PMID: 29855347; PubMed Central PMCID: PMC5984365.
42. Presti RM, Handy SA, Droit L, Ghanounou M, Jacobson M, Shiboski CH, et al. Alterations in the oral microbiome in HIV-infected participants after antiretroviral therapy administration are influenced by immune status. AIDS. 2018; 32(10):1279–87. Epub 2018/06/01. https://doi.org/10.1097/QAD.0000000000001811 PMID: 29851662.
43. Gaboriau-Routhiau V, Rakotobe S,LECuyer E, Mulder I, Lan A, Bridonneau C, et al. The key role of segmented filamentous bacteria in the coordinated maturation of gut helper T cell responses. Immunity. 2009; 31(4):677–89. Epub 2009/10/17. https://doi.org/10.1016/j.immuni.2009.08.020 PMID: 19833089.
44. Arpaia N, Campbell C, Fan X, Dikiy S, van der Veeken J, deRoos P, et al. Metabolites produced by commensal bacteria promote peripheral regulatory T-cell generation. Nature. 2013; 504(7480):451–5. Epub 2013/11/15. https://doi.org/10.1038/nature12726 PMID: 24226773; PubMed Central PMCID: PMC3868984.
45. Qi X, Zhang M, Yang X, Hong N, Yu C. Faecalibacterium prausnitzii upregulates regulatory T cells and anti-inflammatory cytokines in treating TNBS-induced colitis. J Crohns Colitis. 2013; 7(11):e558–68. Epub 2013/05/07. https://doi.org/10.1016/j.crohns.2013.04.002 PMID: 23643066.
46. Atarashi K, Tanoue T, Oshima K, Suda W, Nagano Y, Nishikawa H, et al. Treg induction by a rationally selected mixture of Clostridia strains from the human microbiota. Nature. 2013; 500(7461):232–6. Epub 2013/07/12. https://doi.org/10.1038/nature12331 PMID: 23842501.
47. Ganem D. KSHV-induced oncogenesis. In: Arvin A, Campadelli-Fiume G, Mocarski E, Moore PS, Roizman B, Whitley R, et al., editors. Human Herpesviruses: Biology, Therapy, and Immunophylaxis. Cambridge2007.
48. Niederman R, Buyle-Bodin Y, Lu BY, Robinson P, Naleway C. Short-chain carboxylic acid concentration in human gingival crevicular fluid. J Dent Res. 1997; 76(1):575–9. Epub 1997/01/01. https://doi.org/10.1177/00220345970760010801 PMID: 9042080.
49. Orbe-Orihuela YC, Lagunas-Martinez A, Bahena-Roman M, Madrid-Marina V, Torres-Poveda K, Flores-Alfaro E, et al. High relative abundance of firmicutes and increased TNF-alpha levels correlate with obesity in children. Salud Publica Mex. 2018; 60(1):5–11. Epub 2018/04/25. https://doi.org/10.21149/8133 PMID: 29689651.
50. Krown SE, Testa MA, Huang J. AIDS-related Kaposi’s sarcoma: prospective validation of the AIDS Clinical Trials Group staging classification. AIDS Clinical Trials Group Oncology Committee. J Clin Oncol. 1997; 15(9):3085–92. https://doi.org/10.1200/JCO.1997.15.9.3085 PMID: 9294471.
51. Polizzotto MN, Ulrick TS, Wyvill KM, Alemain K, Peer CJ, Bevans M, et al. Pomalidomide for symptomatic Kaposi's sarcoma in people with and without HIV infection: a phase I/II study. J Clin Oncol. 2016; 34(34):4125–31. https://doi.org/10.1200/JCO.2016.69.3812 PMID: 27863194; PubMed Central PMCID: PMC5477825.
52. Yuan CC, Miley W, Waters D. A quantification of human cells using an ERV-3 real time PCR assay. J Virol Methods. 2001; 91(2):109–17. https://doi.org/10.1016/S0166-0934(00)00244-5 PMID: 11164492.
53. Youssef N, Sheik CS, Krumholz LR, Najer FZ, Roe BA, Elshahed MS. Comparison of species richness estimates obtained using nearly complete fragments and simulated pyrosequencing-generated fragments in 16S rRNA gene-based environmental surveys. Appl Environ Microbiol. 2009; 75(16):5227–36. Epub 2009/06/30. https://doi.org/10.1128/AEM.00592-09 PMID: 19561178; PubMed Central PMCID: PMC2755448.
54. Brown J, Pirrung M, McCue LA. FQC Dashboard: integrates FastQC results into a web-based, interactive, and extensible FASTQ quality control tool. Bioinformatics. 2017. Epub 2017/06/13. https://doi.org/10.1093/bioinformatics/btx373 PMID: 28605449; PubMed Central PMCID: PMC5870778.
55. Caporaso JG, Kuczynski J, Stombaugh J, Bittinger K, Bushman FD, Costello EK, et al. QIIME allows analysis of high-throughput community sequencing data. Nat Methods. 2010; 7(5):335–6. Epub 2010/04/13. https://doi.org/10.1038/nmeth.f.303 PMID: 20383131; PubMed Central PMCID: PMC3156573.
56. Hu J, Koh H, He L, Liu M, Blaser MJ, Li H. A two-stage microbial association mapping framework with advanced FDR control. Microbiome. 2018; 6(1):131. https://doi.org/10.1186/s40168-018-0517-1 PMID: 30045760; PubMed Central PMCID: PMC6060480. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2942,
1
],
[
2942,
6747,
2
],
[
6747,
10628,
3
],
[
10628,
14512,
4
],
[
14512,
15164,
5
],
[
15164,
16334,
6
],
[
16334,
20124,
7
],
[
20124,
20737,
8
],
[
20737,
21336,
9
],
[
21336,
22754,
10
],
[
22754,
27025,
11
],
[
27025,
30884,
12
],
[
30884,
34726,
13
],
[
34726,
37859,
14
],
[
37859,
41049,
15
],
[
41049,
41876,
16
],
[
41876,
46839,
17
],
[
46839,
50800,
18
]
]
} |
fb2247cf310f9dabea2958695a5e56dc670497b8 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 20,
"total-fallback-pages": 0,
"total-input-tokens": 44326,
"total-output-tokens": 14640,
"fieldofstudy": [
"Economics"
]
} | Inclusive Business for Smallholders’ Household Food and Nutrition Security: Disconcerting Results from an Analysis of a French Bean Agri-investment in Kenya
James Wangu1, Ellen Mangnus2, A. C. M. (Guus) van Westen1 and Alphons de Vocht1
Abstract
Inclusive business is regarded as having the potential to improve food security status in the Global South. Despite increased popularity among governments, donors and other development stakeholders, little is known about the approach’s impact on smallholder farmer communities. As such, the above-mentioned inclusive business promise on food security status largely rests on assumptions. This article scrutinises a case of smallholders’ French bean production for export market in Tharaka Nithi County, Kenya. The business model adopted in the initiative is termed inclusive and is intended to enhance food and nutrition security in the community. The empirical findings show that several contextual factors—in particular, access to land and water resources—limit the participation of the majority of farmers in the community. This leads to a notable level of exclusion. Moreover, the company risks negatively influencing local food security when food crops are substituted for an export crop that is not consumed locally. Results of this article demonstrate that while private sector-led development might contribute to higher economic productivity and access to food of better quality, it rarely changes the structural causes of food and nutrition insecurity, which are oftentimes related to access to production resources. We plead for increased scrutiny of the contextual factors when designing and implementing inclusive business models.
1 Department of Human Geography and Spatial Planning, Utrecht University, Utrecht, The Netherlands.
2 Wageningen University & Research Droevendaalsesteeg 4, Wageningen, The Netherlands.
Corresponding author:
James Wangu, Department of Human Geography and Spatial Planning, Utrecht University, Vening Meineszgebouw A, Princetonlaan 8a, 3584 CB Utrecht, The Netherlands.
E-mails: [email protected]; [email protected]
Keywords
Private sector, inclusive business, agribusiness, food and nutrition security, export, Kenya
Introduction
In 2005 the World Business Council for Sustainable Development (WBCSD) launched the concept of ‘Inclusive Business’ (GIZ, 2013). In the Global South, inclusive business in the agricultural sector entails integrating poor and low-income communities into global and regional value chains as producers, processors, workers and/or other forms of business partners. It is considered integral to enhance local livelihoods and food security (Bonnell & Veglio, 2011; Chevrollier et al., 2012; FAO, 2015; Sjauw-Koen-Fa, 2012). Since its launch, it has been promoted as a way of doing business while contributing to development objectives; presented as a unique opportunity to combine business development and growth, on the one hand, with combating poverty and hunger, on the other. Furthermore, by putting private businesses in the drivers’ seat, Global South’s development should not have to solely rely on exhausted public funding sources. The profit orientation of a business is expected to ensure sustainability as opposed to most donor-supported programmes that invariably depend on temporary public funding. Profit-driven businesses are thought to be able to scale up at speed and to a level unattainable by most other development interventions (Wach, 2012).
Against this backdrop, governments and international donors have gratefully accepted the concept in their pro-poor narratives (Zoomers et al., 2017). International organisations, like the Food and Agricultural Organisation (FAO) of the United Nations, WBCSD, the Word Bank and the African Development Bank have all put inclusive business models in the centre of their market-oriented interventions (da Silva & de Souza Filho, 2007; Gupta et al., 2015; Gupta & Vegelin, 2016; IFC, 2012; Shortall, 2008). Indeed, to facilitate the concept, a multilateral alliance has been established, Business Call to Action (BCtA, 2017), bringing donors together to promote inclusive business models to companies.
The potential of inclusive business to contribute to food security, as highlighted by FAO (2015), rests on a business model’s ability to address low-income smallholders’ bottlenecks to reach the market. This notion leads to our definition of ‘inclusiveness’ for this article. Integration in agricultural value chains is expected to improve farmers’ access to capital, offer risk mitigation tools such as insurance, promote knowledge and technology, and market information. All these are the assumed means to enhance farmers’ income through which food security would be improved (FAO, 2015; Golja & Požega, 2012; Ion et al., 2014; Jenkins et al., 2011; Lundy et al., 2014; WBCSD, 2013).
Despite increased popularity among governments¹, donors and other international development stakeholders, little is known about the impact of inclusive business on local communities (Bruni & Santucci, 2016; Chamberlain & Anseeuw, 2017; FAO, 2015; Sjauw-Koen-Fa, 2012; Smith, 2013). Furthermore, while it is presented as a model that contributes to improved local food security,
little is known on whether and how it contributes. As such, to better understand the relationship between inclusive business, and food security and nutrition, this study scrutinises a case of a proclaimed inclusive business initiative. The initiative involves smallholders’ French bean production for export market which is intended to enhance household food security in Tharaka Nithi County, Kenya. Guiding the study are the following research questions. First, who are the farmers that participate as producers in the business model? Second, what is the impact of an inclusive business model producing for distant markets on household food and nutrition security for the farmers in the production area?
The next section discusses the evolution of private sector development in policy and cooperation in the Global South, with the aim of providing the background of inclusive business now embraced as a crucial strategy to address food security. The third section introduces the case study and its context. In the fourth section we detail the research methodology employed. The fifth and sixth sections present the findings on both research questions. We conclude the article with a discussion of the findings and policy considerations.
From Private Sector Engagement to Inclusive Business
Private sector involvement in development interventions is not a new phenomenon. In recurring periods since the emergence of development assistance in the 1960s, the private sector has been positioned as an essential agent in promoting development. Most noticeable was the period between 1985 and 1994, the era of structural adjustment programmes, in which global development actors—the World Bank and International Monetary Fund—compelled African states to rigorously ‘roll back’ their involvement in public services such as extension, research in the agricultural sector, and other activities; presuming a private sector takeover. Aid was provided under strict conditions. In the era of ‘good governance’ in the late 1990s, the state was re-appreciated, this time for its role in overseeing globalisation and in facilitating economic growth via promotion and protection of foreign investment and new regimes of property rights. Structural adjustment gave way to poverty reduction strategy programmes that involved the state in tandem with the market (Craig & Porter, 2006).
Following the global financial crisis of 2007–2008, however, the private sector again obtained a central position in development. Development assistance shifted from supporting civil society to collaborate with—oftentimes financing—companies, philanthropists and investors. ‘Trade not aid’ became the new adage of many donors (European Commission, 2016; Mawdsley, 2017; Ploumen, 2013; Richey & Ponte, 2014). Trade was presented as a considerably equal way of achieving change instead of development aid in which the beneficiaries were usually perceived to be dependent on the provider. Against this backdrop, in 2011, the Busan Partnership Agreement for effective development cooperation referred to the private sector as critical for achieving sustainable development (OECD, 2011). Currently, in the so-called era of ‘retro-liberalism’ (Murray & Overton, 2016, p. 248), governments are actively intervening to promote business interests as crucial instruments for development; a clear departure
from the preceding neo-liberal stance of rolling back the state to allow the market to do its work. The donor programmes supporting the establishment of cross-border agri-value chains linking smallholders to regional and global markets are seen as the contemporary example.
However, merely engaging the private sector as the lead agent of development does not automatically lead to positive outcomes for local communities. If business growth is to improve the livelihood of the many through production support, employment, market uptake and market outlets for smallholders, then it should look beyond the bottom line and incorporate social objectives of inclusion and sustainability in its business models (Deininger, 2011; European Union, 2013; West & Haug, 2017). Responsible investments that consider the impact on the local population and environment at the very least pursue a ‘do no harm’ principle in their operations (Cotula et al., 2009; Teklemariam et al., 2017). Also, public support to the private sector is often made conditional on ‘doing good’, that is, including positive societal impact in the objectives of the business model. Thus, the concept of inclusive business is embraced when a win-win situation for both the company and the local population is guaranteed, which may hamper the scope of their contribution to development.
**Agribusiness and Food Security Context in Keanya**
Kenya provides a compelling research area to assess the role of inclusive agribusiness in fostering food security. The country, especially communities in the rural areas, relies strongly on agriculture for livelihoods and food security. The agricultural sector employs about 70% of the rural inhabitants and accounts for 26% of Kenya’s gross domestic product (GDP) and 65% of the total export revenue (FAO, 2018b). The primary producers (75%) are smallholders that practice rain-fed agriculture on plots averaging 0.2–0.3 hectare (ha) per household (Mbata, 2013; Mohajan, 2014). The traditional agricultural practices mostly employed by these farmers are understood as the main hindrance to economic growth (von Kimakowitz, 2012). The pathway towards improved food and nutrition, according to the government, is through agricultural modernisation.
In 2008, the country launched its long-term development strategy—Vision 2030, which aims for transformation in the agricultural sector, particularly by attracting foreign investments, as the vehicle to attain an annual GDP growth of 10% (GoK, 2007). Towards this endeavour, the Kenyan government has recently adopted several policies, including the National Agricultural Extension Policy (GoK, 2012b), National Horticulture Policy (GoK, 2012c), Agricultural Sector Development Strategy 2010–2020 (GoK, 2010), and National Agribusiness Strategy (GoK, 2012a). Together, these policies are set to improve agricultural extension services, sectoral knowledge transfers, productivity, commercialisation and competitiveness of smallholder farmers. Building on the same logic, in 2011 the government developed the National Food and Nutrition Security Policy that was adopted a year later (GoK, 2011), to ensure an increase in the production of quality and diverse food accessible to all Kenyans. The increased production and
accessibility are to be bolstered by improved storage to minimise post-harvest losses and to facilitate the establishment of well-functioning markets.
Notwithstanding the development efforts in the sector, Kenya still grapples with the problem of food insecurity. In 2011, approximately 10 million Kenyans were chronically food insecure (GoK, 2011). Despite notable improvement, food insecurity remains a significant concern through 2016, considering that an estimated four million people consumed meals lacking in dietary diversity, but made primarily of staples with some green vegetables and oil flavours (WFP, 2016). In 2018, it was reported that approximately 2.4 million people relied on relief food aid (FAO, 2018a). The persistent food crisis in the country has even prompted the government to make food and nutrition security among the crucial target in its ‘Big-Four Agenda’, in which it proposes an increased role of the private actors in the agricultural sector (KEPSA, 2017). In light of these circumstances, several smallholders’ targeted inclusive business initiatives by the Kenyan government and development agencies have been implemented in the recent past, and others are underway across the country.
**Materials and Methods**
**Case Study**
This article scrutinises a business case that claims to be inclusive. It concerns a Kenyan company, referred to as Bean in this article, that engages smallholders in the production and supply of French bean, a high-value horticultural crop for export. The company sources French bean from nine counties across Kenya and the produce is destined for Germany and France. French bean is an essential horticultural crop in the country, where it contributes to the livelihoods of approximately 52,000 smallholders (DGICD, 2018). It is also known to account for 19% of the total horticulture value and 25% of the total volume of vegetables exported from Kenya (RSA, 2018).
Bean is supported by the Facility for Sustainable Entrepreneurship and Food Security (FDOV) of the Dutch Ministry of Foreign Affairs, a financing instrument set to stimulate public–private partnerships (PPPs) with an aim to contribute to food security in developing countries (NEA, 2018). The guiding principle for supporting the French bean scheme run by Bean is based on the idea that, given the extremely small farm sizes in the densely populated eastern Kenya, smallholders should be assisted in growing valuable cash crops. It is supposed that high-value cash crops should lead to enhanced income, which in turn should enable the farmers to meet their basic needs such as food and others. In this initiative, Bean collaborates with an international NGO, a Dutch private agricultural service provider, and a Kenyan government export crop regulation agency. Rolled out in 2016 by the consortium, the initiative is expected to run for five years and involves a total budget of about € 5 million, half of which is FDOV’s contribution. Bean contributes by setting up a French bean processing factory at the export processing zone (EPZ) in Kitengela, south of Nairobi. The other partners contribute through expertise provision.
Through the PPP, Bean’s main goal is to provide 48,500 smallholder farmers with production support and a guaranteed market. The company mission is to improve availability and affordability of quality and nutritious food for local and regional consumers, enhancing households’ food security in the context of a private enterprise. The initiative’s theory of change revolves around removing the barriers to food security in the region, including—(a) low production and productivity attributed to poor agricultural practices, post-harvest losses, inaccessibility of improved technologies and lack of extension delivery systems; (b) market inaccessibility due to poor smallholder’s organisation capacity and lack of market information; and (c) cross-cutting issues such as inability to produce within the quality standards essential to meet the market standards, and addressing weak infrastructure. Bean assumes that by improving access to quality inputs, training farmers on good farming practices, and a guaranteed market, farmers should be able to enhance their income, and thereby improving their ability to afford quality food.
The community of Kaanwa in Tharaka Nithi County was purposively sampled. Tharaka Nithi is reported to be among the counties in Kenya where households are forced to adopt coping strategies due to high levels of food insecurity (WFP, 2016). This includes consuming less preferred or borrowed food, reducing the size or number of meals and/or adults skipping their meal to provide for children. According to the government, even with the support of food aid, the households in the county only manage ‘minimally adequate food consumption but are unable to afford some essential non-food expenditures without engaging in irreversible coping strategies’ (GoK, 2017, p. 7). The high food insecurity prevalence compared to other counties where Bean is operating made the county the logical site for this research.
Bean contacts individual farmers and/or groups of farmers with whom it works on pre-agreed prices, quality and delivery dates arrangement. The company provides farmers with input including seeds, chemicals and fertiliser. Several other services are provided by the other project partners including soil testing and analysis at a subsidised price, knowledge on climate-smart technologies such as drip irrigation, training and advice on soil conservations methods and biological pest control. These services are offered in the form of demonstration plots to all farmers (programme participants and non-participants) in the implementation areas. The NGO trains extension service providers—promoter farmers and trainer of trainers on good agricultural practices (GAPs), who in turn train the farmers. The Kenyan government export crop regulation agency is responsible for facilitating the contracting process between the farmers and Bean. It is also in charge of farmers’ training and certification to ensure they are producing within the set standards—KenyaGAP and GlobalGAP.
To be able to research this contemporary phenomenon that is taking place in a dynamic context, a case study method was found to be most suitable. A case study method allows for an ‘up-close or otherwise in-depth understanding of a small number of cases, set in their real-world contexts’ (Yin, 2011, p. 3). It encourages a context-specific analysis (Zainal, 2007). The French bean intervention studied involves both public and private stakeholders, locally and internationally interacting in an environment that is changing in terms of resources, weather, information and interests. Such empirical inquiry necessitates a unique research
strategy that provides for extensive research methods, especially with regards to
data collection and analysis that a case study provides, fortunately.
Mixed-method data collection tools were employed in this research
including surveys, focus group discussions and in-depth interviews. They
were considered useful for getting a detailed understanding of fundamental
farmers’ dynamics in the research site, especially those triggered by or linked
to the intervention. Secondary data sources, including national statistics,
reports, policy documents, media sources, blogs and other documentations
were consulted to triangulate some findings.
The data collection exercise was conducted from January to April 2017. The
survey data were obtained from a clustered sample comprising three categories of
farmers. As we are interested in understanding the impact of Bean, we compared
the farmers engaged in the company’s business model against those that do not
take part. The latter comprised two categories: farmers with access to water for
irrigation and those without. This categorization assumed that ability to irrigate is
a significant determinant of the type of farming practice, and the resulting
household’s livelihood outcome. The category that was composed of farmers
without access served as a control group in the research.
The total number of farmers’ households that worked with Bean in Kaanwa
during this study was small. Hence, all known participants (25) were recruited for
the survey. The second category, comprising households with access to water for
irrigation but did not work with Bean was a larger group; a random sample of 29
households was obtained from a list of the members of the irrigation projects. A
snowballing sampling method was employed to select 56 households without
access to water for irrigation: 28 of the farmers in the first two categories were
tasked with providing names of at least four farmers without access to water for
irrigation in their neighbourhood. From the list, two households from every four
were randomly selected as respondents. The survey data captured households and
farming characteristics, value chain relations and economic impact, food security
status, savings and credit options, and perceptions on farming.
The focus groups discussions and in-depth interviews involved 60 smallholder
households, randomly sampled from the list of 110 survey respondents; 20
households from each category. The interviews focussed on the farm–firm
relationship, and farmers’ perception on the impact of Bean on income and food
security status. Five key-informants including a representative from Bean, local
agricultural expert, and three irrigation project leaders were also interviewed. The
intention was to enhance understanding of the local agribusiness context.
Respondents in this study were on a voluntary basis, and prior to acquiring any
data, consent was requested and confidentiality assured. In total, 175 individuals
were engaged in this study. Survey data were cleaned, coded and analysed using
STATA (version 13) software. The analysis comprised descriptive statistics and
regression tests. the descriptive statistics included summaries of the variables—
such as land, farming practice, crops grown and income—that characterise the
farming community in the study area. Under the assumption that land is the
primary resource base that dictate smallholders’ participation in Bean business
initiative, a multinomial logistic regression test was conducted with land as the
dependent variable and the three farming groups as predictor variables: farmers
without access to water for irrigation—as a control group (0), farmers with access and working with Bean (1) and farmers with access but not working with Bean (2). To isolate the effect of Bean, we assessed the effect of irrigation on income and on the household dietary diversity score (HDDS) of Bean and non-Bean farmers, with the non-irrigation group as control. Two more multinomial logistic regression tests were carried out to assess whether income varied among the groups. This analysis involved a similar model to the above-mentioned but in this case income and HDDS replaced land as the dependent variable.
The qualitative data from interviews and focus group discussions were coded and analysed using NVIVO software. The resulting themes and patterns were used to triangulate and validate the relevant related results obtained from the statistical analysis.
Findings
Research Site and Farmers Characteristics
Farming is the primary means of livelihood in Kaanwa. The households surveyed own an average of 0.8 ha farm size, with the majority (60%) having less than the average. These farms are bound to shrink further in the future as farmers continue to subdivide their parcels to their children; 96% of survey participants plan to do so. Most of the farmers (65%) produce both for subsistence—own food—consumption and for the market. A third (34%) engages only in subsistence farming and only 1% exclusively in commercial agriculture.
Seasonally, maize and beans predominate the type of crops cultivated in the community. Of the households surveyed, 85% and 70% produced these crops, respectively, in the previous seasons. Both are primarily produced for home consumption, but in case of surplus are sold locally at farmgate and town centres. The commonly produced horticultural crops include bananas (16%), French bean (13%), Sukuma wiki (7%) and tomatoes (7%). Due to the poor rain condition in Kaanwa, horticulture farming is confined to farmers that have access to water for irrigation. All the fruits and vegetables except for French bean are cultivated for the local markets. Other less common crops cultivated include cowpeas, sunflower, tobacco, Miraa (Khat) and millet. Sunflower and tobacco have been newly introduced in the community.
A minority of households have other means of livelihood, including business ownership, formal employment and wage labour, constituting 20%, 12% and 10% of all the study participants, respectively. The formal employment comprises civil service jobs, such as teaching, police and health care. Wage labour involves working in other households’ farms and/or in construction sites.
Inclusion for All? Who Participates in the French Bean Business?
The first question that this study sought response was: who participates as producers in the Bean initiative? In other words, how inclusive is the business
The scheme operated by Bean and partners is, in principle, open to every farmer in Kaanwa who wishes to enrol; it does not actively select participants. However, the first prerequisite of participation is access to water for irrigation. French bean cannot be grown in Kaanwa, and most other parts of the country, without water for irrigation. Kaanwa has two irrigation schemes. One of them is Mbwiru-Mwanjati scheme, initiated by a German Development Bank (KFW), and has 250 members. The project draws its water from river Tongo. Of all members that registered at the scheme’s inception, about 7% dropped out because they were unable to repay the loan. According to a key informant—a senior member of the irrigation project—outside of the dry periods of the year each farmer can adequately farm up to 1 acre of land with water from the scheme. The second irrigation scheme is called Ciambaraga. Out of its initial 200 registered members at the inception, about 50 withdrew from the project, claiming they were unable to meet the costs. This higher rate of attrition may relate to the absence of an adequate grace period for loan repayment in the start-up phase. Also, water access from this project is less reliable than in the case of Mbwiru-Mwanjati. Because of the unreliability, Ciambaraga members are often forced to use water in shifts due to scarcity. This situation is attributed to the many irrigation schemes drawing water from the same river, Naka, according to the key-informant from the scheme.
In both irrigation projects, membership was open to everyone in Kaanwa community. However, it was indicated in the focus group discussions and by the informants, only about 20% of the farming households are part of the mentioned irrigation schemes. Lack of financial capability is the main reason for non-participation, as confirmed by key-informants and in the focus group discussions. To become a member of Mbwiru-Mwanjati and Ciambaraga, each farmer was required to pay Ksh.66,000 and Ksh.120,000, respectively, excluding a monthly fee of Ksh.200 which is meant for the project’s maintenance. While the amount was not a one-time payment, other more immediate household needs such as school fees, farming capital and health care, meant meeting the project costs is a major challenge. Irrigation, thus, is shown to be a significant constraint on partaking in the Bean scheme, raising questions about the inclusiveness of the business model, given that most of the vulnerable households are unable to participate. While it is unreasonable to expect all farmers in Kaanwa to participate in the Bean scheme, inclusive or not, when over three-quarters of the farming population is not able to meet the essential requirement—access to water for irrigation—then the business model must be deemed selective. It is not likely that much can be done about this limitation, as rainfall scarcity imposes a ceiling on the expansion of local irrigation schemes, according to no less than 98% of all farmers interviewed.
This, however, proves not to be the whole story. Survey data also reveal that Bean farmers have more land than others without access to water for irrigation. On average, farmers working with Bean cultivate on average 2.18 acres, compared to 1.65 acres for those not practising irrigation, as indicated in Table 1. The difference observed is statistically significant, as shown in Table 2.
Our conclusion, therefore, is that the business model is not inclusive. It is understood, however, that such business models cannot reach everybody. Nevertheless, in Kaanwa, the inclusiveness is bound by real constraints, access to water and land resources, that make the French bean scheme significantly selective.
French Bean Business Income Outcome
There may be strong limitations to the business model, but then, does it deliver on its promises to those enrolled? What is the impact of Bean’s activities on income? As outlined, Bean scheme aims to achieve enhanced food security by means of enabling farmers to grow a valuable export crop and thus improving their ability to purchase quality food. The first outcome expected is improved income, and a secondary impact is on food and nutrition security.
First, we have a look at current annual income levels for the three subgroups of farmers in Kaanwa: Bean participants, non-Bean irrigating farmers and those without irrigation (Table 3). The survey results show a remarkable advantage of the irrigation farmers not engaged in French bean production compared to ‘Beaners’ and the non-irrigating group—a statistically significant difference, and an insignificant difference between the ‘Beaners’ and non-irrigating group (Table 4).
The difference is such that it cannot be explained by this group’s advantage in land size or by an alternative miracle crop that dwarfs the results of French bean cultivation. Closer scrutiny reveals that the relatively advantaged group of no-Bean irrigation farmers are on average older (52 years, and 43 years for Bean growers), and are more educated (80% as compared to 20% of Bean growers have higher education). This is in line with the decisive differentiating factor—many have formal jobs alongside farming. In fact, 8 out of 19 people in the non-Bean group are employed as teachers or in other government jobs or are enjoying retirement benefits from previous formal employment. For these employees, farming is a side job—a useful addition to salaries that stand out by local standards but are not so
Table 1. Summary Statistics: Average Land Size by Irrigation (In)Accessibility.
| Group | N | mean | sd | cv |
|-------------------|----|-------|------|-----|
| No irrigation | 53 | 1.656 | 1.062| 0.641|
| Irrigation Bean | 23 | 2.185 | 0.942| 0.431|
| Irrigation No Bean| 19 | 3 | 1.871| 0.624|
Source: Survey.
Table 2. Multinomial Logistic Regression Estimating Likely Group as Explained by Land.
| Group | Coef. | St.Err. | t-value | p-value | [95% Conf Interval] | Sig |
|------------------|--------|---------|---------|---------|---------------------|-----|
| No irrigation | – | – | – | – | – | – |
| Irrigation Bean | 0.430 | 0.229 | 1.88 | .060 | −0.018 - 0.878 | * |
| Irrigation No Bean| 0.798 | 0.238 | 3.36 | .001 | 0.332 - 1.264 | *** |
Mean dependent var 0.642 SD dependent var 0.798
Pseudo r-squared 0.077 Number of obs 95.000
Chi-square 14.541 Prob > chi2 0.001
Akaikie crit. (AIC) 181.724 Bayesian crit. (BIC) 191.939
Source: Survey.
Note: ***p <.01, *p <.1.
generous that they enable a comfortable life. This in addition to older age also explains why this relatively privileged group is less inclined to take up the labour-intensive growing of French bean. Furthermore, their schedules do not make it easy to combine these activities. Bean growers and non-irrigating farmers also tend to have side jobs as a livelihood strategy by combining several income sources, often engaging in informal activities and casual labour.
More ominous for the appraisal of the Bean scheme, however, is the observation that participating farmers barely earn more than the excluded group of farmers without access to water irrigation (Tables 3 and 4). A constraint for our analysis is that our survey gives us a single observation in time, limiting the ability to make a reliable comparison between the situations ex-ante and ex-post introduction of Bean. Yet we can draw conclusions from another perspective, and that is the returns to farmers of the French bean endeavour. The following calculations are quite revealing.
The minimum amount of French bean a farmer is required to cultivate within the Bean scheme is one unit of seeds which is equivalent to 4,000 seedlings. A single unit occupies about 0.06 acre plot size. Depending on the amount of land and water available, farmers are free to plant as many units as they deem possible. According to verified accounts from both key-informants (agronomists and different company representatives working in a French bean business), a farmer should optimally be able to produce up to 400 kilogram (kg) from a single unit of seeds. Assuming they manage to produce 400 kg and have been provided with all the inputs required for production (seeds, fertiliser, agri-chemicals, and possibly...
advance for harvesting costs) by Bean, 100 kg/25% of the produce per unit is deducted from the final price to cover these costs. In the case of farmers who only take seeds from the company, the corresponding share would be 10% of the benchmark yield. As confirmed by the Bean representative, few farmers manage to reach the optimal output target. In fact, in the crop cycle prior to this study survey date in Kaanwa, the average yield was 58 kg per unit of seeds, which was bought at Ksh.40 per kg (Ksh.110 ≈ €1). Relative to the cost of production, the average total revenue (Ksh.2,320) per farmer amounted to a loss for the farmer considering the costs of production (see Table 3). In this particular crop cycle, only 20% of the farmers made a profit. The year in question, it should be mentioned, was unfavourable because of drought, so an average year is likely to produce better results. Nonetheless, droughts are not exceptional in the region and can be expected to occur regularly in the future due to the impact of climate change.
Several factors explain this outcome. First, a lack of consistent and reliable access to water needs to be highlighted again. Bean farmers may have access to water for irrigation, but the erratic rainfall patterns mean they are liable to shortages, as mentioned in the interviews. Second, French bean production is capital intensive, as discussed above, and a lack of funds among farmers translates into inadequate investment in the business, hampering crop productivity. Nevertheless, among all the farmers surveyed, Bean farmers spend notably higher proportion of their income on input than farmers from the other two subgroups. The amount spent on hired labour (Table 5) may vary according to household size, but all French bean growers face the need to hire workers for harvesting, the most labour-intensive part of the production process. This needs to take place in a short period of time to avoid quality issues because of French bean’s highly perishable condition. And finally, shortcomings on the side of Bean and consortium partners play a role in the depressing profitability. All farmers maintained that Bean extension services were poor as exemplified by delayed supply of inputs, and/or unavailability of suggested agro-chemicals. Bean acknowledges this problem and puts the blame to their extension officers.
Given the questionable returns, one may wonder why farmers continue to work with Bean. Participants in the scheme mention the provision of seeds, which they imply eases their capital burden. They state that while they may
| Items | Cost/returns (Ksh.) |
|-------|---------------------|
| Seeds (one unit) | 1,000 |
| Manure & fertiliser | 1,200 |
| Agro-chemicals & foliar | 450 |
| Labour (weeding, spraying, top dressing, watering, harvesting) | 6,700 |
| Total costs | 9,350 |
| Total returns from 400 kg | 12,000 |
| Profit (returns–costs) | 2,650 |
*Source:* Survey.
experience loss in some cycles, they remain positive about getting rewarding returns in other cycles. Furthermore, French bean has a short production cycle (45 days), has a certain market and pays relatively better compared to other horticultural crops such as tomato. Tomato takes longer to mature (60–85 days depending on the variety) and its market is highly volatile, as was stressed in focus group discussions. The short French bean cycle yields quick cash in a lump sum which is helpful in meeting farmers’ financial needs. Indeed, the crop is perceived as having some attractions although overall returns are low. With 80% of farmers not making a profit from French bean production in our survey, it can be stated that the Bean initiative does not significantly contribute to improved income in Kaanwa.
**French Bean Business Impact on Food and Nutrition Security**
The ultimate goal of the Bean scheme, we should remember, is to increase food security in this food insecure part of eastern Kenya. For this purpose, raising income was an intermediate objective rather than the final goal. Disappointing outcomes in terms of income, as discussed above, imply that not much can be expected from the anticipated initiative’s theory of change, yet it is still worthwhile to check if impacts on food security can be observed.
From Figure 1 it becomes clear that overall, the majority of the surveyed farmers in Kaanwa (54%) are ‘food insecure’, as measured by the Household Food Insecurity Access Scale (HFIAS; Coates et al., 2015). HFIAS provides an approximation of a household’s food insecurity prevalence. As Figure 1 indicates, food insecurity is most prevalent among farmers without access to water for irrigation, as can be expected. Among irrigating farmers, slightly more farmers are food secure in the group not working with Bean. This will come as no surprise
in view of income data discussed above—their higher income can be expected to translate into better food security status. Figure 1 shows a certain link but does not establish causality between Bean and food security in Kaanwa.
A direct effect of Bean on food security could be expected if farmers would consume part of the French bean they produce. French bean is rich in nutrients and may thus serve as a valuable addition to local diets. The direct impact pathway must be discounted, though, as French bean do not form part of the local dish and are rarely consumed in farmer households. Recently, French bean has found its way to the premium supermarkets shelves, hotels and hospitals in the city (SNV, 2012). In the rural areas, however, people tend to view the vegetable as an immature bean even though local cuisine does recognise the mature seeds of kidney bean and other traditional bean varieties as valuable items for consumption.
A review of the composition of the meals consumed by the households in Kaanwa further details the local nutrition security condition. Much of daily household diets comprise cereals (98% of the households), some dairy products in the form of tea with milk (96%), some vegetables (some kale and tomato; 84%), onions, legumes—primarily bean (66%)—oil and fats (63%) and fruits (50%). As the figures suggest, half of the households do not consume fruits, and those that do can do so only when fruits are in season and available in their individual farms—it is not an item people tend to purchase. Further, nearly half of the households’ meals lack in plant-based proteins (a strength of French bean), and oils and fats. The majority of people hardly consume animal products considering only 23% and 3% can afford meat and eggs, respectively. Spices, condiment and beverages complete the menu (3%). So, a direct contribution of French bean production to local food security must be discounted for cultural reasons (taste), although the vegetable would offer a valuable contribution in terms of nutrition.
An indirect effect on nutrition security can be mainly expected through income. Table 6 below shows the result of the multinomial logistic regression analysis with the three farmer groups (no irrigation, irrigation participating in Bean and other irrigation) as predictor variables and the household dietary diversity score (HDDS; Kennedy et al., 2011) as the dependent variable. HDDS measure the quality of the diet consumed by a household. The result shows no significant relationship between enrolment in the Bean scheme and HDDS compared to non-irrigating farmers. As we have seen, the Bean scheme has little effect on income and thus it is no surprise that it does not contribute to the nutrition diversity in the community. Non-Bean farmers, however, seem to have a significant difference (at less than 10% significance level) in dietary diversity compared to non-irrigating farmers. In the case where participation in business yielded better income for Bean farmers, it would still have little impact on nutrition security. Based on interviews and focus groups discussions, besides spending their income on basic food items such as sugar, salt, cooking oil, onions and staples in difficult times, much of the money goes to other non-food needs including education, healthcare, clothing, labour and agricultural input. Against these backgrounds, for French bean farmers it is right to claim that the scheme exposes participating households to more risk, in the sense that they are less involved in food production. Such is the case of local staples production—maize and bean. According to the survey,
10% of *Bean* scheme participants do not grow staples, all other farmers do. This implies they cannot fall back on subsistence, should their harvest fail as already happened in the previous crop cycle.
**Bean Impact in Local Context**
This case study shows clear limitations to the increasingly popular strategy of fostering rural development through inclusive business models linking smallholders to corporate agribusiness. Some of these constraints may pertain to specific circumstances of the case. The performance of the *Bean* company could improve, and local rainfall conditions vary positively in some cases. Yet other limitations point to shortcomings of the approach itself. First, the limit to inclusiveness. There are pre-conditions for participation in the business model that exclude a considerable part of the community involuntarily. Access to water for irrigation is required for a farmer to engage in French bean production; also, those included tend to have more than the average size of land. This points to a mismatch between the rationale of the development strategy and the needs of local farmers. While the inclusive business model is based on the premise that land scarcity and fragmentation must be countered by adopting more remunerative export crops rather than local staples, local farmers tend to prioritise their family’s subsistence needs before considering a cash crop. Results in this study suggest these farmers may well be right, considering the lack of profit for the *Bean* farmers. In the end, considering not all farmers with irrigation participate in the French bean production—less than 20% of the total local farmers who have taken part in the scheme—it is a severely limited form of inclusiveness. In fact, it is probably realistic to state that any such scheme can reach only part of the community with a reasonable chance of success.
In Kaanwa, the majority of farmers do not have access to water for irrigation—the most food insecure in the community never qualified for being taken on board
**Table 6. Multinomial Logistic Regression Estimating Likely Group Based on HDDS.**
| Groups | Coef. | St.Err. | t-value | p-value | [95% Conf Interval] | Sig |
|-----------------|-------|---------|---------|---------|---------------------|-----|
| No irrigation | – | – | – | – | – | |
| Irrigation | –0.132| 0.182 | -0.72 | .468 | -0.490 | 0.225 |
| Bean | | | | | | |
| Irrigation No | 0.328 | 0.183 | 1.79 | .073 | -0.031 | 0.687 * |
| Bean | | | | | | |
Mean dependent var 0.755 SD dependent var 0.848
Pseudo $r$-squared 0.023 Number of obs 110.000
Chi-square 5.268 Prob > chi2 0.072
Akaike crit. (AIC) 229.752 Bayesian crit. (BIC) 240.554
*Source:* Survey.
*Note:* *p* <0.1.
in the Bean scheme. This again highlights the gap between the offices where development policy on food security and business models are being designed and conditions in the places where these ideas are to be realised. Kaanwa area is prone to drought to make French bean an important crop on which to build local food security, although it may add a useful source of cash income. As shown, luring farmers away from subsistence crops into French bean may increase the risk of food and nutrition insecurity in case of drought or market disruptions, and it is likely that such considerations partially explain why farmers with the smallest plot sizes tend not to join Bean. The implication for policy is that inclusive business models may well be part of a local development strategy but should always be complemented by additional initiatives to take care of those that cannot realistically be integrated into a commercial value chain. There is nothing wrong with inclusive business models as one potential contributor to local rural development. The problem arises when donors and planners stop there and consider the job done by means of this approach alone. In that case, the farmers most in need of the support to improve their households’ food security situation are excluded in such efforts.
What then should be done for those rural people who cannot benefit much from the inclusive business approach—in this case food security? Easy answers are not available, but we may take some leads from Kaanwa case study. We have seen that the most successful farmers (in terms of income generation) are those that combine farming with another revenue source, specifically formal employment. While this is obviously an option open to only some people, it raises the question whether the policy should aim so single-mindedly on encouraging farmers with very small landholdings into agriculture as has implicitly been done in the inclusive business approach and generally in other rural development interventions. Farming has a crucial role in local survival as a fall back option and through subsistence food production. But off-farm employment may eventually offer more opportunities—as in the historical experiences of wealthier countries, where part of rural labour gradually moved to other sectors, in the process also freeing up land that can be consolidated in more sustainable farm units. Tentatively it may be argued that the ambitions of many smallholders in places like Kaanwa are also orientated elsewhere than to local farming. If we take a clue from the expenditure patterns observed in this study, increased income does contribute to better food security, but not as much as one would expect. The reason, as transpired in interviews, was that many people have other priorities, and prefer to invest more in education for their children—perceived as the key to a better life—than in acquiring better food.
The study findings suggest that the interest of the business investment does not align with the intended goal for the farmers, which would entail a guarantee to sufficient and quality food for own consumption, amid scarce water and land resources. Furthermore, for Bean to be able to reach low-income farmers in the French bean business—considered a high-risk investment—public funding is critical. This aligns with Murray and Overton’s (2016) analysis that even though private sector-led development is presented as cost-efficient compared to development aid programmes led by NGOs, in most cases public support is essential. The intervention is packaged as an impact investment expected to enhance food security while being profitable, yet the participation of a private
sector actor is primarily made possible due to the promise of public support. Researchers observe a global trend, which as previously referred to as ‘retro-liberalism’, in which public finance is transferred to the private sector to facilitate a role in development. For example, in Western countries, where following the collapse of global capital markets, ‘retro-liberalism’ is the principal means by which markets were reconstructed through large public stimulus packages. Murray and Overton (2016) claim that the intention of such interventions, in general, is not to address social inequalities and injustice but to reignite capitalism. According to these scholars, similar intentions of self-interest underlie current ‘trade not aid’ regime. Although we cannot claim this for Bean, the case study has made clear that the business interests do not align with the priorities of the community. It, therefore, brings into question whether it is ideal for making the private sector responsible for collective societal goals, such as smallholders’ food and nutrition security, that are often not directly in line with company advantage, especially where public funds are involved.
In sum, the study highlights that livelihood diversification is a critical strategy employed by all farming households. Indeed, besides agriculture, each family has (an) additional income source(s) such as wage labour, formal employment, and/or running a business. According to Hakizimana et al. (2017, p. 510), mainly due to meagre farm sizes, smallholder farmers’ livelihood includes a combination of ‘own petty commodity production and wage labour’. To contribute to local food security, policymakers and parties implementing interventions related to Bean scheme should start from the real-life situation in the field. This means ensuring local priorities and contexts as expressed by the communities are strongly considered where such interventions ought to be implemented. Our case study shows that when the business proposition is leading the intervention and the design, improved food and nutrition security is not guaranteed. In Kaanwa, several factors explain this:
1. The theory of change does not fit the local context, that is, using small parcels of land to cultivate a high-value crop intensively did not provide farmers with a higher income and/or improve their household food and nutrition security status. In fact, this limited them from growing crops that could have bolstered their food security needs.
2. There is no clear link between export of a high-value vegetable that is not consumed locally and improved local food security.
3. Part of the community is excluded. Even if a positive effect had been realised, this would have been only for the minority—the relatively well off—possibly contributing to growing inequality.
Private sector-led development may contribute to higher small farms productivity or even access to better quality food. However, rarely does it change the structural causes of food insecurity, which are often fundamental to access to production resources. Increased scrutiny of various structural factors warrants further study, for instance, paying more attention to the local context. Similarly, more effort needs to be directed towards establishing what determines the diversification strategies adopted by smallholder households and how they complement crop production in meeting food and nutrition needs.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship and/or publication of this article: This work was supported by funding from the Netherlands Organisation for Scientific Research (NWO)—Project Number W 08.250.206.
Notes
1. Kenyan government through its business agency: Kenya Investment Authority, has actively promoted the participation of investors in primary production and value addition via processing of agricultural produce.
2. Kenya experiences two rain seasons. The ‘long rain’ season takes place between March and June, and the ‘short rain’ season occurs from October to December. In the study region, these seasonal patterns are vice versa; ‘long rain’ season takes place from October to December.
References
BCtA. (2017). About BCtA. https://www.businesscalltoaction.org/business-call-action#blocks-body
Bonnell, V., & Veglio, F. (2011). Inclusive business for sustainable livelihoods [Field Actions Science Reports]. The Journal of Field Actions, 5, 1–5.
Bruni, M., & Santucci, F. (2016). Agribusiness at global scale and smallholders. Bulgarian Journal of Agricultural Science, 22(1), 1–9.
Chamberlain, W., & Anseeuw, W. (2017). Contract farming as part of a multi-instrument inclusive business structure: A theoretical analysis. Agrekon, 56(2), 158–172. https://doi.org/10.1080/03031853.2017.1297725
Chevrollier, N., Bults, R., Sprenger, T., Danse, M., Poniatowski, B., & O’Neill, K. (2012). Access to Food and Improved Nutrition at the Base of the Pyramid: Five business interventions to achieve social impact, financial sustainability and scale. BOP Innovation Center and GAIN (Global Alliance for Improved Nutrition).
Coates, J., Swindale, A., & Bilinsky, P. (2015). Household Food Insecurity Access Scale (HFIAS) for Measurement of Food Access: Indicator Guide (version 3). Food and Nutrition Technical Assistance Project (FANTA), Academy for Educational Development.
Cotula, L., Vermeulen, S., Leonard, R., & Keeley, J. (2009). Land grab or development opportunity?: Agricultural investment and international land deals in Africa. IIED/FAO/IFAD.
Craig, D. A., & Porter, D. (2006). Development beyond neoliberalism?: Governance, poverty reduction and political economy. Routledge.
da Silva, C. A., & de Souza Filho, H. M. (2007). Guidelines for rapid appraisals of agrifood chain performance in developing countries. Food and Agriculture Organization of the United Nations (FAO).
Deininger, K. (2011). Challenges posed by the new wave of farmland investment. The Journal of Peasant Studies, 38(2), 217–247. https://doi.org/10.1080/03066150.2011.559007
Directorate General International Cooperation and Development (DGICD)—EuropeAid (2018). Green beans value chain analysis in Kenya.
European Commission. (2016). Aid for trade report 2016: Review of progress by the EU and it’s member states.
European Union. (2013). Agribusiness and development: How investment in the African agri-food sector can help support development.
FAO. (2015). Inclusive business models: Guidelines for improving linkages between producer groups and buyers of agricultural produce.
FAO. (2018a, May). GIEWS—global information and early warning system: Country briefs, Kenya. http://www.fao.org/giews/countrybrief/country.jsp?code=KEN
FAO. (2018b). Kenya at a glance. http://www.fao.org/kenya/fao-in-kenya/kenya-at-a-glance/en/
Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ). (2013). Inclusive business models: Options for support through PSD programmes.
Government of Kenya (GoK). (2007). Kenya vision 2030: A globally competitive and prosperous Kenya.
Government of Kenya (GoK). (2010). Agricultural sector development strategy 2010–2020.
Government of Kenya (GoK). (2011). National food and nutrition security policy.
Government of Kenya (GoK). (2012a). National Agribusiness Strategy: Making Kenya’s agribusiness sector a competitive driver of growth.
Government of Kenya (GoK). (2012b). National agricultural sector extension policy (NASEP).
Government of Kenya (GoK). (2012c). National horticulture policy.
Government of Kenya (GoK). (2017). The 2016 Short Rains Season Assessment Report.
Golja, T., & Požega, S. (2012). Inclusive business—what it is all about? Managing inclusive companies. International Review of Management and Marketing, 2(1), 22–42.
Gupta, J., Pouw, N. R., & Ros-Tonen, M. A. (2015). Towards an elaborated theory of inclusive development. The European Journal of Development Research, 27(4), 541–559.
Gupta, J., & Vegelin, C. (2016). Sustainable development goals and inclusive development. International Environmental Agreements: Politics, law and Economics, 16(3), 433–448.
Hakizimana, C., Goldsmith, P., Nunow, A. A., Roba, A. W., & Biashara, J. K. (2017). Land and agricultural commercialisation in Meru county, Kenya: Evidence from three models. The Journal of Peasant Studies, 44(3), 555–573.
IFC. (2012). Policy note on the business environment for inclusive business models. International Finance Corporation, World Bank Group.
Ion, A., Beyard, K., & Sedaca, S. (2014). Synthesis of trends in public-private partnerships (PPPs) for improving food security and rural development through agriculture report. CARANA Corporation.
Jenkins, B., Ishikawa, E., Geaneotes, A., Baptista, P., & Masuoka, T. (2011). Accelerating inclusive business opportunities: Business models that make a difference. International Finance Corporation, World Bank Group.
Kennedy, G., Ballard, T., & Dop, M. C. (2011). Guidelines for measuring household and individual dietary diversity. FAO.
KEPSA. (2017). Highlights of the big-four agenda of H.E. president Uhuru Kenyatta issued on 12th December, 2017. https://kepsa.or.ke/download/highlights-of-the-big-four-agenda-of-h-e-president-uhuru-kenyatta-issued-on-12th-december-2017/
Lundy, M., Amrein, A., Hurtado, J. J., Becx, G., Zamierowski, N., Rodriguez, F., & Mosquera, E. E. (2014). LINK methodology: A participatory guide to business models that link smallholders to markets. Version 2.0. International Center for Tropical Agriculture.
Mawdsley, E. (2017). Development geography 1: Cooperation, competition and convergence between ‘North’ and ‘South’. Progress in Human Geography, 41(1), 108–117. https://doi.org/10.1177/0309132515601776
Mbata, J. (2013). Agribusiness indicators: Kenya. World Bank.
Mohajan, H. (2014). Food and nutrition scenario of Kenya. American Journal of Food and Nutrition, 2(2), 28–38.
Murray, W. E., & Overton, J. (2016). Retroliberalism and the new aid regime of the 2010s. *Progress in Development Studies, 16*(3), 244–260.
Netherlands Enterprise Agency (NEA). (2018). Facility for sustainable entrepreneurship and food security (FDOV). https://english.rvo.nl/subsidies-programmes/facility-sustainable-entrepreneurship-and-food-security-fdov
OECD. (2011). The Busan partnership for effective development co-operation.
Ploumen, L. (2013). *A world to gain: A new agenda for aid, trade and investment*. Ministry of Foreign Affairs of the Netherlands.
Richey, L. A., & Ponte, S. (2014). New actors and alliances in development. *Third World Quarterly, 35*(1), 1–21. https://doi.org/10.1080/01436597.2014.868979
Research Solutions Africa (RSA). (2018). *Study of the mapping of distributors of fruits and vegetables in Kenya*.
Shortall, S. (2008). Are rural development programmes socially inclusive? Social inclusion, civic engagement, participation, and social capital: Exploring the differences. *Journal of Rural Studies, 24*(4), 450–457.
Sjauw-Koen-Fa, A. (2012). *Framework for an inclusive food strategy: Co-Operatives-A key for smallholder inclusion into value chains*. Rabobank Group.
Smith, W. (2013). How donors engage with business. Overseas Development Institute.
SNV Netherlands Development Organization Kenya. (2012). *The beans value chain in Kenya*.
Teklemariam, D., Azadi, H., Nyssen, J., Lanckriet, S., Asfaha, T. G., Haile, M., Taheri, F., & Witlox, F. (2017). Dynamics of the North–South capital flows or rise of South–South land deals? features of land acquisition in Ethiopia. *Land Degradation & Development, 28*(8), 2389–2407.
von Kimakowitz, E. (2012). The evolution of and current debate in development thinking: What defines good development? (Research Paper Series No. 13/04). Humanistic Management Network.
Wach, E. (2012). Measuring the ‘inclusivity’ of inclusive business. *IDS Practice Papers, 2012*(9), 1–30.
West, J. J., & Haug, R. (2017). The vulnerability and resilience of smallholder-inclusive agricultural investments in Tanzania. *Journal of Eastern African Studies, 11*(4), 670–691.
World Business Council for Sustainable Development (WBCSD). (2013). Scaling up inclusive business: Solutions to overcome internal barriers.
World Food Programme (WFP). (2016). *Kenya—comprehensive food security and vulnerability analysis (CFSVA)*.
Yin, R. K. (2011). Applications of case study research (3rd ed). SAGE Publications.
Zainal, Z. (2007). Case study as a research method. *Jurnal Kemanusiaan, 5*(1), 1–6.
Zoomers, A., Van Noorloos, F., Otsuki, K., Steel, G., & Van Westen, G. (2017). The rush for land in an urbanizing world: From land grabbing toward developing safe, resilient, and sustainable cities and landscapes. *World Development, 92*(C), 242–252. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2109,
1
],
[
2109,
5247,
2
],
[
5247,
8614,
3
],
[
8614,
11884,
4
],
[
11884,
15043,
5
],
[
15043,
18691,
6
],
[
18691,
22301,
7
],
[
22301,
25164,
8
],
[
25164,
28887,
9
],
[
28887,
31771,
10
],
[
31771,
33537,
11
],
[
33537,
36476,
12
],
[
36476,
38456,
13
],
[
38456,
42107,
14
],
[
42107,
45068,
15
],
[
45068,
48754,
16
],
[
48754,
52195,
17
],
[
52195,
55231,
18
],
[
55231,
58827,
19
],
[
58827,
61650,
20
]
]
} |
3e5305e8e93bda91058784ffd65e646f9c53c01f | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 9,
"total-fallback-pages": 0,
"total-input-tokens": 31242,
"total-output-tokens": 9712,
"fieldofstudy": [
"Agricultural And Food Sciences"
]
} | Article
Efficient Protocol for Improving the Development of Cryopreserved Embryonic Axes of Chestnut (*Castanea sativa* Mill.) by Encapsulation–Vitrification
Mariam Gaidamashvili ¹,* Eka Khurtsidze ¹, Tamari Kutchava ¹, Maurizio Lambardi ² and Carla Benelli ²
1 Department of Biology, Faculty of Exact and Natural Sciences, Iv. Javakhishvili Tbilisi State University, 1, Chavchavadze Ave., 0179 Tbilisi, Georgia; [email protected] (E.K.); [email protected] (T.K.)
2 Institute of BioEconomy, National Research Council (CNR/IBE), Sesto Fiorentino, 50019 Florence, Italy; [email protected] (M.L.); [email protected] (C.B.)
* Correspondence: [email protected]
Abstract: An optimized cryopreservation protocol for embryonic axes (EAs) of chestnut (*Castanea sativa* Mill.) has been developed based on the encapsulation–vitrification procedure. EAs of mature seeds were aseptically dissected and encapsulated in alginate beads with or without 0.3% (w/v) activated charcoal (AC). Embedded EAs were dehydrated with Plant Vitrification Solution 2 for different treatment times up to 120 min, followed by direct immersion in liquid nitrogen. Cryopreserved embryonic axes encapsulated with AC showed higher survival (70%) compared to those encapsulated without AC (50%). Sixty-four percent of embryonic axes, from synthetic seeds with AC, subsequently developed as whole plants. Plantlet regrowth was faster in AC-encapsulated EAs and showed enhanced postcryopreservation shoot and root regrowth over 2 cm after five weeks from rewarming. Results indicate that encapsulation–vitrification with activated charcoal added to the beads is an effective method for the long-term preservation of *Castanea sativa* embryonic axes.
Keywords: activated charcoal; alginate; cryopreservation; European chestnut; zygotic embryo
1. Introduction
European chestnut or sweet chestnut (*Castanea sativa* Mill.), belonging to the genus *Castanea*, is dominant in the mountainous forests of Western Georgia (150–1800 m), occupying the highest percentage of areas covered with forests (approx. 75%). Chestnut forests are developed in both West and East Georgia, but to the West of the country, they occupy larger areas. Chestnut trees generally extend from 100 m (Western Georgia) up to 900–1000 m a.s.l., reaching the absolute upper limit at 1400 m in sporadic locations of West and East Georgia [1,2]. According to the official International Union for Conservation of Nature (IUCN) list, *Castanea sativa* has been assessed as Least Concern [3,4]. Because of low self-renewal and pathogenic diseases, the large massifs of chestnut forests in Georgia are on the verge of destruction [5]. Therefore, sweet chestnut has been included in the Red List of Georgia under state Vulnerable (VU), according to the IUCN Red List Categories and Criteria [6,7]. The reason for including *Castanea sativa* in the Red List is the fragmentation and decreased distribution range. Hereafter, the development of efficient conservation measures is essential for both economic and wildlife protection commitments.
The ex situ conservation of chestnut in seed banks is limited due to nonresistance to storage at low-temperature conditions of partially dehydrated recalcitrant seeds [8,9]. Medium- (by in vitro slow-growth storage) and long-term preservation techniques (by cryopreservation in liquid nitrogen (LN)) are currently widely used for selected germplasm collections of various woody perennials [10–12]. In chestnut, Janeiro et al. reported the successful medium-term preservation of chestnut hybrid clones since the mid-1990s [13]. *Castanea* shoot cultures remained viable between 5 and 18 months of slow-growth storage.
at 4–8 °C [14–16]. Depending on storage conditions, up to 82% of explants survived and resumed normal growth [17]. As for cryopreservation, this method has been broadly used for the preservation of different biological materials of chestnut species and hybrid clones, such as shoot tips [18,19], zygotic embryonic axes [8,20,21] and embryogenic cultures [22–24], where desiccation and Plant Vitrification Solution 2 (PVS2)-based vitrification techniques [25] have been practiced on naked explants. On the other hand, since its proposal in the early 1990s [26,27], the use of dehydrated encapsulated explants (generally named synthetic seeds) has become a valid alternative for the cryopreservation of many plant species [28,29]. Following the “encapsulation–dehydration” method, a new variant, termed “encapsulation–vitrification,” was proposed [27,30]. This method combines the advantages of the vitrification and encapsulation of explants, greatly reducing the time required to apply a protocol in comparison to the “encapsulation–dehydration” method. So far, the “encapsulation–vitrification” technique has successfully been applied to the cryopreservation of the shoot tips of several fruit crops [31–33] and embryogenic cell suspensions of grapevine (Vitis spp.) [34].
The present study describes, for the first time, a protocol for the cryopreservation of the excised embryonic axes (EAs) of Castanea sativa L. by using the “encapsulation–vitrification” approach to evaluate its effectiveness in long-term conservation. Furthermore, the addition of activated charcoal (AC) as a component of the artificial matrix of synthetic seeds to diminish the polyphenol toxicity was tested with the aim to facilitate the optimization of the tested new cryoprocedure.
2. Results
Effect of LS and PVS2 Treatments on Survival and Regrowth of Encapsulated EAs
Treatment of alginate-coated EAs with only a loading solution (LS, containing 2 M glycerol and 0.4 M sucrose) for 60 min at 25 °C induced a small reduction of survival from 100% (control, nontreated and noncryopreserved) to 88.9% (time 0); however, treatment with LS positively influenced the survival rate of noncryostored (LN−) encapsulated EAs after treatment with PVS2. EA survival remained between 85.7% (30 and 60 min of treatment) and 81.3% (90 min), with no significant differences in percentage values. Only the survival of EAs treated for 120 min was reduced to 53.8% (Figure 1A). Similar findings were observed with the regrowth rates of noncryopreserved encapsulated EAs. EA regrowth remained in the range of 72.2% to 68.8% (30 and 90 min of treatment), and the regrowth of EAs treated for 120 min was reduced to 46.2% (120 min) (Figure 1B). However, only the loading treatment induced tolerance to ultrarapid cooling in LN, as survival and plantlet regrowth of encapsulated EAs passed from nil to almost 16.7% and 8.3%, respectively (Figure 1A,B, 0 treatment).
The PVS2 treatment duration significantly affected the survival of cryopreserved (LN+) encapsulated EAs. The 30 min treatment with PVS2 induced the highest (50%) survival of EAs. A further increase in PVS2 treatment time to 120 min resulted in a significant decline of cryopreserved EA survival up to a minimum of 13.3% (Figure 1A). Referring to the regrowth of plantlets, derived from “germinated” synthetic seeds, noncryopreserved encapsulated EAs exhibited a decrease in regrowth, ranging from 91.6% in control plants to 46.2% after treatment with PVS2 at 120 min (Figure 1B). Plantlet regrowth rates were significantly reduced after cryogenic storage at –196 °C. The 30 min PVS2 treatment was the most effective in inducing tolerance to ultrarapid cooling in LN, resulting in 50% plantlet regrowth in postcryopreservation (Figure 1B). A further increase in PVS2 treatment time yielded a significant reduction of cryopreserved EA regrowth up to a minimum of 13.3% (Figure 1B).
Survival values were significantly different when EAs were encapsulated in alginate beads containing 0.3% activated charcoal (AC) in the artificial matrix. The best results were achieved after the 30 min treatment with PVS2 with a survival rate of 70% in cryopreserved EAs, significantly higher than in non-AC beads (50%; Figure 2A).
Plants 2021, 10, x FOR PEER REVIEW 3 of 9
Figure 1. Percentages of survival (A) and plantlet regrowth (B) of encapsulated Castanea sativa embryonic axes (EAs) after exposure to Plant Vitrification Solution 2 (PVS2) for increasing times, with (LN+) or without (LN−) subsequent immersion in liquid nitrogen. EAs were encapsulated in WPM medium containing 2.5% (w/v) sodium alginate. Encapsulated EAs were treated for 60 min with a loading solution (2.0 M glycerol, 0.4 M sucrose), followed by treatment with PVS2 at 0 °C for 30–120 min, prior to direct immersion in LN for 1 h. Control EAs received no LS and PVS2 treatments. Within each line (LN− and LN+), data followed by different letters are significantly different at p ≤ 0.05 by LSD test (bars, SE of means).
Figure 2. Effect of activated charcoal on the survival (A) and regrowth (B) of cryopreserved (LN+) chestnut embryonic axes by encapsulation–vitrification. EAs were encapsulated in WPM solution containing 2.5% (w/v) sodium alginate with (AC+) or without (AC−) 0.3% (w/v) activated charcoal in the artificial matrix. Encapsulated EAs were treated for 60 min with LS (2.0 M glycerol, 0.4 M sucrose), followed by treatment with PVS2 at 0 °C for 30–120 min, prior to direct immersion in liquid nitrogen for 1 h. Control EAs received no LS and PVS2 treatments. Within each exposure time, different letters indicate significant differences between AC+ and AC− at p ≤ 0.05 by chi-squared test (bars, SE of means).
The presence of AC significantly influenced regrowth rates in EAs. The regrowth of AC-encapsulated EAs after treatment with PVS2 for increasing times was in the range of 21.4% and 20.5% for 60 and 90 min of treatment, respectively, whereas it was 16.7% for 120 min of treatment, with no significant differences in percentage values (Figure 2B). In comparison, treatment with PVS2 for 30 min yielded 64% regrowth, showing to be the best recovery rate of AC-added cryopreserved encapsulated EAs, significantly higher than EAs encapsulated without AC (50%, Figure 2B).
The shoot and root length data summarized in Table 1 also show the effect of AC on the plantlet regrowth of encapsulated EAs subjected to various vitrification times with PVS2, five weeks after cryostorage, rewarming and plating.
All surviving cryostored EAs produced roots and shoots, and their development was clearly pronounced in both (AC−) and (AC+) synthetic seeds with the 30 min PVS2 treatment time (Table 1). Moreover, it is noteworthy that with AC added to the synthetic seed, an appreciable shoot and root length was highlighted during postcryopreservation
(Figure 3D), with 14.5 and 22.8 cm, respectively, after five weeks, whereas without AC, the cryopreserved plantlets had 9.2 mm and 10.2 mm of the shoot and root length at the same period. The germination of noncryopreserved synthetic seeds started after one week of culture, and it was delayed up to four weeks after cryogenic storage in LN. However, the regrowth initiation time was shorter in encapsulated explants containing AC, i.e., 20 days in postcryopreservation (Figure 3C). After eight weeks of culture, all (AC+) derived plantlets showed well-developed roots and shoots that allowed their transfer in greenhouse conditions (Figure 3D–F).
### Table 1. Effect of activated charcoal (AC) on the plantlet regrowth of encapsulated *Castanea sativa* EAs subjected to various dehydration times with PVS2 following immersion in LN evaluated 5 weeks after cryostorage, rewarming and plating.
| PVS2 (min) | Regrowth * (AC−) | Regrowth * (AC+) |
|------------|------------------|------------------|
| | Shoot Length (mm) | Root Length (mm) | Shoot Length (mm) | Root Length (mm) |
| 0 | 4.0 b | 5.2 b | 5.0 c | 6.3 c |
| 30 | 9.2 a | 10.2 a | 14.5 a | 22.8 a |
| 60 | 6.5 b | 8.8 a | 8.7 b | 9.8 b |
| 90 | 5.9 b | 8.9 a | 8.3 b | 10.4 b |
| 120 | 4.9 b | 7.8 a | 7.9 b | 9.5 b |
*a* Mean of 90 plantlets tested. Data were recorded after 5 weeks of culture following cryopreservation. Statistical analysis in each column was performed by ANOVA. Data followed by different letters are significantly different at $p \leq 0.05$ by LSD test.
Figure 3. Plant regeneration from cryopreserved embryonic axes of *Castanea sativa* by encapsulation–vitrification. (A) Excised embryonic axes (EAs) used for cryopreservation. (B) Encapsulated EAs in 2.5% sodium alginate with (AC+) or without (AC−) 0.3% (w/v) activated charcoal in the artificial matrix. (C) Survived encapsulated (AC+) EAs after cryopreservation and 20 days of postculture. (D) Primary plantlet development 2 weeks after survival assessment cryopreserved by encapsulation–vitrification procedure. (E) Elongated root and shoot 8 weeks of postculture after cryopreservation. (F) Plantlets established under greenhouse conditions 4 weeks after transfer to soil.
3. Discussion
Synthetic seed technology, in addition to fulfilling needs related to micropropagation [35–39], can prove to be an efficient tool for the storage of rare and commercially important species at low temperatures. It has the potential for the medium-term and long-term preservation of plant explants encapsulated in synthetic seeds, without losing viability after immersion in LN when cryopreservation is applied [26–28,40,41].
The “encapsulation–vitrification” cryoprocessure [27] has been used for the cryopreservation of the shoot tips of several woody fruit crops [28,31–33] and embryogenic cell suspensions [34]. Although it requires a long treatment time compared to the vitrification of naked explants, the encapsulation of plant germplasm makes for less damage to samples during the vitrification procedures [42,43]. In our encapsulation–vitrification experiment with chestnut EAs, after treatment with LS for 60 min, the 30 min exposure time of PVS2 showed the best regrowth rate (50%). Optimizing the time of exposure to PVS2 is most important for producing a satisfactory level of regrowth after cryopreservation, and the PVS2 osmoprotection effect can change among different species [28,44]. For example, the duration of PVS2 treatment was up to 200 min in the encapsulated shoot tips of *Dianthus caryophyllus* L. [45].
In the following experiment, the addition of AC into synthetic seeds treated with the same conditions positively affected plantlet initiation and regrowth from chestnut EAs, with the concentration amended with 0.3% (w/v) AC. In a previous research, AC was added in a culture medium to overcome the onset of browning, shortly after the excision of EAs, and promoted their germination [21]. In another study, AC added in the artificial endosperm of synthetic seeds containing somatic embryos of hybrid rice improved their germination and conversion to plantlets [46]. Furthermore, the germination and root development of encapsulated somatic embryos of *Picea glauca* and *Picea mariana* enhanced with the addition of 0.05 gL\(^{-1}\) AC to the beads [47]. Therefore, as also shown in this study, AC represents a component that can improve the development of explants even after their ultrarapid cooling in LN. Indeed, the survival and regrowth rates of cryopreserved encapsulated EAs were markedly increased when AC was included in the bead composition.
It is also noteworthy that the results obtained here showed an improvement in the survival and regrowth of cryostored chestnut EAs by 15% and 10%, respectively, in comparison with a previous study concerning the vitrification procedure of naked EAs [21]. Although the survival and regrowth rates of encapsulated (AC−) EAs were lower than the same parameters obtained in a previous study by the “desiccation–one-step cooling” protocol (70% and 64%, respectively) [21], it should be noted that the overall ratio between embryo survival and plantlet regrowth appreciably improved with the presence of AC in synthetic seeds. Thus, the survival/regrowth ratio of AC-encapsulated EAs after cryopreservation was 91.4% versus 83% obtained by the “desiccation–one-step freezing protocol,” at the best treatment times. Corredoira et al., [20] reported a 63.3% recovery applied to *Cassiopea sativa* zygotic embryos by the desiccation procedure, which was still lower than the regrowth percentage obtained in our experiment with (AC+) encapsulation–vitrification.
The conversion of synthetic seeds into plants after germination is a fundamental aspect of the success of the encapsulation–vitrification technique. In this study, the development of cryopreserved encapsulated EAs with (AC+), after treatment with 30 min PVS2, was faster by 6–7 days with respect to EAs encapsulated without AC. Furthermore, the root and shoot length of (AC+) EAs achieved 22.8 mm and 14.5 mm, respectively, five weeks after rewarming and plating, whereas the (AC−) EAs showed less development at the same period (Table 1). Evident differences were also found in the postcryopreservation initiation times of plantlet formation with respect to previous cryopreservation procedures applied on EAs. Indeed, the plantlet development of cryopreserved EAs synthetic seeds with (AC+) started two weeks earlier than naked vitrified EAs, where the full germination of EAs (expressed as plantlet regrowth) required eight weeks and eight days earlier than desiccated by dehydration–“one-step freezing” EAs [21]. Notably, even root and shoot...
Plants 2021, 10, 231
6 of 9
elongation from encapsulated (AC+) EAs was considerably faster, exceeding 2 cm root and 1.5 cm shoot length in five weeks.
AC seems to play a role to keep nutrients within the artificial matrix, releasing them slowly during the development of embryos. The absorption of detrimental polyphenolic exudates released by encapsulated explants is also facilitated by AC [48].
4. Materials and Methods
4.1. Plant Material
Chestnut fruits were collected in Western Georgia at the beginning of October 2019 from the open-pollinated trees of Castanea sativa. Mature fruits were stored at 4 °C for a maximum of 1 month until use in the cryopreservation trials. For the cryopreservation experiments, the fruits were washed in 2% (v/v) household detergent and rinsed three times under tap water. Then, the pericarp, seed coat and part of the kernel were removed. The remaining embryo axes along with the part of the kernel were surface-sterilized by successive immersion in 70% (v/v) ethanol with a few drops of Tween 20 for 2 min, followed by decontamination with a 10% (v/w) solution of sodium hypochlorite (Sigma-Aldrich®, Saint Louis, MO, USA) for 20 min. After being rinsed in sterile distilled water three times, EAs, composed of the zygotic embryos along with 2–3 mm long cotyledon residuals, were dissected from the seeds (Figure 3A).
4.2. Encapsulation
Dissected EAs were immersed in a calcium-free liquid woody plant medium (WPM) [49] without plant growth regulators, supplemented with 2.5% (w/v) sodium alginate (Bioworld®, Dublin, OH, USA) and 0.3 M sucrose. The mixture (including dissected EAs) was dropped with a sterile pipette into WPM liquid medium containing 100 mM calcium chloride, forming beads about 4–5 mm in diameter (Figure 3B). The drops with EAs were maintained in the solution for 20 min to achieve polymerization.
In one specific experiment, 0.3% (w/v) activated charcoal (AC; Sigma-Aldrich, DARCO®, Saint Louis, MO, USA) was added to the sodium alginate solution to assess its influence on the survival and regrowth rate of encapsulated EAs after PVS2 treatment and subsequent cooling. All operations were performed under sterile conditions. After the incubation period in the complexion agent, the encapsulated explants were rinsed three times in sterile distilled water.
4.3. Encapsulation–Vitrification Technique for EA Cryopreservation
Encapsulated EAs were transferred to LS containing 2.0 M glycerol and 0.4 M sucrose for 60 min at 25 °C, followed by treatment with PVS2 [25] (30% w/v glycerol, 15%, w/v DMSO, 15% w/v ethylene glycol in WPM medium containing 0.4 M sucrose) for 0, 30, 60, 90, 120 min treatment times at 0 °C. Then, synthetic seeds were placed in 2 mL cryovials (5 in each) and immersed in LN for 24 h (LN+). For rewarming, the cryovials were rapidly immersed in a water bath at 40 °C for 2 min. Encapsulated EAs were rinsed in a washing solution containing the WPM liquid medium and 1.2 M sucrose (two times of 10 min each, at 25 °C), and then LN+ and LN− (synthetic seeds without cooling) samples were cultured in test tubes (20 mm × 150 mm) in WPM supplemented with 30 g L−1 sucrose and 0.4 µM 6-benzylaminopurine (BAP). The medium was solidified with 6 g L−1 agar (PlantMedia™, Dublin, OH, USA) and adjusted to pH 5.7 before autoclaving. Cultured tubes were maintained in a growth chamber at 24 ± 0.5 °C under a 16/8 h light/dark regime with an irradiance of 40 µmol m−2 s−1 in cool-white fluorescent light. After 2 and 4 weeks, survival and regrowth were recorded, respectively. After 8 weeks of in vitro culture, plantlets (i.e., “germinated” EAs) were washed thoroughly in running tap water; the root length was measured and transferred to plastic cups filled with a mixture of 100% sphagnum peat/perlite at a ratio of 2:1. The plantlets were relocated for acclimatization in controlled chambers at 23 ± 1 °C under 60 ± 5% moisture content and 16 h photoperiod with an irradiance of 40 µmol m−2 s−1 in cool-white fluorescent light over the following three weeks. After the
emergence of new leaves, the plants were transplanted in bigger pots containing peat, soil and perlite at a ratio of 1:2:1 and transferred to natural greenhouse conditions.
4.4. Data Collection and Statistical Analysis
The total number of embryos used for the experiment was 240. Each treatment, with (AC+) or without (AC−)-activated charcoal, included 50 noncryopreserved (LN−) encapsulated EAs (10 EAs for each condition from 0 to 120 min treatment time) and 50 cryopreserved (LN+) encapsulated EAs (10 EAs for each condition). Control EAs (20 for each, AC− and AC+ treatments), receiving no LS, PVS2 or LN treatment, were also included. For a 0 h-min PVS2 treatment time, encapsulated EAs were only loaded in LS solution and cryopreserved without PVS2 or directly cultured in test tubes for synthetic seed “germination.” Each treatment consisted of 3 replicates, and all experiments were repeated 3 times.
Survival was recorded after two weeks of culture and defined as the percentage of the total number of encapsulated EAs, which showed initial normal germination and development (i.e., root and shoot emission) or only root development. The regrowth of encapsulated EAs was assessed after four weeks of culture. Plant regrowth rate was estimated as a percentage of whole plantlets (retaining normal shoots and roots ≥5 mm in length) developing from encapsulated EAs relative to the total number of synthetic seeds cultured after cryopreservation. Root and shoot length were recorded weekly. Statistical analysis of percentages was performed by ANOVA (when comparing multiple treatments), followed by the LSD test at \( p \leq 0.05 \) for mean separation or chi-squared test (when comparing pairs of treatments). Percentage data used in ANOVA were subjected to arcsine transformation prior to analysis. The bars in the figures represent standard errors (SE) of means.
5. Conclusions
The present study has clearly demonstrated the feasibility of the long-term preservation of \textit{Castanea sativa} germplasm by the encapsulation–vitrification of EAs. The acquisition of suitable dehydration tolerance with PVS2 to survive after the cryopreservation of EA synthetic seeds and their germination and regrowth under optimized conditions (AC+) promoted growth by shortening the development times and limiting the loss of explants; therefore, the overall performance of the cryopreserved EAs appears to be improved in comparison with previous studies. The protocol described in this study will now be tested on a wide range of chestnut cultivars and hybrid clones to achieve the practical long-term cryopreservation of \textit{Castanea} genus germplasm.
**Author Contributions:** Conceptualization, M.G.; methodology, M.G., C.B.; software, E.K.; validation, M.G.; formal analysis, M.G., E.K. and T.K.; investigation, E.K. and T.K.; resources, M.G.; data curation, M.G. and E.K.; writing—original draft preparation, M.G.; writing—review and editing, M.G., C.B., M.L.; visualization, M.G. and E.K., supervision, M.L.; project administration, M.G.; funding acquisition, M.G. All authors have read and agreed to the published version of the manuscript.
**Funding:** This work was supported by Shota Rustaveli National Science Foundation of Georgia (SRNSFG), grant number FR17-444. The APC was funded by Shota Rustaveli National Science Foundation of Georgia (SRNSFG).
**Institutional Review Board Statement:** Not applicable.
**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.
**Acknowledgments:** Shota Rustaveli National Science Foundation of Georgia (SRNSFG) and National Council Research of Italy (Bilateral Project: Developing efficient cryopreservation procedures for the long-term storage of endangered plant genetic resources of Georgia) are acknowledged for financial support.
**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
References
1. Dolukhanov, A. Rastitel’ nost’ Gruzii (Vegetation of Georgia); Metsniereba: Tbilisi, Georgia, 1989; Volume 1. (In Russian)
2. Nakhtursrishvili, G. Forest vegetation of Georgia. In the Vegetation of Georgia (South Caucasus); Nakhtursrishvili, G., Ed.; Springer: Berlin/Heidelberg, Germany, 2013; pp. 35–87.
3. IUCN Red List of Threatened Species. Version 2020-3. Available online: https://www.iucnredlist.org (accessed on 18 December 2020).
4. Barstow, M.; Khela, S. Castanea Sativa. The IUCN Red List of Threatened Species. 2018. Available online: https://dx.doi.org/10.2305/IUCN.UK.2018-1.RLTS.T202948A67740523.en (accessed on 18 December 2020).
5. Tavadze, B.; Supatashvili, A.; Kapanadze, G.; Mamukashvili, T. Pathological status of chestnut stands in Tkibuli region (Georgia). Ann. For. 2012, 5, 21–32.
6. Red List of Georgia. Edict of the President of Georgia #303 on Approval of the Red List of Georgia. Tbilisi. 2006. Available online: https://www.matsne.gov.ge/ka/document/view/97288?publication=0 (accessed on 18 December 2020).
7. IUCN Red List Categories and Criteria 2012 Version 3.1, 2nd ed. Available online: https://www.iucn.org/content/iucn-red-list-categories-and-criteria-version-31-second-edition (accessed on 18 December 2020).
8. Pence, V.C. Desiccation and survival of Aesculus, Castanea and Quercus embryo axis through cryopreservation. Cryobiology 1992, 29, 391–399. [CrossRef]
9. Westengen, O.T.; Jeppson, S.; Guarino, L. Global Ex-Situ Crop Diversity Conservation and the Svalbard Global Seed Vault: Assessing the Current Status. PLoS ONE 2013, 8, e64146. [CrossRef] [PubMed]
10. Lambardi, M.; De Carlo, A. Application of tissue culture to germplasm conservation of temperate broad-leaf trees. In Micropropagation of Woody Trees and Fruits; Jain, S.M., Ishii, K., Eds.; Kluwer Academic: Dordrecht, The Netherlands, 2003; pp. 815–840.
11. Panis, B.; Lambardi, M. Status of cryopreservation technologies in plants (crops and forest trees). In The Role of Biotechnology in Exploring and Protecting Agricultural Genetic Resources; Ruane, J., Sonnino, A., Eds.; FAO: Rome, Italy, 2006; pp. 61–78.
12. Panis, B. Sixty years of plant cryopreservation: From freezing hardy mulberry twigs to establishing reference crop collections for future generations. Acta Hortic. 2019, 1234, 1–8. [CrossRef]
13. Janeiro, L.V.; Vieitez, A.M.; Ballester, A. Cold storage of the in vitro cultures of wild cherry, chestnut and oak. Ann. Sci. For. 1995, 52, 287–293. [CrossRef]
14. Lambaro, M.; Benelli, C.; De Paoli, G.; Battistini, A. Biotecloniche per la conservazione del Castagno. In Proceedings of the Convegno Nazionale Castagno, Firenze, Italy, 25–27 October 2001; Bellini, E., Ed.; Università di Firenze: Marradi (Firenze), Italy, 2001; pp. 86–91.
15. Corredoira, E.; Valladares, S.; Martinez, M.T.; Couselo, J.L.; San Jose, M.C.; Ballester, A.; Vieitez, A.M. Conservación de germoplasma en especies leñosas con técnicas de cultivo in vitro y almacenamiento en frío (Span). J. Rural Dev. 2011, 2, 15–24.
16. Corredoira, E.; Martinez, M.T.; Cernadas, M.J.; San Jose, M.C. Application of biotechnology in the conservation of the genus Castanea. Forests 2017, 8, 394. [CrossRef]
17. Capuana, M.; Di Lonardo, S. In vitro conservation of chestnut (Castanea sativa) by slow growth. In Vitro Cell. Dev. Biol. Plant 2013, 49, 605–610. [CrossRef]
18. Vidal, N.; Sanchez, C.; Jorquera, L.; Ballester, A.; Vieitez, A.M. Cryopreservation of chestnut by vitrification of in vitro-grown shoot tips. In Vitro Cell. Dev. Biol.-Plant 2005, 41, 63–68. [CrossRef]
19. Jorquera, L.; Vidal, N.; Sanchez, C.; Vieitez, A.M. Optimizing conditions for successful plant regeneration from cryopreserved Castanea sativa shoot tips. Acta Hortic. 2005, 693, 511–518. [CrossRef]
20. Corredoira, E.; San Jose, M.C.; Ballester, A.; Vieitez, A.M. Cryopreservation of zygotic embryo axes and somatic embryos of European chestnut. CryoLetters 2004, 25, 33–42. [PubMed]
21. Gaidamashvili, M.; Khurtsidze, E.; Benelli, K.; Lambardi, M. Development of an Efficient ‘One-Step Freezing’ Cryopreservation Protocol for a Georgian Provenance of Chestnut (Castanea sativa Mill.) Zygotic Embryos. Not. Bot. Hortic. Agrobo. 2019, 47, 1047–1054. [CrossRef]
22. Holliday, C.; Merkle, S.A. Preservation of American chestnut germplasm by cryostorage of embryogenic cultures. J. Am. Chestnut Found. 2000, 14, 46–52.
23. San Jose, M.C.; Jorquera, L.; Vidal, N.; Corredoira, E.; Sanchez, C. Cryopreservation of European chestnut germplasm. Acta Hortic. 2005, 693, 225–232. [CrossRef]
24. Vieitez, A.M.; San Jose, M.C.; Corredoira, E. Cryopreservation of zygotic embryonic axes and somatic embryos of European chestnut. In Plant Embryo Cultures: Methods and Protocols, Methods in Molecular Biology; Thorpe, T.A., Yeung, E.C., Eds.; Springer: New York, NY, USA, 2011; Volume 710, pp. 201–213.
25. Sakai, A.; Kobayashi, S.; Oiyama, I. Cryopreservation of nucellar cells of navel orange (Citrus sinensis Osb. var. brasiliensis Tanaka) by vitrification. Plant Cell Rep. 1990, 9, 30–33. [CrossRef] [PubMed]
26. Fabre, J.; Dereuddre, J. Encapsulation-dehydration: A new approach to cryopreservation of Solanum shoot tips. CryoLetters 1990, 11, 413–426.
27. Matsumoto, T.; Sakai, A.; Takahashi, C.; Yamada, K. Cryopreservation of in-vitro grown apical meristems of wasabi (Wasabia japonica) by encapsulation-vitrification method. CryoLetters 1995, 16, 189–196.
28. Sakai, A.; Engelmann, F. Vitrification, encapsulation-vitrification and droplet-vitrification: A review. CryoLetters 2007, 28, 151–172.
29. Kulus, D. Application of synthetic seeds in propagation, storage, and preservation of Asteraceae plant species. In Synthetic Seeds Germplasm Regeneration, Preservation and Prospects; Faisal, A., Alatar, A., Eds.; Springer: Berlin, Germany, 2019; pp. 155–179.
30. Sakai, A.; Hirai, D.; Niino, T. Development of PVS-Based Vitrification and Encapsulation–Vitrification Protocols. In Plant Cryopreservation: A Practical Guide; Reed, B.M., Ed.; Springer: New York, NY, USA, 2008; pp. 33–57.
31. Niino, T.; Sakai, A. Cryopreservation of alginate-coated in vitro grown shoot tips of apple, pear and mulberry. Plant Sci. 1992, 87, 199–206. [CrossRef]
32. Paul, H.; Daigny, G.; Sangwan-Norreel, B.S. Cryopreservation of apple (Malus domestica Borkh.) shoot tips following encapsulation-dehydration or encapsulation-vitrification. Plant Cell Rep. 2000, 19, 768–774. [CrossRef]
33. Wang, Q.; Laamanen, J.; Uosukainen, M.; Valkonen, J.P.T. Cryopreservation of in vitro-grown shoot tips of raspberry (Rubus idaeus L.) by encapsulation–vitrification, and encapsulation–dehydration. Plant Cell Rep. 2005, 24, 280–288. [CrossRef] [PubMed]
34. Wang, Q.C.; Mawassi, M.; Sahar, N.; Li, P.; Violeta, C.-T.; Gafny, R.; Sela, I.; Tanne, E.; Pearl, A. Cryopreservation of grapevine (Vitis spp.) embryogenic cell suspensions by encapsulation–vitrification. Plant Cell Tissue Organ Cult. 2004, 77, 267–275. [CrossRef]
35. Lambardi, M.; Halmagyi, A.; Deliu, C. Cryopreservation of carnation (Dianthus caryophyllus L.) shoot tips by encapsulation-vitrification. Sci. Hortic. 2007, 113, 300–306. [CrossRef]
36. Kumar, M.B.A.; Vakeswaran, V.; Krishnasamy, V. Enhancement of synthetic seed conversion to seedlings in hybrid rice. Plant Cell Tissue Organ Cult. 2005, 81, 97–100. [CrossRef]
37. George, E.F.; Sherrington, P.D. Plant Propagation by Tissue Culture—Handbook and Directory of Commercial Laboratories; Exegetics Ltd.: Eversley, UK, 1984. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3737,
1
],
[
3737,
7982,
2
],
[
7982,
10591,
3
],
[
10591,
13092,
4
],
[
13092,
17595,
5
],
[
17595,
21654,
6
],
[
21654,
25757,
7
],
[
25757,
31321,
8
],
[
31321,
33250,
9
]
]
} |
710a278e233da7fd0f66a52a72088e9524f0faa6 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 5,
"total-fallback-pages": 0,
"total-input-tokens": 18166,
"total-output-tokens": 6282,
"fieldofstudy": [
"Medicine"
]
} | Surgical Decision Making for the Elderly Patients in Severe Head Injuries
Kyeong-Seok Lee, M.D., Jae-Jun Shim, M.D., Seok-Man Yoon, M.D., Jae-Sang Oh, M.D., Hack-Gun Bae, M.D., Jae-Won Doh, M.D.
Department of Neurosurgery, Soonchunhyang University Cheonan Hospital, Cheonan, Korea
Objective: Age is a strong predictor of mortality in traumatic brain injuries. A surgical decision making is difficult especially for the elderly patients with severe head injuries. We studied so-called ‘withholding a life-saving surgery’ over a two year period at a university hospital.
Methods: We collected data from 227 elderly patients. In 35 patients with Glasgow Coma Score 3–8, 28 patients had lesions that required operation. A life-saving surgery was withheld in 15 patients either by doctors and/or the families (Group A). Surgery was performed in 13 patients (Group B). We retrospectively examined the medical records and radiological findings of these 28 patients. We calculated the predicted probability of 6 month mortality (IPM) and 6 month unfavorable outcome (IPU) to compare the result of decision by the International Mission for Prognosis and Analysis of Clinical Trials in TBI (IMPACT) calculator.
Results: Types of the mass lesion did not affect on the surgical decision making. None of the motor score 1 underwent surgery, while all patients with reactive pupils underwent surgery. Causes of injury or episodes of hypoxia/hypotension might have affected on the decision making, however, their role was not distinct. All patients in the group A died. In the group B, the outcome was unfavorable in 11 of 13 patients. Patients with high IPM or IPU were more common in group A than group B. Wrong decisions brought futile cares.
Conclusion: Ethical training and developing decision-making skills are necessary including shared decision making.
Key Words: Prognosis · Decision making · Patient participation · Craniocerebral trauma.
INTRODUCTION
In traumatic brain injuries, age is one of the strongest predictors of mortality and functional outcome\(^\text{[14,22]}\). Older age is associated with poorer outcome\(^\text{[1,11,16,24]}\). 100% mortality rate in elderly patients aged >80 years, with a Glasgow Coma Scale (GCS) of 11 or less, and other studies have reported that surgery on elderly patients with a GCS of 5 or less was incompatible with survival\(^\text{[19]}\). However, an older age is continuously associated with a worsening outcome after traumatic brain injuries\(^\text{[31]}\). The decision to offer life-saving, but not restorative, surgery to patients with severe neurotrauma is difficult\(^\text{[10]}\). Balancing the risk of not doing enough against the risk of doing too much with the result of poor neurological recovery is emotionally challenging, both for health care providers and the families of patients\(^\text{[10]}\).
Decision making is a key competency of surgeons; however, how best to assess decisions and decision makers is not clearly established\(^\text{[41]}\). It is hard to learn or study the good decision making. We studied the frequency and the basis of so-called ‘withholding a life-saving surgery’ over a two year period at a university hospital, in order to analyze how such situations are handled on a daily basis.
MATERIALS AND METHODS
Patient recruitment started in January 2011 and ended in December 2012. During this period, 577 patients were admitted after head injuries. We could collect 227 elderly (over 60 years) patients (Table 1). The level of consciousness was measured by the GCS score on admission. In 35 patients with GCS 3–8, 28 patients had surgical lesion. A life-saving surgery was withheld in 15 patients either by doctors and/or the families of patients (Group A). Surgery was performed in 13 patients (Group B).
We retrospectively examined the medical records and radiological findings of these 28 patients. We obtained the predicted probability of 6 month mortality (IPM) and 6 month unfavorable outcome (IPU) to compare the result of decision by the International Mission for Prognosis and Analysis of Clinical Trials in TBI (IMPACT) calculator.
probability of 6 month mortality (IPM) and 6 month unfavorable outcome (IPU) to compare the result of decision making by the International Mission for Prognosis and Analysis of Clinical Trials in TBI (IMPACT) calculator. IPM and IPU were automatically calculated by the IMPACT calculator at the internet homepage (http://www.tbi-impact.org/?p=impact/calc). The predicted risk of mortality or unfavorable outcome was not used during the patient management. The outcome was measured by the Glasgow outcome scale at the time of discharge.
Statistical analysis was performed using the chi-square test or Fisher’s exact test. For the statistical significance, we divided the etiology into either known or unknown groups. The outcome was also divided into either favorable (good recovery and moderate disability) or unfavorable (from severe disability to death). Differences were considered significant if the probability value was less than 0.05.
Table 1. Treatment of elderly patients with head injuries
| Treatment | GCS 3–8 | GCS 9–12 | GCS 13–15 | Total |
|-----------|---------|----------|-----------|-------|
| Conservative | 22 | 13 | 139 | 174 |
| Burr hole | 3 | 2 | 23 | 28 |
| Craniotomy | 10 | 7 | 8 | 25 |
| Total | 35 | 22 | 170 | 227 |
GCS: Glasgow Coma Scale
Table 2. Clinical characteristics of elderly patients with severe head injuries
| CT | Group A | Group B | Total |
|-------------|---------|---------|-------|
| SDHa | 14 | 9 | 23 |
| tICH | 1 | 2 | 3 |
| SDHc | 0 | 2 | 2 |
| Motor Score | | | |
| 1–3 | 11 | 5 | 16 |
| 4–6 | 4 | 8 | 12 |
| Pupil | | | |
| Both dilated| 13 | 4 | 17 |
| Reactive 1 or 2 | 2 | 9 | 11 |
| Hypoxia/hypotension | yes | 6 | 4 | 10 |
| No | | | |
| Causes | | | |
| Unknown | 3 | 4 | 7 |
| Slip/fall | 9 | 5 | 14 |
| Other | 3 | 4 | 7 |
| GOS | | | |
| D | 15 | 3 | 18 |
| VS/SD | 0 | 8 | 8 |
| MD/GR | 0 | 2 | 2 |
| Total | 15 | 13 | 28 |
Table 3. The predicted probability of 6 month mortality by the IMPACT calculator
| Group A | Group B | Total |
|---------|---------|-------|
| IPMc (%) | | |
| >80 | 11 | 1 | 12 |
| ≤80 | 4 | 12 | 16 |
| IPMt (%) | | |
| >80 | 9 | 3 | 12 |
| ≤80 | 6 | 10 | 16 |
| IPUc | | |
| >80 | 14 | 4 | 18 |
| ≤80 | 1 | 9 | 10 |
| IPUt | | |
| >80 | 12 | 5 | 17 |
| ≤80 | 3 | 8 | 11 |
| Total | 15 | 13 | 28 |
The most common surgical lesion was acute subdural hematoma in both groups (Table 2). Types of the surgical lesion did not affect the surgical decision making. None of the motor score 1 underwent surgery. Although low (1–3) motor score was more common in the group A, this difference could not reach to the statistical significance (p > 0.05, by Fisher test). Pupil response was a major factor of decision making. All patients with reactive pupils underwent surgery. In 18 patients with both dilated pupils, only five of them underwent surgery. The outcome was unfavorable in all (1 death, 3 vegetative states, and 1 severe disability). Causes of injury or episodes of hypoxia/hypotension might affect on the decision making, however, their role was not distinct.
All patients in the group A were expired within 3 days, except one (case 4). In the group B, 10 patients could survive, however the outcome was unfavorable in 8 patients.
The IPM ranged from 54% to 90% in the group A (mean 82%); from 29% to 91% in the group B (mean 54.2%) by core model (IPMc). The IPU ranged from 73% to 97% in the group A (mean 91.5%); from 44% to 98% in the group B (mean 71.0%) by core model (IPUc). Patients with high IPM or IPU were more common in group A than group B, either by core model or core+CT model (Table 3). These differences were statistically significant (p<0.05, by Fisher's test).
We tried to illustrate 6 cases; four patients withheld surgery despite of relatively low probability of mortality or young age, two patients underwent surgery even though the risk of mortality was high.
Case 1
This 81-year-old female patient visited our emergency room.
(ER) with comatose mentality, probably after slip down. She suffered from hypertension and diabetes. She has been taking warfarin after history of stroke. She had a fracture on her left humerus after a slip due to hypoglycemia 10 days before admission. Her GCS score was 7 (E1, V1, M5). Her pupils were reactive on the right side initially, however became bilaterally non-reactive, soon. CT scan revealed an acute subdural hematoma (Fig. 1A). Her family did not want to do surgery due to her age and frequent illness. When we used the best motor score, the IPMc, IPUC, IPMt (core+CT model), and IPUt were 54, 73, 45, and 65 respectively. She died on the next day.
Case 2
This 80-year-old male patient was discovered on the road with comatose mentality. His GCS score was 7 (E1, V1, M5). His pupils were non-reactive on both sides. His vital signs were normal. There were no evidence of coagulopathy in the laboratory tests. CT scan revealed an acute traumatic intracerebral hematoma (Fig. 1B). His family did not want surgery. The IPMc, IPUC, IPUC, and IPUt were 70, 83, 60, and 76, respectively. He died on the next day.
Case 3
This 61-year-old male patient was transferred to our hospital with comatose mentality after falling. His GCS was 3 (E1, V1, M1). Pupils were non-reactive on both sides. Blood pressure was 60/30 mm Hg. He had a history of coronary angiography and received anticoagulant therapy. CT scan revealed an acute subdural hematoma (Fig. 1C). His family agreed to the do-not-resuscitate (DNR) order. The IPMc, IPUC, IPUC, and IPUt were 84, 91, 92, and 96, respectively. He died on the 3rd hospital day (HD).
Case 4
This 67-year-old male patient visited our ER with comatose mentality. He suffered from dementia. His GCS was 3 (E1, VT, M1). Pupils were both dilated. CT scan revealed a bilateral subdural hematoma (Fig. 2). After CT scanning, his respiration was ceased. A ventilator was applied. Although his family agreed to the DNR order on the 3rd HD, no one could remove the ventilator: The IPMc, IPUC, IPUC, and IPUt were 87, 89, 92, and 95, respectively. He died on the 13th hospital day (HD).
Case 5
This 82-year-old male patient transferred to our hospital with comatose mentality. He suffered from hypertension and diabetes. His GCS was 4 (E1, VT, M2). Pupils were both dilated. CT scan revealed a huge subdural hematoma (Fig. 3). The IPMc, IPUC, IPUC, and IPUt were 91, 84, 98, and 96, respectively. His family refused surgical treatment. On the 3rd HD, his motor response became 4. We suggested an endoscopic burr-hole evacuation under local anesthesia. After surgery, the degree of midline shift was much improved (Fig. 3). However, his pupils were dilated and the motor score remained unchanged. His family refused any further treatment unless surgeons guaranteed his recovery. He was transferred to a nursing home in the vegetative state on the 8th HD.
Case 6
This 63-year-old male patient visited our ER with comatose mentality. He slipped down after drinking. His GCS was 4 (E1, V1, M2). Pupils were non-reactive on both sides. CT scan revealed an acute subdural hematoma (Fig. 4). The IPMc, IPUC, IPUC, and IPUt were 75, 90, 77, and 91, respectively. Decompressive craniectomy was performed (Fig. 4). He was discharged with a severe disability on the 17th HD.
DISCUSSION
The most common surgical lesion was acute subdural hematoma in this study. Subdural hematomas became more frequent than epidural hematomas after the age of 50 years²⁰. Since patients having any mass lesion with significant mass effect should be treated operatively, types of the mass lesion did not affect the surgical decision making in both groups. Although none of the motor score of 1 underwent surgery, motor score 2 or 3 did not
decompressive craniectomy. Fig. 4. were a poor chance of survival, a prognosis incompatible with the patient’s wishes and a poor long-term neurologic prognosis. Some patients may consider death to be a preferable outcome to living in a permanent vegetative state or coma. In such situations, withdrawal of life-sustaining therapies may be the most acceptable option of care for families, relatives and medical teams according to patients’ wishes and the philosophy of care. If there is a prediction of an unfavorable outcome of >80%, a surviving patient is likely to remain severely disabled. When the prediction of an unfavorable outcome is greater than 90%, we have an ethical obligation to at least enter into a discussion, not only regarding the likely outcome but also what would have been the patient’s wishes. Although removal of the ventilator can be ethically justified, it is hard to do. In Korea, the Supreme Court sentenced the doctor who permitted a moribund discharge to 1 year and 6 month in prison and 2 year probation. Although the mean IPM and mean IPU of the group A were higher than those of the group B, the outcome was favorable in only 2 patients in the group B. A good decision may be more likely to lead to a better patient outcome, but outcome alone cannot measure the quality of a decision.
Traumatic brain injury in the elderly has been and will be an important burden to the society with longer life expectancy and an aging population, with fall being the commonest cause. Increasing use of antiplatelet and anticoagulation medication is going to further complicate the condition. In making the difficult decision about whether to proceed with a life-saving, but non-restorative surgical intervention, surrogate decision-makers for the patients will need as much objective and reliable information as possible to make a truly informed decision. Clear decisions can and should be made on the basis of straight and thoughtful communication between the senior staff member on call and the residents on duty. Ethical training for neurosurgeons is to be encouraged. To develop decision-making skills, there should be a supportive environment, with good supervision, mentoring, and role modeling. Most medical problems can be treated in multiple different ways. Each treatment option offers different trade-offs in quality of life and mortality, and there is often no objective “best” treatment. Rather, a treatment choice should be made based on an individual’s preferences and willingness to accept risk. The most consistent features about a surgical decision were that it was well informed and well considered. A mathematical model, while limited, does provide an accurate index of injury severity and may be useful in providing cause to pause and consider the long-term implications of life-saving but non-restorative surgical intervention. Shared decision making, i.e., thorough discussion with predicted probability of unfavorable outcome is best used for problems involving medical uncertainty.
preclude surgery. Contrary to the motor score, all patients with reactive pupils underwent surgery. Abnormalities in pupillary reactivity indicate brainstem compression and are strongly associated with poorer outcome. Pupillary reactivity is a more stable variable in the early phase after injury than is the GCS, because it is less prone to influences of sedation and paralysis. Even in patients with dilated pupils and GCS of 3, there is a chance of good outcome. However, aggressive treatment is highly demanding and increases suffering for patients and their families and adds unnecessary medical burden. The outcome was unfavorable in all five patients underwent surgery despite of dilated pupils, in this study. Pupillary response was a major factor of decision making. There is no age-dependent cut-off in brain injured patients. Most studies analyzed the association between age and outcome by use of threshold values, varying from 30 to 60 years of age. Relatively high mortality begins from the age of 55, 70, 75, or 80. Outcome following TBI in “younger elderly” (those less than 75 years old) can be comparable with younger adults with acceptable outcome. It would be better that the life-saving, but not restorative, surgery is reserved for the so-called younger elderly patients with dilated pupils and GCS of 3.
All patients in the group A were expired. In case 4, a ventilator was applied before adequate discussion on the treatment with his family. In case 5, a decision was changed without any eventual benefit. A wrong decision may bring a long futile care. The decision is frequently made under pressure of time. Common reasons to justify withdrawal of life-sustaining therapy were a poor chance of survival, a prognosis incompatible with...
CONCLUSION
Surgical decision making is difficult especially for the elderly patients with severe head injuries. Withholding a life-saving surgery was 15 times in 28 elderly patients with surgical lesions. There were wrong decisions, which brought futile cares. Ethical training and developing decision-making skills are necessary including shared decision making.
References
1. Andrews PJ, Sleeman DH, Statham PE, McQuatt A, Corruble V, Jones PA, et al.: Predicting recovery in patients suffering from traumatic brain injury by using admission variables and physiological data: a comparison between decision tree analysis and logistic regression. J Neurosurg 97: 326-336, 2002
2. Cagetti B, Cosu M, Pau A, Rivano C, Viale G: The outcome from acute subdural and epidural intracranial haematoma in very elderly patients. Br J Neurosurg 6: 227-231, 1992
3. Chamoun RB, Robertson CS, Gopinath SP: Outcome in patients with blunt head trauma and a Glasgow Coma Scale score of 3 at presentation. J Neurosurg 111: 683-687, 2009
4. Chaudhuri K, Malham GM, Rosenfeld JV: Survival of trauma patients with coma and bilateral fixed dilated pupils. Injury 40: 28-32, 2009
5. Frosh DL, Kaplan RM: Shared decision making in clinical medicine: past research and future directions. Am J Prev Med 17: 285-294, 1999
6. Hahm KH, Lee I: Biomedical ethics policy in Korea: characteristics and historical development. J Korean Med Sci 27 Suppl: S76-S81, 2012
7. Honeybul S, Gillett GR, Ho KM, Lind CR: Neurotrauma and the rule of rescue. J Med Ethics 37: 707-710, 2011
8. Honeybul S, Ho K, O’Hanlon S: Access to reliable information about long-term prognosis influences clinical opinion on use of lifesaving intervention. PLoS One 7: e32375, 2012
9. Honeybul S, Ho KM, Lind CR, Gillett GR: Surgical intervention for severe head injury: ethical considerations when performing life-saving but non-restorative surgery. Acta Neurochir (Wien) 153: 1105-1110, 2011
10. Honeybul S, O’Hanlon S, Ho KM: Decompressive craniectomy for severe head injury: does an outcome prediction model influence clinical decision-making? J Neurotrauma 28: 13-19, 2011
11. Hukkelhoven CW, Steyerberg EW, Rampen AJ, Parrae E, Habbema JD, Marshall LF, et al.: Patient age and outcome following severe traumatic brain injury: an analysis of 5600 patients. J Neurosurg 99: 666-673, 2003
12. Kotwica Z, Saracen A: Severe head injuries in the elderly. Hygeia Public Health 45: 197-201, 2010
13. Kuhne CA, Ruchholtz S, Kaiser GM, Nast-Kolb D; Working Group on Multiple Trauma of the German Society of Trauma: Mortality in severely injured elderly trauma patients—when does age become a risk factor? World J Surg 29: 1476-1482, 2005
14. Lingsma HF, Roozenbeek B, Steyerberg EW, Murray GD, Maas AI: Early prognosis in traumatic brain injury: from prophecies to predictions. Lancet Neurol 9: 543-554, 2010
15. Mak CH, Wong SK, Wong GK, Ng S, Wang KK, Lam PK, et al.: Traumatic brain injury in the elderly: is it as bad as we think? Curr Transl Geriatr Exp Gerontol Rep 1: 171-178, 2012
16. MRC CRASH Trial Collaborators, Perel P, Arango M, Clayton T, Edwards P, Komolafe E, et al.: Predicting outcome after traumatic brain injury: practical prognostic models based on large cohort of international patients. BMJ 336: 425-429, 2008
17. Mushkudiani NA, Engel DG, Steyerberg EW, Butcher I, Lu J, Marmarou A, et al.: Prognostic value of demographic characteristics in traumatic brain injury: results from the IMPACT study. J Neurotrauma 24: 259-269, 2007
18. Patel HC, Bouamra O, Woodford M, Yates DW, Lecky FE: Clinical article: mortality associated with severe head injury in the elderly. Acta Neurochir (Wien) 152: 1333-1357, discussion 1357, 2010
19. Pilitsis J, Atwater B, Warden D, Deck G, Carroll J, Smith J, et al.: Outcomes in octogenarians with subdural hematomas. Clin Neurol Neurosurg 115: 1429-1432, 2013
20. Rennie SC, van Rij AM, Jaye C, Hall KH: Defining decision making: a qualitative study of international experts’ views on surgical trainee decision making. World J Surg 35: 1214-1220, 2011
21. Schaller C, Kessler M: On the difficulty of neurosurgical end of life decisions. J Med Ethics 32: 65-69, 2006
22. Steyerberg EW, Mushkudiani N, Perel P, Butcher I, Lu J, McHugh GS, et al.: Predicting outcome after traumatic brain injury: development and international validation of prognostic scores based on admission characteristics. PLoS Med 5: e165; discussion e165, 2008
23. Stocchetti N, Patrono R, Citerio G, Beretta L, Colombo A: Traumatic brain injury in an aging population. J Neurotrauma 29: 1119-1125, 2012
24. Tokutomi T, Miyagi T, Ogawa T, Ono J, Kawamata T, Sakamoto T, et al.: Age-associated increases in poor outcomes after traumatic brain injury: a report from the Japan Neurotrauma Data Bank. J Neurotrauma 25: 1407-1414, 2008
25. Turgeon AF, Lauerier F, Simard JE, Scales DC, Burns KE, Moore L, et al.: Mortality associated with withdrawal of life-sustaining therapy for patients with severe traumatic brain injury: a Canadian multicentre cohort study. CMAJ 183: 1581-1588, 2011 | 2025-03-06T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4137,
1
],
[
4137,
8822,
2
],
[
8822,
12570,
3
],
[
12570,
17360,
4
],
[
17360,
22410,
5
]
]
} |
576a0a9ea444ff1a064e99065f3400794f1c813f | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 12,
"total-fallback-pages": 0,
"total-input-tokens": 40854,
"total-output-tokens": 9028,
"fieldofstudy": [
"Computer Science"
]
} | Keywords: overhead cranes; Monte Carlo simulation; variance reduction; parallel computing
Janusz SZPYTKO, Yorlandys SALGADO DUARTE*
AGH University of Science and Technology
al. Mickiewicza 30, 30-059 Cracow, Poland
*Corresponding author. E-mail: [email protected]
ROBUST SIMULATION METHOD OF COMPLEX TECHNICAL TRANSPORT SYSTEMS
Summary. In the optimization of technical systems focused on a specific functional purpose (reliability, safety, and availability) with the use of simulation methods, an important parameter is the digital simulation time of the research subject. With the complexity of the issue, the digital simulation time increases. The aim of the article is to present a method (combination of parallel computing and variance reduction techniques) of reducing the computer simulation time of the research technical object. An example of the application of the developed method was presented as a result of an experiment conducted for decision making and control processes aimed at optimizing the process of operating overhead cranes in critical conditions. In this paper, selecting parallel batch jobs computation and stratified sampling, we exponentially decreased the simulation time, finding fast and practical solutions and eliminating the time constraint in the search of solutions.
1. INTRODUCTION
Monte Carlo methods consist of generating consecutive random samples with computer algorithms to obtain numerical results. It is usually applied in three types of problems: optimization, numerical integration, and generation of samples from a probability distribution according to Kroese et al [7].
Engineering problems sometimes present countless free or unknown variables, making it suitable to use Monte Carlo methods to simulate the sensitivity of the system according to the unknown variables. Authors like Hubbard and Samuelson [5] highlight that a frequent application in system engineering problems is failure predictions based on Monte Carlo simulations.
According to the law of large numbers and the Central Limit Theorem, the calculation of integrals described by the expected value (centre of mass) of some random variable can be approximated to the empirical mean (also known as sample mean) of independent samples of the variable, and it is known that the dispersion around the expected value is associated with the sample variance and the sample length. Therefore, Monte Carlo methods are often used to solve probabilistically oriented problems, particularly those involving mixture or compound probability distributions. A frequent application is to generate random samples from parameterized probability distributions, commonly known as the Markov chain Monte Carlo (MCMC) sampler.
Examples of practical applications based on the Monte Carlo method are the papers, Monte Carlo simulation model to estimate the reliability of logistics and supply chain networks from Ozkan and Kilic [11]; Monte Carlo tolerance simulation on a prefabricated construction assembly, showing a proactive design tool with several key advantages for prefabricated and offsite construction from Rausch et al [12]; Monte Carlo simulation model to analyze and optimize the time interval between periodic inspections in cold standby systems considering the required availability and the lowest cost possible from Alebrant Mendes and Weber Lorenzoni [1]; and a model to improve cooling load
prediction reliability method in which the input variables are calibrated offline with Monte Carlo simulations and stochastic treatment before inputting them into the prediction model from Fan et al [4].
In the references cited before, all of them highlight the advantages of the simulation approach showing relevant results, but only a few of them show practical implementations within a reasonable CPU time.
The obvious limitation in the simulation approach is clearly time, and we believe that this is the reason why authors sometime just ignore the limitation, as we can observe in the references cited.
But, on the other hand, references such as Dieker et al [3] and Leahu et al [8], work strongly in parallel with methodological solutions on how to allocate time-consuming simulations depending on the length of the queues.
In this paper, we touch on a classical application of Monte Carlo simulation: generating samples from probability distributions fitted to historical degradation data. Specially, we estimate by convolution the loss capacity (risk indicator) given the stochastics availability (including degradation due the operation and maintenance planned) of the analyzed overhead crane system and the minimum set of overhead cranes needed to guarantee the production line. The risk indicator obtained is used as a criterion to evaluate the quality of the maintenance scheduling.
More details about the model adopted in this paper can be found in the references Szpytko and Salgado Duarte [16, 17].
As other applications cited, when the complexity of the simulation increase s to assess a scenario, the computation time needed to generate a robust estimation can be an issue.
Knowing the limitation, in this investigation, we discuss two possible improvements to reduce the simulation time given a fixed scenario. As a starting point, we analyze the simulation time before and after the improvements proposed.
The investigation carried out in this paper is organized as follows: first, a full discussion of the proposed methodological and technical improvements, then the analyzed model adopted is discussed, and then the experiment performed to evaluate the time improvements is described. Finally, some conclusions are drawn to highlight possible outcomes in this research area.
2. MATERIALS AND METHODS
In this investigation, to reduce the Monte Carlo simulation time, the selected improvements are variance reduction (methodological) and parallel computing (technical). Each of them has a wide family of possibilities and definitions; therefore, we specify the selection later with a brief discussion about, starting with the variance reduction.
2.1. Variance reduction
Within Monte Carlo methods, variance reduction is a strategy (by referring to how the random sample is generated or by using some known features of the sampled distribution) applied to increase the accuracy of estimates (e.g., the expected value, in our case) that can be obtained from a computer simulation as highlighted by Botev and Ridder [2].
Each random variable generated from a computer simulation has an associated variance or variability, commonly speaking, that determines the accuracy of the result. To achieve a statistically efficient simulation, that is, to obtain an estimate with smaller confidence intervals for the random variable analyzed, variance reduction techniques can be used.
The main techniques are common random numbers, antithetic variates, control variates, importance sampling, and stratified sampling according to the authors Botev and Ridder [2].
Variance reduction techniques can be commonly separated into groups. First, we have antithetic sampling (for each generated uniform random number defined between [0, 1], its complement is used to compensate large deviations from the mean), stratification (the generated sample is strategically
divided into groups), and common random numbers (usually used when it is necessary to compare two different distributions, for a given common random sample). All these methods improve the variance by sampling the values strategically.
On the contrary, we have conditioning and control variates. These methods use information from the sampler to bound the generated outputs.
Finally, the importance sampling method. This method changes where we take the sample values from, i.e. it intentionally over samples from some regions and then corrects this distortion by re-weighting.
All the techniques are well defined in the literature and possible to implement, but given the model specificities, easy implementation and understanding, the case selected is stratified sampling.
In statistics, stratified sampling is a technique for sampling a population whose essence is to divide the population into subpopulations.
The precision of the estimation of a population statistic from pseudo-random samples, for example the mean, likewise the sample size, depends on the variability between the generated population samples. Consequently, in addition to increasing the sample size, another possible way to increase the precision of the estimate could be by dividing the generated pseudo-random samples into groups, so that the variability within groups is minimal and the variability between groups is maximum. In this way, smaller samples can be selected from each of the groups formed. The groups formed are called strata and the process of stratum formation is known as stratification according to Singh and Singh Mangat [14].
When we are in the presence of simple stratified random sampling, under the assumption that samples from different strata are selected independently, each stratum can be treated as an independent population. The estimation of the stratified mean will be more efficient than the usual simple random samples from different strata are selected independently, each stratum can be treated as an independent population. The estimation of the stratified mean will be more efficient than the usual simple random mean if the variation between the stratum means is large enough in relation to the variation within the stratum. However, the gain in precision also depends on the method used to create the strata. According to Singh and Singh Mangat [14], other points that need to be considered are as follows:
- Determining the number of strata to be constructed (for this point the characteristics of the population guide the decision).
- Allocation of total sample size to different strata (usually all the strata have the same size, but the technique allows for strata with different sizes).
- The choice of strata (the selection is mainly guided by the modelled problem).
This method introduces a challenge, the optimum allocation (or disproportionate allocation). As we can deduce, the allocation depends on the problem to be modelled, therefore, in next sections we discuss the topic by computational experiments after the evaluated scenario is fully described.
Relevant in this investigation is the expected value and the variance of the expected value, because the first reflects the risk value estimation of the system evaluated, and the second define the number of simulations needed to get a robust estimation, both redefined later in the following sections according to the model formulation.
Therefore, as starting point according to Singh and Singh Mangat [14], we can define the mean (Equation 1) and variance (Equation 2) of stratified random sampling as follows:
\[
\bar{y} = \sum_{h=1}^{k} W_h \bar{y}_h \tag{1}
\]
\[
s^2 = \sum_{h=1}^{k} W^2_h \left( \frac{N_h - n_h}{N / n_h} \right) s^2_h \tag{2}
\]
where: \( N_h \) = total number of units in the stratum \( h \);
- \( f_h = n_h / N \) = sampling portion for the stratum \( h \);
- \( n_h \) = number of units selected in the sample from the stratum \( h \);
- \( w_h = N_h / N \) = proportion of the population units falling in the stratum \( h \);
- \( Y_h \) = the value of study variable for the \( i \)-th unit in the stratum, \( i = 1, 2, \ldots, N_h \);
- \( Y_h^* = \sum_{i=1}^{n_h} Y_i \) = stratum total for the estimation variable based on \( N_h \) units;
\[ \bar{y}_h = \frac{1}{N_h} \sum_{i=1}^{N_h} y_{hi} = \text{mean for the estimation variable in the stratum}; \]
\[ \bar{y}_h = \frac{1}{n_h} \sum_{i=1}^{n_h} y_{hi} = \text{stratum sample mean for the estimation variable}; \]
\[ \sigma^2_h = \frac{1}{N_h} \left( \sum_{i=1}^{N_h} y_{hi}^2 - N_h \bar{y}_h^2 \right) = \text{stratum variance based on } N_h \text{ units}; \]
\[ S^2_h = \frac{N_h}{N_h - 1} \sigma^2_h = \text{stratum mean square based on } N_h \text{ units}; \]
\[ s^2_h = \frac{1}{n_h - 1} \left( \sum_{i=1}^{n_h} y_{hi}^2 - n_h \bar{y}_h^2 \right) = \text{sample mean square based on } n_h \text{ sample units drawn from the stratum}. \]
2.2. Parallel computing
Once the proposal for methodological improvement has been defined, we jump to the technological improvement in this section.
Nowadays, parallel computing is relevant to solve complex problems in science and engineering in practical times. Parallelism is a technological strategy to speed up calculations that require a lot of time and memory. According to the Keyes [6], historically, parallelism had acceleration as its main objective, which was characterized by Amdahl's law.
Gustafson-Barsis' law is another perspective of the same idea according to Keyes [6]; in this case, the idea is to measure how to keep time constant when the complexity of the problem increases. This definition is described by scalability. This new law leads the concept of designing systems with an increasing number of processors, a law applied and followed conceptually in this research.
As we know, a system design with these features is expensive when the required number of processors is relatively high, and sometimes the profitability of the software does not exploit all the available power. For this reason, the trend of parallelism is moving toward the era of clusters. For example, networked desktop computer labs can be configured as clusters. Regardless, it is well known that progress continues, at the microprocessor level, parallelism is found in multicores and many cores as highlighted by Trobec et al [18].
At the tip of the iceberg, with the idea of raising parallelism to its maximum expression, are the ideas of Grid and Cloud. These structures centralize computing power and provide access through the network, while also minimizing power consumption. In addition, on the somewhat more distant horizon, there are paradigms such as quantum computing, optical computing, and chips working on biological structures, which have a potential for future parallel computing according to Trobec et al [18].
All the advances described before walk at the same time with the development of methods and algorithms for these systems. It is well known that without effort in the field of software, it is impossible to exploit the computational power of the super modern computers that exist and are available.
The design of efficient and robust algorithms is essential to efficiently solve complex scientific problems where parallelism is in principle necessary.
The continuous trend over time is to increase the number of cores on a single chip. Now a simple conventional desktop computer can have 16 cores. To take full advantage of available capabilities, new tools, new algorithms, and a new way of looking at the programming will be required.
Today, a standard operating system can run different tasks in parallel depending on the available cores. However, in cases of serial software programme, the program or code needs to be restructured and parallelized to take full advantage of the multi-core architecture.
Following the paradigm of parallel computing, the simulation model adopted in this paper (defined later in the following sections), is implemented in MATLAB [10], parallelized, and tested assessing the simulation times.
The reason we selected MATLAB to implement the model is because it allows us, given a well-defined parallel code sequence, to run the simulations using batch jobs and parallel loops at the same time, as shows Fig. 1.
The MATLAB architecture for parallel computing is frequently used by authors from different fields and is well accepted standard software for research purposes in the academy. The literature is full of applications with this tool.
According to the MATLAB online help, using batch jobs allows offloading the execution of long-running calculations in the background, increasing the efficiency of calculation time.
Additionally, Parallel Computing Toolbox of MATLAB enables interactive parallel computing and allows us to speed up independent calculations by running them on multiple workers at the same time. The idea is to use the parfor routines to run for-loop iterations in parallel on independent workers, as the online help describes.
The combined use of batch jobs and parallel loops exponentially increases the performance efficiency of the code and allows us to use all the computation resources on the computer to perform a task, as shown in Fig. 1.
When working interactively in a MATLAB session, we can download the job to a MATLAB work session to run as a batch job, which means we can create a group of workers for their batch job.
The workers can run on the same machine as the client or on a remote cluster machine. With that said, the code implemented in MATLAB can also be run on a cluster if needed.
### 2.3. Improvements proposed
The main outcome of the investigation is to evidence the reduction of the simulation times, given a parameterized scenario and a scheduling solution, for the adopted model, through the application of the methodological and technological improvements described in the sections 2.1 and 2.2.
As a general view of the research conducted, Fig. 2 summarizes the idea proposed, leaving clearly defined the experiment discussed in the following sections.
3. MODEL DESCRIPTION
The model definition under study in this paper is taken from Szpytko and Salgado Duarte [16, 17]. The first reference describes the connection between the raw information sources and the model variables, and the second reference details the optimization model definition and the solution achieved given a parameterized scenario (optimal maintenance scheduling).
As a general description of the model used, we can say that the subject of the mathematical model is to manage the logistics-maintenance process of overhead type cranes operated in critical systems.
A hot rolling mill system in a steel plant, mainly supported by critical overhead cranes operating under hazard conditions and running continuously, was selected as a study case.
The model output (requested as an engineering solution) is an optimal risk-oriented maintenance scheduling for critical overhead cranes.
The model input (created as a formal structured information) is a digital database with historical information related to operation, maintenance-logistics, and management processes of the system.
The optimization model supporting the risk-oriented maintenance scheduling can be defined as stochastic, no-linear, with bounded constraints.
The model inputs are a SCADA (Supervisory Control And Data Acquisition) and SAP (Systems, Applications & Products in Data Processing) systems. Once the inputs are in place, the optimization algorithm proposes maintenance scheduling scenarios to be evaluated in the model proposed by Szpytko and Salgado Duarte [17] (risk oriented) given the corresponding restrictions. The best solution is the scenario with lower risk, and at the same time, the final output of the model.
From the database structure, the proposed optimization model takes the input variables and parameters. The model objective is to minimize the conditional expected value of the convolution function defined as $E[R]$ (risk indicator Capacity Loss), between the overhead crane capacity distribution function of the steel plant affected for the maintenance scheduling defined as $X$ and the necessary load capacity distribution function defined as $Y$ (see reference Szpytko and Salgado Duarte [17], model description).
In the model adopted, the time needed to obtain the optimal solution (maintenance scheduling) depends on the number of evaluations on the objective function (because a heuristic algorithm is used) and within each evaluation, on the number of simulations needed to achieve an accurate estimate of the conditional expected value of the convolution function (the variance of the expected value define
the accuracy of the estimation, and therefore the number of simulations needed to achieve a desired error).
Saying that, Fig. 3 shows an architecture diagram of the model flow and the improvements layers proposed (methodological and technical) to improve the simulation time.
The variance reduction technique layer impacts the computation of the conditional expected value by the Monte Carlo method given a proposed scenario of maintenance scheduling by the optimization algorithm.
In the case of the parallel computing layer, each scenario proposed by the optimization algorithm is independent (each maintenance scheduling scenario), therefore, can be evaluated independently.
Consequently, the logic of the parallel computing improvement is to evaluate block-by-block scenarios depending on the available CPU in the computer.
Fig. 3. Model flow with improvements
4. RESULT AND DISCUSSION
Given the starting point, the previous two cited references Szpytko and Salgado Duarte [16, 17] and a general overview description of the model, additional definitions are added in this section to complete how we improve the simulation times applying the layers proposed in practice.
The parallel computing layer is mainly oriented to the implementation and has no direct implication for the model definition, but the variance reduction technique layer has an impact on the definition of the conditional expected value, which is why it is discussed in depth in this section.
To estimate the conditional expected value of the risk function defined on Szpytko and Salgado Duarte [16], we must know the system distribution function. In this case the construction is possible by sampling random values from individual distribution functions (overhead crane individual distribution functions depend on the historical degradation data, fitting process, and the reliability block diagram system structure), and then following the definitions described on Szpytko and Salgado Duarte [16], the system distribution function can be computed by Monte Carlo method.
As we declare before, the complexity and dimension of the system analyzed is large. Therefore, the simulation time is an issue.
Equations 3, 3a and 3b state the criteria for robust expected value estimation (optimization target) and the clear dependency on number of simulations and the standard deviation of the conditional expected value:
\[
\text{Capacity Loss} = \bar{E}[R] \pm \beta \sigma[R]
\]
\( (3) \)
To compensate for an unnecessary increase in the number of simulations to achieve an accurate estimate (time-consuming), variance reduction techniques are a way to reduce the standard deviation by sampling strategically.
Relevant for this investigation and easy to implement in the model proposed by Szpytko and Salgado Duarte [16] is stratified sampling, as we declare above.
When stratified sampling are used, now adapting the definitions to the nomenclature of the model, the population mean (4) and variance (5) are given by equations 4 and 5:
\[
E[R] = \frac{1}{K} \sum_{i=1}^{K} E[R]_k
\]
\[
V[R] = \frac{1}{N} \sum_{i=1}^{K} \left( \frac{V[E[R]_k]}{N} \right)
\]
where \(E[R]_k\) and \(V[R]_k\) are given by the standard expected value and variance, respectively, for each \(k\) strata; \(K\) = number of strata (in our case, the number depends on the size of the generated sample); \(N\) = size of stratum \(k\) (estimated by experimental calculations).
Consequently, from previous definition, the Capacity Loss indicator estimated by Monte Carlo simulation, and defined in Equation 3, when stratified sampling are used, is defined by equations 6, 6a and 6b:
\[
Capacity Loss = E[R] \pm \frac{\sigma[R]}{\sqrt{K}}
\]
\[
Capacity Loss = \frac{1}{K} \sum_{i=1}^{K} E[R]_k \pm \sqrt{\frac{V[E[R]_k]}{N}} \frac{\sqrt{K}}{\sqrt{K}}
\]
\[
Capacity Loss = \frac{1}{K} \sum_{i=1}^{K} E[R]_k \pm \frac{\sigma[E[R]_k]}{\sqrt{K}}
\]
Visible issues coming when this technique is used, the \(w_k\) estimation, where \(w_k = N/K\) is the population weight of stratum \(k\).
To estimate optimal weights, we use an experimental design and heuristic adjustments as a calibration (finding the optimal weights for the model) following some principles discussed on Mandies et al [9].
The experiment conducted is as follows: given a maintenance scheduling solution for the scenario, taken from Szpytko and Salgado Duarte [17], we compute by Monte Carlo method the conditional expected value (optimization target) and the error \(\beta\) continuously until we achieved the desired robust estimation without any stratification, pure Monte Carlo, saving in the end the string of random numbers needed, then with the same string of random number saved (for reproducibility reasons), by steps, we changed \(N_k\) (size of stratum \(k\)) and we recalculated the conditional expected value and the error \(\beta\) continuously every \(N_k\) simulations until we achieved the desired robust estimation.
Table 1 summarizes the results of the experiment conducted. Table 1 shows the number of simulations needed to achieve the same error for each stratum tested.
Based on Table 1 results and making a conservative decision in order to avoid singularities and be aware about the fact that we never know the sampled population size needed when the error \(\beta\) is fixed, we decided a size of stratum equal to 5, therefore, during the conditional expected value estimation of the following analysis, we grouped every 5 simulations and we estimated the mean, then we estimated the expected value (optimization target) and the error \(\beta\) continuously every group of 5 samples until we achieved the desired robust estimation.
Table 1 clearly shows the decreases in the number of simulations needed (from 1290 to 230 simulations).
### Calibration results
| Cases | Expected Value | Error | Simulations Needed |
|--------|----------------|-----------|--------------------|
| Without| 1042.45 | 0.009995 | 1290 |
| \( N_i = 2 \) | 1046.51 | 0.009992 | 650 |
| \( N_i = 3 \) | 1048.38 | 0.009998 | 420 |
| \( N_i = 4 \) | 1058.69 | 0.009941 | 312 |
| \( N_i = 5 \) | **1059.10** | **0.009904** | **230** |
| \( N_i = 10 \) | 1059.98 | 0.009452 | 170 |
| \( N_i = 15 \) | 1080.42 | 0.008750 | 120 |
| \( N_i = 20 \) | 1080.42 | 0.009678 | 120 |
| \( N_i = 25 \) | 1069.98 | 0.008906 | 100 |
| \( N_i = 30 \) | 1133.33 | 0.001383 | 60 |
| \( N_i = 35 \) | 1087.93 | 0.009099 | 70 |
All the estimations so far are performed with a personal computer i5 5250U 1.6 GHz CPU. The solution time of the optimization problem in this investigation depends on the number of samples (\( N_s \)) needed to ensure the error in each scenario simulated and the number of evaluations (\( E \)) in the objective function to reach the solution.
According with results from Szpytko and Salgado Duarte [17], the average time of one Monte Carlo simulation is \( [(1.955813 \pm 0.072131) \cdot N_s \cdot E] \) seconds.
Visible so far how the number of simulations is reduced by variance reduction (see Table 1). On the other hands, in the case of the number of evaluations, when parallel computing is used, in our case, four scenarios are evaluated at the same time because the features of the i5 5250U 1.6 GHz CPU allow us to run four computations at the same time (see Fig. 3).
Once the decision of the size of the strata has been made, we run the model by moving sensitive variables, in this case the efficiency indicator (see model description in reference Szpytko and Salgado Duarte [16]), evaluating the simulation time with and without applying the improvement layers.
As we declare here, using parallel computing reduces by four the time needed to assess the scenario. Once both improvements are described, below we define the experiment to analyze the simulation times. The experiment consists of changing the efficiency indicator between 80% to 85% and assessing the simulation times needed with, and without variance reduction techniques implemented. In the case of parallel computing the combined impact is divide into all the cases performed by four (i5 5250U 1.6 GHz CPU, has four CPU).
The experiment results shown in Table 2 and Table 3 corroborate the hypothesis of this research. Visible in Table 2 the reduction of the simulation times, only the variance reduction technique decreases the simulation time in almost five times less (depending on the scenario evaluated), and with the addition of the parallel computing, the reduction in time will happen to almost twenty times less.
On the other hand, Table 3 shows for the same scenarios, the estimation of the risk indicator, indicating estimations within confidence intervals in all the cases (robust estimation assumed, \( \varepsilon = 0.01 \)). Of particular interest are the results in Table 2, because when the system is more reliable (loss capacity lower, see Table 3 also) the variance reduction technique is more pertinent, therefore the improvements behave well on a highly reliable system.
Fig. 4 adds other results to confirm the reduction of the simulation times (visual impact). Given a maintenance scheduling scenario, we simulated by Monte Carlo the evaluation without and with stratification, estimating the expected value in both cases.
In this experiment, the efficiency indicator is 75%. Fig. 4 shows the decreases in the simulations needed to achieve a robust estimation when the variance reduction techniques are used.
Simulation time summary
| Scenario | Time (seconds) | Efficiency | Without variance reduction | With variance reduction |
|----------|----------------|------------|---------------------------|------------------------|
| | | 80% | 8609.88 | 1846.05 |
| | | 81% | 7289.31 | 1571.35 |
| | | 82% | 7977.12 | 2052.64 |
| | | 83% | 4767.47 | 896.93 |
| | | 84% | 3531.06 | 722.11 |
| | | 85% | 2129.12 | 696.81 |
Expected value estimation with and without variance reduction
| Scenario | Loss capacity (tons) | Efficiency | Without variance reduction | With variance reduction |
|----------|----------------------|------------|---------------------------|------------------------|
| | | 80% | 111.91 | 111.47 |
| | | 81% | 152.70 | 151.42 |
| | | 82% | 212.68 | 214.56 |
| | | 83% | 307.54 | 313.85 |
| | | 84% | 472.55 | 477.45 |
| | | 85% | 782.20 | 788.51 |
Fig. 4. Normalized risk value distribution with and without stratification
On the contrary, the model analyzed is a highly reliable system as we stated above, so if we do not apply variance reduction techniques, the number of simulations increases exponentially when the system is more reliable.
Simulation-based approaches are powerful for modelling stochastic processes with complex compound functions, but the time to simulate these processes can be a limitation with the current computing power. This paper evidence strategies to decrease the impact of the limitation and clearly improvements for Monte Carlo simulation approaches.
5. CONCLUSION
The paper presents a combination of parallel computing and variance reduction techniques and shows how they can help to reduce the computer simulation time in a case study of practical importance, the decision-making and control processes of overhead cranes operating optimally under critical conditions.
The paper describes two possible improvements to reduce the simulation times of the risk assessment approach, showing consistent results in all the scenarios. The above results confirm, in the presented scenarios, the time reductions of a simulation-based approach. The presented improvements open the way to keep using simulations approach but in a robust way.
Through Cloud Computing (Parallel Computing), complex NP problems based on simulations that seek solutions in exhaustive and complex decision-making diagrams, designed by humans, Cyber-Physical Systems can find fast and practical solutions, as the case presented in this paper, eliminating the time limitation in the search of solutions.
Acknowledgement
The work has been financially supported by the Polish Ministry of Education and Science.
References
1. Alebrant Mendes, A. & Weber Lorenzoni, M. Analysis and optimization of periodic inspection intervals in cold standby systems using Monte Carlo simulation. *Journal of Manufacturing Systems*. 2018. Vol. 49. P. 121-130.
2. Botev, Z. & Ridder, A. *Variance Reduction*. Wiley StatsRef: Statistics Reference Online: 1-6. 2017.
3. Dieker, A.B. & Ghosh, S. & Squillante, M.S. Optimal resource capacity management for stochastic networks. *Operations Research*. 2017. Vol. 65(1). P. 221-241.
4. Fan, C. & Liao, Y. & Zhou, G. & Zhou, X. & Ding, Y. Improving cooling load prediction reliability for HVAC system using Monte-Carlo simulation to deal with uncertainties in input variables. *Energy & Buildings*. 2020. No 110372.
5. Hubbard, D. & Samuelson, D.A. *Modeling Without Measurements*. 2009. OR/MS Today: 28-33.
6. Keyes, D. Parallel numerical algorithms: An introduction. In: *Parallel Numerical Algorithms*. Keyes D.E. & Sameh, A. & Venkatakrishnan, V. (Eds.). Kluwer Academic Publisher. Norwell, MA. 1997.
7. Kroese, D.P. & Brereton, T. & Taimre, T. & Botev, Z.I. Why the Monte Carlo method is so important today. *WIREs Comput Stat*. 2014. Vol. 6(6). P. 386-392.
8. Leahu, H. & Mandjes, M. & Oprescu, A.M. A numerical approach to stability of multiclass queueing networks. IEEE. *Transactions on Automatic Control*. 2017. Vol. 62(10). P. 5478-5484.
9. Mandjes, M. & Patch, B. & Walton, N.S. Detecting Markov chain instability: a Monte Carlo approach. *Stochastic Systems*. 2017. Vol. 7(2). P. 289-314.
10. MATLAB. version 9.7.9.1319299 (R2019b). Natick, Massachusetts: The MathWorks Inc. 2010.
11. Ozkan, O. & Kilic, S.A. Monte Carlo Simulation for Reliability Estimation of Logistics and Supply Chain Networks. *IFAC PapersOnLine*. 2019. Vol. 52(13). P. 2080-2085.
12. Rausch, C. & Nahangi, M. & Haas, C. & Liang, W. Monte Carlo simulation for tolerance analysis in prefabrication and offsite construction. *Automation in Construction*. 2019. Vol. 103. P. 300-314.
13. Salgado Duarte, Y. & Szpytko, J. & del Castillo Serpa, A.M. Monte Carlo simulation model to coordinate the preventive maintenance scheduling of generating units in isolated distributed Power Systems. *Electric Power Systems Research*. 2020. Vol. 182. No. 106237.
14. Singh, R. & Singh Mangat, N. Elements of Survey Sampling. Springer Science + Business Media Dordrecht. 1996.
15. Spall, J.C. Estimation via Markov Chain Monte Carlo. IEEE Control Systems Magazine. 2003. Vol. 23(2). P. 34-45.
16. Szpytko, J. & Salgado Duarte, Y. Integrated maintenance platform for critical cranes under operation: Database for maintenance purposes. Proceeding of 4th IFAC Workshop on Advanced Maintenance Engineering, Services and Technologies Sept. 10-11, 2020. Cambridge, UK, IFAC PapersOnLine. 2020. Vol. 53(3). P. 167-172.
17. Szpytko, J. & Salgado Duarte, Y. Exploitation Efficiency System of Crane based on Risk Management. Proceeding of International Conference on Innovative Intelligent Industrial Production and Logistics. IN4PL 2020. 2-4 November 2020. ISBN: 978-989-758-476-3.
18. Trobec, R. & Vajtersic, M. & Zinterhof, P. (Eds.). Parallel Computing Numerics, Applications, and Trends. Springer. Dordrecht Heidelberg London New York. 2009.
Received 12.12.2019; accepted in revised form 11.05.2021 | 2025-03-06T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3408,
1
],
[
3408,
7287,
2
],
[
7287,
11551,
3
],
[
11551,
15150,
4
],
[
15150,
17487,
5
],
[
17487,
20119,
6
],
[
20119,
22586,
7
],
[
22586,
25805,
8
],
[
25805,
29821,
9
],
[
29821,
32137,
10
],
[
32137,
35515,
11
],
[
35515,
36545,
12
]
]
} |
e03204545a75d1701f8116b06fce94e98e156473 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 15,
"total-fallback-pages": 0,
"total-input-tokens": 56817,
"total-output-tokens": 14419,
"fieldofstudy": [
"Computer Science"
]
} | An expectation transformer approach to predicate abstraction and data independence for probabilistic programs
Ukachukwu Ndukwu and AK McIver
Department of Computing, Macquarie University, NSW 2109 Australia.
{ukndukwu,anabel}@science.mq.edu.au
In this paper we revisit the well-known technique of predicate abstraction to characterise performance attributes of system models incorporating probability. We recast the theory using expectation transformers [8], and identify transformer properties which correspond to abstractions that yield nevertheless exact bound on the performance of infinite state probabilistic systems. In addition, we extend the developed technique to the special case of “data independent” programs [14] incorporating probability. Finally, we demonstrate the subtleness of the extended technique by using the PRISM model checking tool [1] to analyse an infinite state protocol, obtaining exact bounds on its performance.
1 Introduction
Automated analysis of infinite (very large) state systems often relies on abstractions which summarise the essential behaviour as a finite state “anti-refinement” in such a way as to guarantee the desired properties (if indeed they hold). Typically, however, abstractions introduce nondeterminism, and in a probabilistic system this can lead to a high degree of imprecision in the estimated probabilistic properties. The choice of abstraction therefore is critical; some approaches to finding the right one use “abstraction refinement”, sometimes relying on counterexamples of failed attempts to obtain incremental improvements [7].
In this paper we revisit the technique of “predicate abstraction” from the perspective of “expectation transformers”. Predicate abstraction refers to the notion of approximating a system using a given set of predicates: states are grouped together according to the predicates they satisfy (in the given set), and the system is abstracted by tracking only the transformations expressible in the induced equivalence classes. Expectation transformers [8] is a generalisation to probabilistic systems of Hoare/Dijkstra-style semantic reasoning [20] — predicates are replaced by real-valued functions of the state. The approach is equivalent to operational models of programming based on Markov-Decision Processes, but results in a convenient proof system for verifying general properties of probabilistic programs.
In particular we are able to characterise, using expectation transformers, a simple criterion for when an abstraction gives exact quantitative analysis for probabilistic properties. The criterion is sufficient to identify when predicate abstraction introduces no additional nondeterminism. A typical class of programs where this is effective is the so-called class of “data independent” programs [14]. A program is data independent whenever its control structure does not depend on the precise values of the data. Wolper [14] first identified this as a class of interesting programs amenable to verification via model checking [9]. In addition to Wolper’s idea, we consider the notion of probabilistic data independence where the probabilistic choice cannot be dependent on the data. In general, the idea we propose here is aimed
*The authors acknowledge support from (I) The Australian Commonwealth Endeavour International Postgraduate Research Scholarship (E-IPRS) Fund, and (II) The Australian Research Council (ARC) Grant Number DP0879529.
An expectation transformer approach to predicate abstraction
at constructing abstractions which result in no loss of information especially when probability plays a crucial role in the performance analysis of infinite state system models. Such abstractions are said to be “information-preserving” since they suffice as exact representations of their original systems.
Using the expectation transformer approach we prove the “folk theorem” (see [4]) for probabilistic systems: that data independent programs can be treated with predicate abstraction yielding exact results on threshold properties such as “the probability that a set of states has been reached in at most \( k \) steps”.
In particular our contributions in this paper are:
(i) The development of a technique which permits the application of predicate abstraction to probabilistic programs using expectation transformers;
(ii) An establishment of a criterion for identifying abstractions which do not lose information;
(iii) We show how the developed technique and criterion can be applied to data independent programs especially when probability plays a crucial role;
(iv) And finally, a demonstration of the technique on a case study of a system with potential infinite state behaviour.
This paper is structured as follows: In Sec. 2 we summarise the expectation transformer semantics for probabilistic programs, Sec. 3 is the development of the technique for predicate abstraction using expectation transformers. In Sec. 4 we show how the technique can be applied to identify when predicate abstraction yields exact thresholds for infinite state systems; we then explore the special treatment of data independent programs. In Sec. 5 we illustrate the technique by model checking the Rabin’s choice coordination problem (also known as the distributed consensus) [13]; this is a protocol which has the potential to require unbounded resources on its performance and therefore cannot be verified directly with a model-checking approach. However the theory of Sec. 4 shows that the results we obtain using its information-preserving abstraction are nevertheless exact interpretations of its performance.
1.1 Summary of notation
Function application is represented by a dot, as in \( f \cdot x \) (rather than \( f(x) \)). We use an abstract state space \( S \). Given predicate \( \text{Pred} \) we write \( [\text{Pred}] \) for the characteristic function mapping states satisfying \( \text{Pred} \) to 1 and to 0 otherwise, punning 1 and 0 with “True” and “False” respectively. Whenever \( e, e' \) are real-valued functions over \( S \) we write \( e + e' \), \( e \sqcup e' \), \( e \sqcap e' \) for the pointwise addition, maximum and minimum. Moreover \( \alpha \times e \) is \( e \) scaled by the real \( \alpha \).
For commutative operator \( \odot \), we use \( (\odot x \in X \cdot f \cdot x) \) for the comprehension which applies \( \odot \) between all instances \( f \cdot x \) as \( x \) ranges over \( X \). For example, \( x \in [0, 1] \cdot x^2 \) gives the maximum value of \( x^2 \) as \( x \) ranges over the closed interval \([0, 1]\).
2 Probabilistic program semantics and expectation transformers
When programs incorporate probability, their properties can no longer be guaranteed “with certainty”, but only “up to some probability”. For example the program
\[
\text{inc} \triangleq x := \frac{x}{2} \odot (\frac{1}{2} \oplus x + 1)
\]
sets the integer-valued variable \( x \) to \( x/2 \) (the result of the integer division) only with probability \( 1/2 \) — in practice this means that if the statement (1) were executed a large number of times, and the number
of times that \(x\) was halved or increased tabulated, roughly \(1/2\) of them would record \(x\) as having been halved (up to well-known statistical confidence \([15]\)).
The probabilistic guarded command language pGCL \([8]\) and its associated quantitative logic were developed to express such programs and to derive their probabilistic properties by extending the classical assertional style of programming. Programs in the pGCL are modeled (operationally) as functions (or transitions) which map initial states in \(S\) to (sets of) probability distributions over final states — the program at \([1]\) for instance has a single transition which maps any initial state \(x = k_0\) to a (single) final distribution; we represent that distribution as a function \(d\), evaluating to \(1/2\) when \(x = k_0/2\) or \(x = k_0+1\).
Since properties are now quantitative we express them via a logic of real-valued functions, or expectations. For example, the property “if the initial state satisfies \(x = 0 \lor x = 2\), then the final value of \(x\) is 1 with probability \(1/2\)” can be expressed as the expected value of the function \([x = 1]\) with respect to \(d\), which evaluates to \(1/2 \times 1 + 1/2 \times 0 = 1/2\), when \(x\) is initially 2 for example.
Direct appeal to the operational semantics quickly becomes impractical for all but the simplest programs — better is the equivalent transformer-style semantics which is obtained by rationalising the above calculation in terms of expected values rather than transitions, and the explanation runs as follows. Writing \(\mathcal{E}S\) for the set of all (non-normalised) functions from \(S\) to the interval \([0, 1]\), which we call the set of expectations, we say that the expectation \([x = 1]\) has been transformed to the expectation \([x = 0 \lor x = 2]\)/2 by the program inc \([1]\) above so that they are in the relation “1/2 is the expected value of \([x = 1]\)” with respect to inc’s result distribution whenever \(x\) is initially either 0 or 2”. More generally given a program \(\text{Prog}\), an expectation \(e\) in \(\mathcal{E}S\) and a state \(s \in S\), we define \(\text{wp.Prog.e.s}\) to be the expected value of \(e\) with respect to the result distribution of program \(\text{Prog}\) if executed initially from state \(s\). We say that \(\text{wp.Prog}\) is the expectation transformer relative to \(\text{Prog}\). In our example that allows us to write
\[
[x = 0 \lor x = 2]/2 = \text{wp}(x := x/2 \lor x := x+1).[x = 1].
\]
In the case that nondeterminism is present, execution of \(\text{Prog}\) results in a set of possible distributions and we modify the definition of \(\text{wp}\) to take account of this — indeed we define \(\text{wp.Prog.E.s}\) so that it delivers the least-expected value with respect to all distributions in the result set. The transformers \([8]\) give rise to a complete characterisation of probabilistic programs with nondeterminism, and they are sufficient to express many performance-style properties, including the probability that an event occurs, the expected time that it occurs, and long-run average of the number of times it occurs over many repeated executions of the system.
In Fig. \(\text{I}\) we set out the semantics for the pGCL, a variation of Dijkstra’s GCL with the addition of probabilistic choice. All the programming features have been defined previously elsewhere, and (apart from probabilistic choice) have interpretations which are merely adapted to the real-valued context. For example, nondeterminism, as explained above, is interpreted demonically and can be thought of as being resolved by a “minimal-seeking demon”, providing guarantees on all program behaviour, such as is expected for total correctness. Probabilistic choice, on the other hand, selects the operands at random with weightings determined by the probability parameter \(p\). Iteration is defined by a least fixed point of a monotone expectation-to-expectation function.\(^1\)
We end this section with a discussion of a simple performance property: a probabilistic analysis of the number of iterations until termination. Given a loop do \(G \rightarrow \text{Prog} od\) which executes the program \(\text{Prog}\) until \(G\) becomes false, we can compute the probability that the loop has executed no more than \(k\)
\(^1\)Well-definedness is guaranteed by, for example, restricting the expectations to lie in the real interval \([0, 1]\) or to complete the reals with \(\infty\). These issues have been discussed elsewhere \([8]\).
An expectation transformer approach to predicate abstraction
\[
\begin{array}{ll}
skip & \text{wp.skip}\cdot E \doteq E \\
abort & \text{wp.abort}\cdot E \doteq 0 \\
assignment & \text{wp.(x := f)}\cdot E \doteq E[x := f] \\
sequential\ composition & \text{wp.(r \circ r')}\cdot E \doteq \text{wp.r.}\text{wp.r'}\cdot E \\
probabilistic\ choice & \text{wp.}(r \oplus r')\cdot E \doteq p \times \text{wp.r.E} + (1-p) \times \text{wp.r'}\cdot E \\
nondeterministic\ choice & \text{wp.}(r \mathbin{\parallel} r')\cdot E \doteq \text{wp.r.E} \sqcap \text{wp.r'}\cdot E \\
Boolean\ choice & \text{wp.}(\text{if } G \text{ then } r \text{ else } r')\cdot E \doteq [G] \times \text{wp.r.E} + [-G] \times \text{wp.r'}\cdot E \\
Iteration & \text{wp.}(\text{do } G \rightarrow r \text{ od})\cdot E \doteq (\mu X \cdot [-G] \times E + [G] \times \text{wp.r.X}) \\
\end{array}
\]
\(E\) is an expectation in \(\mathcal{E}S\), and \(f\) is a function of the state, and \(\sqcap\) is pointwise minimum. The real \(p\) is restricted to lie between 0 and 1, and the term \((\mu X \ldots)\) refers to the least fixed point with respect to \(\leq\), which we lift to real-valued functions. Commands are ordered using \textit{refinement}, so that more refined programs improve probabilistic results, thus \(P \sqsubseteq Q\) \iff \((\forall E \in \mathcal{E}S \cdot \text{wp.P.E} \leq \text{wp.Q.E})\); note also that the \textit{monotone} property of \text{wp} is such that if \(E \leq E'\) then \(\text{wp.P.E} \leq \text{wp.P.E'}\), where \(P, Q\) are program commands and \(E, E'\) are expectations.
\[\text{Figure 1: Structural definitions of wp for the pGCL.}\]
\[\text{times on termination as:}\]
\[
\text{wp.do } G \rightarrow \text{Prog}; n := n+1 \text{ od.}[n \leq k],
\]
where \(n\) is a fresh variable, not occurring in \(\text{Prog}\). Informally, if \(n\) is initialised to 0 before the execution of the loop and is incremented after each execution of \(\text{Prog}\), this expresses the (minimum) probability that its value on exiting the loop does not exceed \(k\). When no nondeterminism is present the expression in (2) computes an exact bound for expected performance; when it is present it computes the greatest lower bound. However upper bounds can be calculated using a maximum interpretation of nondeterminism but we do not discuss that interpretation here.
In this section we have summarised an expectation transformer approach to probabilistic semantics. In many cases, especially for performance, the exact analysis of the system in this style is impractical; an alternative to model checking, however this is not viable for very large or infinite systems. Predicate abstraction is a popular approach to approximating such programs, and in the next section we develop the expectation transformer approach to predicate abstraction for probabilistic programs.
\section{Abstract expectation transformers}
Predicate abstraction is a standard technique for defining abstractions of transition systems. In this section we will show how to define it for probabilistic programs using expectation transformers. The approach is inspired by Ball’s formalisation of predicate abstraction for standard sequential programs using weakest precondition semantics \cite{Ball:1991}.
Let \(\Phi\) be a (finite) set of predicates over the state space \(S\). The standard predicate abstraction over \(\Phi\) is induced by the equivalence class:
\[
s \sim_{\Phi} s' \iff (\forall \phi \in \Phi \cdot \phi.s = \phi.s').
\]
Given a transition system \(T\) over \(S\), the abstract transition system \(T/\sim_{\Phi}\) takes the equivalence classes given by \(S/\sim_{\Phi}\) as the state space, and their transitions \(s \rightarrow t\) in \(T/\sim_{\Phi}\) provided that there exists a
transition $s \rightarrow t$ in $T$. The probabilistic generalisation is somewhat more complicated to define. On the other hand the expectation transformer semantics characterises operational behaviour, and the approach we take here is to define the abstract transition system over $S/\Phi$ using a generalisation of Ball’s idea.
Let cubes,$\Phi$ be the (finite) set of (non trivial) minimal predicates formed by taking negations and conjunctions of predicates in $\Phi$. The set cubes,$\Phi$ corresponds to the (set of) equivalence classes $S/\sim_\Phi$, and represents the abstract state space of the abstraction induced by $\Phi$. Let cubed$_\Phi : \mathcal{E}S \rightarrow \mathcal{E}S$ be defined
$$\text{cubed}_\Phi.e \triangleq (\biguplus c \in \text{cubes}.\Phi \cdot (\biguplus \lambda [c] \leq e \cdot \lambda [c])) \cdot$$
(3)
We note that cubed$_\Phi.e$ is unique and would usually be a linear combination of the elements of cubes,$\Phi$, hence making it the sum of scaled cubes over the latter. Consequently, cubed.$e$ is the weakest approximation of $e$ with respect to the granularity expressible by conjunctions, negations and disjunctions in $\Phi$. We say that $e$ is cubed relative to $\Phi$ exactly when $e = \text{cubed}_\Phi.e$. Note that cubed.cubed.$e = \text{cubed}.e$ and that sums, maxima and minima of cubed expressions are still cubes, i.e.
$$3(a) \quad \text{cubed}.(\text{cubed}.e + \text{cubed}.e') = (\text{cubed}.e + \text{cubed}.e') \cdot$$
$$3(b) \quad \text{cubed}.(\text{cubed}.e \sqcup \text{cubed}.e') = (\text{cubed}.e \sqcup \text{cubed}.e') \cdot$$
$$3(c) \quad \text{cubed}.(\text{cubed}.e \sqcap \text{cubed}.e') = (\text{cubed}.e \sqcap \text{cubed}.e') \cdot$$
Definition 1. Given a pGCL program Prog, and a set of predicates $\Phi$, and expectation $e$ over $S$ we define the abstract weakest expectation relative to $\Phi$ as:
$$\text{wp}_\Phi.\text{Prog}.e \triangleq \text{cubed}_\Phi.\text{wp}.\text{Prog}.e \cdot$$
We write Prog$_\Phi$ for the corresponding abstract program operating over the abstract system $S/\Phi$. This implies that Prog$_\Phi$ is determined by $\text{wp}_\Phi.\text{Prog}$.
As an example, consider the program inc at (11) operating under arithmetic modulo 4. The underlying state space is defined by $0 \leq x < 4$; consider now the set $\Phi$ consisting of the single predicate $x = 0 \lor x = 2$; the set of cubes cubes.$\Phi \triangleq \{(x = 0 \lor x = 2), (x = 1 \lor x = 3)\}$, implying that the induced predicate abstraction has two states. We can see now that
$$\text{wp}_\Phi.\text{inc}.[x = 1 \lor x = 3] = [x = 0 \lor x = 2]/2 \cdot$$
and
$$\text{wp}_\Phi.\text{inc}.[x = 0 \lor x = 2] = [x = 1 \lor x = 3]/2 \cdot$$
which is consistent with the abstraction in Fig. 2 where each abstract state has a probability of at least 1/2 of being transformed to the other state, with the remaining probability being assigned to a nondeterministic update.
The next lemma sets out some properties of the abstract expectation transformer.
Lemma 1. Let Prog,Prog’ be programs, $\Phi,\Phi’$ sets of predicates, and $e,e’$ expectations and $\alpha$ a real. The following inequalities apply:
\begin{align*}
(1) \quad \text{wp}_\Phi.\text{Prog}.e & \leq \text{wp}.\text{Prog}.e \\
(2) \quad \text{wp}_\Phi.\text{Prog}.e & \leq \text{wp}_\Phi.\text{Prog}.e \\
(3) \quad \text{wp}_\Phi.\text{Prog}.e + \text{wp}_\Phi.\text{Prog}.e’ & \leq \text{wp}_\Phi.\text{Prog}.(e + e’) \\
(4) \quad \alpha \times \text{wp}_\Phi.\text{Prog}.e & = \text{wp}_\Phi.\text{Prog}.(\alpha \times e) \\
(5) \quad (\text{wp}_\Phi.\text{Prog}.e - 1) \sqcup 0 & \leq \text{wp}_\Phi.\text{Prog}.(e - 1 \sqcup 0) \\
\end{align*}
Proof: The inequalities and equalities all follow from arithmetic and Def. 1.
The transition system on the left represents the program `inc` over the state space defined by $0 \leq x < 4$, using arithmetic modulo 4. Each solid black arrow occurs with probability $1/2$. The transition system on the right is the abstraction based on $\Phi = \{ x = 0 \lor x = 2 \}$. Here we can see non-determinism (indicated by dotted lines) is introduced after any transition which divides the value of $x$ by 2.
Figure 2: The transition system for `inc` and an abstraction.
Lem. 1 confirms our intuition that (1) the properties measured with respect to the abstraction are no more than with respect to the original program; (2) finer-grained abstractions give more accurate results and (3,4,5) $wp_\Phi.Prog$ corresponds to a well-defined probabilistic transition system [8].
For standard transitions systems (with no probability) an abstract system $Prog\Phi$ is determined directly from the control structure and assignment statements. This corresponds to $wp_\Phi$ distributing through the program operators. For probabilistic systems this is not the case. For example, the program $inc;inc$ (with addition modulo 4) we may compute $3[x = 0 \lor x = 2]/4 \leq wp_\Phi.(inc;inc).[x = 0 \lor x = 2]$, whereas $[x = 0 \lor x = 2]/4 = wp_\Phi.inc.(wp_\Phi.inc).[x = 0 \lor x = 2]$, implying that the non-determinism introduced at each abstract transition will increase the inaccuracy. Comparing with Fig. 2 we see that non-determinism is introduced at each abstract transition, and this could be resolved in the abstract system in such a way that there is only $1/4$ chance of returning to the initial abstract state.
The following lemma shows that $wp_\Phi$ only distributes through non-determinism, and only sub-distributes through sequential composition and probabilistic choice.
**Lemma 2.** Let $Prog, Prog'$ be programs, $\Phi, \Phi'$ sets of predicates, and $e, e'$ expectations. The following inequalities apply:
\[
\begin{align*}
(4) \quad wp_\Phi.(Prog \parallel Prog').e &= wp_\Phi.Prog.e \sqcap wp_\Phi.Prog'.e \\
(5) \quad wp_\Phi.Prog.(wp_\Phi.Prog').e &\leq wp_\Phi.(Prog;Prog').e \\
(6) \quad wp_\Phi.(Prog \oplus wp_\Phi.Prog').e &\leq wp_\Phi.(Prog \oplus Prog').e
\end{align*}
\]
**Proof:** The inequalities and equalities all follow from arithmetic and Def. 1.
Lem. 2 implies that whenever non-determinism is introduced, the analysis of a program abstracted at each program statement could be too coarse to verify a desired quantitative threshold. This is not a problem when the abstraction does not introduce non-determinism. Consider the program
\[
twoFlip \triangleq x := 0 \oplus x := 1 ; y := 0 \oplus y := 1 ,
\]
and the set of predicates $\Phi \triangleq \{ x = y, x \neq y \}$. The resulting transition system over the state space defined by $x$ and $y$ is set out in Fig. 3 together with the abstraction induced by this $\Phi$.
The transition system (labelled with probabilities) on the left represents the program `twoFlip` over the state space defined by variables `x` and `y`, each of which can take 0 or 1 value in the states `(x,y)`. Each branch is executed with the probability that it occurs; only the transitions from `x = y = 0` are illustrated, with transitions from the remaining states similarly calculated. The transition system on the right represents the abstraction which only keeps track of whether `x` and `y` are equal or not. Since no nondeterminism is introduced, properties at that level of granularity can be accurately calculated using this abstraction.
Figure 3: The transition system for `twoFlip` and an abstraction.
Observe how no nondeterminism has been introduced in this abstraction — since indeed $wp_{\Phi}.(\text{twoFlip;twoFlip}) = wp_{\Phi}.(\text{twoFlip}) \cdot wp_{\Phi}.(\text{twoFlip})$. Intuitively this tells us that properties which can be stated at the granularity of $\Phi$ can be computed accurately from its corresponding abstraction. In the next section we formalise our intuition using expectation transformers.
4 Information-preserving abstractions and expected time to terminate
In this section we introduce “information-preserving” abstractions and study how they apply to the computation of exact bounds on performance-style properties of probabilistic programs.
As we saw above, an abstraction which does not introduce nondeterminism preserves the exact behaviour of the program at the granularity of the chosen set of predicates. Programs which do not exhibit nondeterminism or aborting behaviours satisfy the special properties that:
\[
\begin{align*}
wp_{\Phi}.\text{Prog.}(e + e') &= wp_{\Phi}.\text{Prog}.e + wp_{\Phi}.\text{Prog}.e' \\
wp_{\Phi}.\text{Prog.}(1 - e) &= 1 - wp_{\Phi}.\text{Prog}.e
\end{align*}
\]
for any deterministic pGCL program command `Prog`, set of predicate $\Phi$, and expectations $e, e'$. The next definition formalises that idea in terms of expectation transformers.
**Definition 2.** Given a deterministic program `Prog` and a set of predicates $\Phi$, we say that the abstraction induced by $\Phi$ is information-preserving if:
\[
wp_{\Phi}.\text{Prog.}[c] = \text{wp.Prog.}[c],
\]
for all $c \in \text{cubes.}\Phi$.
To see Def. 2 in action, observe that
wp.\(inc\).[x = 0 \lor x = 2]
= \([x = 0 \lor x = 3]/2 + [x = 1]
\neq \([x = 1 \lor x = 3]/2
= \text{wp}.\(inc\).[x = 0 \lor x = 2],
implying the introduction of a nondeterministic branch at the abstract state corresponding to \(x = 1 \lor x = 3\).
A more efficient way to check for information-preservation is simply to check that \(\text{wp}.\(Prog\).[\Phi]\) is cubed for all \(\Phi \in \Phi\); the next lemma shows that this is sound.
**Lemma 3.** Let \(Prog\) be a deterministic (probabilistic) program, and let \(\Phi\) be a set of predicates. If \(\text{wp}.\(Prog\).[\Phi]\) is cubed for all \(\Phi \in \Phi\) then the abstraction induced by \(\Phi\) is information-preserving.
**Proof:** We need to show that \(\text{wp}.\(Prog\).[c] = \text{wp}_\Phi.\(Prog\).[c]\) for all \(c \in \text{cubed.}\Phi\). Note that each such \(c\) is generated from negations and conjunctions, so all we need show is that for predicates \(\psi, \psi'\) such that \(\text{wp}.\(Prog\).[\psi]\) and \(\text{wp}.\(Prog\).[\psi']\) are cubed, then so too are \(\text{wp}.\(Prog\).(1 - [\psi])\) and \(\text{wp}.\(Prog\).[\psi \land \psi']\).
The result follows since \(\text{wp}.\(Prog\).[\psi \land \psi'] = (\text{wp}.\(Prog\).[\psi] + \text{wp}.\(Prog\).[\psi'] - 1) \uplus 0\), and \(\text{wp}.\(Prog\).(1 - [\psi]) = 1 - \text{wp}.\(Prog\).[\psi]\), and the fact that sums and inverses of cubed expressions are still cubed.
As mentioned above, a key characterising property of information-preserving abstractions is that they generate no new nondeterminism. A probabilistic program exhibits no (demonic) nondeterminism if its expectation transformer semantics distributes addition. The next lemma shows this for information-preserving abstractions.
**Lemma 4.** Let \(Prog\) be a deterministic (probabilistic) program, and let \(\Phi\) be a set of predicates inducing an information-preserving abstraction on \(Prog\). The predicate transformer \(\text{wp}_\Phi.\(Prog\)\) is deterministic on cubed expectations.
**Proof:** The result follows by showing that \(\text{wp}.\(Prog\) = \text{wp}_\Phi.\(Prog\)\) on cubed expressions. Assume first that \(c, c' \in \text{cubed.}\Phi\), and that \(\lambda, \lambda'\) are reals. We reason as follows:
\[
\begin{align*}
\text{wp}_\Phi.\(Prog\).([\lambda \psi] + [\lambda' \psi']) & \leq \text{wp}.\(Prog\).([\lambda \psi] + [\lambda' \psi']) & \text{Lem.}\(\ref{lem:wp}\)(1) \\
& = \lambda \text{wp}.\(Prog\).[\psi] + \lambda' \text{wp}.\(Prog\).[\psi'] & \text{Prog is deterministic} \\
& = \lambda \text{wp}_\Phi.\(Prog\).[\psi] + \lambda' \text{wp}_\Phi.\(Prog\).[\psi'] \text{ Prog is information-preserving} \\
& \leq \text{wp}_\Phi.\(Prog\).([\lambda \psi] + [\lambda' \psi']) . & \text{Lem.}\(\ref{lem:wp}\)(3)
\end{align*}
\]
Observe finally that the equality generalises for expressions consisting of finite sums of cubes, and the fact that there are only finitely many distinct cubes whenever \(\Phi\) is finite.
In particular we can now see that information-preserving abstractions compute exact results for all cubed expressions:
**Corollary 1.** Let \(Prog\) be a deterministic (probabilistic) program, and let \(\Phi\) be a set of predicates inducing an information-preserving abstraction on \(Prog\). Then \(\text{wp}_\Phi.\(Prog\).[e] = \text{wp}.\(Prog\).[e]\) whenever \(e\) is cubed.
**Proof:** Follows since if \(e\) is cubed then it is a finite sum of scaled cubes, and by Lem.\(\ref{lem:wp}\) \(\text{wp}_\Phi.\(Prog\)\) distributes addition.
### 4.1 Computing abstractions component-wise
The above notions assume that the abstraction is calculated wholesale on the program \(Prog\); in practice it may be more efficient to calculate the abstraction by computing it relative to, and on smaller components of the program, however as Lem.\(\ref{lem:wp}\) (5,6) indicate, additional inaccuracies can creep in wherever the abstraction is computed from program components.
Fortunately this does not occur in the case of information-preserving abstractions: Lem. 4 is key to verifying that information-preserving abstractions are determined from their components alone, provided that they themselves are also information-preserving. In practical terms this means that in a transition system, provided each transition preserves the information, then so will the abstraction. In our predicate transformer framework, we need to show that wpφ distributes sequential composition and probabilistic choice.
**Lemma 5.** Let Prog, Prog′ be deterministic (probabilistic) programs, and let Φ be a set of predicates inducing an information-preserving abstraction on each. The following inequalities apply:
\[(5') \quad \wp_{\Phi}.\text{Prog}(\wp_{\Phi}.\text{Prog}') = \wp_{\Phi}.(\text{Prog};\text{Prog}')\]
\[(6') \quad \wp_{\Phi}.\text{Prog} \oplus \wp_{\Phi}.\text{Prog}' = \wp_{\Phi}.(\text{Prog} \oplus \text{Prog}').\]
**Proof:** Follows easily from Lem. 4 since \(\wp_{\Phi}.\text{Prog}\) and \(\wp_{\Phi}.\text{Prog}'\) are both cubed expressions.
### 4.2 Computing average performance
Significantly, we can now compute expected performance profiles exactly from the abstraction.
**Lemma 6.** Let Prog be a deterministic program, and information-preserving with respect to Φ, and suppose that \(G. (n ≤ k) ∈ Φ\), where k is an integer. The following equalities hold:
\[\wp.(\text{do } G → \text{Prog}; n := n + 1 \text{ od}).[n ≤ k] = \wp.(\text{do } \hat{G} → \text{Prog}_φ; n := n + 1 \text{ od}).[n ≤ k],\]
where \(\hat{G}\) represents the abstraction of G in \(S/ ~_{\Phi}\).
**Proof:**
Let \(N \equiv \wp.(\text{do } G → \text{Prog}; n := n + 1 \text{ od}).[n ≤ k],\) and \(N_φ \equiv \wp.(\text{do } \hat{G} → \text{Prog}_φ; n := n + 1 \text{ od}).[n ≤ k].\) By Lem. 7(1), and monotonicity of the programming language Fig. 1, we see that \(N_φ ≤ N\). To show that \(N ≤ N_φ\) we note that \(N\) and \(N_φ\) are both least fixed points of monotone expectation-to-expectation functions. We use the least fixed point property of functions over partially-ordered sets, i.e. that if \(f.x ≤ x\) then \(μ.f ≤ x\). Applied to \(N\) and \(N_φ\) we establish that \(N_φ\) satisfies the least fixed point equation for \(N\) as follows:
\[
N_φ = \wp.(\text{do } \hat{G} → \text{Prog}_φ; n := n + 1 \text{ od}).[n ≤ k],
\]
Thus the result now follows since \(N\) is the least fixed point of the function \((λx. [G] × n + [−G] × \wp.\text{Prog}.x).\)
For the “see below” part, note that \(N_φ\) is itself a fixed point, satisfying: \(N_φ = [G] × [n ≤ k] + [−G] × \wp_{\Phi}.\text{Prog}_φ.\text{N}_φ.\) It now follows that \(N_φ\) is cubed since \(\wp_{\Phi}.\text{Prog}_φ\) is, for any expression \(e\).
More generally exact bounds can be computed even when the program exhibits finitely-branching nondeterminism.
**Corollary 2.** Let Prog₁...Progₘ be deterministic and information-preserving with respect to Φ. Let \(G ∈ Φ\), and \(n\) a fresh variable. The following equalities hold:
\[\wp.\text{Prog}.(\text{do } G → (\text{Prog}_1 [\ldots [\text{Prog}_m \text{ od}).[n ≤ k] \]
\[= \wp.\text{Prog}.(\text{do } G → (\text{Prog}_1 [\ldots [\text{Prog}_m [\text{od}.[n ≤ k].\]
**Proof:** The proof is similar to Lem. 6 since nondeterminism distributes by Lem. 7(4).
The significance of Cor. 2 is that whenever the abstraction is known to be information-preserving component-wise over a program (or transition system), then exact performance can be carried out on the abstraction. An important class of such programs are the so-called “data independent” systems, to which we turn in the next section.
4.3 Data independence
A program is said to be data independent (with respect to a data type $X$) \cite{14} if it cannot perform operations involving specific values of the type: specifically it can only input, output, store and make comparisons using any relational operator $\Theta \in \{=, <, \leq, >, \geq, \ldots\}$. Wolper points out that many distributed protocols fall into this category — he shows that such systems can be model checked accurately. In fact, if we extend this informal definition to probabilistic programs such that all probabilistic choices are constants, then our results above imply that there is an abstraction which can be used to compute performance properties exactly, namely the abstraction induced by predicates $\Psi \equiv \{x \Theta y : \forall x, y \text{ program variables of same type}\}$. We use this intuition to define a simple characterisation of data independent programs: they are the programs which are information-preserving with respect to $\Psi$ (with informal definition above).
**Definition 3.** Let $Prog$ be a deterministic pGCL program with variables $x_1 \ldots x_m$. We say that $Prog$ is data independent with respect to $x_1 \ldots x_m$ provided that $Prog$ is information-preserving with respect to the abstraction induced by $\Psi$, where $\Psi$ is the set of predicates containing all expressions of the form $x_i \Theta y_j$ for all $1 \leq i, j \leq m$.
Note that this characterisation of data independence can be generalised to programs $Prog_1 \parallel \ldots \parallel Prog_n$ which exhibit nondeterministic behaviour by ensuring that the deterministic components $Prog_i$ comply with Def. 3. Note that this definition shares some similarities with Wolper’s denotational characterisation \cite{14}, in that Def. 3 captures the idea that properties expressible at the granularity of $\Phi$ are shared by both $Prog_\Phi$ and $Prog$. It does not deal with general types however, as does Lazic \cite{16}.
With Def. 3 we can now conclude that data independent probabilistic programs have abstractions which preserve performance bounds.
**Lemma 7.** Let $Prog$ be a data independent program. Then the expected number of iterations do $G \rightarrow Prog$ od may be computed exactly using the abstract program $Prog_\Psi$ whenever $G \in \Psi$, where $\Psi$ is defined in Def. 3.
The practical implication of Lem. 7 (which follows directly from Lem. 6 and Cor. 2) is that performance (and correctness) of data independent programs can be analysed exactly using model checking. In the next section we give an example to illustrate this idea.
5 Case study: Rabin’s distributed consensus
We illustrate the effectiveness of our technique on the Rabin’s choice-coordination problem \cite{13}. The state space generated on execution of the algorithm is potentially infinite hence limiting the scope of model checking on verifying liveness properties (such as termination conditions) relating to its overall performance. As we will see, even though the algorithm is not quite data independent, there does exist an information-preserving abstraction demonstrating that exact numerical analysis is still possible on its performance.
5.1 Informal description
A group of tourists are to decide between two meeting places (which are not of interest to us). A major constraint is that they may not communicate as a group; nor is there a central “authority” (e.g. a tour guide) whose decision overrides theirs.
Each tourist carries a notepad on which he will write various numbers; outside each of the two potential meeting places is a noticeboard on which various messages will be written. Initially the number zero appears on all the notepads and on the two noticeboards.
Each tourist decides independently (demonically) which meeting place to visit first, after which he strictly alternates his visits between them. At each place he looks at the noticeboard, and if it displays “here”, he goes inside. If it does not display “here” it will display a number instead, in which case the tourist compares that number $K$ with the number $k$ on his notepad and does one of the following:
- if $k < K$ — The tourist writes $K$ on his notepad (erasing $k$), and goes to the other place.
- if $k > K$ — The tourist writes “here” on the noticeboard (erasing $K$), and goes inside.
- if $k = K$ — The tourist chooses $K'$, the next even number larger than $K$, and then flips a coin: if it comes up heads, he increases $K'$ by a further one. He then writes $K'$ on the noticeboard and on his notepad (erasing $k$ and $K$), and goes to the other place.\(^2\)
A key characterisation of the Rabin’s algorithm, which has also been proved elsewhere \[^{[8]}\] is that, on termination all the tourists’ will converge at the same meeting place, and that happens with probability 1. However it is not always the case that an “observer” can witness every state of the program that will lead to termination. For example, it is possible that the tourists will forever (according to an observer) keep updating their notepads and the noticeboards without deciding on a meeting place. This enforces an unbounded state behaviour on the algorithm. Nonetheless, given the unbounded state nature of the algorithm, our theoretical results still permit us to study a suitable performance attribute of the system: the expected number of rounds (or steps) of the protocol until termination (analogous to convergence).
5.2 A pGCL snapshot of the Rabin’s algorithm
Fig. 4 (on the next page) gives an overview of the Rabin’s choice-coordination problem in the pGCL. We call the two meeting places “left” and “right” as we discuss it and refer to the notations\(^3\) accordingly. Bag $lout$ ($rout$) is the bag of numbers held by tourists waiting to look at the left (right) noticeboard; bag $lin$ ($rin$) is the bag of numbers held by tourists who have already decided on the left (right) alternative; number $L$ ($R$) is the number on the left (right) noticeboard. Initially there are $A$ ($B$) tourists on the left (right), all holding the number zero; no tourist has yet made a decision, and both noticeboards show zero.
Execution is as follows: if some tourists are still undecided (so that $lout$ ($rout$) is not yet empty), select one: the number he holds is $l$ ($r$). If some tourist has already decided on this alternative (so that $lin$ ($rin$) is not empty), this tourist does the same; otherwise any of the three possibilities discussed above is executed.
5.3 Computing average performance of the Rabin’s algorithm
In this section we discuss properties of the Rabin’s algorithm sufficient for an analysis of its average performance. Since the unbounded state nature of the algorithm limits the scope of model checking on the
---
2 For example if $K$ is 2 or 3, then $K'$ becomes 4 and then possibly 5.
3 \([...]\) — bag (multiset) brackets; \(\square\) — empty bag; \(\lfloor n \rfloor^N\) — bag containing $N$ copies all of value $n$; take $n$ from $b$ — a program command which chooses an element demonically from non-empty bag $b$, assigns it to $n$ and removes it from $b$; add $n$ to $b$ — add element $n$ to bag $b$; $n$ — the “conjugate” of $n$, it is $n + 1$ if $n$ is even and $n - 1$ if $n$ is odd; $\#b$ — the number of elements in a bag $b$.
An expectation transformer approach to predicate abstraction
\[ l_{\text{out}}, r_{\text{out}} \triangleq [[0]]^A, [[0]]^B; \]
\[ l_{\text{in}}, r_{\text{in}} \triangleq \square, \square; \]
\[ L, R \triangleq 0, 0; \]
\textbf{do} \ l_{\text{out}} \neq \square \rightarrow \textbf{take} \ l \text{ from } l_{\text{out}}; \]
\textbf{if} \ l_{\text{in}} \neq \square \text{ then add } l \text{ to } l_{\text{in}} \text{ else } \]
\text{if} \ l_{\text{in}} \neq \square \text{ then add } l \text{ to } l_{\text{in}} \text{ else } \]
\text{if} \ r_{\text{in}} \neq \square \text{ then add } r \text{ to } r_{\text{in}} \text{ else } \]
\text{fi} \]
\textbf{fi} \]
\textbf{od} \]
Figure 4: The Rabin’s choice coordination algorithm in the pGCL (adapted from [8]).
original system, we must therefore compute a suitable abstraction prior to performance analysis. Nevertheless, with our proposed technique, it is possible to defeat the overhead incurred by model checking the unbounded state system just by model checking its information-preserving abstraction.
One performance property of interest is captured by computing the minimum probability \( P_{\text{min}} \), that within a finite number of steps \( T \), the tourists eventually converge at the same meeting place on termination. In the logic PCTL [6], directly supported by the PRISM tool, we express this property as
\[ P_{\text{min}} = \exists [\text{true} \ U \leq T \ (\#l_{\text{in}} = N) \ | \ (\#r_{\text{in}} = N)], \quad (5) \]
where \( N \) represents the total number of tourists who will initially decide on where to meet \( i.e. N = A + B \).
Similarly, with the reward structures [12] of the PCTL we compute the expected number of rounds of the protocol until termination. Again, this will be done on the protocol’s abstraction using the specification:
\[ R_{\text{min}} | R_{\text{max}} = \exists [F \ (\#l_{\text{in}} = N) \ | \ (\#r_{\text{in}} = N)]. \quad (6) \]
The parameters \( R_{\text{min}} \) and \( R_{\text{max}} \) respectively represent the expected minimum and maximum rewards (expected number of rounds) until the tourists eventually converge at the same meeting place. We note that states where the tourists have not yet met the convergence condition are worth a reward value of one.
In the sections that follow, we explain how we identify essential behaviours of the algorithm that will permit the construction of an information-preserving abstraction upon which the analysis can be performed.
5.4 An information-preserving abstraction
As earlier stated, even though the Rabin’s algorithm has unbounded state behaviour, it is not data independent since the probabilistic update increments the variables \( L, R \ etc. \) However there is still an
information-preserving abstraction, which we will now describe.
Observe that although the noticeboard values are incremented, they always maintain $|L - R| \leq 2$. In terms of the algorithm, the only information that needs to be preserved is the value $L - R$ and whether $L, R$ are odd or even. Finally, the relative values of the tourists’ numbers to $L$ and $R$ also need to be recorded, as well as their location. This generates an information-preserving abstraction. In practice, we characterise the relationship between the noticeboard values using a fresh variable we call $slot$, which can only take values in $\{0, 1, 2\}$ — since the noticeboard values and hence the notepad values can only lie in one of these slots for any given state of the system. We define the slot variable as follows and interpret transitions in the abstract state with respect to the slot values:
$$
slot \triangleq \begin{cases}
0 & \text{if } L = R \\
1 & \text{if } L = R - 2 \lor R = L - 2 \\
2 & \text{if } L = \overline{R}.
\end{cases}
$$
In the section that follows, we explain the performance results derived from the information-preserving abstraction. The results nevertheless give a precise summary of the performance of the original system.
### 5.5 Experimental results
We model the abstract behaviour discussed above for the base cases of even and odd number of tourists ($N = 2, 3$) in the PRISM language, and similarly analyse the performance results as captured by the properties in (5) and (6), using the experimentation facility of the tool. A similar model construction and analysis for larger values of $N$ is also possible by repeating the same technique although very laborious.
Fig. 5 captures the performance characteristics of the information-preserving abstraction of the Rabin’s algorithm. It clearly establishes the termination property of the unbounded state system using just its abstraction: note that both graphs converge to probability 1. In the original unbounded state system, achieving this is practically impossible. See the original model in the compendium of case studies at [1].
We also observe (Fig. 6) that the expected minimum and maximum number of rounds until termination can be model checked, and hence nevertheless gives an exact bound on the number of steps required for the unbounded state system to terminate. Again, in the unbounded state system, the result of computing \( R_{\text{max}} \) for example is \textit{infinity}, which in the PRISM tool is interpreted to mean that it is not possible for a terminating (or convergence) condition to be reached.
6 Discussion
While some probabilistic program logics allow programs to be compared even at abstract levels, for example using the techniques in [5,3], the underlying logic of the pGCL supports the notion of program refinement and hence compositionality. This makes it easy to relate refinement over concrete states to their abstract counterparts and furthermore with the other probabilistic program logics, given any context.
Other approaches seek to use variations of counterexample guided predicate abstraction [10, 7] to automate finding sets of predicates which generate finer abstractions. One way to see the relationship with our approach would be to note that when an abstraction is observed to be information-preserving (according to Lem. 3 for example) then further refinement is unnecessary. Kwiatkowska et al. [11] propose an approach to estimate the accuracy of the analysis implied by any abstraction, confirming that for information-preserving abstractions the analysis is exact.
On the application level, one way to see the usefulness of our technique is in the recent research direction of linking proof-based verification with model checking for probabilistic systems [18, 19]. Since proof-based verification can cope with proofs over infinite state systems, a key challenge with this technique is then the identification and constructing of information-preserving abstractions upon which a model checking algorithmic verification can be performed. This is still an open problem.
7 Conclusion and future work
In this paper we have developed the theory of predicate abstraction for probabilistic programs within the framework of expectation transformers. We have similarly established a criterion to help discover when abstractions do not lose information especially for probabilistic programs; and we have demonstrated the applicability of the results to data independent programs (or at least their approximations).
Whilst our theoretical approach identifies when a set of predicates is information-preserving, it does not provide assistance for finding one. Even though we have computed the abstraction by hand, we quickly remark that applying the manual construction technique for \( N > 3 \) would seem a laborious task. Note that our technique results in a huge success for verifying the termination condition of the algorithm when compared with the concrete system as modeled in the compendium of case studies at the URL [1].
However, a future direction for this work would be to develop an automated strategy which would construct abstractions “on the fly”, given that our theoretical framework is rich enough to provide intuitions to identifying sets of suitable predicates to aid the construction of information-preserving abstractions.
Acknowledgement: The authors are grateful to the anonymous reviewers for their helpful comments.
References
[1] PRISM: Probabilistic Symbolic Model Checker. URL: http://www.prismmodelchecker.org/
[2] Tarski A. A Lattice-theoretic Fixpoint Theorem and its Applications. Pacific Journal of Mathematics, 5:285-309, 1955.
[3] Jou C. C. and Smolka S.A. Equivalences, Congruences, and Complete Axiomatizations for Probabilistic Processes. In J. Baeten and J. Klop, Editors, CONCUR 90 1st Int. Conf. on Concurrency Theory, Number 458 in LNCS Vol. 94:1-28, Springer, 1990.
[4] Fudenberg D. and E. Maskin. The Folk Theorem in Repeated Games with Discounting or with Incomplete Information. Econometrica, 1986.
[5] Larsen K. G. and Skou A. Bisimulation Through Probabilistic Testing. Information and Computation, 94:1-28, 1991.
[6] Hansson H. and Jonsson B. A Logic for Reasoning about Time and Reliability. In Formal Aspects of Computing, 6(5):512-535, 1994.
[7] Hermanns H., Wachter B., and Zhang L. Probabilistic CEGAR. In Proc. of the 20th international Conference on Computer Aided Verification, Princeton, NJ, USA, 2008.
[8] McIver A. K. and Morgan C. C. Abstraction, Refinement and Proof for Probabilistic Systems. Monographs in Computer Science, Springer, Verlag, 2004.
[9] Clarke E. M., Grumberg O., and Peled D. A. Model Checking. MIT Press, 1999.
[10] Kattenbelt M., Kwiatkowska M., Norman G., and Parker D. Abstraction Refinement for Probabilistic Software. In Proc. of 10th International Conference on Verification, Model Checking and Abstract Interpretation (VMCAI ’09), Springer, 2009.
[11] Kwiatkowska M., Norman G., and Parker D. Game-based Abstraction for Markov Decision Processes. In Proc. of 3rd International Conference on Quantitative Evaluation of Systems (QEST’06), pages 157-166, IEEE CS Press, 2006.
[12] Kwiatkowska M., Norman G., and Parker D. Stochastic Model Checking. In Proc. of SFM’07 vol. 4486 LNCS, 220-270, Springer, 2007.
[13] Rabin M. O. The Choice Coordination Problem. Acta Informatica vol. 17, 121 - 134, 1982.
[14] Wolper P. Expressing Interesting Properties of Programs in Propositional Temporal Logic. In Proc. of the 13th Annual Symposium on Principles of Programming Languages, pp. 184 - 193, ACM, 1986.
[15] Grimmett G. R. and Welsh D. Probability: An Introduction. Oxford Science Publications, 1986.
[16] Lazić R. A Semantic Study of Data Independence with Applications to Model Checking. DPhil Thesis, Oxford University Computing Laboratory, 1999.
[17] Ball T. Formalizing Counterexample-driven Refinement with Weakest Preconditions. Technical Report MSR-TR-2004-134, Microsoft Research, Redmond, WA 98052, USA, 2004.
[18] Ndukwu U. Quantitative Safety: Linking Proof-based Verification with Model Checking for Probabilistic Systems. In Proc. First International Workshop on Quantitative Formal Methods (QFM 2009), Eindhoven, Netherlands, 2009.
[19] Ndukwu U. Generating Counterexamples for Quantitative Safety Specifications in Probabilistic B. Submitted to the Journal of Logic and Algebraic Programming (JLAP) URL: http://web.science.mq.edu.au/~ukndukwu/counterexamples.pdf 2010.
[20] Dijkstra E. W. A Discipline of Programming. Prentice Hall International, Englewood Cliffs, N.J., 1976. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3454,
1
],
[
3454,
7120,
2
],
[
7120,
11673,
3
],
[
11673,
15453,
4
],
[
15453,
19221,
5
],
[
19221,
22112,
6
],
[
22112,
24440,
7
],
[
24440,
28379,
8
],
[
28379,
32016,
9
],
[
32016,
35484,
10
],
[
35484,
39319,
11
],
[
39319,
42066,
12
],
[
42066,
43758,
13
],
[
43758,
47535,
14
],
[
47535,
50688,
15
]
]
} |
df6c4e711aedc01f20e3ae6f335287499fa496c4 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 24,
"total-fallback-pages": 0,
"total-input-tokens": 68560,
"total-output-tokens": 16504,
"fieldofstudy": [
"Physics"
]
} | Topological correlations and breaking of fermionic antisymmetry of electrons in FQHE
Janusz Jacak
Department of Quantum Technologies,
Faculty of Fundamental Problems of Technology,
Wrocław University of Science and Technology,
Wyb. Wyspiąskiego 27, 50-370 Wrocław, Poland
(Dated: November 23, 2020)
Abstract
Highly nonlocal interparticle correlations in quantum Hall states of 2D charged system exposed to the perpendicular strong magnetic field are detailed by application of the commensurability condition upon path-integral quantization approach and examined by Monte-Carlo Metropolis simulations in perfect consistence with exact diagonalization of the Coulomb interaction in small models and with experimental data. In this way refined filling rate hierarchy in the lowest Landau level fully explains the experimentally collected features for FQHE at filling ratios predicted by the conventional composite fermion model as well as for those beyond the composite fermion model but also visible in the experiment. The trial wave functions for FQHE states are proposed using a systematic topological method revealing the different symmetry for different correlated states depending on filling fraction. A violation of fermionic antisymmetry for 2D electrons is evidenced for some filling rates related to specific correlations in FQHE.
Keywords: Monte-Carlo-Metropolis simulation, Commencurability condition in 2D, FQHE hierarchy, Composite fermions, Correlations in quantum Hall states
* [email protected]
I. INTRODUCTION
Since the discovery of the fractional quantum Hall effect (FQHE) \cite{1} and its description by the famous Laughlin function \cite{2,3}, the problem of the long-range interparticle correlations on the plane between electrons exposed to the perpendicular quantizing strong magnetic field is still puzzling as manifesting itself at only specially chosen electron densities corresponding to the so-called FQHE hierarchy, whereas in close vicinity of these selected filling ratios correlations completely disappear. Experimentally noticed hierarchy of FQHE states embraces up to date ca. 70 various fractional fillings, majority of them in the lowest Landau level (LLL) including its spin polarization \cite{4,5}. The similar correlations at some selected fractional fillings are observed also in higher LLs, though the hierarchy from the LLL is not repeated there \cite{6–10}. The structure of the FQHE hierarchy is robust against a crystalline structure of the material which has been recently confirmed in graphene monolayer \cite{11–14} and in bilayer graphene \cite{15–19}, though in graphene the different Landau level quantization is induced by pseudo-relativistic Dirac-like electron dynamics \cite{20}.
The correlations in FQHE in the LLL duplicate correlations at the completely filled LLs corresponding to IQHE at filling fraction \( \nu = 1, 2, \ldots \), which manifests itself in vanishing of the longitudinal resistivity \( R_{xx} \) and quantization of the Hall conductivity \( \sigma_{yx} = \frac{j_x}{U_y} = \frac{e^2 \nu}{h} \), as visible in experiments \cite{4}. The closeness between correlations corresponding to IQHE and FQHE is phenomenologically addressed upon the concept of composite fermions (CFs) \cite{21} to mapping of FQHE at fractional fillings of the LLL onto IQHE of completely filled consecutive LLs in reduced magnetic field screened by the mean field of auxiliary fluxes pinned to composite fermions \cite{22}. In this way the main line of CF hierarchy is established, \( \nu = \frac{n}{(q-1)n+1} \) \( (n = 1, 2, 3, \ldots, q\text{-odd integer}) \).
Nevertheless, the full insight into correlations in FQHE is still incomplete as there are observed many fractions out of the main line of CF hierarchy (e.g., \( \nu = \frac{4}{11}, \frac{5}{13}, \frac{3}{10}, \frac{3}{8}, \frac{5}{17}, \frac{6}{17}, \frac{7}{11}, \frac{4}{11}, \ldots \) \cite{4}. At theses fractions \( R_{xx} \) typically is reduced in comparison to insulating phase but does not vanish which is in opposition to vanishing of \( R_{xx} \) at filling fractions from the main CF series (cf. Fig. \ref{fig:1}). This corresponds apparently to distinct type of correlations at these extraordinary filling fractions. They cannot be referred to higher LL correlations of CFs in resultant magnetic field screened by the averaged field of flux tubes pinned to electrons as it has been argued for the main line of CFs. Instead, an attempt to explain \( \frac{4}{11} \) and \( \frac{5}{13} \) states as
the secondary generation of FQHE of the primary generation CFs is proposed in the paper [23], which resolves itself to hypothetical dressing with interaction of already dressed CFs. This needs, however, the vague assumption that the residual interaction of primary CFs is still strong at some selected filling fractions, whereas in very close vicinity of these filling rates this additional interaction disappears. This is in conflict with an assertion [22] that CFs are some kind of quasiparticles which cannot change their residual interaction by a very delicate shift of the electron concentration. Similarly, some earlier study of the unconventional CF states out of the main series were not conclusive, cf. Refs [24–26]. These troubles of the theory indicate that the structure of FQHE correlations goes beyond the concept of CFs efficient only for the main CF series. In the present paper we clarify this situation and explain the correlations of FQHE in topological terms both for CFs and beyond, in fully consistence with the experimental observations [4, 27, 28].
FIG. 1. Measured $R_{xx}$ in the fragment of the LLL in GaAs 2DEG (after [4]), in color are indicated fractions at which the similar level of $R_{xx}$ is achieved— its nonzero value indicates that not all electrons are involved in the correlated state, as for correlations given by (4) with $x > 1$
The CF fermion structure has been referred to dressing of ordinary electrons with Coulomb interaction in some analogy to quasiparticle formation frequent in solids [22]. The rigorous way to define quasiparticles (e.g., Landau quasiparticles in metals) is their definition as poles of the retarded single-particle Green function [29] which is unavoidably linked with the requirement of the mass operator continuity [29]. This condition is, however, not fulfilled for 2D Coulomb interacting electrons in strong magnetic field and called originally by Laughlin as the ‘particle separation quantization’ [2, 3] which precludes the rigorous
The FQHE observation was more recently supplemented by rich experimental data in graphene [11–14] revealing the different hierarchy of FQHE in higher LLs in comparison to the LLL thus in disagreement with CF predictions. In conflict with CF predictions are also recent observations of FQHE at even denominator fractions in bilayer graphene including the most pronounced one at $\nu = -\frac{1}{2}$ (in the LLL for holes in the valence band) [15]. Though in graphene one deals with so-called relativistic version of the Landau level structure [20], the FQHE in graphene is convergent with that one in the conventional semiconductor 2DEG. In particular, the incompressible FQHE state at $\nu = \frac{1}{2}$ has been observed formerly in conventional semiconductor bilayer Hall system [30, 31]. This state is distinct than the compressible Hall metal state observed in the monolayer system at $\nu = \frac{1}{2}$. The latter found an explanation in terms of CFs but the former one (in the bilayer system) not.
All these listed above features indicate that the correlations in FQHE are distinct at various filling rates and still need clarification beyond the phenomenological CF model. Noticeably, none of the hypothetical secondary CF (with strong residual interaction) hierarchy states [23] is found in graphene as of yet, probably due to their lower stability in comparison to states from the main CF line as is also visible in experiments in the conventional 2DEG [4, 32, 33].
The aim of the present paper is to summarize an alternative insight into correlated 2D Hall states which can be gained by utilization of the path integral quantization with topological methods [34, 35] taking advantage of nonlocal technique of braid group approach [36, 37] allowing for exploration of odd topology of 2D space. It is commonly approved that behind the FQHE stands topological oddness of planar correlations of interacting electrons in 2D charged multiparticle system at magnetic field. The FQHE correlations are nonlocally conditioned by the topological factors and cannot be reduced to only local effects of interaction.
In which manner the perpendicular magnetic field causes in a planar charged interacting multiparticle system the strongly nonlocal topological effect? The answer resolves itself to braid group analysis [34, 35, 37] demonstrating constraints imposed on the braid trajectories needed for the path integral quantization by the cyclotron effect [38]. The braids are homotopy classes of closed multiparticle trajectories (including particle indistinguishability) describing particle exchanges in the multiparticle configuration space. The configuration
space and braid trajectories is not referred to real dynamics of quantum particles (without trajectories at all). Braids are classical objects used in topology terms upon the precisely defined braid group construction adjusted to requirements of the path integral quantization \[34, 36, 37, 39, 40\]. The braid group approach is not a single particle model, it is an essentially multiparticle nonlocal theory especially dedicated to recognize topological interparticle correlations. One-dimensional unitary representations (1DURs) of braids define weights for Feynman path integral of nonohomotopic path sectors (enumerated just by the braid group elements). Various 1DURs of the braid group allow for distinction of various quantum statistics of the same classical particles: fermions, bosons, anyons, composite fermions, composite bosons and composite anyons \[34, 37, 39, 40\]. As the multiparticle configuration spaces are not simply-connected then their homotopy groups (the $\pi_1$ homotopy groups for multiparticle configuration space are called braid groups \[37\]) are nontrivial with also multiple 1DURs, especially rich in the case of 2D charged particles at magnetic field presence. The strong magnetic field discriminates particle trajectories in 2D and allows for braid particle exchanges exclusively when planar cyclotron orbits fit to particle separation \[38, 41\]. The resulting commensurability condition reproduces hierarchy of FQHE in full agreement with experimental observations. Related details are presented in the following paragraphs.
II. COMMENSURABILITY CONDITION AND FQHE HIERARCHY IN THE LLL
For the completely filled LLL, $\nu = \frac{N}{N_0} = 1$, one deals with the commensurability condition:
$$\frac{S}{N} = \frac{S}{N_0} = \frac{hc}{eB_0},$$
where $S$ is the surface of the 2D sample, $B_0$ is the magnetic field corresponding to the completely filled LLL in the planar system with $N = N_0$ electrons, $N_0 = \frac{B_0 S}{hc/e}$ is the degeneracy of LLs (for $B = B_O$) and $\frac{hc}{e}$ is the quantum of the magnetic field flux. The bare Landau kinetic energy $E_n = \hbar \omega_c(n + \frac{1}{2})$ with cyclotron energy $\omega_c = \frac{eB}{mc}$, here for $B = B_0$. The Eq. (1) states that the surface per single particle, $\frac{S}{N}$, equals to the cyclotron orbit size being the orbit size corresponding to single flux quantum, $\frac{hc}{eB_0}$. Despite the cyclotron orbits are meaningless quantumly, the surface of cyclotron orbit $\frac{hc}{eB_0}$ is still well defined and when this orbit fits to the surface per single particle $\frac{S}{N}$ we deal with the commensurability as given by
Above observations are especially important for the description of the multiparticle system in terms of the braid group. The so-called full braid group for \( N \) particle system on the plane \( \mathbb{R}^2 \) \[36, 37\] defines exchanges between classical particles (which positions can be associated, on the other hand, with arguments of a multiparticle wave-function—any choice of its argument, \( z_1, \ldots, z_N \), corresponds to certain distribution of \( N \) classical particles on the plane). The generators of the full braid group, \( \sigma_i, i = 1, \ldots, N - 1 \), describe exchanges of neighboring particles (\( i \)th with \( (i + 1) \)th at certain particle enumeration, arbitrary, however, due to particle indistinguishability imposed on the system by dividing of the configuration space of distinguishable particles by the permutation group \( S_N \) \[36, 37\]).
1DURs of the braid group (defined on the braid group generators \( \sigma_i \)) determine quantum particles (bosons, fermions, anyons, composite fermions, composite bosons and composite anyons \[37\]). Various 1DURs, \( e^{i\alpha_l}, \alpha_l \in [-\pi, \pi) \), \( l \) enumerates braids, define different unitary weights for nonhomotopic (not continuously linked) sectors of trajectories in the domain of the Feynman path integral \[34, 35\],
\[
I(z_1, \ldots, z_N, t; z_1', \ldots, z_N', t') = \sum_{l \in \pi_1(\Omega)} e^{i\alpha_l} \int d\lambda_l e^{iS[\lambda_l(z_1, \ldots, z_N, t; z_1', \ldots, z_N', t')]} / \hbar, \tag{2}
\]
where, \( I(z_1, \ldots, z_N, t; z_1', \ldots, z_N', t') \) is the propagator, i.e., the matrix element of the evolution operator in position representation which determines the probability of quantum transition from the point \( z_1, \ldots, z_N \) in time instant \( t \) to other point in the configuration space \( z_1', \ldots, z_N' \) in time instant \( t' \), \( d\lambda_l \) is the measure in the path space sector enumerated by braid group element \( l \in \pi_1(\Omega) \), \( \pi_1(\Omega) \) is the first homotopy group of the configuration space \( \Omega \) (it is just called the braid group), \( \Omega = (M^N - \Delta) / S_N \), \( M \) is 2D plane here, \( M^N \) is \( N \)-fold normal product, \( \Delta \) is the collection of diagonal points in the normal product (when at least two coordinates \( z_i \) coincide) subtracted in order to assure particle number conservation, the quotient structure by the permutation group \( S_N \) accounts for quantum indistinguishability of particles, \( S[\lambda_l(z_1, \ldots, z_N, t; z_1', \ldots, z_N', t')] \) is the classical action for the trajectory \( \lambda_l \) joining selected points in the configuration space \( \Omega \) between time instances \( t, t' \) and lying in \( l \)th sector of trajectory space. The whole space of trajectories is decomposed into disjoint sectors enumerated by braid group element index \( l \) because to any trajectory at arbitrary time instant between \( t \) and \( t' \) one can adjoin the closed trajectory loop—the braid from the
braid group. As braids are nonhomotopic thus resulted trajectories $\lambda_l$ are also nonhomotopic, i.e., they cannot be transformed one into another one in a continuous manner (in Fig. 2 an example of nonhomotopic paths with additional braid loops are visualised for 2-particle system). This discontinuous decomposition of the domain of the path integral into disjoint sectors (topologically inequivalent) precludes a definition of the path measure $d\lambda$ uniformly on the whole space of paths and for each sector the measure $d\lambda_l$ must be defined separately and finally the contributions of all sectors must be summed with unitary factors $e^{i\alpha_l}$ (unitarity is caused by the causality constraint). It has been proved that these unitary factors establish the 1DUR of the braid group. Distinct unitary weights in the path integral (i.e., distinct 1DURs of the braid group) determine different sorts of quantum particles corresponding to the same classical ones. Braids describe particle exchanges, thus their 1DURs assign quantum statistics. Equivalently, the 1DUR of a particular braid defines a phase shifts of the multiparticle wave function $\Psi(z_1, \ldots, z_N)$ when its arguments $z_1, \ldots, z_N$ (classical coordinates of particles on the plane) mutually exchange according to this braid (let us emphasize that in 2D these exchanges are not permutations).
FIG. 2. Example of nonhomotopic trajectories obtained by addition of various braids to 2-particle trajectory
Turning back to the condition (1), we notice that it guarantees particle exchanges of uniformly distributed $N$ particles along cyclotron braid trajectories. At magnetic field presence no other trajectories are admitted, thus for the definition of the braid group the commensurability condition (1) must be fulfilled. The braid group can be then implemented and the statistics can be defined. The system is quantumly correlated on whole plane and the correlation is expressed by the commensurability condition (1).
The condition (1) defines the correlation only when particles are uniformly distributed with fixed and constant separation. This is a case for triangle Wigner lattice of 2D elec-
trons on the plane being the classical stable distribution for repulsing electrons located on uniform jellium at $T = 0$. For free noninteracting particles the condition (1) does not define correlation because particle separation is not fixed in the gas and is not commensurate with the cyclotron orbit. The same happens when the temperature grows and the kinetic energy destroys the classical Wigner lattice—then the correlated IQHE is substituted by only completely filled LLL without any topological correlations. IQHE is the specific collective state for completely filled LLL. It is nonlocally correlated multiparticle state at low temperature when the kinetic energy is suppressed due the massive degeneracy of the LLL and the Wigner crystal distribution can be used as the classical model of the lowest energy state caused solely by interaction. When temperature reaches cyclotron energy $\hbar \omega_c$ this correlation would be disrupted.
If $N_0$ is shifted (by the magnetic field change) then the correlation (1) is destroyed, because for $N \neq N_0$ the size of cyclotronic orbits does not fit to interparticle separation and electrons cannot exchange one with another one. Similar correlations for more complicated commensurability situations can be, however, recognized resulting in FQHE which will be described below.
If the external magnetic field grows, $B > B_0$, then one deals with the fractional filling of the LLL $\nu < 1$ ($N < N_0$ as the degeneracy has changed suitably to $B$, $N_0 = \frac{BS e}{hc}$). One can determine the correlated states at fractional fillings generalizing the genuine pattern (1) assuring particle mutual exchanges despite the cyclotron orbits are too short to match neighboring particles at $B > B_0$. At fractional fillings of the LLL the cyclotron orbits $\frac{h}{eB}$ are smaller than $\frac{S}{N}$ (as $B > B_0$) and cyclotron orbits cannot reach neighboring particles, which precludes ordinary cyclotron braid exchanges. For establishing of any correlated state the particle exchanges are, however, necessary to define statistics of quantum particles. This can be achieved by multiloop cyclotron orbits and related braid exchanges with additional loops. Exclusively in 2D multiloop cyclotron orbits have larger size in comparison to singlelooped ones at the same magnetic fields [38, 41]. This very peculiar property of a planar charged system follows from the distribution of the external field $B$ flux $\frac{BS}{N}$ per particle among all loops of the multiloop cyclotron orbit all located in the same plane (contrary to 3D, where each loop can lie on different surface and for 3D multiloop orbit the transpassing flux grows with number of loops, whereas in 2D does not). The condition for commensurability (1) at
magnetic field $B > B_0$ attains thus the more general form:
$$\frac{BS}{N} = (q - 1) \frac{hc}{e} \pm \frac{hc}{ey},$$
where: $q$ is the number of loops of single cyclotron orbit, $q$ must be odd integer in order to ensure that the corresponding braid is a particle exchange. The braid generator with $n$ additional loops, $\sigma_i^n$, corresponds to $2n + 1 = q$ loop cyclotron orbit \[37, 38\] (at magnetic fields the braids in 2D are built form half-pieces of cyclotron orbits provided that these orbits accurately fit to neighboring particle separation at the uniform particle distribution).
In the above formula $x \geq 1$ (integer) indicates the commensurability of $q - 1$ single loops from $q$-loop cyclotron orbit to every $x$th particle on the plane (it follows from the relation $\frac{BS}{N/x} = \frac{hc}{e}$, which is the condition \[1\] for fraction $N/x$ of particles). $y \geq x$ (also integer) indicates the commensurability of the last loop of the $q$-loop orbit with every $y$th particle. The sign $\pm$ indicates the same or opposite (of eight-figure-shape) orientation of the last i.e., $q$th loop.
In Eq. \[3\] we note that its left-hand-side, $\frac{BS}{N} = \frac{N_0}{N} \frac{hc}{e}$, because $N_0 = \frac{BS}{hc/e}$. Hence, from \[3\] we obtain the following conditions for the general hierarchy of correlated states in the LLL describing the FQHE hierarchy,
$$\nu = \frac{N}{N_0} = \frac{x y}{(q-1)y \pm x}, \text{ for band electrons},$$
$$\nu = 1 - \frac{x y}{(q-1)y \pm x}, \text{ for band holes},$$
For $x = 1$ the hierarchy \[4\] gives the FQHE hierarchy in the LLL derived by Jain \[21\] upon the effective model of CFs defined as electrons associated with localized on classical particles of auxiliary field flux tubes with even number of flux quanta each. For $x > 1$ the general hierarchy \[4\] is beyond the ability of the CF model and gives ratios for FQHE observed in experiment outside the main CF hierarchy. The comparison with the experimental data is summarized in Fig. \[3\].
It is clear thus that CFs simulate by fixed flux quanta additional loops for cyclotron orbits in the simplest case of the commensurability ($x = 1$). Therefore CFs comprise nonlocal topological information on multiloop size in effective flux tubes imagined to be pinned to particles. The reduction of the external field by the average field of pinned to electrons flux tubes as assumed in CF model results in fact in enhancement of cyclotron orbit size what is actually needed for fulfilment of the braid commensurability requirement. The CF
FIG. 3. Comparison of the hierarchy (4) with all measured fractional filling rates for FQHE features in the LLL (spin polarized). The hierarchy series acc. (4) for several $y$ each are displayed, filling rates beyond the main CF hierarchy are shown in red (Hall metal state fraction $1/2$ is marked).
The general Hall metal hierarchy in the LLL has thus the form:
\[
\begin{align*}
\nu &= \frac{x}{q-1}, \quad \text{for electrons,} \\
\nu &= 1 - \frac{x}{q-1}, \quad \text{for holes.}
\end{align*}
\]
(5)
Note that Hall metal correlation can manifest itself at fractions not necessarily with even denominators (for \(x\) even, beyond the Jain CF concept), similarly as the hierarchy (4) displays fractions both with odd and even denominators in compliance with the experimental observations (cf. Figs 1 and 3). Some fractions are repeated in various lines of the general hierarchy (4). This fact reveals the possibility of various types of commensurability of multiloop cyclotron orbits with interparticle separation \(\frac{S}{N}\). The advantage of one commensurability over the others (alternative ones at the same filling ratio) is related with energy minimization, i.e., with the minimization of the Coulomb interaction being distinct for different commensurability patterns.
| \(q\) | \(x\) | \(y\) | filling ratios acc. hierarchy (4) |
|------|------|------|----------------------------------|
| 3 | 1 | 1...10 | \(\frac{1}{3}, \frac{2}{5}, \frac{3}{7}, \frac{4}{9}, \frac{5}{11}, \frac{6}{13}, \frac{7}{15}, \frac{8}{17}, \frac{9}{19}, \) \(\ldots\) |
| | | | \(\frac{10}{21}, \frac{2}{3}, \frac{3}{5}, \frac{4}{7}, \frac{5}{9}, \frac{6}{11}, \frac{7}{13}, \frac{8}{15}, \frac{9}{17}, \) \(\ldots\) |
| 5 | 1 | 1...5 | \(\frac{1}{5}, \frac{3}{13}, \frac{2}{7}, \frac{3}{5}, \frac{4}{7}, \frac{5}{9}, \frac{6}{11}, \frac{7}{13}, \frac{8}{15}, \frac{9}{17}, \) \(\ldots\) |
| 5 | 2 | 2...5 | \(\frac{2}{5}, \frac{3}{7}, \frac{4}{9}, \frac{5}{11}, \frac{3}{7}, \frac{5}{9}, \frac{6}{11}, \frac{1}{3}, \frac{5}{7}, \frac{8}{11}, \ldots\) |
| 5 | 3 | 3...5 | \(\frac{3}{5}, \frac{12}{23}, \frac{15}{22}, \frac{12}{23}, \frac{15}{22}, \frac{10}{21}, \frac{2}{3}, \frac{7}{13}, \frac{8}{23}, \frac{1}{13}, \frac{2}{17}, \ldots\) |
| 7 | 2 | 2...5 | \(\frac{3}{7}, \frac{5}{11}, \frac{4}{7}, \frac{5}{11}, \frac{2}{7}, \frac{3}{5}, \frac{4}{7}, \frac{5}{9}, \frac{1}{3}, \frac{7}{15}, \frac{9}{17}, \frac{11}{16}, \frac{3}{5}, \frac{7}{11}, \frac{9}{17}, \ldots\) |
TABLE I. Exemplary filling rates for FQHE in the LLL according to the hierarchy given by Eq. (4) for various types of commensurability of cyclotron multilooped orbits; fractions obtained for \(x > 1\) (i.e., out of main line of CF model) are marked in red, though some of them coincide with CF-like hierarchy as an alternative commensurability (indicated in blue).
III. WAVE FUNCTIONS FOR CORRELATED FQHE STATES AND MONTE-CARLO METROPOLIS ASSESSMENT OF AN ACTIVATION ENERGY
For the simple line of the hierarchy (4) with \(x = y = 1\), i.e., \(\nu = \frac{1}{q}\), \(q - odd\), the corresponding wave function has been given by Laughlin in the form [2]:
\[
\Psi_{q}(z_1, z_2, \ldots, z_N) = A \prod_{i,j,i>j}^{N,N} (z_i - z_j)^q e^{-\frac{\sum |z_i|^2}{l^2}},
\]
where \(z_i = x_i + iy_i\) is \(i\)th particle classical position (arguments of the quantum multiparticle wave function) on the complex plane, \(l = \sqrt{\frac{hc}{eB}}\) is the magnetic length, and the product \(\prod_{i,j,i>j}^{N,N} (z_i - z_j)^q\) is the Jastrow polynomial—the generalization of the Vandermonde polynomial at \(q = 1\), \(A\) is an appropriate normalization constant. The \(q\)-fold zero structure of the Jastrow polynomial keeps particles apart stronger than for Vandermonde polynomial, and thus diminishes the Coulomb interaction energy.
The function (6) transforms itself according to the selected 1DUR of the cyclotron braid
generated by $q$–loop cyclotron braid subgroup with generators $\sigma_i^q$. For the 1DUR of the full braid group given by $\sigma_i \rightarrow e^{i\alpha}$ with $\alpha = \pi$ (fermions) one can get the phase shift for the exchange of neighbors $i$th and $j = (i + 1)$th as $(z_i - z_j)^q \rightarrow (z_j - z_i)^q e^{iq\pi}$ as required by the 1DUR of the generator $\sigma_i^q \rightarrow e^{iq\pi}$ (note that for more distant particles $i$th and $j$th, their exchange braid is given by the braid group element $(\sigma_i \cdots \sigma_{j-2} \cdot \sigma_{j-1} \cdot \sigma_{j-2}^{-1} \cdots \sigma_i^{-1})^q$ with 1DUR representation again equal to $e^{iq\pi}$ consistent with the form of the Laughlin function). Let us emphasize also, that the braid group generated by $\sigma_i^q, i = 1, \ldots, N - 1$ is the subgroup of the full braid group generated by $\sigma_i, i = 1, \ldots, N - 1$ (these subgroup is called as the cyclotron braid subgroup $[37]$). The given above 1DUR of the cyclotron braid subgroup is the 1DUR of the full braid group, $\sigma_i \rightarrow e^{i\alpha}$, reduced to the subgroup.
For the hierarchy $[41]$ the generators (describing elementary exchanges) of the appropriate more complicated cyclotron braid subgroups are defined as follows (for $\pm$ in $[41]$):
\begin{equation}
\psi_i^{q;x,y,+} = \left(\sigma_i \cdot \sigma_{i+1} \cdot \cdots \cdot \sigma_{i+x-2} \cdot \sigma_{i+x-1} \cdot \sigma_{i+x-2}^{-1} \cdot \cdots \cdot \sigma_{i+1}^{-1} \cdot \sigma_i^{-1}\right)^q \cdot \left(\sigma_i \cdot \sigma_{i+1} \cdot \cdots \cdot \sigma_{i+y-2} \cdot \sigma_{i+y-1} \cdot \sigma_{i+y-2}^{-1} \cdot \cdots \cdot \sigma_{i+1}^{-1} \cdot \sigma_i^{-1}\right),
\end{equation}
and
\begin{equation}
\psi_i^{q;x,y,-} = \left(\sigma_i \cdot \sigma_{i+1} \cdot \cdots \cdot \sigma_{i+x-2} \cdot \sigma_{i+x-1} \cdot \sigma_{i+x-2}^{-1} \cdot \cdots \cdot \sigma_{i+1}^{-1} \cdot \sigma_i^{-1}\right)^q \cdot \left(\sigma_i \cdot \sigma_{i+1} \cdot \cdots \cdot \sigma_{i+y-2} \cdot \sigma_{i+y-1} \cdot \sigma_{i+y-2}^{-1} \cdot \cdots \cdot \sigma_{i+1}^{-1} \cdot \sigma_i^{-1}\right)^{-1},
\end{equation}
with 1DURs (for $\alpha = \pi$) $e^{iq\pi}$ (for $+$) and $e^{i(q-2)\pi}$ (for $-$) (with supplement of the above notation for $x(y) = 1, \sigma_i \cdot \sigma_{i+1} \cdot \cdots \cdot \sigma_{i+x-2} \cdot \sigma_{i+x-1} \cdot \sigma_{i+x-2}^{-1} \cdot \cdots \cdot \sigma_{i+1}^{-1} \cdot \sigma_i^{-1} = \sigma_i$). Examples of these braid generators are depicted in Fig. 4.
The related modification of the Jastrow polynomial in the Laughlin function $[6]$ is thus
as follows:
\[ \Psi_{q}^{x,y,+} (z_1, z_2, \ldots, z_N) = \]
\[ A_{N,N/x} \prod_{i,j=1; i<j \mod x + (j-1)x} (z_i - z_{i \mod x + (j-1)x})^{q-1} \]
\[ \times A_{N,N/y} \prod_{i,j=1; i<j \mod y + (j-1)y} (z_i - z_{i \mod y + (j-1)y}) e^{-N \sum_{i} \frac{|z_i|^2}{4l^2}}. \]
(8)
The above functions for the Jain-like hierarchy \((x = 1)\) attains the form,
\[ \Psi_{q}^{x=1,y,+} (z_1, z_2, \ldots, z_N) = \]
\[ A_{N,N} \prod_{i,j=1; i<j} (z_i - z_j)^{q-1} \]
\[ \times A_{N,N/y} \prod_{i,j=1; i<j \mod y + (j-1)y} (z_i - z_{i \mod y + (j-1)y}) e^{-N \sum_{i} \frac{|z_i|^2}{4l^2}}. \]
(9)
The functions (8) are proposed as the trial wave functions for correlated states for filling rates (4) for which elementary exchanges of particles are defined by braids (7) and generalize the Laughlin function (6) for the case when \(x, y > 1\).
The energy gain in the Laughlin state is due to the reducing of the Coulomb repulsion energy \( \sum_{i,j,i>j} \frac{e^2}{|z_i-z_j|} \). It is clear that the energy reducing with the function (8) is the weaker the higher \(x\) is (for the same \(q\) and \(y\)). It follows from the dilution of correlated particles for \(x > 1\) (the correlation concerns every \(x\)th electron only) as expressed in modified Laughlin-type function (8) by reducing of the domain of the product. This leads to the diminishing of the repulsion energy gain due to the averaging of the Coulomb energy, \( \sum_{i,j,i>j} \frac{e^2}{|z_i-z_j|} \), with the wave function (8) instead of (6) (or (9)) because \(q-1\) fold zero in these functions prevents approaching not all electrons in the case of function (8) but only its \(1/x\) fraction (opposite to the case of function (6) or (9) for which \(x = 1\)). Therefore more stable are states with
lower $x$. Thus states with $x = 1$ energetically prevail over states with $x > 1$ and are more stable.
To confront the activation energy values obtained from numerical exact diagonalization of electron interaction for different fillings corresponding to FQHE in small models [43], the numerical calculation of expectation value of interaction energy for newly proposed functions (8) and (9) were performed by the Monte Carlo integration method upon the Metropolis scheme [44-46]. The Monte Carlo Metropolis method [46] is especially effective in assessment of the multi-argument integrals (we have utilized this method to models with 200 particles in circular symmetry of boundary conditions). For 400-fold argument (for 200 particles on 2D plane) of the multiple integral the standard Monte Carlo procedure are ineffective but the required accuracy can be achieved by utilization of appropriately modified version originally formulated by Metropolis for large scale chemical thermodynamic modeling [46]. To quantum Hall systems the method has been adopted in [44] but ranged to only analysis of Laughlin type states for $\nu = \frac{1}{3}$ and $\frac{1}{5}$. We have developed the method of Metropolis also to other states from FQHE hierarchy and related to trial wave functions given by Eqs. (8) and (9). Effectiveness of the applied multi-variable integration method consists in searching of local maxima of the integrand expression by the self-organized optimal random walk. As the integrand is proportional to the density of probability of distribution of $N$ particles, the resulted pattern of local maxima distribution (probability distribution for particle positions) reveals correlations imposed on the multiparticle system by the wave function under examination. These patterns of correlations can be easy compared with respect to the expectation value of particle electric interaction. The final averaged value of Coulomb energy, specific for each trial wave function form, is obtained by averaging over $10^7$ repetitions of the maxima searching steps avoiding few thousands initial steps to suppress an initial configuration influence. Next the procedure to assess of the expectation value of interaction energy in the thermodynamic limit is performed repeating the calculus for different $N$ and by the linear extrapolation with respect to $\frac{1}{\sqrt{N}} \to 0$ [44] in the range for $N = 4$ up to 200 in our study. Some exemplary results revealing the very good consistence of the determined Monte Carlo Metropolis expectation value of interparticle interaction energy (including interaction with positive jellium) with the energy found by the exact diagonalization of the Coulomb interaction in small models for FQHE [43] are presented in Table II.
TABLE II. Comparison of energy values obtained by exact diagonalization and by Monte Carlo simulation for some exemplary filling fractions for FQHE (Monte Carlo Metropolis simulation for trial wave functions derived by the cyclotron braid group method given by Eqs. (8) and (9), for $N = 200$ particles).
| $q$ | $x$ | $y$ | hierarchy fraction, $\nu = N/N_0$ | energy from Monte Carlo simulation for functions, according to Eq. (8) | energy from exact diagonalization |
|-----|-----|-----|-----------------------------------|-------------------------------------------------|--------------------------------|
| 3 | 1 | 2 | $\frac{2}{5}$ | $-0.432677$ | $-0.432804$ |
| 3 | 1 | 3 | $\frac{3}{7}$ | $-0.441974$ | $-0.442281$ |
| 3 | 1 | 4 | $\frac{4}{9}$ | $-0.464744$ | $-0.474422$ |
| 3 | 1 | 5 | $\frac{5}{11}$ | $-0.451056$ | $-0.450797$ |
| 5 | 1 | 2 | $\frac{2}{9}$ | $-0.342379$ | $-0.342742$ |
| 5 | 1 | 3 | $\frac{3}{13}$ | $-0.348134$ | $-0.348349$ |
| 5 | 1 | 4 | $\frac{4}{17}$ | $-0.351857$ | $-0.351189$ |
Turning back to the commensurability condition (4) it should be commented that for multiloop cyclotron orbits, none of the loop cannot be featured in general, thus each loop may be accommodated to the particle separation independently. Thus, for $q$-looped orbit one would deal with the ordered series $x_1 \leq x_2 \leq \cdots \leq x_q$ simplified in (4) to $x_1 = \cdots = x_{q-1} = x$, $x_q = y$. Apparently, the Coulomb repulsion minimization prefers $x_1 = \cdots = x_{q-1} = x$ for which the minimization domain restriction (resulting in weaker interaction energy reducing) is more convenient than for distinct distributions of $x_i$. This explains the choice of the uniform behavior of $q - 1$ loops (i.e., $x_1 = \cdots = x_{q-1} = x$) but this is not a rule and for many fractions various energetically competitive commensurability opportunities might be considered in principle.
The another observation related to various types of correlation identified by the commensurability criterion agrees with experimental data for the longitudinal resistivity $R_{xx}$ (cf. Fig. 1), which is zero for states with all correlated particles (i.e., with $x = 1$), whereas the residual its value grows with $x > 1$ probably due to scattering on portion of non-correlated electrons.
IV. SYMMETRY OF WAVE FUNCTIONS FOR FQHE STATES AND VIOLATION OF FERMIONIC ANTISYMMETRY AT SOME FILLING FRACTIONS
1DURs of cyclotron braid subgroup generated by the commensurability condition (3) defines symmetry of the corresponding wave function for multiparticle correlated state referred to FQHE at particular filling rates (4) given by the commensurability constraint. The wave function must transform according to 1DUR of the braid when the argument of the wave function \( z_1, \ldots, z_N \) change position on the plane according to this particular braid. This symmetry property together with condition that the wave function in the LLL must be a holomorphic function (then uniquely defined by its nodes), the shape of the function can be determined up to an invariant to braid transformation term. This term must be the same one as for the gas system because the gas wave functions span the Hilbert space for wave functions with interactions. This invariant term has the form of exponent, \( \exp(-\sum_{i=1}^{N} |z_i|^2/4l^2) \). Thus, all the symmetry resolves itself to the polynomial part of the wave function as have been defined for various commensurabilities (3) by the expressions (8) or (9).
It must be emphasized that the arbitrariness in choice of 1DURs corresponding to a plethora of composite anyons (being at the presence of strong magnetic the generalization of ordinary anyons in planar system without the quantizing magnetic field) can be reduced by assumption of fermion type particles corresponding to original covering full braid group. In this way we may address only to composite fermions and their wave function symmetry (the name related to CFs may be justified by the limit \( x = 1 \) in the commensurability condition (3)), i.e., 1DUR= \( e^{i\pi} \) for original generators \( \sigma_i \) of the full braid group. Though the resulted 1DURs for cyclotron subgroups have the form \( e^{ip\pi} \) with \( p \) odd integer, the corresponding wave functions have not fermionic antisymmetry property in general. This is caused by the fact that new generators of corresponding cyclotron subgroups do not define simple exchange of function variables as it was in 3D space described by the permutation group. In 2D braid groups and their cyclotron subgroups are far more complicated than the permutation group. Therefore the wave function satisfying symmetry requirements imposed by the 1DUR of the cyclotron subgroup induced by the particular commensurability condition has not a simple fermionic antisymmetry property (as visible in expressions (8) and (9)). This feature is first time clearly explained in terms of presented above formalism but contains a significant exceptional conclusion on violation of fermionic antisymmetry property of electrons creating...
many FQHE states. This is a fundamental difference with respect to 3D correlated states described obligatory by the antisymmetric functions for fermions. In 2D we have proved that this antisymmetry can be violated.
It must be emphasized that we have thus elucidated an embarrassing fact upon the CF theory when the conjecture on the form of the corresponding wave function has been formulated by utilization of wave functions for completely filled higher LLs \([22]\). To avoid poles present in wave functions for LLs, some heuristic procedures were then applied to model finally the holomorphic multiparticle function for fractional filling of the LLL without any poles. Despite many attempts none rigorous method for this artificial procedure of projection onto LLL has been proved and the method was uncertain and frequently searched empirically by energy minimization. The procedure of this projection onto LLL violates, however in an uncontrolled manner, the fermionic antisymmetry of electrons described by the final function. Nevertheless, the rigorous definition of the wave function symmetry according to the appropriate 1DUR of the cyclotron braid subgroup generated by a specific commensurability condition solves the problem of symmetry of electrons at FQHE accurately and completely. Worth noting is the fact that it is a sole case of violation of fermionic antisymmetry of electrons in condensed matter. This significant and fundamental quantum property is born by also exceptional multiloop cyclotron topology and related nonlocal correlations exclusively in 2D space.
V. SUMMARY
In conclusion we state that the commensurability condition selects the filling ratios for correlated states which coincide very well with the experimentally observed FQHE hierarchy. All experimentally noted fractional fillings for incompressible states in the LLL are predicted theoretically in this way. Worth noting is that all experimentally observed fractions for FQHE beyond the so-called main CF hierarchy are reproduced by the braid group commensurability condition (in Fig. \(3\) twelve such fractions are indicated). Remarkably, all the experimentally observed filling fractions for FQHE and for Hall metal states are accounted for in the hierarchy given by Eqs. \((4)\) and \((5)\) in the LLL. The presented braid group topological approach to FQHE hierarchy is complete, i.e., the cyclotron braid subgroup appropriate to each selected fraction is defined by the explicit construction of relevant subgroup
generators in compliance with the specific commensurability criterion. The forms of generators allow then for shaping of the generalization of Laughlin function polynomial multiplier which must transform according to the 1DUR of related cyclotron braid generators. Utilizing the form of this wave function generalization one can conclude on comparative stability of various correlated states via assessment of their energies, i.e., of interaction energy averaged over these trial wave functions. The averaged energies obtained by Monte-Carlo Metropolis method of multivariable function integration are in agreement with experimental data and coincide with exact diagonalization of Coulomb energy in small models.
Let us emphasize that the commensurability condition based on the braid group approach has nonlocal and topological character. The role of the Coulomb interaction is of the primary importance for this method. FQHE is the nonperturbative effect induced by the Coulomb interaction of electrons in planar geometry at strong perpendicular magnetic field presence. In this context worth mentioning is that in Hall systems the Coulomb interaction does not lead to a continuous mass operator which precludes the quasiparticle concept [29, 47]. Instead we deal here with particle separation quantization [2, 3] expressed in the form of the famous Laughlin function and Haldane interaction pseudopotentials [48, 49]. The related nonlocal and nonperturbative correlations cannot be expressed by any local or single-particle approach. The conventional CF model [22] though uses single-particle notion of effective composite particles is in fact also nonlocal one due to involvement of effective auxiliary field flux quanta pinned to electrons. This construction is essentially nonlocal and nonperturbative and displays the topological deep property of the 2D charged system in magnetic field clearly different than the concept of quasiparticles common in solids [29]. CFs are neither quasiparticles nor effective single-particle excitations in the Hall system. They reveal rather the long range correlation between 2D electrons expressed by auxiliary condition of quantization of fictitious flux pinned to each electron in order to form the composite particle. The character of this long range correlation is clearly explained in terms of the braid group approach and related commensurability condition allowing for clarifying of the structure of CFs and of constraints imposed on the CF model, on the other hand. This is especially well visible in the manifestation of FQHE at these fractions which cannot be explained by the conventional CF model but are quite naturally embraced by the commensurability condition (e.g., LLL fractions with even denominator like \( \frac{3}{8} \), \( \frac{4}{10} \), or LLL fraction \( \frac{1}{2} \) in bilayer Hall system revealing FQHE and not Hall metal as in monolayer.
system \([15, 50]\)). The braid group commensurability condition allows for explanation of full experimentally observed hierarchy of filling fractions corresponding to FQHE including those admitted by the CF model as a particular limiting situation. The convincing example is also the possibility of an explanation in braid group terms of the unconventional FQHE observed in the LLL of bilayer graphene at even denominator fractions \([15]\) beyond the prediction of the CF model \([50]\). Note that the presented commensurability braid group approach is efficient also in higher LLs \([51, 52]\) where constraints limiting applicability of the CF model are even more severe than those in the LLL \([53]\).
The most important feature related with symmetry of carriers involved in correlations of FQHE and revealed by the developed approach of cyclotron braid subgroups induced by the specific commensurability condition in 2D charged interacting systems at magnetic field presence is a possible violation of fermionic antisymmetry property of 2D electrons. This fundamental observation is unique in condensed matter and closely related with specific topology of 2D space resulted in FQHE.
**ACKNOWLEDGMENTS**
Supported by the NCN projects P.2011/02/A/ST3/00116 and P.2016/21/D/ST3/00958.
---
[1] D. C. Tsui, H. L. Störner, and A. C. Gossard. Two-dimensional magnetotransport in the extreme quantum limit. *Phys. Rev. Lett.*, 48:1559, 1982.
[2] R. B. Laughlin. Anomalous quantum Hall effect: an incompressible quantum fluid with fractionally charged excitations. *Phys. Rev. Lett.*, 50:1395, 1983.
[3] R. B. Laughlin. Quantized motion of three two-dimensional electrons in a strong magnetic field. *Phys. Rev. B*, 27:3383, 1983.
[4] W. Pan, H. L. Störmer, D. C. Tsui, L. N. Pfeiffer, K. W. Baldwin, and K. W. West. Fractional quantum Hall effect of composite fermions. *Phys. Rev. Lett.*, 90:016801, 2003.
[5] J. K. Jain. A note contrasting two microscopic theories of the fractional quantum hall effect. *Indian Journal of Physics*, 88:915, 2014.
[6] J. P. Eisenstein, M. P. Lilly, K. B. Cooper, L. N. Pfeiffer, and K. W. West. New physics in high Landau levels. *Physica E*, 6:29, 2000.
[7] M. Dolev, Y. Gross, R. Sabo, I. Gurman, M. Heiblum, V. Umansky, and D. Mahalu. Characterizing neutral modes of fractional states in the second Landau level. *Phys. Rev. Lett.*, 107:036805, 2011.
[8] R. L. Willett. The quantum Hall effect at 5/2 filling factor. *Rep. Prog. Phys.*, 76:076501, 2013.
[9] W. Pan, K. W. Baldwin, K. W. West, L. N. Pfeiffer, and D. C. Tsui. Spin transition in the \( \nu = 8/3 \) fractional quantum Hall effect. *Phys. Rev. Lett.*, 108:216804, 2012.
[10] J. S. Xia, W. Pan, C. L. Vicente, E. D. Adams, N. S. Sullivan, H. L. Störmer, D. C. Tsui, L. N. Pfeiffer, K. W. Baldwin, and K. W. West. Electron correlation in the second Landau level: a competition between many nearly degenerate quantum phases. *Phys. Rev. Lett.*, 93:176809, 2004.
[11] F. Amet, A. J. Bestwick, J. R. Williams, L. Balicas, K. Watanabe, T. Taniguchi, and D. Goldhaber-Gordon. Composite fermions and broken symmetries in graphene. *Nat. Commun.*, 6:5838, 2015.
[12] C. R. Dean, A. F. Young, P. Cadden-Zimansky, L. Wang, H. Ren, K. Watanabe, T. Taniguchi, P. Kim, J. Hone, and K. L. Shepard. Multicomponent fractional quantum Hall effect in graphene. *Nature Physics*, 7:693, 2011.
[13] B. E. Feldman, B. Krauss, J. H. Smet, and A. Yacoby. Unconventional sequence of fractional quantum Hall states in suspended graphene. *Science*, 337:1196, 2012.
[14] B. E. Feldman, A. J. Levin, B. Krauss, D. A. Abanin, B. I. Halperin, J. H. Smet, and A. Yacoby. Fractional quantum Hall phase transitions and four-flux states in graphene. *Phys. Rev. Lett.*, 111:076802, 2013.
[15] D. K. Ki, V. I. Falko, D. A. Abanin, and A. Morpurgo. Observation of even denominator fractional quantum Hall effect in suspended bilayer graphene. *Nano Lett.*, 14:2135, 2014.
[16] A. Kou, B. E. Feldman, A. J. Levin, B. I. Halperin, K. Watanabe, T. Taniguchi, and A. Yacoby. Electron-hole asymmetric integer and fractional quantum hall effect in bilayer graphene. *Science*, 345:55, 2014.
[17] Patrick Maher, Lei Wang, Yuanda Gao, Carlos Forsythe, Takashi Taniguchi, Kenji Watanabe, Dmitry Abanin, Zlatko Papić, Paul Cadden-Zimansky, James Hone, Philip Kim, and Cory R.
Dean. Tunable fractional quantum hall phases in bilayer graphene. *Science*, 345:61, 2014.
[18] Youngwook Kim, Dong Su Lee, Suyong Jung, Viera Skákalová, T. Taniguchi, K. Watanabe, Jun Sung Kim, and Jurgen H. Smet. Fractional quantum hall states in bilayer graphene probed by transconductance fluctuations. *Nano Lett.*, 15:7445, 2015.
[19] Georgi Diankov, Chi-Te Liang, François Amet, Patrick Gallagher, Menyoung Lee, Andrew J. Bestwick, Kevin Tharratt, William Coniglio, Jan Jaroszynski, Kenji Watanabe, Takashi Taniguchi, and David Goldhaber-Gordon. Robust fractional quantum hall effect in the n=2 landau level in bilayer graphene. *Nature Comm.*, 7:13908, 2016.
[20] M. O. Goerbig. Electronic properties of graphene in a strong magnetic field. *Rev. Mod. Phys.*, 83:1193, 2011.
[21] J. K. Jain. Composite-fermion approach for the fractional quantum Hall effect. *Phys. Rev. Lett.*, 63:199, 1989.
[22] J. K. Jain. *Composite Fermions*. Cambridge UP, Cambridge, 2007.
[23] S. Mukherjee, S. S. Mandal, Y.H. Wu, A. Wójc, and J. K. Jain. Enigmatic 4/11 state: A prototype for unconventional fractional quantum hall effect. *Phys. Rev. Lett.*, 112:016801, 2014.
[24] P. Sitko, K.-S. Yi, and J. J. Quinn. Composite fermion hierarchy: Condensed states of composite fermion excitations. *Phys. Rev. B*, 56:12417, 1997.
[25] K. Park and J. K. Jain. Mixed states of composite fermions carrying two and four vortices. *Phys. Rev. B*, 62:R13274, 2000.
[26] A. Wójc, K.-S. Yi, and J. J. Quinn. Fractional quantum hall states of clustered composite fermions. *Phys. Rev. B*, 69:205322, 2004.
[27] W. Pan, K. W. Baldwin, K. W. West, L. N. Pfeiffer, and D. C. Tsui. Fractional quantum hall effect at landau level filling \( \nu = 4/11 \). *Phys. Rev. B*, 91:041301, 2015.
[28] N. Samkharadze, I. Arnold, L. N. Pfeiffer, K. W. West, and G. A. Csáthy. Observation of incompressibility at \( \nu = 4/11 \) and \( \nu = 5/13 \). *Phys. Rev. B*, 91:081109, 2015.
[29] A. A. Abrikosov, L. P. Gorkov, and I. E. Dzialoshinskii. *Methods of Quantum Field Theory in Statistical Physics*. Dover Publ. Inc., Dover, 1975.
[30] Y. W. Suen, L. W. Engel, M. B. Santos, M. Shayegan, and D. C. Tsui. Observation of a \( \nu=1/2 \) fractional quantum Hall state in a double-layer electron system. *Phys. Rev. Lett.*, 68:1379, 1992.
[31] J. P. Eisenstein, G. S. Boebinger, L. N. Pfeiffer, K. W. West, and S. He. New fractional quantum Hall state in double-layer two-dimensional electron systems. Phys. Rev. Lett., 68:1383, 1992.
[32] N. Samkharadze, I. Arnold, L. N. Pfeiffer, K. W. West, and G. A. Csáthy. Observation of incompressibility at $\nu = 4/11$ and $\nu = 5/13$. Phys. Rev. B, 91:081109, 2015.
[33] W. Pan, K. W. Baldwin, K. W. West, L. N. Pfeiffer, and D. C. Tsui. Fractional quantum hall effect at landau level filling $4/11$. Phys. Rev. B, 91:041301(R), 2015.
[34] Y. S. Wu. General theory for quantum statistics in two dimensions. Phys. Rev. Lett., 52:2103, 1984.
[35] M. G. Laidlaw and C. M. DeWitt. Feynman functional integrals for systems of indistinguishable particles. Phys. Rev. D, 3:1375, 1971.
[36] J. S. Birman. Braids, Links and Mapping Class Groups. Princeton UP, Princeton, 1974.
[37] J. Jacak, R. Gonczarek, L. Jacak, and I. Jóźwiak. Application of Braid Groups in 2D Hall System Physics: Composite Fermion Structure. World Scientific, 2012.
[38] J. Jacak, I. Jóźwiak, and L. Jacak. New implementation of composite fermions in terms of subgroups of a braid group. Phys. Lett. A, 374:346, 2009.
[39] T. D. Imbo, C. S. Imbo, and C. S. Sudarshan. Identical particles, exotic statistics and braid groups. Phys. Lett. B, 234:103, 1990.
[40] E. C. G. Sudarshan, T. D. Imbo, and T. R. Govindarajan. Configuration space topology and quantum internal symmetries. Phys. Lett. B, 213:471, 1988.
[41] J. Jacak, I. Jóźwiak, L. Jacak, and K. Wieczorek. Cyclotron braid group structure for composite fermions. J. Phys: Cond. Matt., 22:355602, 2010.
[42] H. L. Stormer, R. R. Du, W. Kang, D. C. Tsui, L. N. Pfeiffer, K. W. Baldwin, and K. W. West. The fractional quantum Hall effect in a new light. Semicond. Sci. Technol., 9:1853, 1994.
[43] Ajit C. Balram, Csaba Tőke, A. Wójs, and J. K. Jain. Fractional quantum hall effect in graphene: Quantitative comparison between theory and experiment. Phys. Rev. B, 92:075410, 2015.
[44] Orion Ciftja and Carlos Wexler. Monte carlo simulation method for lauglin-like states in a disk geometry. Phys. Rev. B, 67:075304, 2003.
[45] R. Morf and B. I. Halperin. Monte carlo evaluation of trial wavefunctions for the fractional quantized hall effect: Spherical geometry. *Z. Phys. B Condensed Matter*, 68:391, 1987.
[46] N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. M. Teller, and E. Teller. Equation of state calculations by fast computing machines. *J. Chem. Phys.*, 21:1087, 1953.
[47] L. D. Landau and E. M. Lifshitz. *Quantum Mechanics: Non-relativistic Theory*. Nauka, Moscow, 1972.
[48] S. Das Sarma and A. Pinczuk. *Perspectives in Quantum Hall Effects: Novel Quantum Liquids in Low-Dimensional Semiconductor Structures*. Wiley, New York, 1997.
[49] F. D. M. Haldane. Fractional quantization of the Hall effect: A hierarchy of incompressible quantum fluid states. *Phys. Rev. Lett.*, 51:605, 1983.
[50] J. Jacak and L. Jacak. Explanation of nu=-1/2 fractional quantum hall state in bilayer graphene. *Proc. R. Soc. A*, 472:20150330, 2016.
[51] A. Kumar, G. A. Csáthy, M. J. Manfra, L. N. Pfeiffer, and K. W. West. Nonconventional odd-denominator fractional quantum hall states in the second landau level. *Phys. Rev. Lett.*, 105:246808, 2010.
[52] E. Kleinbaum, A. Kumar, L. N. Pfeiffer, K. W. West, and G. A. Csáthy. Gap reversal at filling factors 3 + 1/3 and 3 + 1/5: Towards novel topological order in the fractional quantum hall regime. *Phys. Rev. Lett.*, 114:076801, 2015.
[53] J. Jacak and L. Jacak. The commensurability condition and fractional quantum hall effect hierarchy in higher landau levels. *JETP Letters*, 102:19, 2015. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
1522,
1
],
[
1522,
4539,
2
],
[
4539,
6549,
3
],
[
6549,
9216,
4
],
[
9216,
11863,
5
],
[
11863,
14935,
6
],
[
14935,
17130,
7
],
[
17130,
19909,
8
],
[
19909,
22481,
9
],
[
22481,
22782,
10
],
[
22782,
23767,
11
],
[
23767,
26420,
12
],
[
26420,
29035,
13
],
[
29035,
30803,
14
],
[
30803,
33575,
15
],
[
33575,
36431,
16
],
[
36431,
39228,
17
],
[
39228,
41742,
18
],
[
41742,
44656,
19
],
[
44656,
46712,
20
],
[
46712,
49006,
21
],
[
49006,
51320,
22
],
[
51320,
53480,
23
],
[
53480,
55013,
24
]
]
} |
dc45d082175d8bfd8f8329bb587174922a65c885 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 7,
"total-fallback-pages": 0,
"total-input-tokens": 25244,
"total-output-tokens": 9159,
"fieldofstudy": [
"Computer Science"
]
} | Abstract—The fifth generation of wireless cellular networks (5G) is expected to be the infrastructure for emergency services, natural disasters rescue, public safety, and military communications. 5G, as any previous wireless cellular network, is vulnerable to jamming attacks, which create deliberate interference to hinder the communication of legitimate users. Therefore, jamming 5G networks can be a real threat to public safety. Thus, there is a strong need to investigate to what extent these networks are vulnerable to jamming attacks. For this investigation, we consider the 3GPP standard released in 2017, which is widely accepted as the primary reference for the deployment of these networks. First, we describe the key elements of 5G New Radio (NR) architecture, such as different channels and signals exchanged between the base station and user equipment. Second, we conduct an in-depth review of the jamming attack models and we assess the 5G NR vulnerabilities to these jamming attacks. Then, we present the state-of-the-art detection and mitigation techniques, and we discuss their suitability to defeat smart jammers in 5G wireless networks. Finally, we provide some recommendations and future research directions at the end of this paper.
Index Terms—Smart Jamming, 5G New Radio, Frequency Hopping Spread Spectrum, Game Theory, Direct Sequence Spread Spectrum, Timing Channels, Machine Learning.
I. INTRODUCTION
The fifth generation of wireless cellular networks, 5G, promises faster data rates and reliable service delivery. It is expected to enable many cutting-edge technologies such as internet-of-things (IoTs), self-driving cars, and smart cities. In 2017, 3GPP released the specification of 5G New Radio (NR), which has been the primary reference for the deployment of these networks. 5G NR architecture is built upon five fundamental pillars: New radio spectrum, massive MIMO/beamforming, multi-connectivity, network flexibility, and high level of security. 5G is operable on a new radio spectrum from below 1 GHz to up to 100 GHz. The 5G NR physical layer uses orthogonal frequency division modulation (OFDM) with a cyclic prefix on the downlink and either the OFDM or discrete Fourier transform-spread OFDM for uplink. The 5G NR frame is of 10 ms duration, in which there are ten sub-frames and fourteen OFDM symbols. 5G NR supports both the frequency division multiplexing FDD and time division multiplexing TDD modes [1].
As any wireless cellular networks, 5G networks are built upon open sharing in which the communication medium is the free space making them prone to interference, which is one of the fundamental causes of degradation of the performance of wireless networks. If the level of obstruction is high, the receivers are not able to decode the transmitted signals. This weakness can be used by some adversary nodes to cause intentional interference and hinder legitimate user's communication over specific wireless channels. This is well-known as jamming attacks.
Jamming attacks pose serious risks to public communication services [2], [3]. In early 1900, jamming attacks were used in military battles. Nowadays, jamming attacks can be launched to hinder public communication services. Several jammer devices are available in the market at a low cost. In addition, the most sophisticated jamming attacks can be implemented with a price as low as 1k$ using low-cost software-defined radio tools, and some primary programming skills. Furthermore, 5G is expected to be the infrastructure for emergency services, natural disasters rescue, public safety, and military communications making jamming attacks a real threat.
Therefore, one of the central requirements of 5G NR is a high level of security and resilience to jamming attacks [3]. 5G is expected to enhance the security of wireless networks and fix the breaches of long-term evolution (LTE) or 4G networks, especially the resilience to jamming attacks. In 2017, 3GPP released the 5G NR standard. Before deploying these cellular networks, one needs to investigate to what extent the 5G standard released by 3GPP is resilient to jamming attacks, and build security protocol that can be incorporated in this standard. Thus, it is essential to determine under which conditions (e.g., the power of the jammer, the duty cycle, etc.), the jammer can take down the communication channel; and to determine which anti-jamming techniques are suitable for 5G NR.
In this paper, we aim to provide an in-depth study of these issues, assess the risks of jamming attacks on 5G NR, and suggest some possible future research directions on how to efficiently tackle this problem. The main inputs of this article can be summarized as follows:
- Description of the key elements of 5G NR architecture
- Review of jamming attack models and strategies
• Assessment of jamming vulnerabilities of the 5G NR
• Review of detection and mitigation techniques and discussion of their suitability to defeat smart jammers in 5G
• Conclusions, recommendations, and future research directions
The remaining sections of this paper are arranged as follows. Section II describes the 5G NR architecture. Section III defines jamming attack strategies, types, and models. The overall 5G NR vulnerability to jamming attacks assessment is provided in Section IV. Section V revisits briefly the state-of-the-art anti-jamming and mitigation techniques and discusses their suitability for 5G NR. Section VI summarizes the results of this work and draws the conclusions.
II. 5G NR ARCHITECTURE
5G NR is operable on from low to very high-frequency bands (0.6-30GHz). It gives ultra-wide carrier bandwidth, which can be up to 100 MHz in below 6 GHz and up to 400 MHz in higher than 6 GHz. In 5G NR, there are several physical channels. For instance, for the downlink, there is downlink shared channel (PDSCH), Broadcast channel (PBCH), and downlink control channel (PDCCH). For the uplink, there is uplink shared channel (PUSCH), uplink control channel (PUCCH), and Random access channel (PRACH).
The physical layer of 5G NR includes many types of signaling reference signals and synchronization pilots exchanged on both downlink and uplink. For instance, the base station uses the primary synchronization signal (PSS) and secondary synchronization signal (SSS) for downlink frame synchronization and conveying cell-ID to user equipment (UE). The PSS has three possible combinations, while the SSS has 336 combinations. Each of the PSS and SSS consists of an m sequence of length 127, and is mapped to a set of 127 subcarriers within the same OFDM symbol, different OFDM symbols, respectively. The use of the Gold sequence, which is formed by combining two orthogonal m-sequences, enables the UE to differentiate between several base stations on the same carrier at a low signal-to-noise ratio (SNR) [4].
5G NR enables scalable NR numerology to address different radio spectrum, bandwidths, and services. For instance, subcarrier spacing (SCS) of 15, 30, 60, and 120 kHz is specified for macro coverage, small cell, indoor, and mmWave, respectively. 5G NR frame is similar to the one of 4G/LTE with some slight modifications. One slot in the 5G NR frame is composed of 14 symbols, and the slot length is dependent upon the CSC. Mini-slot is comprised of 2, 4, and 7 symbols, which can be allocated for shorter transmissions. Slots can be aggregated for more extended periods of communication. The OFDM symbol contains PSS, PBCH, and SSS.
For the coding schemes, 5G NR uses low-density parity check (LDPC) for the data channel and polar coding for the control channel. It has been shown that LDPC codes perform well when used for error correction for small chunks of data. Polar coding, on the other hand, can achieve performance close to the Shannon limit, but it has to be used with large pieces of data.
Another feature of 5G NR is the use massive Multiple input multiple output (MIMO) to enhance the coverage and the capacity of wireless cells [5]–[8].
III. JAMMING ATTACKS
Jammers can be defined as malicious wireless nodes planted by an adversary to cause intentional interference in wireless cellular networks. Depending on its attacking strategy, one can identify several types of jammers. In the following, we provide the most common types of jammers and describe their strategies.
1) Regular jammer: In this type, the jammers tend to not follow any MAC protocol before continually injecting radio frequency signals without gaps in between, which can be either legitimate bit sequences or random bit sequences to interfere with legitimate transmitted signals over a wireless channel. Subsequently, these bits occupy the transmission channel to starve transmissions initiated by legitimate nodes. This type of attack uses enormous power, which drains the battery life of the malicious node due to its continuous transmission of radio signals. Regular jammers, consequently, require a high amount of power to carry out this attack. On the other hand, regular jammers do not need to monitor the activity of legitimate users [9]–[11].
2) Delusive jammer: In this type, known also as deceptive, jammers continuously inject legitimate sequence of bits into the communication channel. This type of jammer often misleads the receiver to believe that this is a message from a legitimate source. It forces the receiver to wait in the listening states. In comparison with regular jammer, delusive jammer tends to be quite challenging to detect because of the similarity between the fake signal and the legitimate one [10], [11].
3) Random jammer: Different from both regular and deceptive jammers, random jammers conserve their energy by alternating between active and idle states. During the jamming process, the malicious node jams for a predetermined period before turning off its radio. After a while, it reactivates the jamming process from the sleep mode and continually follows that pattern. During the jamming mode, it can exhibit either regular or deceptive jamming feature, while during the idle state, it conserves energy and therefore reducing its power consumption [9]–[11].
4) Responsive jammer: All of the three previous jamming strategies discussed before are active jammers, as they attempt to block the communication channel, regardless of the activity pattern of the legitimate nodes. An alternative
to active jammers to reduce its power consumption is to be a quick responsive jammer, known also as reactive jammers, which can be a more power-efficient method. Responsive jammers continually monitor the communication channel, and transmit only of the transmitter is active [9], [10], [12], [13]. Responsive jammers minimize power consumption despite the monitoring activity for the power required is far less than the one necessary to jam a communication channel. For instance, the authors of [7] launched a jamming attacks using deep learning.
5) Go-next jammer: This jammer is selective because it targets one frequency channel at a time. If the transmitter detects the existence of a jammer over the frequency channel and hops to the next frequency, this kind of jammer follows on the transmitter and goes to the next frequency channel. Due to its selective nature, go-next jammer conserves its energy. Notwithstanding, if the transmitter performs fast rate frequency hopping, the jammer’s energy can be wasted because of the successive hops [11].
6) Control channel jammers: This jammer targets the control channel to block the exchange between the transmitter and the receiver before initiation of the communication. Control channel jammer can be of several types and can cause a denial of service and even denies nodes access to the network [11], [14].
IV. VULNERABILITY OF NR TO JAMMING ATTACKS
A. Vulnerableness of the PBCH to Jamming
Base station PBCH are assigned symbols within two slots of each other if the carrier is below 3 GHz and within four slots if the carrier frequency is above 3 GHz [4]. As higher the sub-carrier spacing (SCS), the duration of one slot is smaller, and the selective jamming duty cycle is lower. Consequently, a selective jammer can target the PBCH using a shallow duty cycle as the symbols are close to each other in both cases. This design is a vulnerability in design even that the use of higher frequency does not propagate for a long-distance making the jammer getting closer to the mobile station to launch its attacks. A localization-based detection technique can identify the source of the jammer and stop it. However, if the jammer is mobile, the anti-jamming has to monitor the mobility of the jammer to detect the next jammer positions. The longer the monitoring process, the harmful the jamming is going to be.
B. Vulnerableness of PDCCH to Jamming
CORESET is a set of physical resources (i.e, a specific area on 5G NR Downlink Resource Grid) and a set of parameters that are used to carry PDCCH. It is equivalent to LTE PDCCH area (the first fourth OFDM symbols in a subframe). But in LTE PDCCH region, the PDCCH always spread across the whole channel bandwidth, but NR CORESET region is localized to a specific region in the frequency domain. Jamming PDCCH channel is far more complicated than jamming PBCH channel. To jam PDCCH, the jammer has to cram all the possible locations in which the PDCCH resides, assuming that the jammer does not have any knowledge of the CORESET freq-domain. However, the jammer can intercept and decode the CORESET freq-domain, which gives the jammer and advantage to jam specific sub-carrier, using a small duty cycle depending upon the value of CORESET-time-duration. The question is how long it can take to intercept and decode the CORESET.
C. Vulnerableness of PUCCH to Jamming
The PUCCH has an option for intra-slot hopping, which can provide some protection against selective jammer, but the robustness of this defense mechanism is dependent on the hopping rate. Also, this information is available to the jammer as the 5G standard is public. Thus, knowing the intra-slot hopping gives the jammer an advantage to jam PUCCH at low cost. Furthermore, the PUCCH is modulated with MPSK (m=2 or 4) and polar code or just repetition code as an error coding scheme depending on the number of bits to be transmitted. Polar codes are well known by their low protection against jamming attacks.
D. Vulnerableness of RACH to Jamming
The random access (RA) procedure is the uplink transmission of a random access preamble by the UE on a dedicated RACH. After the reception of a preamble, the base station estimates temporal synchronization parameters and allocates radio resources for further communication with the UE. The synchronization parameters and allocation of radio resources are then communicated to the UE that initiated the RA procedure within a specified time after the RA preamble transmission [15]. This information is broadcasted on PRACH, which takes the form of a Zadoff-Chu sequence that embeds a value used to identify the UE temporarily. Despite the large number of possible locations, and the high complexity needed to determine the positions in real-time, jamming PRACH is still feasible [4]. Furthermore, if the jammer does not succeed in determining these locations, it can flood the channel with an invalid preamble as the 5G NR does not specify what it should be done in this scenario.
E. Vulnerableness of Massive MIMO to Jamming
Many research studies (industrial and academic) showed that massive MIMO are vulnerable to jamming attacks. Jamming MIMO systems targets the channel estimation of these systems. By targeting the channel estimation procedure, an adversary may launch active jamming attacks against...
unsuspecting users. The authors of [16] presented several jamming methodologies for SVD-based MIMO systems, including a powerful and practical channel rank attack. The authors presented several attack strategies to undermine Alamouti STBC-based MIMO scheme. Such attacks have been proven feasible by way of analysis, simulations, and real-world experimentation. Additionally, the attack strategies presented are general and remains valid for massive MIMO systems. Therefore, preserving accurate channel estimation under jamming attacks is quintessential to gain the desired performance enabled by massive MIMO. Hence, there is an urgent need for developing techniques for accurate channel state estimation whose performance is not impacted by the presence of jammers, or at least it considers the presence of the jammer and estimate the channel state information from affected pilot samples.
F. Robustness of 5G NR Channel Coding to Jamming
3GPP specifications for 5G radio standard includes polar coding and LDPC coding techniques. Polar coding which uses the channel polarization to split the channel into good channel and worse channel and transmit only on the good channel, presents several advantages but have some limitations. LDPC codes on the other hand if used with large block, the complexity of the decoder increase exponentially. Most of the control channels use polar coding as error coding scheme, and the data channels use LDPC coding. It has been shown that polar coding is vulnerable to jammers. At $0\, dB$ SNR, the bit error rate is so high. Likewise, LDPC coding are vulnerable to jamming attacks.
V. ANTI-JAMMING IN 5G
In this section, we review the anti-jamming techniques. We divided this section into three parts: the first part presents jamming detection methods, the second part deals with mitigation methods, and the last part provide a discussion on the effectiveness of these techniques in tackling jamming attacks in 5G.
A. Detection of Jamming Attacks
Detection of smart-jamming attacks is feasible by monitoring any excess amount of energy on a specific physical channel (e.g. using masking) or any sudden change in the performance of the communication over this channel. A common strategy in jamming detection is the use of a threshold with some performance metrics such as the packet delivery ratio (PDR), packet drop ratio (PDR), bit error rate (BER), and signal-to-noise ratio (SNR). These techniques monitor the level of these metrics during the absence and the presence of jamming attacks and set manually the threshold for detection. Threshold based detection are only efficient when we are dealing with constant jammer. In addition, because of the wireless environment dynamics, these methods have a high false alarm.
Another detection category is statistical based [17], [18]. The concept of these techniques often uses historical data and compute some statistic to distinguish jammed signal from a non-jammed signal. Statistical detection of jamming is investigated with different forms of jamming attacks, and can achieve high accuracy when dealing with constant jammers.
The last category is machine learning based. Several machine learning techniques such as random forest, decision tree, adaptive boosting, support vector machine, and expectation maximization are investigated in detecting jamming attacks [19]. Recently, deep learning which is a special case of machine learning is heavily investigated to detect jammers [7]. Deep learning can detect jammers with high accuracy. Nevertheless, deep learning presents some limitations. There is no public dataset that can be used to train machine learning models. Most of the proposed methods generated dataset using simulations and only a few papers have conducted real-world setup to collect data. It is hard to foretell the performance of these detection techniques under a real jamming attacks.
B. Mitigation of Jamming Attacks
5G networks are going to use frequency higher than 30 GHz. Jamming these bands are not likely to happen because the jammer need high level of power to jam these bands [4]. In addition, 5G is going to implement techniques such as direct sequence spread spectrum (DSSS) and frequency hopping spread spectrum (FHSS) [20]. One need to investigate to which extent these techniques can act against jamming at the 5G physical layer.
1) Direct Sequence Spread Spectrum: spread spectrum can provide protection against interfering jamming signals with finite power. This technique purposely makes the information bearing signal occupy a bandwidth larger that of the minimum necessary to transmit it. Thus, the signal is transmitted through the channel undetected by an eavesdropper. In direct sequence spread spectrum (DSSS), the data signal is multiplied with a pseudo-noise (PN) sequence. The data signal is a narrow band and the PN sequence is wideband making the product nearly have a spectrum as same as PN which plays the spreading code role. The resilience of this technique against jamming attacks depends upon the spreading factor. An example of the protection that DSSS provides is shown by the authors of [21], in which a BPSK modulated signal is considered and if $N = 4095$ and the BER is not to exceed $10^{-5}$, the authors showed the data at the receiver can be detected reliably even the jamming power is more than 400 times the received signal power. This example shows that direct sequence spread spectrum is powerful against interference jamming. Nevertheless, DSSS has some limitations such as the larger bandwidth required and the
In the pilot phase, the base station estimates the jammer’s presence and resorts to reinforcement learning techniques to estimate the jammer’s optimal strategy. Scheduling can achieve better performance, expressed as a game between the jammer and the legitimate user. The switching costs of the jammer are determined using Deep reinforcement learning, which means one symbol rate is transmitted during several frequency hops.
Using frequency hopping has many limitations. For instance, using a slow-rate hopping does not provide a robust protection against smart jammers. In this framework, the jammer switches to another frequency instead of instantaneously. One characteristic of Frequency-hop Spread Spectrum (FHSS) techniques is that they are powerful yet impose some practical limits because of the capabilities of the physical devices used to generate the PN sequence. Specifically, it may turn that the generated PN sequence is still not large enough to overcome the effects of some jammers, which is the same cases, resort to different strategies. One way to get around this problem is by randomly hopping the data modulated carrier from one frequency to another [22], [23]. In this type of spread spectrum, the spectrum of the transmitted signal is spread sequentially instead of instantaneously. One characteristic of FHSS is the hopping rate, based on which one can distinguish between two types:
- Slow-rate hopping: Several modulated symbols are conveyed within one frequency hop before it hops to the next frequency.
- Fast-rate hopping: is the converse of the slow-frequency hopping, which means one symbol rate is transmitted during several frequency hops.
Using frequency hopping has many limitations. For instance, using a slow-rate hopping does not provide a robust protection against smart jammers. In this case, the jammer can find the next hop before the transmitter switches to the next frequency; and using a fast-rate hopping can decrease the performance of the communication channel as it becomes hard to synchronize the transmitter and the receiver. In addition, frequency hopping requires a pre-shared key between the transmitter and the receiver to agree on the hopping pattern, exchanging the keys can be intercepted by an eavesdropper. The authors of [24] proposed a secretive adaptive frequency hopping scheme for 5G. The authors of [25] proposed a pseudo-random time hopping for anti-jamming in 5G wireless networks. The authors analytically evaluated the performance of the proposed scheme by determining the jamming probability, the switching rate, and bit error rate. The authors of [26] proposed a frequency hopping for 5G mmWave. Yet, there is need to evaluate the impact of frequency hopping on outage probability for 5G mmWave.
3) **Game Theory:** Game theory is another anti-jamming technique that aims to find the optimal strategy to defeat jammers [8], [27]–[29]. Legitimate users can avoid the jamming attacks by proactively hopping among accessible channels and thereby minimizing the payoff function. The anti-jamming in this context is expressed as a game between the legitimate user and the jammer. Game theory can be used to find the optimal strategy to cope with a jammer such as hopping to the next frequency. Several researchers have shown that it is possible to achieve the Nash equilibrium, meaning that the transmitter can find the optimal strategy to cope with the jammer.
4) **Timing channels:** The timing channel restores the communication between legitimate users under jamming attacks instead of frequency hopping. The timing channel is reinstated over the jammed channel using the timing patterns of attacker [30], [30]. This information enables the transmitter to transmit only when the jammer is in the idle-state. The timing channel requires the detection step before the creation of the timing channel.
5) **UAVs and Reinforcement learning:** An unmanned aerial vehicle (UAV) aided 5G wireless communication framework is yet another anti-jamming strategy. The UAVs are used as a relay scheme if the base station is heavily jammed. The UAVs use deep reinforcement learning techniques to determine the optimal relay policy for mobile users in 5G cellular networks. Examples of solutions based on UAVs and Deep reinforcement learning have been proposed in [6], [31]–[34]. This solution could be useful because of the flexibility the UAVs give to the network to avoid the jammer. Nevertheless, this framework faces many challenges, and UAVs themselves are also vulnerable to jammers, and its power supply is limited.
6) **Suppression of Jammers:** Massive MIMO suppression is a potential technique that can be enhanced and used to deal with jamming, and it does not require any change in the 5G NR specification [35]. The solution to interference is to build robust channel coding schemes that can correct packets corrupted by the jammers. This strategy can exhaust the jammer.
To improve the resilience of 5G systems to jamming attacks, the authors of [36] proposed a jamming-resistant receiver scheme. The prominent feature of this proposed scheme is that, in the pilot phase, the base station estimates estimate both the jamming channel. The jamming channel estimate is then exploited to build linear receiver filters that reject the impact of the jamming signal. The authors of [37] proposed a mitigation technique based on random matrix theory.
7) **Scheduling and Deep Learning:** 5G wireless network functions SDN and NFV alongside deep learning can help in building intelligent dynamic radio resource allocation and scheduling, which can significantly reduce the risk of jamming attacks. For instance, the authors of [38] proposed the joint power control and scheduling problem in jammed networks under minimum QoS constraints without any prior about the jammer positions. If combined with Deep learning to learn the strategy of the jammer, scheduling can achieve better performance.
C. Effectiveness of Mitigation Techniques and Future Research Directions
Direct spread spectrum techniques can achieve high protection against jamming attacks. However, the complexity of this technique can be an obstacle to the implementation of this technique. Frequency hopping techniques are not suitable to counter jammers because the latter can predict the next channels, and any exchange between the receiver and the transmitter can be intercepted [20]. Machine learning based anti-jamming schemes are not practical for some applications in 5G wireless networks as it requires long training time and it requires building large comprehensive dataset to have reliable detection accuracy. Timing channels can be reliable if combined with excellent detection technique. UAVs based mitigation is a promising solution, but further investigation of the practical issues should be considered.
Thus, further research studies on anti-jamming techniques are highly needed. The cyber-security requirement of 5G NR has to be embedded in the initial design of these networks. In this way, one can ensure a low-cost deployment, in contrast, to develop solutions to deal with future failures. For instance, base station must implement anti-jamming techniques. For example, if the exchange between the base station and user equipment, the base station should provide a spatial retreat, movement, time, and network reconfiguration.
Another future research direction is the use of a deep learning based approach can be used as anti-jamming. To train deep learning, a large comprehensive dataset is needed. For that, a real-world setup is needed. To collect data, different jammers should be considered. Data collection should be done under both scenarios, under jamming and under the normal scenario. The built dataset can be used to train and test deep learning techniques. Then, these techniques can be combined with sensing to detect the strategy of the jammer and actively select the communication channel that is not under jamming attacks. Deep learning models have to be trained to recognize a jamming signal from a legitimate user on the fly. Only if that can be performed, then the legitimate user can identify the pattern of the jammer and dedicate its strategy and accordingly define a mitigation hopping.
VI. CONCLUSION
In this paper, we presented an in-depth study of the vulnerability of 5G wireless systems to smart jamming attacks. We reviewed different types of jammers, and we showed that 5G NR enhanced the resilience of wireless cellular network to jamming attacks, primarily because of its flexible and dynamic resource allocation, yet 5G NR is still far from being secure against jamming attacks. The anti-jamming strategies presented in this paper address simple jamming attacks and are not very suitable for 5G NR. Thus, further research studies on anti-jamming techniques are highly needed.
REFERENCES
[1] M. Agiwal, A. Roy, and N. Saxena, “Next generation 5g wireless networks: A comprehensive survey,” IEEE Communications Surveys & Tutorials, vol. 18, no. 3, pp. 1617–1655, 2016.
[2] R. Khan, P. Kumar, D. N. K. Jayakody, and M. Liyanage, “A survey on security and privacy of 5g technologies: Potential solutions, recent advancements and future directions,” IEEE Communications Surveys & Tutorials, 2019.
[3] S. Vadlamani, B. Eksioglu, H. Medal, and A. Nandi, “Jamming attacks on wireless networks: A taxonomic survey,” International Journal of Production Economics, vol. 172, pp. 76–94, 2016.
[4] M. Lichtman, R. Rao, V. Marojevic, J. Reed, and R. P. Jover, “5g nr jamming, spoofing, and sniffing: threat assessment and mitigation,” in 2018 IEEE International Conference on Communications Workshops (ICC Workshops). IEEE, 2018, pp. 1–6.
[5] A. Tsiotis, D. Xenakis, N. Passas, and L. Merakos, “On jamming and black hole attacks in heterogeneous wireless networks,” IEEE Transactions on Vehicular Technology, 2019.
[6] L. Xiao, X. Lu, D. Xu, Y. Tang, L. Wang, and W. Zhuang, “Uav relay in vanets against smart jamming with reinforcement learning,” IEEE Transactions on Vehicular Technology, vol. 67, no. 5, pp. 4087–4097, 2018.
[7] T. Erpek, Y. E. Sagduyu, and Y. Shi, “Deep learning for launching and mitigating wireless jamming attacks,” IEEE Transactions on Cognitive Communications and Networking, vol. 5, no. 1, pp. 2–14, 2018.
[8] Q. Wang, T. Nguyen, K. Pham, and H. Kwon, “Mitigating jamming attack: A game-theoretic perspective,” IEEE Transactions on Vehicular Technology, vol. 67, no. 7, pp. 6063–6074, 2018.
[9] K. Pelechrinis, M. Iliofotou, and S. V. Krishnamurthy, “Denial of service attacks in wireless networks: The case of jammers,” IEEE Communications surveys & tutorials, vol. 13, no. 2, pp. 245–257, 2010.
[10] W. Xu, W. Trappe, Y. Zhang, and T. Wood, “The feasibility of launching and detecting jamming attacks in wireless networks,” in Proceedings of the 6th ACM international symposium on Mobile ad hoc networking and computing. ACM, 2005, pp. 46–57.
[11] K. Grover, A. Lim, and Q. Yang, “Jamming and anti-jamming techniques in wireless networks: a survey,” International Journal of Ad Hoc and Ubiquitous Computing, vol. 17, no. 4, pp. 197–215, 2014.
[12] D. Nguyen, C. Sahin, B. Shishkin, N. Kandasamy, and K. R. Dandekar, “A real-time and protocol-aware reactive jamming framework built on software-defined radios,” in Proceedings of the 2014 ACM workshop on Software radio implementation forum. ACM, 2014, pp. 15–22.
[13] M. Strasser, B. Danev, and S. Capkun, “Detection of reactive jamming” in sensor networks,” ACM Transactions on Sensor Networks (TOSN), vol. 7, no. 2, p. 16, 2010.
[14] X. Wei, Q. Wang, T. Wang, and J. Fan, “Jammer localization in multihop wireless network: a comprehensive survey,” IEEE Communications Surveys & Tutorials, vol. 19, no. 2, pp. 765–799, 2016.
[15] G. Schreiber and M. Tavares, “5G new radio physical random access preamble design,” 2018 IEEE 5G World Forum (5GWF), pp. 215–220, 2018.
[16] R. Miller and W. Trappe, “On the vulnerabilities of csi in mimo wireless communication systems,” IEEE Transactions on mobile Computing, vol. 11, no. 8, pp. 1386–1398, 2011.
[17] O. Osanaiye, A. Alfa, and G. Hancke, “A statistical approach to detect jamming attacks in wireless sensor networks,” Sensors, vol. 18, no. 6, p. 1691, 2018.
[18] A. Marttinen, A. M. Wyglinski, and R. Jantti, “Statistics-based jamming” detection algorithm for jamming attacks against tactical manets,” in 2014 IEEE Military Communications Conference. IEEE, 2014, pp. 501–506.
[19] O. Punal, I. Aktas, C.-J. Schnelke, G. Abidin, K. Wehrle, and J. Gross, “Machine learning-based jamming detection for ieee 802.11.” Design and experimental evaluation,” in Proceeding of IEEE International
D. Fang, Y. Qian, and R. Q. Hu, “Security for 5G mobile wireless networks,” IEEE Access, vol. 6, pp. 4850–4874, 2017.
S. Haykin, Communication systems. John Wiley & Sons, 2008.
S. Djuraev, J.-G. Choi, K.-S. Sohn, and S. Y. Nam, “Channel hopping scheme to mitigate jamming attacks in wireless lans,” EURASIP Journal on Wireless Communications and Networking, vol. 2017, no. 1, p. 11, 2017.
M. Strasser, C. Popper, and S. Capkun, “Efficient uncoordinated FHSS anti-jamming communication,” in Proceedings of the tenth ACM international symposium on Mobile ad hoc networking and computing. ACM, 2009, pp. 207–218.
Y. Li, B. Kaur, and B. Andersen, “Denial of service prevention for 5G,” Wireless Personal Communications, vol. 57, no. 3, pp. 365–376, 2011.
N. Adem, B. Hamdaoui, and A. Yavuz, “Pseudorandom time-hopping anti-jamming technique for mobile cognitive users,” in 2015 IEEE Globecom Workshops (GC Wkshps). IEEE, 2015, pp. 1–6.
S. Talarico and M. C. Valenti, “Frequency hopping on a 5G millimeterwave uplink,” in 2015 49th Asilomar Conference on Signals, Systems and Computers. IEEE, 2015, pp. 333–337.
B. Wang, Y. Wu, K. R. Liu, and T. C. Clancy, “An anti-jamming stochastic game for cognitive radio networks,” IEEE Journal on Selected Areas in Communications, vol. 29, no. 4, pp. 877–889, 2011.
L. Jia, Y. Xu, Y. Sun, S. Feng, and A. Anpalagan, “Stackelberg game approaches for anti-jamming defence in wireless networks,” IEEE Wireless Communications, vol. 25, no. 6, pp. 120–128, 2018.
D. Yang, G. Xue, J. Zhang, A. Richa, and X. Fang, “Coping with a smart jammer in wireless networks: A stackelberg game approach,” IEEE Transactions on Wireless Communications, vol. 12, no. 8, pp. 4038–4047, 2013.
W. Xu, W. Trappe, and Y. Zhang, “Anti-jamming timing channels for wireless networks,” in Proceedings of the first ACM conference on Wireless network security. ACM, 2008, pp. 203–213.
Y. Gwon, S. Dastangoo, C. Fossa, and H. Kung, “Competing mobile network game: Embracing antijamming and jamming strategies with reinforcement learning,” in 2013 IEEE Conference on Communications and Network Security (CNS). IEEE, 2013, pp. 28–36.
S. Machuzak and S. K. Jayaweera, “Reinforcement learning based anti-jamming with wideband autonomous cognitive radios,” in 2016 IEEE/CIC International Conference on Communications in China (ICCC). IEEE, 2016, pp. 1–5.
X. Lu, L. Xiao, C. Dai, and H. Dai, “UAV-aided cellular communications with deep reinforcement learning against jamming,” arXiv preprint arXiv:1805.06628, 2018.
X. Liu, Y. Xu, L. Jia, Q. Wu, and A. Anpalagan, “Anti-jamming communications using spectrum waterfall: A deep reinforcement learning approach,” IEEE Communications Letters, vol. 22, no. 5, pp. 998–1001, 2018.
H. Akhlaghpasand, E. Bjornson, and S. M. Razavizadeh, “Jamming suppression in massive MIMO systems,” IEEE Transactions on Circuits and Systems II: Express Briefs, 2019.
T. T. Do, E. Bjornson, E. G. Larsson, and S. M. Razavizadeh, “Jamming-resistant receivers for the massive MIMO uplink,” IEEE Transactions on Information Forensics and Security, vol. 13, no. 1, pp. 210–223, 2017.
J. Vinogradova, E. Bjornson, and E. G. Larsson, “Detection and mitigation of jamming attacks in massive MIMO systems using random matrix theory,” in 2016 IEEE 17th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC). IEEE, 2016, pp. 1–5.
S. D’Oro, E. Ekici, and S. Palazzo, “Optimal power allocation and scheduling under jamming attacks,” IEEE/ACM Transactions on Networking, vol. 25, no. 3, pp. 1310–1323, 2016. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4832,
1
],
[
4832,
10384,
2
],
[
10384,
15718,
3
],
[
15718,
21299,
4
],
[
21299,
27262,
5
],
[
27262,
33994,
6
],
[
33994,
37564,
7
]
]
} |
56f51fb7f74e1e7676d432062652cd47bbba824e | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 12,
"total-fallback-pages": 0,
"total-input-tokens": 42084,
"total-output-tokens": 15962,
"fieldofstudy": [
"Biology"
]
} | Discovery and characterization of ionic liquid-tolerant thermophilic cellulases from a switchgrass-adapted microbial community
John M Gladden¹²*, Joshua I Park¹²⁷, Jessica Bergmann⁵, Vimalier Reyes-Ortiz¹, Patrik D’haeseleer³, Betania F Quirino⁵⁶, Kenneth L Sale¹², Blake A Simmons¹² and Steven W Singer¹⁴
Abstract
**Background:** The development of advanced biofuels from lignocellulosic biomass will require the use of both efficient pretreatment methods and new biomass-deconstructing enzyme cocktails to generate sugars from lignocellulosic substrates. Certain ionic liquids (ILs) have emerged as a promising class of compounds for biomass pretreatment and have been demonstrated to reduce the recalcitrance of biomass for enzymatic hydrolysis. However, current commercial cellulase cocktails are strongly inhibited by most of the ILs that are effective biomass pretreatment solvents. Fortunately, recent research has shown that IL-tolerant cocktails can be formulated and are functional on lignocellulosic biomass. This study sought to expand the list of known IL-tolerant cellulases to further enable IL-tolerant cocktail development by developing a combined in vitro/in vivo screening pipeline for metagenome-derived genes.
**Results:** Thirty-seven predicted cellulases derived from a thermophilic switchgrass-adapted microbial community were screened in this study. Eighteen of the twenty-one enzymes that expressed well in *E. coli* were active in the presence of the IL 1-ethyl-3-methylimidazolium acetate ([C2mim][OAc]) concentrations of at least 10% (v/v), with several retaining activity in the presence of 40% (v/v), which is currently the highest reported tolerance to [C2mim][OAc] for any cellulase. In addition, the optimum temperatures of the enzymes ranged from 45 to 95°C and the pH optimum ranged from 5.5 to 7.5, indicating these enzymes can be used to construct cellulase cocktails that function under a broad range of temperature, pH and IL concentrations.
**Conclusions:** This study characterized in detail twenty-one cellulose-degrading enzymes derived from a thermophilic microbial community and found that 70% of them were [C2mim][OAc]-tolerant. A comparison of optimum temperature and [C2mim][OAc]-tolerance demonstrates that a positive correlation exists between these properties for those enzymes with an optimum temperature >70°C, further strengthening the link between thermotolerance and IL-tolerance for lignocellulosic glycoside hydrolases.
**Keywords:** Cellulase, Ionic liquid, Thermophilic, Biofuel
some form of chemical or physical pretreatment to permit enzymes or chemicals to gain access to and hydrolyze the plant polymers into fermentable sugars [4,6,8]. This study focuses on this challenge and discloses the discovery and characterization of biomass-deconstructing enzymes that are more compatible with certain forms of biomass pretreatment solvents than the current commercially available enzyme cocktails.
The recalcitrance of lignocellulosic biomass has been a difficult hurdle to overcome, but promising new technologies using certain ionic liquids (ILs) that have come about in the last decade indicate that we are well on our way to moving past this barrier [9,10]. Pretreating biomass with certain classes of ILs, most notably those with imidazolium-based cations, can be more efficient and tunable than other existing forms of pretreatment, and technoeconomic analysis of IL-pretreatment suggests that there are potential routes to economic viability [8,11]. One remaining issue with this technology that needs to be addressed to maximize efficiency and reduce capital costs is the incompatibility of ILs with cellulase cocktails derived from filamentous fungi. These enzyme cocktails can be strongly inhibited by certain ILs, such as 1-ethyl-3-methylimidazolium acetate [C2mim][OAc], necessitating expensive and inefficient washing steps to remove residual IL from the biomass prior to addition of enzymes [8,12-14]. One solution to this issue is to develop enzyme cocktails that are tolerant to ILs. Fortunately, it has been shown that certain thermophilic bacterial cellulase enzymes can tolerate high levels of the IL [C2mim][OAc], and in fact these enzymes have been used to develop an IL-tolerant cellulase cocktail called JTherm [13-17]. It has been further demonstrated that JTherm can be used in a one-pot IL pretreatment and saccharification bioprocessing scheme that eliminates the need to wash the pretreated biomass with water, significantly reducing the number of process steps [18].
The next step toward an economically viable IL-based bioprocessing scheme for the conversion of lignocellulosic biomass to biofuels will be to further integrate and improve all components of the process. For IL-tolerant cellulase cocktails, this includes reducing enzyme loadings by reformulating the cocktails to achieve greater saccharification efficiency; a modification predicted by technoeconomic modeling to substantially reduce overall costs in a biorefinery [11]. Enzyme cocktail reformulation will require screening through an expansive list of IL-tolerant cellulase enzymes to identify those that enhance saccharification efficiency under conditions likely to be found in a biorefinery. However, the number of known IL-tolerant cellulase enzymes, specifically those tolerant to [C2mim][OAc], is quite small, an issue that hampers cocktail reformulation efforts. The goal of this study was therefore to discover and characterize an expanded set of [C2mim][OAc]-tolerant cellulase enzymes to enable future development of highly efficient IL-tolerant biomass-deconstructing enzyme cocktails. This study focused on thermophilic organisms based on clues provided by previous studies that indicate that thermostolerance may be positively correlated with IL-tolerance. Hence, we can leverage a naturally evolved physiological characteristic of an enzyme and use it as a proxy to discover enzymes with a non-natural industrially relevant characteristic, such as IL-tolerance.
This concept drove recent work where complex compost-derived microbial communities were cultivated on switchgrass under thermophilic conditions to enrich for organisms that produce mixtures of IL-tolerant cellulases and xylanases [14]. The community was composed of several abundant bacterial populations related to Thermus thermophilus, Rhodothermus marinus, Paenibacillus, Thermobacillus and an uncultivated lineage in the Gemmatimonadetes phylum [19]. The glycoside hydrolases from this community were found to have high optimum temperatures (approximately 80°C) and tolerated relatively high levels of [C2mim][OAc] compared to commercial cellulase cocktails (>50% activity in the presence of 30% (v/v) [C2mim][OAc]). Therefore, these communities provide a rich reservoir of potential enzyme targets to develop thermophilic and IL-tolerant cellulase cocktails to be used in lignocellulosic biofuel production platforms. To discover the genes that encode these IL- and thermo-tolerant enzymes, metagenomic and proteomic analysis was conducted on the community [14,19]. The analysis identified a variety of genes encoding potential cellulose and hemicellulose-degrading enzymes, a subset of which were assembled into complete open reading frames (ORFs) from the metagenome. To validate the concept that thermostolerance can be used as an engine for discovery of IL-tolerant enzymes, this study expressed and characterized 37 of these predicted cellulase genes from the metagenome using both cell-free and in vivo Escherichia coli (E. coli) expression systems. Both expression methods were employed to determine which method is most suitable for rapid and efficient screening of metagenome-derived gene sets. We found that several of the ORFs encode IL- and thermo-tolerant cellulase enzymes, including enzymes with activities that are stimulated in the presence of ILs.
Results
Identification of cellulases in a switchgrass-adapted metagenome
The metagenome of a thermophilic switchgrass-degrading bacterial community was curated for genes with cellulase-related annotations or homology to sequences for cellulase enzymes deposited in the CAZy database (http://www.cazy.org/), including β-glucosidases (BG), cellobiohydrolases (CBH), and endoglucanases (Endo). A total of nineteen predicted BGs, two CBHs, and sixteen Endos were
identified that appeared to be complete ORFs (Table 1; see Methods). The top BLASTP hit for each identified cellulase is indicated in Table 1, including the maximum identity and source organism of the top hit in GenBank. Many of the ORFs are homologous to those found in isolates that cluster with abundant community members, such as \textit{Rhodothermus marinus}, \textit{Paenibacillus}, \textit{Thermobacillus} and \textit{Gemmatimonadetes}. Many of the ORFs fall into sequence bins assigned to these organisms in the metagenome that are consistent with the phylogenetic affiliation predicted by the BLASTP search (Table 1, Additional file 1, and D’Haeseleer et al. 2013 [19]). Several of the ORFs in Table 1 contained sequencing errors or were identified as fragments and were manually corrected/assembled (see Methods for details). For J08/09 and J38/39, the manual assembly resulted in two closely related proteins, and therefore both versions were tested.
**Cell-free and E. coli expression and screening of predicted cellulase genes**
Each of the 37 predicted metagenome-derived cellulase genes were synthesized and cloned into a custom vector for \textit{in vitro} cell-free expression using a T7 promoter/terminator-based system [20]. Each gene was expressed \textit{in vitro} and screened for Endo, CBH and BG activity (Table 2). For comparison to the cell-free system, each gene was then cloned into the pDEST17 vector for expression in \textit{E. coli} and screened for the same activities (Table 2). There was a large degree of overlap in terms of active genes detected between the two expression methods, but the \textit{E. coli}-based screen detected activity from a larger subset of genes than the cell-free screen (26 versus 19). BG activity was detected for 15 of the 19 predicted BGs, and none of these enzymes showed Endo activity, consistent with their annotation assigned by the JGI and D’Haeseleer et al. [19]. Furthermore, 12 of these 15 positive candidates exhibited CBH activity on, indicating that these enzymes have activity on glucose oligomers with n ≥2. For the predicted Endos, activity was detected for 11 of the 16 candidates. In addition to Endo activity, 7 of the 11 Endos also had BG and/or CBH activity. No activity was detected for the two predicted CBH genes.
**Activity profile of cellulases**
Of the 37 enzymes in the initial screen, 15 of the 19 BGs and 6 of the 16 Endos were expressed at sufficient quantities to profile in greater detail. The activity of each enzyme was measured at temperatures ranging from 45 to 99°C, pH between 4.0 and 8.0, and IL concentrations ranging from 0 to 40% [C2mim][OAc] (v/v). These data were then plotted and optimal temperature/pH and IL-tolerance was determined for each enzyme (Table 3). To illustrate the dynamic activity range of each enzyme, the temperature, pH and IL concentration ranges that gave greater than 80 or 50% activity compared to the optimal activity are also reported in Table 3. All of the enzymes were active at elevated temperature, but the range of optimum temperatures (Topt) was broad, ranging from 45 to 95°C. The enzymes were divided into two groups: seven enzymes with a Topt within 5° of 70°C and another seven near 90°C. Of the remaining enzymes, five had a Topt below 70°C and two had an intermediate Topt of 80°C. The enzymes also showed a similar clustering around optimal pH values (pHopt), with fourteen enzymes having a slightly acidic pHopt between 5.0 and 6.0 and the remaining seven enzymes having a pHopt between 6.5 and 7.5. However, many of these enzymes were active over a broad pH range, and all but J16 retained ≥50% activity at pH 7.0. Five of the enzymes were more than 80% active at the highest pH tested of 8.0, indicating that these enzymes also tolerate slightly alkaline conditions.
Enzyme activity was profiled at temperatures (Temp) between 45 and 95°C, pH between 4 and 8, and IL concentrations between 0 and 40% (v/v) of 1-ethyl-3-methylimidazolium acetate [C2mim][OAc]. The temperature and pH that elicited the highest activity is indicated in the row for optimum temperature (Topt) and optimal PH values (pHopt), respectively. Temperature and pH ranges that permitted greater than 80% and 50% activity are indicated below the optimum value. Ionic liquid (IL)-tolerance is indicated as the maximum concentration of [C2mim][OAc] that permits at least 80% and 50% enzyme activity (that is, a value of 15 in the 80% row would indicate that 15% (v/v) of [C2mim][OAc] is the maximum concentration of [C2mim][OAc] that can be used to retain at least 80% enzyme activity). Most enzymes showed a steady decline in activity with increasing IL concentrations. *Maximum (Max) activity in IL is reported as the highest fold change of activity in the presence of IL compared to water and the values in brackets are the IL concentrations (v/v) in which that highest activity was achieved. Values less than 1 indicate the enzyme is less active in IL than in water, and values greater than 1 indicate the enzyme had increased activity in the presence of IL.
Surprisingly, most of the enzymes (16 of the 21 tested) showed an initial increase in activity in the presence of [C2mim][OAc] compared to water (0% IL), with a 15 to 500% enhancement in activity that eventually declined at higher [C2mim][OAc] concentrations (Table 3). This phenomenon is illustrated in the row labeled “Max activity in IL” in Table 3 that lists the highest fold change in activity in the presence of [C2mim][OAc]. For example, enzyme J16 was found to be five times more active in 10% (v/v) [C2mim][OAc] than in water. The majority of the enzymes were active in at least 20% (v/v) [C2mim][OAc] and maintained greater than 50% activity. Six of
Table 1 Predicted cellulase enzymes identified in the switchgrass-adapted metagenome
| Gene ID | IMG gene ID | GH family | Predicted function | Max identity (%) | Genbank accession number | Top Blast-hit organism | Metagenome bin* |
|---------|-------------|-----------|--------------------|------------------|--------------------------|------------------------|-----------------|
| J01 | 2061974227 | 3 | β-glucosidase | 42 | ZP_06970881.1 | Ktedonobacter racemifer DSM 44963 | Paenibacillus |
| J02 | 2061976655 | 3 | β-glucosidase | 97 | YP_003321925.1 | Thermobaculum terrenum | Thermobaculum |
| J03 | 2061976732 | 3 | β-glucosidase | 96 | YP_003322827.1 | Thermobaculum terrenum | Thermobaculum |
| J04 | 2061977694 | 1 | β-glucosidase | 62 | ZP_10205923.1 | Rhodanobacter thiocyanodys LCS2 | Gemmatimonadetes |
| J05 | 2061979262 | 3 | β-glucosidase | 44 | YP_002760449.1 | Gemmatimonas aurantiaca T-27 | Gemmatimonadetes |
| J06 | 2061979786 | 1 | β-glucosidase | 61 | NP_242789.1 | Bacillus halodurans C-125 | Not binned |
| J07 | 2061980390 | 1 | β-glucosidase | 66 | YP_003323667.1 | Thermobaculum terrenum ATCC BAA-798 | Not binned |
| J08 | 2062002762 | 1 | β-glucosidase | 99 | YP_003323667.1 | Thermobaculum terrenum ATCC BAA-798 | Not binned |
| J09 | 2062007681 | 3 | β-glucosidase | 75 | YP_823953.1 | Candidatus Solibacter usitatus Ellin6076 | Not binned |
| J10 | 2062002993 | 3 | β-glucosidase | 77 | ZP_09004353.1 | Paenibacillus lactis 154 | Not binned |
| J11 | 2062005533 | 3 | β-glucosidase | 42 | ZP_06970881.1 | Ktedonobacter racemifer DSM 44963 | Not binned |
| J12 | 2062006736 | 3 | β-glucosidase | 94 | YP_003291338.1 | Rhodothermus marinus DSM 4252 | Rhodothermus |
| J13 | 2062007625 | 1 | β-glucosidase | 93 | YP_003318753.1 | Sphaerobacter thermophilus DSM 20745 | Sphaerobacter |
| J14 | 2062008681 | 3 | β-glucosidase | 97 | YP_003324065.1 | Thermobaculum terrenum ATCC BAA-798 | Sphaerobacter |
| J15 | 2062012385 | 3 | β-glucosidase | 75 | YP_823953.1 | Candidatus Solibacter usitatus Ellin6076 | Not binned |
| J16 | 2062018481 | 3 | β-glucosidase | 100 | YP_004824792.1 | Rhodothermus marinus SG0.5JP17-172 | Rhodothermus |
| J17 | 2062019328 | 3 | β-glucosidase | 71 | ZP_08918857.1 | Thermobacillus composti KWC4 | Paenibacillus |
| J18 | 2062019735 | 1 | β-glucosidase | 99 | AAN05441.1 | Thermus sp. IB-21 | Thermus |
| J19 | 2062026722 | 1 | β-glucosidase | 72 | YP_002522957.1 | Thermomicrobium roseum DSM 5159 | Thermomicrobium |
| J21 | 2061975668 | 9 | Endoglucanase | 54 | YP_002759529.1 | Gemmatimonas aurantiaca T-27 | Gemmatimonadetes |
| J22 | 2061976479 | 8 | Endoglucanase | 72 | BAF49077.1 | Paenibacillus sp. W-61 | Paenibacillus |
| J23 | 2061977143 | 5 | Endoglucanase | 32 | ZP_09216417.1 | Gordonia amarae NBRC 15530 | Sphaerobacter2 |
| J24 | 2061979932 | 9 | Endoglucanase | 54 | ACJ68032.1 | Paenibacillus provencensis | Paenibacillus |
| J25 | 2061986269 | 12 | Endoglucanase | 98 | YP_004824941.1 | Rhodothermus marinus SG0.5JP17-172 | Rhodothermus2 |
| J26 | 2061990001 | 12 | Endoglucanase | 100 | YP_004824941.1 | Rhodothermus marinus SG0.5JP17-172 | Not binned |
| J27 | 2061990054 | 5 | Endoglucanase | 35 | ZP_09309733.1 | Rhodococcus pyridinivorans AK37 | Sphaerobacter2 |
| J28 | 2061994288 | 5 | Endoglucanase | 98 | YP_003323917.1 | Thermobaculum terrenum ATCC BAA-798 | Sphaerobacter |
| J29 | 2062006179 | 5 | Endoglucanase | 52 | BAJ2227.1 | Paenibacillus sp. KSM-NS46 | Paenibacillus |
| J30 | 2062016312 | 9 | Endoglucanase | 54 | ZP_08919343.1 | Thermobacillus composti KWC4 | Not binned |
| J31 | 2062017860 | 5 | Endoglucanase | 57 | ZP_08873206.1 | Vermineprobacter arporrectodeae | Not binned |
| J32 | 2062025020 | 5 | Endoglucanase | 96 | YP_003320228.1 | Sphaerobacter thermophilus DSM 20745 | Not binned |
| J33 | 2062027867 | 8 | Endoglucanase | 72 | ZP_04851456.1 | Paenibacillus sp. oral taxon 786 str. D14 | Not binned |
| J34 | 2062029826 | 6 | Endoglucanase | 37 | ZP_06416445.1 | Frankia sp. EUN1f | Thermobaculum |
| J35 | 2062032441 | 5 | Endoglucanase | 35 | ZP_08873206.1 | Vermineprobacter arporrectodeae | Not Binned |
the enzymes (J03, J05, J16, J25, J26 and J36) maintained more than 80% activity in 35 to 40% [C2mim][OAc]. Only a single enzyme, J15, lost activity at low [C2mim] [OAc] concentrations. The BG enzymes J5 and J16 and Endo enzymes J26 and J36 showed the highest increase in activity in the presence of [C2mim][OAc]. To examine the relationship of IL-tolerance to potential halotolerance, their activity was measured in equal molar concentrations of [C2mim][OAc] and sodium acetate (NaOAc) (Figure 1A-B). Each of these enzymes also showed greater or equal activity in the presence of NaOAc, despite this salt buffering the solution at a more basic pH, which tends to be outside the optimal activity range for these enzymes (in water), especially J16 (Figure 1C-D).
The $T_{opt}$ and $pH_{opt}$ values of these enzymes were compared to their IL-tolerance to determine whether either of these properties positively correlates with high IL-tolerance. A plot of the optimum temperature or pH of the enzyme versus the highest concentration of [C2mim][OAc] in which the enzyme retains ≥80% of its activity was examined for clustering of values that would indicate that a particular range of pH or temperature positively correlates with high IL-tolerance. Of these two properties, only the $T_{opt}$ showed any discernible correlation with high IL-tolerance (Figure 2). It appears that a $T_{opt}$ >70°C is a positive indicator of high IL-tolerance. Enzymes with a $T_{opt}$ ≤70°C have only an 18% probability of being highly tolerant to [C2mim][OAc], whereas enzymes with a $T_{opt}$ >70°C have a 78% chance of being highly IL-tolerant (see Figure 2 legend for details).
**Discussion**
Developing IL-tolerant enzymatic mixtures for cellulose hydrolysis will permit the advancement of technologies that combine IL-based pretreatment using [C2mim] [OAc] with enzymatic hydrolysis. This type of process intensification will be critical for the development of cost-competitive lignocellulosic biofuel technologies [11]. However, there are few IL-tolerant enzymes known and more must be discovered before these technologies can be matured to the point of large-scale implementation in a biorefinery. This study was based on the hypothesis that thermotolerance and IL-tolerance are correlated, and therefore sought to expand the list of known IL-tolerant enzymes by identifying, expressing, and characterizing multiple thermophilic biomass deconstructing enzymes sourced from a single compost-derived microbial community that was previously used as a test bed for comparing IL and thermotolerance [14,19]. In the course of this study, we compared cell-free and *in vivo* *E. coli* expression methods for rapidly (and with high fidelity) screening through predicted enzyme candidates to narrow down the list of targets to functional and properly annotated enzymes. Results from this study elicit several interesting conclusions regarding the utility of *in vitro* versus *in vivo* screening methods, the activity of the recombinant enzymes versus the native enzymes from the parent microbial community, and the hypothesis that thermotolerance and IL-tolerance are correlated.
Comparison of the cell-free and *in vivo* *E. coli* screens yielded several observations: 1) both screens work well at quickly screening through candidate genes to identify functional genes; 2) the screens produce similar results in regards to predicted annotation and 3) the cell-free screen is more rapid (24 hours) compared to the *in vivo* screen (5 days); however, 4) the cell-free screen missed about 27% of the positive candidates (19 versus 26), and 5) the cell-free screen will eventually require porting into an *in vivo* expression system to conduct more detailed enzyme profiling. In light of these observations, the cell-free screen would be advantageous if the number of candidates to screen is large, as it is more rapid and less labor-intensive than the *in vivo* screen, whereas the *in vivo* screen would be more advantageous in smaller screens as it provides greater returns and enables more detailed characterization. Overall, the assigned annotation of each enzyme accurately reflected its measured activity. Several enzymes showed activity on multiple substrates, but in most cases the highest measured activity matched the annotation of the enzyme.
After the initial screening, there were 21 promising enzyme targets (15 BG and 6 Endo) to profile in more detail for optimum temperature, pH and IL-tolerance. The profiles revealed that the enzymes are indeed thermotolerant ($T_{opt}$ between 45 and 95°C), and the two clusters of optimum temperatures observed for these enzymes (70 and 90°C) mirror the pattern seen in the profile of the native enzymes produced by the parent community from which these genes were isolated, except that the
Table 2 Screen of predicted glycoside hydrolase enzymes for β-glucosidase, endoglucanase, and cellobiohydrolase activity
| Gene ID | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 |
|---------|---|---|---|---|---|---|---|---|---|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|
| Endo | | | | | | | | | | + | + | + | + | + | + | + | + | + | + | + | + | + | + | Cell-free |
| CBH | | | | | | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + | Cell-free |
| βG | | | | | | | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + | + | Cell-free |
Cell-free and in vivo expressed enzymes are labeled in the far right column. Enzyme activities are as follows: Endoglucanase (Endo), cellobiohydrolase (CBH), and β-glucosidase (βG). Detection of enzymatic activity is indicated with a + for positive and a blank cell for negative.
| Gene ID | 01 | 02 | 03 | 05 | 06 | 07 | 08 | 09 | 11 | 14 | 15 | 16 | 17 | 18 | 19 | 24 | 25 | 26 | 29 | 30 | 36 |
|---------|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|----|
| \( T_{\text{opt}} \) (°C) | 45 | 90 | 75 | 70 | 65 | 70 | 90 | 90 | 60 | 70 | 70 | 80 | 60 | 95 | 80 | 55 | 95 | 95 | 65 | 50 | 95 |
| Temp (°C) \( \geq 80\% \) activity | 45–50 | 80–90 | 75 | 60–75 | 55–70 | 65–70 | 80–90 | 80–90 | 55–60 | 60–75 | 65–70 | 75–80 | 55–60 | 85–95 | 70–85 | 50–60 | 80–95 | 85–95 | 55–70 | 45–55 | 75–95 |
| \( \text{pH}_{\text{opt}} \) | 6 | 7 | 5.5 | 7 | 6 | 6 | 5.5 | 5.5 | 6 | 6 | 6.5 | 5 | 6 | 5.5 | 6.5 | 7.5 | 7.5 | 7.5 | 6 | 6 |
| Temp (°C) \( \geq 50\% \) activity | 6–6.5 | 5.5–8 | 4.5–6 | 6–7.5 | 5–6.5 | 6–7.5 | 4–8 | 4–8 | 5–6.5 | 5.5–7.5 | 6–7 | 5–5.5 | 5–7 | 4.5–7 | 4.5–7.5 | 5.5–7.5 | 4–8 | 5.5–8 | 5.5–8 | 5–7 | 6–7.5 |
| \( \text{pH} \geq 80\% \) activity | 5–7 | 4.5–8 | 4.5–7 | 5.5–8 | 4.5–7.5 | 5.5–8 | 4–8 | 4–8 | 4.5–7 | 5–8 | 6–7.5 | 5–6.5 | 5–7.5 | 4–8 | 4.5–8 | 5–8 | 4–8 | 5–8 | 4.5–8 | 5.5–8 |
| \( \text{pH} \geq 50\% \) activity | 0 | 15 | 40 | 35 | 5 | 10 | 0 | 0 | 10 | 5 | 0 | 35 | 0 | 5 | 10 | 10 | 40 | 30 | 5 | 25 | 35 |
| IL% (v/v) \( \geq 100\% \) activity | 0 | 25 | 40 | 40 | 5 | 15 | 5 | 5 | 15 | 10 | 0 | 40 | 0 | 15 | 20 | 10 | 40 | 35 | 10 | 30 | 35 |
| IL% (v/v) \( \geq 80\% \) activity | 15 | 35 | 40 | 40 | 5 | 15 | 20 | 20 | 20 | 20 | 0 | 40 | 5 | 30 | 5 | 30 | 15 | 40 | 35 | 15 | 35 |
| IL% (v/v) \( \geq 50\% \) activity | 0.68 (5) | 1.1 (5) | 1.2 (40) | 2.1 (15) | 1.3 (5) | 1.2 (5) | 0.87 (5) | 0.89 (5) | 1.5 (15) | 1.2 (5) | 0.45 (5) | 5 (10) | 0.5 (5) | 1.2 (5) | 1.1 (5) | 2.1 (5) | 1.9 (15) | 2.5 (15) | 1.4 (5) | 2.5 (15) | 2 (25) |
Max activity in IL* indicated the highest activity in row \( T_{\text{opt}} \) and \( \text{pH}_{\text{opt}} \), respectively. Temperature and pH ranges that permitted greater than 80% and 50% activity are indicated below the optimum value. IL-tolerance is indicated as the maximum concentration of [C2mim][OAc] that permits at least 80% and 50% enzyme activity (i.e. a value of 15 in the 80% row would indicate that 15% (v/v) of [C2mim][OAc] is the maximum concentration of [C2mim][OAc] that can be used to retain at least 80% enzyme activity). Most enzymes showed a steady decline in activity with increasing IL concentrations. *Max activity in IL is reported as the highest fold change of activity in the presence of IL compared to water and the () indicates the IL concentration (v/v) in which that highest activity as achieved. Values less than 1 indicate the enzyme is less active in IL than in water while values greater than 1 indicate the enzyme has increased activity in the presence of IL.
**Figure 1** Plot of enzyme activity in the presence of 0 to 40% 1-ethyl-3-methylimidazolium acetate ([C2mim][OAc]) or an equal molarity of sodium acetate (NaOAc). Relative activity is based on activity in water (0% ionic liquid (IL) value). (A) Two IL-tolerant β-glucosidases and (B) two IL-tolerant endoglucanases were profiled. The pH was determined at each concentration of (C) [C2mim][OAc] and (D) NaOAc. Error bars represent one standard deviation (they are too small to be visualized on C and D).
**Figure 2** A plot highlighting the correlation between thermotolerance and ionic liquid (IL)-tolerance of the enzymes shown in Table 3. The plot shows the maximum [C2mim][OAc] concentration that permits ≥80% enzyme activity compared to water versus the optimum temperature ($T_{opt}$) of the enzyme. There are two overlapping data points at 95°C, 35% IL. Enzymes with high IL-tolerance are defined as the enzymes that can tolerate ≥20% (v/v) [C2mim][OAc] or greater (above horizontal line). The enzymes fall into two clusters: the black polygon where 78% (7/9) of the enzymes with a $T_{opt} > 70°C$ have high IL-tolerance, and the grey polygon where 82% (9/11) of the enzymes with a $T_{opt} ≤ 70°C$ have low or no IL-tolerance. Only 18% (2/11) of the enzymes with a $T_{opt} ≤ 70°C$ have high IL-tolerance.
native enzymes had their had two $T_{opt}$ peaks 10° lower than the heterologous enzymes (60 and 80°C) [14]. It is unclear why this may be. Perhaps the community produces a complex mixture of enzymes, the average of which results in observed $T_{opt}$ at around 60 and 80°C, or the community only expresses a complement of enzymes with $T_{opt}$ near 60 and 80°C.
The native enzymes produced by the parent microbial community were also [C2mim][OAc]-tolerant, a trait mirrored by the majority of enzymes profiled in this study. However, unlike the recombinant enzymes in this study, the native cellulase enzymes were not observed to have an increase in activity in the presence of ILs [14].
Many of the enzymes in this study showed an increase in activity in the lower range of [C2mim][OAc] concentrations tested (0.3 to 0.9 M), some several fold higher than the activity in water. The fact that several of these enzymes also showed increased activity in the presence of NaOAc suggests that these enzymes may require the presence of salt for optimal activity. The increase in activity with NaOAc was not as high for enzyme J16 as in the corresponding concentration of IL, which is likely due to the more basic pH of NaOAc and the lower pH optimum of J16 (pH 5.0). This phenomenon was less apparent for the other enzymes tested, but generally the enzymes demonstrated relatively higher levels of activity in the presence of [C2mim][OAc] compared to NaOAc. This apparent IL- and salt-tolerance is not surprising, considering that these enzymes are similar to those derived from thermotolerant and slightly halo-tolerant organisms like *Rhodothermus marinus*, which requires salt and grows optimally in about 0.3 M NaCl [21]. Unlike many fungal enzymes, these cellulases tend to prefer more neutral pH (6.0 or 7.0), and many retained more than 80% activity at the highest pH tested of 8.0. [C2mim][OAc] buffers around neutral pH in the range of concentrations tested, a property that may further ameliorate tolerance to this IL by several of the enzymes tested. The affinity of these enzymes for more neutral pH may reflect their origin; for example, *R. marinus* grows optimally at pH 7.0 [21].
The mechanisms of IL-tolerance are not well understood; few enzymes have been investigated for IL-tolerance in general and there are no studies that have looked at a large enough set of enzymes with a single type of IL, such as [C2mim][OAc], to do any type of thorough comparative analysis. The 21 enzymes characterized in this study had varying degrees of [C2mim][OAc]-tolerance and therefore provide an opportunity to look for correlations between IL-tolerance and other characteristics of the enzymes, that is, $T_{opt}$ and pH ranges. Of those two properties, there only appears to be a correlation between IL-tolerance and $T_{opt}$ consistent with the conclusion from studies of other thermosttolerant enzymes. A comparison of the IL-tolerance and $T_{opt}$ revealed that the enzymes with a $T_{opt}$ >70°C tend to have a higher probability of tolerating high concentrations of [C2mim][OAc]. This indicates that evolution towards higher $T_{opt}$ frequently alters the properties of an enzyme in a manner that also promotes tolerance to [C2mim][OAc]. The data from this study also indicates that the correlation is not simply between thermostolerance and IL-tolerance but more specifically hyperthermotolerance and IL-tolerance. Only a single enzyme studied with a $T_{opt}$ <70°C displayed appreciable levels of IL-tolerance. This observation helps explain why enzymes from filamentous fungi used in commercial cellulase cocktails do not display IL-tolerance; there are no known hypertherophilic filamentous fungi. Furthermore, future studies aimed at studying the mechanisms of [C2mim][OAc]-tolerance may benefit from a refined hypothesis that hyperthermotolerance (>70°C) is correlated with IL-tolerance.
The results presented here can also be used to comment on the general strategy used to identify enzymes with a particular set of characteristics, in this case IL-tolerance. The microbial community from which these enzymes were derived was originally established under the premise that organisms endowed with a particular functionality could be selectively enriched in abundance from a complex microbial community by cultivation under defined conditions. This selective enrichment could then help researchers target organisms and genes with a desired set of characteristics. In this case, the desired functionality was production of cellulase enzymes and the desired characteristic was thermo- and IL-tolerant cellulase enzymes. This strategy was implemented by cultivating a microbial community derived from green-waste compost under thermophilic conditions with plant biomass as a sole carbon source [14]. The native enzymes produced by this community were both thermo- and IL-tolerant and so were the recombinant enzymes derived from this community, suggesting that selective cultivation is a good method for discovering enzymes that function under a desired set of conditions.
**Conclusions**
The enzymes characterized in this report are some of the most tolerant to [C2mim][OAc] reported to date [12,14,15,17]. Tolerance to this particular IL is of increasing interest as it is currently one of the most effective and well-studied ILs for pretreatment of lignocellulosic biomass [22]. Recent efforts to develop IL-tolerant cellulase cocktails and to incorporate these cocktails into one-pot pretreatment and saccharification bioprocessing schemes show that IL-tolerant enzymes can be used to develop new technologies to deconstruct biomass, and open up the technological landscape for lignocellulosic biorefineries [18].
The enzymes described in this report can be used to further those technologies.
**Methods**
**Manual cellulase gene assembly**
Although most of the full-length ORFs in Table 1 were taken directly from the metagenome, several were manually reconstructed from fragmented genes identified in the assembly of the metagenomic dataset. The following ORFs were manually assembled: J03 had an incorrectly predicted start codon. The start of this ORF was moved 5’ to match the start of its top BLAST hit. J08/09 are two versions of a single ORF composed of four gene fragments from the metagenome (IMG gene IDs 2061981261, 2062002762, 2062037967, 2061992858), which all have very high homology with a predicted BG from *Thermobaculum terrestrum* ATCC BAA-798 [GenBank: ACZ42845.1]. J08 is an assembly of 2061981261 (N-terminus), 2062002762 (C-terminus), and ACZ42845.1 (sequence that encodes AAVITENGAYPDE inserted between the two sequences), and J09 is a compilation of 2062037967, 2061992858, and the same fragment from ACZ42845.1 assembled in the same order as J08. Overall, J08 and J09 differ by 5 AA. The same situation applies to J10, which is assembled from 2062002992 (N-terminus), 2062002993 (C-terminus), and a middle fragment (sequence encoding NAVKVTAAA) from ACX65411.1, a glycoside hydrolase family 3 protein from *Geobacillus sp.* Y412MC1.
J11 was also assembled in the same manner; two consecutive ORFs (2062005533 and 2062005534) were merged with a fragment encoding (YVR) derived from a glycoside hydrolase family 3 protein from *Ktedonobacter racemifer* DSM 44963 (EFH83601.1). J38/39 are two versions of two consecutive ORFs (2062019305, and 2062019306), which may be separated by a single base pair frame-shift or a larger deletion. J38 is a merger of the two ORFs by inserting a single base pair to encode a leucine codon at residue 103. J39 is a merger of the two ORFs with a 316 base pair insertion at the same location derived from a BG from *Paenibacillus sp.* JDR-2 (ACT00588.1), to repair the glycoside hydrolase family 3 N-terminal domain.
**Gene synthesis and cloning**
Each gene was codon-optimized for expression in *E. coli* and synthesized by Genscript (Piscataway, NJ, USA). They were then cloned into a modified pUC57 vector constructed at Genscript, pUC57CFv1, with an added T7 promoter and terminator, as well as gateway attB1/attB2 sequences flanking the ORF, and a 8 × C-terminal 8 × His and Strep-tag II dual tag. There was an in-frame NheI-XhoI cloning site added between the attB1/attB2 sequences to place the ORFs into the pUC57CFv1 vector. The added vector sequences were cloned into the pUC57 vector at the EcoRI and SacI sites. Synthesized ORFs were then cloned into the pUC57CFv1 vector at the NheI-XhoI sites. The synthesized genes in the pUC57CFv1 vector were transformed in to TOP10 *E. coli* for storage at −80°C.
The T7, Gateway attB1/attB2 and His tag sequences added to pUC57 are:
```
GAATTCTAAATATAGACTCAGCATATTAGAGGAGACCCACAACGTTTCCTCTAGAAATAATTGTGTTTA
ACTTTAAGAAGGAGATATACTAGACAAAGGTGTGATACAAAAAGCAGGCTTCGCTAGCCCAATCCAATCTCGAGGACCACCCACTTTCTTGGTACAAAGTGGTCCAT
CATCACCACATCAGGATTACAATAACTAGCATAACCCCTTTGGGCCTCTAAGCGGTCTTGAGGGGTTT
TTTGGAGAAGCTCTTCTGTTAGG
```
**In vitro and in vivo expression of cellulases**
Each of the 37 cellulases was expressed in vitro using the RTS 100 *E. coli* 100 Hy cell-free expression Kit (Roche Diagnostics, Mannheim, Germany, Catalogue Number 03 186 148 001), using 0.5 μg of vector and following the manufacturer’s instructions. The lyophilized plasmids were dissolved in DNase/RNase-free water before use. The *in vitro* protein expression was performed at 30°C for six hours. The expression products were used immediately for enzyme assay reactions.
To validate the enzyme activity results of *in vitro* protein expression and assays, the cellulase genes were cloned into the low-copy bacterial expression plasmid pDEST17 by Gateway cloning techniques following the manufacturer’s instructions (Invitrogen). The sequences of all cloned genes in the pDONR221 and pDEST17 vectors were verified by DNA sequencing (Quintara Biosciences; Albany, CA, USA). All cellulase genes in the pDEST17 vector, except J24 and J29, were transformed into BL21(DE3)Star *E. coli* (Invitrogen, Carlsbad, CA, USA). The J24 and J29 genes in the pDEST17 vector were transformed into the T7 Express I*®* *E. coli* strain (New England BioLabs, Ipswich, Massachusetts, USA) to attenuate the basal level of cellulase expression during the growth phase prior to induction of protein expression. This was done because the expression vectors containing J24 and J29 were toxic to TOP10 and BL21(DE3)Star strains of *E. coli*, presumably due to the leaky activation of the T7 promoter. Bacterial cultures were grown in 96-deep well-plates in 800 μL of Luria-Bertani (LB) Miller broth containing carbenicillin (50 μg/ml) in each well. The overnight cultures of *E. coli* were inoculated to fresh LB medium containing Overnight Express Autoduction System 1 (Calbiochem, San Diego, CA, USA) reagent and carbenicillin. In the autoinduction medium, the bacterial cultures were incubated at 37°C with constant shaking at 200 rpm for the first four hours. Then the cultures were grown at 30°C for 18
hours with constant shaking at 200 rpm. The cell pellets were harvested by centrifugation at 6,000 g for 30 minutes, and then stored at −20°C. Each of the frozen cell pellets was thawed and resuspended in 0.1 mL of BugBuster containing lysozyme (1 mg/mL), benzonase (25 U/mL) and phenylmethanesulfonylfluoride (PMSF) (1 mM). After 30 minutes of incubation at room temperature, the cell lysates were centrifuged at 4,000 g for 30 minutes at 4°C. The soluble protein extracts (supernatants) were filtered through 0.45-μm syringe filters, and then used for enzymatic assays.
**Enzyme assays for in vitro and in vivo screens**
The enzyme activities of the in vitro protein expression products from the pUC57CFE1 vector were screened on the following substrates: 4-nitrophenyl-β-D-glucopyranoside (pNPC, 5 mM), 4-nitrophenyl-β-D-cellobioside (pNPC, 5 mM), and 1% carboxymethyl cellulose (Sigma Aldrich). Each enzyme reaction mixture containing one of these substrates and 5 μL of in vitro expression product or soluble extract from E. coli cell lysates (before or after induction) was done in 50 mM sodium acetate buffer at pH 5 in a total volume of 50 μL. The final concentration of 4-nitrophenol-labeled substrate (pNPC, or pNPC) was 5 mM, and that of carboxymethyl cellulose (CMC) was 1% in each reaction. The enzymatic reaction was done at 50°C for 16 hours. For the reaction mixtures containing CMC, a 3.5-dinitrosalicylic acid (DNS) assay was used to quantify hydrolyzed products. For the reaction mixtures containing pNPC, or pNPC, an equal volume of 2% sodium carbonate (Na2CO3) was added prior to measuring absorbance at 420 nm to detect hydroyzed 4-nitrophenol.
**Enzyme assays for activity profiling of cellulases**
To profile the enzyme activity of positive cellulases in the screen, each enzyme was expressed in vivo as described above, except the culture volume was scaled to 50 mL. For each enzyme assay, 5 to 20 ul of lysate was used to ensure that each enzyme had an activity that fell within the linear range of the activity assay. Enzymes J1 to J19 were screened using pNPC (5 mM final concentration) and enzymes J21 to J39 were screened using CMC (1% w/v final concentration) in a 100-ul reaction volume. Each value reported in Table 3 is from the average of triplicate reactions. For the temperature profile, the reaction was set up using 50 mM 2-(N-morpholino)ethanesulfonic acid (MES) buffer pH 6.5, and reactions were run for 15 to 60 minutes, depending on enzyme activity, at 5° increments from 45 to 99°C. For the pH profile, the reactions were run at approximately 10°C below the optimal temperature of each enzyme in 100 mM NaOAc 50 mM MES and 50 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES) buffers between pH 4.0 and 8.0. The buffers were made by mixing two aliquots of the aforementioned buffer set to either pH 4.0 (buffer A) or 8.0 (buffer B) in 10% increments, starting from 0% B to 100% B, giving 11 points total between pH 4.0 and 8.0. For IL-tolerance profiles, the reactions were run without added buffer in IL concentrations between 0 and 40% w/v [C2mim][OAc] at approximately 10°C below the optimal temperature of each enzyme. Reaction times were set to keep the values within the linear range of detection. For some enzymes, the same reaction was set up substituting an equal molar amount of NaOAc for [C2mim][OAc]. Figure 1C-D shows the pH at each concentration of IL and molar equivalent concentrations of NaOAc.
### Additional file
**Additional file 1: Table giving detailed annotation information on predicted cellulases.** Shown are predicted cellulase genes investigated in this report, detailing additional annotations for each gene, including E.C. number and glycoside family predictions.
### Abbreviations
- BG: β-glucosidase; [C2mim][OAc]: 1-ethyl-3-methylimidazolium acetate;
- CBH: cellobiohydrolase; Endo: Endoglucanase; GH: Glycoside hydrolase;
- HEPES: 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid; IL: Ionic liquid;
- MES: 2-(N-morpholino)ethanesulfonic acid; NaOAc: Sodium acetate;
- ORF: Open reading frame; PMSF: phenylmethanesulfonylfluoride;
- pNPC: 4-nitrophenyl-β-D-glucopyranoside; pNPG: 4-nitrophenyl-β-D-cellobioside;
- Na2CO3: Sodium carbonate; Topt: Optimum temperature.
### Competing interests
The authors declare that they have no competing interests.
### Authors’ contributions
JMG carried out most of the research and wrote the manuscript. JMG and JIP designed the study, designed the in vitro expression vector and cloned all cellulase genes to E. coli expression vector, and conducted the in vitro and in vivo screens. JB and VR profiled enzyme activities, PD identified and repaired the 37 cellulase ORFs. BFG advised JB. KLS advised JIP and VR and helped visualize Ttop and IL-tolerance. SWS and BAS advised JG. All authors read, provided edits, and approved the manuscript.
### Acknowledgements
This work conducted by the Joint BioEnergy Institute was supported by the Office of Science, Office of Biological and Environmental Research, of the US Department of Energy under Contract No. DE-AC02-05CH11231. JCB was supported by fellowship 9721/11-8 from CAPES Foundation, Ministry of Education of Brazil.
### Author details
1. Physical Biosciences Division, Lawrence Berkeley National Laboratory, Joint BioEnergy Institute (JBEI), 1 Cyclotron Road, Berkeley, CA 94720, USA.
2. Biological and Materials Science Center, Sandia National Laboratories, Livermore, CA, USA.
3. Physical and Life Sciences Directorate, Lawrence Livermore National Laboratory, Livermore, CA, USA.
4. Department of Geochemistry & Department of Ecology, Earth Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA, USA.
5. Department of Genomics and Department of Energy, Earth Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA, USA.
6. Embrapa-Agroenergia, Brasilia DF 70701-901, Brazil.
7. Current address: Department of Biological Sciences, Takeda California, Inc., San Diego, CA, USA.
Received: 14 October 2013 Accepted: 15 January 2014
Published: 29 January 2014
References
1. Laboratory ORNL: U.S. Billion-Ton Update: Biomass Supply for a Bioenergy and Bioproducts Industry. US DOE Energy Efficiency and Renewable Energy web site; 2011. http://www1.eere.energy.gov/bioenergy/pdfs/billion_ton_update.pdf.
2. Wiselogel AE, Agblevor FA, Johnson DK, Deutch S, Fennell JA, Sanderson MA: Compositional changes during storage of large round switchgrass bales. Bioresource Technol 1996, 56:103–109.
3. Wald ML: U.S. Backs Project to Produce Fuel From Corn Waste. New York Times, New York edition; July 7th 2011.
4. U.S. DOE: Using Fermentation and Catalysis to Make Fuels and Products: BIOCHEMICAL CONVERSION. US DOE Energy Efficiency and Renewable Energy web site; 2010. http://www1.eere.energy.gov/bioenergy/pdfs/biochemical_four_page.pdf.
5. Gladden JM, Eichorst SA, Hazen TC, Simmons BA, Singer SW: Advanced biofuel production in microbes. Biotechnol J 2010, 5:147–162.
6. Tadesse H, Luque R: Improved pretreatment and saccharification of switchgrass. Biofuels Bioprod Biorefin 2011, 5:647–675.
7. Nakayama S, Yoshio K, Kadokura T, Nakazato A: Butanol production from crystalline cellulose by cocultured clostridium thermocellum and clostridium saccharoperbutylicacetonicum N1-4. Appl Environ Microbiol 2011, 77:6470–6475.
8. Li C, Knierim B, Manisseri C, Arora R, Scheller HV, Auer M, Vogel KP, Steen EJ, Kang Y, Bokinsky G, Hu Z, Schirmer A, McClure A, Del Cardayre SB, Steen EJ, Kang Y, Bokinsky G, Hu Z, Schirmer A, McClure A, Del Cardayre SB: Dynamics of a lignocellulosic ethanol biorefinery with ionic liquid pretreatment. Bioresource Technol 2010, 101:4900–4906.
9. Zhao H, Jones CI, Baker GA, Xia S, Dubaij O, Person V: Regenerating cellulase from ionic liquids for an accelerated enzymatic hydrolysis. J Biotechnol 2009, 139:47–54.
10. Tadesse H, Luque R: Advances on biomass pretreatment using ionic liquids: An overview. Energ Environ Sci 2011, 4:3913–3929.
11. Klein-Marcuscharner D, Simmons BA, Blanch HW: Techno-economic analysis of a lignocellulosic ethanol biorefinery with ionic liquid pre-treatment. Biofuels Bioprod Biorefin 2011, 5:562–569.
12. Turner MB, Spear SK, Huddleston JG, Holbrey JD, Rogers RD: Ionic liquid salt-induced inactivation and unfolding of cellulase from Trichoderma reesei. Green Chem 2003, 5:443–447.
13. Park JJ, Steen EJ, Burd H, Evans SS, Redding-Johnson AM, Barth T, Benke PI, D’Haeseleer P, Sun N, Sale KL, Kealing JD, Lee TS, Petzold CJ, Mukhopadhyay A, Singer SW, Simmons BA, Gladden JW: A thermophilic ionic liquid-tolerant cellulase cocktail for the production of cellulosic biofuels. PLoS One 2012, 7:e37010.
14. Gladden JM, Allgaier M, Miller CS, Hazen TC, Simmons BA, Singer SW: Glycoside hydrolase activities of thermophilic bacterial consortia adapted to switchgrass. Appl Environ Microbiol 2011, 77:5804–5812.
15. Bormann HS, Kretschmer S, Jframberg L, Gudmundsdottir SM, Halla S, Addo K, Gladden JM: Proteogenomic analysis of a thermophilic bacterial consortium adapted to deconstruct switchgrass. PLoS ONE 2013, 8:e68465.
16. Rosenberg AH, Lade BN, Chui DS, Lin SW, Dunn JJ, Studer FW: Vectors for selective expression of cloned DNAs by T7 RNA polymerase. Gene 1987, 56:125–135.
17. Zhang T, Datta S, Eichler J, Ivanova N, Axen SD, Kerfeld CA, Chen F, Kyridis N, Hugenholtz P, Cheng JF, Sale KL, Simmons BA, Rubin E: Identification of a haloalkaliphilic and thermostable cellulase with improved ionic liquid tolerance. Green Chem 2011, 13:2083–2090.
18. Shi J, Gladden JM, Sathitsuksanoh N, Kambarn P, Sandovol L, Mira D, Zhang S, George A, Singer SW, Simmons BA, Singh S: One-pot ionic liquid pretreatment and saccharification of switchgrass. Green Chem 2013, 15:2579–2589.
19. D’Haeseleer P, Gladden JM, Allgaier M, Chain PSG, Tringe SG, Malfatt SA, Aldrich JT, Nicora CD, Robinson EW, Paia-Tolc L, Hugenholtz P, Simmons BA, Singer SW: Protoeogenomic analysis of a thermophilic bacterial consortium adapted to deconstruct switchgrass. PLoS ONE 2013, 8:e68465. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2546,
1
],
[
2546,
8374,
2
],
[
8374,
14120,
3
],
[
14120,
19801,
4
],
[
19801,
24628,
5
],
[
24628,
25839,
6
],
[
25839,
28607,
7
],
[
28607,
29923,
8
],
[
29923,
35660,
9
],
[
35660,
40912,
10
],
[
40912,
47002,
11
],
[
47002,
50976,
12
]
]
} |
38f27306379616d691f296b3c28c2683cdb1ae5b | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 6,
"total-fallback-pages": 0,
"total-input-tokens": 20740,
"total-output-tokens": 7194,
"fieldofstudy": [
"Medicine",
"Biology"
]
} | Association of Interleukin-1B gene Polymorphism with
*H. pylori* infected Dyspeptic Gastric Diseases
and Healthy Population
Furkhanda Kalsoom¹, Sajjad-ur-Rahman²,
Muhammad Shahid Mahmood³, Tahir Zahoor⁴
**ABSTRACT**
**Objective:** The aim of study was to investigate the association of IL 1B gene polymorphism with involvement of *H. pylori* and other gastric diseases.
**Methods:** Blood samples of dyspeptic patients were collected from endoscopy department of Allied Hospital Faisalabad from January 2017 to January 2019 and were qualitatively assayed for serological detection of CagA *H. pylori* antibodies. PCR followed by direct sequencing was performed for proinflammatory IL-1B gene polymorphism detection. Sequence analysis was performed in software SnapGene viewer for haplotypes.
**Results:** Demographic characteristics of seropositive patients showed maximum 25% gastritis in age groups of 20-40 years and 41-60 years, predominantly (41.7%) in females. While in seronegative patient’s gastritis (33.3%) was found in age group of 20-40 years mainly in males (41.7%). Among studied groups, higher expression of IL-1B-511 genotype (33.3%) polymorphism was found in healthy individuals as compared to *H. pylori* seropositive (25%) and seronegative (8.3%). While IL-1B-31 genotype showed maximum 33.3% polymorphism rate in seropositive gastric diseased group. Moreover, haplotypes frequencies IL-1B-511CC and IL-1B-31TT were predominantly (20%) found in seropositive gastric diseased group.
**Conclusions:** In *H. pylori* seropositive patients, gastric disease was commonly found, however, gastric disease was not only associated with *H. pylori* as seronegative patients were also carrying gastric complications. Interleukin IL-1B polymorphism was partially associated with *H. pylori* infection in studied dyspeptic population.
**KEYWORDS:** Cytotoxic antigen gene A (Cag A), Dyspepsia, *H. pylori*, Gastritis, Peptic ulcer.
How to cite this: Kalsoom F, Sajjad-ur-Rahman, Mahmood MS, Zahoor T. Association of Interleukin-1B gene Polymorphism with *H. pylori* infected Dyspeptic Gastric Diseases and Healthy Population. Pak J Med Sci. 2020;36(4):825-830.
doi: https://doi.org/10.12669/pjms.36.4.1883
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
**INTRODUCTION**
*Helicobacter pylori* (*H. pylori*) is a Gram-negative rod shape bacterium that colonizes stomach of about 50% of world’s population.¹ It can colonize the gastric mucosal environment for years if left untreated and increase the risk of gastric disease development.² Frequent physiological changes by this bacterium are gastritis, peptic ulcer (PU) and less commonly gastric cancer.³ *H. pylori* has been classified as class I carcinogen by the World Health Organization.⁴ *H. pylori* is considered as a major cause of peptic ulcer and mucosa associated lymphoid tissue lymphoma or gastric cancer.⁵ Although, gastritis to gastric cancer development
is a rare state, various studies reported that IL-1 and tumor necrosis factor alpha (TNF-α) polymorphism along with H. pylori infection are predisposing risk factors for gastric carcinoma. H. pylori infection induces proinflammatory host response in stomach and leads to release of different cytokines or interleukins, more frequently IL-1B, IL-1A, IL-6, IL-8, IL-10, and TNF-α. Interleukin polymorphism increases the production of mucosal cytokine (IL-1β) level that ultimately reduces the acid (HCl) secretion in the stomach and causes gastric inflammation.
Various studies have demonstrated that expression of IL-1B gene is frequently influenced by two allelic variants, IL-1B-511 and IL-1B-31, which are associated with IL-1B transcription. The polymorphism of these two genotypes has a synergistic effect on phenotypic change that increases the production of cytokine level and results in predisposition of gastritis. Several studies have reported that chronic gastritis is an established precursor of gastric adenocarcinoma with involvement of cytokine gene polymorphism. A study described the association of genotypes (IL-1 or IL-8) polymorphism and H. pylori infection and reported their combined effect for the risk of gastric carcinogenesis. In addition, haplotypes (TT, CC or CT) variable frequencies have an association with gastritis and gastric cancer development. The aim of our study was to elucidate the potential association of H. pylori infection with IL-1B gene polymorphism existence in infected population in district Faisalabad, Pakistan.
**METHODS**
**Study subjects:** A total of 240 dyspeptic patients were examined through endoscopy for presence of gastric disease at tertiary care hospital, Allied Faisalabad. This study was conducted from January 2017 to January 2019. A structured questionnaire was designed to collect demographic data of enrolled patients. Age of participants was categorized in three groups; 20-40 years, 41-60 years and 61-90 years including both (male and female) genders. The patients unable to complete the endoscopy procedure were excluded due to failure of clinical indication and informed consent was signed by patient or patient’s attendant for blood sample collection. Prior to conduct study approval was obtained from local health committee Allied Hospital Faisalabad (D.No.194/ORIC dated January 1, 2019). The bioethics committee UAF also approved the study protocol. A group of healthy volunteer individuals was also enrolled to compare interleukin (IL-1B) gene polymorphism among clinically gastric diseased and healthy population at district Faisalabad.
**H. pylori Serological Examination:** All selected dyspeptic patients were initially screened for presence of H. pylori infection on the basis of antibodies (Ab/s) detection. All blood samples were processed on H. pylori ‘One Step Test device (CTK BIOTECH, San Diego, CA 92121 USA). DNA Extraction: A total of 36 samples, 12 from each selected group (Healthy, H. pylori seropositive and seronegative gastric diseased) were processed for IL-1B gene polymorphism as only H. pylori seropositive were entertained with comparison of others. Host genomic DNA extraction was performed by using commercially available kit (HiPura Blood Genomic DNA Kit) as described previously, and extracted DNA samples were stored at 4ºC for genotyping.
**Genotyping for IL-1B gene Polymorphism:** All extracted DNA samples were processed further for IL-1B genotyping and amplification was done through PCR using commercially available PCR kit (Thermo Scientific™ K0171). Polymorphism frequency was analyzed by processing the PCR products on 2% gel electrophoresis, visualizing a single band of specific base pair sizes in patients and healthy group samples. Specific primer sets and PCR conditions were used as followed in a previous study. The sequences of reverse and forward primers and PCR conditions are given in (Table-I).
For allelic variant analysis (IL-1B-511 C/T and IL-1B-31 T/C allele haplotypes), PCR products were sent to Eurofins Genomics, USA. Sequencing results were annotated in software SnapGene (version 4.2.1) for evaluation of allelic haplotype (CC, TC and TT) frequencies.
**RESULTS**
**Segregation of Selected Dyspeptic Population:** Out of 240 dyspeptic patients, 70 (29.8%) were diagnosed with gastric diseases, of which 25 were having gastritis, 29 were with esophageal fundal varices (EFV), 12 showed red sign in stomach with gastro-pathy while only 2 patients were having peptic ulcer and 2 patients showed gastro esophageal reflux disease (GERD). The categorical distribution of dyspeptic gastric diseased patients is shown in (Fig.1).
Demographic characteristic of gastric diseased patients showed that 50% of population were found in second age group, 28.6% in first while only 21.4% were found in third age group. Gender wise distribution showed 55.7% were females.
Seroprevalence of H. pylori: In overall selected dyspeptic population, a low seroprevalence (7.0%) of H. pylori was found, while among 70 gastric diseased patients, 16 (23%) were positive for CagA H. pylori antibodies and others 54 (77%) were seronegative. Most of the seropositive gastric diseased patients (43.8%) were found in the first age group and predominantly were females (56.2%).
Characteristics of selected clinical and healthy groups: Studied subjects (n=36) were found with variable clinical characteristics. Table-I. Maximum gastritis 33.3% cases of seronegative gastric diseased group were found in the first age group. While, 25% of gastritis cases were found in other two age groups of seropositive and seronegative gastric diseased cases respectively. In healthy group, maximum 41.7% individuals were found in first and second age group. Gender wise distribution showed that majority of the cases (58.3%) were females in seronegative and healthy groups. EFV cases were found maximum 16.7% in first and second age groups of seropositive and seronegative gastric diseased patient respectively. PU occurrence rate of 8.3% was found in first age group of seronegative while in other two age groups of seropositive patients. Demographic distribution of three groups selected for genotype IL-1B polymorphism is shown in (Table-II).
Genotype and Allelic variant distribution of IL-1B polymorphism in clinical and healthy Individuals: Out of 36 cases, 23 showed IL-1B gene polymorphism on targeted positions (-511 and -31). Maximum 25% polymorphism was found at IL-1B-511CT in H. pylori seropositive gastric diseased cases, while for IL-1B-31TC genotype
Table-I: PCR primers and conditions.
| Genotype polymorphism | Primers | PCR conditions |
|------------------------|---------|----------------|
| IL-1B-511C/T | F 5′-GC CT GA AC CC TG CA TA CC CC GT-3′<br>R 5′-GG AA TC TT CC CA CT TA CA GA TGG-3′ | Denaturation of DNA for 10 minutes at 95°C, 40 cycles for 1 minute at 95°C, annealing at 55°C extension at 72°C and final extension for 5 minute at 72°C. |
| IL-1B-31T/C | F 5′-AG AA GC TT CC ACCAATAC TC-3′<br>R 5′-AC TA AC TT TA GG GT GT CAG-3′ | 10 min denaturation at 94°C, 36 cycles for 2 minute at 94°C, annealing at 54°C, extension at 72°C and final extension at 72°C for 5 min. |
Table-II: Demographic distribution of clinical and healthy individuals for IL-1B Genotype.
| Parameters | Healthy group | H. pylori Seronegative Gastric diseased group | H. pylori Seropositive Gastric diseased group |
|------------|---------------|-----------------------------------------------|-----------------------------------------------|
| | No. of Subjects (%) | No. of cases (%) | No. of cases (%) | No. of cases (%) |
| Age | Gastritis | PU* | EFV** | Gastritis | PU* | EFV** |
| 20-40 Years | 5 (41.7) | 4 (33.3) | 1 (8.3) | - | 3 (25) | - | 2 (16.7) |
| 41-60 Years | 5 (41.7) | 3 (25) | - | 2 (16.7) | 3 (25) | 1 (8.3) | - |
| 61-90 Years | 2 (16.7) | 1 (8.3) | - | 1 (8.3) | 2 (16.7) | 1 (8.3) | - |
| Gender | Male | 5 (41.7) | 5 (41.7) | - | - | 3 (25) | 1 (8.3) | 2 (16.7) |
| | Female | 7 (58.3) | 3 (25) | 1 (8.3) | 3 (25) | 5 (41.7) | 1 (8.3) | - |
| | | | | | | | | |
*Peptic Ulcer (PU), **Esophageal Fundal Varices (EFV).
polymorphism existence was high 16.7% in both gastric diseased cases. Twenty-five percent of gastritis patients showed polymorphism for both (IL-1B-511 and IL-1B-31) genotypes whereas, only 8.3% of EFV patients showed polymorphism for IL-1B-31T/C. Healthy individuals also showed 33.3% and 16.7% polymorphism at IL-1B-511C/T and IL-1B-31T/C promoter sites respectively. The detail of IL-1B gene polymorphism is presented in (Table-III). All samples positive for IL-1B-511 T/C genotype polymorphism, produced a single band of 155bp size amplicon while IL-1B-31 C/T positive samples exhibited a single band of 448bp in PCR products as visualized on 2% gel electrophoresis.
Haplotype C-T allelic variants in clinical and healthy groups: Two loci in IL-1B gene (IL-1B-511C/T and IL-1B-31T/C) containing three main haplotypes (CC, CT and TT) were analyzed in the studied population. Out of 23 individuals (clinical and healthy) with polymorphic IL-1B genotype, the persistence of polymorphism was compared between seronegative and seropositive groups. For IL-1B-511, seronegative healthy individuals showed maximum 33.3% frequency of CT haplotype, whereas, high 20% of CC haplotype carriers were found in seropositive gastric diseased patients. While, 14.3% CT haplotype frequency was found for IL-1B-511 in gastric diseased seronegative patients. Individuals carrying IL-1B-31 genotype were found to have maximum 28.6% haplotype CC in seronegative gastric diseased cases as it was 20% among seropositive gastric diseased cases. Similarly, the frequency of haplotype TT was also prevalent in 20% cases of seropositive gastric disease and CT allelic variant was 14.3% among seronegative gastric diseased cases. The polymorphism of both genotypes; IL-1B-511 and IL-1B-31 was simultaneously identified among seronegative gastric diseased cases with 28.6% TT haplotype frequency. While, 20% CC haplotype was found in seropositive cases followed and 10% CT haplotype. Seronegative gastric diseased cases (14.3%) were found with CC haplotype. Haplotype distribution in clinical cases and healthy individuals was mentioned in (Table-IV).
**DISCUSSION**
In the present study out of total dyspeptic population, 29.2% were found with gastric
---
**Table-III:** Comparative genotype frequencies in relation IL-1B gene polymorphism in seropositive gastric diseased patient and healthy group.
| Genotype | Seronegative Healthy group No. of individuals (%) | Seronegative Gastric Diseased group No. of cases (%) | Seronegative Gastric Diseased group No. of cases (%) |
|----------------|--------------------------------------------------|----------------------------------------------------|----------------------------------------------------|
| | Gastritis | PU | EFV | Gastritis | PU | EFV |
| IL-1B-511C/T | 4 (33.3) | 1 (8.3) | - | 3 (25) | - | - |
| IL-1B-31T/C | 2 (16.7) | 2 (16.7) | - | 1 (8.3) | 2 (16.7) | 1 (8.3) | 1 |
| IL-1B-511C/T plus IL-1B-31 T/C | - | 3 (25) | - | 3 (25) | - | - (8.3) |
| Negative | 6 (50) | 2 (16.7) | 1 (8.3) | 2 (16.7) | 1 (8.3) | 1 (8.3) | - |
*Peptic Ulcer (PU), **Esophageal Fundal Varices (EFV).
**Table-IV:** Genotype and haplotype allelic distribution of IL-1B polymorphism in clinical patients with gastric disease and healthy individuals.
| Genotype | Haplotype | Healthy individuals (%) | Seronegative Gastric Diseased Cases (%) | Seropositive Gastric Diseased Cases (%) |
|----------------|-----------|-------------------------|----------------------------------------|----------------------------------------|
| IL-1B-511 | CC | 1 (16.7) | - | 2 (20) |
| | CT | 2 (33.3) | 1 (14.3) | - |
| | TT | 1 (16.7) | - | 1 (10) |
| IL-1B-31 | CC | - | 2 (28.6) | 2 (20) |
| | CT | 1 (16.7) | 1 (14.3) | - |
| | TT | 1 (16.7) | - | 2 (20) |
| IL-1B 511plus IL-1B-31 | CC | - | 1 (14.3) | 2 (20) |
| | CT | - | - | 1 (10) |
| | TT | - | 2 (28.6) | - |
In this study, we evaluated the relevance of sequence context of IL-1B gene polymorphs with haplotypic frequency predominance in the selected population. Three main haplotype variables (CC, TC and TT) with various rates were found in the promoter/enhancer sites of IL-1B (-511CT and -31TC). The frequencies of C/T polymorphism were identified by annotating the genotyping sequence results in software Snapgene viewer (version 4.2.1). In this particular study, for IL-1B-511T/C polymorphism, CC and TT haplotype frequencies were in 16.7%, while CT was found in 33.3% of healthy individuals. However, in the H. pylori seropositive patients with gastric diseases, risk was increased 20% for IL-1B-511 CC haplotype as compared to H. pylori seronegative carrying 14.3% CT haplotype. IL-1B-31 CC haplotype was observed higher 28.6% in seronegative gastric diseased group followed by 20% in H. pylori seropositive gastric diseased cases. While haplotypes TT and TC (16.7%) were found predominant in healthy individuals. Combined polymorphism of IL-1B-511 plus IL-1B-31 genotypes results showed that TT (28.6%) was high in H. pylori seronegative gastric diseased patients while in H. pylori seropositive patients, haplotype CC (20%) was observed. In China, a study conducted by Hong et al. described that interleukin IB-511TT or T alelic variants had a link for gastric carcinogenesis. Another study conducted in Germany, showed polymorphism of IL-1B gene was without any involvement of carcinogenesis. A comparative study investigated the haplotype frequencies (CC, TC and TT) in four Asian populations with gastric diseases and found Japan and Thailand population at higher risk of H. pylori infection as compared with China and Vietnamese population. In the present study, a potential association of allelic variants of IL-1B gene with gastritis was found. Study polymorphism results showed predominance of CC and TC was found in gastric diseased population of district Faisalabad. While a study conducted in Korea, reported a higher frequency of TT haplotype in -511 and -1RN positions polymorphism with an association of gastric diseases.
CONCLUSIONS
The present study showed only a higher expression of IL-1B-31 polymorphism in H. pylori infected gastric diseased patients as compared to IL-1B-511 polymorphism. So a partial involvement of studied gene contribution was observed. Both IL-1B-511 and 31 positions were not equally responded for gastric diseases development. The patients with
age 41-60 years were found at higher risk of *H. pylori* infection. Although the studied gastric diseased patients were frequently found positive for *H. pylori*, it is not only the risk for gastric complications. The *H. pylori* positive population having IL-1B polymorphism was presented with gastritis, EFV and PU that increase the risk of adenocarcinoma. It is suggested that upon diagnosis with *H. pylori* infection, the in time therapy be considered as compulsory to the patients. Moreover, IL-1B-511 and IL-1B-31 CT polymorphism that was also observed in healthy individuals may increase the risk of getting *H. pylori* infection.
**Acknowledgements:** Authors committee is thankful to Dr. Moghees Ather, Head of Endoscopy department, Allied Hospital Faisalabad, Pakistan for providing relaxation to collect data and gastric samples during endoscopy. The authors would like to thank ORIC, UAF for funding support.
**Source of Funding:** ORIC, UAF.
**Conflicts of interest:** None.
**REFERENCES**
1. Das D, Abbasi M, Akabar M, Nepal A, Islam MA, Ayuba A, et al. A clinical review on the pathogenesis and management "Helicobacter pylori" infection. Int J Adv Res Biol Sci 2016;3:18-30. doi: 10.22192/ijarbs.2016.03.10.004
2. Elosnaiha R. Detection of *H. Pylori* infection on dyspepsia patients. InOCP Conference Series: Earth Env Sci. 2018;125(1):2014. doi: 10.1088/1755-1315/125/1/01201
3. Sadeghi RN, Damavand B, Vahedi M, Mohebbi SR, Zojazi H, Molaei M, et al. Detection of p53 common intron polymorphisms in patients with gastritis lesions from Iran. Asian Pac J Cancer Prev. 2013;14(1):91-96. doi: 10.7037/apcj.2013.14.1.91
4. Yang XJ, Si RH, Liang YH, Ma BQ, Jiang ZB, Wang B, et al. Aminotransferase levels. J Pak Med Assocc. 2003;53(2):59-62.
5. Venerito M, Selgrad M, Malfertheiner P. Helicobacter pylori: gastric cancer and extragastric malignancies—clinical aspects. Helicobacter. 2013;18:49-39. doi: 10.1111/hel.12078
6. Zhao Y, Wang JW, Tanaka T, Hoseno A, Ando R, Tokudome S. Detection of *H. Pylori* by polymerase chain reaction through inhibition of autophagy pathway. World J Gastroenterol. 2016;22:3579-3598. doi: 10.3748/wjg.v22.i15.3978
7. Venerito M, Slorgal M, Malfertheiner P. Helicobacter pylori: gastric cancer and extragastric malignancies—clinical aspects. Helicobacter. 2013;19(4):39-43. doi: 10.1111/hel.12078
8. Zhao Y, Wang JW, Tanaka T, Hoseno A, Ando R, Tokudome S. Association between TNF-α and IL-1β genotypes vs Helicobacter pylori infection in Korea. World J Gastroenterol. 2013;19(46):8756-8763. doi: 10.3748/wjg.v19.i46.8758
9. Abbas Z, Moatter T. Interleukin (IL) 1b and IL-10 gene polymorphism in chronic hepatitis C patients with normal or elevated alanine aminotransferase levels. J Pak Med Assoc. 2003;53(2):59-62.
10. Ramis IB, Vianna JS, Goncalves CV, von Groll A, Dellagostin OA, da Silva PE. Polymorphisms of the IL-6, IL-8 and IL-10 genes and gastric carcinoma in a Mexican population. Pathol Oncol Res. 2017;23(4):873-880. doi: 10.1007/s12253-017-0191-9
11. Kulambetova GN, Imanbekova MK, Logvinenko AA, Sukashv AT, Filipenko ML, Ramancolov EM. Association of cytokine gene polymorphisms with gastritis in a Kazakh population. Asian Pac J Cancer Prev. 2014;15(18):7763-7768. doi: 10.7314/ACPJ.2014.15.18.7763
12. Murphy G, Thornton J, McManus R, Swan N, Ryan B, O’Morain CA, et al. Association of gastric disease with polymorphisms in the inflammatory related genes IL-1B, IL-1RN, IL-10, TNF and TLR4. Eur J Gastroenterol Hepatol. 2009;21(6):630-635. doi: 10.1097/EJG.0b013e3282f3406a
13. Hong JR, Cho WC, Wang AJ, Wu NH. Helicobacter pylori infection syndrome with IL-1β gene polymorphisms potentially contributes to the carcinogenesis of gastric cancer. Int J Med Sci. 2013;10(4):298-303. doi: 10.7150/ijms.14239
14. Wex T, Leodolter A, Bornschein J, Kuester D, Kaehne T, Kropf IA. Genetic difference in interleukin-1 beta polymorphisms among four Asian populations: an analysis of the Asian paradox between *H. pylori* infection and Helicobacter pylori related gastric cancer. J Exp Clin Cancer Res. 2010;29:55-69.
15. Matsuura N, Imanix T, Sato Y, Tomtitchong P, Tajiri T, Miki M, et al. Genetic difference in interleukin-1 beta polymorphisms among four Asian populations: an analysis of the Asian paradox between *H. pylori* infection and gastric cancer incidence. J Exp Clin Cancer Res. 2003;22:47-55.
16. Kim JJ, Kim N, Hwang S, Kim JY, Kim JY, Choi Y, et al. Relationship of interleukin-1β levels and gastroesophageal reflux disease in Korea. J Gastroenterol Hepatol. 2013;28(1):90-98. doi: 10.1111/j.1440-1746.2012.07274.x
**Author’s Contribution:** FK and SUR did study design and conceived study concepts. FK did data acquisition, quality control of data, data analysis, interpretation and manuscript writing. FK and SUR did data analysis, interpretation and statistical analysis. SUR, MSM and TZ did manuscript editing and review. All authors have approved the final version to be published. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3164,
1
],
[
3164,
8068,
2
],
[
8068,
11486,
3
],
[
11486,
16320,
4
],
[
16320,
18811,
5
],
[
18811,
23792,
6
]
]
} |
9cd26b5e338ad1edce30935f2ed3137ba22f29cd | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 42,
"total-fallback-pages": 0,
"total-input-tokens": 133033,
"total-output-tokens": 29943,
"fieldofstudy": [
"Biology"
]
} | Dynamic effects of genetic variation on gene expression revealed following hypoxic stress in cardiomyocytes
Michelle C. Ward¹#$, Nicholas E. Banovich²*, Abhishek Sarkar², Matthew Stephens²,³, Yoav Gilad¹,²#
¹Department of Medicine, University of Chicago, Chicago, IL, USA
²Department of Human Genetics, University of Chicago, Chicago, IL, USA
³Department of Statistics, University of Chicago, Chicago, IL, USA
* These authors contributed equally to the work
$Present address: Department of Biochemistry and Molecular Biology, University of Texas Medical Branch, Galveston, TX, USA
*Present address: Integrated Cancer Genomics Division, Translational Genomics Research Institute, Phoenix, AZ, USA
#Correspondence should be addressed to M.C.W ([email protected]), and Y.G ([email protected]).
Abstract
One life-threatening outcome of cardiovascular disease is myocardial infarction, where cardiomyocytes are deprived of oxygen. To study inter-individual differences in response to hypoxia, we established an in vitro model of induced pluripotent stem cell-derived cardiomyocytes from 15 individuals. We measured gene expression levels, chromatin accessibility, and methylation levels in four culturing conditions that correspond to normoxia, hypoxia and short or long-term re-oxygenation. We characterized thousands of gene regulatory changes as the cells transition between conditions. Using available genotypes, we identified 1,573 genes with a cis expression quantitative locus (eQTL) in at least one condition, as well as 367 dynamic eQTLs, which are classified as eQTLs in at least one, but not in all conditions. A subset of genes with dynamic eQTLs is associated with complex traits and disease. Our data demonstrate how dynamic genetic effects on gene expression, which are likely relevant for disease, can be uncovered under stress.
Introduction
Cardiovascular disease (CVD), which ultimately damages heart muscle, is a leading cause of death worldwide (WHO, 2018). CVD encompasses a range of pathologies including myocardial infarction (MI), where ischemia or a lack of oxygen delivery to energy-demanding cardiomyocytes results in cellular stress, irreparable damage and cell death. Genome-wide association studies (GWAS) have identified hundreds of loci associated with coronary artery disease (Nikpay et al., 2015), MI, and heart failure (Shah, 2019), indicating the potential contribution of specific genetic variants to disease risk. Most disease-associated loci do not localize within coding regions of the genome, often making inference about the molecular mechanisms of disease challenging. That said, because most GWAS loci fall within non-coding regions, these variants are thought to have a role in regulating gene expression. One of the main goals of the Genotype-Tissue Expression (GTEx) project has been to bridge the gap between genotype and organismal level phenotypes by identifying associations between genetic variants and intermediate molecular level phenotypes such as gene expression levels (Consortium et al., 2017). The GTEx project has identified tens of thousands of expression quantitative trait loci (eQTLs); namely, variants that are associated with changes in gene expression levels, across dozens of tissues including ventricular and atrial samples from the heart. However, the GTEx project has reported that most eQTLs are shared across tissues, suggesting that they probably do not contribute to disease in a tissue-specific manner.
It is becoming increasingly evident that many genetic variants that are not associated with gene expression levels at steady state, may be found to impact dynamic programs of gene expression in specific contexts. This includes specific developmental stages (Strober et al., 2019), or specific exposure to an environmental stimulus such as endoplasmic reticulum stress (Dombroski et al., 2010), hormone treatment (Maranville et al., 2011), radiation-induced cell death (Smirnov et al., 2012), vitamin D exposure (Kariuki et al., 2016), drug-induced cardiotoxicity (Knowles et al., 2018), and response to infection (Alasoo et al., 2018; Barreiro et al., 2012; Caliskan et al., 2015; Kim-Hellmuth et al., 2017; Manry et al., 2017; Nedelec et al., 2016). The studies of context-specific dynamic eQTLs highlight the need to determine the effects of genetic variants in the relevant environment. Therefore, if we are to fully understand the effects of genetic variation on disease, we must assay disease-relevant cell types and disease-relevant perturbations. Most of the aforementioned studies were performed in whole blood or immune cells, which means that there are many cell types and disease-relevant states that have yet to be explored.
With advances in pluripotent stem cell technology, we can now generate otherwise largely inaccessible human cell types through directed differentiation of induced pluripotent stem cells (iPSCs) reprogrammed from easily accessible tissues such as fibroblasts or B-cells. One of the advantages of iPSC-derived cell types as a model system is that the environment can be controlled, and thus we can specifically test for genetic effects on molecular phenotypes in response to controlled perturbation. This is particularly useful for studies of complex diseases such as CVD, which result from a combination of both genetic and environmental factors.
The heart is a complex tissue consisting of multiple cell types, yet the bulk of the volume of the heart is comprised of cardiomyocytes (Donovan, 2019; Pinto et al., 2016), which are particularly susceptible to oxygen deprivation given their high metabolic activity. iPSC-derived cardiomyocytes (iPSC-CMs) have been shown to be a useful model for studying genetic effects on various cardiovascular traits and diseases, as well for studying gene regulation (Banovich et al., 2018; Benaglio et al., 2019; Brodehl et al., 2019; Burridge et al., 2016; de la Roche et al., 2019; Ma et al., 2018; McDermott-Roe et al., 2019; Panopoulos et al., 2017; Pavlovic et al., 2018; Ward and Gilad, 2019).
In humans, coronary artery disease can lead to MI (Dzau et al., 2006) which results in ischemia and a lack of oxygen delivery to energy-demanding cardiomyocytes. Given the inability of cardiomyocytes to regenerate, this cellular stress ultimately leads to tissue damage. Advances in
treatment for MI, such as surgery to restore blood flow and oxygen to occluded arteries, have improved clinical outcomes. However, a rapid increase in oxygen levels post-MI can generate reactive oxygen species leading to ischemia-reperfusion (I/R) injury (Giordano, 2005). Both MI and I/R injury can thus ultimately influence the amount of damage in the heart. iPSC-CMs allow us to mimic the I/R injury process in vitro by manipulating the oxygen levels that cardiomyocytes are exposed to in vivo.
We thus designed a study aimed at developing an understanding of the genetic determinants of the response to a universal cellular stress, oxygen deprivation, in a disease-relevant cell type, mimicking a disease-relevant process. To do so, we established an in vitro model of oxygen deprivation (hypoxia) and re-oxygenation in a panel of iPSC-CMs from 15 genotyped individuals (Banovich et al., 2018). We collected data for three molecular level phenotypes: gene expression, chromatin accessibility and DNA methylation to understand both the genetic and regulatory responses to this cellular stress. This framework allowed us to identify eQTLs that are not evident at steady state, and assess their association with complex traits and disease.
**Results**
We differentiated iPSC-CMs from iPSCs of 15 Yoruba individuals that were part of the HapMap project (Banovich et al., 2018). To obtain a measure of variance associated with the differentiation process, and to more effectively account for batch effects, we replicated the iPSC-CM differentiation from three individuals three times, yielding 21 differentiation experiments in total. The proportion of iPSC-CMs in each cell culture was enriched by metabolic purification (see Methods). iPSC-CMs were matured by electrical pulsing and maintenance in cell culture for 30 days. On Day 30, the median cardiomyocyte purity was 81% (40-97% range), determined by flow cytometry as the proportion of cells that were positive for the cardiac-specific marker, TNNT2 (FigS1; STable; See methods).
We studied the response of the iPSC-CMs to hypoxia and re-oxygenation (Fig1A). To do so, we first cultured the iPSC-CMs at oxygen levels that are close to physiological oxygen levels (10% oxygen - Condition A) for seven days. We then subjected the iPSC-CMs to six hours of hypoxia (1% oxygen - Condition B), followed by re-oxygenation for 6 hours (10% oxygen - Condition C), or 24 hours (10% oxygen - Condition D) as previously described (Ward and Gilad, 2019). Oxygen levels were reproducibly controlled in cell culture (Fig1B, STable). In order to determine whether
Figure 1: iPSC-derived cardiomyocytes elicit a cellular response to hypoxia. (A) Experimental design of the study. Cardiomyocytes differentiated from iPSCs (iPSC-CMs) from 15 Yoruba individuals were cultured in normoxic conditions (10% oxygen condition A) and subjected to 6 hours of hypoxia (1% oxygen condition B) followed by 6 and 24 hours of re-oxygenation (10% oxygen conditions C and D). Immunocytochemistry of a representative cardiomyocyte culture where green: TNNT2; blue: nuclei. (B) Peri-cellular oxygen levels of each condition. Each point represents one individual undergoing the oxygen stress experiment. (C) Relative levels of BNP, a marker for cardiac stress, released into cell culture media. Asterisk denotes a statistically significant difference in BNP release (*p < 0.05, **p < 0.005). (D) Molecular phenotypes collected from each individual in each condition.
the cardiomyocytes were affected by the changes in oxygen levels, we measured the enzymatic activity of released lactate dehydrogenase throughout the experiment, as a proxy for cytotoxicity. We also measured released BNP, a clinical marker of heart failure (Maeda et al., 1998). As expected, both cytotoxicity ($P = 0.01$, FigS2) and BNP ($P = 5 \times 10^{-6}$, Fig1C) levels increased following hypoxia and long-term re-oxygenation.
With this system established, we sought to understand the contribution of the global gene regulatory response to the molecular and cellular response to hypoxia and re-oxygenation. To do so, we collected global gene expression data (using RNA-seq; $n=15$), chromatin accessibility data (using ATAC-seq; $n = 14$), and DNA methylation data (using the EPIC arrays; $n = 13$; Fig1D) in each condition. With these data we studied both the gene regulatory response to oxygen perturbation, as well as the interaction of the response with the underlying genotype of the assayed individuals.
**Gene expression changes in response to hypoxia and re-oxygenation**
We first sought to identify those genes important for regulating the response by analyzing the gene expression (RNA-seq) data. We processed samples in batches as described in STable and mapped and filtered sequencing reads to prevent allelic mapping biases (FigS3; TableS1; See methods; (van de Geijn et al., 2015)). We observed that one sample (18852A) was a clear outlier when comparing read counts for 18,226 autosomal genes across all samples, and thus excluded it from further analysis (FigS4). We filtered out genes with low expression levels (See methods) to yield a final set with data from 12,347 expressed genes (See methods). We performed a number of correlation-based analyses using the data from the technical replicates (FigS5), and confirmed that the quality of the data is high and that, in line with our flow cytometry data, our iPSC-CMs express a range of cardiomyocyte marker genes including $MYH7$ and $TNNT2$ (FigS6).
We took advantage of the fact that we have replicate experiments from three individuals to correct the data for unwanted variation (See methods; (Risso et al., 2014)). Following this procedure, our samples clustered both by oxygen level and individual (FigS7). To identify genes that respond to hypoxia and re-oxygenation, we first tested for differential expression between pairs of conditions using a linear model with a fixed effect for ‘condition’, a random effect for ‘individual’, and four unwanted factors of variation, learned from the data, as covariates. At an FDR of 10%, we identified thousands of genes that are differentially expressed between conditions ($A$ vs. $B = 4,983$, 2759).
B vs. C = 6,311; B vs. D = 6,792; A vs. D = 2,835; FigS8A-B). We used Cormotif (Wei et al., 2015) to classify 2,113 genes (17% of all expressed genes) as responding to hypoxia (Fig2, FigS8C-D). Response genes are enriched for genes previously identified to respond to hypoxia in a Caucasian population of individuals (Chi-squared test; P < 2.2x10^-16; (Ward and Gilad, 2019)), and are highly enriched for gene ontologies in transcription-related processes (See methods, modified Fisher's exact test; P = 1x10^-19).
Figure 2: Hundreds of dynamic eQTLs are revealed following hypoxia and re-oxygenation. (A) Expression levels of Jun, a response gene, during the course of the experiment. (B) Expression levels of actinin, a non-response gene. (C) The proportion of all expressed genes that are classified as response genes (green), and non-response genes (magenta). (D) QQ plot illustrating an enrichment of associations between genetic variants and gene expression levels in each condition. Numbers represent the number of eGenes in each condition. (E) An example of a shared eQTL, HEPB2. (F) Heatmap illustrating the 367 SNPs that are classified as dynamic eQTLs. Each row represents a SNP that is an eQTL in at least one condition. Color represents the strength of the association p-value. (G) Examples of each of the two dynamic eQTL categories. Top panel: genes that become an eQTL following hypoxia e.g. ZNF845. Bottom panel: genes that are an eQTL at baseline but not following hypoxia e.g. RFC2. (H) The proportion of baseline eQTLs (all those identified in condition A), and dynamic eQTLs that are also response genes.
Dynamic eQTLs are revealed following hypoxia
Having established that oxygen stress initiates a transcriptional response affecting thousands of genes, we sought to identify eQTLs, either before or after oxygen stress. Using the combined haplotype test (CHT), an approach that leverages allele-specific information in small sample sizes (See methods; (van de Geijn et al., 2015)), we identified 1,573 genes with eQTLs (eGenes) in at least one condition (q-value < 0.1; A: 613; B: 564; C: 564 and D: 464; Fig2D). We refer to the 613 eGenes identified in condition A as baseline eGenes. We confirmed that genes whose expression levels are not significantly associated with a SNP in any condition show a largely uniform distribution of P-values in each condition (FigS9A), suggesting that, as expected given our study design, the overall power to detect eQTLs is similar across conditions.
Our goal was to identify dynamic eQTLs, which are either revealed or suppressed as the cells transition between conditions. Due to the small sample size of our study, we have incomplete power to detect eQTLs in any condition; thus, a naïve comparison of eQTLs classified as ‘significant’ across conditions will result in an over-estimation of the number of dynamic eQTLs. To address this challenge, we first considered eQTLs identified using a q-value < 0.1 in at least one condition, and visualized the P-value distributions of the corresponding eQTL associations in all other conditions. These P-value distributions are expected to be uniform if we had complete power to detect eQTLs in any condition (because in that theoretical case, even a naïve comparison of eQTLs classified as ‘significant’ across conditions will result in the identification of true condition-specific eQTLs). Due to incomplete power, this is clearly not the case (FigS9B); however, this distribution allowed us to choose a lenient secondary P-value-based cutoff, where values deviate from the uniform distribution, to classify dynamic QTLs (FigS9B).
We specifically focused on two dynamic scenarios. First, we defined suppressed eQTLs, as eQTLs that are identified in condition A at a q-value < 0.1 but not in any of the other conditions, with a P-value greater than 0.15 (37 instances; Fig2F). Second, we defined induced eQTLs as eQTLs identified in conditions B, C or D at a q-value < 0.1, but not in A, with a P-value greater than 0.15 (330 instances; Fig2G). This set of 367 dynamic eQTLs corresponds to 328 unique dynamic eGenes (See methods). While our choice of the particular statistical cutoffs is somewhat arbitrary, we can evaluate the false discovery rate associated with our chosen cutoff. Based on the P-value distributions of the corresponding eQTL associations in all other conditions, we estimate that our approach to classify dynamic eQTLs is associated with a false discovery rate of 48%. The
relatively high FDR associated with our choice of statistical cutoffs does not indicate that these loci are not eQTLs; rather it means that if we had a larger sample size, roughly half of our dynamic eQTLs should have been classified as eQTLs in more conditions, potentially in all of them.
We next wanted to determine whether the dynamic eQTLs we identified in iPSC-CMs are also eQTLs in primary heart tissue. To do so, we compared our list of eGenes to eGenes identified in tissue from two locations in the primary heart - left ventricle (LV) and atrial appendage (AA) - from hundreds of individuals in the GTEx study (Consortium et al., 2017). We found that the majority of eGenes present in all conditions ('shared', n=61) in our study are also eGenes in heart tissue (60% in LV, and 67% in AA). A smaller proportion of dynamic eGenes were also classified as primary heart tissue eGenes (38% in both LV and AA). Indeed, shared eGenes are more likely to be eGenes in heart tissue than dynamic eGenes (Chi-squared test; \( P < 0.005 \); FigS1A-B). Overall, 98% of dynamic eGenes are an eGene in at least one of 14 selected tissues assayed by the GTEx consortium, supporting our analysis indicating context-dependent inter-individual variation in expression (FigS1C).
**Dynamic eGenes are enriched for response genes and transcription factors**
To determine whether dynamic eGenes coincide with expression changes of the same genes following hypoxia, we integrated the results of our eQTL and differential expression analyses. In line with our previous findings, baseline eGenes, as well as LV and AA eGenes found in primary tissue (GTEx), are depleted for response genes (Chi-squared test; \( P < 0.02 \), (Ward and Gilad, 2019)). However, we found a significant enrichment in response genes amongst dynamic eGenes (61 of 328 genes; Chi-squared test; \( P = 0.03 \)) when compared to baseline eGenes, suggesting that dynamic eQTLs often impact the regulation of genes that respond to hypoxic stress.
Given that thousands of genes are differentially expressed in response to hypoxia, and many of these genes correspond to transcription-related processes, we next investigated the role of transcription factors, which are likely to drive transcriptome differences in our system. We found a significant enrichment of annotated transcription factors amongst the genes responding to oxygen stress compared to non-response genes (327 of 1,639 annotated human TFs, chi-squared test; \( P < 2 \times 10^{-16} \); FigS11). Given that stress affects transcription factor expression, we asked whether dynamic eGenes are also enriched for transcription factors. Indeed, transcription factors are enriched in dynamic eGenes compared to baseline eGenes (35 TFs; \( P = 0.004 \)), including MITF.
and PPARα, both of which are TFs that have been previously implicated in hypoxic response (Feige et al., 2011; Narravula and Colgan, 2001).
**Chromatin accessibility changes following hypoxia and re-oxygenation**
We next asked whether the hundreds of transcription factor expression changes following hypoxic stress are accompanied by global chromatin accessibility changes. To examine this, we performed ATAC-seq experiments to identify regions of open chromatin (we were only able to collect these data from 14 of the 15 individuals; STable). We filtered the ATAC-seq reads to include only those reads that map to the nuclear genome, and do not show allelic mapping biases (See methods, FigS12). We identified a set of open chromatin regions in each sample, and merged samples across individuals within each condition. Genomic regions identified as accessible in each condition were then merged to yield a set of 128,672 open chromatin regions across conditions (with a median length of 312 bp). Regions with low read counts were filtered out, resulting in a final set of 110,128 regions. Analysis of various metrics revealed the data to be of good quality (FigS13-14).
We sought to identify chromatin regions that are differentially accessible across pairs of conditions. Using a sensitive adaptive shrinkage based approach with a False Sign Rate of 10% (Stephens, 2017) we could not detect changes in accessibility between baseline and hypoxia; however, we identified 831 differentially accessible regions (DARs) between hypoxia and short-term re-oxygenation (BC-DARs; 429 regions with increased accessibility and 402 with decreased accessibility), and 71 DARs between hypoxia and long-term re-oxygenation (BD-DARs; Fig3A). There is a strong correlation in effect sizes between hypoxia and short-term re-oxygenation (BC-DARs), and hypoxia and long-term re-oxygenation (BD-DARs; Spearman correlation = 0.74), and 59 of the 71 BD-DARs are amongst the 831 BC-DARs, suggesting that most regions have returned to baseline levels of accessibility by the first re-oxygenation condition. We therefore considered the 831 BC-DARs, henceforth DARs, in further analysis. Manual inspection of accessibility levels at individual DARs, identified between hypoxia and re-oxygenation, shows that many of these regions appear to have small changes in accessibility between the baseline and hypoxic conditions, which is opposite to the direction of the effect between hypoxia and re-oxygenation. This includes a region within the intron of the FOXO1 gene, a master regulator of the oxidative stress response (Fig3B). Global analysis reveals that there is a strong anti-correlation in the effect size between
these pairs of conditions across regions (Spearman correlation = -0.62; sign test $P = 4.6 \times 10^{-14}$; FigS15A-C).
**Figure 3:** Chromatin accessibility changes following hypoxia and re-oxygenation. (A) Numbers of chromatin regions that are differentially accessible (DARs) between pairs of conditions. (B) Chromatin accessibility levels at a chromatin region within a FOXO1 intron. (C) Expression levels of the hypoxia-responsive gene ADM following hypoxia. (D) Chromatin accessibility levels at a DAR, overlapping an induced HIF1$\alpha$-bound region, close to the ADM gene.
Linking chromatin accessibility changes with gene expression changes
Changes in chromatin accessibility likely cannot directly explain the thousands of gene expression changes that occur following hypoxia. However, we found that when considering a 50 kb window around the TSS of expressed genes, DARs are enriched near response genes compared to non-response genes (Chi-squared test; $P = 0.03$). 113 of 2,113 response genes have a DAR within 50 kb of the TSS. This set includes an accessible region, overlapping a HIF1α site, within 500 bp of the 3’ end of the classic hypoxia response gene, ADM (Fig3C-D).
We asked whether the changes in chromatin accessibility coincide with the appearance of dynamic eQTLs. We found that DARs are no more likely to be near dynamic eGenes than shared or baseline eGenes. In line with previous estimates of the proportion of eQTLs in open chromatin regions, 24 baseline eQTL SNPs (613 total SNPs) and 19 dynamic eQTL SNPs (367 total SNPs) overlap with accessible chromatin regions (Consortium et al., 2017). One dynamic eQTL SNP overlaps a DAR, near the actin filament binding protein gene, FGD4. This gene was also shown to be differentially expressed between children with congenital heart defects where the defect leads to a chronic hypoxic state (cyanotic disease), and children with a similar defect but where oxygen levels are not affected (acyanotic disease; (Ghorbel et al., 2010)).
To directly test whether there are genetic effects on chromatin accessibility, independent of gene expression, we sought to identify chromatin accessibility QTLs (caQTLs) i.e. genetic variants located within the 128,672 accessible regions, which coincide with different levels of accessibility based on genotype. We identified few caQTLs per condition (q-value < 0.1; A: 10, B: 1, C: 7, D: 6; FigS16A). Six of these caQTLs are classified as dynamic caQTLs i.e. induced or suppressed in response to hypoxia using the same definitions as used for the dynamic eQTLs, and include regions at the TSS of the mRNA decapping enzyme gene DCPS, and a region within 100 kb of the C1Orf99 gene (FigS16B-C). These results suggest that gene expression changes, which respond to stress in a genotype-dependent or independent manner, occur largely in the absence of chromatin accessibility changes.
Genomic features associated with differentially accessible regions (DARs)
We next wanted to determine what distinguishes DARs from constitutively accessible regions. To do so, we investigated three classes of genomic features: 1) promoter- and enhancer-associated
marks, 2) transcription factor binding locations, and 3) underlying DNA sequence features. We found that DARs are more likely to overlap TSS than constitutively accessible regions (43% overlap vs. 11% overlap; $P < 2 \times 10^{-16}$; Fig4A) suggesting that DARs may be involved in the gene regulatory response. Indeed, DARs are more likely to coincide with active histone marks in left ventricle heart tissue than constitutively accessible regions (H3K4me3: 57% overlap DARs vs. 24% overlap constitutively accessible regions; chi-squared test; $P < 2.2 \times 10^{-16}$; H3K4me1: 86% overlap DARs vs. 52% overlap constitutively accessible regions; $P < 2.2 \times 10^{-16}$; Fig4B).
To determine whether sequence-specific hypoxia-responsive transcription factors associate with differentially accessible chromatin, we integrated DARs with published chromatin immunoprecipitation followed by high-throughput sequencing (ChIP-seq) data for the well-studied hypoxia-inducible factors HIF1α and HIF2α (Schodel et al., 2011). Based on our inference, 234 of the 356 annotated HIF1α sites (66%), and 150 of the 301 HIF2α sites (50%) overlap with all accessible chromatin regions. We found that HIF1α- and HIF2α-bound regions are more likely to overlap the 831 differentially accessible regions than the 109,275 constitutively accessible regions (Fisher test; $P = 0.03$; Fig4C).
We next took an unbiased approach to identify transcription factor binding motifs that are enriched in DARs compared to all accessible regions (See methods). We found two motifs to be enriched in DARs compared to all regions (Fig4D). Motif 1 ($P = 2 \times 10^{-2}$) is recognized by HTF4 and TFE2, both of which are non-response genes in our system. Motif 2 ($P = 6.2 \times 10^{-42}$) is posited to be recognized by ZN770, E2F3, and E2F4. Both ZN770 and E2F4 are response genes in our system. DARs arise between the hypoxia and re-oxygenation conditions, and E2F4 expression increases following re-oxygenation, suggesting that it may be involved in the response. To test this hypothesis, we obtained a published ChIP-seq data set for E2F4 (Lee et al., 2011), and overlapped the 16,245 E2F4-bound regions with DARs and constitutively accessible regions. E2F4-bound regions are significantly enriched in DARs compared to constitutively accessible regions (Chi-squared test; $P < 2.2 \times 10^{-16}$; Fig4D). E2F4 is important for survival following ischemia in neurons, and has been suggested to be an anti-apoptotic factor in cardiomyocytes (Dingar et al., 2012; Iyirhiaro et al., 2014).
To identify additional sequence features that associate with DARs, we asked whether transposable elements (TE), a potential source of regulatory sequence subjected to chromatin-level regulation, are enriched in these sites (Du et al., 2016; He et al., 2019). We found that while three of the main
Figure 4: Differentially accessible regions are enriched for active chromatin features. (A) The proportion of differentially accessible regions (DARs) and constitutively accessible regions (CARs) that overlap with annotated TSS. (B) The proportion of DARs and CARs that overlap with the locations of histone marks determined by ChIP-seq in human heart tissue (Consortium, 2012). (C) The proportion of DARs and CARs that overlap with HIF1α and HIF2α binding locations determined by ChIP-seq in a breast cancer cell line (Schodel et al., 2011). (D) The most significant motif identified to be differentially enriched in DARs compared to all ARs that is putatively recognized by ZNF770, E2F3 and E2F4. We classify E2F4 as a response gene and therefore determined the proportion of DARs and CARs that overlap with E2F4 ChIP-seq binding locations identified in a human LCL line (Lee et al., 2011). (E) The proportion of DARs and CARs that overlap four major transposable element (TE) classes – LTR, LINE, SINE, DNA. The proportion of DARs and CARs that overlap with CpG islands (CGIs) that is proximal to the TSS (+/- 2 kb from the TSS), and distal to the TSS. Asterisk denotes a statistically significant difference between DARs and CARs (*p < 0.05, **p < 0.005, ***p<0.0005).
TE classes, LINEs, LTRs, and DNA elements are similarly enriched in DARs compared to constitutively accessible regions; SINEs are specifically enriched in DARs ($P = 6.5 \times 10^{-6}$; Fig4E). There is an enrichment of both Alu and MIR SINE family members in DARs ($P = 3 \times 10^{-5}$ and $P = 0.006$). AluS elements, and the AluSq and AluSp sub-families, are particularly enriched within the Alu family ($P = 0.007$ and $P = 0.02$ respectively). A different cellular stress, heat shock, has previously been shown to remodel chromatin accessibility at Alu elements in cervical cancer cells (Kim et al., 2001).
As Alu and MIR TE sequences are notably CpG dense (Medstrand et al., 2002), we next asked about the enrichment of CpG-dense CpG islands (CGIs) in our differentially accessible regions. We found that CpG islands are enriched in DARs compared to constitutively accessible regions, whether these regions fall within 1 kb of TSS, which are typically enriched for CGIs, or not ($P < 2.2 \times 10^{-16}$; Fig4F).
**DNA methylation state at stress-responsive genes and chromatin regions**
Genes with CGI promoters are thought to allow flexibility in TSS choice compared to genes without CGI promoters (Carninci et al., 2006), and to allow for the rapid induction of gene expression in response to stimuli (Ramirez-Carrozzi et al., 2009). We therefore asked whether this promoter feature is enriched in the stress response genes. Indeed, we find that response genes are more likely to have CGI promoters than non-response genes (Chi-squared test; $P = 0.002$).
Given the enrichment of CpG islands in gene promoters and chromatin regions that are responsive to stress, we asked whether this feature corresponded to differences in CpG DNA methylation levels in these same regions. We measured global DNA methylation levels at 766,658 CpG sites in all conditions from 13 of our individuals (missing data from two of the individuals), together with 24 replicate samples from three individuals (See methods, STable). We found the expected bimodal distribution of DNA methylation Beta-values across CpGs (Beta-values represent the ratio of intensities between the methylated and unmethylated alleles; FigS17A). Additional analyses indicated the data to be of good quality (FigS17-18).
To determine whether steady-state DNA methylation levels mark genes or regions that will change their expression level in response to stress, we investigated baseline DNA methylation levels in the promoters of genes classified as response genes and non-response genes, as well as DARs and
constitutively accessible regions. To do so, we assessed the DNA methylation level at CpGs within 200 bp upstream of the TSS in the baseline condition. The majority of the assayed CpGs were hypomethylated with a median Beta-value of less than 0.2 across genes and regions (Fig 5A). While there is no difference in median DNA methylation levels between response and non-response genes, we found that the median DNA methylation level is lower in DARs compared to all accessible regions (ARs) (*p < 0.05, ***p < 0.0005).
**Figure 5: Minimal DNA methylation changes following hypoxia.** (A) Mean DNA methylation levels (Beta-values) in the baseline condition (A) at CpGs within 200 bp upstream of the TSS of response genes and non-response genes, and within differentially accessible regions (DARs) and all accessible regions (ARs) (*p < 0.05, ***p < 0.0005). (B) Numbers of differentially methylated CpGs within CpG islands across pairs of conditions. (C) DNA methylation levels at a differentially methylated CpG within an intron of the EGR2 response gene. (D) EGR2 expression during the course of the experiment.
accessible regions \( (P < 2.2 \times 10^{-16}) \). These data suggest that the underlying DNA sequence features of these regions can affect their epigenetic profile, and that responsive chromatin regions may have specific epigenetic profiles which poise them for rapid response to stress.
**DNA methylation levels are largely stable following hypoxia and re-oxygenation**
Given that DNA methylation levels can associate with gene expression levels, we asked whether any CpGs are differentially methylated during the course of the oxygen perturbation experiment, which induces thousands of gene expression changes. When considering all 766,658 CpGs we did not find any differentially methylated CpGs (DMCpGs) across any pair of conditions (10% FDR; FigS19A), and all p-values are estimated to be true null p-values (\( \pi_0 = 1 \) when estimated by q-value across all pairs of conditions; FigS19B). We found this to be the case when considering two estimates of DNA methylation levels as input: Beta-values or M-values (log\(_2\) ratio of intensities of methylated versus unmethylated alleles), or when only considering the 32,794 CpGs that are located 200 bp upstream of the TSS (data not shown). Because of the CpG island enrichment in our response gene promoters, we then selected only those 143,587 CpGs present within CpG islands. Although we found no difference in CpG methylation between baseline and hypoxia, and the short-term re-oxygenation conditions, we identified four DMCpGs between the baseline and long-term re-oxygenation conditions. This set includes a CpG in the intron of the \( EGR2 \) response gene, which shows increased DNA methylation levels over time (Fig4C-D). Methylation at CpG islands within the intron of \( EGR2 \) has been shown to confer enhancer activity in cancer cells (Unoki and Nakamura, 2003). If we only select CpGs located within the promoters of the 2,113 response genes, we find one DMCpG within the promoter of the \( FTSJ2 \) gene, a rRNA methyltransferase, that is differentially methylated between hypoxia and long-term re-oxygenation. Selecting CpGs located only within the 831 DARs reveals two DMCpGs between baseline and hypoxia, and one DMCpG between baseline and long-term re-oxygenation. Changes in DNA methylation are therefore not likely to be a major mechanism behind gene expression or chromatin accessibility changes following six hours of hypoxia.
**Dynamic eQTLs associate with traits and disease**
Finally, we wanted to determine whether any of the dynamic eQTL SNPs or genes that we identified are also associated with complex traits or disease. We first searched within a catalog of genetic variants associated with traits assayed in GWAS for overlap with our dynamic eQTL SNPs.
(Buniello et al., 2019) and found an induced dynamic eQTL SNPs that is also associated with a measured phenotype - varicose veins (Table1A; Fig6A).
**Figure 6: Dynamic eQTls associate with SNPs and genes implicated in complex disease.**
(A) Example of a GWAS-implicated SNP that is also a dynamic eQTL SNP. *RNF166* is a dynamic eGene and the associated SNP is implicated in the presence of varicose veins. *RNF166* expression levels are stratified by genotype in each condition. (B) Expression levels of *RNF166* during the course of the experiment following aggregation of all individuals. (C) Example of a stroke and myocardial infarction (MI) GWAS-implicated gene that is also a dynamic eGene (*ZC3HC1*). (D) Expression levels of *ZC3HC1* during the course of the experiment following aggregation of all individuals.
We next took an orthogonal approach, using the same GWAS catalog, to specifically investigate three phenotypes that are associated with cardiovascular function or response to oxygen deprivation: MI, heart failure, and stroke (See methods). Six of our dynamic eGenes are also implicated in these disease states by GWAS (Table1B). This list includes the DNA damage and apoptosis factor *ZC3HC1*, which is implicated in MI and stroke (Fig6B). Importantly, *ZC3HC1* is not an eGene in LV or AA, but the SNP-gene pair is an eQTL in other tissues. These results suggest that perturbation studies in relevant cell types can give insight into the molecular basis for the genetic association with complex traits and disease, which might not be gleaned from the study of post-mortem tissues.
Table 1: Dynamic eQTL SNPs and genes associated with GWAS traits.
(A) Dynamic eQTL (deQTL) SNPs associated with GWAS traits. The eQTL SNP, associated gene, type of eQTL based on whether the eQTL is induced or suppressed in response to hypoxia, tested GWAS trait, whether the eQTL SNP is an eSNP in GTEx left ventricle or atrial appendage, and whether it is an eGene in GTEx ventricle or atrial appendage tissue. (B) Dynamic eGenes implicated in three relevant GWAS traits - heart failure, myocardial infarction (MI), and stroke. N.T: not tested.
| deQTL SNP | ENSGID | deQTL gene | type | trait | eSNP | eGene |
|-----------|--------------|------------|---------|-------------|------|-------|
| rs8053350 | ENSG00000158717 | RNF166 | induced | Varicose veins | Yes | Yes |
| deQTL SNP | ENSGID | deQTL gene | type | trait | eSNP | eGene |
|-----------|--------------|------------|---------|-------------|------|-------|
| rs60026782 | ENSG00000134317 | GRHL1 | suppressed | Heart failure | No | Yes |
| rs187463073 | ENSG00000136111 | TBC1D4 | induced | Heart failure | No | Yes |
| rs20128042 | ENSG00000186197 | EDARADD | induced | Heart failure | N.T | Yes |
| rs10242432 | ENSG00000091732 | ZC3HC1 | induced | MI/Stroke | No | No |
| rs75401776 | ENSG00000096401 | CDC5L | induced | Stroke | No | Yes |
| rs11612275 | ENSG00000171840 | NINJ2 | induced | Stroke | No | Yes |
Discussion
Studying gene expression across individuals in response to stress can reveal latent effects of genetic variation, which may contribute to higher-order phenotypes and disease. In order to understand the effects of genetic variation in a disease-relevant cell type and a disease-relevant process, we differentiated cardiomyocytes from a panel of genotyped individuals, and subjected them to hypoxia and re-oxygenation. We found hundreds of eQTLs that are revealed or suppressed following hypoxic stress (dynamic eQTLs), several of which have been associated with phenotypes measured in GWAS.
Steady-state and dynamic eQTLs may help understand CVD
Attempts have been made to identify genetic variants that associate with gene expression levels and CVD phenotypes in easily accessible biological samples such as blood. However, less than half of CVD/MI GWAS loci are associated with an eQTL in whole blood when thousands of individuals are tested (Joehanes et al., 2017). To determine the effects of genetic variation on gene expression specifically in the heart, more targeted studies have taken advantage of left ventricle tissue (Consortium et al., 2017; Koopmann et al., 2014), left atrium tissue (Lin et al., 2014; Sigurdsson et al., 2017), and right atrial appendage tissue (Consortium et al., 2017) obtained during cardiac surgeries or post-mortem. Using fewer than a hundred individuals, a handful of identified eQTL SNPs correspond to SNPs associated with cardiac traits, thus linking specific
genes to organismal-level phenotypes. A compelling example is the association between MYOZ1 expression and atrial fibrillation (Lin et al., 2014; Sigurdsson et al., 2017). Across tissues, the GTEx consortium reported that ~50% of eQTLs are also associated with variation in other measured complex traits (Consortium et al., 2017), and Heinig et al. have shown that 20% of left ventricle eQTLs relate to heart-associated loci (Heinig et al., 2017). However, these variants, identified in healthy individuals, are unlikely to represent all genetic variants that have consequences on disease. Indeed, Heinig et al. identified 100 dilated cardiomyopathy–specific eQTLs (not seen in healthy individuals) in a case-control study of 97 individuals with dilated cardiomyopathy and 108 healthy donors (Heinig et al., 2017). Similarly, by collecting samples pre- and post-surgically-induced ischemia, Stone et al. identified genetic associations that are only detected under stress (Stone et al., 2019). While these studies provide a set of gene targets for further investigation, there are many loci that remain unexplained.
The heart is a complex tissue consisting of multiple cell types. The effects of some genetic variants in specific cell types might well be masked when considering heterogeneous tissue samples. As we are now able to direct iPSCs towards a cardiac fate, we can test for genetic effects on specific cell types such as cardiomyocytes (Panopoulos et al., 2017). As one would expect, iPSC-CMs are better suited to study cardiovascular traits than the immortalized B-cells or iPSCs from which they are derived (Banovich et al., 2018). However, given the high degree of eQTL sharing across diverse tissues (GTEx), identifying eQTLs in the disease-relevant terminal cell type at steady state may not give substantial insight into disease biology. A significant advantage of using iPSC-CMs is that these cells provide a system to interrogate gene expression dynamics.
Cellular stressors that perturb gene expression levels and the cell state can unmask additional layers of regulatory variation (Alasoo et al., 2018; Barreiro et al., 2012; Caliskan et al., 2015; Kim-Hellmuth et al., 2017; Knowles et al., 2018; Manry et al., 2017; Nedelec et al., 2016). Intermediate developmental cell states can similarly provide insight into GWAS loci where eQTL analysis in terminal cell types cannot (Strober et al., 2019). Further evidence for the notion that steady state eQTLs may have limited applicability to disease states comes from our previous work, where we used a comparative evolutionary approach to investigate the response to stress. We showed that genes that have a conserved response to oxygen stress in iPSC-CMs from both humans and chimpanzees, and are therefore likely relevant for disease, are depleted for eQTLs identified in heart tissue (Ward and Gilad, 2019).
In the current study, by subjecting iPSC-CMs from a panel of individuals to perturbation (oxygen deprivation), we were able to identify a dynamic eQTL SNP (rs8053350) that is associated with varicose veins, and the level of \textit{RNF166} expression (Fukaya et al., 2018). This SNP falls within an intron of the \textit{PIEZO1} gene. Varicose veins are associated with a risk for developing deep vein thrombosis and other vascular diseases (Chang et al., 2018). When we performed an analogous analysis focused on genes previously associated with three relevant traits – MI, heart failure and stroke, we identified a novel heart eGene, \textit{ZC3HC1}, encoding the NIPA protein, which is implicated in MI, coronary artery disease and ischemic stroke (Consortium, 2011; Nikpay et al., 2015; Schunkert et al., 2011). This dynamic eQTL SNP is also associated with bronchodilator responsiveness in chronic obstructive pulmonary disease (Hardin et al., 2016).
\textbf{Mechanisms behind response genes and dynamic eQTLs}
Changes in gene expression can associate with other molecular-level phenotypes. The response to hypoxia is mediated by the HIF1α transcription factor (Samanta and Semenza, 2017), but given that there are hundreds of HIF1α binding locations and thousands of differentially expressed genes, regulation by this factor alone cannot directly explain all the transcriptional changes. We explored two additional molecular phenotypes in the context of oxygen deprivation - the locations and level of accessibility of open chromatin regions, and DNA methylation levels. We did not find either to contribute substantially to the gene expression response we observed. There are minimal changes in accessibility following hypoxia, which is in contrast to observations of studies that considered stimulation of immune cell types (Alasoo et al., 2018; Calderon et al., 2019; Pacis et al., 2015). This could reflect cell type specificity in response to stress, or the specificity of the cellular response to different stressors. Indeed, despite large gene expression changes in response to various stimulants in endothelial cells, there are a relatively small number of differentially accessible regions (Findley et al., 2019). We speculate that the transcriptional response to oxygen stress could result in the induction of transcription factors, which bind already accessible regions of open chromatin, and that cells are primed for a quick response to this universal cellular stress. Indeed, it has been shown that chromatin contacts exist between HIF1α binding sites and hypoxia-inducible genes in the normoxic state (Platt et al., 2016). Conversely, it has been suggested that hypoxia results in the induction of HIF1α, and significant changes in histone methylation (Batie et al., 2019). As we did not measure histone marks in our system, these changes may occur in the absence of chromatin accessibility changes, but we also cannot rule out the possibility that the
choice of a single timepoint following six hours of hypoxia, or insufficient statistical power in our sample size, contributed to the minimal differences in accessibility that we observed.
Using an approach designed to measure small effect sizes between conditions, we did identify a set of 831 DARs between hypoxia and short-term re-oxygenation that are enriched for marks of active chromatin, CpG islands, and TEs. These regions do not appear to explain many of the gene expression differences we observed. Hypoxia and oxidative damage are likely to also affect the genome in ways that do not directly impact gene expression. Indeed, the distribution of oxidative DNA damage sites varies across the genome following stress such that TEs and active chromatin regions are enriched for DNA damage, while promoters are depleted (Poetsch et al., 2018). We found enrichment for TEs, specifically Alu SINE elements, in DARs. Interestingly, TEs, and DNA transposons in particular, are also enriched in regions that become accessible in macrophages in response to bacterial infection; suggesting sequence-specific effects of TEs in response to different cellular stressors (Bogdan, 2019). Alu elements have previously been found to associate with the response to stress in other contexts. Serum starvation induces binding of TFIIIC, which recruits RNA polymerase III, to Alu elements (Ferrari et al., 2019), and heat shock increases chromatin accessibility around Alu elements (Kim et al., 2001).
There are several studies, which suggest that DNA methylation levels are dynamic and change in response to stressors such as hypoxia. We did not find any notable differences in DNA methylation levels pre- and post-hypoxia and re-oxygenation, which suggests that like chromatin accessibility, DNA methylation levels do not make large contributions to changes in gene expression levels or the appearance of dynamic eQTLs in our system. Many of the DNA methylation changes that have been described in response to hypoxia occur in chronic and intermittent hypoxia, and not acute hypoxia as investigated in our study (Hartley et al., 2013; Robinson et al., 2012; Watson et al., 2014). DNA methylation levels are also altered in response to other stressors such as bacterial infection (Pacis et al., 2015); however, the importance of timing is highlighted by the fact that, in this system, gene expression responses precede DNA methylation changes (Pacis et al., 2019). It is also important to note that our study considers baseline oxygen levels to be 10% oxygen, which is closer to physiological oxygen levels (5-13%) than atmospheric oxygen levels (21%; (Brahimi-Horn and Pouyssegur, 2007; Carreau et al., 2011; Jagannathan et al., 2016). Most studies define normoxia as 21% oxygen saturation, and while this likely leads to larger effect size differences in known hypoxia response genes following hypoxia, these comparisons may not give meaningful insight into the in vivo state.
One can speculate about different mechanisms that might lead to the appearance or disappearance of dynamic eQTLs. In the context of the immune response, it has been shown that the same response variants affect both gene expression and chromatin accessibility (Alasoo et al., 2018). This is in line with the general notion that changing cellular environments results in differences in chromatin accessibility at transcription factor binding sites, which leads to gene expression changes. We found that this does not appear to be a major mechanism in our system as there are minimal changes in accessibility following hypoxia. We observed that there is an enrichment of response genes amongst dynamic eQTLs suggesting that the change in environment results in a change in expression levels that is dependent on the associated genotype. We also find enrichment for TFs amongst response genes and dynamic eQTLs, suggesting that dynamic eQTLs can appear through secondary *trans* effects.
**Potential limitations of our model**
To understand the effects of genetic variation on human heart tissue, and how this variation might contribute to the MI and I/R injury etiologies of CVD, we carefully perturbed oxygen levels that cardiomyocytes in culture are exposed to. This *in vitro* approach is by design a model system, and therefore will likely not fully recapitulate the *in vivo* state. However, we previously found that out of 2,549 genes that respond to hypoxia in iPSC-CMs from humans and chimpanzees, only 16% are differentially expressed between iPSC-CMs and heart tissue (Pavlovic et al., 2018; Ward and Gilad, 2019). This suggests that our *in vitro* system is applicable to heart tissue. There is still a possibility that the dynamic eQTLs that we identify in our *in vitro* system are not physiologically relevant.
Our study comprised a small number of individuals (15), far fewer than what is typical for identifying eQTLs. Our work is therefore a first step towards understanding the effects of genetic variation on gene expression in response to stress. Nevertheless, with a small number of individuals we were able to identify a couple of hundred of dynamic eQTLs that are revealed or suppressed under stress, suggesting that this paradigm is worth exploring further in larger cohorts. Under the simplifying assumption of a single causal variant, we determined that we have ~6% power to detect an effect which explains ~38% of the heritability, and an equal false positive rate to call it a dynamic eQTL (See methods; FigS20). This suggests that the impact of stress on genotype-dependent effects on gene expression will be even greater in studies which have higher power to detect smaller effects of genotype. For perspective, early eQTL studies were similarly
powered to our study, using 70 individuals; yet these studies still led to important insights opening an avenue of research focused on assaying the consequences of genetic variation by RNA-seq (Pickrell et al., 2010).
In summary, there have been few studies assessing the effects of genetic variation in response to CVD-relevant perturbations in cardiomyocytes. Here we profiled the response to oxygen deprivation in cardiomyocytes from a panel of genotyped individuals. We find that eQTLs can appear and disappear in response to oxygen deprivation, and that some of these eQTLs have effects on relevant complex traits and disease.
Materials and Methods
Cardiomyocyte differentiation from iPSCs
We used fifteen individuals from the Yoruba YRI HapMap population. iPSCs were reprogrammed from lymphoblastoid cell lines (Banovich et al., 2018). iPSCs were maintained in a feeder-independent state in Essential 8 Medium (A1517001, ThermoFisher Scientific, Waltham, MA, USA) with Penicillin/Streptomycin (30002, Corning, NY, USA) on Matrigel hESC-qualified Matrix (354277, Corning, Bedford, MA, USA) at a 1:100 dilution. Cells were passaged at ~70% confluence every 3-4 days with dissociation reagent (0.5 mM EDTA, 300 mm NaCl in PBS), and seeded with ROCK inhibitor Y-27632 (ab12019, Abcam, Cambridge, MA, USA).
Cardiomyocyte differentiations were performed largely as previously described (Ward and Gilad, 2019), except the duration and concentration of the Wnt agonist and antagonist differed for this panel of individuals, which included only human samples. Briefly, on Day 0, iPSC lines at 70-100% confluence in 100 mm plates were treated with 12 µM GSK3 inhibitor CHIR99021 trihydrochloride (4953, Tocris Bioscience, Bristol, UK) in 12 ml Cardiomyocyte Differentiation Media [500 mL RPMI1640 (15-040-CM ThermoFisher Scientific), 10 mL B-27 Minus Insulin (A1895601, ThermoFisher Scientific), 5 mL Glutamax (35050-061, ThermoFisher Scientific), and 5 mL Penicillin/Streptomycin)], and a 1:100 dilution of Matrigel. 24 hours later, on Day 1, the media was replaced with Cardiomyocyte Differentiation Media. 48 hours later, on Day 3, 2 µM of the Wnt inhibitor Wnt-C59 (5148, Tocris Bioscience), diluted in Cardiomyocyte Differentiation Media, was added to the cultures. Cardiomyocyte Differentiation Media was replaced on Days 5, 7, 10, and 12. Cardiomyocytes were purified by metabolic purification by the addition of glucose-free, lactate-
containing media (Purification Media) [500 mL RPMI without glucose (11879, ThermoFisher Scientific), 106.5 mg L-Ascorbic acid 2-phosphate sesquimagnesium salt (sc228390, Santa Cruz Biotechnology, Santa Cruz, CA, USA), 3.33 ml 75 mg/ml Human Recombinant Albumin (A0237, Sigma-Aldrich, St Louis, MO, USA), 2.5 mL 1 M lactate in 1 M HEPES (L(+)Lactic acid sodium (L7022, Sigma-Aldrich)), and 5 ml Penicillin/Streptomycin] on Days 14, 16 and 18. 1.5 million cardiomyocytes were re-plated per well of a 6-well plate on Day 20 in Cardiomyocyte Maintenance Media [500 mL DMEM without glucose (A14430-01, ThermoFisher Scientific), 50 mL FBS (S1200-500, Genemate), 990 mg Galactose (G5388, Sigma-Aldrich), 5 mL 100 mM sodium pyruvate (11360-070, ThermoFisher Scientific), 2.5 mL 1 M HEPES (SH3023701, ThermoFisher Scientific), 5 mL Glutamax (35050-061, ThermoFisher Scientific), 5 mL Penicillin/Streptomycin]. iPSC-CMs were matured in culture for a further 10 days with Cardiomyocyte Maintenance Media replaced on Days 23, 25, 27, 28 and 30.
On Day 25, iPSC-CMs were transferred to a 10% oxygen environment (representative of in vivo levels) in an oxygen-controlled incubator (HERAcell 150i CO\textsubscript{2} incubator, ThermoFisher Scientific). From Day 27 onwards, iPSC-CMs were pulsed at a voltage of 6.6 V/cm, frequency of 1 Hz, and pulse frequency of 2 ms using an IonOptix C-Dish & C-Pace EP Culture Pacer to further mature the cells and synchronize beating.
Flow cytometry
Purity of the cardiomyocyte cultures was assessed ~Day 30 as previously described (Ward and Gilad, 2019). Briefly, cells were stained with Zombie Violet Fixable Viability Kit (423113, BioLegend), and PE Mouse Anti-Cardiac Troponin T antibody (564767, clone 13-11, BD Biosciences, San Jose, CA, USA), and analyzed on a BD LSRFortessa Cell Analyzer together with negative control samples of iPSCs, and iPSC-CMs that are incubated without the troponin antibody, or without either the troponin antibody or viability stain.
Hypoxia experiment
On Day 31/32 iPSC-CMs were subjected to the hypoxia experiment. At time = 0, condition A samples remained at 10% O\textsubscript{2} (normoxia), while samples for conditions B, C and D were transferred to an incubator set at 1% O\textsubscript{2} (hypoxia). After 6 hours, conditions A and B were harvested while plates C and D were returned to normoxic oxygen conditions. Plate C was harvested 6 hours
following the hypoxic treatment, and Plate D was harvested 24 hours following the hypoxic treatment. Oxygen levels, experienced by the cells in culture, were measured in cultures from each experimental batch using an oxygen-sensitive sensor (SP-PSt3-NAU-D5-YOP, PreSens Precision Sensing GmbH, Regensburg, Germany), optical fiber (NWDV29, Coy, Grass Lake, MI, USA), and oxygen meter (Fibox 3 Transmitter NWDV16, Coy).
Material collection
Cell culture media for ELISA and cytotoxicity assays
Aliquots of cell culture media from each experiment were centrifuged at 10 000 rpm for 10 min at 4°C to remove cellular debris. The supernatant was stored at -80°C until further use.
Nuclei for ATAC-seq
Cardiomyocytes from each well of a 6-well plate were washed twice with cold PBS on ice before collection by manual scraping in 1.5 ml PBS. 200 µl of cells were pelleted by centrifugation at 500 g for 5 min. Cell pellets were re-suspended in 50 µl cold ATAC-seq lysis buffer (10 mM Tris-HCl pH 7.4, 10 mM NaCl, 3 mM MgCl₂, 0.1% Igepal CA630, dH₂O). Nuclei were pelleted by centrifugation at 500 g for 5 min at 4°C. Nuclei were re-suspended in 50 µl transposition mix (25 µl 2xTD buffer, 2.5 µl Tn5 transposase, 22.5 µl nuclease-free dH₂O) from the Nextera DNA sample kit (FC-121-1031, Illumina). The transposition reaction was performed at 37°C for 30 min. Transposed DNA was purified with Qiagen MinElute Kit (28004, Qiagen, MD, USA), re-suspended in 12 µl elution buffer, and stored at -20°C.
Cell pellets for RNA-seq & DNA methylation arrays
Cells from each well of a 6-well plate were washed twice with cold PBS on ice before collection by manual scraping in 1.5 ml PBS. Cells were pelleted by centrifugation at 7 000 rpm for 8 min at 4°C, flash-frozen and stored at -80°C.
RNA/DNA extraction
RNA and DNA were extracted from the same frozen cell pellets using the ZR-Duet DNA/RNA MiniPrep kit (D7001, Zymo, CA, USA) according to the manufacturer’s instructions. All four conditions from three or four individuals were extracted in the same batch. RNA samples had a median RIN score of 8.5 (STable).
RNA-seq library prep
500 ng of RNA were used to prepare sequencing libraries using the Illumina TruSeq RNA Sample Preparation Kit v2 (RS-122-2001 & -2002, Illumina). Libraries were pooled into five master mixes containing 12 or 16 samples. Each pool was sequenced 50 bp, single-end on the HiSeq2500 or HiSeq4000 according to the manufacturer’s instructions.
DNA methylation array
9 chips (8 samples per chip) with 60-1000 ng DNA were bisulfite-converted and processed on an Illumina Infinium MethylationEPIC array at the University of Chicago Functional Genomics facility.
ATAC-seq
We performed ATAC-seq in 14 of the 15 individuals we had gene expression data for (STable for details). ATAC-seq libraries were prepared using the Illumina Nextera DNA sample kit. Libraries were amplified for 10-16 cycles depending on the amplification rate of each library. Each library was amplified in a PCR reaction containing 10 µl DNA, 10 µl dH2O, 15 µl NMP (PCR master mix), 5 µl PPC (PCR primer cocktail), and 5 µl index N5, 5 µl index N7). PCR conditions were set at 72°C for 5 min, 98°C for 30 sec, 98°C for 10 sec, 63°C for 30 sec, 72°C for 1 min, repeat steps 3-5 4x and hold at 4°C. The number of cycles per library was determined using a qPCR side reaction as described in Buenrostro et al. (Buenrostro et al., 2013). Libraries were purified using Agencourt AMPure XP beads (A63880, Beckman Coulter, IN, USA), and bioanalyzed to determine library quality. 12 or 16 samples were pooled together to generate four master mixes. Each master mix was sequenced 50 bp paired-end on the HiSeq4000 according to the manufacturer’s instructions.
Lactate dehydrogenase activity assay
Lactate dehydrogenase activity (LDH) was measured in 5 µl cell culture media using the Lactate Dehydrogenase Activity Assay Kit (MAK066, MilliporeSigma, MO, USA) according to the manufacturer’s instructions. Each sample was assayed in triplicate. LDH activity was measured as the difference in absorbance prior to the addition of the substrate, and 10 min after the initiation of the enzymatic reaction, calculated relative to a standard curve. Measurements are standardized relative to A, and reported as A (A-A), B (B-A), C (C-B) and D (D-B).
BNP ELISA
125 µl of cell culture media was assayed to quantify the level of secreted BNP using the Brain Natriuretic Peptide EIA kit (RAB0386, MilliporeSigma). Each sample was assayed in duplicate on two 96-well plates. BNP levels were quantified relative to a standard curve using 4- and 5-parameter logistic models using the R package drc. Measurements are standardized relative to A, and reported as A (A-A), B (B-A), C (C-B) and D (D-B).
RNA-seq analysis
Reads were aligned to hg19 using subread align (Liao et al., 2013). The mapped reads were then reprocessed to reduce reference bias for downstream analyses using the WASP pipeline (van de Geijn et al., 2015). Briefly, reads overlapping polymorphisms segregating in our population were remapped to the genome using the true read, and a version of the read with the alternative allele. Only reads that mapped uniquely to the same locations with both possible alleles were kept. The median number of reads across conditions was similar (A: 34,353,716; B: 33,493,298; C: 33,883,532; D: 38,147,083). The number of filtered reads mapping to genes was quantified using featureCounts within subread (Liao et al., 2014). We obtained measurements for 19,081 genes. Sample 18852A was an outlier when considering read count correlations between pairs of samples, and was therefore removed prior to subsequent analyses.
Differential expression analysis
We selected autosomal genes for downstream analysis (18,226). Log2-transformed counts per million were calculated (Robinson, 2010), and genes with a mean log2cpm < 0 were excluded. We
used the fact that we have replicate data from three individuals to remove unwanted variation in our data. We used the RUVs function in the RUVSeq package in R (Risso et al., 2014) to identify such factors. By manual inspection, our data segregated by individual or condition after correction with four factors. For the differential expression analysis, we excluded sample replicate one to avoid the outlier sample and randomly selected replicate two, instead of replicate three, for individuals with replicate samples. We used the RUV factors as covariates in our differential expression analysis using the TMM-voom-limma pipeline (Law et al., 2014; Robinson et al., 2010; Smyth, 2004). We used fixed effects for each condition (A, B, C, D), the RUVs factors as covariates, and a random effect for individual, which was implemented using duplicateCorrelation. Genes with a Benjamini and Hochberg FDR < 0.1 are classified as differentially expressed (Benjamini and Hochberg, 1995).
-Gene expression trajectory analysis
To identify response genes, we used the Cormotif package in R (Wei et al., 2015) to jointly model pairs of tests. We used TMM-normalized log_{2}cpm values as input and considered the following pairs of tests: A vs B, B vs. C and B vs. D to determine which genes are changing their expression during the course of the experiment. The best fit was determined to correspond to two correlation motifs or clusters using BIC and AIC. We classified genes as response genes if the probability of differential expression between conditions was > 0.5 in all pairs of tests.
eQTL identification
To map eQTLs, we analyzed the same samples considered in the differential expression analysis. Given the sample size in this study, we utilized the combined haplotype test (CHT) to identify eQTLs (van de Geijn et al., 2015). This test models both allelic imbalance and total read depth at a region to identify QTLs. We require 50 total counts and 10 ten allele-specific counts for each gene, and tested variants 25 kb upstream and 25 kb downstream of the TSS, resulting in 1,040,874 shared tests (A: 1,215,476; B: 1,211,099, C: 1,224,612, D: 1,201,078). As previously reported, we found that null p-values for the CHT were not calibrated in our data. To calibrate the p-values, we estimated the null distribution of the CHT by permuting the data 100 times and fitting a Beta distribution to the permuted p-values for each SNP-gene pair (previously proposed by (Delaneau et al., 2017; Ongen et al., 2016)). We then computed an adjusted p-value for each SNP-gene pair by taking the CDF of the fitted Beta distribution, evaluated at the reported CHT p-value.
To call significant eQTLs, we estimated q-values for the set of adjusted p-values for each phenotype, and took tests with q < 0.1. The number of eGenes in each condition was determined by taking the most significant SNP-gene association in each condition (i.e. the top SNP). We defined dynamic eQTLs as either: 1) significant only in A (q < 0.1 in A and permutation-adjusted p > 0.15 in B and C and D; suppressed eQTL); 2) significant in at least one of B, C, or D (q < 0.1) and not nominally significant in A (adjusted p > 0.15; induced eQTL).
**Power analysis**
For QTL mapping, we assume a linear model
\[ y_i = x_i \beta + \epsilon_i \quad \epsilon_i \sim N(0, \sigma^2) \]
where \( y_i \) denotes the phenotype of individual \( i \) and \( x_i \) denotes the genotype of individual \( i \) at a single SNP of interest. We estimate an effect size \( \hat{\beta} \)
\[ \hat{\beta} \sim N \left( \beta, \frac{\sigma^2}{n} \right) \]
where \( n \) is the sample size. Let \( \lambda = \beta / \sigma \) be the standardized effect size. Then,
\[ \hat{\lambda} \sim N \left( \lambda, \frac{1}{n} \right) \]
and
\[ \text{Power}(\lambda, \alpha, n) = \Phi \left( \Phi^{-1}(\alpha/2) + \lambda \sqrt{n} \right) \]
where \( \alpha \) denotes the significance level and \( \Phi \) denotes the standard Gaussian CDF. To simplify the analysis, we consider \( \alpha = 0.05/20000 = 2.5 \times 10^{-6} \) (i.e., Bonferroni correction; this is equivalent to controlling the FDR when all tests are null, and is conservative otherwise). Assume there is a single causal variant. Then, the phenotypic variance explained is:
\[ h^2 = \frac{\lambda^2}{\lambda^2 + 1} \]
We defined a dynamic eQTL as either significant only in A, or significant (after Bonferroni correction, in this analysis) in one of B, C, or D and not significant in A. To estimate the false
positive rate of dynamic eQTL calling, we asked what was the probability of a SNP passing this definition, assuming the standardized effect size $\lambda$ was identical in all four conditions. We then computed phenotypic variance explained, power to detect an eQTL, and false positive rate to call a dynamic eQTL for every choice of standardized effect size $\lambda$.
**Overlapping response genes and eGenes with existing gene sets**
- **Gene ontology analysis**
Gene set enrichment analysis was performed on response genes, and a background set of all expressed genes using the DAVID genomic annotation tool (Huang da et al., 2009a, b). GO Terms related to Biological Processes were selected, and those with a Benjamini-Hochberg controlled FDR < 0.05 were designated as significantly enriched. Each of the five significantly enriched processes relates to transcription (“DNA-templated transcription”, “DNA-templated regulation of transcription”, “DNA-templated negative regulation of transcription”, “negative regulation of transcription from RNA polymerase II promoter”, “positive regulation of transcription from RNA polymerase II promoter”). The most significantly enriched GO terms related to Molecular Functions include “transcription factor activity, sequence-specific DNA binding”, “nucleic acid binding” and “DNA binding”.
- **Transcription factors**
A list of 1,637 annotated human TFs was obtained from (Lambert et al., 2018), and intersected with our gene sets.
- **GTEx eQTLs**
eQTLs in LV and AA, and twelve other randomly selected tissues (adipose, brain cortex, colon, lung, liver, muscle, pancreas, pituitary, skin, spleen, thyroid, whole blood) were downloaded from v7 in the GTEx portal (www.gtexportal.org). eGenes were selected at 5% FDR in each tissue, and intersected with our gene categories.
**ATAC-seq analysis**
Paired-end sequencing reads were aligned to hg19 using bowtie2 with default settings (Langmead and Salzberg, 2012). Reads were filtered using Picard Tools (https://broadinstitute.github.io/picard/) to remove duplicate reads, and reads mapping to the
mitochondrial genome. Reads were then remapped using the WASP pipeline as described above. We retained a similar median number of reads across conditions (A: 28,998,060; B: 33,662,261; C: 30,161,640; D: 34,534,416). Across conditions, there is no significant difference in the number of mapped reads, number of regions identified, or fraction of reads mapped to open chromatin regions (FigS13A-C). All libraries, across conditions, show the expected fragment size distribution, enrichment of reads at transcription start sites (TSS), and footprints at well-defined CTCF motifs (FigS13D-F). Correlation analysis of read counts between pairs of samples revealed clustering by individual and condition (FigS14). As expected, the correlation of read counts between samples at the 10,633 regions overlapping the TSS is higher than the correlation across all regions (median rho = 0.83 vs. 0.56). Pairs of samples from the same condition are marginally more correlated in their accessibility profiles than pairs of samples across all conditions (median rho = 0.84 vs. 0.83 at the TSS).
Identification of accessible chromatin regions
To generate a unified list of regions with accessible chromatin across conditions and samples, we first used MACS2 (Zhang et al., 2008) to identify peaks within each sample independently. Next, we used BEDtools (Quinlan and Hall, 2010) with the multiIntersectBed function to identify overlapping peaks within each condition separately. Within each condition, we retained peaks with support from more than three individuals and used the mergeBed function to create a condition-specific consensus. We then combined and merged the bed files across the four conditions to make a final consensus file containing all the filtered accessible regions. The number of reads mapping to accessible chromatin regions was quantified using featureCounts within subread (Liao et al., 2014).
Identification of differentially accessible regions (DARs)
The 128,673 open chromatin regions associated with count data were filtered to include only those regions on the autosomes, and those which had mean log2cpm values > 0 for each region. First, to identify differentially accessible regions we used the same limma framework described above for the RNA-seq data. To test for differences between conditions, a linear model with a fixed effect for condition was used together with a random effect for individual. We did not identify any significantly differentially accessible regions with a Benjamini and Hochberg FDR < 0.1. To identify regions with small effect size differences between conditions we used an adaptive shrinkage method.
implemented in the ashr package in R (Stephens, 2017). We used the regression estimates (regression coefficients, posterior standard errors, and posterior degrees of freedom) generated by limma to calculate a posterior mean (shrunken regression coefficients), FDR, and False Sign Rate (FSR, probability that the sign of the effect size is wrong). We considered regions to be differentially accessible at FSR < 0.1. We denote regions that are not differentially accessible as constitutively accessible regions.
**Overlap of DARs with genomic features**
- **TSS**
Transcription start sites were obtained from the UCSC Table Browser (http://genome.ucsc.edu/cgi-bin/hgTables) using ‘txStart’ from Ensembl genes (Karolchik et al., 2004). TSS were defined based on the TSS of the 5’ most transcript on the sense strand and 3’ most transcript on the anti-sense strand. TSS regions, and subsequent genomic features, were intersected with DARs and constitutively accessible regions requiring a 1 bp overlap using bedtools intersect (Quinlan and Hall, 2010).
- **Histone marks**
We obtained histone mark data (.bed files) for human heart tissue from the ENCODE consortium (Consortium, 2012; Davis et al., 2018) ENCODE portal, (https://www.encodeproject.org). We selected H3K4me3 (Experiment ENCSR181ATL), H3K4me1 (Experiment ENCSR449FRQ), H3K36me3 (Experiment ENCSR799KLF), H3K27me3 (Experiment: ENCSR613PPL), and H3K9me3 (Experiment ENCSR803MVC) ChIP-seq data from heart left ventricle tissue from a 51-year-old female individual (Biosample ENCBS684IAD).
- **Transcription factor binding locations**
We obtained ChIP-seq data for the hypoxia-responsive factors HIF1α and HIF2α assayed in the MCF-7 breast cancer cell line (Schodel et al., 2011), and E2F4 in the GM06990 lymphoblastoid cell line (Lee et al., 2011). Co-ordinates of the 356 HIF1α, 301 HIF2α and 16,245 E2F4 bound regions were converted from hg18 to hg19 using the liftOver tool in the Galaxy platform (http://galaxyproject.org/; (Afgan et al., 2018)).
-Motif enrichment analysis in DARs
We obtained sequences for all accessible regions and differentially accessible regions using the Galaxy platform (Afgan et al., 2018). We used the MEME-ChIP tool within The MEME Suite (Bailey et al., 2009; Machanick and Bailey, 2011) in Differential Enrichment mode to identify motifs differentially enriched in DARs compared to all accessible regions.
-TEs
We obtained repeat annotations from the RepeatMasker track (Jurka, 2000; Smith, 2010) from the UCSC Table browser (Karolchik et al., 2004). We intersected the Repeatmasker track with our accessible regions and reported those elements where 50% of their length overlaps a DAR or constitutively accessible region. We stratified TEs by TE class: LINE, SINE, DNA and LTR, and then by TE family and type within the SINE class.
-CpG islands
We obtained CpG island annotations from the UCSC Table Browser, and overlapped these regions with DARs and constitutively accessible regions.
caQTL identification
The caQTLs were identified in the same manner as described for the eQTLs. However, in the caQTL analyses, we limited tested SNPs to those falling within the peak regions from our consensus file, as opposed to testing variants within 25 kb of the region. Dynamic caQTLs were identified as for dynamic eQTLs.
DNA methylation analysis
To allow for accurate quantification of DNA methylation levels we removed probes overlapping SNPs with a minor allele frequency of > 0.1, and only retained probes with a detection p-value of > 0.75 across samples. Beta-values (ratio of methylated probe intensity and overall probe intensity, and bounded between 0-1) were quantile normalized using lumiN, and, when appropriate, converted to M-values (log₂ ratio of intensities of methylated probe versus unmethylated probe) using lumi (Du et al., 2008).
The methylation level of CpGs coincides with the expected distribution based on their annotated genomic location i.e. low levels of DNA methylation in CpG islands, and higher levels in CpG island shores, and CpG island shelves respectively (FigS17B). Correlation analysis across all pairs of samples, including replicate samples, reveals clustering primarily by individual rather than condition (FigS18).
To measure the DNA methylation level at gene set promoters, we selected CpGs 200 bp upstream of the TSS (TSS200 defined on the array). We considered all CpGs when overlapping with DARs.
Identification of differentially methylated CpGs (DMCpGs)
Differentially methylated CpGs were identified using the same limma framework as described for the RNA-seq data. Analysis was run using both Beta-values and M-values.
Integration with GWAS-implicated genes
We intersected the Reference SNP cluster ID of our dynamic QTLs with the 158,654 SNPs in the NHGRI-EBI GWAS Catalog available from the UCSC Table Browser (Buniello et al., 2019) in August 2019.
We also considered the ‘mapped genes’ results from GWAS from thee relevant traits: myocardial infarction (EFO_0000612, 89 genes), heart failure (EFO_0003144, 164 genes) and stroke (EFO_0000712, 255 genes), downloaded from the NHGRI-EBI GWAS Catalog in August 2019. Gene lists were intersected with our response eGenes.
Supplemental material
Supplemental material: Document containing Supplemental Figures 1-20, and Table S1.
Supplemental tables: Document containing Supplemental Tables listing experimental batches, oxygen levels, RIN scores, response genes, eGenes in A, B, C, D, dynamic eGenes, DARs, caQTLs in A, B, C, D, dynamic caQTLs and DMCpGs.
Data Access
All RNA-seq, ATAC-seq and DNA methylation data have been deposited in the Gene Expression Omnibus (www.ncbi.nlm.nih.gov/geo/) under accession number GSE144426.
Acknowledgements
We thank Kristen Patterson and Amy Mitrano for experimental assistance, David Knowles and Benjamin Strober for preliminary data exploration, all members of the Gilad lab and Luis Barreiro for helpful discussions, and Natalia Gonzales for comments on the manuscript. We thank the Genomics Core Facility at the University of Chicago for sequencing the libraries and processing the DNA methylation arrays. We thank The Genotype-Tissue Expression (GTEx) Project, supported by the Common Fund of the Office of the Director of the National Institutes of Health, and by NCI, NHGRI, NHLBI, NIDA, NIMH and NINDS, for providing data. The data used for the analyses described in this manuscript were obtained from the GTEx portal v7 on May 24th 2018. We thank the ENCODE Consortium and the Bernstein Lab at Broad for generating and making the histone mark ChIP-seq data available. M.C.W was supported by an EMBO Long-Term Fellowship (ALTF 751-2014), and the European Commission Marie Curie Actions. N.E.B was supported by NIH grant AG044948. This work was funded by an NIH grant from NHLBI (HL092206) to Y.G.
Contributions
M.C.W, N.E.B and Y.G conceived and designed the study. M.C.W and N.E.B performed experiments. M.C.W, N.E.B, and A.S analyzed the data. M.C.W and Y.G wrote the paper. M.S and Y.G supervised the work.
Disclosure Declaration
The authors declare no competing interests.
References
Afgan, E., Baker, D., Batut, B., van den Beek, M., Bouvier, D., Cech, M., Chilton, J., Clements, D., Coraor, N., Gruning, B.A., et al. (2018). The Galaxy platform for accessible, reproducible and collaborative biomedical analyses: 2018 update. Nucleic Acids Res 46, W537-W544.
Alasoo, K., Rodrigues, J., Mukhopadhyay, S., Knights, A.J., Mann, A.L., Kundu, K., Consortium, H., Hale, C., Dougan, G., and Gaffney, D.J. (2018). Shared genetic effects on chromatin and gene expression indicate a role for enhancer priming in immune response. Nat Genet 50, 424-431.
Bailey, T.L., Boden, M., Buske, F.A., Frith, M., Grant, C.E., Clementi, L., Ren, J., Li, W.W., and Noble, W.S. (2009). MEME SUITE: tools for motif discovery and searching. Nucleic Acids Res 37, W202-208.
Banovich, N.E., Li, Y.I., Raj, A., Ward, M.C., Greenside, P., Calderon, D., Tung, P.Y., Burnett, J.E., Myrthil, M., Thomas, S.M., et al. (2018). Impact of regulatory variation across human iPSCs and differentiated cells. Genome Res 28, 122-131.
Barreiro, L.B., Tailleux, L., Pai, A.A., Gicquel, B., Marioni, J.C., and Gilad, Y. (2012). Deciphering the genetic architecture of variation in the immune response to Mycobacterium tuberculosis infection. Proc Natl Acad Sci U S A 109, 1204-1209.
Batie, M., Frost, J., Frost, M., Wilson, J.W., Schofield, P., and Rocha, S. (2019). Hypoxia induces rapid changes to histone methylation and reprograms chromatin. Science 363, 1222-1226.
Benaglio, P., D’Antonio-Chronowska, A., Ma, W., Yang, F., Young Greenwald, W.W., Donovan, M.K.R., DeBoever, C., Li, H., Drees, F., Singhal, S., et al. (2019). Allele-specific NKX2-5 binding underlies multiple genetic associations with human electrocardiographic traits. Nat Genet 51, 1506-1517.
Benjamini, Y., and Hochberg, Y. (1995). Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society Series B (Methodological) 57, 289-300.
Bogdan, L.B., L.; Bourque, G. (2019). Transposable elements have contributed human regulatory regions that are activated upon bacterial infection. bioRxiv.
Brahimi-Horn, M.C., and Pouyssegur, J. (2007). Oxygen, a source of life and stress. FEBS Lett 581, 3582-3591.
Brodehl, A., Ebbinghaus, H., Deutsch, M.A., Gummert, J., Gartner, A., Ratnavadivel, S., and Milting, H. (2019). Human Induced Pluripotent Stem-Cell-Derived Cardiomyocytes as Models for Genetic Cardiomyopathies. Int J Mol Sci 20.
Buenrostro, J.D., Giresi, P.G., Zaba, L.C., Chang, H.Y., and Greenleaf, W.J. (2013). Transposition of native chromatin for fast and sensitive epigenomic profiling of open chromatin, DNA-binding proteins and nucleosome position. Nat Methods 10, 1213-1218.
Buniello, A., MacArthur, J.A.L., Cerezo, M., Harris, L.W., Hayhurst, J., Malangone, C., McMahon, A., Morales, J., Mountjoy, E., Sollis, E., et al. (2019). The NHGRI-EBI GWAS Catalog of published genome-wide association studies, targeted arrays and summary statistics 2019. Nucleic Acids Res 47, D1005-D1012.
Burridge, P.W., Li, Y.F., Matsa, E., Wu, H., Ong, S.G., Sharma, A., Holmstrom, A., Chang, A.C., Coronado, M.J., Ebert, A.D., et al. (2016). Human induced pluripotent stem cell-derived cardiomyocytes recapitulate the predilection of breast cancer patients to doxorubicin-induced cardiotoxicity. Nat Med 22, 547-556.
Calderon, D., Nguyen, M.L.T., Mezger, A., Kathiria, A., Muller, F., Nguyen, V., Lescano, N., Wu, B., Trombetta, J., Ribot, J.V., et al. (2019). Landscape of stimulation-responsive chromatin across diverse human immune cells. Nat Genet 51, 1494-1505.
Caliskan, M., Baker, S.W., Gilad, Y., and Ober, C. (2015). Host genetic variation influences gene expression response to rhinovirus infection. PLoS Genet 11, e1005111.
Carninci, P., Sandelin, A., Lenhard, B., Katayama, S., Shimokawa, K., Ponjavic, J., Semple, C.A., Taylor, M.S., Engstrom, P.G., Frith, M.C., et al. (2006). Genome-wide analysis of mammalian promoter architecture and evolution. Nat Genet 38, 626-635.
Carreau, A., El Hafny-Rahbi, B., Matejuk, A., Grillon, C., and Kieda, C. (2011). Why is the partial oxygen pressure of human tissues a crucial parameter? Small molecules and hypoxia. J Cell Mol Med 15, 1239-1253.
Chang, S.L., Huang, Y.L., Lee, M.C., Hu, S., Hsiao, Y.C., Chang, S.W., Chang, C.J., and Chen, P.C. (2018). Association of Varicose Veins With Incident Venous Thromboembolism and Peripheral Artery Disease. JAMA 319, 807-817.
Consortium, E.P. (2012). An integrated encyclopedia of DNA elements in the human genome. Nature 489, 57-74.
Consortium, G.T., Laboratory, D.A., Coordinating Center -Analysis Working, G., Statistical Methods groups-Analysis Working, G., Enhancing, G.g., Fund, N.I.H.C., Nih/Nci, Nih/Nhgri, Nih/Nimh, Nih/Nida, et al. (2017). Genetic effects on gene expression across human tissues. Nature 550, 204-213.
Consortium, I.K.C. (2011). Large-scale gene-centric analysis identifies novel variants for coronary artery disease. PLoS Genet 7, e1002260.
Davis, C.A., Hitz, B.C., Sloan, C.A., Chan, E.T., Davidson, J.M., Gabdank, I., Hilton, J.A., Jain, K., Baymuradov, U.K., Narayanan, A.K., et al. (2018). The Encyclopedia of DNA elements (ENCODE): data portal update. Nucleic Acids Res 46, D794-D801.
de la Roche, J., Angsutararux, P., Kempf, H., Janan, M., Bolesani, E., Thiemann, S., Wojciechowski, D., Coffee, M., Franke, A., Schwanke, K., et al. (2019). Comparing human iPSC-cardiomyocytes versus HEK293T cells unveils disease-causing effects of Brugada mutation A735V of NaV1.5 sodium channels. Sci Rep 9, 11173.
Delaneau, O., Ongen, H., Brown, A.A., Fort, A., Panousis, N.I., and Dermitzakis, E.T. (2017). A complete tool set for molecular QTL discovery and analysis. Nat Commun 8, 15452.
Dingar, D., Konecny, F., Zou, J., Sun, X., and von Harsdorf, R. (2012). Anti-apoptotic function of the E2F transcription factor 4 (E2F4)/p130, a member of retinoblastoma gene family in cardiac myocytes. J Mol Cell Cardiol 53, 820-828.
Dombroski, B.A., Nayak, R.R., Ewens, K.G., Ankener, W., Cheung, V.G., and Spielman, R.S. (2010). Gene expression and genetic variation in response to endoplasmic reticulum stress in human cells. Am J Hum Genet 86, 719-729.
Donovan, M.K.R.D.A., A.; D'Antonio, M.; Frazer, K.A. (2019). Cellular deconvolution of GTEx tissues powers eQTL studies to discover thousands of novel disease and cell-type associated regulatory variants. bioRxiv.
Du, J., Leung, A., Trac, C., Lee, M., Parks, B.W., Lusis, A.J., Natarajan, R., and Schones, D.E. (2016). Chromatin variation associated with liver metabolism is mediated by transposable elements. Epigenetics Chromatin 9, 28.
Du, P., Kibbe, W.A., and Lin, S.M. (2008). lumi: a pipeline for processing Illumina microarray. Bioinformatics 24, 1547-1548.
Dzau, V.J., Antman, E.M., Black, H.R., Hayes, D.L., Manson, J.E., Plutzky, J., Popma, J.J., and Stevenson, W. (2006). The cardiovascular disease continuum validated: clinical evidence of improved patient outcomes: part I: Pathophysiology and clinical trial evidence (risk factors through stable coronary artery disease). Circulation 114, 2850-2870.
Feige, E., Yokoyama, S., Levy, C., Khaled, M., Igras, V., Lin, R.J., Lee, S., Widlund, H.R., Granter, S.R., Kung, A.L., et al. (2011). Hypoxia-induced transcriptional repression of the melanoma-associated oncogene MITF. Proc Natl Acad Sci U S A 108, E924-933.
Ferrari, R., de Llobet Cucalon, L.I., Di Vona, C., Le Dilly, F., Vidal, E., Lioutas, A., Oliete, J.Q., Jochem, L., Cutts, E., Dieci, G., et al. (2019). TFIIIC Binding to Alu Elements Controls Gene Expression via Chromatin Looping and Histone Acetylation. Mol Cell.
Findley, A.S., Richards, A.L., Petrini, C., Alazizi, A., Doman, E., Shanku, A.G., Davis, G.O., Hauff, N., Sorokin, Y., Wen, X., et al. (2019). Interpreting Coronary Artery Disease Risk Through Gene-Environment Interactions in Gene Regulation. Genetics 213, 651-663.
Fukaya, E., Flores, A.M., Lindholm, D., Gustafsson, S., Zanetti, D., Ingelsson, E., and Leeper, N.J. (2018). Clinical and Genetic Determinants of Varicose Veins. Circulation 138, 2869-2880.
Ghorbel, M.T., Cherif, M., Jenkins, E., Mokhtari, A., Kenny, D., Angelini, G.D., and Caputo, M. (2010). Transcriptomic analysis of patients with tetralogy of Fallot reveals the effect of chronic hypoxia on myocardial gene expression. J Thorac Cardiovasc Surg 140, 337-345 e326.
Giordano, F.J. (2005). Oxygen, oxidative stress, hypoxia, and heart failure. J Clin Invest 115, 500-508.
Hardin, M., Cho, M.H., McDonald, M.L., Wan, E., Lomas, D.A., Coxson, H.O., MacNee, W., Vestbo, J., Yates, J.C., Agusti, A., et al. (2016). A genome-wide analysis of the response to inhaled beta2-agonists in chronic obstructive pulmonary disease. Pharmacogenomics J 16, 326-335.
Hartley, I., Elkhoury, F.F., Heon Shin, J., Xie, B., Gu, X., Gao, Y., Zhou, D., and Haddad, G.G. (2013). Long-lasting changes in DNA methylation following short-term hypoxic exposure in primary hippocampal neuronal cultures. PLoS One 8, e77859.
He, J., Fu, X., Zhang, M., He, F., Li, W., Abdul, M.M., Zhou, J., Sun, L., Chang, C., Li, Y., et al. (2019). Transposable elements are regulated by context-specific patterns of chromatin marks in mouse embryonic stem cells. Nat Commun 10, 34.
Heinig, M., Adriaens, M.E., Schafer, S., van Deutekom, H.W.M., Lodder, E.M., Ware, J.S., Schneider, V., Felkin, L.E., Creemers, E.E., Meder, B., et al. (2017). Natural genetic variation of the cardiac transcriptome in non-diseased donors and patients with dilated cardiomyopathy. Genome Biol 18, 170.
Huang da, W., Sherman, B.T., and Lempicki, R.A. (2009a). Bioinformatics enrichment tools: paths toward the comprehensive functional analysis of large gene lists. Nucleic Acids Res 37, 1-13.
Huang da, W., Sherman, B.T., and Lempicki, R.A. (2009b). Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc 4, 44-57.
Iyirihiro, G.O., Zhang, Y., Estey, C., O'Hare, M.J., Safarpour, F., Parsanejad, M., Wang, S., Abdel-Messih, E., Callaghan, S.M., During, M.J., et al. (2014). Regulation of ischemic neuronal death by E2F4-p130 protein complexes. J Biol Chem 289, 18202-18213.
Jagannathan, L., Cuddapah, S., and Costa, M. (2016). Oxidative stress under ambient and physiological oxygen tension in tissue culture. Curr Pharmocol Rep 2, 64-72.
Joehanes, R., Zhang, X., Huan, T., Yao, C., Ying, S.X., Nguyen, Q.T., Demirkale, C.Y., Feolo, M.L., Sharopova, N.R., Sturcke, A., et al. (2017). Integrated genome-wide analysis of expression quantitative trait loci aids interpretation of genomic association studies. Genome Biol 18, 16.
Jurka, J. (2000). Repbase update: a database and an electronic journal of repetitive elements. Trends Genet 16, 418-420.
Kariuki, S.N., Maranville, J.C., Baxter, S.S., Jeong, C., Nakagome, S., Hrusch, C.L., Witonsky, D.B., Sperling, A.I., and Di Rienzo, A. (2016). Mapping Variation in Cellular and Transcriptional Response to 1,25-Dihydroxyvitamin D3 in Peripheral Blood Mononuclear Cells. PLoS One 11, e0159779.
Karolchik, D., Hinrichs, A.S., Furey, T.S., Roskin, K.M., Sugnet, C.W., Haussler, D., and Kent, W.J. (2004). The UCSC Table Browser data retrieval tool. Nucleic Acids Res 32, D493-496.
Kim, C., Rubin, C.M., and Schmid, C.W. (2001). Genome-wide chromatin remodeling modulates the Alu heat shock response. Gene 276, 127-133.
Kim-Hellmuth, S., Bechheim, M., Putz, B., Mohammadi, P., Nedelec, Y., Giangreco, N., Becker, J., Kaiser, V., Fricker, N., Beier, E., et al. (2017). Genetic regulatory effects modified by immune activation contribute to autoimmune disease associations. Nat Commun 8, 266.
Knowles, D.A., Burrows, C.K., Blischak, J.D., Patterson, K.M., Serie, D.J., Norton, N., Ober, C., Pritchard, J.K., and Glad, Y. (2018). Determining the genetic basis of anthracycline-cardiotoxicity by molecular response QTL mapping in induced cardiomyocytes. Elife 7.
Koopmann, T.T., Adriaens, M.E., Moerland, P.D., Marsman, R.F., Westerveld, M.L., Lal, S., Zhang, T., Simmons, C.Q., Baczko, I., dos Remedios, C., et al. (2014). Genome-wide identification of expression quantitative trait loci (eQTLs) in human heart. PLoS One 9, e97380.
Lambert, S.A., Jolma, A., Campitelli, L.F., Das, P.K., Yin, Y., Albu, M., Chen, X., Taipale, J., Hughes, T.R., and Weirauch, M.T. (2018). The Human Transcription Factors. Cell 175, 598-599.
Langmead, B., and Salzberg, S.L. (2012). Fast gapped-read alignment with Bowtie 2. Nat Methods 9, 357-359.
Law, C.W., Chen, Y., Shi, W., and Smyth, G.K. (2014). voom: Precision weights unlock linear model analysis tools for RNA-seq read counts. Genome Biol 15, R29.
Lee, B.K., Bhinge, A.A., and Iyer, V.R. (2011). Wide-ranging functions of E2F4 in transcriptional activation and repression revealed by genome-wide analysis. Nucleic Acids Res 39, 3558-3573.
Liao, Y., Smyth, G.K., and Shi, W. (2013). The Subread aligner: fast, accurate and scalable read mapping by seed-and-vote. Nucleic Acids Res 41, e108.
Lee, Y., Smyth, G.K., and Shi, W. (2014). featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923-930.
Lin, H., Dolmatova, E.V., Morley, M.P., Lunetta, K.L., McManus, D.D., Magnani, J.W., Margulies, K.B., Hakonarson, H., del Monte, F., Benjamin, E.J., et al. (2014). Gene expression and genetic variation in human atria. Heart Rhythm 11, 266-271.
Ma, N., Zhang, J.Z., Itzhaki, I., Zhang, S.L., Chen, H., Haddad, F., Kitani, T., Wilson, K.D., Tian, L., Shrestha, R., et al. (2018). Determining the Pathogenicity of a Genomic Variant of...
Uncertain Significance Using CRISPR/Cas9 and Human-Induced Pluripotent Stem Cells. Circulation 138, 2666-2681.
Machanick, P., and Bailey, T.L. (2011). MEME-ChIP: motif analysis of large DNA datasets. Bioinformatics 27, 1696-1697.
Maeda, K., Tsutamoto, T., Wada, A., Hisanaga, T., and Kinoshita, M. (1998). Plasma brain natriuretic peptide as a biochemical marker of high left ventricular end-diastolic pressure in patients with symptomatic left ventricular dysfunction. Am Heart J 135, 825-832.
Manry, J., Nedelec, Y., Fava, V.M., Cobat, A., Orlova, M., Thuc, N.V., Thai, V.H., Laval, G., Barreiro, L.B., and Schurr, E. (2017). Deciphering the genetic control of gene expression following Mycobacterium leprae antigen stimulation. PLoS Genet 13, e1006952.
Maranville, J.C., Luca, F., Richards, A.L., Wen, X., Witonsky, D.B., Baxter, S., Stephens, M., and Di Rienzo, A. (2011). Interactions between glucocorticoid treatment and cis-regulatory polymorphisms contribute to cellular response phenotypes. PLoS Genet 7, e1002162.
McDermott-Roe, C., Lv, W., Maximova, T., Wada, S., Bukowy, J., Marquez, M., Lai, S., Shehu, A., Benjamin, I., Geurts, A., et al. (2019). Investigation of a dilated cardiomyopathy-associated variant in BAG3 using genome-edited iPSC-derived cardiomyocytes. JCI Insight 4.
Medstrand, P., van de Lagemaat, L.N., and Mager, D.L. (2002). Retroelement distributions in the human genome: variations associated with age and proximity to genes. Genome Res 12, 1483-1495.
Narravula, S., and Colgan, S.P. (2001). Hypoxia-inducible factor 1-mediated inhibition of peroxisome proliferator-activated receptor alpha expression during hypoxia. J Immunol 166, 7543-7548.
Nedelec, Y., Sanz, J., Baharian, G., Szpiech, Z.A., Pacis, A., Dumaine, A., Grenier, J.C., Freiman, A., Sams, A.J., Hebert, S., et al. (2016). Genetic Ancestry and Natural Selection Drive Population Differences in Immune Responses to Pathogens. Cell 167, 657-669 e621.
Nikpay, M., Goel, A., Won, H.H., Hall, L.M., Willenborg, C., Kanoni, S., Saleheen, D., Kyriakou, T., Nelson, C.P., Hopewell, J.C., et al. (2015). A comprehensive 1,000 Genomes-based genome-wide association meta-analysis of coronary artery disease. Nat Genet 47, 1121-1130.
Ongen, H., Buil, A., Brown, A.A., Dermitzakis, E.T., and Delaneau, O. (2016). Fast and efficient QTL mapper for thousands of molecular phenotypes. Bioinformatics 32, 1479-1485.
Pacis, A., Mailhot-Leonard, F., Tailleux, L., Randolph, H.E., Yotova, V., Dumaine, A., Grenier, J.C., and Barreiro, L.B. (2019). Gene activation precedes DNA demethylation in response to infection in human dendritic cells. Proc Natl Acad Sci U S A 116, 6938-6943.
Pacis, A., Tailleux, L., Morin, A.M., Lambourne, J., Maclsaac, J.L., Yotova, V., Dumaine, A., Danckaert, A., Luca, F., Grenier, J.C., et al. (2015). Bacterial infection remodels the DNA methylation landscape of human dendritic cells. Genome Res 25, 1801-1811.
Panopoulos, A.D., D'Antonio, M., Benaglio, P., Williams, R., Hashem, S.I., Schuld, B.M., DeBoever, C., Arias, A.D., Garcia, M., Nelson, B.C., et al. (2017). iPSCORE: A Resource of 222 iPSC Lines Enabling Functional Characterization of Genetic Variation across a Variety of Cell Types. Stem Cell Reports 8, 1086-1100.
Pavlovic, B.J., Blake, L.E., Roux, J., Chavarria, C.I., and Gilad, Y. (2018). A Comparative Assessment of Human and Chimpanzee iPSC-derived Cardiomyocytes with Primary Heart Tissues. bioRxiv.
Pickrell, J.K., Marioni, J.C., Pai, A.A., Degner, J.F., Engelhardt, B.E., Nkadori, E., Veyrieras, J.B., Stephens, M., Gilad, Y., and Pritchard, J.K. (2010). Understanding mechanisms underlying human gene expression variation with RNA sequencing. Nature 464, 768-772.
Pinto, A.R., Ilinykh, A., Ivey, M.J., Kuwabara, J.T., D'Antoni, M.L., Debuque, R., Chandran, A., Wang, L., Arora, K., Rosenthal, N.A., et al. (2016). Revisiting Cardiac Cellular Composition. Circ Res 118, 400-409.
Platt, J.L., Salama, R., Smythies, J., Choudhry, H., Davies, J.O., Hughes, J.R., Ratcliffe, P.J., and Mole, D.R. (2016). Capture-C reveals preformed chromatin interactions between HIF-binding sites and distant promoters. EMBO Rep 17, 1410-1421.
Poetsch, A.R., Boulton, S.J., and Luscombe, N.M. (2018). Genomic landscape of oxidative DNA damage and repair reveals regioselective protection from mutagenesis. Genome Biol 19, 215.
Quinlan, A.R., and Hall, I.M. (2010). BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26, 841-842.
Ramirez-Carrozzi, V.R., Braas, D., Bhatt, D.M., Cheng, C.S., Hong, C., Doty, K.R., Black, J.C., Hoffmann, A., Carey, M., and Smale, S.T. (2009). A unifying model for the selective regulation of inducible transcription by CpG islands and nucleosome remodeling. Cell 138, 114-128.
Risso, D., Ngai, J., Speed, T.P., and Dudoit, S. (2014). Normalization of RNA-seq data using factor analysis of control genes or samples. Nat Biotechnol 32, 896-902.
Robinson, C.M., Neary, R., Levendale, A., Watson, C.J., and Baugh, J.A. (2012). Hypoxia-induced DNA hypermethylation in human pulmonary fibroblasts is associated with Thy-1 promoter methylation and the development of a pro-fibrotic phenotype. Respir Res 13, 74.
Robinson, M.D., McCarthy, D.J., and Smyth, G.K. (2010). edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics 26, 139-140.
Samanta, D., and Semenza, G.L. (2017). Maintenance of redox homeostasis by hypoxia-inducible factors. Redox Biol 13, 331-335.
Schodel, J., Oikonomopoulos, S., Ragoussis, J., Pugh, C.W., Ratcliffe, P.J., and Mole, D.R. (2011). High-resolution genome-wide mapping of HIF-binding sites by ChIP-seq. Blood 117, e207-217.
Schunkert, H., Konig, I.R., Kathiresan, S., Reilly, M.P., Assimes, T.L., Holm, H., Preuss, M., Stewart, A.F., Barbatic, M., Gieger, C., et al. (2011). Large-scale association analysis identifies 13 new susceptibility loci for coronary artery disease. Nat Genet 43, 333-338.
Shah, S.H., A.; Roselli, C.; Honghuang, L.; Sveinbjornsson, G.; Fatemifar, G. ... (2019). Genome-wide association study provides new insights into the genetic architecture and pathogenesis of heart failure. bioRxiv.
Sigurdsson, M.I., Saddic, L., Heydarpour, M., Chang, T.W., Shekar, P., Aranki, S., Couper, G.S., Sherman, S.K., Muehlschlegel, J.D., and Body, S.C. (2017). Post-operative atrial fibrillation examined using whole-genome RNA sequencing in human left atrial tissue. BMC Med Genomics 10, 25.
Smirnov, D.A., Brady, L., Halasa, K., Morley, M., Solomon, S., and Cheung, V.G. (2012). Genetic variation in radiation-induced cell death. Genome Res 22, 332-339.
Smith, A., Hubley, R. & Green, P. (2010). RepeatMasker.
Strober, B.J., Elorbany, R., Rhodes, K., Krishnan, N., Tayeb, K., Battle, A., and Gilad, Y. (2019). Dynamic genetic regulation of gene expression during cellular differentiation. Science 364, 1287-1290.
Unoki, M., and Nakamura, Y. (2003). Methylation at CpG islands in intron 1 of EGR2 confers enhancer-like activity. FEBS Lett 554, 67-72.
van de Geijn, B., McVicker, G., Gilad, Y., and Pritchard, J.K. (2015). WASP: allele-specific software for robust molecular quantitative trait locus discovery. Nat Methods 12, 1061-1063.
Ward, M.C., and Gilad, Y. (2019). A generally conserved response to hypoxia in iPSC-derived cardiomyocytes from humans and chimpanzees. Elife 8.
Watson, C.J., Collier, P., Tea, I., Neary, R., Watson, J.A., Robinson, C., Phelan, D., Ledwidge, M.T., McDonald, K.M., McCann, A., et al. (2014). Hypoxia-induced epigenetic modifications are associated with cardiac tissue fibrosis and the development of a myofibroblast-like phenotype. Hum Mol Genet 23, 2176-2188.
Wei, Y., Tenzen, T., and Ji, H. (2015). Joint analysis of differential gene expression in multiple studies using correlation motifs. Biostatistics 16, 31-46.
WHO (2018). World Health Statistics 2018: Monitoring health for the SDGs.
Zhang, Y., Liu, T., Meyer, C.A., Eeckhoute, J., Johnson, D.S., Bernstein, B.E., Nusbaum, C., Myers, R.M., Brown, M., Li, W., et al. (2008). Model-based analysis of ChIP-Seq (MACS). Genome Biol 9, R137. | 2025-03-06T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
792,
1
],
[
792,
3478,
2
],
[
3478,
6337,
3
],
[
6337,
8945,
4
],
[
8945,
9827,
5
],
[
9827,
12556,
6
],
[
12556,
14183,
7
],
[
14183,
17060,
8
],
[
17060,
19848,
9
],
[
19848,
22544,
10
],
[
22544,
23128,
11
],
[
23128,
25706,
12
],
[
25706,
28569,
13
],
[
28569,
29842,
14
],
[
29842,
32423,
15
],
[
32423,
33581,
16
],
[
33581,
36329,
17
],
[
36329,
37935,
18
],
[
37935,
40936,
19
],
[
40936,
43817,
20
],
[
43817,
46793,
21
],
[
46793,
49763,
22
],
[
49763,
52539,
23
],
[
52539,
54983,
24
],
[
54983,
57401,
25
],
[
57401,
59179,
26
],
[
59179,
61142,
27
],
[
61142,
63314,
28
],
[
63314,
65975,
29
],
[
65975,
67831,
30
],
[
67831,
69933,
31
],
[
69933,
72580,
32
],
[
72580,
74598,
33
],
[
74598,
76434,
34
],
[
76434,
78318,
35
],
[
78318,
81474,
36
],
[
81474,
85289,
37
],
[
85289,
89285,
38
],
[
89285,
93327,
39
],
[
93327,
97237,
40
],
[
97237,
100696,
41
],
[
100696,
101447,
42
]
]
} |
f94903be45fe7d71e54c2dbde6c84875efc850c4 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 10,
"total-fallback-pages": 0,
"total-input-tokens": 31032,
"total-output-tokens": 9498,
"fieldofstudy": [
"Medicine"
]
} | The Effectiveness of Metagenomic Next-Generation Sequencing in the Diagnosis of Prosthetic Joint Infection: A Systematic Review and Meta-Analysis
Jun Tan¹, Yang Liu², Sabrina Ehnert³, Andreas K. Nüssler³, Yang Yu¹, Jianzhong Xu¹* and Tao Chen¹*
¹ Department of Orthopedic Surgery, The First Affiliated Hospital of Zhengzhou University, Zhengzhou, China, ² Department of Clinical Sciences, Orthopedics, Faculty of Medicine, Lund University, Lund, Sweden, ³ Department of Trauma and Reconstructive Surgery, BG Trauma Center Tübingen, Siegfried Weller Institute for Trauma Research, Eberhard Karls University Tübingen, Tübingen, Germany
Background: A prosthetic joint infection (PJI) is a devastating complication following total joint arthroplasties with poor prognosis. Identifying an accurate and prompt diagnostic method is particularly important for PJI. Recently, the diagnostic value of metagenomic next-generation sequencing (mNGS) in detecting PJI has attracted much attention, while the evidence of its accuracy is quite limited. Thus, this study aimed to evaluate the accuracy of mNGS for the diagnosis of PJI.
Methods: We summarized published studies to identify the potential diagnostic value of mNGS for PJI patients by searching online databases using keywords such as “prosthetic joint infection”, “PJI”, and “metagenomic sequencing”. Ten of 380 studies with 955 patients in total were included. The included studies provided sufficient data for the completion of 2-by-2 tables. We calculated the sensitivity, specificity, and area under the SROC curve (AUC) to evaluate the accuracy of mNGS for the diagnosis of PJI.
Results: We found that the pooled diagnostic sensitivity and specificity of mNGS for PJI were 0.93 (95% CI, 0.83 to 0.97) and 0.95 (95% CI, 0.92 to 0.97), respectively. Positive and negative likelihood ratios were 18.3 (95% CI, 10.9 to 30.6) and 0.07 (95% CI, 0.03 to 0.18), respectively. The area under the curve was 0.96 (95% CI, 0.93 to 0.97).
Conclusion: Metagenomic next-generation sequencing displays high accuracy in the diagnosis of PJI, especially for culture-negative cases.
Keywords: metagenomics, next-generation sequencing, clinical diagnosis and treatment, arthroplasty, infection disease, prosthetic joint infection
INTRODUCTION
Prosthetic joint infection (PJI), noted as a devastating complication of prosthetic joint implantation, accounts for 25% of failed knee arthroplasties and 15% of failed hip arthroplasties (Bozic et al., 2010; Rietbergen et al., 2016). PJI after joint arthroplasty has extreme adverse effects on cost and quality of life. In recent years, with the increasing number of cases, the proportion of its cost in the healthcare budgets is also increasing (Källala et al., 2018). It is estimated that each episode of prosthetic infection costs the health service over 20,000 pounds (Vanhegan et al., 2012).
To date, the timely and accurately diagnosis of PJI is still challenging, especially for the identification of pathogenic microorganisms. Although many methods have emerged for establishing the diagnosis, none has been universally accepted (Moojen et al., 2014). Nowadays, traditional blood testing, such as white blood cell (WBC) count, serum erythrocyte sedimentation rate (ESR), and serum C-reactive protein concentration (CRP) are being widely performed for diagnosing PJI in clinics. Nonetheless, these inflammatory markers are nonspecific for PJI, and sometimes they may even be normal in severe cases of joint infections (Nodzo et al., 2015). In addition, routine microbial culture has also been widely used to identify causative organisms in PJI, but it has a significantly high false-negative rate (Rak et al., 2013; Yoon et al., 2017). It has been reported that approximately 40% of culture-negative cases meet the clinical diagnostic criteria for PJI, which might be due to the restricted growth conditions of specific pathogens and the widespread use of antibiotics (Tande and Patel, 2014). In recent years, matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS) has emerged for the identification of bacterial in clinical laboratories (Peel et al., 2015). The MALDI-TOF MS process is rapid, sensitive and economical in terms of labor and costs involved, in which microbes are identified using either intact cells or cell extracts. Although it shows high accuracy for the direct identification of Gram-negative bacteria from blood culture, the accuracy for Gram-positive bacteria is moderate (Ruiz-Aragón et al., 2018). Therefore, the research and development of a reliable technique for the diagnosis of PJI are urgent for both patients and clinicians.
In recent years, researches on PJI diagnosis have switched to mNGS. mNGS is a rapidly developing technology in terms of both pathogenic microorganism detection and data analysis. It has been shown to play important roles in the diagnosis of cancers, genetic diseases, and infectious diseases (Kwon et al., 2019; Wilson et al., 2019). Compared to PCR, mNGS does not have limitations on the detection of specific pathogens, and it can detect almost all pathogens, such as bacteria, fungi, viruses, and parasites. Furthermore, it allows thousands or even billions of DNA fragments to be sequenced independently at the same time, and its consequence is confirmed through comparison with a dedicated pathogen database (Schlaberg et al., 2017; Gu et al., 2019).
mNGS has shown high value in the diagnosis of pathogens of many infectious diseases. In a study on tuberculous meningitis, the diagnostic sensitivity of mNGS based on cerebrospinal fluid was 84.44%, which was much higher than the 22.2% of traditional cerebrospinal fluid culture (Yan et al., 2020). Another study showed that the sensitivity of mNGS was much higher than that of traditional culture in the pathogen diagnosis of mixed lung infections (97.2% vs 13.9%, P<0.01) (Wang et al., 2019).
In 2019, a systematic review discussed the sequencing assays for the diagnosis of PJI and showed a low statistical power, owing to a few studies regarding mNGS was involved in this study. Additionally, this review failed to analyze and evaluate the accuracy and diagnostic value of mNGS for PJI (Li et al., 2019). Herein, we incorporated the latest clinical trials for this systematic review to summarize published studies about mNGS. We also performed a meta-analysis to investigate its diagnostic accuracy for PJI.
MATERIALS AND METHODS
The protocol for this review was registered with the PROSPERO database, registration number CRD42020193251. We strictly adhered to the standards of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) in reporting the findings of this review (Moher et al., 2015).
Search Strategy
We carefully searched for longitudinal studies (prospective or retrospective case-control, prospective cohort, retrospective cohort, case-cohort, nested-case control trials) reporting on the use of mNGS for PJI in MEDLINE, EMBASE, China National Knowledge Internet (CNKI), and Cochrane Library databases from inception to July 2021. A systematic literature search was performed to obtain all of the published articles focusing on mNGS diagnosis of PJI. Vocabulary and syntax were adjusted according to the different databases. We mainly used “prosthetic joint infection”, “periprosthetic joint infection”, “PJI”, “prosthesis-related infections”, “prosthesis infection”, “infection”, and “metagenomic sequencing”, “mNGS”, “metagenomic next-generation sequencing”, “shotgun metagenomics”, “genomics”, “genetic diagnosis”, “sequencing”, as the search target keywords. The exact retrieval strategy is demonstrated in Supplementary S1. Reference lists of the retrieved articles were manually scanned for all relevant additional studies and review articles.
Study Selection
The screening was performed in two stages, title and abstract screening, followed by full-text screening. A gold standard for diagnosing PJI has not been established, and different studies may adopt different reference standards. Among these reference standards, Musculoskeletal Infection Society (MSIS) (Parvizi et al., 2011) and Infectious Diseases Society of America (IDSA) (Osmon et al., 2013) are commonly used. We included studies with different reference standards, and then investigated the heterogeneity between MSIS and IDSA as reference standards through subgroup analysis. Two researchers independently reviewed the title and abstract of each study to select those likely to meet the inclusion criteria. In the
The initial stage of the screening, 10 to 12 articles were used to confirm the agreement between the researchers. To achieve at a consensus, any discrepancy was resolved by discussion or with the assistance of a third reviewer. After full-text screening, a list of excluded studies with reasons for exclusion was performed.
Studies were considered eligible for inclusion if they met the following criteria: (1) patients with suspected PJ; following primary or revision total hip or knee arthroplasty; (2) focus on mNGS-based diagnosis of PJI; (3) the diagnosis of PJI was confirmed by MSIS or IDSA; (4) false positive (FP), true positive (TP), false negative (FN), and true negative (TN) were provided to construct the $2 \times 2$ contingency table. Articles were excluded based on the following criteria: (1) Irrelevant reviews, letters, personal opinions, book chapters, and meeting abstracts; (2) insufficient data to calculate sensitivity and specificity; (3) mNGS and PJI were not studied.
**Quality Assessment**
The quality of the included studies was evaluated by two researchers using the revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS)-2 (Whiting et al., 2011), which is comprised of four key domains that focus on patient selection, index test, reference standard, flow, and timing. Signaling questions were applied to assess the risk of bias and clinical applicability. The overall risk of bias and applicability was summarized as low, high, or unclear.
**Data Extraction**
Two reviewers independently extracted the data from the included studies using a standardized form. Data extraction included the following items: last name of the first author; publication year; study population and regions; false and true positives and negatives; sample site and type; reference standard and study design. To deal with absent or unclear data, we tried to contact the study authors.
**Statistical Analysis**
Overall pooled sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and the corresponding 95% CI for the diagnosis of PJI were calculated using a bivariate meta-analysis framework. We also tested the pooled diagnostic value of mNGS through the summary receiver operating characteristic (SROC) curve and the area under the SROC curve (AUC). We assessed heterogeneity among the studies using the chi-squared and I2 tests. Moreover, subgroup and sensitivity analyses were undertaken to explore the potential sources of heterogeneity. All analyses were undertaken by using RevMan 5.4 (The Nordic Cochrane Centre, The Cochrane Collaboration, London, UK 2020) and Stata 15.0 (Stata Corporation, College Station, TX, USA), and a value of $P < 0.05$ was considered statistically significant.
**RESULTS**
The selection process was shown in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flowchart in Figure 1. 380 relevant articles were identified for initial review by systematically searching in the aforementioned databases. Of the identified 380 articles, 253 duplicates were excluded. Then, 106 articles were excluded due to inappropriate article types (reviews, comments, or letters). After reading the remaining 21 articles in full text, seven were excluded due to insufficient data, and four were excluded due to not being an original diagnostic study. Ten studies were finally included in this meta-analysis (Ivy et al., 2018; Thoendel et al., 2018; Huang et al., 2019; Zhang et al., 2019; Cai et al., 2020; Fang et al., 2020; Huang et al., 2020; Wang et al., 2020; He et al., 2021; Yu et al., 2021).
These 10 studies, including a total of 955 patients, were published between 2018 and 2021. Among the included studies, two (Huang et al., 2019; Yu et al., 2021) were conducted prospectively, and the other studies were conducted prospectively. Eight studies (Ivy et al., 2018; Huang et al., 2019; Cai et al., 2020; Fang et al., 2020; Huang et al., 2020; Wang et al., 2020; He et al., 2021; Yu et al., 2021) collected synovial fluid samples before any clinical treatment, six studies (Thoendel et al., 2018; Huang et al., 2019; Zhang et al., 2019; Fang et al., 2020; Wang et al., 2020; He et al., 2021) obtained sonication fluid and two studies (Cai et al., 2020; He et al., 2021) selected periprosthetic tissue for mNGS. The MSIS criteria were used in seven studies (Huang et al., 2019; Zhang et al., 2019; Cai et al., 2020; Fang et al., 2020; Huang et al., 2020; He et al., 2021; Yu et al., 2021), and the other three studies (Ivy et al., 2018; Thoendel et al., 2018; Wang et al., 2020) adopted the IDSA as the only reference standard. Among the ten studies analyzed, nine studies (Thoendel et al., 2018; Huang et al., 2019; Zhang et al., 2019; Cai et al., 2020; Fang et al., 2020; Huang et al., 2020; Wang et al., 2020; He et al., 2021; Yu et al., 2021) focused on both hip and knee while one study (Ivy et al., 2018) only enrolled knee arthroplasty (Table 1). A graphical summary of the methodological assessment based on the QUADAS-2 quality assessment for the 10 studies is shown in Figures 2A, B. Included studies were assessed with the QUADAS-2 guidelines, and detailed information is shown in Supplementary Table S1. The majority of studies had a low risk of bias for patient selection, reference standard, flow, and timing. For index test bias, six studies were at an unclear risk because the information was insufficient to ensure that the index test results were interpreted without knowledge of the results of the reference standard. Most of the studies in this meta-analysis raised low concerns about applicability.
The sensitivity and specificity of mNGS for diagnosing PJI are shown in Figure 3. The pooled sensitivity was 0.93 (95% CI, 0.83–0.97), specificity was 0.95 (95% CI, 0.92–0.97), positive likelihood ratio was 18.3 (95% CI, 10.9–30.6), negative likelihood ratio was 0.07 (95% CI, 0.03–0.18), and DOR was 247 (95% CI, 84–723). Moreover, we plotted the SROC curve to evaluate diagnostic accuracy (Figure 4). AUC was 0.96 (95% CI, 0.93–0.97), suggesting a unique superior diagnostic accuracy of mNGS.
The performance of mNGS in both culture-positive and culture-negative is indicated in Table 2. In all 565 specimens tested in all publications, 375 (66.4%) were culture-positive and 190 (33.6%) were culture-negative. In 375 culture-positive
specimens, the pathogens identified by culture were also detected by metagenomics in 340 (90.1%) cases. In 190 specimens considered as culture-negative, potential pathogens were detected in 103 (54.2%) using metagenomics.
For sensitivity analysis, the goodness of fit and bivariate normality showed that a random-effects bivariate model is suitable (Figures 5A, B). Influence analysis identified that the studies of Thoendel et al. (Thoendel et al., 2018), Ivy et al. (Ivy et al., 2018), and Yu et al. (Yu et al., 2021) were the most dominant studies in weight (Figure 5C). Outlier detection implied that the studies of Ivy et al. (Ivy et al., 2018) and Yu et al. (Yu et al., 2021) might be the reason for the heterogeneity (Figure 5D). The Spearman correlation coefficient of sensitivity and 1-specificity was 0.418, and the P-value was 0.229, indicating that heterogeneity may not be caused by the threshold effect (Supplementary S2). We conducted an univariable meta-regression analysis based on the characteristics of the ten studies to explore the potential sources of heterogeneity. We found that sensitivity was affected by ethnicity, sample site, and study design, while specificity was influenced by ethnicity, sample type, and reference standard (Figure S1).
We performed subgroup analysis according to the results of univariate meta-regression to further investigate the sources of heterogeneity. If $I^2 < 50\%$, or $P > 0.05$, heterogeneity in this subgroup was defined as low. Between these subgroup analyses, ethnicity, sample type, and reference standard showed low heterogeneity (Table 3).
TABLE 1 | Characteristics of the studies that were included.
| Study | Country | Patients | Study design | Sample site(s) | Reference standard | Sample type | Antibiotics* | TP | FP | FN | TN |
|------------------|---------|----------|--------------|-----------------------------|--------------------|------------------------------------------|--------------|------|----|----|----|
| Thoendel et al., 2018 | USA | 408 | Prospective | Hip and knee | IDSA | Sonication fluid | Yes | 251 | 7 | 62 | 188|
| Ivy et al., 2018 | USA | 168 | Prospective | Knee | IDSA | Synovial fluid | Yes | 72 | 4 | 35 | 57 |
| Zhang et al., 2019 | China | 37 | Prospective | Hip and knee | MSIS | Sonication fluid | Yes | 24 | 1 | 0 | 12 |
| Huang et al., 2019 | China | 35 | Prospective | Hip and knee | MSIS | Synovial and sonication fluid | Yes | 20 | 1 | 0 | 14 |
| Cai et al., 2020 | China | 44 | Prospective | Hip and knee | MSIS | Periprosthetic tissue and synovial fluid | No | 21 | 2 | 1 | 20 |
| Wang et al., 2020 | China | 63 | Prospective | Hip and knee | IDSA | Synovial and sonication fluid | No | 43 | 1 | 2 | 17 |
| Huang et al., 2020 | China | 70 | Prospective | Hip and knee | MSIS | Synovial fluid | Yes | 47 | 1 | 2 | 20 |
| Fang et al., 2020 | China | 38 | Prospective | Hip and knee | MSIS | Synovial and sonication fluid | Yes | 24 | 0 | 1 | 13 |
| He et al., 2021 | China | 59 | Prospective | Hip and knee | MSIS | Synovial, sonication fluid and tissues | Yes | 38 | 1 | 2 | 18 |
| Yu et al., 2021 | China | 33 | Retrospective | Hip and knee | MSIS | Synovial fluid | Yes | 13 | 1 | 8 | 11 |
TP, true positive; FP, false positive; FN, false negative; TN, true negative; MSIS, Musculoskeletal Infection Society; IDSA, Infectious Disease Society of America guidelines.
*Only antibiotics before sampling are considered here.
Moreover, The Deeks’ funnel plot asymmetry test of pooled DOR with a P-value of 0.20 indicated no significant publication bias (Figure S2).
DISCUSSION
Although mNGS has demonstrated an encouraging value in the diagnosis of pathogens of various infectious diseases, especially for diagnosing tuberculous meningitis and chlamydia psittaci pneumonia, consensus for its clinical application of PIJ diagnosis has still not yet been achieved (Chen et al., 2020; Yan et al., 2020). A former meta-analysis suggested that sequencing assays have the potential to improve the clinical diagnosis of PIJ, especially for culture-negative cases, but the diagnostic value and accuracy of mNGS in PIJ were still unclear (Li et al., 2019). According to our literature search, no previous systematic review or meta-analysis about mNGS in the diagnosis of PIJ has been published, which makes it necessary to explore and fill this gap.
Our findings suggested that mNGS had a high accuracy in PIJ diagnostics, with a pooled sensitivity of 0.93, a pooled specificity of 0.95, and an AUC of 0.96. The pooled PLR was 18.3, indicating that the probability of an accurate diagnosis of PIJ increased by 18.3-fold with positive mNGS testing. Moreover, NLR was 0.07, implying that the probability of a PIJ decreased by 93% when the studied mNGS was negative.
Li et al. (Li et al., 2019) showed that the sensitivity, specificity, and AUC of sequencing assays were 0.81, 0.94, and 0.94, respectively. The pooled sensitivity and specificity were both lower than the data of our study (0.81 vs 0.93; 0.94 vs 0.95). The AUC, which is usually used to indicate overall accuracy, was also lower than our study (0.94 vs. 0.96), supporting the idea that mNGS might be more effective in the diagnosis of PIJ than other sequencing assays. There are several potential reasons for the higher sensitivity and AUC in our study: our study only focused on the diagnostic accuracy of mNGS, while Li et al. used different sequencing methods, including Sanger sequencing, Sequencing by Synthesis and NGS methods. mNGS technology can simultaneously and independently detect pathogens and multiple target genes in the same clinical samples without the need of pre-amplify target sequences (Gu et al., 2019). The ability of mNGS to effectively identify most pathogens in the joint fluid of PIJ may have contributed to this result.
In another study of broad-range PCR-based (BR-PCR) diagnosis of PIJ (Wang et al., 2020), the pooled sensitivity and specificity were 0.82 and 0.94, respectively, which were both also lower than in our analysis (0.82 vs 0.93; 0.94 vs 0.95). These results were likely caused by different sequencing procedures between mNGS and BR-PCR. The outstanding advantage of mNGS is unbiased sampling, which can broadly identify known and unexpected pathogens and even discover new organisms in an unbiased approach (Gu et al., 2019). BR-PCR is based on the V3-V4 region of 16S rDNA, which can only identify some pathogens at the genus level and may miss the causative pathogens in polymicrobial infection and fungal infections (Dabrowski et al., 2017).
According to subgroup analysis, the effectiveness of mNGS in the diagnosis of PIJ among Asians seems to have a better sensitivity than that of Caucasians (0.94 vs. 0.77), while the specificity in Caucasians was slightly higher than in Asians (0.96 vs. 0.95). In fact, the total number of Caucasians studies was much larger than that of Asians (576 vs. 379) and a different platform was used to perform mNGS in the included study. We assume that this may cause the significant difference in sensitivity and specificity between Asians and Caucasians. Therefore, it is necessary to carry out more high-quality clinical trials of different ethnicities to explore racial differences in mNGS. Besides, the significant differences among sample types were considered as the main source of heterogeneity in specificity. Sequencing of sonication fluid seems to have a better specificity than other sample types, while multiple sample types sequencing had better sensitivity than other sample types. In fact, the ultrasonic lysis method can peel the thickened joint fluids that are difficult to centrifuge, ultrasonic lysis fluids could achieve a 20-fold higher
concentration of microbial cells after centrifugation and increase the sensitivity of diagnosis. Nevertheless, the ultrasonic lysis procedure may introduce exogenous microbial cells and nucleic acid fragments. Therefore, the additional pathogenic bacteria detected in the ultrasonic lysate should be further verified by specific PCR or other methods to exclude the possibility of exogenous contamination.
Our results showed that the sensitivity of MSIS was better than IDSA (0.930 vs. 0.787), while the specificity of MSIS was lower than IDSA (0.939 vs. 0.956). However, some information important for determining the cases with low virulence levels may be missed by using different reference standards and therefore resulting in the wrong grouping method. For that reason, a common and widely accepted reference standard should be established to help to minimize classification bias.
The main pathogenic microorganisms of PJI obtained by mNGS were Staphylococcus epidermidis (25.1%, 139/553) and Staphylococcus aureus (17.5%, 97/553), which is similar to the common microbiological causes of PJI reported by Tande et al. (Tande and Patel, 2014). It is noteworthy that metagenomics is able to detect most pathogens identified by culture (90.1%) as well as many that were not detected by culture. This occurs...
particularly in the culture-negative PJI group in which potential pathogens were detected in 54.2% of cases. This result supported the idea that mNGS is a powerful tool to identify PJI pathogens that are difficult to detect in culture-negative infections. Importantly, mNGS will become more accurate and offer more comprehensive microbiologic diagnosis as the technology evolves.
Helping clinical decision-making is the most important value of mNGS. Likelihood ratios and post-test probabilities are useful for clinicians, as they could show the probability that a patient has or does not have PJI, given a negative or positive test result. We also summarized the positive likelihood ratios and negative likelihood ratios to judge the clinical applicability of mNGS for diagnosis (Figure S3). PLR >10 and NLR <0.1 represent a high diagnostic accuracy (Wacker et al., 2013). We found that the articles of mNGS from Wang et al. (Wang et al., 2020), Huang et al. (Huang et al., 2020), Fang et al. (Fang et al., 2020), and He et al. (He et al., 2021) had high diagnostic accuracy and clinical applicability. When the pre-test probability was set at 50%, the post-test probability for a positive test result was 95%. When the negative likelihood ratio was set at 0.07, the post-test probability was reduced to 7% for a negative test result (Figure S4).
mNGS offers a novel approach to diagnose clinical infectious diseases and address current pitfalls in clinical management. Although the valuable insights of mNGS have already been derived, its use in the diagnosis of PJI is still in its infancy and many challenges still exist (Han et al., 2019). In particular, it is difficult to detect pathogenic virulence and drug sensitivity, which limits its role in guiding the rational selection of antibiotics. Another challenge is no comprehensive and unified background bacteria identification strategy, making interpretation of the sequencing results difficult. It seems inevitable to mix microbial gene sequences during sampling and laboratory testing, which makes it difficult to identify the real pathogen. Moreover, the high cost and lack of timeliness also limit the clinical applications of this technology.
In addition, several limitations of this meta-analysis should be emphasized. It is hard to elucidate whether the sample site had a decisive influence on diagnostic accuracy since the raw data were not provided in the published articles and we cannot divide the data into hip and knee to eliminate heterogeneity. Future studies should focus on the differences in diagnostic accuracy associated with potential sources of heterogeneity, including different arthroplasty sites. Secondly, the gold standard for diagnosing
| Study | CP-PJI | Organisms identified by metagenomics | CN-PJI | Organisms identified by metagenomics |
|------------------------|--------|--------------------------------------|--------|--------------------------------------|
| Cai et al., 2020 | 16 | 16 (100%) | 6 | 5 (83.3%) |
| Wang et al., 2020 | 35 | 33 (94.3%) | 10 | 10 (100%) |
| Huang et al., 2019 | 13 | 12 (92.3%) | 7 | 6 (85.7%) |
| Thoendel et al., 2018 | 115 | 109 (94.8%) | 98 | 43 (43.9%) |
| Ivy et al., 2018 | 82 | 69 (84.1%) | 25 | 4 (16.0%) |
| Huang et al., 2020 | 39 | 37 (94.9%) | 10 | 10 (100%) |
| Zhang et al., 2019 | 17 | 17 (100%) | 7 | 7 (100%) |
| He et al., 2021 | 34 | 34 (100%) | 6 | 4 (66.7%) |
| Fang et al., 2020 | 18 | 18 (100%) | 6 | 6 (100%) |
| Yu et al., 2021 | 6 | 5 (83.3%) | 15 | 8 (53.3%) |
| Total | 375 | 340 (90.1%) | 190 | 103 (54.2%) |
Listed are the numbers of samples that were detected by metagenomics in culture-positive and culture-negative PJI samples. CP-PJI, culture-positive prosthetic joint infection. CN-PJI, culture-negative prosthetic joint infection.
PJIs have not been established and we included studies according to different reference standards, which may result in misdiagnosis for PJIs (Liu et al., 2018). Thirdly, an antibiotic-free interval before sampling may enhance the ability to detect the causative organism, but through our univariable meta-regression and subgroup analysis, we still could not conclude that antibiotics were the main source of heterogeneity. Finally, studies with positive results are more likely to be published, which can amplify the overall diagnostic accuracy.
CONCLUSIONS
To the best of our knowledge, our study is the first meta-analysis that evaluates the clinical usability of mNGS in the diagnosis of PJIs. Our study indicated that mNGS has a superior diagnostic accuracy for PJIs and may be particularly useful for culture-negative cases. This systematic review provides effective support for the diagnostic performance of mNGS, which can provide clinicians with recommendations for accurate and effective diagnosis of PJIs and antibiotics treatment. Meanwhile, large-sized and good-quality studies should be conducted to verify our results and to confirm the clinical value of mNGS in PJI patients.
AUTHOR CONTRIBUTIONS
JX and TC were responsible for the idea and concept of the paper. JT and TC built the database. JT and YY analyzed the data. JT wrote the manuscript. YL, SE, and AN critically reviewed and revised the manuscript. All authors contributed to the article and approved the submitted version.
SUPPLEMENTARY MATERIAL
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcimb.2022.875822/full#supplementary-material
REFERENCES
Bozic, K. J., Kurtz, S. M., Lau, E., Ong, K., Chiu, V., Vail, T. P., et al. (2010). The Epidemiology of Revision Total Knee Arthroplasty in the United States. Clin. Orthop. Relat. Res. 468 (1), 45–51. doi: 10.1007/s11999-009-0945-0
Cai, Y., Fang, X., Chen, Y., Huang, Z., Zhang, C., Li, W., et al. (2020). Metagenomic Next Generation Sequencing Improves Diagnosis of Prosthetic Joint Infection by Detecting the Presence of Bacteria in Periprosthetic Tissues. Int. J. Infect. Dis. 96, 573–578. doi: 10.1016/j.ijid.2020.05.125
Chen, X., Cao, K., Wei, Y., Qian, Y., Liang, J., Dong, D., et al. (2020). Metagenomic Next-Generation Sequencing in the Diagnosis of Severe Pneumonia Caused by Chlamydia Pneumonia. Infection 48 (4), 535–542. doi: 10.1007/s13311-020-01429-0
Dabrowski, P., Jurskiewicz, J., Czernicki, Z., Koszewska, W., and Jasielski, P. (2017). Polymerase Chain Reaction Based Detection of Bacterial 16S rRNA Gene in the Cerebrospinal Fluid in the Diagnosis of Bacterial Central Nervous System Infection in the Course of External Cerebrospinal Fluid Drainage. Comparison
SUPPLEMENTARY MATERIAL
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcimb.2022.875822/full#supplementary-material
TABLE 3 | Subgroup analysis of mNGS.
| Subgroup | Number of studies | Pooled sensitivity (95% CI) | Pooled specificity (95% CI) | P | I² |
|----------|------------------|-----------------------------|-----------------------------|---|---|
| Ethnicity | | | | | |
| Asians | 8 | 0.94 (0.90-0.96) | 0.96 (0.89-0.97) | 0.001/0.948 | 72%/0.0% |
| Caucasians | 2 | 0.77 (0.73-0.81) | 0.96 (0.92-0.98) | 0.008/0.341 | 85.9%/0.0% |
| Sample type | | | | | |
| Sonication fluid | 2 | 0.816 (0.770-0.856) | 0.963 (0.926-0.983) | 0.001/0.506 | 90.2%/0.0% |
| Synovial fluid | 3 | 0.709 (0.629-0.781) | 0.932 (0.857-0.975) | 0.001/0.976 | 86.8%/0.0% |
| Multiple samples | 5 | 0.961 (0.916-0.985) | 0.943 (0.871-0.981) | 0.778/0.743 | 0.0%/0.0% |
| Reference standard | | | | | |
| IDSA | 3 | 0.787 (0.747-0.823) | 0.956 (0.925-0.977) | 0.000/0.617 | 88.5%/0.0% |
| MSIS | 7 | 0.930 (0.886-0.961) | 0.939 (0.879-0.975) | 0.251/0.902 | 24.0%/0.0% |
| Study design | | | | | |
| Prospective | 8 | 0.832 (0.800-0.860) | 0.953 (0.926-0.972) | 0.000/0.859 | 86.9%/0.0% |
| Retrospective | 2 | 0.805 (0.651-0.912) | 0.882 (0.636-0.985) | 0.000/0.138 | 92.0%/54.4% |
CI, confidence interval; MSIS, Musculoskeletal Infection Society; IDSA, Infectious Disease Society of America guidelines.
| 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2269,
1
],
[
2269,
8582,
2
],
[
8582,
14986,
3
],
[
14986,
16595,
4
],
[
16595,
23445,
5
],
[
23445,
24758,
6
],
[
24758,
26107,
7
],
[
26107,
29314,
8
],
[
29314,
34105,
9
],
[
34105,
34105,
10
]
]
} |
725a515acc53ac650c82306dd47606c185484467 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 7,
"total-fallback-pages": 0,
"total-input-tokens": 38338,
"total-output-tokens": 8373,
"fieldofstudy": [
"Medicine"
]
} | CASE REPORT
Fooled by the fragments: vitamin B12 deficiency masquerading as thrombotic thrombocytopenic purpura
Pegah Jahangiria, Rachel Hicksb, Prabjot K. Bathb and Christopher J. Haasc,d
aMedStar Health Internal Medicine Residency Program, Medstar Health, Baltimore, MD, USA; bDepartment of Medicine, Saba University School of Medicine, Dutch Caribbean; cMedStar Health Internal Medicine Program, MedStar Health, Baltimore, MD, USA; dDepartment of Medicine, Georgetown University School of Medicine, Washington, D.C., USA
ABSTRACT
Background: Thrombotic thrombocytopenic purpura (TTP) is a hematological emergency requiring prompt plasmapheresis. Conversely, vitamin B12 deficiency is a relatively benign diagnosis that can mimic microangiopathic hemolytic anemia, characterized by the presence of anemia, thrombocytopenia, indirect hyperbilirubinemia, markers of hemolysis, and schistocytes. This case series highlights the association of vitamin B12 deficiency and its TTP-like presentations.
Cases: The first case describes a 72-year-old man with shortness of breath and weakness. Diagnostics were notable for pancytopenia, schistocytes, and a low reticulocyte index. Intriguingly, total bilirubin was only mildly elevated however LDH and Haptoglobin were elevated and low, respectively. Additional diagnostic workup demonstrated an undetectable B12, elevated methylmalonic acid and elevated homocysteine. Initiation of B12 supplementation resolved his pancytopenia.
The second case describes a 57-year-old man with chest tightness, dyspnea on exertion, and night sweats. Diagnostic evaluation demonstrated pancytopenia, schistocytes, a low reticulocyte index, and a remarkably low B12. He had associated high methylmalonic acid and homocysteine levels, confirming the diagnosis. B12 supplementation resolved his pancytopenia.
Conclusion: The polysymptomatic presentation of vitamin B12 deficiency-induced pseudo-thrombotic microangiopathy highlights the vitamin’s role in essential physiological cellular functions. Rapid recognition of the underlying etiology of microangiopathic hemolytic anemia is necessary as treatment approaches diverge greatly.
1. Background
Thrombotic thrombocytopenic purpura (TTP) is a microangiopathic hemolytic anemia caused by deficiency of ADAMTS13, a disintegrin and von Willebrand factor metalloproteinase with a thrombospondin type 1 motif, member 13. Whether the deficiency is due to congenital, genetic mutations or acquired through the development of auto-antibodies, the resultant enzyme insufficiency impedes the breakdown of von Willebrand factor (vWF) multimers and causes progressive multimer accumulation on endothelial surfaces. This incites intravascular platelet adhesion, microthrombi formation, and subsequent erythrocyte shearing and fragmentation. Together, this pathophysiology is responsible for the classic clinical pentad of fever, thrombocytopenia, microangiopathic hemolytic anemia, and end-organ damage (i.e., renal failure, neurological manifestations). Management requires prompt initiation of plasmapheresis, designed to provide normal ADAMTS13 or remove ADAMTS13 inhibitory autoantibodies in cases of congenital or acquired TTP, respectively.
Vitamin B12 (cobalamin) deficiency, in contrast, is a relatively benign nutritional deficiency that may result in a macrocytic, megaloblastic anemia, pancytopenia, and neurological sequelae such as subacute combined degeneration of the myelinated dorsal columns of the spinal cord. The etiology may be related to poor nutritional intake or poor absorption in the setting of a variety of causes – alcohol abuse, atrophic gastritis, pernicious anemia, and inflammatory bowel disease. Laboratory diagnostics demonstrate elevated levels of methylmalonic acid (MMA) and homocysteine, precursors that accumulate secondary to reduced function of cobalamin-requiring enzymes, methylmalonyl coenzyme A mutase and methionine synthase, respectively. In rare cases, cobalamin deficiency may mimic a microangiopathic hemolytic anemia such as TTP, with laboratory diagnostics demonstrating anemia,
thrombocytopenia, elevated lactate dehydrogenase (LDH), and low haptoglobin, with a critically distinguishing feature, reticulocyte hypoproliferation. In the case of cobalamin deficiency, macrocytic, immature, rigid red blood cells undergo intramedullary hemolysis resulting in a hemolytic anemia. Additionally, there is a hypothesized role for elevated homocysteine levels causing intravascular hemolysis due to activation of the clotting cascade and associated endothelial cell dysfunction with resultant clot formation. This case series will highlight the association of vitamin B12 deficiency, nuclear-cytoplasmic desynchrony, intramedullary hemolysis, thrombocytopenia, and venous thrombosis, with specific focus on thombotic thrombocytopenic purpura (TTP)-like presentations.
2. Case report 1
A 72-year-old gentleman with a history of hypertension, hyperlipidemia, uncontrolled diabetes mellitus, prior deep venous thrombosis/pulmonary embolism, alcohol use disorder, and benign prostatic hyperplasia presented to the emergency department with shortness of breath and generalized fatigue. He reported usual health until 1 week prior to presentation at which time he noted exertional shortness of breath with minimal ambulation as well as a productive cough, nasal congestion, and sore throat, without chest discomfort, edema, fever, or chills. He furthermore denied hematochezia, melena, easy bruising, paresthesia, or gait disturbance, but did report chronic alcohol use, approximately 1–2 beers every other day, and taking a baby aspirin daily.
Examination revealed a fatigued-appearing African-American male with conjunctival pallor who was afebrile, mildly tachycardic (100–110 beats per minute), and hypertensive (SBP 160–180 mmHg), with a preserved ambient-air oxygen saturation. Cardiopulmonary examination was notable only for a flow murmur best appreciated at the upper sternal border with delayed capillary refill. There was no evidence of hepatosplenomegaly, peripheral edema, nor rash. Neurological examination was unremarkable with preserved strength, gait, proprioception, sensation (fine touch and pain), and reflexes. Diagnostic evaluation revealed pancytopenia (white blood cell count of 3.6 k/μL, a hemoglobin of 4.6 g/dL, and a platelet count of 74 k/μL) (Table 1). His marked anemia was notably macrocytic (MCV 128 fl) with an elevated distribution width (RDW 19.5% [normal range, 11.5 to 15.5]) and reticulocyte hypoproliferation (1.8%; Reticulocyte Index 0.23). Peripheral blood smear demonstrated no atypical lymphocytes, nor any evidence of abnormal platelets; however, red blood cell morphology included ovalocytes, target cells, moderate schistocytes, and teardrop cells. Basophilic stippling and hyper-segmented neutrophils were also seen (Figure 1). Metabolic panel demonstrated a mildly elevated AST (65 units/L) with a normal ALT, a mildly elevated total and indirect bilirubin (1.2 mg/dL [0.1–1.0 mg/dL] and 0.86 mg/dL, respectively), and a normal alkaline phosphatase (76 units/L). Lactate dehydrogenase was markedly
| Table 1. Significant values on pre-hospitalization, presentation, hospital course and follow-up after vitamin B12 repletion for Case 1. |
|---------------------------------------------------------------|
| **Pre-hospitalization (2 years prior)** | **Presentation** | **Hospital course** | **First follow-up in 1 month** | **Second follow-up in 6 months** |
|--------------------------------------|-----------------|-------------------|-----------------------------|---------------------------------|
| WBC (k/μL) | 4.8 | 3 | 4 | 4.9 | Not performed |
| Hgb (gm/dL) | 9.9 | 4.7 | 7.6 | 13.7 | Not performed |
| Hct (%) | 27.5 | 13.2 | 21.9 | 40.3 | Not performed |
| MCV (FL) | 128.5 | 128.2 | 107.9 | 96 | Not performed |
| Platelets (k/μL) | 152 | 73 | 59 | 134 | Not performed |
| RBC count (million/μL) | 2.14 | 1.03 | 2.03 | 4.2 | Not performed |
| % of Retic | 1.8 | | 4.9 | | |
| Retic Index | 0.23 | | 2.6 | | |
| LDH (units/L) [87–241 units/L] | 3034 | | 2079 | | |
| Haptoglobin (mg/dL) | <8 | <8 | | | |
| B12 level (pg/mL) [0.3–1.5 pg/mL] | 65 | <60 | >6000 | 190 | |
| MMA (umol/L) [0.0–4.0 umol/L] | 8.81 | | | | |
| Homocysteine (umol/L) [3.2–10.7 umol/L] | | | | | |
elevated (3000 units/L) with a low serum haptoglobin (<8 mg/dL). Prothrombin time, partial thromboplastin time, INR, and D-Dimer were mildly elevated (20.3 sec, 40.2 sec, 1.7, and 1.15 mcg/mL, respectively), with a normal fibrinogen. There was no evidence of a gamma gap. Diagnostic imaging included a CT of the chest, abdomen, and pelvis that showed no evidence of pulmonary embolism, nor organomegaly/adenopathy, but the presence of a shrunken, nodular appearing liver and left lingular pneumonia.
In the setting of the noted anemia, schistocytes on peripheral smear, thrombocytopenia, hyperbilirubinemia, markedly elevated LDH, low haptoglobin, and coagulopathy, there was concern that the patient’s presentation could represent a microangiopathic hemolytic anemia, specifically thrombotic thrombocytopenic purpura; however, the lack of concomitant renal disease, neurological symptoms, skin lesions, reticulocyte hyperproliferation, and the presence of marked macrocytosis seemed to make this etiology less likely. Longitudinal review of his laboratory diagnostics demonstrated a chronic macrocytosis (MCV 128 fL) with borderline anemia (10–12 g/dL) and borderline thrombocytopenia (150 k/µL) (Table 1). Furthermore, the patient’s vitamin B12 was noted to be low (65 ng/mL) 2 years prior to presentation, but he was not initiated on B12 supplementation at that time. On the current presentation, his vitamin B12, MMA, and homocysteine were subsequently checked and confirmed the diagnosis of marked vitamin B12 deficiency (B12 < 60 pg/mL, MMA 8.81 umol/L, homocysteine 33.0 umol/L). Intrinsic factor antibody, anti-parietal cell antibody, and H. pylori testing were notably negative, as were additional viral etiologies.
The patient initially received three units of packed red blood cells for symptomatic anemia and was initiated on intramuscular 1000 mcg vitamin B12 injection as well as folate and iron pills in anticipation of folate and iron depletion upon resumption of erythropoiesis. At the time of discharge, the patient’s liver function tests had completely resolved (AST 21 units/L, ALT 24 units/L), and he was discharged on a high-dose vitamin B12 supplementation (1000 mcg IM weekly for 4 weeks) and instructed to follow-up with outpatient hematology 1 month later. At this follow-up visit, the patient’s laboratory data revealed significant improvement in the vitamin B12 level (above 6000 pg/mL) with resolution of his pancytopenia (white blood cell count of 4.9 k/µL, hemoglobin of 13.7 gm/dL [MCV 107.3 fl] and platelet count of 134 k/µL). At that point, he was instructed to continue vitamin B12 shots once a month with serial vitamin B12 levels and CBC monitoring. The plan was to transition back to oral vitamin B12 supplementation, but unfortunately the patient stopped vitamin B12 supplementation. Six months after the first presentation he was once again noted to have vitamin B12 deficiency (190 pg/mL [normal range 200–900 pg/mL]), with associated return of adjunctive symptoms including shortness of breath and dyspnea on exertion. CBC nor MMA or homocysteine were collected.
**3. Case report 2**
A 57-year-old African-American gentleman with a history of obesity and prostate cancer status-post prostatectomy presented with a one-week history of chest tightness, dyspnea on exertion, and night sweats. He reported usual health until 1 month prior at which time he noted a decreasing exercise capacity, attributing his symptoms to work-related...
stress as a university professor. His symptoms culminated in the week prior to presentation as an inability to complete his standard 35-minute treadmill regimen and marked dyspnea on climbing a flight of steps. Review of systems was positive for intermittent night sweats of six months duration and notably negative for palpitations, hemoptysis, hematochezia/melena, adenopathy, and easy bruising. His family history was remarkable for acute lymphocytic leukemia and chronic myeloid leukemia in second-degree relatives.
Physical examination revealed a well-appearing African-American male who was hemodynamically stable. Sclera were anicteric and no evidence of adenopathy (occipital, cervical, axillary, or inguinal) was appreciated. Cardiopulmonary as well as abdominal examination was unremarkable. Diagnostic evaluation demonstrated pancytopenia (white blood cell count of 3.9k/μL, a normocytic anemia with a hemoglobin of 6.8 gm/dL [MCV 96.1 fL], and a platelet count of 122 k/μL) (Table 2). Reticulocyte count, reticulocyte index and immature platelet fraction were 0.4%, 0.07 and 4.3%, respectively, indicative of ineffective erythropoiesis. Smaer demonstrated moderate anisocytes, occasional teardrop cells and schistocytes, and a few enlarged platelets. Metabolic panel demonstrated a normal creatinine, an elevated AST (363 units/L) and ALT (121 units/L), a normal alkaline phosphatase, a normal total bilirubin, and normal gamma gap (3.2 gm/dL). Additional diagnostic workup for evaluation of the noted pancytopenia and transaminitis revealed negative serologies for HIV, Hepatitis A/B/C, but a history of EBV infection (EBV VCA IgG >750.0 u/mL, EBV nuclear antigen IgG 518 u/mL, EBV IgG/early antigen 34.1 u/mL, negative IgM). Serum protein electrophoresis, serum immunoglobulins, and kappa/lambda light chains were normal. ESR and CRP were notably elevated at 51 mm/hr and 6.49 mg/L, respectively. D-Dimer was notably elevated to 9.83 mcg/mL, with a PT 15.8 sec (normal 11–15 sec), PTT 36.6 sec (normal 25–40 sec), and INR 1.3. LDH was markedly elevated to >4000 units/L with an undetectable haptoglobin (<8 mg/dL), seemingly concerning for hemolysis, in light of the noted anemia. Diagnostic imaging including chest radiograph and CT of the chest, abdomen, and pelvis remained unremarkable. In the setting of the noted anemia, schistocytes, thrombocytopenia, low haptoglobin, elevated LDH, and elevated D-Dimer, there was minor concern that this could represent an atypical presentation of microangiopathic hemolytic anemia. Given the atypical presentation, iron studies and vitamin B12 levels were sent which demonstrated an iron level 167 μg/dL (normal 60–170 μg/dL), TIBC 187 μg/dL (normal 240–450 μg/dL), iron saturation 89% (normal 20–50%), consistent with anemia of chronic disease, and a remarkably low vitamin B12 (64 pg/mL), likely contributing to the normal MCV. He had an associated high methylmalonic acid (16.87 umol/L, normal 0.0–0.4 umol/L) and homocysteine (140.0 umol/L), confirming the diagnosis. Intrinsic factor antibodies were negative and anti-parietal cell antibodies were within normal limits (12.2 units, normal <24.9 units).
The patient initially received a transfusion of packed red blood cells for his symptomatic anemia and was initiated on high-dose parenteral cobalamin as well as folate and iron in anticipation of depletion following erythropoiesis. At the time of discharge, liver function tests had begun to resolve (AST 113 units/L, ALT 133 units/L), and he was discharged on high-dose vitamin B12 supplementation regimen (1000 mcg IM weekly for 4 weeks) and was instructed to follow-up with outpatient hematology. Follow-up laboratory diagnostics one-week after discharge showed resolution of the pancytopenia (white blood cell count of 5.1 k/μL, hemoglobin of 9.3 gm/dL [MCV 99.0 fL], and platelet count of 221 k/μL). He was instructed to continue taking weekly vitamin B12 shots for a total of four weeks followed by monthly treatment. Five months after his first follow-up, his CBC continued to demonstrate complete resolution of pancytopenia (WBC count of 5.4 k/μL, hemoglobin of 13.1 gm/dL, hematocrit of 41.2% (MCV 86.9 fL), and a platelet count of 205 k/μL).
Table 2. Significant values on presentation, hospital course and follow-up after vitamin B12 repletion for Case 2.
| Presentation | Hospital course | First follow-up in 1 week | Second follow-up in 5 months |
|--------------|----------------|--------------------------|-----------------------------|
| WBC count (k/μL) | 3.9 | 4.1 | 5.1 | 5.4 |
| Hgb (gm/dL) | 6.8 | 7.8 | 9.3 | 13.1 |
| Hct (%) | 19.5 | 23.1 | 29.3 | 41.2 |
| MCV(fL) | 96.1 | 97.9 | 99 | 88.9 |
| Platelets (k/μL) | 122 | 77 | 221 | 205 |
| RBC count (million/μL) | 1.8 | 2.36 | 4.2 | 4.74 |
| % of Retic | 0.4 | 0.07 | 4.9 | 2.96 |
| Retic index (units/μL) | >4000 | | | |
| LDH (units/μL) | | | | |
| Haptoglobin (mg/dL) | | | | |
| B12 level (μg/mL) | | | | |
| MAA (umol/L) | 16.87 | | | |
| Homocysteine (umol/L) | 140 | | | |
4. Discussion
Vitamin B12 is a water-soluble vitamin with an essential role in DNA synthesis, hematopoiesis, and myelination. Vitamin B12 is a necessary cofactor for the conversion of methylmalonyl coenzyme A to succinyl coenzyme A via methylmalonyl coenzyme A mutase (Figure 2), enabling the breakdown of odd-chained fatty acids and some amino acids. It is also a required cofactor for methionine synthase, which catalyzes the conversion of homocysteine to methionine, ultimately allowing for the generation of tetrahydrofolate, a biologically active form of folate needed for DNA synthesis (Figure 2). Insufficient levels of vitamin B12 lead to reduced function of methylmalonyl coenzyme A mutase and methionine synthase, with a resultant accumulation of the precursors methylmalonyl coenzyme A and homocysteine, respectively. Classically, vitamin B12 deficiency manifests as a macrocytic, megaloblastic anemia and in severe cases, subacute combined degeneration of the posterior columns of the spinal cord. In rare cases, severe vitamin B12 deficiency can present as a pseudo-microangiopathic hemolytic anemia/TTP, characterized by thrombocytopenia and hemolytic anemia (i.e., elevated LDH, low haptoglobin, hyperbilirubinemia, and schistocytes). In contrast to classic microangiopathic hemolytic anemia/TTP, vitamin B12-related pseudo-TTP presents as a macrocytic, megaloblastic anemia with reticulocyte hypoproliferation, elevated levels of homocysteine and MMA, and a low vitamin B12. Management of this masked deficiency requires only vitamin B12 supplementation in contrast to initiation of plasmapheresis required for true TTP.
4.1. Megaloblastic anemia and hypersegmented neutrophils
As previously noted, vitamin B12 is critically involved in DNA synthesis, catalyzing the conversion of homocysteine to methionine, while at the same time acting as a methyl-group acceptor to allow for the regeneration of tetrahydrofolate, a biologically active form of folate required for DNA synthesis. Defective DNA synthesis leads to nucleo-cytoplasmic asynchrony [1] wherein the cell is unable to progress from the G2 growth stage to mitosis, leading to progressive cell growth without division. On blood smear, this presents as erythrocytic megaloblastosis (i.e., large, dysfunctional red blood cells), and neutrophil hypersegmentation (greater than five nuclear lobes) [2]. With chronic deficiency, further dyspoiesis may occur, with resultant leukopenia and thrombocytopenia [3]. Indeed, our patients’ megaloblastic, macrocytic anemia, leukopenia, and thrombocytopenia reflect the cumulative effects of ongoing vitamin B12 deficiency. While thrombocytopenia and anemia are both classic findings in TTP, leukopenia is less commonly described. One case report highlighted leukopenia as a result of bone marrow necrosis in TTP, though blood smear revealed normal leukocyte morphology [4], again differentiating true TTP from vitamin B12 deficiency.
4.2. Hemolysis
Both cases showed evidence of hemolytic anemia, with characteristically elevated LDH and low haptoglobin. The hemolytic anemia seen in vitamin B12 deficiency may be due to a combination of extramedullary and
Figure 2. Vitamin B12 (Cobalamin) in cellular metabolism.
intramedullary hemolysis. The hypercoagulability and endothelial dysfunction associated with hyperhomocysteinemia may lead to erythrocyte fragmentation comparable to that observed in microangiopathic hemolytic anemia/TTP [5,6]. In the cases described, however, bilirubin levels were normal to minimally elevated, suggesting an intramedullary process deviating from that of true TTP.
Intramedullary hemolysis secondary to paurfolate deficiency and subsequently impaired DNA synthesis and cellular division manifests in one of two mechanisms. One mechanism may be that immune reticulocytes arrest in development and are destroyed within the marrow [7,8]. An alternative mechanism is impaired marrow egression. Normally, RBCs leave the marrow by squeezing through sinu- soids; however, a larger cell, as occurs with vitamin B12-associated nucleo-cytoplasmic asynchrony, would impede this process with resultant microcirculation entrapment and intramedullary cell lysis.
Both intra- and extra-medullary hemolysis are compounded by the increased erythrocyte membrane rigid- ity observed in vitamin B12 deficiency [9]. While still under investigation, altered TCA and methionine cycles are thought to be involved [9,10]. Free radicals accumulate following inhibition of the mitochondrial respiratory chain due to reduced methylenalynal CoA mutase activity. These free radicals damage the cellular phospholipid bilayer, particularly polyunsaturated fatty acids, a process that is further exacerbated by the expenditure of glutathione in free radical reduction: glutathionylcobalamin functions as a precursor to both physiologic coenzyme vitamin B12 forms, perpetuating the deficiency [10]. Therefore, diversion of glutathione to reducing free radicals further reduces vitamin B12 availability. Impairment of the TCA cycle may also divert acetyl CoA, the final product of fatty acid catabolism, to cholesterol synthesis thereby increasing membrane cholesterol concentrations. Methylcobalamin, through its interaction with methio- nine synthase, is a regulator of TCA and methionine reactions including phospholipid methylation; a deficiency of this coenzyme would result in S-adenosylmethionine accumu- lation, an intermediate with a feedforward inhibitory role in methylation reactions. The combined impaired phospholipid production and increased cholesterol synthesis may additively decrease membrane fluidity, contributing to hemolysis [9]. This multifactorial pathophysiology starkly contrasts the hemolysis of TTP, which is a direct consequence of erythrocytes shearing against platelet clots secondary to thrombotic microangiopathy.
4.3. Thrombotic microangiopathy
Like that of hemolysis, the mechanism of thrombotic microangiopathy in vitamin B12 deficiency is multifactorial, but most closely tied to hyperhomocysteinemia. This further distinguishes B12 deficiency from TTP, in which thrombotic microangiopathy occurs due to vWF accumulation and subsequent platelet aggregation. Homocysteine elevation is an independent risk factor for atherosclerotic disease, causing hypercoagulability through endothelial damage, coagulation activation, and impaired nitric oxide response [11]. Hyperhomocysteinemia is associated with increased oxidative stress via the formation of superoxide and hydrogen peroxide radicals; those with nutritional deficiencies, as likely seen in the first patient case, will lack the antioxidant vitamins C and E required to reduce these radicals. Radicals cause damage directly to endothelium, facilitating the formation of atherosclerosis and a platelet plug. They further contribute to coagulation by inactivating nitric oxide. Homocysteine also impedes the anti-aggregating properties of nitric oxide indirectly, by reducing tissues’ responsiveness to l-arginine, a precursor to nitric oxide [5]. These mechanisms were best appreciable in our first patient’s history of thrombosis in the setting of chronically depleted vitamin B12 levels.
4.4. Neurologic manifestations
Classically, vitamin B12 deficiency is associated with subacute combined degeneration, a condition char-acterized by damage to the dorsal columns of the spinal cord with resultant loss of vibration sense/ proprioception, spastic paresis, and gait abnor- malities. Mechanistically, this occurs due to lack of conversion of the three-carbon methylmalonyl-CoA to four-carbon succinyl-CoA, causing a subsequent accumulation of three-carbon propionate. Some suggest this results in the formation and incorporation of 15- and 17-carbon fatty acids into neuronal lipids, predisposing to myelin and neuronal breakdown [12]. Notably, individuals with inherited deficits in methylmalonyl coenzyme-A mutase have highly variable neurological phenotypes without evidence of predisposition to SCD [13], purporting an alternative role of cobalamin in this symptomatology.
Although SCD is highly associated with B12 defi- ciency, non-specific neurologic findings may also occur, including altered mental status and delirium. This presentation is comparable to that seen in TTP, wherein patients present with altered mental status, stroke, and seizures as a result of thrombosis of cerebral vasculature and focal vasconstriction due to reversible cerebrovascular constriction syndrome [14]. Neurologic issues were not observed in either case presented, with both patients demonstrating a benign neurologic exam. Few case studies discussing pseudo-TTP in the setting of vitamin B12 deficiency mention concomitant neurologic issues,
including generalized weakness (potentially attributable to anemia), altered mental status [15-18] and paresthesia [11,17]. These non-specific findings may be attributable to a combination of hypercoagulability and impaired myelin synthesis. We were unable to find any cases with frank descriptions of subacute combined degeneration with simultaneous pseudo-TTP, drawing into question individual factors influencing clinical presentation in the setting of vitamin B12 deficiency.
5. Conclusion
Vitamin B12 deficiency-induced pseudothrombotic microangiopathy is a rare condition that resembles the clinical features of TTP. This polysymptomatic presentation highlights the vitamin’s role in essential physiological cellular functions. Rapid recognition of the underlying etiology of microangiopathic hemolytic anemia is necessary as treatment approaches diverge greatly.
Disclosure statement
No potential conflict of interest was reported by the authors.
References
[1] Stabler SP. Vitamin B12 deficiency. N Engl J Med. 2013;368(2):149–160.
[2] Farrelly SJ, O’Connor KA. Hypersegmented neutrophils and oval macrocytes in the setting of B12 deficiency and pancytopenia. Case Rep. 2017; 2017.
[3] Bhattacharjee A, Samuel AE. Vitamin B12 deficiency in a patient presenting with dyspnea: a case report. Adv J Emergency Med. 2019;3(2):2.
[4] Parekh HD, Reese JA, Cobb PW, et al. Bone marrow necrosis discovered in a patient with suspected thrombotic thrombocytopenic purpura. Am J Hematol. 2015;90(3):264.
[5] Nappo F, De Rosa N, Marfella R, et al. Impairment of endothelial functions by acute hyperhomocysteinemia and reversal by antioxidant vitamins. Jama. 1999;281 (22):2113–2118.
[6] Kollipara VK, Brine PL, Gemmel D, et al. A case of asymptomatic pancytopenia with clinical features of hemolysis as a presentation of pernicious anemia. J Community Hosp Intern Med Perspect. 2016;6(4):32493.
[7] Bailey M, Maestas T, Betancourt R, et al. Cause of thrombotic thrombocytopenia purpura-(TTP-) like syndrome, vitamin B12 deficiency: interpretation of significant pathological findings. Case Rep Hematol. 2019;2019. DOI:10.1155/2019/1529306
[8] Tran PN, Tran M-H. Cobalamin deficiency presenting with thrombotic microangiopathy (TMA) features: a systematic review. Transfus Apheresis Sci. 2018;57(1):102–106.
[9] Özcan Ö, Ipcioglu OM, Gültepe M, et al. Altered red cell membrane compositions related to functional vitamin B12 deficiency manifested by elevated urine methylmalonic acid concentrations in patients with schizophrenia. Ann Clin Biochem. 2008;45(1):44–49.
[10] Pezacka E, Green R, Jacobsen DW. Glutathionyckobalamin as an intermediate in the formation of cobalamin coenzymes. Biochem Biophys Res Commun. 1990 Jun 15;169(2):443–450.
[11] Vanoli J, Carrer A, Martorana R, et al. Vitamin B 12 deficiency-induced pseudothrombotic microangiopathy without macrocytosis presenting with acute renal failure: a case report. J Med Case Rep. 2018;12(1):1–5.
[12] Metz J. Cobalamin deficiency and the pathogenesis of nervous system disease. Annu Rev Nutr. 1992 Jul;12 (1):59–79.
[13] Shevell MI, Matiaszuk N, Ledley FD, et al. Varying neurological phenotypes among mut° and mut− patients with methylmalonylCoA mutase deficiency. Am J Med Genet A. 1993;45(5):619–624.
[14] Xiao Z, Deng L, Steinberg L. Focal neurologic manifestation as initial presentation of thrombotic thrombocytopenic purpura. Ann Clin Case Rep. 2019;4:1725.
[15] Franks AM, Bannister T, Jarrell A, et al. Cobalamin deficient thrombotic microangiopathy: a case of TTP or pseudo-TTP. West Virginia Med J OA. 2018 Apr;26:3578.
[16] Kandel S, Budhathoki N, Pandey S, et al. Pseudo-thrombotic thrombocytopenic purpura presenting as multi-organ dysfunction syndrome: A rare complication of pernicious anemia. SAGE Open Med Case Rep. 2017 Jun;3(5):2050313X17713149.
[17] Walter K, Vaughn J, Martin D. Therapeutic dilemma in the management of a patient with the clinical picture of TTP and severe B 12 deficiency. BMC Hematol. 2015 Dec 1;15(1):16.
[18] Tadakamalla AK, Talluri SK, Besur S. Pseudo-thrombotic thrombocytopenic purpura: a rare presentation of pernicious anemia. N Am J Med Sci. 2011 Oct;3(10):472. | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4105,
1
],
[
4105,
9939,
2
],
[
9939,
13420,
3
],
[
13420,
18442,
4
],
[
18442,
21676,
5
],
[
21676,
27193,
6
],
[
27193,
31365,
7
]
]
} |
e245f9ad2080acd3994432478498288b3a43a68d | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 21,
"total-fallback-pages": 0,
"total-input-tokens": 69297,
"total-output-tokens": 22912,
"fieldofstudy": [
"Environmental Science"
]
} | High Species Richness and Extremely Low Abundance of Cumacean Communities Along the Shelf and Slope of the Gulf of Guinea (West Africa)
Anna Stepien*, Krzysztof Pabis¹, Robert Sobczyk¹ and Bjorn Serigstad²
¹ Department of Invertebrate Zoology and Hydrobiology, University of Łódź, Łódź, Poland, ² Center for Development Cooperation in Fisheries, Institute of Marine Research, Bergen, Norway
The Gulf of Guinea belongs to the most scarcely sampled marine basins in the oceans of the world. We have analyzed diversity and distribution patterns of cumacean communities on the shelf and slope, along the coast of Ghana. The material was collected in October and November of 2012 using a van Veen grab (0.1 m²) on nine transects. Six stations were located at each transect (25, 50, 100, 250, 500, and 1,000 m). Sixty-three species of Cumacea were recorded with Leucon and Eocuma as the most speciose genera, with 12 and eight species, respectively. Comparisons of species richness with literature data pointed that the Ghanaian coast hosts very diverse communities. About 95% of species were new to science, and the number of cumacean species known from the West Africa increased by over 100%. Nevertheless, most of the species had extremely low abundance, 13 singletons and 15 doubletons were found. Mean density of cumaceans was estimated at only 1.5 ind./0.1 m². Species accumulation curve did not reach the asymptotic level, suggesting undersampling, despite the fact that sampling effort was high (250 samples). The highest species richness was recorded in the inner shelf (25–50 m) and on the slope (1,000 m). Cluster analysis separated shallow water communities from deeper regions on the shelf and upper slope. The most unique species composition was found at 1,000 m. Principal component analysis showed the importance of oxygen, sediments, and human-related disturbance for distribution of cumacean communities. In the shallows, oxygen content and presence of gravel were the most important factors structuring communities. In the deeper bottom areas (250–1,000 m), cumacean fauna was affected by local pollution, mainly by higher concentration of barium, other heavy metals, and THC.
Keywords: Cumacea, depth gradient, diversity, pollution, Gulf of Guinea
INTRODUCTION
Continental margins constitute about 11% of the oceans of the world and are shaped by a complex set of environmental factors that are dynamically changing along a depth gradient (Levin and Sibuet, 2012). They are characterized by high habitat heterogeneity and belong to the most important marine biodiversity hot spots (Danovaro et al., 2009; Menot et al., 2010; Levin and Sibuet, 2012). At the same time, continental margins belong to areas of special economic interests,
such as fishery and oil industry (Menot et al., 2010). This makes them one of the most interesting natural laboratories for studies of biodiversity, ecosystem services, and environmental gradients as well as influence of human activities and climate change on marine biota (Levin and Sibuet, 2012; Birchenough et al., 2015). On the other hand, recent analysis based on over 10 million records obtained mostly from the Ocean Biogeographic Information System (OBIS), revealed a strong sampling bias in the marine biodiversity assessment. There is a large gap in the knowledge about marine fauna associated with tropical areas, and it is visible not only in the deep sea (bathyal and abyssal) but also on the shelf, with average number of sampling events an order of magnitude lower than in northern and southern mid latitudes (Menegotto and Rangel, 2018). The authors of this research pointed out a lack of scientific infrastructure and funding for marine research in developing tropical countries as the main reason of this situation. It is highly visible in the case of African marine fauna.
The West African continental margin belongs to the most scarcely sampled regions. Most of the available studies were focused on the shallows and based on low sampling effort (Buchanan, 1957; Longhurst, 1958, 1959; Bassindale, 1961; Le Loeff and Intês, 1999; Bamikole et al., 2009). There is a particular lack of ecological research based on quantitative sampling and lack of detailed biodiversity inventories based on species level identification. The deep sea communities of the Gulf of Guinea are almost completely neglected in earlier research, with the exception of the areas affected by organic discharge from the Congo River (e.g., Gaever et al., 2009; Galéron et al., 2009; Menot et al., 2009) and most recent studies from Gabon (Friedlander et al., 2014) and Ghana (Pabis et al., 2020; Sobczyk et al., 2021). At the same time, the Gulf of Guinea is facing serious problems associated with various types of anthropogenic impacts, such as pollution events associated with the oil industry (Scheren et al., 2002; Ayamdoo, 2016), but those problems are only scarcely studied and need further research based on the analysis of various taxonomic groups (Pabis et al., 2020; Sobczyk et al., 2021). Influence of heavy metals, hydrocarbons, and other pollutants might have a substantial influence on the composition, diversity, and abundance of benthic communities (Olsgard and Gray, 1995; Gomez-Gesteira et al., 2003; Stark et al., 2020). However, there are no studies demonstrating the influence of anthropogenic disturbance on Cumacea.
Based on literature data, Le Loeff and Cosel (1998) listed only 1,440 benthic species from the large part of the West African coast, starting from the Mauretanies and ending in the Namibia (up to 200 m depth), although the study was focused mostly on megafauna (corals, echinoderms, and decapod crustaceans) as well as polychaetes and bivalves. Analysis of the same set of samples as in the cumacean study of the authors revealed 253 species of Polychaeta (Sobczyk et al., 2021), only from the small fragment of the Ghanaian coast, placing this area amongst the important biodiversity hot spots for those marine annelids. We can expect that similar hidden biodiversity can be encountered for many other taxonomic groups, especially so important like small peracarid crustaceans.
Cumacea are classified as one of the orders of Peracarida. With about 1,400 recognized species (Gerken, 2018), this order is on the third place in terms of species richness within Peracarida, after Ampipoda (9,500 species) and Isopoda (about 6,000 species), and together with Tanaidacea (about 1,400 species). Their true diversity is vastly underestimated, mostly because of taxonomic expertise bias (Appeltans et al., 2012). As all peracarids Cumacea are small brooders with no planktonic larvae, they borrow in the surface layer of the sediment (Pilar-Cornejo et al., 2004). They are often found in the first few centimeters of sediments and occur from the intertidal, down to abyssal depths (Watling and Gerken, 2021). Cumacea are significant element of benthic communities, that in particular regions or depth zones (e.g., deep sea and tropical areas) might be one of the most diverse groups of crustaceans (Jones and Sanders, 1972; Cartes et al., 2003; Doti et al., 2020). For example, at the upper slope off Portugal, Cumacea together with Isopoda reached the highest number of the species (Cunha et al., 1997); while in the Angola Basin off Namibia, they were the third most abundant group of Peracarida, after Isopoda and Tanaidacea (Brandt, 2005). Moreover, some species might reach locally high abundance, even up to 500 individuals per square meter, both on the shelf, and in the deep sea (Bishop, 1982; Swaileh and Adeling, 1995). Cumacea also play an important role in the trophic webs, especially as food source for fish and some macroinvertebrates, such as decapods (Cartes, 1993; Watling and Gerken, 2021). For example, Diastylis rathkei might constitute even 35% of the diet of flounder (Swaileh and Adeling, 1995).
Available studies demonstrated that cumaceans display preferences to particular grain size, which makes them a good indicator of sediment type (Dixon, 1944; Wieser, 1956; Jones, 1976). Some species are known to be sensitive to environmental stress. Two dominant species in Algeciras Bay (Camella limicola and Nannastacus unguiculatus) were strongly influenced by hydrodynamism, sedimentation, and water turbidity (Alfonso et al., 1998). Some species of Cumacea are also good indicators of eutrophication, and have been proposed as organisms appropriate for biomonitoring (Corbera and Cardell, 1995; Ateş et al., 2014). Nevertheless, studies on biology and ecology of particular species or distribution patterns and structure of cumacean communities are still scarce, especially in the deep sea (e.g., Gage et al., 2004; Pabis and Blazewicz-Paszko wycz, 2011; Corbera and Sorbe, 2020 and references therein).
The knowledge about cumacean fauna of the Ghanian coast is highly scattered and based mostly on taxonomic publications (e.g., Băcescu, 1961, 1972; Day, 1975, 1978, 1980; Mühlenhardt-Siegel, 1996, 2000; Petrescu, 1998). So far, only 154 species of Cumacea are known from the whole African coast (Watling and Gerken, 2021), which makes 11% of the world fauna (Gerken, 2018). From West Africa, 59 species have been recorded, mostly from the continental shelf (Watling and Gerken, 2021), and only three were found on the Ghanaian coast (Jones, 1956; Petrescu, 2018). There are no quantitative studies on cumacean communities conducted on the African coast. Most earlier studies were focused on taxonomy.
Biodiversity assessments of the tropical continental margins are among the most important priorities of the current
marine science (Menegotto and Rangel, 2018). Therefore, the main aim of this study was to assess cumacean diversity on the continental shelf and slope of the Gulf of Guinea (25–1,000 m depth, Ghanaian coast) and compare it with literature data. We hypothesize that the Ghanaian continental margin hosts speciose cumacean communities with many species new to science. We wanted also to analyse the influence of various natural (e.g., oxygen, sediment type, salinity, and temperature) and anthropogenic (heavy metals and hydrocarbons) factors on the diversity and distribution patterns of those crustaceans. We hypothesize that local pollution might lower the abundance and diversity of those small crustaceans with limited dispersal abilities.
MATERIALS AND METHODS
Study Area
The Gulf of Guinea is a large open basin located in West Africa, influenced by a complex set of currents (Guinea Current, Benguela Current, South Equatorial Counter Current; Ukwe et al., 2003, 2006) and upwelling events (Nieto and Melin, 2017). The northern part of the Gulf of Guinea is influenced by seasonal upwelling, bringing nutrient-rich mid-depth waters to the surface and increasing the primary production (Binet and Marchal, 1993). The southern part depends rather on nutrient input from land drainage and river flood, mostly the Volta River, which is the only large river system located along the almost 600-km-long Ghanaian coast (Buchanan, 1957; Ukwe et al., 2003). The Gulf of Guinea is classified as a province in the Tropical Atlantic Realm, with rich fishery resources as well as large oil and gas reserves, and its sectors (e.g., north, central, and south) are considered a separate ecoregion (Spalding et al., 2007). The heterogeneity of habitats on continental margins has influence on high diversity of habitats for benthic fauna. At the same time, industrialization and the oil industry create numerous sources of disturbance that can potentially affect marine communities (Germain and Armengol, 1999; Owusu-Boadi and Kuitunen, 2002).
Sampling
The material was collected in October and November of 2012 from the board of RV Fridtjof Nansen. Nine transects were distributed along the whole coast of Ghana (Figure 1). Six stations were designated at each transect (25, 50, 100, 250, 500, and 1,000 m). Five replicate samples were collected at each station using a van Veen grab (0.1 m²) supported with the Video Assisted Multi Sampler (VAMS), allowing for appropriate sediment penetration. The samples were washed using 0.3 mm mesh size and preserved in 4% formaldehyde solution. The material was collected in the framework of the Oil for Development (OfD) program, and supported by the Food and Agriculture Organization of the United Nations (FAO).
Environmental Factors
Physical and chemical properties of the sediment and water were also analyzed at each station. Temperature, conductivity, and oxygen level were measured using a Seabird 911 CTD Plus and SBE 21 Seacat thermosalinograph. Sediment structure (percentage content of gravel, sand, and silt) was also analyzed. Diameter of particles was calculated using the equations of Buchanan (1984), and Folk and Ward (1957). Level of total hydrocarbons (THC), toxic metals: arsenic (As), barium (Ba), cadmium (Cd), chromium (Cr), copper (Cu), nickel (Ni), lead (Pb), zinc (Zn), and content of total organic matter (TOM) were also measured. Total hydrocarbon content was analyzed using a gas chromatograph with a flame ionization detector (GC/FID), as outlined in the Intergovernmental Oceanographic Commission, Manuals and Guides No. 11, UNESCO Intergovernmental Oceanographic Commission (1982) while toxic metals contents were analyzed via inductively coupled plasma-atomic emission spectrometry (ICP-AES) (Jarvis and Jarvis, 1992; Elezz et al., 2018). Total organic matter was determined as the weight loss in a 2–3-g dried sample (dried at 105°C for 20 h) after 2 h of combustion at 480°C.
Analysis of Diversity and Abundance
Specimens were identified at the morphospecies level (Wägele, 2005). We have calculated species richness (S—number of species per sample), diversity (Shannon Index) and evenness (Pielou Index) (Magurran, 2012) as well as abundance [ind./0.1 m²] for each sample. Mean values with standard deviations (SD) of those indices were calculated for each depth zone and for the whole material. Statistical differences between the depth zones were assessed by non-parametric Kruskal–Wallis test. Post-hoc testing was done by Dunn’s test. This part of the analysis was performed using a STATISTICA 13 package (StatSoft, 2006).
Species accumulation curves averaging over 999 permutations were created using the PRIMER package. The curve plotted the cumulative number of different species observed as each new sample was added (Clarke and Warwick, 2001).
We have also assessed the number of rare species recorded in the material. The number of singletons (species represented by only one individual in the whole material, in all collected samples) and doubletons (represented by two individuals), and the number of unique species (species found in one sample only) and duplicates (species found in two samples only) were also calculated. Additionally, we have calculated the number of species recorded only in a given depth zone or given transect as well as the number of species common to given depth zones and transects. Frequency of occurrence (F—percentage of samples where a species was found in total number of samples) was calculated for each species.
Multivariate Analysis
Hierarchical agglomerative clustering analysis, based on the Bray–Curtis formula, was performed to check for faunistic similarity among the stations. For the analysis, mean values of abundance of each species calculated for each station were used. Data were square root-transformed, and the group average method was used. A SIMPROF test with 1% significance level was performed to check the multivariate structure within groups. This part of the analysis was performed using a Primer package (Clarke and Warwick, 2001).
The R software was used for all analyses of environmental factors influence on cumacean communities (R Core Team,
We used the Pearson correlation matrix (corplot::corplot; Wei and Simko, 2017) to assess pair-wise cross-correlation between each environmental variable. Based on strong correlation ($r > 0.6$), we included six variables from the initial set of 19 variables into further analysis, assuming they have an ecologically important role in explaining the richness of cumaceans.
Finally, six variables: Ba, Cd, THC, oxygen, gravel, and salinity were added into further analysis (Figure 2). For full list of environmental variables, please see Supplementary Table 1. Salinity was used in principal component analysis (PCA) only. Yeo-Johnson power transformation [caret::preProcess(); Kuhn, 2020] was used for reducing deviations linked with unequal ranges off selected factors (e.g., Ba). Next, PCA was performed to show dissimilarities in species composition among transects and stations [vegan::rda(); Oksanen et al., 2019]. Additionally, ranges of salinity were added to the PCA [vegan::ordisurf(); Oksanen et al., 2019] to demonstrate salt content relations in arrangement of stations in ordination space.
Species richness and PCA axis were used to fit generalized linear models (glm; for species richness) or linear models (lm; for PCA axis 1 and axis 2) with five environmental variables (Ba, Cd, THC, oxygen, and gravel) as fixed effects using the stats4 package [stats4::lm(), stats4::glm(); R Core Team, 2020]. Poisson distribution was used for species richness. To choose the best fitted models based on corrected Akaike Information Criterion (AICc), the dredge function was used (MuMIn::dredge; Barton, 2018). To calculate estimates of function slopes for the models with $\Delta$AICc < 2, model averaging was employed [MuMIn package model.avg(), confset95p(), and avgmod.95p()]. The RSquareAdj function (vegan::RSquareAdj; Oksanen et al., 2019) was computed to reveal how much variance was explained by averaged models for PCA axes 1 and 2.
Hierarchical partitioning function (hier.part::hier.part(); Walsh and Mac Nally, 2013) for species richness as well as PCA axes 1 and axis 2 was used for checking the independent effect (%) of each environmental variable and its joint contribution to all other predictors. To compute it, goodness-of-fit measures for all model combinations with all predictors, with Gaussian (for PCA axes 1 and 2) or Poisson distribution (for species richness) were used. Statistical significance of the relative contribution of each predictor were determined by randomization test [hier.part::rand.hp()] with implementing $P$-values and $z$-scores (confidence limit $< 0.95$).
RESULTS
Diversity and Abundance
Altogether, 63 species (391 individuals) of Cumacea were identified. They represented 13 genera and six families (Table 1). It is assumed that 95% of species (60 species) are new to science. The most speciose genera were: Leucon (12 species), Eocuma (8), Iphinoe, and Diastaylis (both genera with seven species). The most abundant genera were: Eocuma...
(95 individuals), *Bodotria* (60 individuals), and *Leucon* (50 individuals). Together, they constituted over half of the material. The most speciose and abundant families were: *Bodotridae* (23 species, 235 individuals), *Leuconidae* (15 species, 66 individuals), *Nannastacidea* (10 species, 67 individuals), and *Diastylidae* (10 species, 20 individuals). A large number of rare species were also recorded. In the whole material, 13 singletons and 15 doubletons were found. Seventeen species were found only in one sample, and 14 in two samples only. Frequency of occurrence of species in the whole material was extremely low. Only five species had frequency higher than 4%. The species with the highest frequency of occurrence in the whole material was *Eocuma* sp. 7 that was found only in 7% of the samples (Table 1). The species accumulation curve did not reach the asymptotic level, suggesting undersampling of the studied area (Figure 3).
The mean density of cumaceans calculated for all collected samples equalled to only 1.5 ind./0.1 m$^2$. General species richness and abundance decreased along a depth gradient. The highest number of species was found at 25 and 50 m with 17 and 28 species recorded, respectively (Table 2). Moreover, 15 species were common in those two depth zones (Table 3). On the outer shelf and upper slope, the number of species was lower and increased again to 19 species at 1,000 m stations (Table 2). It is also the depth zone with the most unique fauna, as 14 out of 19 species were recorded only here (Table 2). The general number of species was similar in most of the transects (Table 2). The highest number of species was found in a transect G6 (29 species), while the lowest species richness was recorded in transects G8 and G9 with seven and 10 species, respectively (Table 2). The highest number of species common with other transects was recorded in transect G6, but generally there was no clear pattern observed (Table 4).
Mean species richness and diversity per sample were the highest on the shallows (25 m – number of species per sample 1.04 ± 1; Shannon Index 0.2 ± 0.3, 50 m – number of species per sample 1.4 ± 1.2, Shannon Index 0.4 ± 0.4) and at 250 m (number of species per sample 1.1 ± 1, Shannon Index 0.3 ± 0.3) (Figure 4). Evenness was the highest at 1,000 m [0.9 ± 0.03 (Figure 4)]. Mean abundance changed along the depth gradient, and the highest values were observed at 50 (2.2 ± 2.2 ind./0.1 m$^2$) and 250 m (1.8 ± 1.8 ind./0.1 m$^2$). Below 250 m, it decreases with increasing depth (Figure 4).
**Cluster Analysis**
Four groups were distinguished in the cluster analysis although at low similarity level (20% or less), but all were significantly differentiated by the SIMPROF (Figure 5). Inner shelf areas were clearly separated from the outer shelf and slope showing strong depth zonation of cumacean communities. Two clusters (B and C) of grouped samples were collected at depth 25–50 m. The next two clusters consist of samples collected at depth 100–500 (cluster A) and 500–1,000 m (cluster D). The clusters differ in family and genera composition, number of species, and frequency of the species.
In the samples grouped in cluster A, 22 species were found, and eight of them belong to family Bodotridae, and six to family Diastylidae. Genus *Diastylis* was represented by six species and genus *Iphione* and *Eocuma* by three species each. The highest frequency of occurrence (56%) was observed for *Campylapsis* sp 2 and *Eocuma* sp 7.
Eleven species were found in samples forming cluster B, and nine of them represent family Bodotridae. The most speciose genera were *Bodotria* with five species and *Iphinoe* with two species. *Vaunthampsonia* sp 1 was present in 80% of the samples, and *Bodotria* sp 2 and *Eudorellopsis* sp 1 were present in 60% of the samples.
### TABLE 1 | List of species with total abundance, frequency of occurrence in samples, and depth range.
| Family | Genus | Number of specimens | Frequency [%] | Depth range [m] |
|------------|---------------------|---------------------|---------------|-----------------|
| Bodotriidae| *Eocuma* sp. 1 | 10 | 2.0 | 25–50 |
| Bodotriidae| *Eocuma* sp. 2 | 14 | 2.8 | 50 |
| Bodotriidae| *Eocuma* sp. 3 | 13 | 2.8 | 25–50 |
| Bodotriidae| *Eocuma* sp. 4 | 5 | 1.6 | 25–50 |
| Bodotriidae| *Eocuma* sp. 5 | 4 | 1.6 | 25–150 |
| Bodotriidae| *Eocuma* sp. 6 | 8 | 2.4 | 25–50 |
| Bodotriidae| *Eocuma* sp. 7 | 37 | 7.2 | 100 |
| Bodotriidae| *Eocuma* sp. 8 | 4 | 1.2 | 250 |
| Bodotriidae| *Bodotria* sp. 1 | 16 | 2.8 | 25–50 |
| Bodotriidae| *Bodotria* sp. 2 | 9 | 3.2 | 25–50 |
| Bodotriidae| *Bodotria* sp. 3 | 17 | 4.0 | 25–250 |
| Bodotriidae| *Bodotria* sp. 4 | 14 | 2.4 | 25–50 |
| Bodotriidae| *Bodotria* sp. 5 | 3 | 0.8 | 25–50 |
| Bodotriidae| *Bodotria* sp. 6 | 1 | 0.4 | 50 |
| Bodotriidae| *Cycluspis* sp. 1 | 1 | 0.4 | 1,000 |
| Bodotriidae| *Iphinoe* sp. 1 | 4 | 1.6 | 50–100 |
| Bodotriidae| *Iphinoe* sp. 2 | 31 | 4.8 | 25–100 |
| Bodotriidae| *Iphinoe* sp. 3 | 2 | 0.8 | 25 |
| Bodotriidae| *Iphinoe* sp. 4 | 1 | 0.4 | 50 |
| Bodotriidae| *Iphinoe* sp. 5 | 2 | 0.8 | 50 |
| Bodotriidae| *Iphinoe* sp. 6 | 1 | 0.4 | 50 |
| Bodotriidae| *Iphinoe* sp. 7 | 2 | 0.8 | 250 |
| Bodotriidae| *Vaunthompsonia* sp. 1 | 14 | 3.2 | 25–100 |
| Diastylidae| *Diastylis* sp. 1 | 1 | 0.4 | 250 |
| Diastylidae| *Diastylis* sp. 2 | 3 | 0.8 | 50–100 |
| Diastylidae| *Diastylis* sp. 3 | 4 | 1.6 | 50–250 |
| Diastylidae| *Diastylis* sp. 4 | 1 | 0.4 | 50 |
| Diastylidae| *Diastylis* sp. 5 | 2 | 0.8 | 100 |
| Diastylidae| *Diastylis* sp. 6 | 1 | 0.4 | 250 |
| Diastylidae| *Diastylis* sp. 7 | 3 | 1.2 | 50–1,000 |
| Diastylidae| *Makrokylindrus* sp. 1 | 1 | 0.4 | 1,000 |
| Diastylidae| *Makrokylindrus* sp. 2 | 2 | 0.8 | 1,000 |
| Diastylidae| *Makrokylindrus* sp. 3 | 2 | 0.4 | 1,000 |
| Lampropidae| *Lampropidae* sp. 1 | 14 | 2.4 | 250 |
| Lampropidae| *Lampropidae* sp. 2 | 7 | 1.2 | 250 |
| Leuconidae| *Eudorella* sp. 1 | 2 | 0.8 | 50–100 |
| Leuconidae| *Eudorella* sp. 2 | 2 | 0.4 | 1,000 |
| Leuconidae| *Eudorellopsis* sp. 1 | 4 | 1.6 | 25–50 |
| Leuconidae| *Leucon* (Epileucon) sp. 1 | 18 | 2.8 | 500–1,000 |
| Leuconidae| *Leucon* (Epileucon) sp. 2 | 1 | 0.4 | 1,000– |
| Leuconidae| *Leucon* (Macrauloleucon) sp. 3 | 6 | 2.4 | 100–500 |
| Leuconidae| *Leucon* (Macrauloleucon) sp. 4 | 5 | 1.2 | 500 |
| Leuconidae| *Leucon* (Macrauloleucon) sp. 5 | 1 | 0.4 | 1,000 |
| Leuconidea| *Leucon* (Crymoleucon) sp. 6 | 3 | 0.8 | 1,000 |
| Leuconidea| *Leucon* (Leucon) sp. 7 | 2 | 0.8 | 500 |
| Leuconidea| *Leucon* (Leucon) sp. 8 | 7 | 1.2 | 500 |
| Leuconidea| *Leucon* (Leucon) sp. 9 | 2 | 0.8 | 1,000 |
| Leuconidea| *Leucon* (Leucon) sp. 10 | 3 | 1.2 | 500–1,000 |
| Nannastacidae| *Campylaspis* sp. 1 | 7 | 2.0 | 25–1,000 |
| Nannastacidae| *Campylaspis* sp. 2 | 20 | 5.6 | 50–250 |
| Nannastacidae| *Campylaspis* sp. 3 | 15 | 4.4 | 25–500 |
TABLE 1 | Continued
| Family | Genus | Number of specimens | Frequency [%] | Depth range [m] |
|-------------|---------------------|---------------------|---------------|-----------------|
| Nannastacidae | Campylaspis sp. 4 | 4 | 1.2 | 25 |
| Nannastacidae | Campylaspis sp. 5 | 1 | 0.4 | 1,000 |
| Nannastacidae | Cumella sp. 1 | 2 | 0.4 | 250 |
| Nannastacidae | Cumella sp. 2 | 2 | 0.4 | 1,000 |
| Nannastacidae | Cumella sp. 3 | 2 | 0.8 | 500 |
| Nannastacidae | Cumella sp. 4 | 5 | 1.6 | 1,000 |
| Nannastacidae | Nannastacidae sp. 1 | 9 | 2.8 | 500–1,000 |
| Pseudocumatidae | Pseudocumatidae sp. 1 | 2 | 0.8 | 1,000 |
| | indet sp. 1 | 2 | 0.8 | 500 |
| | indet sp. 1 | 1 | 0.4 | 1,000 |
FIGURE 3 | Species accumulation curve for cumacean fauna sampled at the Gulf of Guinea.
Within cluster C, 28 species in total were observed, and 16 species belong to family Bodotridae. The most speciose genera were Eocuma with six species and Bodotria with five species. Iphinoe sp 2 and Eocuma sp 2 were characterized by the highest frequency of occurrence, which was 73 and 56%, respectively.
In the samples from cluster D, 25 species were recorded, and 10 belong to family Leuconidae and seven to Nannastacidae. Leucon was the most speciose genus (10 species), followed by Cumella (three species) and Campylaspis (three species). Nannastacidae sp. 1 had the highest frequency of occurrence, and it was present in 40% of the samples.
Influence of Physical and Chemical Factors on Cumacean Communities
PCA1 and PCA2 axes explained about 20% of variance. The first axis (10.7% variance explained) showed high dissimilarity between stations located at 100 m, and all other sites followed dissolved oxygen and salinity gradient. Three groups were established in the PCA mostly along the PCA2 axis (9% variance explained) (Figure 6). The first one (lower part of gradient) contained shallow water samples (25–50 m), characterized by higher concentration of oxygen and gravel. Here, a sandy type of substratum with relatively high contribution of gravel (depth zone 25–50 m) was noticed. Content of Ba, Cd, as well as THCs was significantly lower. The second group (higher part of gradient) contained samples from 250–1,000 m deep; and here, the samples were characterized by higher concentration of Ba, Cd, and THC. Lower concentration of oxygen and lower content of gravel and sand were noted here. Bottom deposits constituted mostly of silt. Salinity reached low to average values with high range (34.8–35.6‰). The third group contained samples from 100 m depth. The samples were distinguished from the other groups by high salinity content with low values range (35.7–35.8‰), and bottom substrate was dominated by sand and silt.
A set of two most parsimonious (with ΔAICc < 2) linear models for PCA axis 1 revealed that high content of gravel...
TABLE 2 | Total number of species in each depth zone/transect and number of species recorded only in a given depth zone/transect.
| Depth zone | Number of unique species | Total number of species | Percentage of unique species |
|------------|--------------------------|-------------------------|-----------------------------|
| 25m | 2 | 17 | 11.7 |
| 50m | 7 | 28 | 25.0 |
| 100m | 2 | 11 | 18.1 |
| 250m | 7 | 15 | 46.6 |
| 500m | 5 | 10 | 50.0 |
| 1,000m | 14 | 19 | 73.6 |
**Transect**
| Transect | G1 | G2 | G3 | G4 | G5 | G6 | G7 | G8 | G9 |
|----------|----|----|----|----|----|----|----|----|----|
| G1 | 1 | | | | | | | | |
| G2 | 1 | | | | | | | | |
| G3 | 2 | | | | | | | | |
| G4 | 8 | | | | | | | | |
| G5 | 2 | | | | | | | | |
| G6 | 1 | | | | | | | | |
| G7 | 4 | | | | | | | | |
| G8 | 2 | | | | | | | | |
| G9 | 0 | | | | | | | | |
TABLE 3 | Species common between the depth zones.
| Depth zone | 25 | 50 | 100 | 250 | 500 |
|------------|----|----|-----|-----|-----|
| 25 | | | 15 | | |
| 50 | 2 | | 7 | | |
| 100 | 3 | 6 | | | 4 |
| 250 | | 6 | | 4 | |
| 500 | 1 | 1 | 1 | 2 | |
| 1,000 | 1 | 2 | 0 | 1 | 3 |
TABLE 4 | Species common between the transects.
| Transect | G1 | G2 | G3 | G4 | G5 | G6 | G7 | G8 |
|----------|----|----|----|----|----|----|----|----|
| G1 | | | | | | | | |
| G2 | 8 | | | | | | | |
| G3 | 7 | 8 | | | | | | |
| G4 | 8 | | | | | | | |
| G5 | 7 | | | 8 | | | | |
| G6 | 15 | | 12 | 12| 11 | 13 | | |
| G7 | 8 | 2 | 5 | | 4 | 4 | 8 | |
| G8 | 2 | | 4 | | 3 | 2 | | 2 |
| G9 | 4 | 5 | 4 | 3 | 5 | 9 | 3 | 1 |
(estimate slope: $-0.30$, $p = 0.05$) as well as oxygen (estimate slope: $0.25$, $p < 0.001$) shaped species composition along the axis. The influence of gravel was negatively correlated with species composition along axis 1. However, higher concentration of oxygen dissolved in water had a positive influence on it (Table 5, Figure 7A, Supplementary Table 2). The model explained about 31% of total variance. Hierarchical partitioning revealed that only the influence of oxygen (relative contribution: 64.3%) was significant for PCA axis 1 (Figure 8).
Form the three models best describing species composition along PCA axis 2 (containing Ba, THCs, oxygen, and gravel) we noted statistical significance of oxygen, gravel, and Ba. The higher content of oxygen (estimate slope: $-0.16$, $p < 0.001$) and gravel (estimate slope: $-0.39$, $p = 0.002$) had a negative influence on species composition along axis 2, while Ba (estimate slope: $0.21$, $p = 0.01$) enhanced it (Table 5, Figure 7B, Supplementary Table 2). The model explained about 62% of total variance of cumaceans. The relative contribution of each factor supports the previous results. The influence of barium, gravel, and oxygen (independent effect: 29.3, 29.7, and 19%, respectively) was statistically significant (Figure 8).
A set of four most parsimonious models with $\Delta$AICc < 2 best explained richness of cumaceans species. Ba, gravel, and oxygen were included into best averaged model; however, only the adverse effect of Ba (estimate slope: $-0.37$, $p = 0.04$) was statistically significant and caused decrease in species richness (Table 5, Figure 9, Supplementary Table 2). Based on the results of hierarchical partitioning, we found that the influence of Ba and oxygen (relative contribution: 24 and 32.1%, respectively) was significantly correlated with species richness (Figure 8).
**DISCUSSION**
**Species Richness**
Despite very low total abundance, the cumacean species richness on the Ghanaian coast was very high, and the species accumulation curves still showed substantial undersampling (Figure 3). Taking into account generally high sampling effort (much larger than in most of other cumacean studies, see Table 6) and large number of sampled stations, it can be assumed that great species rarity was the main reason behind this result. Large number of singletons and doubletons as well as large number of species recorded in a very low number of samples was typical feature of cumacean communities along the Ghanaian coast (Table 1). Moreover, the primary analysis indicates that 95% of collected species are new to the science. The results demonstrated the highly underestimated diversity of those crustaceans in the Gulf of Guinea, even compared with the global diversity of Cumacea, which was estimated at about 1,400 species (Gerken, 2018). After this study, the list of known cumacea from the coast of Guinea increased from 3 to 66 species (Watling and Gerken, 2021), which is a significant result for about 500-km long part of the coastline.
Comparisons with other studies of cumacean species richness are difficult because of large discrepancies in type of gear used, scale of the sampling area, sampling effort, and studied depth range, not to mention the differences in local environmental conditions or geological history of various regions (Table 6). Nevertheless, the number of 63 species is comparable even with much larger areas that were sampled extensively for a very long time. For example, in the whole Antarctic waters, 86 species of cumacea were recorded (De Broyer and Danis, 2011). Extensive
analysis of literature data resulted in the list of 172 species recorded from Iberian waters (Atlantic and Mediterranean) including 142 species found in bathyal (200–3,000 m) (Corbera, 1995). The current list of all Mediterranean cumaceans (such as Tyrrenhian, Adriatic, Aegea, and Levantine Seas) includes 99 species (Coll et al., 2010). Analysis of large set of 122 epibenthic sledge samples collected in the deep Atlantic (500–4,000 m depth) revealed the presence of 225 species, although from a large number of almost 56,000 individuals (Gage et al., 2004). On the other hand, only 29 species were recorded from a whole region of tropical Eastern Pacific (Jarquin-Gonzalez and Garcia-Madel, 2013) and only 34 species from the whole Chilean coast (Thiel et al., 2003). Even in the intensively sampled, large open system of the Bay of Biscay, the number of recorded cumacean species was lower than in this study. In the subtidal zone (up 63 m depth) 18 species were recorded in over 100 samples collected using the van Veen grab (Cacabelos et al., 2010; Corbera et al., 2013; Corbera and Galil, 2016). At the deeper areas of the bay in Kostarrenkala area, 42 species in total were collected (13 species were found at 170 m, six species at 300 m, nine species at 400 m, 18 species at 724 m, and 24 species at 1,000 m) (Frutos and Sorbe, 2014). We have analyzed sampling effort and cumacean species richness from 39 different sampling campaigns (Table 6). In majority of the studies, the number of species was lower than 35, even if the sampling effort was high, and even if an epibenthic sledge or other gears collecting large number of individuals and species were used. Nevertheless, it is worth mentioning that this study was conducted in a relatively wide depth range. Cumaceans have low dispersal potential (Jones and Sanders, 1972; Pilar-Cornejol et al., 2004), therefore, large depth range sample (25–1,000 m), together with large diversity of microhabitats and differences in environmental conditions, could result in recognition of a larger number of species. It is clearly visible in the analysis of species common in different depth zones (Table 3) and in the results of the cluster analysis (Figure 5).
Based on current data, we cannot postulate that the Gulf of Guinea is a hot spot of cumacean diversity, although that kind of assumptions is likely possible. There were previous suggestions that this region might be an important center of cumacean diversity. In the deeper parts of the Angola Bay (5,125–5,415 m), 45 species were recorded in just seven epibenthic sledge samples (Brandt, 2005; Mühlenhardt-Siegel, 2005), while Bochert and Zettler (2011) described 16 additional species from the shelf of the Angolan and Namibian waters. High diversity of Cumacea on the equator was already mentioned by Jones and Sanders (1972) and later supported by large scale latitudinal analysis of the deep Atlantic cumacean richness, although authors declare that it is difficult to say if this pattern is related to geological and evolutionary history (e.g., glaciation in the northern hemisphere) or differences in more recent changes in local ecological conditions (e.g., high productivity in tropical areas; Gage et al., 2004). Confrontation of those observed patterns with the knowledge about large sampling bias in the tropical marine waters (Menegotto and Rangel, 2018) demonstrates that many important questions regarding the distribution patterns and diversity are still open and need further comprehensive studies. Based on the current data, it is impossible to discuss about the dependencies between local (e.g., Ghanaian coast) and regional (e.g., whole Atlantic African coast, West Africa, and Gulf of Guinea) species pools (Witman et al., 2004) or provide any generalizations about factors influencing diversity on a larger scale. At the same time, we did not observe large differences in species composition on intermediate scale (between investigated transects). There were some transects with very low (G8 – 7 species) or very high (G6 – 29 species) total number of species, but at the same time the number of species unique to a given transect was very low (Table 2),
and there was no clear spatial pattern in species common to different transects, even from opposite parts of the Ghanaian coast (Table 4). Those differences are rather not related to distance between the transects but are most probably due to the influence of local environmental conditions as shown in the PCA analysis.
**Distribution Patterns and Diversity on a Background of Environmental Conditions**
Oxygen content and sediment type (especially content of gravel) drive species composition and diversity especially in the 25–50 m stations. Well-oxygenated water and elevated primary production may increase species richness (Levin and Sibuet, 2012; McCallum et al., 2015). On the other hand, in previous studies, Pabis et al. (2020) reported low oxygen concentration at 250–500 m depth on the coast of Ghana, and those factors might also cause decrease in cumacean abundance and species richness in this depth zone, although the pattern is not clear, and visible only at 500 m. Lower oxygen concentrations might be caused by sinking organic matters resulting from seasonal upwelling at the Ghanaian coast (Nieto and Mélin, 2017).
There is no evidence that increased salinity may reduce the abundance and species richness of cumacea. We suspect that higher salinity values at 100 m were a result of seasonal and oceanographic factors such as upwelling events, bottom currents, temperature, or rainy seasons (Ukwe et al., 2003; Djagoua et al., 2011; Nieto and Mélin, 2017). Martin et al. (2010) showed that decrease in salinity may increase the activity of cumaceans in water column, although we have observed only slight differences in salinity along the coast of Ghana. Therefore, based on available data, we cannot speculate about its influence on cumacean communities.
Earlier studies support our result, pointing substrate grain size and organic matter content as the most important drivers of cumacean assemblages (Corbera and Cardell, 1995; Dos Santos and Pires-Vanin, 1999; Cristales and Pires-Vanin, 2014; Corbera and Sorbe, 2020). In the study of the shallow water communities of the Persian Gulf, the presence of gravel also had a positive
influence on cumacean fauna (Martin et al., 2010). The positive effect of gravel on the shallow water communities was also confirmed in the study of polychaete functional diversity in the Ghanaian waters (Sobczyk et al., 2021). The presence of coarser sediment fractions might increase habitat complexity and heterogeneity for small infaunal invertebrates such as cumaceans or various groups of polychaetes, resulting in higher number of microhabitats and/or ecological niches and increased diversity (Sebens, 1991; Carvalho et al., 2017). We also have to take into account interactions with other benthic organisms occurring in the shallows. Generally, the abundance and diversity of benthic fauna of the Ghanaian coast were highest in the 25–50 m depth range (Pabis et al., 2020). This fact might increase the diversity of mutual interactions between various organisms, for example, because of higher level of sediment bioturbation, which could influence oxygenation of the sediment and food availability (Aller and Cochran, 2019). Such conclusions are supported by high abundance and diversity of burrowing polychaetes recorded in this depth zone along the Ghanaian coast (Sobczyk et al., 2021).
Sediment character might be crucial for cumacean survival, as it can be strictly related to the feeding strategy and respiratory mechanism (Dixon, 1944 in Dos Santos and Pires-Vanin, 1999). Cumacea feed on microorganism (especially diatoms) and/or detritus (Błązewicz-Paszkowycz and Ligowski, 2002). It is assumed that mud-dwellers filter small particles of suspension, while sand dwellers scrub food from sand grains. However, studies on Cumella vaulgaris demonstrated that the attractiveness of a particular substratum depends on the amount and type of food (Wieser, 1956). The type of substrate is also suggested to have some impact on filter apparatus appearance in some cumaceans. Species that live in muddy sediment have the filter apparatus equipped with finely feathered bristles that allow easier water flow. For example, members of the Diastylis are known to have filter apparatus adapted to catching small particles of food from water (Dixon, 1944). Nevertheless, the knowledge on diet, habitat preferences, and other aspects of the cumacean biology is extremely scarce. We know nothing about the ecology and biology of majority of genera, and it is impossible to link the results with any data about the biology of particular species recorded in West Africa.
Slope communities were also affected by disturbance associated with the influence of barium, other heavy metals, and hydrocarbons that are associated mainly with increasing activities of petroleum companies. The oil industry (e.g., Jubilee Oil Field) combined with pollution from other sources such as the dyeing industry, leaks from crude oil storage, and inputs of polluted fresh water have an important influence on the Gulf of Guinea (Acquah, 1995; Owusu-Boadi and Kuitunen, 2002; Scheren et al., 2002; Ayamdoo, 2016; Hanson and Kwarteng, 2019). For example, between 2009 and 2011, there was a spill of oil-based mud in Ghanaian waters, and the control of pollution in this region remains poor and not well-documented, although it is considered to continuously increase (Ay amdoo, 2016). Moreover, Ghana is importing barite for the dyeing industry (Sobczyk et al., 2021). Larger concentration of Ba on the slope is also not surprising because of the influence of pressure on the solubility of barite (Neff, 2002). At the same time, elevated levels of barium were not visible in all slope stations, but only on part of the transects (Pabis et al., 2020), confirming that pollution has local anthropogenic origin. Despite the fact that cumacean abundance and species richness per sample were generally low along the whole depth range, we have noticed decreased values in the slope samples, where muddy sediments are characterized by higher content of barium, other toxic metals, and hydrocarbons (Pabis et al., 2020). Those factors might influence benthic communities (e.g., Olsgard and Gray, 1995; Gomez-Gesteira et al., 2003; Stark et al., 2020), and it is known that heavy metals might accumulate in cumacean bodies (Swaileh and Adelung, 1995). Ba and other heavy metals may affect development and cause decrease in abundance of benthic invertebrates (Lira et al., 2011), or influence embryos of crustaceans and bivalves (Macdonald et al., 1988). Similar effects were described for hydrocarbons (Main et al., 2015; Honda and Suzuki, 2020). Nevertheless, there are only scarce data about exact doses of various pollutants that could influence particular species or taxonomic groups of benthic organisms (Lira et al., 2011). We already noticed in the earlier study (Pabis et al., 2020) that levels of Ba and other toxic metals in the Gulf of Guinea were close to background levels according to OSPAR and KLIF (Norwegian Pollution Authority) guidelines (OSPAR, 2017), although literature data from other regions demonstrated that even low concentrations of Ba and other pollutants might influence benthic communities (Olsgard and Gray, 1995). The influence of local pollution associated with oil exploration in Ghanaian waters and dying industry was also visible in the study on polychaete functional diversity that was based on the same set
| Response variable | Model | df | logLik | AICc | ΔAICc | Weight |
|------------------|-------|----|--------|------|-------|--------|
| Site scores along PCA ordination axis 1 | Gravel + Oxygen | 4 | −10.68 | 30.3 | 0.00 | 0.310 |
| | Ba + Gravel + Oxygen | 5 | −10.30 | 32.0 | 1.73 | 0.313 |
| Site scores along PCA ordination axis 2 | Ba + Gravel + Oxygen | 5 | 4.87 | 1.7 | 0.00 | 0.409 |
| | Ba + Gravel + Oxygen + THC | 6 | 5.66 | 2.7 | 1.02 | 0.245 |
| | Ba + Cd + Gravel + Oxygen | 6 | 5.34 | 3.3 | 1.66 | 0.178 |
| Richness of cumaceans | Ba + Oxygen | 3 | −90.00 | 186.6 | 0.00 | 0.155 |
| | Ba + Gravel + Oxygen | 4 | −88.87 | 186.8 | 0.16 | 0.143 |
| | Ba | 2 | −91.76 | 187.8 | 1.20 | 0.085 |
| | Ba + Gravel | 3 | −90.84 | 188.3 | 1.67 | 0.067 |
of samples (Sobczyk et al., 2021). Patterns observed in cumacean study are very similar, although not that obvious and strong as in the case of polychaetes, which is most probably caused by generally very low abundance of those crustaceans. Moreover, polychaetes are considered perfect model organisms for various studies on ecosystem response to natural or anthropogenic changes and disturbances (Olsgard et al., 2003; Giangrande et al., 2005), and it is not surprising that they are good indicators of disturbance.
Cumaceans are small benthic brooders with limited dispersal potential. Therefore, they are considered to be sensitive to changes in environmental factors (Corbera and Cardell, 1995; Alfonso et al., 1998), although there are no data on influence of pollution on their communities, except those of one study.
FIGURE 8 | Relative contribution of each environmental factor to shared variability of full models testing for effects of environmental factors on species composition (expressed as site scores along PCA ordination axes (1 and 2) and richness of cumaceans. Predictors that had significant effect on response variables are given in white. Plus (+) signs express positive impact of predictors on response variables, and minus (-) signs express negative influences. For full predictor names, see Supplementary Table 1.
FIGURE 9 | Visualization of generalized linear model testing for effects of environmental factors on species richness of cumaceans. Phrase “n.a.” means that environmental factor was not included in a set of the most parsimonious models. Phrase “n.s.” means its explanatory power was not significant despite the fact that means environmental factor was included in a set of the most parsimonious models.
showing decrease in abundance in the polluted site (de-la-Ossa-Carretero et al., 2012). However, studies on similar small peracarid crustaceans such as Tanaidacea demonstrated that they might be good indicators of disturbance processes (Guerra-García and García-Gómez, 2004). The influence of local pollution on the Ghanaian coast was visible even in the case of higher taxa, although the taxonomic level of phyla and orders is normally not sufficient for meaningful assessments of ecosystem health (Pabis et al., 2020). Moreover, we have to remember that despite the fact that Ba was a significant factor in the analysis, other variables such as hydrocarbons and other heavy metals such as Cd, Cu, and Ni could also be responsible for combined influence on cumacean communities (Sobczyk et al., 2021). In such cases, it is difficult to unequivocally assess the influence of one out of multiple stressors on benthic communities (Borja et al., 2011; Lenihan et al., 2018), even by advanced multivariate analysis and especially when we analyse communities of less abundant taxa-like cumaceans. The results of the PCA are not strong, since first PC axis explained only 10% of variance, which is due to very low abundance, large number of singletons, and highly patchy distribution of majority of species. Nevertheless, the results are supported by analysis based on the abundance of macrozoobenthic higher taxa and polychaete communities (Pabis et al., 2020; Sobczyk et al., 2021). Moreover, similar results of the PCA are sometimes sufficient for
TABLE 6 | Sampling effort, species richness, and total abundance of Cumacea from various studies.
| Area | Gear | Total number of samples | Depth [m] | Number of individuals | Number of species | Abundance | References |
|-----------------------------|-----------------------------|-------------------------|-----------|-----------------------|-------------------|-----------|------------------------------------------------|
| off Santos, SE Brazil | Box corer (0.1 m²) | 21 | 10–100 | 919 | 24 | nd | Cristales and Pires-Vanin, 2014 |
| E Mediterranean Sea | Box corer, epibenthic sledge, beam trawl | 161 | 45–4,398 | nd | 29 | nd | Mühlenhardt-Siegel, 2009 |
| W Mediterranean Sea, coast of Barcelona | Dredge (0.1 m²) | 40 | 5–70 | nd | 22 | 0–613 indv/m² | Corbera and Cardell, 1995 |
| Ross Sea | Dredge | 19 | 84–515 | 5,287 | 28 | nd | Rehm et al., 2007 |
| Tropical Eastern Pacific | Grabbing | 13 | Max 10 | 378 | 29 | nd | Jarquin-Gonzalez and Garcia-Madrigal Mdel, 2013 |
| Algeciras Bay, Giblartar Strait | Scuba diving | 25 | Shallow | 2,058 | 3 | nd | Alfonso et al., 1998 |
| Bermuda | Scuba diving | 23 | 1–6 | 825 | 7 | nd | Petrescu and Sterrer, 2001 |
| California coast (Dillon Beach) | Scuba diving | 20 | 1–21.5 | 952 | 12 | 1–209 indv/0.04 m² | Gladfelter, 1975 |
| Puerto Morelos Reef National Park Mexico | Scuba diving | nd | 3–12 | 177 | 30 | nd | Monroy-Velázquez et al., 2017 |
| W Mediterranean Sea, Creixell beach | Sledge | 1,800 | 0.5–3 | N/A | 6 | nd | San Vicente and Sorbe, 1999 |
| Bay of Seine, English Channel | Sledge | 38 | 8–13 | N/A | 5 | 352.6–15.5 indv/100 m³ | Wang and Dauvin, 1994 |
| Hendaya and Creixell beaches, Bay of Biscay | Sledge | 132 | Up to 10 | N/A | 5 | 0.1–96.9 indv/ 5 m² | San Vicente and Sorbe, 2001 |
| Portuguese coast | Sledge | 5 | 21–299 | 24 | nd | 14–61 indv/100 m² | Cunha et al., 1997 |
| Beagle Channel, Argentinian coast | Sledge, dredge | 18 | 25–665 | 15,662 | 25 | nd | Mühlenhardt-Siegel, 1999 |
| South Shetland Island, Trinity Islands | Sledge | 24 | 45–649 | 1,236 | 25 | 1–289 | Corbera, 2000 |
| Bellingshausen Sea, Antarctic Peninsula | Sledge | 26 | 85–1,870 | 557 | 35 | 4.2–652.2 indv/1000 m² | Corbera et al., 2008; San Vicente et al., 2009 |
| Mediterraneaean Sea | Sledge | 27 | 100–4,000 | 1,505 | 33 | nd | Reyss, 1973 |
| Faïland Island | Sledge | 3 | 103–202 | 8,074 | 13 | nd | Doi et al., 2020 |
| Kostarrenkala area, Bay of Biscay | Sledge | 10 | 175–1,000 | 1,476 | 37 | nd | Frutos and Sorbe, 2014 |
| NE Greenland | Sledge | 8 | 197–2,681 | 7,888 | 24 | nd | Brandt, 1997 |
| E Mediterranean Sea, SW Balearic Island, Algerian Basin, | Sledge and bottom trawl | 6 sledges, 12 trawls | 249–1,620 | 7,888 | 24 | nd | Cartes et al., 2003 |
| Cap Ferret Canyon, Bay of Biscay | Sledge | 13 | 346–1,099 | 1,885 | 42 | 2.8–55.8 indv/100 m² | Corbera and Sorbe, 2020 |
| Cap Ferret Canyon, Bay of Biscay | Sledge | 12 | 386–420 | 472 | 9 | 2.1–32.2 indv/100 m² | Sorbe and Elizáide, 2014 |
| E Mediterranean Sea, Catalan Sea | Sledge | 21 | 389–1,859 | 2,747 | 32 | nd | Cartes and Sorbe, 1997 |
| Capbreton area, Bay of Biscay | Sledge | nd | 500–797 | N/A | N/A | nd | Frutos and Sorbe, 2017 |
| Capbreton canyon (site A and B), Bay of Biscay | Sledge and box corer | 17 box corer and 17 sledges | A: 923–1,002, B: 971–1,027 | N/A | A: sledges: 8 species, box corer - 2 species B: sledges: 16 species, box: 4 species | nd | Marquiegui and Sorbe, 1999 |
(Continued)
TABLE 6 | Continued
| Area | Gear | Total number of samples logic | Depth [m] | Number of individuals | Number of species | Abundance | References |
|------|------|--------------------------------|---------|----------------------|------------------|----------|------------|
| Angola Basin | Sledge | 7 | 5,125–5,415 | 479 | 45 | nd | Brandt, 2005 |
| E Mediterranean Sea, Catalan Sea | Trawl with net and sledge | 35 | 398–1,808 | 3,159 | | Upper slope - 5 the most abundant species; middle slope - 6 most abundant, lower slope - 7 | Cartes and Sorbe, 1997 |
| W Mediterranean Sea, coast of Israel | Trawl | nd | 1,241–1,557 | 575 | 12 | nd | Corbera and Gall, 2016 |
| Ría de Pontevedra, Galicia coast | van Veen grab (0.056 m²) | 135 | Subtidal | 473 (2.7% of collected peracarids) | 14 | nd | Lourido et al., 2008 |
| Ría de Vigo, Galicia coast | van Veen grab (0.056 m²) | 145 | 0–28.2 | Nd | 4 | nd | Cacabelos et al., 2010 |
| SE Brazilian continental shelf | van Veen grab (0.1 m²), dredge and beam-trawl | 108 samples | 10–124 | 1,587 | 19 | nd | Dos Santos and Pires-Vanin, 1999 |
| Mexico, Bay of All Saints | van Veen grab (0.1 m²) | 60 | <15 | Nd | 12 | 1–124 indiv/0.1m² | Donath-Hernández, 1987 |
| E Mediterranean Sea, Bay of Biarres | van Veen grab (600 cm² = 0.06 m²) | nd | 15 | Nd | 10 | Max 333 indiv/m² | Corbera et al., 2013 |
| Persian Gulf, Iranian coast | van Veen grab (0.1 m²) | 15 | up to 30 | 232 | 8 | nd | Martin et al., 2010 |
| Admiralty Bay, Antarctic | van-Veen grab (0.1 m²) | 105 | 20–500 | 685 | 11 | nd | Pabis and Błazewicz-Paszkowycz, 2011 |
| Mobile Bay, Alabama, Gulf of Mexico | nd | 3,150 | 2.5–6 | Nd | 5 | Up to: 69 indiv/m² for Oxuurostylissmithi, 11 indiv/m² for Leucon americanus, 6 indiv/m² for Cyclaspisvarians and for Eudorella monodon | Modlin and Dardeau, 1987 |
A description of ecological patterns (Sarthou et al., 2010), although they have to be treated cautiously.
Nevertheless, it is, to some point, surprising that we have noticed two peaks in the general number of species, one in the shallows and one in the 1,000 m (Table 2), where the influence of Ba and hydrocarbons was the highest. Moreover, the cumacean fauna recorded at 1,000 m stations was also the most unique. Those facts might be associated with the general pattern showing that bathyal is the main biodiversity hot spot for benthic fauna due to higher habitat diversity (Danovaro et al., 2009; Rex and Etter, 2010). High diversity of bathyal cumacean communities was already demonstrated in many previous studies (e.g., Corbera, 1995; Gage et al., 2004; Corbera and Sorbe, 2020 and citations therein).
Distribution of genera and/or families along a depth gradient might also be at least partially explained by earlier studies on cumacean evolution and phylogeny, although we also know very little about those important problems (Gerken, 2018 and references therein). There are only scarce data about the possible origin of various families or genera and their affinities to given depth zones or regions. For example, Bodotriidae are classified as typical shallow water crustaceans (Day, 1978; Mühlénhardt-Siegel, 1996; Petrescu, 1998), while members of Leuconidae are classified as typical shallow water crustaceans (Day, 1978; Mühlénhardt-Siegel, 1996; Petrescu, 1998), while members of Leuconidae are classified as typical shallow water crustaceans (Day, 1978; Mühlénhardt-Siegel, 1996; Petrescu, 1998), while members of Leuconidae are classified as typical shallow water crustaceans (Day, 1978; Mühlénhardt-Siegel, 1996; Petrescu, 1998).
preferred sediment type might extend the bathymetric range of cumaceans, recognized as shallow water, even to upper bathyal depths (Corbera and Sorbe, 2020), demonstrating that similar generalizations are still far from being conclusive. Brandt et al. (2012) summarized the information about widely distributed peracarid crustaceans. According to this analysis, there are at least 48 eurybathic cumacean species in the deep sea, and at that least 25 have a very wide geographic distribution (two or more oceans), although we have to remember that those numbers could substantially change after detailed molecular studies.
The results suggest high level of undescribed cumacean diversity in West African waters. Future biodiversity studies should be focused on benthal communities, especially in areas not affected by human related disturbance processes, and explore a wider depth range. The use of dredges or epibenthic sledge could also allow to collect a larger number of individuals than point scale samplers such as the van Veen grab. Probably, the most appropriate sampling strategy should include the use of both quantitative and semiquantitative methods, as it was already demonstrated in case of tanaidaceans (Żoziwiak et al., 2020). The hypothesis of the high diversity of cumacean fauna in tropical African waters still cannot be verified because of strong sampling bias. The great rarity, small population densities, and high level of patchiness in the distribution of particular species suggest the necessity of sampling at larger number of stations, allowing for more comprehensive biodiversity inventory of those small crustaceans. The high diversity of Cumacea observed in this study showed that small peracarids should be included in future research, especially since the pressure of human activities in large marine ecosystems such as the Gulf of Guinea could lead to substantial loss in marine diversity yet unknown. There is also a great need for further taxonomic studies on the region. They could help to accelerate the further analysis of ecological interactions occurring in West African seabed ecosystems, because they constitute an important base of any ecological research and biodiversity inventories.
DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.
AUTHOR CONTRIBUTIONS
AS: concept of the paper, taxonomic identification, statistical analysis, and manuscript writing. KP: concept of the paper, statistical analysis, writing of the manuscript, sampling design, and coordination of benthic ecology in the project. RS: statistical analysis. BS: sampling design, proofreading, and head of the FAO project. All authors contributed to the article and approved the submitted version.
FUNDING
The sampling cruise and the environmental data analysis within this program were funded by the Norwegian Agency for Development Cooperation (NORAD) (Oil for Development Program—OFD) and by the Food and Agriculture Organization of the United Nations (FAO). AS and KP were also supported by University of Lodz.
ACKNOWLEDGMENTS
The authors want to thank Dr. Ute Mühlenhardt-Siegel for helping with identifying species.
SUPPLEMENTARY MATERIAL
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars.2021.703547/full#supplementary-material
REFERENCES
Acquah, P. (1995). Natural Resources Management and Sustainable Development: The Case of the Gold Sector in Ghana. Geneva: United Nations Conference on Trade and Development (UNCTAD).
Alfonso, M. I., Bandera, M. E., López-González, P. J., and Garcia-Gómez, J. C. (1998). The cumacean community associated with a seaweed as a bioindicator of environmental conditions in the Algeciras Bay (Strait of Gibraltar). Cah. Biol. Mar. 39, 197–205.
Allen, R. C., and Cochran, J. K. (2019). The critical role of bioturbation for particle dynamics, priming potential, and organic C remineralization in marine sediments: local and basin scales. Front. Earth Sci. 7:157. doi: 10.3389/feart.2019.00157
Appelants, W., Ahyong, S. T., Anderson, G., Angel, M. V., Artois, T., Bailly, N., et al. (2012). The magnitude of global marine species diversity. Cur. Biol. 22, 2189–2202. doi: 10.1016/j.cub.2012.09.036
Atçık, A. S., Katagan, T., Sezgin, M., Özdeik, H. G., Berber, S., and Bulut, M. (2014). The effects of some domestic pollutants on the cumacean (Crustacea) community structure at the coastal waters of the Dardanelles, Turkey. Arthropods 3, 27–42. Available online at: http://www.iaees.org/publications/journals/arthropods/onlineversion.asp
Ayamdo, N. A. (2016). Protecting the Gulf of Guinea in an oil boom: regulating offshore 618petroleum pollution in a divided world. J. World Energy Law Bus. 619, 219–232. doi: 10.1093/jwel/jww007
Băcescu, M. (1961). Contribution à la connaissance du genre Makroklindrus Stebbing (Crustacea: Cumacea). Rev. Biol. 6, 325–333.
Băcescu, M. (1972). Archaeoecuma and Schizocuma, new genera of Cumacea from the American tropical waters. Rev. Roum. Biol. Academia RepubliciiSocialiste Romania 17, 241–250.
Băcescu, M., and Petrescu, I. (1991). New Cumacea (Crustacea, Peracarida) from the littoral waters of Brazil. Trav. Mus. Natl. Hist. Nat. 31, 327–340.
Bamikole, W., Ndubuisi, A., Ochuko, A. P., and Olamide-Olaronke, O. P. (2009). Macrobrachic crustaceans in the littoral waters of Lagos lagoon, Nigeria. Res. J. Biol. Sci. 4, 272–276. Available online at: https://medwelljournals.com/abstract/?doi=rjbsci.2009.272.276
Bartoń, K. (2018). MuMIn: Multi-model inference. R package version 1.42.1. Available online at: https://CRAN.R-project.org/package=MuMIn (accessed January 15, 2021).
Bassindale, R. (1961). On the marine fauna of Ghana. Proc. Zool. Soc. Lond. 137, 481–510. doi: 10.1111/j.1469-7998.1961.tb06086.x
Binet, D., and Marchal, E. (1993). “The large marine ecosystem of shelf areas in the Gulf of Guinea: long-term variability induced by climatic changes,” in Large
description of six new species. *Beaufortia* 50, 197–222. Available online at: https://repository.naturalis.nl/pub/504899
Mühlendarft-Siegell, U. (2005). Cumacea species (Crustacea: Peracarida) from the Deep-Sea Expedition DIVA-1 with RV "Meteor" to the Angola Basin in July 2000. Families Lampropidae, Bodotiidae. Org. Divers. Evol. 5, 113–130. doi: 10.1007/BF02358098
Mühlendarft-Siegell, U. (2009). Cumacea (Crustacea, Peracarida) in the deep Mediterranean, with the description of one new species. Zootaxa 2096, 413–432. doi: 10.11646/zootaxa.2096.1.25
Mühlendarft-Siegell, U. (2011). New and known species of the family Leuconeidae (Cumacea, Peracarida) from Antarctic deep-sea basins. Zootaxa 3117, 1–68. doi: 10.11646/zootaxa.3117.1.1
Neff, J. M. (2002). Bioaccumulation in Marine Organisms: Effect of Contaminants From Oil Well Produced Water. Amsterdam: Elsevier. doi: 10.1016/B978-008043716-3/50002-6
Nieto, K., and Mélun, F. (2017). Variability of chlorophyll-a concentration in the Gulf of Guinea and its relation to physical oceanographic variables. *Prog. Oceanogr.* 151, 97–115. doi: 10.1016/j.pocean.2016.11.009
Oksanen, J., Blanche, F. G., Friendy, M., Kindt, R., Legendt, P., McGlinn, D., et al. (2019). *Vegan: Community Ecology Package. R package version 2. 5-4*. Available online at: https://CRAN.R-project.org/package=vegan (accessed November 28, 2020).
Olsgard, F., and Gray, J. S. (1995). A comprehensive analysis of *"GrigoreAntipa" Museum of Natural History from the coasts of Tanzania* Part 1. Family Bodotriidae. *Trav. Mus. Natl. Hist. Nat. GrigoreAntipa* 40, 227–310.
Pabis, K., and Błazewicz-Paszkowycz, M. (2011). Distribution and diversity of *San Vicente*, C., Munilla, T., Corbera, J., Sorbe, J. C., and Ramos, A. (2009). Suprabenthic fauna from the Bellingshausen Sea and western Antarctic Peninsula: spatial distribution and community structure. *Sci. Mar.* 73, 357–368. doi: 10.3989/scimar.2009.73n2357
San Vicente, C., and Sorbe, J. C. (1999). Spatio-temporal structure of the suprabenthic community from Creel shelf (western Mediterranean). *Acta Oecol.* 20, 377–389. doi: 10.1016/S1146-697X(99)0129-0
San Vicente, C., and Sorbe, J. C. (2001). Temporal changes in the structure of the suprabenthic community from Hendaya beach (southern Bay of Biscay): a comparison with a northwesternmediterranean beach community. *Bol. Inst. Esp. Oceanogr.* 17, 107–120. Available online at: https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.537.6244&rep=rep1&type=pdf
Sarthou, C., Larpin, D., Fonty, E., Pavoine, S., and andPonge, J.-F. (2010). Stability of plant communities along a tropical inselberg ecotone in French Guiana (South America). *Flora* 205, 682–694. doi: 10.1016/j.flora.2010.04.005
Scheren, P. A., Ibe, A. C., Janssen, F. J., and Lemmens, A. M. (2002). Environmental pollution in the Gulf of Guinea – a regional approach. *Mar. Pollut. Bull.* 44, 633–641. doi: 10.1016/S0025-326X(01)0305-8
Sebens, K. P. (1991). "Habitat structure and community dynamics in marine benthic systems” in Habitat Structure: The physical arrangement of objects in space,” eds S. S. Bell, E. D. McCoy, and H. R. Mushinsky (Dordrecht: Springer Netherlands), 211–234. doi: 10.1007/978-94-011-3076-9_11
Sobczyk, R., Czrotek, P., Serigstad, B., and Pabis, K. (2021). Modelling of polychaete functional diversity: a Large Marine Ecosystem response to multiple natural factors and human impacts on the West African continental margin. *Sci. Total Environ.* 792:148075. doi: 10.1016/j.scitotenv.2021.148075
Sorbe, J. C., and Elizalde, M. (2014). Temporal changes in the structure of a slope suprabenthic community from the Bay of Biscay (NE Atlantic Ocean). *Deep Sea Res. II Top. Stud. Oceanogr.* 106, 179–191. doi: 10.1016/j.dsr2.2013.09.041
Spalding, M. D., Fox, H. E., Allen, G. R., Davidson, N., Ferdaña, Z. A. , Finlayson, M., et al. (2007). Marine ecoregions of the world: a biogeoregionalization of coastal and shelf areas. *Biogeosciences* 5, 573–583. doi: 10.1516/B5 7007
Stark, J. S., Mohammad, M., McMinn, A., and Ingels, J. (2020). Diversity, abundance, spatial variation, and human impacts in marine miobenthic Nematoide and Copepod communities at Casey Station, East Antarctica. *Front. Mar. Sci.* 7:480. doi: 10.3389/fmars.2020.00480
StatSoft (2006). *Elektroniczny Podrecznik Statystyki PL*. Krakow
Thiel, M., González, E. R., Balanda, M. J., Haye, P., and Watling, L. (2003). "Diversity of Chilean peracarids (Crustacea: Malacostraca)," in *Contributiones al Estudio de los Crustaceos del Pacifico Este*, ed M. E. Hendrickx (Mexico: Instituto de Ciencias del Mar y Limnología, UNAM), 177–189.
Ukwe, C. N., Ibe, C. A., Alo, B. I., and Yumkella, K. K. (2003). Achieving a paradigm shift in environmental and living resources management in the Gulf of Guinea: the large marine ecosystem approach. *Mar. Pollut. Bull.* 33, 103–107. doi: 10.1016/S0025-326X(94)00258-B
Wägele, J. W. (2005). *Foundations of Phylogenetic Systematics*. München: Verlag Dr. Friedrich Pfeil.
Walsh, C., and Mac Nally, R. (2013). *hier.part: Hierarchical Partitioning. R package version 1.0-4*. Available online at: https://CRAN.R-project.org/package=hier.part (accessed March 10, 2020).
Wang, Z., and Dauvin, J. C. (1994). The suprabenthic crustacean fauna of the infralittoral fine sand community from the Bay of Seine (Eastern English Channel): composition, swimming activity and diurnal variation. *Cah. Biol. Mar.* 35, 135–155.
Watling, L., and Gerken, S. (2021). *World Cumacea Database*. Available online at: http://www.marinespecies.org/cumaceaon~2021-04-27 (accessed July 7, 2020).
Wei, T., and Simko, V. (2017). *R Package “Corrplot”: Visualization of a Correlation Matrix (Version 0.84)*. Available online at: https://guthub.com/taiyun/corrplot (accessed February 12, 2020).
Wieser, W. (1956). Factors influencing the choice of substratum in *Cumella vulgaris* Hart (Crustacea, Cumacea). *Limnol. Oceanogr.* 1, 274–285. doi: 10.4319/lo.1956.1.4.0274
Witman, J. D., Etter, R. J., and Smith, F. (2004). The relationship between regional and local species diversity in marine benthic communities: a global perspective. *Proc. Natl. Acad. Sci. U. S. A.* 101, 15664–15669. doi: 10.1073/pnas.0404300101
**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
**Publisher's Note:** All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Copyright © 2021 Stępień, Pabis, Sobczyk and Serigstad. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2755,
1
],
[
2755,
9627,
2
],
[
9627,
15822,
3
],
[
15822,
18821,
4
],
[
18821,
22666,
5
],
[
22666,
27508,
6
],
[
27508,
30819,
7
],
[
30819,
34711,
8
],
[
34711,
37028,
9
],
[
37028,
41221,
10
],
[
41221,
43379,
11
],
[
43379,
49456,
12
],
[
49456,
50281,
13
],
[
50281,
52746,
14
],
[
52746,
58632,
15
],
[
58632,
62203,
16
],
[
62203,
68290,
17
],
[
68290,
68290,
18
],
[
68290,
68290,
19
],
[
68290,
73317,
20
],
[
73317,
75612,
21
]
]
} |
5ca916f2346d1f068c62feaf25e71fd0ba5cc807 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 25,
"total-fallback-pages": 0,
"total-input-tokens": 82766,
"total-output-tokens": 18536,
"fieldofstudy": [
"Physics"
]
} | Lifetime of metastable states and suppression of noise in Interdisciplinary Physical Models
B. Spagnolo\textsuperscript{a}, A. A. Dubkov\textsuperscript{b}, A. L. Pankratov\textsuperscript{c}, E. V. Pankratova\textsuperscript{d}, A. Fiasconaro\textsuperscript{a,c,f}, A. Ochab-Marcinek\textsuperscript{e}
\textsuperscript{a}Dipartimento di Fisica e Tecnologie Relative, Università di Palermo CNISM - Unità di Palermo, Group of Interdisciplinary Physics\textsuperscript{f} Viale delle Scienze, I-90128 Palermo, Italy
\textsuperscript{b}Radiophysics Department, Nizhny Novgorod State University, 23 Gagarin ave., 603950 Nizhny Novgorod, Russia
\textsuperscript{c}Institute for Physics of Microstructures of RAS, GSP-105, Nizhny Novgorod, 603950, Russia
\textsuperscript{d}Mathematical Department, Volga State Academy, Nesterov street 5, Nizhny Novgorod, 603600, Russia
\textsuperscript{e}Marian Smoluchowski Institute of Physics, Jagellonian University Reymonta 4, 30059 Krakw, Poland
\textsuperscript{f}Mark Kac Center for Complex Systems Research, Jagellonian University Reymonta 4, 30059 Krakw, Poland
Transient properties of different physical systems with metastable states perturbed by external white noise have been investigated. Two noise-induced phenomena, namely the noise enhanced stability and the resonant activation, are theoretically predicted in a piece-wise linear fluctuating potential with a metastable state. The enhancement of the lifetime of metastable states due to the noise, and the suppression of noise through resonant activation phenomenon will be reviewed in models of interdisciplinary physics: (i) dynamics of an overdamped Josephson junction; (ii) transient regime of the noisy FitzHugh-Nagumo model; (iii) population dynamics.
PACS numbers: 05.10.-a, 05.40.-a, 87.23.Cc
\textsuperscript{*} Presented at the 19\textsuperscript{th} Marian Smoluchowski Symposium on Statistical Physics, Kraków, Poland, May 14-17, 2006
\textsuperscript{†} e-mail: [email protected]
\textsuperscript{‡} http://gip.dft.unipa.it
1. Introduction
Metastability is a generic feature of many nonlinear systems, and the problem of the lifetime of metastable states involves fundamental aspects of nonequilibrium statistical mechanics. Nonequilibrium systems are usually open systems which strongly interact with environment through exchanging materials and energy, which can be modelled as noise. The investigation of noise-induced phenomena in far from equilibrium systems is one of the approaches used to understand the behaviour of physical and biological complex systems. Specifically, the relaxation in many natural complex systems proceeds through metastable states and this transient behaviour is observed in condensed matter physics and in different other fields, such as cosmology, chemical kinetics, biology and high energy physics [1]-[9]. In spite of such ubiquity, the microscopic understanding of metastability still raises fundamental questions, such as those related to the fluctuation-dissipation theorem in transient dynamics [10].
Recently, the investigation of the thermal activated escape in systems with fluctuating metastable states has led to the discovery of resonancelike phenomena, characterized by a nonmonotonic behavior of the lifetime of the metastable state as a function of the noise intensity or the driving frequency. Among these we recall two of them, namely the resonant activation (RA) phenomenon [11]-[21], whose signature is a minimum of the lifetime of the metastable state as a function of a driving frequency, and the noise enhanced stability (NES) [3], [17]-[27]. This resonancelike effect, which contradicts the monotonic behavior predicted by the Kramers formula [28, 29], shows that the noise can modify the stability of the system by enhancing the lifetime of the metastable state with respect to the deterministic decay time. Specifically when a Brownian particle is moving in a potential profile with a fluctuating metastable state, the NES effect is always obtained, regardless of the unstable initial position of the particle. Two different dynamical regimes occur. These are characterized by: (i) a monotonic behavior with a divergence of the lifetime of the metastable state when the noise intensity tends to zero, for a given range of unstable initial conditions (see for detail Ref. [24]), which means that the Brownian particle will be trapped into the metastable state in the limit of very small noise intensities; (ii) a nonmonotonic behavior of the lifetime of the metastable state as a function of noise intensity. The noise enhanced stability effect implies that, under the action of additive noise, a system remains in the metastable state for a longer time than in the deterministic case, and the escape time has a maximum as a function of noise intensity. We can lengthen or shorten the mean lifetime of the metastable state of our physical system, by acting on the white noise intensity. The noise-induced stabilization, the noise induced slowing
down in a periodical potential, the noise induced order in one-dimensional map of the Belousov-Zhabotinsky reaction, and the transient properties of a bistable kinetic system driven by two correlated noises, are akin to the NES phenomenon [22].
In this paper we will review these two noise-induced effects in models of interdisciplinary physics, ranging from condensed matter physics to biophysics. Specifically in the first section, after shortly reviewing the theoretical results obtained with a model based on a piece-wise linear fluctuating potential with a metastable state, we focus on the noise-induced effects RA and NES. In the next sections, we show how the enhancement of the lifetime of metastable states due to the noise and the suppression of noise, through resonant activation phenomenon, occur in the following interdisciplinary physics models: (i) dynamics of an overdamped Josephson junction; (ii) transient regime of the noisy FitzHugh-Nagumo model; (iii) population dynamics.
2. The model
As an archetypal model for systems with a metastable state and strongly coupled with the noisy environment, we consider the one-dimensional overdamped Brownian motion in a fluctuating potential profile
$$\frac{dx}{dt} = -\frac{\partial [U(x) + V(x) \eta(t)]}{\partial x} + \xi(t),$$
where $x(t)$ is the displacement of the Brownian particle and $\xi(t)$ is the white Gaussian noise with the usual statistical properties: $\langle \xi(t) \rangle = 0$, $\langle \xi(t) \xi(t') \rangle = 2D \delta(t-t')$. The variable $\eta(t)$ is the Markovian dichotomous noise, which takes the values $\pm 1$ with the mean flipping rate $\nu$. The potential profile $U(x) + V(x)$ corresponds to a metastable state, and $U(x) - V(x)$ corresponds to an unstable one, with a reflecting boundary at $x \to -\infty$ and an absorbing boundary at $x \to +\infty$ (see Fig. 1). Starting from the well-known expression for the probability density of the process $x(t)$
$$P(x,t) = \langle \delta(x-x(t)) \rangle$$
and using the auxiliary function $Q(x,t)$
$$Q(x,t) = \langle \eta(t) \delta(x-x(t)) \rangle,$$
and the Eq. (1), we obtain the set of differential equations
$$\frac{\partial P(x,t)}{\partial t} = \frac{\partial}{\partial x} \left[ U'(x) P + V'(x) Q \right] + D \frac{\partial^2 P}{\partial x^2},$$
The average lifetime of Brownian particles in the interval \((L_1, L_2)\), with the initial conditions \(P(x, 0) = \delta(x - x_0)\) and \(Q(x, 0) = \pm \delta(x - x_0)\), is
\[
\tau(x_0) = \int_0^\infty dt \int_{L_1}^{L_2} P(x, t| x_0, 0) \, dx = \int_{L_1}^{L_2} Y(x, x_0, 0) \, dx, \tag{5}
\]
where \(Y(x, x_0, s)\) is the Laplace transform of the conditional probability density \(P(x, t| x_0, 0)\). After Laplace transforming Eqs. (4), with above-mentioned initial conditions and using the method proposed in Refs. [25, 26, 30], we can express the lifetime \(\tau(x_0)\) as
\[
\tau(x_0) = \int_{L_1}^{L_2} Z_1(x, x_0) \, dx, \tag{6}
\]
where \(Z_1(x, x_0)\) is the linear coefficient of the expansion of the function \(sY(x, x_0, s)\) in a power series in \(s\). By Laplace transforming the auxiliary function \(Q(x, t)\) in \(R(x, x_0, s)\), and expanding the function \(sR(x, x_0, s)\) in similar power series, we obtain the following closed set of integro-differential equations for the functions \(Z_1(x, x_0)\) and \(R_1(x, x_0)\)[17]
\[
DZ'_1 + U'(x) Z_1 + V'(x) R_1 = -\theta(x - x_0),
\]
\[
DR'_1 + U'R_1 + V'Z_1 = 2\nu \int_{-\infty}^{x} R_1 dy \mp \theta(x - x_0), \tag{7}
\]
where \(\theta(x - x_0)\) is the Heaviside step function, and \(R_1(x, x_0)\) is the linear coefficient of the expansion of the function \(sR(x, x_0, s)\). We put equal to zero the probability flow at the reflecting boundary \(x = -\infty\). These general
equations (6) and (7) allow to calculate the average lifetime for potential profiles with metastable states. We may consider two mean lifetimes $\tau_+(x_0)$ and $\tau_-(x_0)$, depending on the initial configuration of the randomly switching potential profile: $U(x) + V(x)$ or $U(x) - V(x)$. The average lifetime (6) is equal to $\tau_+(x_0)$, when we take the sign "−" in the second equation of system (7), and vice versa for $\tau_-(x_0)$.
We consider now the following piece-wise linear potential profile
$$U(x) = \begin{cases} +\infty, & x < 0 \\ 0, & 0 \leq x \leq L \\ k(L-x), & x > L \end{cases} \quad (8)$$
and $V(x) = ax \quad (x > 0, \quad 0 < a < k)$. Here we consider the interval $L_1 = 0, L_2 = b$, with $b > L$. After solving the differential equations (7) with the potential profile (8), and by choosing the initial position of Brownian particles at $x = 0$, we get the exact mean lifetime
$$\tau_- (x_{in} = 0) = \frac{b}{k} + \frac{\nu L^2}{\Gamma^2} + \frac{a}{2\nu \Gamma^4} f(D, \Gamma, \nu), \quad (9)$$
where $\Gamma = \sqrt{a^2 + 2\nu D}$. Here $f(D, \Gamma, \nu)$, which is also a function of the potential parameters $a, b, L, k$, has a complicated expression in terms of parameters $D, \Gamma$ and $\nu$ [17]. It is worthwhile to note that Eq. (9) was derived without any assumptions on the white noise intensity $D$ and on the mean rate of flippings $\nu$ of the potential.
To look for the NES effect, which is observable at very small noise intensity [3, 23], we derive the average life time in the limit $D \to 0$
$$\tau_- (x_{in} = 0) = \tau_0 + \frac{D}{a^2} g(q, \omega, s) + o(D), \quad (10)$$
where $\omega = \nu L/k, \quad q = a/k$ and $s = 2\omega (b/L - 1) / (1 - q^2)$ are dimensionless parameters. Here
$$g(q, \omega, s) = \frac{3q^2 + 4q - 5}{2(1 - q^2)} + 2\omega \frac{3q^2 + q - 3}{q(1 - q^2)} \frac{2\omega^2}{q^2} + se^{-s} \frac{q^3(1 + q^2)}{(1 + q)(1 - q^2)} + (1 - e^{-s}) \frac{q(1 - q^2 - 2q^3)}{2(1 - q^2)} \quad (11)$$
and
$$\tau_0 = \frac{2L}{a} + \frac{\nu L^2}{a^2} + \frac{b - L}{k} - \frac{q(1 - q)}{2\nu} \left(1 - e^{-s}\right) \quad (12)$$
is the mean lifetime in the absence of white Gaussian noise \((D = 0)\). The condition to observe the NES effect can be expressed by the inequality
\[
g(q, \omega, s) > 0. \tag{13}
\]
The main conclusions from the analysis of the inequality \((13)\) are: (i) the NES effect occurs at \(q \approx 1\), i.e. at very small steepness \(k - a = k(1 - q)\) of the reverse potential barrier for the metastable state: for this potential profile, a small noise intensity can return particles into potential well, after they crossed the point \(L\); (ii) for a fixed mean flipping rate, the NES effect increases when \(q \to 1\), and (iii) for fixed parameter \(q\) the effect increases when \(\omega \to 0\), because Brownian particles have enough time to move back into potential well.
Under very large noise intensity \(D\), the Brownian particles "do not see" the fine structure of potential profile and move as in the fixed potential \(U(x) = -kx\). Therefore the average life time decreases with noise intensity, tending to the value \(b/k\) as follows from Eq. \((9)\) in the limit \(D \to \infty\). In Fig. 2 we show the plots of the normalized mean lifetime \(\tau_-(x_{in} = 0)/\tau_0\), Eq. \((4)\), as a function of the noise intensity \(D\) for three values of the dimensionless mean flipping rate \(\omega = \nu L/k\): 0.03, 0.01, 0.005. The values of the parameters of the potential profile are: \(L = k = 1, \ a = 0.995, \ b = 2\). The maximum value of the average lifetime and the range of noise intensity values, where NES effect occurs, increases when \(\omega\) decreases. By using exact Eq. \((9)\) we have also investigated the behaviour of the mean lifetime \(\tau_-(x_{in} = 0)\) as a function of switchings mean rate \(\nu\) of the potential profile. In Fig. 3 we plot this behaviour for six values of the noise intensity, namely: \(D = 0.08, 0.1, 0.13, 0.16, 0.2, 0.25\). At very slow flippings \((\nu \to 0)\) we obtain...
Fig. 3. Semilogarithmic plot of the mean lifetime $\tau_-(0)$ vs the dimensionless mean flipping rate $\omega = \nu L / k$ for seven noise intensity values. Specifically from top to bottom on the right side of the figure: $D = 0.08, 0.1, 0.13, 0.16, 0.2, 0.25$. The other parameters are the same as in Fig. 2.
$$\tau_-(\nu \to 0)(x_{in} = 0) \simeq \tau_d - \frac{D}{a^2} \left(1 - e^{-aL/D}\right),$$
(14)
i.e. the average lifetime of the fixed unstable potential $U(x) - ax$. Here $\tau_d = L/a + (b - L)/(k + a)$ is the deterministic time at zero frequency ($\nu = 0$). While for very fast switchings ($\nu \to \infty$) we obtain
$$\tau_-(\nu \to \infty)(x_{in} = 0) \simeq \frac{b}{k} + \frac{L^2}{2D},$$
(15)
i.e. the mean lifetime for average potential $U(x)$. All limiting values of $\tau_-(x_{in} = 0)$, expressed by Eqs. (14) and (15), are shown in Fig. 3. At intermediate rates the average escape time from the metastable state exhibits a minimum at $\omega = 0.1$, which is the signature of the resonant activation (RA) phenomenon [11]-[15]. If the potential fluctuations are very slow, the average escape time is equal to the average of the crossing times over upper and lower configurations of the barrier, and the slowest process determines the value of the average escape time [11]. In the limit of very fast fluctuations, the Brownian particle "sees" the average barrier and the average escape time is equal to the crossing time over the average barrier. In the intermediate regime, the crossing is strongly correlated with the potential fluctuations and the average escape time exhibits a minimum at a resonant fluctuation rate. Specifically, for $D \ll 1$ and the parameter values of the potential ($a = 0.995, L = k = 1, b = 2$), we obtain from Eqs. (14) and (15): $\tau_-(\nu \to 0)(x_{in} = 0) \simeq 1.5 - D/2$, and $\tau_-(\nu \to \infty)(x_{in} = 0) \simeq 2 + 1/(2D)$, that is
$$\tau_-(\nu \to 0)(x_{in} = 0) \ll \tau_-(\nu \to \infty)(x_{in} = 0)$$
(16)
which is consistent with the physical picture for which we have at zero frequency of switchings the unstable initial configuration of the potential (see Fig. 1), and at very fast switchings \((\nu \to \infty)\) the average configuration of the potential, which in our case has not barrier. For \(D \gg 1\), because
\[
\lim_{D \to \infty} D \left( 1 - e^{-a L / D} \right) = a L ,
\]
we have \(\tau_{-}(\nu \to 0)(x_{in} = 0) = b/(k + a) \approx 1\), and \(\tau_{-}(\nu \to \infty)(x_{in} = 0) = 2\). So, for the noise intensity values used in our calculations shown in Fig. 3, ranging from \(D = 0.08\) to \(D = 0.25\), the limiting values for the average lifetime are: \(\tau_{-}(\nu \to 0)(x_{in} = 0) \approx 1.38 \div 1.46\), and \(\tau_{-}(\nu \to \infty)(x_{in} = 0) \approx 4 \div 8\), which are consistent with the limiting values shown in Fig. 3, and evaluated directly from exact expression (9).
Moreover, in Fig. 3 a new resonance-like behaviour, is observed. The mean lifetime of the metastable state \(\tau_{-}(x_{in} = 0)\) exhibits a maximum, between the slow limit of potential fluctuations (static limit) and the RA minimum, as a function of the mean fluctuation rate of the potential, \(\omega\). This maximum occurs for a value of the barrier fluctuation rate on the order of the inverse of the time \(\tau_{up}(D)\) required to escape from the metastable fixed configuration
\[
\tau_{up}(D) = \frac{b - L}{k - a} - \frac{L}{a} + \frac{D \left( e^{a L / D} - 1 \right)}{a^2 (1 - q)} .
\]
Specifically we observe that this maximum increases with decreasing noise intensity \(D\) and at the same time the position of the maximum is shifted towards lower values of the dimensionless mean flipping rate \(\omega\). In fact from Eq. (18) we have that the average time required to escape from the metastable fixed configuration \(\tau_{up}\) increases, consequently the corresponding rate of the barrier fluctuations \(\omega_{max} \approx 1/\tau_{up}(D)\) decreases, as shown in Fig. 3. We can also estimate the value of the maximum \((\tau_{max}(x_{in} = 0))\) and its position \((\omega_{max})\), by expanding Eqs. (10)-(12) in a power series up to the second order in \(\omega\). Using the same parameter values of the potential we have: \(\omega = \nu, s = 2\nu/(1 - q^2), se^{-s} \approx s - s^2,\) and \(1 - e^{-s} \approx s - s^2/2\). We obtain finally:
\[
\tau_{-}(x_{in} = 0) \approx 2.5 + (98.7)D + \omega[51 + (347.4)D] - (2 \times 10^6)\omega^2 D .
\]
For \(D = 0.1\) we have: \(\tau_{max} \approx 12.3\) and \(\omega_{max} \approx 2 \times 10^{-4}\), which are an estimate of the coordinates of the maximum of the corresponding curve in Fig. 3. From small noise intensity \(D \to 0\), from Eq. (19), we obtain:
\[
\tau_{max} \approx 0.6 \times 10^{-3} + O(D) .
\]
The maximum of the average lifetime \(\tau_{max}\), therefore, increases when the noise intensity decreases as shown in Fig. 3.
This suggests that, the enhancement of stability of metastable state is strongly correlated with the potential fluctuations, when the Brownian particle "sees" the barrier of the metastable state [3, 17, 23]. When the average time to cross the barrier, that is the average lifetime of the metastable state, is approximately equal to the correlation time of the fluctuations of the potential barrier, a resonance-like phenomenon occurs. In other words, this new effect can be considered as a NES effect in the frequency domain. It is worthwhile to note that the new nonmonotonic behaviour shown in Fig. 3 is in good agreement with experimental results observed in a periodically driven Josephson junction (JJ) [20]. In this very recent paper the authors experimentally observe the coexistence of RA and NES phenomena. Specifically they found (see Fig.3 of the paper [20]) that the maximum increases with decreasing bias current and at the same time the position of the maximum is shifted towards lower values of $\omega$. A decrease of the bias current causes (see next section on transient dynamics of a JJ) a decrease of the slope of the potential profile, which corresponds to a decreasing parameter $k$ in our model (Eq. (8)). Therefore, the average lifetime maximum $\tau_{-\text{max}}$ increases and as a consequence the time required to escape from the metastable fixed configuration $\tau_{\text{up}}(D)$ increases too. Consequently, the corresponding rate of the barrier fluctuations $\omega_{\text{max}} \simeq 1/\tau_{\text{up}}(D)$ decreases, as observed experimentally. Of course a more detailed analysis of the JJ system as a function of the temperature, that is the noise intensity, should add more interesting results.
Finally we note that in the frequency range $\omega \in (10^{-5} \div 10^{-3})$, for fixed values of the mean flipping rate, an overlap occurs in the curves for different values of the noise intensity. A nonmonotonic behavior of $\tau_{-}(x_{\text{in}} = 0)$ as a function of the noise intensity is observed, as we expect in the transient dynamics of metastable states [3, 17, 23].
3. Transient dynamics in a Josephson junction
The investigation of thermal fluctuations and nonlinear properties of Josephson junctions (JJ) is very important owing to their broad applications in logic devices. Superconducting devices, in fact, are natural qubit candidates for quantum computing because they exhibit robust, macroscopic quantum behavior [31]. Recently, a lot of attention was devoted to Josephson logic devices with high damping because of their high-speed switching [18, 32]. The rapid single flux quantum logic (RSFQ), for example, is a superconductive digital technique in which the data are represented by the presence or absence of a flux quantum $\Phi_0 = h/2e$, in a cell which comprises Josephson junctions. The short voltage pulse corresponds to a single flux quantum moving across a Josephson junction, that is a $2\pi$ phase
flip. This short pulse is the unit of information. However the operating temperatures of the high-Tc superconductors lead to higher noise levels by increasing the probability of thermally-induced switching errors. Moreover during the propagation within the Josephson transmission line fluxon accumulates a time jitter. These noise-induced errors are one of the main constraints to obtain higher clock frequencies in RSFQ microprocessors [32].
In this section, after a short introduction with the basic formulas of the Josephson devices, the model used to study the dynamics of a short overdamped Josephson junction is described. The interplay of the noise-induced phenomena RA and NES on the temporal characteristics of the Josephson devices is discussed. The role played by these noise-induced effects, in the accumulation of timing errors in RSFQ logic devices, is analyzed.
The Josephson tunneling junction is made up of two superconductors, separated from each other by a thin layer of oxide. Starting from Schrödinger equation and the two-state approximation model [33], it is straightforward to obtain the Josephson equation
$$\frac{d\phi(t)}{dt} = \frac{2eV(t)}{\hbar},$$
(20)
where $\phi$ is the phase difference between the wave function for the left and right superconductors, $V(t)$ is the potential difference across the junction, $e$ is the electron charge, and $\hbar = \hbar/2\pi$ is the Planck’s constant. A small junction can be modelled by a resistance $R$ in parallel with a capacitance $C$, across which is connected a bias generator and a phase-dependent current generator, $I_0\sin\phi$, representing the Josephson supercurrent due to the Cooper pairs tunnelling through the junction. Since the junction operates at a temperature above absolute zero, there will be a white Gaussian noise current superimposed on the bias current. Therefore the dynamics of a short overdamped JJ, widely used in logic elements with high-speed switching and corresponding to a negligible capacitance $C$, is obtained from Eq. (20) and from the current continuity equation of the equivalent circuit of the Josephson junction. The resulting equation is the following Langevin equation
$$\omega \frac{d\phi(t)}{dt} = -\frac{du(\phi)}{d\phi} - i_F(t),$$
(21)
valid for $\beta \ll 1$, with $\beta = 2eI_cR^2C/\hbar$ the McCamber–Stewart parameter, $I_c$ the critical current, and $i_F(t) = I_F/I_c$, with $I_F$ the random component of the current. Here
\[u(\varphi, t) = 1 - \cos \varphi - i(t)\varphi, \quad \text{with} \quad i(t) = i_0 + f(t), \quad (22)\]
is the dimensionless potential profile (see Fig. 4), \(\varphi\) is the difference in the phases of the order parameter on opposite sides of the junction, \(f(t) = A \sin(\omega t)\) is the driving signal, \(i = \frac{I}{I_c}\), \(\omega_c = \frac{2eR_NI_c}{h}\) is the characteristic frequency of the JJ, and \(R_N\) is the normal state resistance (see Ref. [33]). When only thermal fluctuations are taken into account [33], the random current may be represented by the white Gaussian noise: \(\langle i_F(t) \rangle = 0\), \(\langle i_F(t)i_F(t + \tau)\rangle = \frac{2D}{\omega_c} \delta(\tau)\), where \(D = \frac{2e^2kT}{hI_c} = \frac{I_T}{I_c}\) is the dimensionless intensity of fluctuations, \(T\) is the temperature and \(k\) is the Boltzmann constant. The equation of motion Eq. (21) describes the overdamped motion of a Brownian particle moving in a washboard potential (see Fig. 4). A junction initially trapped in a zero-voltage state, with the particle localized in one of the potential wells, can escape out of the potential well by thermal fluctuations. The phase difference \(\varphi\) fluctuates around the equilibrium positions, minima of the potential \(u(\varphi)\), and randomly performs jumps of \(2\pi\) across the potential barrier towards a neighbor potential minimum. The resulting time phase variation produces a nonzero voltage across the junction with marked spikes. For a bias current less than the critical current \(I_c\), these metastable states correspond to "superconductive" states of the JJ. The mean time between two sequential jumps is the life time of the superconductive metastable state [25]. For an external current greater than \(I_c\), the JJ junction switches from the superconductive state to the resistive one and the phase difference slides down in the potential profile, which now...
has not equilibrium steady states. A Josephson voltage output will be generated in a later time. Such a time is the switching time, which is a random quantity. In the presence of thermal noise a Josephson voltage appears even if the current is less than the critical one \((i < 1)\), therefore we can identify the lifetime of the metastable states with the mean switching time \([18, 25]\).
For the description of our system, i.e. a single overdamped JJ with noise, we will use the Fokker-Planck equation for the probability density \(W(\varphi, t)\), which corresponds to the Langevin equation \([21]\):
\[
\frac{\partial W(\varphi, t)}{\partial t} = -\frac{\partial G(\varphi, t)}{\partial \varphi} = \omega_c \frac{\partial}{\partial \varphi} \left\{ \frac{d\mu(\varphi)}{d\varphi} W(\varphi, t) + D \frac{\partial W(\varphi, t)}{\partial \varphi} \right\}.
\]
The initial and boundary conditions of the probability density and of the probability current for the potential profile \([22]\) are as follows: \(W(\varphi, 0) = \delta(\varphi - \varphi_0)\), \(W(+\infty, t) = 0\), \(G(-\infty, t) = 0\). Let, initially, the JJ is biased by the current smaller than the critical one, that is \(i_0 < 1\), and the junction is in the superconductive state. The current pulse \(f(t)\), such that \(i(t) = i_0 + f(t) > 1\), switches the junction into the resistive state. An output voltage pulse will appear after a random switching time. We will calculate the mean value and the standard deviation of this quantity for two different periodic driving signals: (i) a dichotomous signal, and (ii) a sinusoidal one. We will consider different values of the bias current \(i_o\) and of signal amplitude \(A\). Depending on the values of \(i_o\) and \(A\), as well as values of signal frequency and noise intensity, two noise-induced effects may be observed, namely the resonant activation (RA) and the noise enhanced stability (NES). Specifically the RA effect was theoretically predicted in Ref. \([11]\) and experimentally observed in a tunnel diode \([13]\) and in underdamped Josephson tunnel junctions \([16, 20]\), and the NES effect was theoretically predicted in \([22, 23, 25]\) and experimentally observed in a tunnel diode \([3]\) and in an underdamped Josephson junction \([20]\). The RA and NES effects, however, have different role on the behavior of the temporal characteristics of the Josephson junction. They occur because of the presence of metastable states, in the periodic potential profile of the Josephson tunnel junction, and the thermal noise. Specifically, the RA phenomenon minimizes the switching time and therefore also the timing errors in RSFQ logic devices, while the NES phenomenon increases the mean switching time producing a negative effect \([18]\).
### 3.1. Temporal characteristics
Now we investigate the following temporal characteristics: the mean switching time (MST) and its standard deviation (SD) of the Josephson
junction described by Eq. (21). These quantities may be introduced as characteristic scales of the evolution of the probability $P(t) = \frac{\varphi_2}{\varphi_1} \int_{\varphi_1}^{\varphi_2} W(\varphi,t) d\varphi$, to find the phase within one period of the potential profile of Eq. (22). We choose therefore $\varphi_2 = \pi$, $\varphi_1 = -\pi$ and we put the initial distribution on the bottom of a potential well: $\varphi_0 = \arcsin(i_0)$. A widely used definition of such characteristic time scales is the integral relaxation time (see the paper by Malakhov and Pankratov in Ref [12]). Let us summarize shortly the results obtained in the case of dichotomous driving, $f(t) = \text{A} \text{sign} (\sin(\omega t))$. Both MST and its SD do not depend on the driving frequency below a certain cut-off frequency (approximately $0.2\omega_c$), above which the characteristics degrade. In the frequency range from 0 to $0.2\omega_c$, therefore, we can describe the effect of dichotomous driving by time characteristics in a constant potential. The exact analytical expression of the first two moments of the switching time are [18]
$$\tau_c(\varphi_0) = \frac{1}{D\omega_c} \left[ \int_{\varphi_0}^{\varphi_2} e^{u(y)} dx \int_{\varphi_1}^{\varphi_2} e^{-u(y)} d\varphi dx + \int_{\varphi_1}^{\varphi_2} e^{-u(y)} d\varphi \int_{\varphi_2}^{\infty} e^{u(y)} d\varphi \right], \quad (24)$$
and
$$\tau_2(\varphi_0) = \tau_c^2(\varphi_0) - \int_{\varphi_0}^{\varphi_2} e^{-u(y)/D} H(x) dx - H(\varphi_0) \int_{\varphi_1}^{\varphi_0} e^{-u(y)/D} dx, \quad (25)$$
where $H(x) = \frac{2}{(D\omega_c)^2} \int_x^{\infty} e^{u(z)/D} dz \int_y^{\varphi_2} e^{-u(y)/D} dy \int_{\varphi_0}^{\varphi_0} e^{u(z)/D} dz dy dv$. The asymptotic expressions of the MST and its standard deviation (SD), obtained in the small noise limit ($D \ll 1$), agree very well with computer simulations up to $D = 0.05$ [18]. Therefore, not only low temperature devices ($D \leq 0.001$), but also high temperature devices may be described by these expressions. If the noise intensity is rather large, the phenomenon of NES may be observed in our system: the MST increases with the noise intensity. Here we note that it is very important to consider this effect in the design of large arrays of RSFQ elements, operating at high frequencies. To neglect this noise-induced effect in such nonlinear devices it may lead to malfunctions due to the accumulation of errors.
Now let us consider the case of sinusoidal driving. The corresponding time characteristics may be derived using the modified adiabatic approximation [14, 18]
$$P(\varphi_0,t) = \exp \left\{ - \int_{0}^{t} \frac{1}{\tau_c(\varphi_0,t')} dt' \right\}, \quad (26)$$
with \( \tau_c(\varphi_0, t') \) given by Eq. (24), after inserting in this equation the time dependent potential profile \( u(\varphi, t) \) of Eq. (22). Using the relation \( \tau = \int_0^{+\infty} P(\varphi_0, t) \, dt \) we calculate the MST. We focus now on the current value \( i = 1.5, \) because \( i = 1.2 \) is too small for high frequency applications. In Fig. 5 the MST and its SD as a function of the driving frequency, for three values of the noise intensity \( (D = 0.02, 0.05, 0.5) \), for a bias current \( i_0 = 0.5, \) and \( A = 1 \) are shown. We note that, because \( \varphi_0 = \arcsin(i_0) \) depends on \( i_0 \), the switching time is larger for smaller \( i_0 \). However, great bias current values \( i_0 \), in the absence of driving, give rise to the reduction of the mean life time of superconductive state, i.e. to increasing storage errors (Eq. (24)). Therefore, there must be an optimal value of bias current \( i_0 \), giving minimal switching time and acceptably small storage errors. We observe the phenomenon of resonant activation: MST has a minimum as a function of driving frequency.
**Fig. 5.** The MST and its SD vs frequency for \( f(t) = A \sin(\omega t) \) (computer simulations) for three values of the noise intensity. Namely: Long-dashed line - \( D = 0.02 \), short-dashed line - \( D = 0.05 \), solid line - \( D = 0.5 \). The value of the bias current is \( i_0 = 0.5 \), and the total current is \( i = 1.5 \).
The approximation (26) works rather well below \( 0.1 \omega_c \), that is enough for practical applications (see the inset of Fig. 3 in ref. [18]). It is interesting to see that near the minimum the MST has a very weak dependence on the noise intensity (as it is clearly shown in the \( \tau \) behavior of Fig. 5 for three values of the noise intensity), i.e. in this signal frequency range the noise is effectively suppressed. This noise suppression is due to the resonant activation phenomenon: a minimum appears in the MST and SD, when the escape process is strongly correlated with the potential profile oscillations. A noise suppression effect, but due to the noise, is reported in Ref. [34]. We observe also the NES phenomenon. There is a frequency range in Fig. 5 around \( (0.2 \div 0.48)\omega_c \) for \( i_0 = 0.5 \), where the switching
Fig. 6. The MST vs noise intensity for \( f(t) = A \sin(\omega t) \) and for three values of the driving frequency. Namely: \( \omega = 0.1 \) (long-dashed line), \( \omega = 0.3 \) (short-dashed line), \( \omega = 0.4 \) (solid line). Inset: The standard deviation (SD) vs noise intensity for the same values of driving frequency \( \omega \).
Time increases with the noise intensity. To see in more detail this effect we report in Fig. 6 the MST \( \tau(D) \) and its SD \( \sigma(D) \) vs the noise intensity \( D \), for three fixed values of the driving frequency, namely: \( \omega = 0.1, 0.3, 0.4 \). Both quantities have nonmonotonic behaviour and the great values of \( \sigma(D) \) near the maximum of \( \tau(D) \) confirm that the only information on the MST is not sufficient to fully unravel the statistical properties of the NES effect [24]. A detailed analysis of the PDF of the lifetime during the transient dynamics is required. This is subject of a forthcoming paper. Simulations for different bias current values [18] show that the NES effect increases for smaller \( i_0 \) because the potential barrier disappears for a short time interval within the driving period \( T = 2\pi/\omega \) and the potential is more flat [3]. The noise, therefore, has more chances to prevent the phase to move down and the switching process is delayed. This effect may be avoided, if the operating frequency does not exceed \( 0.2 \omega_c \). Besides the SD and MST (see Fig. 5) have their minima in a short range of values of \( \omega \) [18]. Close location of minima of MST and its SD means that optimization of RSFQ circuit for fast operation will simultaneously lead to minimization of timing errors in the circuit.
4. Dynamics of a FHN stochastic model
4.1. Suppression of noise and noise-enhanced stability effect
Case I. Let us fix the value of the noise intensity and analyze. The analysis of the stochastic properties of neural systems is of particular importance since it plays an important role in signal transmission [10], [35–41]. Bio-
logically realistic models of the nerve cells, such as widely-known Hodgkin-Huxley (HH) system [35], are so complex that they provide little intuitive insight into the neuron dynamics that they simulate. The FitzHugh-Nagumo (FHN) model, however, which is one of the simplified modifications of HH, is more preferable for investigation [41]. Nevertheless many effects observed in neural cells are qualitatively contained in FHN model. Because of this the FHN model has got wide dissemination in the last few years. There has been a lot of papers where the influence of noise on the encoding sensitivity of a neuron in the framework of FHN model has been analyzed. A broad spectrum of noise-induced dynamical effects, which produce ordered periodicity in the output of the FHN system, has been discovered. Among these effects we cite the coherence resonance [36] and the stochastic resonance (SR) [37]. All these investigations deal with neuron dynamics with subthreshold signals, and with an enhancement of a weak signal through the noise. The presence of noise in the case of a strong periodic forcing, however, has a detrimental effect on the encoding process [38]-[40]. For suprathreshold signals the noise always lowers the information transmission, and the SR effects disappear [38, 39]. However, as it was shown in recent papers of Stocks [40], this is only true for a single element threshold system. In neuronal arrays the noise can significantly enhance the information transmission when the signal is predominantly suprathreshold. It is the effect of suprathreshold stochastic resonance.
Here we analyze the effect of noise in a single neuron subjected to a strong periodic forcing. We investigate therefore, the influence of noise on the appearance time of a first spike, or the mean response time, in the output of FHN model with periodical driving in suprathreshold regime. As it was mentioned before, the role of noise for a strong driving is negative. In this case noise suppresses the response of a neuron, that leads to delay of transmission of an external information. But we show that, this negative influence of noise on the spike generation can be significantly minimized.
We analyze the dependencies of the mean response time (MRT) on both driving frequency and noise intensity. We find that, MRT plotted as a function of the driving frequency shows a resonant activation-like phenomenon. The noise enhanced stability (NES) effect is also observed here. It is shown that MRT can be increased due to the effect of fluctuations. We note that NES has nothing to do with the typical SR, where the maximum of signal to noise ratio as a function of noise intensity is observed. There are many differences between these effects concerning the neuron dynamics. First of all the SR is related to the output of the neuron in stationary dynamical regime and concerns the signal-to-noise ratio, while the NES describes the transient dynamical regime of a neuron and concerns the mean response time. In addition there is difference in the nature of the response: We
investigate the case of a strong driving, where the SR effects disappear.
4.2. Deterministic dynamical regime
The dynamic equations of the FitzHugh-Nagumo model with additive periodic forcing are
\[ \begin{align*}
\dot{x} &= x - x^3/3 - y + A \sin(\omega t) \\
\dot{y} &= \epsilon(x + I),
\end{align*} \right. \tag{27} \]
where \( x \) is the voltage, \( y \) is the recovery variable, and \( \epsilon \) is a fixed small parameter \((\epsilon = 0.05)\). In the absence of both external driving and noise, there is only one steady state of the system (27), that is \( x_0 = -I; \ y_0 = -I + I^3/3 \). The choice of the constant \( I \), therefore, fully specifies the location of equilibrium state in the phase space \((x, y)\). Here we consider \( I = 1.1 \).
In our simulations we assume that the initial conditions for each realization are the same, that is the system is in its stable equilibrium point (the rest state) \((x_0, y_0)\) at the initial time \(t_0\). We would like to note, here, that even if we consider sinusoidal driving, we investigate the time of appearance of the first spike only. We are interested in the capability of our system to detect an external input. This means to minimize the detection time and to get a neuron response that would be more robust to the noise action. After generation of the first spike, that is after approaching of the boundary \( x = 0 \), we break the realization off, and start a new one with the same initial conditions \((x_0, y_0)\).
For our system the threshold value of the driving amplitude required for spike generation is \( A_{th} \sim 0.05 \). In our simulations we choose \( A = 0.5 \). Thus, the frequency range where the signal of such amplitude is suprathreshold is: \( \Omega : \omega \in (0.013 \div 1.9) \) \([19]\). Inside this region \( \Omega \) the response time of a neuron has a minimum as a function of the driving frequency. System does not respond outside the range \( \Omega \). Here, a subthreshold oscillation occurs (see Fig. 7(b)). In Fig. 7(c) the time series of the output voltage \( x \) for a suprathreshold signal is shown.
4.3. Suprathreshold stochastic regime
Actually, there are many factors that make the environment noisy in the neuron dynamics. Among them we cite the fluctuating opening and closure of the ion channels within the cell membrane, the noisy presynaptic currents, and others (see, for example, Ref. [36]). We consider two different cases in which the noise is added to the first or the second equation of the system (27):
Fig. 7. (a) The response time dependence versus the frequency of periodic driving for the deterministic case, \(A = 0.5\). Examples of output trajectory for two values of driving frequency invoking two different kinds of oscillations: (b) subthreshold for \(\omega = 0.01\), and (c) suprathreshold for \(\omega = 0.02\).
**Case I** The variable that corresponds to the membrane potential is subjected to fluctuations \([37, 40]\). In this case, the first equation of the system (27) becomes the following stochastic differential equation
\[
\dot{x} = x - x^3/3 + A \sin(\omega t) - y + \xi(t);
\]
(28)
**Case II** The recovery variable associated with the refractory properties of a neuron is noisy \([36]\). Here, the second equation of the system (27) becomes
\[
\dot{y} = \epsilon(x + I) + \xi(t),
\]
(29)
In Eqs. (28) and (29), \(\xi(t)\) is a Gaussian white noise with zero mean and correlation function \(\langle \xi(t)\xi(t+\tau) \rangle = D\delta(\tau)\). For numerical simulations we use the modified midpoint method and the noise generator routine reported in Ref [42].
The mean response time (MRT) of our neuronal system is obtained as the mean first passage time at the boundary \(x = 0\): \(\tau = <T> = \frac{1}{N} \sum_{i=1}^{N} T_i\), where \(T_i\) is the response time for \(i\)-th realization. To obtain smooth average for all the noise values investigated, we need different number of realizations \(N\) in above considered cases. Namely, \(N = 5000\) in case I, and \(N = 15000\) in case II, specifically when the noise intensities are comparable with the value of the parameter \(\epsilon = 0.05\). It is worth noting here that parameter \(T_i\) characterizes the delay of the systems’ response, and has a non-zero value even in the deterministic case, because of the non-instantaneous neuronal
response. In our investigation we consider a strong driving, so the noise increases the time of appearance of the first spike and leads to an additional delay of the signal detection.
the MRT dependence on the driving frequency. In the small noise limit $D \to 0$, a typical behavior (see Fig. 8) with perpendicular walls disposed at the frequencies corresponding to the boundaries of the region $\Omega$ was found. By increasing the noise intensity, these walls go down. We observe a resonant activation-like phenomenon: The MRT exhibits a minimum as a function of the driving frequency, which is almost independent of the noise intensity.
Fig. 8. (a) The mean response time dependence versus the frequency of periodic driving for case $I$, for four values of the noise intensity, namely: $D = 0.005, 0.03, 0.07, 0.5$. The right solid lines give the theoretical values of $\tau$ for fixed bistable potential. Inset: frequency range where the noise enhanced stability effect is observed. (b) The standard deviation of the response time dependence vs frequency of periodic driving for case $I$, for the same values of $D$.
In the same figure (b) the standard deviation (SD) of the response time versus the frequency of the periodic driving shows a minimum. Therefore, the noise has minimal effect in the same range of driving frequencies for MRT and its SD. In a narrow frequency range ($\omega \in (0.6 \div 1.3)$) (see Fig. 8), we found a nonmonotonic behavior of the MRT as a function of the noise intensity. Here the noise enhanced stability effect is observed (see the inset of Fig. 8). Out of this range the MRT monotonically decreases with increasing noise intensity. For larger noise intensities the MRT dependence on driving frequency takes a constant-like behavior in the range of the investigated frequency values ($\omega \in [10^{-4} \div 10]$). Here, the dynamics of the system is mainly controlled by the noise, and the frequency of periodic driving does not affect significantly the neuron response dynamics. By numerical simulations of our system we find that, for large noise intensities, the MRT
coincides with that calculated by standard technique for a Brownian particle moving in a bistable fixed potential \[44\]
\[
\tau = \frac{2}{D} \int_{x_0}^{0} e^{\phi(x)/D} \int_{\infty}^{x} e^{-\phi(y)/D} dy \, dx.
\]
(30)
The theoretical values reported in Fig. 8(a) agree with the limiting values of $T$ for $\omega \to 0$ and $\omega \to \infty$. For $\epsilon \ll 1$ in fact, $x$ is a fast variable and $y$ is a slow variable, so $\dot{y} \simeq 0$ and this case can be recast as an escape problem from a one-dimensional double well in both limiting cases. In fact when $\omega \to 0$ we have a fixed bistable potential, and for $\omega \to \infty$ we have an average fluctuating potential, which coincides with the fixed one. This is well visible in Fig. 8(a) for $D = 0.07$. For $D = 0.5$ the MRT tends to be almost independent on the parameter $\omega$.
Fig. 9. The mean response time dependence versus the noise intensity for case I, for three different values of driving frequency: $\omega = 0.001$, $\omega = 1$ and $\omega = 5$. Solid line gives the theoretical values of $\tau$ for fixed bistable potential.
In Fig. 9 the MRT versus the noise intensity, for three values of the driving frequency, is shown. We see the nonmonotonic behavior for $\omega = 1$, which is a signature of the NES effect. It is interesting to note that even in this system, whose global dynamics cannot be described as the motion of a Brownian particle in a potential profile (because of the coupling between the two stochastic differential equations describing our system) a phase transition-like phenomenon, with respect to the driving frequency parameter $\omega$, occurs. In fact we have nonmonotonic and monotonic behavior depending on the value of $\omega$. We expect similar behavior, if we fix the driving frequency and we change the value of the amplitude $A$ of the driving force \[3, 17\].
Case II. In this case, for noise intensity values greater than \( \epsilon = 0.05 \), the recover variable can be approximated by a Wiener process \( \dot{y} \approx \xi(t) \). This process acts now as a noise source in the same double well potential according the following stochastic differential equation
\[
\dot{x} = x - x^3/3 + A \sin(\omega t) + W(t),
\]
where \( W(t) \) is the Wiener process with the usual statistical properties: \( \langle W(t) \rangle = 0 \), and \( \langle W^2(t) \rangle = t \).
In Fig. 10(a) the curve with diamonds shows the results of this approximation for \( D = 0.5 \). We found again a resonant activation-like phenomenon (see Fig. 10), which is independent of the noise intensity, as in case I, until \( D \) reaches the value of parameter \( \epsilon \). The minimum tends to disappear for greater noise intensities. Here a certain frequency range \( (\omega \in (0.019 \div 1.6)) \), larger than in previous case, exists where an increasing noise intensity leads to a monotonic growth of the MRT. Out of this range the MRT monotonically decreases with increasing noise intensity, as in case I. In Fig. 10(b) the standard deviation of the response time versus the driving frequency is shown. Also in this case II, the SD shows a minimum in the same frequency range of that found for MRT. We have found, therefore, a parameter region where there is minimization of the MRT and its SD, that is "suppression of noise".
We also observe that the saturation level reached in each case is different. Particularly in case II it is greater than in case I, because the MRT is calculated with respect to the membrane voltage \( x \) and with different noise
sources. Therefore, in phase space the variable $x$ reaches, in case $I$, in a minor average time the boundary $x = 0$, according to Eq. (28). While in case $II$ the variation of $x$ depends on the dynamics of the $y$ coordinate and takes much more time to reach the same boundary.
5. A stochastic model for cancer growth dynamics
In this last section we shortly summarize some of the main results obtained with a stochastic model for cancer growth dynamics (see Ref. [21] for more details). Most of tumoral cells bear antigens which are recognized as strange by the immune system. A response against these antigens may be mediated either by immune cells such as T-lymphocytes or other cells like macrophages. The process of damage to tumor proceeds via infiltration of the latter by the specialized cells, which subsequently develop a cytotoxic activity against the cancer cell-population. The series of cytotoxic reactions between the cytotoxic cells and the tumor tissue have been documented to be well approximated by a saturating, enzymatic-like process whose time evolution equations are similar to the standard Michaelis-Menten kinetics [45, 46]. The T-helper lymphocytes and macrophages, can secrete cytokines in response to stimuli. The functions that cytokines induce can both "turn on" and "turn off" particular immune responses [47, 48]. This "on-off" modulating regulatory role of the cytokines is here modelled through a dichotomous random variation of the parameter $\beta$, which is responsible for regulatory inhibition of the population growth, by taking into account the natural random fluctuations always present in biological complex systems.
The dynamical equation of this biological system is
$$\dot{x} = -\frac{dU^\pm(x)}{dx} + \xi(t), \quad (32)$$
where $\xi(t)$ is a Gaussian process with $\langle \xi(t) \rangle = 0$, $\langle \xi(t) \xi(t') \rangle = D\delta(t - t')$, and
$$U^\pm(x) = -\frac{x^2}{2} + \frac{\theta x^3}{3} + (\beta_0 \pm \Delta)(x - \ln(x + 1)), \quad (33)$$
is the stochastic double well Michaelis-Menten potential with one the minima at $x = 0$. Here $x(t)$ is the concentration of the cancer cells. The process $\beta = (\beta_0 \pm \Delta)$ can change the relative stability of the metastable state of the potential profile [46]. We note that the RA and NES phenomena act counter to each other in the cancer growth dynamics: the NES effect increases in an unavoidable way the average lifetime of the metastable state (associated to a fixed-size tumor state), while the RA phenomenon minimizes this lifetime. Therefore it is crucial to find the optimal range of parameters in which the
positive role of resonant activation phenomenon, with respect to the cancer extinction, prevails over the negative role of NES, which enhances the stability of the tumoral state. These are just the main results of the paper [21], that is both NES and RA phenomena are revealed in a biological system with a metastable state, with a co-occurrence region of these effects. In this coexistence region the NES effect, which enhances the stability of the tumoral state, becomes strongly reduced by the RA mechanism, which enhances the cancer extinction. In other words, an asymptotic regression to the zero tumor size may be induced by controlling the modulating stochastic activity of the cytokines on the immune system.
6. Conclusions
Natural systems are open to the environment. Consequently, in general, stationary states are not equilibrium states, but are strongly influenced by dynamics, which adds further challenge to the microscopic understanding of metastability. The investigation of two noise-induced effects in far from equilibrium systems, namely the RA and NES phenomena, has revealed interesting peculiarities of the dynamics of these systems. Specifically the knowledge of the parameter regions where the RA and NES can be revealed allows:
- to optimize and to suppress timing errors in practical RSFQ devices, and therefore to significantly increase working frequencies of RSFQ circuits;
- to optimize the operating range of a neuron, and therefore to realize high rate signal transmission with the suppression of noise;
- to maximize or minimize the extinction time in cancer growth population dynamics.
7. Acknowledgments
This work was supported by MIUR, INFM-CNR and CNISM, Russian Foundation for Basic Research (Projects No. 05-01-00509 and No. 05-02-19815), and ESF(European Science Foundation) STOCHDYN network. E.V.P. also acknowledges the support of the Dynasty Foundation.
REFERENCES
[1] J.D. Gunton, M. Droz, *Introduction to the Theory of Metastable and Unstable States*, Springer, Berlin, 1983.
[2] A. J. Leggett, *Phys. Rev. Lett.* **53**, 1096 (1984); M. Muthukumar, *ibid.* **86**, 3188 (2001).
[3] R. N. Mantegna and B. Spagnolo, *Phys. Rev. Lett.* **76**, 563 (1996); *Int. J. Bifurcation and Chaos* **4**, 783 (1998).
[4] A. Strunia, N. Tetradis, *JHEP* **11**, 023 (1999).
[5] P.G. Debenedetti, F.H. Stillinger, *Nature* **410**, 267 (2001).
[6] R. H. Victora, *Phys. Rev. Lett.* **63**, 457 (1989).
[7] Y.W. Bai et al, *Science* **269**, 192 (1995).
[8] O. A. Tretiakov, T. Gramespacher, and K. A. Matveev, *Phys. Rev. B* **67**, 073303 (2003).
[9] M. Gleiser, R.C. Howell, *Phys. Rev. Lett.* **94**, 151601 (2005).
[10] G. Parisi, *Nature* **433**, 221 (2005); S. Kraut and C. Grebogi, *Phys. Rev. Lett.* **93**, 250603 (2004); H. Larralde and F. Leyvraz, *ibid.* **94**, 160201 (2005); G. Baez et al., *ibid.* **90**, 135701 (2003).
[11] C. R. Doering and J. C. Gadoua, *Phys. Rev. Lett.* **69**, 2318 (1992).
[12] M. Bier and R. D. Astumian, *Phys. Rev. Lett.* **71**, 1649 (1993); P. Pechukas and P. Hänggi, *ibidem* **73**, 2772 (1994); J. Iwaniszewski, *Phys. Rev. E* **54**, 3173 (1996); M. Boguñá, J. M. Porra, J. Masoliver, and K. Lindenberg, *ibidem* **57**, 3990 (1998); M. Bier, I. Derenyi, M. Kostur, D. Astumian, *ibidem* **59**, 6422 (1999); A. L. Pankratov and M. Salerno, *Phys. Lett. A* **273**, 162 (2000); A. N. Malakhov and A. L. Pankratov, *Adv. Chem. Phys.* **121**, 357 (2002).
[13] R. N. Mantegna and B. Spagnolo, *Phys. Rev. Lett.* **84**, 3025 (2000); *J. Phys. IV (France)* **8**, 247 (1998).
[14] A. L. Pankratov and M. Salerno, *Phys. Lett. A* **273**, 162 (2000).
[15] B. Dybiec, E. Gudowska–Nowak, *Phys. Rev. E* **66**, 026123 (2002).
[16] Y. Yu and S. Han, *Phys. Rev. Lett.* **91**, 127003 (2003).
[17] A. A. Dubkov, N. V. Agudov and B. Spagnolo, *Phys. Rev. E* **69**, 061103 (2004).
[18] A. L. Pankratov and B. Spagnolo, *Phys. Rev. Lett.* **93**, 177001 (2004).
[19] E. V. Pankratova, A. V. Polovinkin, and B. Spagnolo, *Physics Letters A* **344**, 43-50 (2005).
[20] G. Sun et al., *Phys. Rev. E* **75**, 021107(4) (2007).
[21] A. Fiasconaro and B. Spagnolo, A. Ochab-Marcinek and E. Gudowska-Nowak, *Phys. Rev. E* **74**, 041904(10) (2006); A. Ochab-Marcinek, E. Gudowska-Nowak, A. Fiasconaro and B. Spagnolo, *Acta Physica Polonica B* **37**(5), 1651 (2006).
[22] J. E. Hirsch, B. A. Huberman, and D. J. Scalapino, *Phys. Rev. A* **25**, 519 (1982); I. Dayan, M. Gitterman, and G. H. Weiss, *ibidem* **46**, 757 (1992); R. Wackerbauer, *Phys. Rev. E* **59**, 2872 (1999); D. Dan, M. C. Mahato, and A. M. Jayannavar, *ibidem* **60**, 6421 (1999); A. Mielke, *Phys. Rev. Lett.* **84**, 818 (2000); C. Xie and D. Mei, *Chin. Phys. Lett.* **20**, 813 (2003).
[23] N. V. Agudov and B. Spagnolo, *Phys. Rev. E* **64**, 035102(R) (2001).
[24] A. Fiasconaro, B. Spagnolo and S. Boccaletti, *Phys. Rev. E* **72**, 061110(5) (2005); A. Fiasconaro, D. Valenti, B. Spagnolo, *Physica A* **325**, 136-143 (2003).
[25] A. N. Malakhov and A.L. Pankratov, *Physica C* **269**, 46 (1996).
[26] N. V. Agudov and A. N. Malakhov, *Phys. Rev. E* **60**, 6333 (1999).
[27] B. Spagnolo, A. A. Dubkov, and N. V. Agudov, *Eur. Phys. J. B* **40**, 273-281 (2004); B. Spagnolo, A. A. Dubkov, N. V. Agudov, *Acta Physica Polonica B* **35**, 1419 (2004).
[28] H. A. Kramers, *Physica* **7**, 284 (1940).
[29] P. Hänggi, P. Talkner, and M. Borkovec, *Rev. Mod. Phys.* **62**, 251 (1990).
[30] A. N. Malakhov, *Chaos* **7**, 488 (1997).
[31] Y. Makhlin, G. Schön, and A. Shnirman, *Rev. Mod. Phys.* **73**, 357 (2001); Y. Yu et al., *Science* **296**, 889 (2002).
[32] T. Ortlepp, H. Toepfer and H. F. Ulhmann, *IEEE Trans. Appl. Supercond.* **13**, 515 (2003); V. Kaplunenko, *Physica C* **372-376**, 119 (2002).
[33] A. Barone and G. Paterno, *Physics and Applications of the Josephson Effect*, Wiley, 1982.
[34] J. M. Vilar and J. M. Rubí, *Phys. Rev. Lett.* **86**, 950 (2001).
[35] S. Lee, A. Neiman, and S. Kim, *Phys. Rev. E* **57**, 3292 (1998).
[36] A.S. Pikovsky and J. Kurths, *Phys. Rev. Lett.* **78**, 775 (1997); B. Lindner and L. Schimansky-Geier, *Phys. Rev. E* **60**, 7270 (1999).
[37] A. Longtin, D. Chialvo, *Phys. Rev. Lett.* **81**, 4012 (1998); L. Gammaitoni, P. Hänggi, P. Jung, and F. Marchesoni, *Rev. Mod. Phys.* **70**, 254 (1998).
[38] A.R. Bulsara and A. Zador, *Phys. Rev. E* **54**, R2185 (1996).
[39] J.E. Levin and J.P. Miller, *Nature* **380**, 165 (1996).
[40] N.G. Stocks and R. Mannella, *Phys. Rev. E* **64**, 030902 (2001); N.G. Stocks, D. Allingham, and R.P. Morse, *Fluctuation and Noise Lett.* **2**, L169 (2002).
[41] R. Fitzhugh, *Biophys. J.* **1**, 445-466 (1961); J.S. Nagumo, S. Arimoto, and S. Yoshizawa, *Proc. Inst. Radio Engineers* **50**, 2061-2070 (1962).
[42] W. Press, B. Flannery, S. Teukolsky, and W. Vetterling *Numerical Recipes in C*, Cambridge University Press, Cambridge, 1993.
[43] L.A. Pontryagin, A.A. Andronov, and A.A. Vitt, *Zh. Eksp. Teor. Fiz.* **3**, 165 (1933).
[44] C. W. Gardiner, *Handbook of Stochastic Methods*, (Springer, Berlin, 2004).
[45] R.P. Garay and R. Lefever, *J. Theor. Biol.* **73**, 417 (1978); R. Lefever, W. Horsthemke, *Bull. of Math. Biol.* **41**, 469 (1979).
[46] A. Ochab-Marcinek and E. Gudowska-Nowak, *Physica A* **343**, 557 (2004).
[47] A. Mantovani, A. Sica, et al., *Trends Immunol.* **25**, 677 (2004); A. Mantovani, P. Allavena, A. Sica, *Eur. J. Cancer* **40**, 1660 (2004).
[48] R. L. Elliott and G. C.Blobe, *J. Clin. Oncol.* **23**, 2078 (2005). | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2042,
1
],
[
2042,
5023,
2
],
[
5023,
7327,
3
],
[
7327,
8774,
4
],
[
8774,
10888,
5
],
[
10888,
12831,
6
],
[
12831,
14817,
7
],
[
14817,
17745,
8
],
[
17745,
20716,
9
],
[
20716,
23175,
10
],
[
23175,
25117,
11
],
[
25117,
28072,
12
],
[
28072,
30780,
13
],
[
30780,
33116,
14
],
[
33116,
35176,
15
],
[
35176,
38252,
16
],
[
38252,
40794,
17
],
[
40794,
42614,
18
],
[
42614,
44749,
19
],
[
44749,
46665,
20
],
[
46665,
48376,
21
],
[
48376,
51018,
22
],
[
51018,
53046,
23
],
[
53046,
55852,
24
],
[
55852,
58571,
25
]
]
} |
0ce2deef1fb6c2f0282e5e4bae684f5520142a27 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 10,
"total-fallback-pages": 0,
"total-input-tokens": 33542,
"total-output-tokens": 13291,
"fieldofstudy": [
"Medicine",
"Biology",
"Psychology"
]
} | Serum neurofilament light protein correlates with unfavorable clinical outcomes in hospitalized patients with COVID-19
Mercedes Prudencio1,2†, Young Erben3†, Christopher P. Marquez4, Karen R. Jansen-West3, Camila Franco-Mesa3, Michael G. Heckman5, Launia J. White5, Judith A. Dunmore1, Casey N. Cook1,2, Meredith T. Lilley1, Yuping Song1, Caroline F. Harlow4, Björn Oskarsson6, Katharine A. Nicholson7, Zbigniew K. Wszolek8, LaTonya J. Hickson8, John C. O’Horo9,10, Jonathan B. Hoyne4, Tania F. Gendron1,2, James F. Meschia6*, Leonard Petrucelli1,2*
Brain imaging studies of patients with COVID-19 show evidence of macro- and microhemorrhagic lesions, multifocal white matter hyperintensities, and lesions consistent with posterior reversible leukoencephalopathy. Imaging studies, however, are subject to selection bias, and prospective studies are challenging to scale. Here, we evaluated whether serum neurofilament light chain (NFL), a neuroaxonal injury marker, could predict the extent of neuronal damage in a cohort of 142 hospitalized patients with COVID-19. NFL was elevated in the serum of patients with COVID-19 compared to healthy controls, including those without overt neurological manifestations. Higher NFL serum concentrations were associated with worse clinical outcomes. In 100 hospitalized patients with COVID-19 treated with remdesivir, a trend toward lower NFL serum concentrations was observed. These data suggest that patients with COVID-19 may experience neuroaxonal injury and may be at risk for long-term neurological sequelae. Neuroaxonal injury should be considered as an outcome in acute pharmacotherapeutic trials for COVID-19.
INTRODUCTION
Since the World Health Organization declared a global pandemic on 11 March 2020, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has caused more than 2.8 million deaths worldwide and more than 550,000 deaths within the United States (1). Although common symptoms of coronavirus disease 2019 (COVID-19) include fever, cough, fatigue, and shortness of breath, the manifestations of COVID-19 can vary widely. For example, some patients can develop pneumonia, acute respiratory distress syndrome, myocardial injury, arrhythmias, and/or multiorgan failure (2), and it is increasingly recognized that SARS-CoV-2 can cause neurologic signs (3–10). Clinical studies observed neuropsychiatric manifestations in up to 70% of patients with COVID-19, including young adults and patients in whom respiratory symptoms have long resolved (10–12). Brain imaging studies detected diverse lesions including perfusion abnormalities with or without acute infarctions, cerebral venous thrombosis, macro- and microhemorrhages, multifocal white matter and basal ganglia lesions, meningeal enhancement, central pontine myelinolysis, posterior reversible encephalopathy syndrome, and neuritis (13–16). Moreover, brain autopsy studies found SARS-CoV-2 RNA or proteins in various neuroanatomical regions of patients with COVID-19, with astrogliosis, activation of microglia, and infiltration of cytotoxic T lymphocytes noted in many cases (17, 18).
Given that neurologic manifestations are now considered common features of COVID-19, we sought to examine the utility of serum neurofilament light chain (NFL) in assessing the frequency, severity, and clinical consequences of neuronal injury associated with SARS-CoV-2 infections warranting hospitalization. Recently, we determined that blood NFL concentrations are associated with radiographic markers of brain tissue damage, as well as indicators of neurological, functional, and cognitive status in patients with ischemic or hemorrhagic stroke (19). On the basis of these findings and others (20) establishing NFL as a marker of axonal injury, we investigated whether serum NFL was elevated in hospitalized patients with COVID-19 and whether it correlated with clinical outcomes and disease severity.
RESULTS
Serum NFL is elevated in hospitalized patients with COVID-19
We measured NFL in 488 serum samples longitudinally collected from 142 hospitalized patients with COVID-19 and in cross-sectional serum samples from 55 healthy controls. A summary of patient demographics, treatments, and clinical outcomes is shown in Table 1. Characteristics of control individuals are presented in table S1. The median age in patients with COVID-19 was 62 years (range, 22 to 99 years), and 57.7% were male (82 patients). In control individuals, the median age was 62 years (range, 32 to 84 years) and 50.9% were male (28 patients). Comparing mean NFL concentrations for each of the 142 patients with COVID-19 to NFL concentrations of the 55 controls revealed that serum NFL was significantly higher in patients (median, 29.4 pg/ml; range, 3.4 to 1538.4 pg/ml) than controls.
1Department of Neuroscience, Mayo Clinic, Jacksonville, FL 32224, USA. 2Neuroscience Graduate Program, Mayo Clinic Graduate School of Biomedical Sciences, Jacksonville, FL 32224, USA. 3Division of Vascular and Endovascular Surgery, Mayo Clinic, Jacksonville, FL 32224, USA. 4Department of Laboratory Medicine and Pathology, Mayo Clinic, Jacksonville, FL 32224, USA. 5Division of Biomedical Statistics and Informatics, Mayo Clinic, Jacksonville, FL 32224, USA. 6Division of Neurology, Mayo Clinic, Jacksonville, FL 32224, USA. 7Division of Biomedical Statistics and Informatics, Mayo Clinic, Jacksonville, FL 32224, USA. 8Division of Neurology, Mayo Clinic, Jacksonville, FL 32224, USA. 9Division of Cardiology, Mayo Clinic, Jacksonville, FL 32224, USA. 10Division of Pulmonary and Critical Care Medicine, Mayo Clinic, Rochester, MN 55905, USA.
*Corresponding author. Email: [email protected] (J.F.M.); [email protected] (L.P.). †These authors contributed equally to this work.
This finding was consistent after correcting for age and sex and when examining the first, last, minimum, or maximum NFL measurement per patient (Table 2). To assess how many patients with COVID-19 had elevated NFL according to a prespecified cutoff, we determined the number of patients with an NFL concentration of at least 3 SDs above the group mean NFL concentration of control individuals. For this analysis, we focused on maximum NFL measurements given that NFL concentrations fluctuate between time points, as demonstrated for 35 of the patients with longitudinal NFL data (Fig. 1B), such that using mean NFL concentrations could mask our ability to discern elevated NFL concentrations at one or more time points. About 34% of patients with COVID-19 (48 individuals) had NFL concentrations higher than the cutoff. We additionally determined
| Variable | N | Median (minimum, maximum) or no. (%) |
|----------|---|-------------------------------------|
| ICU, intensive care unit; BMI, body mass index; CKD, chronic kidney disease; LOS, length of hospital stay; mRS, modified Rankin scale. | 142 | |
| Patient characteristics | 142 | 62 (22, 99) |
| Age at admission (years) | 142 | 62 (22, 99) |
| Sex (male) | 142 | 82 (57.7%) |
| Race | 142 | |
| White | 96 (67.6%) |
| Black | 32 (22.5%) |
| Asian | 12 (8.5%) |
| Other or not disclosed | 2 (1.4%) |
| BMI | 142 | 30.0 (15.3, 80.1) |
| Obesity | 142 | 71 (50.0%) |
| CKD | 142 | 17 (12.0%) |
| Length of time from admission to first NFL (days) | 142 | 1 (0, 32) |
| Number of NFL measurements | 142 | |
| 1 | 42 (29.6%) |
| 2 | 20 (14.1%) |
| 3 | 29 (20.4%) |
| 4 | 24 (16.9%) |
| 5 | 8 (5.6%) |
| >5 | 19 (13.4%) |
| Treatments | 142 | |
| Monoclonal antibody | 37 (26.1%) |
| Remdesivir | 106 (74.6%) |
| Dexamethasone therapy | 116 (81.7%) |
| Convalescent plasma | 44 (31.0%) |
| Outcomes | 142 | |
| Mechanical ventilation (intubation) | 30 (21.1%) |
| ICU admission | 54 (38.0%) |
| LOS (days) | 9 (2, 119) |
| Death | 13 (9.2%) |
| mRS at discharge | 142 | |
| 0 | 1 (0.7%) |
| 1 | 40 (28.2%) |
| 2 | 35 (24.6%) |
| 3 | 33 (23.2%) |
| 4 | 15 (10.6%) |
| 5 | 7 (4.9%) |
| 6 | 11 (7.7%) |
the number of patients with an NFL concentration of at least 2 SDs above the mean control concentration and found that 53% of patients (76 individuals) met this criterion.
**Elevated serum NFL associates with worse clinical outcomes in hospitalized patients with COVID-19**
We next examined the relationship between serum NFL concentrations and diverse clinical outcomes: mechanical ventilation (intubation), admission to the intensive care unit (ICU), length of hospital stay (LOS), and modified Rankin scale (mRS) at discharge. As shown in Table 3, in unadjusted analyses and in analyses adjusting for age, sex, body mass index (BMI), and chronic kidney disease (CKD), serum NFL concentrations were significantly higher in patients who needed mechanical ventilation, who were admitted to the ICU, who had a longer LOS, and who had a higher mRS at discharge (all \( P < 0.001 \)). Furthermore, we found a positive correlation between NFL and time from admission to final blood draw (Spearman’s \( r = 0.50, P < 0.001 \)); thus, we also performed similar analyses adjusting for time from admission to final blood draw (table S2) and observed results consistent with findings reported in Table 3.
**Longitudinal serum NFL concentrations in hospitalized patients with COVID-19 and their association with treatment type**
Among the 142 patients with COVID-19, longitudinal NFL data were available for 100 individuals for whom serum NFL was measured for up to 17 time points. As anticipated, temporal changes in NFL differed among patients, and this was true of the 35 patients with longitudinal NFL measures and for whom mean NFL concentrations were considered elevated (Fig. 1C). For instance, in some individuals, serum NFL rose over time, whereas for others NFL remained relatively consistent or fluctuated to varying degrees.
Since patients were administered a range of potential therapies including monoclonal antibody treatment, remdesivir, dexamethasone, and convalescent plasma therapy (Table 1), we next examined whether these treatments influenced NFL. Comparisons of serum NFL according to specific treatments for COVID-19 are displayed in Table 4. After adjusting for age, sex, BMI, and time from admission to blood draw, as well as correcting for multiple analyses (\( P < 0.0063 \) considered as significant), a tendency toward lower serum NFL in patients who received remdesivir was observed in multivariable analysis (final NFL concentration per patient, \( P = 0.008 \); Table 4).
**DISCUSSION**
COVID-19 is associated with diverse neurological injuries, but questions regarding their incidence, severity, and long-term consequences, and whether such injuries can be mitigated by acute intervention, remain unanswered (21, 22). Toward addressing these important questions, we examined the utility of serum NFL in determining the frequency, severity, and clinical consequences of neuronal injury in hospitalized patients with COVID-19. We show that...
serum NFL is elevated in patients with COVID-19 and that ~34% of patients have mean NFL concentrations of at least 3 SDs above the group mean of control individuals. In addition, elevated serum NFL correlates with worse clinical outcomes, such as the need for mechanical ventilation (intubation), ICU admission, longer lengths of hospitalization, and worse functional outcomes. We further observed a tendency of lower serum NFL in patients with COVID-19 treated with remdesivir. In aggregate, these findings suggest that a considerable proportion of hospitalized patients with COVID-19 suffer neuronal injury, the degree of which associates with disease severity.
We reported that patients with COVID-19 can develop a myriad of neurologic symptoms including headaches, encephalopathy, and seizures (13), providing indication for head imaging. Nevertheless, the great majority of patients with COVID-19 (86.7%) show no intracranial abnormalities by imaging (13). It also bears mentioning that retrospective studies using brain imaging approaches in patients with COVID-19 suffer from selection bias, in particular omitting patients with mild or no obvious neurological symptoms and patients too unstable or having contraindications to undergo magnetic resonance imaging. This selection bias, combined with the uncertainty of the sensitivity of brain imaging to neuroaxonal injury (12), suggests that neuronal injury or neurodegeneration in patients with COVID-19 may be underappreciated based on imaging studies alone. Given findings from the present study, we believe that measuring serum NFL in patients with COVID-19 will facilitate the detection of neuronal injury that may otherwise be overlooked.
Our findings that serum NFL is elevated in our cohort of 142 patients hospitalized with COVID-19, and that higher serum NFL associates with worse clinical outcomes, are in line with previous reports. For example, one study observed elevated serum NFL in 28 patients with mild-to-moderate COVID-19 (23), whereas another found that, among 47 patients with COVID-19, plasma NFL was higher in the 18 patients with severe disease (24). Yet, another reported serum NFL to be elevated in 29 critically ill patients with COVID-19 and to positively associate with an unfavorable outcome (25). Last, cerebrospinal fluid (CSF) NFL was elevated in two of six hospitalized patients with moderate or severe COVID-19 (26). Our data are also congruent with the growing consensus that SARS-CoV-2 causes potentially damaging neurological problems. However, how this occurs remains unclear. Similarly to other respiratory viruses that have neuroinvasive capacities (27, 28), SARS-CoV-2 can spread to the central nervous system (17, 18). Postulated pathways by which this occurs include its retrograde axonal transport along the sensory and olfactory nerves in the cribiform plate, its invasion of endothelial cells by interacting with the angiotensin-converting enzyme 2 receptor, and its ability to alter tight junction proteins formed by endothelial cells (29). Nevertheless, despite the ability of SARS-CoV-2 to invade the brain, the neurological signs and symptoms it causes are believed to more likely result from systemic reactions such as hypoxemia, hypercoagulability, systemic inflammation, and multiorgan failure (18, 26, 30). Therapies available for COVID-19 include dexamethasone (31), monoclonal antibody toward the spike protein of SARS-CoV-2 (32), remdesivir (a viral replication inhibitor) (33, 34), and convalescent plasma (35, 36). In our cohort of patients with COVID-19, a tendency of lower serum NFL concentrations was seen with the use of remdesivir compared to its non-use. Although this association needs to be confirmed in larger patient cohorts and independently replicated, these findings underscore the utility of incorporating subacute measurements of NFL during hospitalization and in randomized drug trials.
Strengths of our study include the relatively substantial number of patients with COVID-19, the longitudinal examination of serum NFL, and our analysis of associations of NFL with clinical outcomes and treatment type. Our study also has limitations. Most patients in our cohort were hospitalized for a considerable length of time such that data for patients with shorter hospital stays (due to less severe disease) were comparatively limited. There were also delays from time of suspected infection or the time of first symptoms to blood draw. Our analyses used the time of hospital admission as the baseline time point; however, the length of time from symptom onset to admission likely varies among patients. Last, our analyses assessing associations of COVID-19 treatments with serum NFL should be viewed as exploratory. The small sample sizes together with differences in timing of treatment in relation to symptom onset, in severity of symptoms once treatment was initiated, and in duration of treatment from its initiation to blood draw for NFL measurements may have hampered our efforts to detect associations of interest. To rigorously address this question would require a randomized trial in which NFL concentrations are longitudinally measured in patients before and after treatment.
Overall, we show that serum NFL was elevated in hospitalized patients with COVID-19 and correlated with worse clinical outcomes. Given findings from the present study, we believe that measuring serum NFL in patients with COVID-19 will facilitate the detection of neuronal injury that may otherwise be overlooked.
Table 3. Associations between serum NFL and outcomes in patients with COVID-19. For descriptive summaries of NFL concentration and for ease of presentation, LOS was categorized using the sample median, whereas mRS at discharge was categorized using a predefined cutoff of interest. Associations of intubation, ICU admission, LOS, and mRS at discharge (all as independent variables) with NFL concentration (as the dependent variable) were evaluated using linear regression models (for the continuous mean NFL per patient, minimum NFL per patient, and maximum NFL per patient variables) and logistic regression models (for the binary occurrence of an NFL value >25th percentile and occurrence of an NFL value >75th percentile variables). Regression coefficients are interpreted as the change in the mean NFL outcome measure (on the natural logarithmic scale) corresponding to the presence of the given characteristic (intubation and ICU admission), to each doubling of LOS, and to each 1-unit increase in mRS at discharge. CI, confidence interval; CKD, chronic kidney disease. P values <0.0025 were considered as statistically significant after applying a Bonferroni correction for multiple testing.
| Association examined | Median (minimum, maximum) NFL or no. (%) for the given group | Association measure | Unadjusted analysis | Adjusting for age, sex, BMI, and CKD |
|----------------------|-------------------------------------------------------------|---------------------|---------------------|----------------------------------|
| | | | Estimate (95% CI) | P value | Estimate (95% CI) | P value |
| | | | | | | |
| **Mean NFL per patient** | | | | | | |
| No intubation (N = 112) | 19.7 (3.4, 972.7) | Intubation (N = 30) | 107.4 (15.9, 1538.4) | Regression coefficient | 1.61 (1.16, 2.06) | <0.001 | 1.66 (1.24, 2.07) | <0.001 |
| **Minimum NFL per patient** | | | | | | |
| No ICU admission (N = 88) | 16.6 (2.2, 585.6) | ICU admission (N = 54) | 50.3 (3.1, 1538.4) | Regression coefficient | 1.25 (0.78, 1.72) | <0.001 | 1.31 (0.87, 1.75) | <0.001 |
| **Maximum NFL per patient** | | | | | | |
| No ICU admission (N = 88) | 22.2 (3.4, 1359.7) | ICU admission (N = 54) | 111.5 (15.9, 2131.0) | Regression coefficient | 1.75 (1.27, 2.24) | <0.001 | 1.79 (1.35, 2.24) | <0.001 |
| NFL value >25th percentile | 74 (66.1%) | 30 (100.0%) | Odds ratio | N/A* | <0.001 | N/A* | N/A* |
| **NFL value >75th percentile** | | | | | | |
| No ICU admission (N = 88) | 19 (13.4%) | ICU admission (N = 54) | 151 (63.3%) | Odds ratio | 11.17 (4.45, 28.03) | <0.001 | 10.64 (3.80, 29.80) | <0.001 |
| **Association between ICU admission and** | **Mean NFL per patient** | | | | | |
| No ICU admission (N = 88) | 18.7 (3.4, 972.7) | ICU admission (N = 54) | 52.2 (4.8, 1538.4) | Regression coefficient | 1.10 (0.69, 1.50) | <0.001 | 1.13 (0.76, 1.49) | <0.001 |
| **Minimum NFL per patient** | | | | | | |
| No ICU admission (N = 88) | 15.8 (2.2, 585.6) | ICU admission (N = 54) | 33.2 (3.1, 1538.4) | Regression coefficient | 0.82 (0.41, 1.23) | <0.001 | 0.86 (0.48, 1.24) | <0.001 |
| **Maximum NFL per patient** | | | | | | |
| No ICU admission (N = 88) | 20.4 (3.4, 1359.7) | ICU admission (N = 54) | 64.5 (7.9, 2131.0) | Regression coefficient | 1.23 (0.80, 1.66) | <0.001 | 1.26 (0.87, 1.65) | <0.001 |
| NFL value >25th percentile | 55 (62.5%) | 49 (90.7%) | Odds ratio | 5.88 (2.13, 16.25) | <0.001 | 9.84 (3.00, 32.29) | <0.001 |
| **NFL value >75th percentile** | | | | | | |
| No ICU admission (N = 88) | 10 (11.4%) | ICU admission (N = 54) | 24 (44.4%) | Odds ratio | 6.24 (2.67, 14.59) | <0.001 | 8.00 (3.08, 20.75) | <0.001 |
| **Association between LOS and** | **Mean NFL per patient** | | | | | |
| LOS ≤ 9 days (N = 79) | 15.9 (3.4, 972.7) | LOS > 9 days (N = 63) | 47.8 (4.8, 1538.4) | Regression coefficient | 0.81 (0.60, 1.01) | <0.001 | 0.74 (0.55, 0.93) | <0.001 |
| **Minimum NFL per patient** | | | | | | |
| LOS ≤ 9 days (N = 79) | 13.9 (2.2, 585.6) | LOS > 9 days (N = 63) | 33.4 (3.1, 1538.4) | Regression coefficient | 0.66 (0.44, 0.87) | <0.001 | 0.60 (0.40, 0.80) | <0.001 |
| **Maximum NFL per patient** | | | | | | |
| LOS ≤ 9 days (N = 79) | 18.6 (3.4, 1359.7) | LOS > 9 days (N = 63) | 66.4 (7.9, 2131.0) | Regression coefficient | 0.89 (0.67, 1.10) | <0.001 | 0.82 (0.62, 1.02) | <0.001 |
| NFL value >25th percentile* | 45 (57.0%) | 59 (93.7%) | Odds ratio | 5.34 (2.53, 11.24) | <0.001 | 4.90 (2.27, 10.62) | <0.001 |
| **NFL value >75th percentile** | | | | | | |
| LOS ≤ 9 days (N = 79) | 8 (10.1%) | LOS > 9 days (N = 63) | 26 (41.3%) | Odds ratio | 3.17 (1.91, 5.24) | <0.001 | 3.24 (1.92, 5.47) | <0.001 |
continued on next page
outcomes. These data further cement the field’s recognition of the neurological manifestations caused by SARS-CoV-2 infection. Nevertheless, our understanding of the long-term implications is limited. It has been shown that patients previously hospitalized with COVID-19 display a wide array of neurological symptoms months after discharge (10, 12, 37), but few studies have systematically followed patients over time, likely because of the inherent difficulties in doing so. However, on the basis of our present data, we posit that longitudinal measurements of serum NFL would provide an efficient means to identify and quantify neurological injury in hospitalized patients with COVID-19. Serum NFL may also be useful for monitoring end-stage organ disease progression and recovery, aiding in the identification of risk factors and clinical features that contribute to COVID-19–associated neurological signs, and indicating neuroaxonal injury in COVID-19 drug trials.
**MATERIALS AND METHODS**
**Study design**
The goal of this study was to investigate serum NFL as a biomarker of neuroaxonal injury in patients with COVID-19. NFL was measured using the NF-Light digital immunoassay from Quanterix. Our primary analyses were to determine whether serum NFL is elevated in hospitalized patients with COVID-19 and whether NFL associates with clinical outcomes and treatment type. We included 142 patients with COVID-19 (no randomization) for whom serum (488 samples) was collected cross-sectionally or longitudinally during hospitalization. Our study also included serum from 55 healthy controls. NFL was measured in a blinded manner. Sample sizes were based on what was available when the study was initiated and not on sample size calculations. Biological samples were obtained when residual blood was available from patients with approval by the ethics committee.
**Study subjects**
A total of 488 serum samples from 142 patients hospitalized with COVID-19 were included in this study. The Mayo Clinic Neurological, Vascular and Neurovascular Events With SARS-CoV-2 Study [MC NEWS; Institutional Review Board (IRB) #20-003457] was queried to identify a cohort of individuals with COVID-19. MC NEWS included patients across the three major campuses of Mayo Clinic and the Mayo Clinic Health System, with hospitals in Arizona, Florida, Minnesota, and Wisconsin. Fifty-five healthy controls without COVID-19 and no neurological condition were additionally included. This group was composed of 22 healthy individuals from the general population, 19 healthy spouses or caregivers of patients with ataxia (N = 15) or amyotrophic lateral sclerosis (ALS; N = 4), and 14 unaffected relatives of patients with ataxia (N = 4) or ALS (N = 10) and who lacked disease-associated gene mutations. Serum samples were obtained under IRB approval through the following protocols: “Investigating biomarkers, disease mechanisms and treatments for spinocerebellar ataxia and nucleotide repeat diseases,” IRB#17-006033; “Investigating the Genetic and Phenotypic Presentation of Spinocerebellar Ataxia and Nucleotide Repeat Diseases,” IRB#16-009414; “Biospecimen Biorepository for the Study of ALS, ALS-FTD and Similar Neurodegenerative Disorders,” IRB#13-004314; “Pilot Evaluation of Neurofilament Heavy Form (NF-H) as a Potential Biomarker of Axonal Loss in Amyotrophic Lateral Sclerosis (ALS),” IRB#10-003592; “Clinical & Genetic Studies in ALS, Suspected ALS, and Other Neurodegenerative Motor Neuron Disorders,” IRB#07-005711; “Biospecimen Collection to Investigate the Causes of ALS,” IRB#15-001187; “The DIALS (Dominant Inherited ALS) Network,” IRB#2017P000485; and “COVID-19 cytokine storm project,” IRB#20-003661. For patients with COVID-19, the following demographic and clinical information was abstracted using the shared electronic medical record (Epic): age, sex, BMI, CKD status, and COVID-19 treatment information was abstracted using the shared electronic medical record (Epic).
| Association examined | Median (minimum, maximum) NFL or no. (%) for the given group | Association measure | Unadjusted analysis | Adjusting for age, sex, BMI, and CKD |
|----------------------|-------------------------------------------------------------|---------------------|---------------------|-------------------------------------|
| | mRS at discharge ≤3 (N = 109) | mRS at discharge >3 (N = 33) | | |
| Mean NFL per patient | 19.5 (3.4, 405.3) | 91.9 (17.9, 1538.4) | Regression coefficient: 0.56 (0.46, 0.67) | <0.001 | 0.54 (0.42, 0.66) | <0.001 |
| Minimum NFL per patient | 14.7 (2.2, 304.9) | 47.8 (3.1, 1538.4) | Regression coefficient: 0.49 (0.38, 0.61) | <0.001 | 0.45 (0.33, 0.58) | <0.001 |
| Maximum NFL per patient | 21.4 (3.4, 779.0) | 105.0 (18.1, 2131.0) | Regression coefficient: 0.60 (0.48, 0.71) | <0.001 | 0.58 (0.46, 0.71) | <0.001 |
| NFL value >25th percentile* | 71 (65.1%) | 33 (100.0%) | Odds ratio: 3.67 (2.17, 6.22) | <0.001 | 3.00 (1.71, 5.29) | <0.001 |
| NFL value >75th percentile | 16 (14.7%) | 18 (54.5%) | Odds ratio: 2.27 (1.66, 3.11) | <0.001 | 3.04 (2.02, 4.58) | <0.001 |
*When evaluating the association between intubation and NFL value >25th percentile, logistic regression was not possible owing to the occurrence of a zero cell count. Therefore, the P value in unadjusted analysis results from Fisher’s exact test, and multivariable analysis was not performed.
summarized descriptively but was not analyzed as an outcome in association analyses owing to the small number of patients who died (13 of 142) and because death is included in the mRS at discharge outcome (note: 2 of the 13 patients who died did so after discharge; hence, only 11 patients had an mRS score of 6 at discharge). Age, race, and sex were collected for the control individuals.
**NFL measurements**
Serum NFL concentrations were measured in duplicate in a blinded fashion on a Simoa HD-X analyzer using an NF-Light digital immunoassay (Quanterix, catalog no. 103186) according to the manufacturer’s instructions. Briefly, samples were thawed and cleared by centrifugation at 10,000g for 5 min, transferred to 96-well plates, and run in duplicate using a 4x instrument dilution. Four samples were included in each run to monitor potential variability among assays, along with appropriate calibrators and controls provided by the manufacturer. NFL concentrations were interpolated from the calibration curve using a 1/y² weighted four-parameter logistic curve fit. Samples with NFL concentrations that exceeded the range of the assay were retested using an appropriate at-bench dilution in addition to the 4x instrument dilution. Samples with coefficients of variation above 20% were also retested.
**Table 4. Comparison of serum NFL concentrations according to COVID-19 treatment.** Regression coefficients, 95% CIs, and P values result from linear regression models, where NFL was the dependent variable. Regression coefficients are interpreted as the difference in mean NFL concentration (on the natural logarithmic scale) between patients who had the given treatment and patients who did not have the given treatment. P values <0.0063 were considered as statistically significant after applying a Bonferroni correction for multiple testing.
| COVID-19 treatment | N | Median (minimum, maximum) NFL concentration | Adjusting for time from admission to blood draw | Adjusting for age, sex, BMI, and time from admission to blood draw |
|-------------------|-----|---------------------------------------------|-----------------------------------------------|---------------------------------------------------------------|
| | | Regression coefficient (95% CI) | P value | Regression coefficient (95% CI) | P value |
| Monoclonal antibody treatment | | | | |
| First NFL value per patient | | | | |
| No treatment 125 | 22.5 (2.9, 1538.4) | 1.00 (reference) | N/A | 1.00 (reference) | N/A |
| Treatment 17 | 20.0 (3.5, 647.3) | -0.34 (−0.91, 0.23) | 0.29 | -0.47 (−1.00, 0.05) | 0.074 |
| Final NFL value per patient | | | | |
| No treatment 107 | 31.1 (2.2, 1538.4) | 1.00 (reference) | N/A | 1.00 (reference) | N/A |
| Treatment 35 | 33.4 (3.1, 1465.8) | -0.38 (−0.85, 0.09) | 0.12 | -0.39 (−0.84, 0.07) | 0.094 |
| Remdesivir treatment | | | | |
| First NFL value per patient | | | | |
| No treatment 92 | 21.2 (2.9, 872.3) | 1.00 (reference) | N/A | 1.00 (reference) | N/A |
| Treatment 50 | 23.7 (3.4, 1538.4) | -0.25 (−0.68, 0.18) | 0.25 | -0.24 (−0.64, 0.15) | 0.23 |
| Final NFL value per patient | | | | |
| No treatment 42 | 32.3 (3.4, 1359.7) | 1.00 (reference) | N/A | 1.00 (reference) | N/A |
| Treatment 100 | 32.1 (2.2, 1538.4) | -0.47 (−0.89, −0.05) | 0.029 | -0.56 (−0.97, −0.15) | 0.008 |
| Dexamethasone treatment | | | | |
| First NFL value per patient | | | | |
| No treatment 103 | 21.4 (2.9, 1538.4) | 1.00 (reference) | N/A | 1.00 (reference) | N/A |
| Treatment 39 | 22.5 (3.4, 872.3) | 0.27 (−0.16, 0.69) | 0.22 | 0.26 (−0.13, 0.64) | 0.19 |
| Final NFL value per patient | | | | |
| No treatment 66 | 29.3 (2.2, 1538.4) | 1.00 (reference) | N/A | 1.00 (reference) | N/A |
| Treatment 76 | 37.8 (3.1, 1465.8) | -0.11 (−0.51, 0.29) | 0.59 | -0.08 (−0.46, 0.31) | 0.69 |
| Convalescent plasma treatment | | | | |
| First NFL value per patient | | | | |
| No treatment 119 | 20.3 (2.9, 1538.4) | 1.00 (reference) | N/A | 1.00 (reference) | N/A |
| Treatment 23 | 33.4 (9.4, 647.3) | -0.24 (−0.35, 0.25) | 0.43 | -0.25 (−0.78, 0.28) | 0.35 |
| Final NFL value per patient | | | | |
| No treatment 107 | 27.1 (2.2, 1538.4) | 1.00 (reference) | N/A | 1.00 (reference) | N/A |
| Treatment 35 | 40.0 (3.1, 1186.2) | -0.01 (−0.04, 0.45) | 0.98 | -0.02 (−0.41, 0.45) | 0.93 |
NFL was measured in serum collected at a single time point for 42 individuals (29.6% of patients) and in serum collected at multiple time points for the remaining patients. For patients who underwent longitudinal serum collection, the median number of NFL measurements was 4 (range, 2 to 17).
**Statistical analysis**
Continuous variables were summarized with the sample median and range. Categorical variables were summarized with number and percentage of patients. Comparisons of NFL concentrations between controls and patients with COVID-19 were made using a Wilcoxon rank sum test in unadjusted analysis and using a stratified van Elteren Wilcoxon rank sum test (39) in adjusted analysis, where the test was stratified by both age as a four-level categorical variable (based on sample quartiles) and sex. For comparisons of NFL between the control group and patients with COVID-19, single–time point NFL concentrations of controls were compared to the following five NFL values in patients: (i) NFL in serum from the first blood draw, (ii) NFL in serum from the last blood draw, (iii) mean NFL concentration in all serum samples for a given patient, (iv) minimum NFL concentration per patient, and (v) maximum NFL concentration per patient. In patients alone, we also assessed the strength of association between the length of time from admission to blood draw and NFL concentrations by estimating Spearman’s correlation coefficient and by obtaining a \( P \) value from a mixed effects linear regression model that included a random effect for patients. NFL was examined on the natural logarithmic scale in this and all subsequently described analyses owing to its skewed distribution.
When assessing associations of NFL concentrations with outcomes (need for mechanical ventilation/intubation, ICU admission, LOS, and mRS at discharge), we considered the five following patient-specific NFL variables: (i) mean NFL concentration per patient, (ii) minimum NFL concentration per patient, (iii) maximum NFL concentration per patient, (iv) occurrence of an NFL concentration \( >25 \)th percentile, and (v) occurrence of an NFL concentration \( >75 \)th percentile. The 25th and 75th percentile cutoffs were calculated using all NFL concentrations in our cohort of patients with COVID-19.
Given that outcomes often occurred either before measurement of any NFL values (need for mechanical ventilation/intubation or ICU admission) or at a very similar time point as measurement of some NFL values (LOS and mRS at discharge), performing an analysis that examined the ability of NFL to predict these outcome measures was not possible given the data. Therefore, to assess whether NFL associates with poor outcomes in patients with COVID-19, we assessed associations between outcomes (intubation, ICU admission, LOS, and mRS at discharge) and each of the five aforementioned NFL variables using linear regression models (mean, minimum, and maximum NFL per patient) and logistic regression models (NFL \( >25 \)th percentile and NFL \( >75 \)th percentile), where the NFL variables were examined as the dependent variables in the regression models and NFL values were included regardless of temporal relationship to time of start of mechanical ventilation (intubation) or time of ICU admission. Regression coefficients and 95% confidence intervals (CIs) were estimated from linear regression models, whereas odds ratios (ORs) and 95% CIs were estimated from logistic regression models. Unadjusted models were first examined, and these were followed by multivariable models that were adjusted for age, sex, BMI, and CKD. The latter was included given that blood NFL concentrations are affected by renal function (40). Length of time from admission to blood draw was not initially taken into account in multivariable models because this variable can, to some extent, also be thought of as an outcome measure (because a longer time from admission to blood draw indicates a patient with a longer hospitalization and likely worse outcomes) and therefore may be on the causal pathway between NFL and outcomes (41). However, in a secondary analysis, we did examine the sensitivity of our results to additional multivariable adjustment for length of time from admission to final blood draw.
We examined whether specific COVID-19 treatments (monoclonal antibody, remdesivir, dexamethasone, and convalescent plasma) associated with a lower or higher serum NFL as follows. First, to satisfy the statistical assumption of independent measurements, we performed two separate analyses using a single NFL value per patient (because many patients had multiple NFL values), with the first analysis using the NFL concentration from the first blood draw for each patient and the second analysis using the NFL concentration in the final blood draw for each patient. For each of these two analyses, we compared serum NFL concentrations between patients who did and did not receive the given treatment (only NFL values that were measured after the given treatment were used for patients in a given “treatment” group) using linear regression models. Owing to the very strong confounding influence of time from admission to blood draw, we first adjusted models for this variable alone and then subsequently additionally adjusted for age, sex, and BMI. Regression coefficients and 95% CIs were estimated.
We used a Bonferroni correction for multiple testing separately for each family of similar statistical tests. After applying this correction, \( P \) values <0.0025 were considered as statistically significant when evaluating associations of NFL with outcomes (four different outcomes were examined for five NFL variables), and \( P \) values <0.0063 were considered as statistically significant when assessing associations of NFL with COVID-19 treatment (four different treatments were examined for two NFL variables). \( P \) values <0.05 were considered as statistically significant in all other analyses. All statistical tests were two-sided and were performed using SAS (version 9.4; SAS Institute Inc.).
**SUPPLEMENTARY MATERIALS**
stm.sciencemag.org/cgi/content/full/13/602/eabi7643/DC1
Tables S1 and S2
View/request a protocol for this paper from Bio-protocol.
**REFERENCES AND NOTES**
1. COVID-19 Dashboard, https://coronavirus.jhu.edu/map.html.
2. H. A. Rothan, S. N. Byrareddy, The epidemiology and pathogenesis of coronavirus disease (COVID-19) outbreak. J. Autoimmun. 109, 102433 (2020).
3. L. Mao, M. Wang, S. Chen, Q. He, J. Chang, C. Hong, Y. Zhou, D. Wang, Y. Li, H. Jin, B. Hu, Neurological manifestations of hospitalized patients with COVID-19 in Wuhan, China: A retrospective case series study. medRxiv 2020.02.22.20026500 (2020).
4. A. Avula, K. Nalleballe, N. Narula, S. Sapozhnikov, V. Danud, S. Toorn, A. Glaser, D. Elsayegh, COVID-19 presenting as stroke. Brain Behav. Immun. 87, 115–119 (2020).
5. R. Bryouti, M. E. Adams, L. Benjamin, H. Cohen, S. F. Farmer, T. Y. Goh, F. Humphries, H. R. Jager, N. A. Losseff, R. J. Perry, S. Shah, R. J. Simister, D. Turner, A. Chandratheva, D. J. Werling, Characteristics of ischaemic stroke associated with COVID-19. J. Neurol. Neurosurg. Psychiatry 91, 889–891 (2020).
6. T. Gonzalez-Pinto, A. Luna-Rodriguez, A. Moreno-Estebanez, G. Agirre-Betita, A. Rodriguez-Antigued, M. Ruiz-Lopez, Emergency room neurology in times of COVID-19: Malignant ischemic stroke and SARS-CoV2 infection. Eur. J. Neurol. 27, e35–e36 (2020).
COVID-19 patients with and without neurological complications. K. Melmed, S. Meropol, A. B. Troxel, E. Petkova, T. Wisniewski, L. Balcer, C. Morrison, S. Yaggi, S. Galetta, A prospective study of long-term outcomes among hospitalized COVID-19 patients with and without neurological complications. J. Neurol. Sci. 426, 117486 (2021).
L. Mao, H. Jin, M. Wang, Y. Hu, S. Chen, Q. He, J. Chang, H. Zhang, T. Fang, J. Li, X. Long, Q. Zhang, X. Fang, X. Lv, D. Zhang, Y. Sun, N. Li, S. Hu, Z. Lin, B. Hu, Clinical time course of COVID-19, its neurological manifestation and some thoughts on its management. Stroke. 1. J. A. Frontera, D. Yang, A. Lewis, P. Patel, C. Medicherla, V. Arena, T. Fang, A. Andino, M. Greenway, Y. Erben, J. F. Huang, J. L. Siegel, C. J. Lamb, M. K. Badi, A. Sakusic, N. Gopal, L. Mao, H. Jin, M. Wang, Y. Hu, S. Chen, Q. He, J. Chang, H. Zhang, T. Fang, J. Li, X. Long, Q. Zhang, X. Fang, X. Lv, D. Zhang, Y. Sun, N. Li, S. Hu, Z. Lin, B. Hu, Clinical time course of COVID-19, its neurological manifestation and some thoughts on its management. Stroke. 1. J. A. Frontera, D. Yang, A. Lewis, P. Patel, C. Medicherla, V. Arena, T. Fang, A. Andino, M. Greenway, Y. Erben, J. F. Huang, J. L. Siegel, C. J. Lamb, M. K. Badi, A. Sakusic, N. Gopal, L. Mao, H. Jin, M. Wang, Y. Hu, S. Chen, Q. He, J. Chang, H. Zhang, T. Fang, J. Li, X. Long, Q. Zhang, X. Fang, X. Lv, D. Zhang, Y. Sun, N. Li, S. Hu, Z. Lin, B. Hu, Clinical time course of COVID-19, its neurological manifestation and some thoughts on its management. Stroke. 1. J. A. Frontera, D. Yang, A. Lewis, P. Patel, C. Medicherla, V. Arena, T. Fang, A. Andino, M. Greenway, Y. Erben, J. F. Huang, J. L. Siegel, C. J. Lamb, M. K. Badi, A. Sakusic, N. Gopal, L. Mao, H. Jin, M. Wang, Y. Hu, S. Chen, Q. He, J. Chang, H. Zhang, T. Fang, J. Li, X. Long, Q. Zhang, X. Fang, X. Lv, D. Zhang, Y. Sun, N. Li, S. Hu, Z. Lin, B. Hu, Clinical time course of COVID-19, its neurological manifestation and some thoughts on its management. Stroke. 1. J. A. Frontera, D. Yang, A. Lewis, P. Patel, C. Medicherla, V. Arena, T. Fang, A. Andino, M. Greenway, Y. Erben, J. F. Huang, J. L. Siegel, C. J. Lamb, M. K. Badi, A. Sakusic, N. Gopal, L. Mao, H. Jin, M. Wang, Y. Hu, S. Chen, Q. He, J. Chang, H. Zhang, T. Fang, J. Li, X. Long, Q. Zhang, X. Fang, X. Lv, D. Zhang, Y. Sun, N. Li, S. Hu, Z. Lin, B. Hu, Clinical time course of COVID-19, its neurological manifestation and some thoughts on its management. Stroke. 1. J. A. Frontera, D. Yang, A. Lewis, P. Patel, C. Medicherla, V. Arena, T. Fang, A. Andino, M. Greenway, Y. Erben, J. F. Huang, J. L. Siegel, C. J. Lamb, M. K. Badi, A. Sakusic, N. Gopal, L. Mao, H. Jin, M. Wang, Y. Hu, S. Chen, Q. He, J. Chang, H. Zhang, T. Fang, J. Li, X. Long, Q. Zhang, X. Fang, X. Lv, D. Zhang, Y. Sun, N. Li, S. Hu, Z. Lin, B. Hu, Clinical time course of COVID-19, its neurological manifestation and some thoughts on its management. Stroke.
Biohaven Pharmaceuticals Inc. (BHV4157-206 and BHV3241-301), and Neuraly Inc. (NLY01-PD-1) grants as well as a co-PI of the Mayo Clinic APDA Center for Advanced Research. L.P. is a consultant for Expansion Therapeutics. The other authors declare that they have no competing interests. **Data and materials availability:** All data associated with this study are present in the paper or the Supplementary Materials. Requests for clinical data should be addressed to L.P. and J.F.M. and will be made available through a data transfer agreement. This work is licensed under a Creative Commons Attribution 4.0 International (CC BY 4.0) license, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/. This license does not apply to figures/photos/artwork or other content included in the article that is credited to a third party; obtain authorization from the rights holder before using this material.
---
**Citation:** M. Prudencio, Y. Erben, C. P. Marquez, K. R. Jansen-West, C. Franco-Mesa, M. G. Heckman, L. J. White, J. A. Dunmore, C. N. Cook, M. T. Lilley, Y. Song, C. F. Harlow, B. Oskarsson, K. A. Nicholson, Z. K. Wszolek, L. J. Hickson, J. C. O’Horo, J. B. Hoyne, T. F. Gendron, J. F. Meschia, L. Petrucelli, Serum neurofilament light protein correlates with unfavorable clinical outcomes in hospitalized patients with COVID-19. Sci. Transl. Med. 13, eabi7643 (2021). | 2025-03-06T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
5771,
1
],
[
5771,
7926,
2
],
[
7926,
10898,
3
],
[
10898,
16432,
4
],
[
16432,
23299,
5
],
[
23299,
29017,
6
],
[
29017,
35319,
7
],
[
35319,
42791,
8
],
[
42791,
45726,
9
],
[
45726,
47254,
10
]
]
} |
827c3c419e28173d81eeeb9d53c55e6b753eee1a | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 15,
"total-fallback-pages": 0,
"total-input-tokens": 60633,
"total-output-tokens": 20794,
"fieldofstudy": [
"Physics"
]
} | Central galaxy growth and feedback in the most massive nearby cool core cluster
G. A. Ogrean1*, N. A. Hatch2,6, A. Simionescu3, H. Böhringer3, M. Brüggen1, A. C. Fabian4, N. Werner5
1Jacobs University Bremen, Campus Ring 1, D-28759 Bremen, Germany
2Leiden Observatory, Niels Bohrweg 2, 2333 CA, The Netherlands
3Max-Planck-Institut für extraterrestrische Physik, Giessenbachstrasse, D-85741 Garching, Germany
4Institute of Astronomy, Madingley Road, Cambridge CB3 0HA
5KIPAC, Stanford University, 382 Via Pueblo Mall, Stanford, CA 94305, USA
6School of Physics and Astronomy, The University of Nottingham, University Park, Nottingham NG7 2RD, UK
26 March 2010
ABSTRACT
We present multi-wavelength observations of the centre of RXCJ1504.1-0248 – the galaxy cluster with the most luminous and relatively nearby cool core at z ∼ 0.2. Although there are several galaxies within 100 kpc of the cluster core, only the brightest cluster galaxy (BCG), which lies at the peak of the X-ray emission, has blue colours and strong line-emission. Approximately 80 M⊙ yr⁻¹ of intracluster gas is cooling below X-ray emitting temperatures, similar to the observed UV star formation rate of ∼140 M⊙ yr⁻¹. Most star formation occurs in the core of the BCG and in a 42 kpc long filament of blue continuum, line emission, and X-ray emission, that extends southwest of the galaxy.
The surrounding filamentary nebula is the most luminous around any observed BCG. The number of ionizing stars in the BCG is barely sufficient to ionize and heat the nebula, and the line ratios indicate an additional heat source is needed. This heat source can contribute to the Hα-deduced star formation rates (SFRs) in BCGs and therefore the derived SFRs should only be considered upper limits.
AGN feedback can slow down the cooling flow to the observed mass deposition rate if the black hole accretion rate is of the order of 0.5 M⊙ yr⁻¹ at 10% energy output efficiency.
The average turbulent velocity of the nebula is vturb ∼ 325 km s⁻¹ which, if shared by the hot gas, limits the ratio of turbulent to thermal energy of the intracluster medium to less than 6 percent.
Key words: galaxies: clusters: general – galaxies: clusters: individual: RXCJ1504.1-0248 – cooling flows – ionized gas
1 INTRODUCTION
Cool core clusters are characterised by bright central peaks in their X-ray surface brightness profiles, cool and radially-increasing core temperatures, and short central cooling times. Unless the gas is reheated, the large amount of energy radiated in X-rays implies several hundred solar masses of gas cools below 10⁷ K every year (Cowie & Binney 1977; Fabian & Nulsen 1977). However, X-ray spectra from XMM-Newton and Chandra show cool-core clusters lack strong cooling lines, so less gas is cooling below X-ray emitting temperatures than is expected from the X-ray luminosity (Peterson et al. 2001, 2003; Böhringer et al. 2001; Peterson & Fabian 2006; McNamara & Nulsen 2007). The implication is that the intracluster medium (ICM) must be heated.
On the other hand, large quantities of cool gas and dust have been found in brightest cluster galaxies at the centre of cool-core clusters. Searches conducted at optical and infrared wavelengths reveal both warm and cool gas reservoirs (e.g. Edge 2001; Salomé & Combes 2003; Hatch et al. 2005; Jaffe, Bremer & Baker 2005; Johnstone et al. 2007). The BCGs in cool-core clusters generally have anomalous blue colours implying recent star formation, and emit in recombination and low-ionization lines (McNamara 1997; Peres
et al. 1998; Cavagnolo et al. 2008). These observations imply a relationship between the properties of the cluster and the central galaxy, but this interaction has not yet been understood or quantified precisely.
RXCJ1504.1-0248 is one of the most massive cool core clusters, with a prominent central X-ray brightness peak, and a short central cooling time. A classical cooling flow model leads to a mass deposition rate of $1400 - 1900 \, M_\odot \, yr^{-1}$, and the brightest cluster galaxy emits strong low ionization emission lines (Böhringer et al. 2005). These properties make this cluster as a prime target to study the relationship between the ICM, the brightest cluster galaxy, and a possible central active galactic nucleus (AGN).
Here we report results of a multi-wavelength study of the core of RXCJ1504.1-0248. It lies at $z = 0.2153$, which allows for a detailed spatial analysis of the BCG. We present results of integral field unit (IFU) spectroscopic observations aimed at understanding the morphology, kinematics and ionization state of the line-emitting nebula surrounding the BCG. We also present an analysis of X-ray spectra obtained with the XMM-Newton Reflection Grating Spectrometer (RGS) and European Photon Imaging Cameras (EPIC) and compare the ICM mass deposition rate to the star formation rate of the BCG. Sections 2 and 3 describe the observations and the data reduction process, while in section 4 we present the properties of the BCG, the nebula, and the cluster core cooling rate. In section 5 we combine the results from the multi-wavelength observations to quantify the relationship between the ionized gas, the BCG and the cluster core. We summarize our findings in section 6.
Unless otherwise stated, the following cosmological parameters have been used: $\Omega_m = 0.3$, $\Omega_\Lambda = 0.7$, $H_0 = 70 \, km \, s^{-1} \, Mpc^{-1}$. For a redshift $z = 0.2173$ (the redshift of the BCG), the scale is $3.519 \, kpc \, arcsec^{-1}$.
## 2 OBSERVATIONS
### 2.1 Optical images
We observed RXCJ1504.1-0248 on 2006 March 1 with the Visible Multi-Object Spectrograph (VIMOS; LeFevre et al. 2003) on the 8.2-m UT3 of the Very Large Telescope (VLT) at ESO Paranal in Chile, operated in imaging mode, using the R-band filter. The VIMOS field of view consists of four optical channels (also referred to as “quadrants”), which in imaging mode cover a field of view of approximately $7' \times 8'$. The CCD size is $2048 \times 2440$ px$^2$, therefore yielding a pixel size of 0.205 arcsec. Twelve scientific exposures were taken, each with an integration time of $\sim 90$ s.
### 2.2 Optical integral field spectroscopy
On 2006 April 5-6, the central galaxy of the cluster was also observed with VIMOS in Integral Field Unit (IFU) mode with the LR-red and LR-blue grisms. Both grisms provide a $54 \times 54$ arcsec$^2$ field of view, divided into 6400 optical fibres that offer a spatial sampling of 0.67 arcsec/fibre. The fibre to fibre distance on the sky is about 1 arcsec and the dead space between fibres is below $\sim 0.1$ arcsec. Thus, the optical fibres, coupled to an array of 6400 micro lenses, ensure a nearly continuous sky coverage. All quadrants of the spectrograph are divided into four pseudoslits, each organized into a $20 \times 20$ array of micro lenses and holding five $20 \times 4$ fibre modules.
The LR-red grism, used together with the OS-red filter, samples the wavelength range $5000 - 9150$ Å, with a dispersion of approximately $7.3 \, Å \, pixel^{-1}$ and a spectral resolution of 260. The LR-blue grism, used together with the OS-blue filter, covers the wavelength range $4000 - 6700$ Å, with a dispersion and spectral resolution of approximately $5.3 \, Å \, pixel^{-1}$ and 220, respectively. Eight 450-second exposures were taken in each setting, at averaged airmasses in the range $1.093 - 1.717$, and with a pointing dither of about $(\Delta \alpha, \Delta \delta) = (+3 \, arcsec, -3 \, arcsec)$. A summary of these observations is provided in Table 1.
### 2.3 X-ray observations
RXCJ1504.1-0248 was furthermore observed with XMM-Newton on 2007 January 22, for 39 ks. We extracted a lightcurve for each of the EPIC detectors separately and excluded the time periods in the observation when the count rate deviated from the mean by more than 3 $sigma$ in order to remove flaring from soft protons (Pratt & Arnaud 2002). We did not find significant time intervals affected by flares, such that after this cleaning, the net effective exposure was $\sim 37.4$ ks and $\sim 28.5$ ks for EPIC/MOS and EPIC/pn, respectively.
## 3 DATA REDUCTION
### 3.1 VIMOS Imaging
The imaging frames were reduced with the VIMOS Pipeline Recipes package, version 1.0, following all the steps described in section 5.5 of the VIMOS Pipeline User’s Guide$^1$, including flat fielding and bias subtraction. Flux calibration was applied using observations of standard stars in the L98 field$^2$ taken on the same night as the science frames.
### 3.2 VIMOS IFU
#### 3.2.1 Reduction of the IFU data cubes
The data reduction process was performed using the VIMOS Interactive Pipeline and Graphical Interface version 1.3 (Scodeggio 2005). Both the red and the blue datacubes underwent the usual VIMOS IFU reduction steps (Zanichelli 2005; Covone et al. 2006), including adjustment of the "first guess" parameters of the instrumental model in order to correct for optical and spectral distortions, spectra extraction, cosmic-ray cleaning, wavelength calibration and flux calibration. The LR-red data were calibrated using the standard
---
1. ftp://ftp.eso.org/pub/cpl/vimos/vimos-pipeline-manual.pdf
2. http://www3.cadc-ccda.hia-iha.nrc-cnrc.gc.ca/community/STETSON/standards/
star EG-274, observed immediately after the science observations were taken. Unfortunately, no standard star was observed at the same time as the observations taken with the LR-Blue grism. Therefore, the flux calibration of the LR-blue spectra was done using observations of the standard star LTT-3884 taken on 2006 May 27.
Each fibre of the IFU has a slightly different transmission. To obtain accurate fluxes we corrected for different transmission between fibers. We measured the flux in the 7246 Å sky line in the red data and 5577 Å sky line in the blue data, and the spectrum from each fibre was normalized so that the flux in the sky line equals the median of the sky line flux from all the fibres. At this stage, spectra from fibres with low transmission were removed as they were too noisy.
A sky spectrum was created for each fibre from the median of 20 nearby spectra in the red frame (or 30 fibres in the blue frame) which contained little or no line emission. These sky spectra were then subtracted from the calibrated spectra. Finally the spectra were trimmed to the wavelength range 3885–6225˚A and 5635–8348Å, for the blue and red gratings respectively.
The absolute flux calibration of the IFU data after the VIPGI reduction is uncertain as our total flux measurement of [OII] is a factor of ~ 2 lower than the slit spectroscopy of (Bühringer et al. 2005). Therefore we flux calibrate the IFU data using all stars and galaxies of known magnitude within the field-of-view. We find the SDSS magnitudes of these objects are consistent with those measured from our VIMOS image. The IFU spectra of the stars and galaxies are convolved with g′, r′ and i′ and VIMOS R-band filters, and the flux calibration is determined using SDSS magnitudes. This self-consistent calibration results in spectra which are 2.5 times brighter than spectra calibrated using VIPGI.
To increase the signal-to-noise ratio the spectra were binned using a Voronoi-2D binning algorithm \(^3\) (Cappellari & Copin 2003). We use the weighted Voronoi tessellation as laid out by Diehl & Statler (2006) to have a signal-to-noise ratio of 15 in the H\(\beta\) line flux measured in the blue frame.
### 3.2.2 Spectral emission line fitting
The emission lines were measured by fitting a linear continuum plus Gaussian model profile to each line using the \texttt{IDL} package MPFIT \(^4\), which performs a Levenberg-Marquardt least-squares fit.
The doublet [OII] \(\lambda\lambda3727, 3729\) is never resolved, therefore it is fit with a single Gaussian. The H\(\alpha\) line and the nearby [NII] \(\lambda\lambda6548, 6583\) doublet are resolved and the complex is fitted with three Gaussians, with the integrated flux of the [NII] \(\lambda6548\) line fixed to be a third of the [NII] \(\lambda6584\) line. Similarly [OII] \(\lambda6363\) is fixed to be a third of the flux of [OIII] \(\lambda4360\), and [OIII] \(\lambda5004\) is fixed at a third of the flux of [OIII] \(\lambda5007\).
In the red wavelength domain, the spectra are strongly affected by fringing beyond 7000 Å. Consequently the signal-to-noise ratio at the wavelength of H\(\alpha\) is not as high as that achieved at the wavelength of [OII].
---
\(^3\) using \texttt{IDL} codes from http://www-astro.physics.ox.ac.uk/~mxc//idl/#binning
\(^4\) http://cow.physics.wisc.edu/~craigm/idl/fitting.html
Figure 1. Extinction map \([E(B - V)]\) of RXJ1504 derived from the ratio of \(H\alpha/H\beta\) measured in each bin. The centre of the galaxy is marked by a cross. Black bins indicate areas where [OII] is detected, but H\(\beta\) is either not detected or the \(H\alpha/H\beta\) ratio implies \(E(B - V) = 0\).
#### 3.2.3 Extinction along the line of sight
We assume that there are two sources of dust which lead to extinction and reddening of the spectra. The first source is a screen of foreground dust at the redshift of the BCG and the other source is dust within our Galaxy. We first correct the spectra for Galactic extinction of \(E(B-V) = 0.108\) (Schlegel et al. 1998), then blue-shift the spectra to the rest-frame of the BCG, and correct for extinction caused by dust in the BCG.
The extinction caused by dust within the BCG is determined from the observed \(H\alpha/H\beta\) ratio. The \(H\alpha/H\beta\) flux is a good indicator of extinction along the line of sight, as shorter wavelengths are scattered stronger by dust particles. Therefore, in dusty regions, values of \(H\alpha/H\beta\) are significantly above the theoretical ratio of 2.87 derived by Osterbrock & Ferland (2006) for Case B recombination at \(T_e = 10,000\) K. We assume the \(H\alpha\) flux is depleted due to 2Å of \(H\beta\) stellar absorption (the mean found in a sample of star forming galaxies, Buat et al. 2002) and we correct the \(H\alpha/H\beta\) ratio for this before calculating the extinction.
The \(E(B - V)\) colour excess in each bin was computed using the equations from Moustakas, Kennicutt Jr. & Tremonti (2006),
\[
E(H\beta - H\alpha) \equiv -2.5 \log \left[ \frac{(H\alpha/H\beta)_{\text{theo}}}{(H\alpha/H\beta)_{\text{obs}}} \right]
\]
\[
E(B - V) \equiv \frac{E(H\beta - H\alpha)}{k_{H\beta} - k_{H\alpha}},
\]
where \(k_{H\beta} - k_{H\alpha} = 1.278\) for \(R_V = 3.2\). The spatial distribution of the extinction is plotted in Fig. 1. The central region of the galaxy is dustier than the outskirts, but the dust is patchy and there is no overall smooth pattern. The spectra in each bin was blue-shifted to the rest-frame and then corrected for extinction using the equation of Calzetti et al. (2000) with the \(R_V = 3.2\) extinction law.
3.3 Reduction of the XMM-Newton data
We reduced the XMM-Newton data using the 7.0.0 version of the Science Analysis System (SAS); the standard analysis methods using this software are described in e.g. Watson et al. (2001).
We subtracted the instrumental background from the EPIC spectra using closed-filter observations which were normalized to match our observation using the count rates in the hard energy band (10 – 12 keV for MOS, 12 – 14 keV for pn) outside of the field of view (OoFoV). Out-of-time events were subtracted from the EPIC/pn data using the standard SAS prescription for the extended full frame mode. The cosmic X-ray background (CXB) obtained from spectra extracted from an annulus between 9.5' and 12.5' (outside the cluster’s r500) is modeled with three components: two thermal components to account for the local hot bubble (LHB) emission (kT = 0.08 keV) and for the Galactic halo (GH) emission (kT ~ 0.2 keV), as described by Kuntz & Snowden (2000), and a power-law to account for the integrated emission of unresolved point sources and for possible contamination from the residual soft proton particle background. Since different detectors can be affected differently by soft protons, we leave the power-law indices and normalizations free between EPIC/MOS1, MOS2 and pn. The temperature of the LHB was frozen to 0.08 keV, while the temperature of the GH and the spectrum normalizations of the thermal components were free in the fit, but constrained to be the same for all three EPIC detectors. The spectra were fit in the 0.4–7. keV energy band.
The RGS spectra were extracted following the method described by Tamura et al. (2001). We modeled the background using the standard background model available in SAS (rgsbkgmodel, González-Riestra et al. 2000). The cluster spectra were extracted from a region which is 2' wide in the cross-dispersion direction of the instrument. The line emission observed with the RGS from extended sources is broadened by the spatial extent of the source along the dispersion direction. In order to account for the line broadening in the spectral modeling, we convolve the line spread function (lsf) model with the surface brightness profile of the source along the dispersion direction. In order to account for the line broadening in the spectral modeling, we convolve the line spread function (lsf) model with the surface brightness profile of the source along the dispersion direction. We fit the 1st order spectra in the wavelength band of 8–25˚ at the 1σ level.
Freezing the Galactic absorbing column density (nH) to 6 × 10^{20} cm^{-2}, as determined from Dickey & Lockman (1990) from 21 cm observations, gives very poor fits to the cluster EPIC spectra. Therefore, we froze the nH to 8.7 × 10^{20} cm^{-2}, the best-fit value determined by fitting EPIC spectra of the central 5' in combination with ROSAT All-Sky Survey (RASS) spectra to better constrain the emission in the soft X-ray band. This value agrees within the 90% confidence interval with the nH value computed from the 100 μm IR data, using the Boulanger et al. (1996) IR-nH correlation function (more details will be provided by Moïs et al., in prep).
| Table 2. XMM-Newton spectral fit results for the central region of RXCJ 1504.1-0248. The abundance values are given with respect to solar abundances (Lodders 2003). Errors are quoted at the 1σ level. |
|-----------------|-----------------|-----------------|
| kT (keV) | O/Fe | Ne/Fe | Mg/Fe | Si/Fe | S/Fe | Fe | M (M⊙/yr) |
| RGS | EPIC | RGS | EPIC | RGS | EPIC | RGS | EPIC |
| 8.8^{+1.3}_{-1.0} | 5.6^{+0.06}_{-0.05} | 0.47 ± 0.25 | 0.47 (fixed) | 0.71 ± 0.35 | 0.71 (fixed) | 0.57 ± 0.32 | 0.57 (fixed) | 0.79 ± 0.12 | 0.53 ± 0.15 | 1.77^{+0.94}_{-0.56} | 0.47^{+0.01}_{-0.01} | 68^{+57}_{-56} | 78^{+26}_{-19} |
| χ^2 / d.o.f. | 467/481 | 3903/2494 |
4 RESULTS
4.1 The mass deposition rate from X-ray spectra
To determine the amount of ICM gas cooling out of the X-ray emitting temperature range, we extracted the EPIC spectrum from a circle of radius 140 kpc (0.67') centred on the cluster centre. This is the cooling radius obtained by Böhringer et al. (2005) for a Hubble constant H₀ of 70 km s^{-1} Mpc^{-1}.
The spectrum is fitted with a Galactically absorbed single temperature collisionally ionized optically thin mekal plasma model plus a classical cooling flow model (with the lower cutoff temperature fixed to 0.1 keV). To ensure a correct propagation of errors due to uncertainties in background determination, the EPIC source spectra are fit in parallel with the CXB spectra obtained from an outer annulus. The normalization of the cluster spectral components were set to zero for the CXB data sets, while the normalizations of the background models were fixed for the source analysis to a ratio corresponding to the relative sizes of the extraction regions for the source and CXB only.
The metal abundances were coupled between the single temperature and cooling flow models both for fitting the EPIC and the RGS data. The O/Fe, Ne/Fe and Mg/Fe ratios were fixed in the EPIC fit based on the best-fit values obtained from the RGS spectra, since the energy resolution and effective area calibration of EPIC around the energy of the O, Mg and Ne lines are too poor to allow a reliable determination of their abundances with EPIC alone, especially for such a hot cluster. The Si and S abundances of the hot plasma can only be measured with EPIC because the emission lines of these elements lie outside the RGS energy band. The fit results are summarized in Table 2. Errors are quoted at the 1σ level. Abundance ratios are given in proto-solar units (Lodders 2003).
The high luminosity and the highly peaked surface brightness distribution of the cooling core in RXCJ1504.1-0248 allow us to obtain relatively sensitive XMM-Newton RGS spectra of this cluster despite its large distance. We show in Fig. 2 the spectrum obtained by combining data from the two RGS detectors. This is one of the most distant galaxy cluster spectra obtained with RGS. The cause for the discrepancy between the temperatures measured with EPIC and RGS is the different spectral extraction region (the RGS
The cool-core of RXCJ1504.1-0248
extraction region is effectively 10′-long in the dispersion direction and 2′-wide in the cross-dispersion direction, and the temperature profile of the cluster rises steeply outside the cooling radius of 0.67″. Because of the higher temperature measured by RGS, the measured absolute abundances are also high, but the abundance ratios are typically robust.
The best-fit values of the mass deposition rates from the EPIC and RGS data are 78 and 68 M⊙ yr⁻¹ respectively. The 3σ upper limits are 156 and 239 M⊙ yr⁻¹, respectively.
The so-called “cooling flow problem” remains in that the mass deposition rates obtained from X-ray spectra are significantly less than those inferred from the X-ray luminosity of between 1400 and 1900 M⊙ yr⁻¹ for $H_0 = 70$ km s⁻¹ Mpc⁻¹. This implies that significant heat input is needed to reduce the cooling from 1400–1900 down to at most 150–240 solar masses per year (∼10%). The bolometric luminosity calculated by Böhringer et al. (2005) is $4.3 \times 10^{45}$ ergs s⁻¹, out of which more than 70% ($L_{Xc} = 3 \times 10^{45}$ ergs s⁻¹) is emitted from inside the cooling radius. Assuming 10% of the gas is cooling and forming stars, more than 90% of $L_{Xc}$ must thus be supplied by a heat source, amounting to $2.7 \times 10^{45}$ ergs s⁻¹.
4.2 Galaxies within the cluster core
The cluster core contains the BCG and a number of fainter galaxies marked B1–B6 in the VIMOS R-band image (Fig. 3). Spectra of these galaxies are shown in Fig. 3 and the BCG spectrum in Fig. 4. The light from B1 fell on some bad IFU fibres so we were not able to extract a spectrum. All spectra were corrected for Galactic extinction of $E(B-V) = 0.108$ (Schlegel et al. 1998) and the BCG was corrected for additional intrinsic extinction of $E(B-V) = 0.211$ derived from the measured $Hα/Hβ$ ratio of 3.68 (assuming 2A of Hβ absorption). The BCG emits strongly in recombination and low ionization lines, and the observed and extinction-corrected emission-line ratios are given in Table 4.
Redshifts of galaxies B2–B6 were measured using the strong stellar absorption lines Ca ii H and K near 4000Å. These lines are not visible in the BCG spectrum so the [OⅡ] emission line is used instead. Redshifts are listed in Table 3, together with the strength of the 4000Å break (D4000) and the rest-frame $B - R$ colour.
Galaxies B2, B3, B5 and B6 lie within a few hundred km s⁻¹ of the redshift of the BCG therefore these galaxies are situated within the cluster. B4 is blue-shifted by almost 2000 km s⁻¹ compared to the BCG and is either a galaxy in the infall region of the cluster or a foreground galaxy.
D4000 is defined as the ratio of the average continuum between the rest-frame 4050–4250Å and 3750–3950Å (Bruzual A. 1983). However, this wavelength range includes strong emission lines visible in the spectrum of the BCG. To remove emission-line contamination from the D4000 measurement, we alter our definition for the BCG to the ratio between rest-frame 4115–4250Å and 3750–3840Å.
The strength of this break, D4000, is typically greater than 1.6 in elliptical galaxies (Kauffmann 2003), even if the elliptical galaxies reside at redshifts of up to $z \sim 1.1$ (Pasquale et al. 2006). The cluster galaxies B2–B6 have D4000 values typical of ellipticals, but the BCG has a much lower D4000, similar to late-type galaxies.
B2 is the only galaxy, apart from the BCG, to exhibit [OⅡ] line emission. Although it lies close to the emission-line nebula that surrounds the BCG, the kinematics of the emission line gas (see section 4.4.4) imply the line is emitted from gas within B2 and not from the BCG nebula.
4.3 The brightest cluster galaxy
4.3.1 Mass
The BCG of RXCJ1504.1-0248 is a massive galaxy: its $K_{mag}$ is 13.130 (2MASS). The mass-to-light ratios of Cappellari et al. (2006) imply a stellar mass of $\sim 6 \times 10^{12}$ M⊙. This is twice the mass of the nearby d galaxy M87. Although no K-correction has been applied, the galaxy is red at these wavelengths with $H - K = 0.546$, so the K-corrected $K_{mag}$ may be even larger. The dominant source of error in the derived mass is likely to be the large scatter in the mass-to-light ratio.
Figure 3. **Left:** $R$–band VIMOS image of the entire field-of-view of the VIMOS IFU centered on the brightest cluster galaxy (labelled BCG). A number of other nearby bright galaxies have been labelled B1–B6. **Right:** Spectra of galaxies B2–B6. Most absorption features above 6000 Å are due to atmospheric absorption. Stellar absorption Ca K and H lines are detected in all galaxies and [Oii] is seen in the spectrum of B2.
Figure 5. Continuum images of the BCG from left to right: (a) unsharp-mask of the $R$–band VIMOS image showing the NE and SW filament; (b) continuum from the rest-frame $\sim5400$ Å in log units of erg cm$^{-2}$ s$^{-1}$ Å$^{-1}$ arcsec$^{-2}$; (c) continuum from the rest-frame $\sim3200$ Å in log units of erg cm$^{-2}$ s$^{-1}$ Å$^{-1}$ arcsec$^{-2}$; (d) map of the [OII] emission with colourbar in log scale in units of erg cm$^{-2}$ s$^{-1}$ arcsec$^{-2}$; (e) rest-frame $B - R$ continuum colour and (f) D4000. The noisy regions (where the flux per binned region is less than the rms noise) are coloured black in (b)-(d) and white in regions (e) and (f). The galaxy B2 is located at (-20,0) in the IFU images.
The cool-core of RXCJ1504.1-0248
Galaxy | BCG | B2 | B3 | B4 | B5 | B6
---|---|---|---|---|---|---
Redshift | 0.2173 | 0.2176 | 0.2151 | 0.210 | 0.221 | 0.216
Velocity shift relative to BCG (km s\(^{-1}\)) | – | 65 | –550 | –1885 | 800 | –385
Projected distance (kpc) | – | 24.6 | 26.3 | 70.7 | 68.8 | 85.5
D4000 | 1.28 (1.11)\(^2\) | 1.77 | 3.84 | 1.79 | 2.85 | 1.80
\(B - R\) colour\(^1\) (mag) | –0.12 (–0.40)\(^2\) | 0.0 | 0.13 | 0.43 | 0.43 | 0.23
Table 3. Properties of the brightest cluster galaxy and galaxies nearby in projection. Redshifts are determined from the Ca II H and K absorption lines. Projected distances are given in kpc, as observed from the dust-uncorrected continuum. \(B - R\) colours are measured from regions free from any emission line contamination. \(^1\) The \(B - R\) colour is the magnitude difference between 5600-5800 Å and 6900-7100Å. \(^2\) Values in the brackets are derived from the extinction-corrected spectrum.
Figure 6. Emission line ratios of the nebula of the BCG from left to right: [O\textsc{iii}]/H\(\beta\), [O\textsc{ii}]/H\(\beta\), [N\textsc{ii}]/H\(\alpha\). Binned regions are only displayed where [O\textsc{iii}], H\(\beta\), [N\textsc{ii}] and H\(\alpha\) fluxes are greater than 5\(\times\)10\(^{-17}\) erg cm\(^{-2}\) s\(^{-1}\) arcsec\(^{-2}\). All scales are in log.
Figure 7. Optical spectrum of RXCJ1504.1-0248 with emission-lines masked out (black solid line) and the best-fit models overplotted in the dashed line. The top panel shows a 2-component fit to the SED consisting of an elliptical galaxy template plus a B3V stellar template. The coloured lines show the relative luminosity from each component. The bottom panel shows a 3-component fit to the SED consisting of an elliptical galaxy template, an O5V stellar template and a B8V stellar template. Both fits give similar reduced-\(\chi^2\) and are thus equally likely.
Table 4. Emission line ratios for the observed and the extinction-corrected spectrum of the BCG. The ratio of H\(\alpha\)/H\(\beta\) is greater than 2.87 due to our assumption that the H\(\beta\) emission line is depleted due to 2Å of stellar absorption.
4.3.2 Stellar populations
The BCG has a different spectral shape compared to the other cluster galaxies (B2–B6). Whereas the cluster galaxies have red \(B - R\) colours and high D4000, typical of elliptical galaxies, the BCG has blue \(B - R\) colours and shallow D4000. These spectral features, in addition to the strong emission lines, means the BCG hosts a strong source of ultraviolet (UV) emission such as an AGN or a young stellar population.
We determine the stellar composition of the BCG by modeling the optical spectral energy distribution (SED). Because the BCG is so massive we assume the underlying galaxy consists of an old stellar population which has the
SED of an elliptical galaxy. The elliptical template from the Kinney-Calzetti Atlas (Kinney et al. 1996) is used to model this component.
The blue continuum was modeled by a power-law continuum (in which the slope and normalization were free parameters) or early-type stellar templates (O5V, O9V, B0V, B3V, B5-7V, B8V and A0V) from the Pickles Atlas (Pickles 1998). The normalization of the elliptical and stellar templates were free parameters.
The emission lines were masked for the least-squared fitting procedure. Since the red part of the spectrum is contaminated by fringing, we only modeled the continuum between 3400–5600˚A and assign constant errors to the entire wavelength range of the optical SED. The errors were adjusted to obtain a reduced $\chi^2$ of approximately $\chi^2_{red} \sim 1$ for the best fit model.
The elliptical plus power-law model resulted in a poor fit to the BCG SED ($\chi^2_{red}=13$) as the break between 3600–4000˚A was not adequately fit. Thus the blue continuum is not due to an AGN component.
The best fit to the data, with a $\chi^2_{red}=1.8$, resulted from the combination of the elliptical template and a B3V stellar template. This fit is shown in the top panel of Fig. 7 together with the relative contributions of each component to the galaxy’s luminosity.
An elliptical plus O-star template is rejected with a $\chi^2_{red}=6.5$. The O-star template is similar to the AGN power-law SED, and the poor fit results from the same inadequate fit to the 3600-4000˚A break.
An O5V star emits 5 orders of magnitude more ionizing flux than a B3V star, thus to estimate the maximum ionizing flux allowed by the optical SED, we fit the optical SED with a 3 component model: an elliptical template, plus an O5V stellar template, and an additional stellar template from the O5V to A0V range. The best fit to the data, with a $\chi^2_{red}=1.5$, results from the elliptical + O5V + B8V model. The bottom panel of Fig. 7 shows the model overplotted on the optical SED of the BCG, and the relative contributions of each component.
The difference in $\chi^2_{red}$ between these two models results from the poor fit to the Balmer absorption lines. The observed SED contains strong emission lines so we are unable to fit any absorption lines to the data. The B3V stellar template has stronger absorption lines than the 3-component model and thus results in a larger $\chi^2_{red}$. We therefore cannot distinguish which model is more plausible from the $\chi^2_{red}$.
In Table 5 we list the relative flux contributions from the 2 and 3 component fits, and the number of ionizing stars (stellar types O5V and B3V) that lie in the BCG. The monochromatic 4500˚A luminosities of an O5V and B3V star are $1.1 \times 10^{34}$ and $4.8 \times 10^{32}$ erg s$^{-1}$, respectively (Kurucz 1993). B8V and later stellar types that compose the elliptical galaxy template emit negligible amounts of ionizing photons.
For both models approximately 60% of the light at 4500˚A comes from the young stellar population and only 40% comes from the elliptical galaxy. However, at longer wavelengths the light from the old stellar population dominates, whilst at 3400˚A the young population emits $\sim 90\%$ of the galaxy’s luminosity.
The H$\alpha$ luminosity resulting from stellar photoionization is calculated from eq.1 in Allen (1995), using the Panagia (1973) ionizing fluxes and assuming unity covering fraction. The H$\alpha$ luminosity expected for the 2- and 3-component models is $6 \times 10^{40}$ and $3.4 \times 10^{43}$ erg s$^{-1}$, respectively.
The total extinction-corrected H$\alpha$ luminosity of the BCG nebula is $3.4 \times 10^{43}$ erg s$^{-1}$, therefore between 0.2-100% of the H$\alpha$ emitted by the galaxy is ionized by the young stellar population.
The 3-component fit gives an upper limit to the amount of ionizing photons emitted by the young stellar population because it is forced to contain the largest possible number of O5V stars and a covering fraction of unity is assumed. Therefore the H$\alpha$ luminosity from this model is also an upper limit. So whilst it is possible that all the observed H$\alpha$ results from stellar photoionization, there is ample room for additional sources to contribute to the ionization of the nebula.
### 4.3.3 Distribution of the stellar populations and nebula
Fig. 5a displays an unsharp-mask image of the BCG and B2, created by subtracting a smoothed image from the VIMOS R-band image. The galaxy is not smooth, but contains a bright filament that traverses the prominent nuclear region and extends NE to SW across the galaxy.
The IFU data is used to visualize the same R-band continuum without any emission line contamination from the H$\beta$, [Oiii] and [Nii] emission lines (Fig. 5b). The centre of the BCG is estimated from this image and marked by a cross. The red continuum is centrally concentrated in an elliptical shape and there are no bright filaments extending southwest and northeast. Therefore the large filament must be due to the line emission that falls within the R-band passband. The nearby galaxy B2 is also visible at position (−20, 0).
Fig. 5c shows the BCG as seen in the shortest wavelength emission measured in the IFU spectra (3885–4030˚A), which in the rest-frame of the cluster is $\sim$3190–3310˚A, falling approximately in the U-band. This wavelength range does not include any bright emission lines so the light is emitted from the young stellar population. Both NE and SW filaments are prominent in the blue continuum image, implying they are the locations of the recent star formation.
Fig. 5d maps (in [Oii]) the emission line nebula that surrounds the BCG and shows that both the NE and SW filaments are clearly visible. The brightest region is the galaxy nucleus, and the general shape of the nebula follows the blue continuum.
We highlight the differences between the blue and red continuum emission with Figs.5e and f which display the rest-frame $B – R$ continuum, and the strength of the 4000˚A break, D4000. Fig. 5e shows that the BCG generally has blue colours, but the nucleus and SW filament are clearly bluer than the rest of the galaxy. We note that the colour variations cannot all result from the applied extinction correction. The highest H$\alpha$/H$\beta$ ratios were observed in the nuclear region (see the E($B − V$) map in Fig. 1), which translates into a large extinction correction. Therefore the enhanced blue colours from the nucleus may result from an excessive extinction correction, however, the dust is patchy and did not extend along the SW filament.
The variation in D4000 across the BCG is shown in Fig. 5f. D4000 is greater than 1.5 in the eastern region of
E + O5V + B8V model gives an upper limit on the Hα with D4000 < 1.3, much lower than observed in elliptical galaxies. D4000 is low in the nuclear region (~1.1) and decreases smoothly down the SW filament of young stars to the western tip of the emission line nebula, 42 kpc away. D4000 is low in these regions, independent of the extinction correction, supporting the above finding that the central and SW parts of the galaxy host the young stars. In summary, the old stellar population of the BCG lies within an elliptical shape. Extending beyond this smooth profile are two filaments, extending both NE and SW, which contain young, blue stars. These filaments are visible in the U-band continuum and in line emission. There is a smooth decrease in the strength of D4000 for 42 kpc along the SW filament implying progressively younger populations along the filament, or a larger fraction of younger stars compared to the underlying older stellar population.
### 4.4 The ionized nebula surrounding the BCG
#### 4.4.1 Luminosity
The flux in the [OIII] line from the entire ionized nebula is 9.8 × 10^{-14} erg s^{-1} cm^{-2} so it is more luminous than any nebula in the Crawford et al. (1999) sample of BCGs (however it should be noted that the Crawford et al. 1999 fluxes luminosities are measured within slits and may be incomplete). Correcting the spectrum for reddening and extinction boosts the [OIII] luminosity to 4.7 × 10^{43} erg s^{-1} making it the most luminous nebula around a BCG yet observed.
A Galaxy Evolution Explorer (GALEX) GR4 observation gives $M_{NUV} = 18.78$ and $M_{FUV} = 18.36$ (AB) for the BCG. In both NUV and FUV images, the emission is spatially extended along the direction of the filament. The observed FUV flux is brighter than expected given the NUV flux as dust extinction generally produces NUV>FUV. Since the Lyα emission line is shifted into the FUV passband, it is likely that the observed FUV emission is enhanced because of contamination from Lyα.
The UV slope β, defined as $f_{\lambda} = \lambda^\beta$, ranges between −2.1 and −2.6 for a galaxy which has been continuously forming stars for more than a Gyr (Leitherer et al. 1999). Almost all other stellar population models result in a higher β, therefore we can assume that $\beta > -2.6$ in this BCG. Thus from the observed NUV flux, we derive an excess flux of $1.3 × 10^{-13}$ erg cm^{-2} s^{-1} within the observed FUV passband. Correcting for Galactic and intrinsic extinction, we derive a lower limit of the Lyα emission of $3 × 10^{-12}$ erg cm^{-2} s^{-1}, translating into a luminosity of $3.8 × 10^{44}$ erg s^{-1}. This is approximately a factor of 10 greater than the Hα flux, and results in a Lyα/Hα ratio similar to the theoretical case A prediction.
#### 4.4.2 Mass and volume-filling factor
The X-ray derived temperature and central electron density of the ICM are $kT \approx 5.6$ keV and $n_{e0} = 0.13$ cm^{-3} (see Table 2 and Böhringer et al. 2005). In a pure hydrogen, completely ionized gas approximation, the ICM pressure is therefore $\sim 2 \times 10^{-9}$ dyn cm^{-2}. Thus, assuming hydrostatic equilibrium between the 10,000 K nebula and the ICM, the electron density inside the nebula is $n_{e} \sim 850$ cm^{-3}.
The mass of the ionized gas is given by
$$M = \frac{L_{H\alpha} m_p}{n_{e} \alpha_{H\alpha}^0 h \nu_{H\alpha}},$$
where $n_{e}$ is the electron density, $\alpha_{H\alpha}^0$ is the effective recombination coefficient and $h \nu_{H\alpha}$ is the energy of a photon at the frequency of Hα (Osterbrock & Ferland 2006). For $n_{e} \sim 850$ cm^{-3}, the equation above yields a mass of approximately $2.5 \times 10^7$ M_{⊙}.
Assuming this mass is distributed inside a sphere, the radius of the sphere is $\sim 50$ pc. On the other hand, the emission in this limited region is spread over a circle of radius $\sim 20$ kpc. Assuming again a spherical distribution, the gas fills a volume of $\sim 3.4 \times 10^{13}$ pc³. Hence, the filling factor is approximately $1.5 \times 10^{-8}$, a value comparable to those found by Hatch et al. (2007) for four BCGs at different redshifts. This low filling fraction suggests either a clumpy or filamentary distribution of the optically-emitting gas.
#### 4.4.3 Source of ionization
Shocks, conduction from the X-ray emitting ICM and UV ionization from hot young stars or the AGN may all play a role in heating and ionizing the gas. But the line emission may also be heated by secondary electrons from more energetic particles or cosmic rays, proposed by Ferland et al. (2009) and shown to reproduce the observed line ratios of BCG nebulae. Here we study the emission line ratios in order to determine the dominant ionization and excitation source of the gas.
Baldwin, Phillips & Terlevich (1981) designed a method to discriminate between Hii regions, power-law ionization and shock heating. The BPT diagrams (Baldwin et al. 1981)
| Model | % E | % O5V | % B3V | % B8V | no. of ionizing stars $\times 10^6$ | Hα flux $10^{43}$ erg s^{-1} |
|--------------|-----|-------|-------|-------|-----------------------------------|-------------------------------|
| E + B3V | 37 | – | 63 | – | 596 | 0.006 |
| E + O5V + B8V| 43 | 27 | – | 30 | 4.5 | 3.4 |
Table 5. Stellar populations in the BCG. The optical SED of the BCG is best fit by an old stellar population (modeled by an elliptical galaxy template, E), plus a large component of young stars having a B3V SED (E + B3V). In a 3-component fit which is forced to include an O5V stellar population, the SED is best fit by a composition of an elliptical galaxy, plus O5V and B8V stars (E + O5V + B8V). The percentage of light emitted at 4500Å by each component is given in columns % E, % O5V, % B3V and % B8V. The number of ionizing stars is the number of O5V or B3V stars, and column 7 lists the Hα flux resulting from photoionization by these stars. The E + O5V + B8V model gives an upper limit on the Hα flux produced by stellar photoionization.
we use are displayed in Fig. 8. The nebula exhibits a variety of ratios depending on the location, but no region is dominated by AGN heating, and few regions are dominated by stellar UV ionization. Rather, these diagnostic diagrams show that the gas is ionized by multiple sources. The spatial distribution of the $\text{[NII]}/\text{H}\alpha$, $\text{[OII]}/\text{H}\beta$, and $\text{[OIII]}/\text{H}\beta$ are shown in Fig. 6. We use these figures to explore whether different ionization sources are located in certain regions.
The $\text{[OIII]}/\text{H}\beta$ flux ratio, a good indicator of the ionization parameter, decreases radially towards the outskirts of the nebula, but is slightly offset from the centre of the galaxy. The $\text{[OIII]}$ may be ionized by the AGN in the nucleus. This distribution is not shared by any other line ratio, but is similar to the distribution of the $\text{[OII]}$ flux.
The $\text{[OII]}/\text{H}\beta$ ratio is relatively constant around the nucleus and along the protruding filament. South of this filament, the ratio suddenly increases to $\gtrsim 0.7$, while East from the nucleus and towards B2 it decreases below 0.5. These low $\text{[OII]}/\text{H}\beta$ ratios are consistent with stellar UV photoionization. If an AGN dominates the ionization in the central part of the BCG, not only would we expect higher $\text{[OII]}/\text{H}\beta$ ratios at that position, but also a gradual decrease of this line ratio towards the outskirts, where photoionization from massive stars should surmount AGN ionization. This is not observed, so we rule out this hypothesis.
The $\text{[OII]}/\text{H}\beta$ flux is higher towards the South, suggesting the existence of a harder, non-stellar excitation source in this region. Fig. 9 displays the X-ray surface brightness contours from Börhringer et al. (2005) overplotted on top of the VIMOS image. Two X-ray maxima are evident: one overlapping the BCG center, the other coinciding with the region of high $\text{[OII]}/\text{H}\beta$. Thus this hard ionization source is also visible in X-rays.
East of the nucleus, the $\text{[NII]}/\text{H}\alpha$ ratio varies rapidly between -0.35 and 0.15 (in logarithmic space), which is too large to be due to stellar photoionization. Possible heating mechanisms include shock heating or power-law ionization. To the West, however, this flux ratio is consistently low up to a distance of $\sim 15$ kpc from the centre; then there is a slight increase again in the SW tip of the nebula. The region of low $\text{[NII]}/\text{H}\alpha$ flux ratios coincides with that of a low D4000 and a blue B-R colour, suggesting the gas in this region is ionized by UV radiation from young stars. However the BPT diagram (Fig. 8a) indicates that this region is heated by more than one source, so an additional source competes with the UV from young stars to ionize the gas.
Fig. 10 shows the three line-ratios as a function of line flux. The line ratios saturate at high luminosities, whereas fainter regions are characterized by more variability. This trend has also been observed in other luminous BCGs by Wilman et al. (2006) who postulate that the line ratios saturate when star formation is proceeding at such a high rate that stellar UV dominates the photoionization of the gas.
In conclusion stellar UV is a plausible ionizing source, but it cannot act alone: harder ionizing sources seem to be also present. The stellar UV ionization is concentrated along the SW filament, whilst the ionizing sources of the nuclear
---
**Figure 8.** Diagnostic emission-line diagrams showing $\text{[NII]} \lambda 6584/\text{H}\alpha$ vs. $\text{[OIII]} \lambda 5007/\text{H}\beta$ and $\text{[OII]} \lambda 3727/\text{H}\beta$ vs. $\text{[OIII]} \lambda 5007/\text{H}\beta$ in logarithmic space. Separation lines used to differentiate between AGN, composite and stellar ionization are from Kewley et al. (2006) and Lamareille et al. (2004), respectively.
**Figure 9.** VIMOS R-band image with overplotted contours of the X-ray surface brightness from a Chandra image of Börhringer et al. (2005).
and other regions remain unknown. The ionizing source of the nuclear region is of particular importance as this region is brightest in both the U-band and line emission. Our data point towards ionization by multiple sources, including stellar UV.
4.4.4 Gas kinematics
Fig. 11a presents the line-of-sight kinematics of the nebula, as derived from the relativistic Doppler shifts of the [OII] emission lines. All velocities were computed with respect to the median redshift of the [OII] λ3727 emission (z = 0.2173).
The galaxy is broken up into distinct regions. The galaxy B2 has a gas velocity of −50 km s$^{-1}$, and is clearly distinct from the redshifted gas that lies to the East of the BCG and in the nuclear region, which has a velocity between 0 and 220 km s$^{-1}$. To the West of the nuclear region the gas is blue-shifted up to −220 km s$^{-1}$, although at the westernmost tip of the nebula the gas velocity again changes direction and is red-shifted by 200 km s$^{-1}$. The central part of the SW filament is blue-shifted, whilst the SW tip is redshifted, thus the filament must either be stretching or collapsing on itself. Similar kinematics can be seen in NGC 1275 in the Perseus cluster (Hatch et al. 2006), and in the models of Pope et al. (2008). The length of the SW filament is 42 kpc, implying a dynamical age of $\sim 10^8$ yr, so the line-emitting gas must be long-lived.
The dispersion map of the gas is displayed in Fig. 11b. All line-widths are $\sigma$-widths, and not FWHM (full width at half maximum). The $\sigma$-widths of the [OII] and Hβ lines used for creating the dispersion map of the gas were corrected for instrumental broadening of 375 km s$^{-1}$ as measured from the nearby 5755Å skyline. White bins in this figure are regions in which the spectral resolution is too low to determine a dispersion. A peak in the dispersion is seen slightly West of the nuclear region. The dispersion here is 250-300 km s$^{-1}$ whilst the median velocity dispersion is $\sim 200$ km s$^{-1}$. The SW filament has the lowest velocity dispersion. These strong internal motions may result from the superposition of distinct components along the line of sight.
\[
\begin{array}{cccccc}
\text{Line} & \mathcal{F} & \text{L} & \text{SFR} & \text{Ext.-cor.} & \text{Ext.-cor.} \\
\text{[OII]} & 9.8 & 1.3 & 83 & 36.9 & 4.7 & 314 \\
\text{Hα} & 14 & 1.8 & 141 & 26.2 & 3.4 & 262 \\
\text{FUV} & 1.10 & - & 15.1 & 11.8 & - & 136 \\
\end{array}
\]
\textbf{Table 6.} Fluxes ($\mathcal{F}$) are stated in units of $10^{-14}$ erg cm$^{-2}$ s$^{-1}$. FUV flux density is in units of $10^{-27}$ erg cm$^{-2}$ Hz$^{-1}$. Luminosities (L) are stated in units of $10^{43}$ erg s$^{-1}$ and star formation rates (SFR) are given in units of M$_{\odot}$ yr$^{-1}$.
### 4.5 Star formation rate of the BCG
The FUV light from a galaxy is dominated by emission from young O and B type stars so we can use the rest-frame FUV emission to estimate the star formation rate. This estimate of the SFR does not assume that the line emission has been ionized by the young stellar population. Since the NUV and FUV emission is spatially extended, approximately in the same direction as the filament, it is likely to be emitted from a spatially extended source, such as stars, rather than the central active nucleus.
Assuming all of the 1800Å flux is emitted from a population of hot young stars, we use the observed NUV emission (equivalent to the rest-frame 1870Å) to obtain a star formation rate. The NUV flux of $1.1 \times 10^{-27}$ erg cm$^{-2}$ Hz$^{-1}$ is corrected for Galactic extinction of $E(B - V) = 0.108$ and internal extinction of $E(B - V) = 0.211$. We use the relation between star formation rate and rest-frame FUV flux (observed NUV flux) derived by Salim (2007) assuming a Salpeter mass function.
\[
\text{SFR}_{\text{FUV}} = 1.08 \times 10^{-28} L_{\text{FUV}} M_{\odot} \text{yr}^{-1}. \tag{4}
\]
This equation is calibrated for FUV emission centered on 1528Å, whilst the rest-frame wavelength of the light observed through the NUV filter is 1866Å. However Salim (2007) show that this will have a negligible effect on the conversion between FUV luminosity and star formation rate. Star formation rates derived from the observed NUV flux are given in Table 6. Kennicutt (1998) derive a relation that results in 30% greater star formation rate for a given NUV flux.
The Hα and [OII] luminosity may also be used to estimate the star formation rate in the BCG under the assumption that the gas is ionized by UV photons from hot young stars. Table 6 lists the total flux, extinction corrected fluxes and luminosities of both [OII] and Hα, together with the derived star formation rates. Star formation rates (SFR) are computed from the [OII] and Hα luminosities according to the empirical relation of Kewley, Geller & Janssen (2004),
\[
\text{SFR}_{\text{Kewley}}^{\text{[OII]}} = 6.58 \times 10^{-42} L_{\text{[OII]}} M_{\odot} \text{yr}^{-1}, \tag{5}
\]
and the theoretical equation of Kennicutt (1998),
\[
\text{SFR}_{\text{Kewley}}^{\text{Hα}} = 7.8 \times 10^{-42} L_{\text{Hα}} M_{\odot} \text{yr}^{-1}, \tag{6}
\]
respectively, where $L_{\text{[OII]}}$ and $L_{\text{Hα}}$ are expressed in units of erg s$^{-1}$.
The FUV-derived star formation rates (calculated from the observed NUV emission) are 2–3 times lower than the rates derived from the Hα and [OII] emission lines. In sec-
tion 4.3.2 we show the number of O and B stars present in the BCG is barely sufficient to power the Hα emission. The discrepancy between the star formation rates derived from FUV and emission line luminosity reinforces this conclusion and implies that the [Oii], Hα and other emission lines are heated by an additional source. Star formation rates of BCGs derived from line emission must only be considered upper limits of the true star formation rates.
5 DISCUSSION
5.1 Brightest cluster galaxy growth
We present a number of observations which imply the BCG is rapidly forming stars. These include extended FUV emission, blue galaxy colours atypical of elliptical galaxies, shallow D4000, and some line ratios indicating stellar UV as a possible ionization source. The IFU continuum and line ratio maps indicate this star formation occurs mainly in the galaxy core and in a 42 kpc filament stretching from the core to the SW.
The rest-frame FUV emission is likely to be the most reliable estimator of the star formation rate, thus setting it at \( \sim 136 \, M_\odot \, yr^{-1} \). This approximately agrees with the mass deposition rate derived from the X-ray spectra (\( \sim 80 \, M_\odot \, yr^{-1} \)). Therefore, the current star formation may be fueled by condensing intracluster gas.
The current star formation rate is large, and together with the BCG of Abell 1835, it is the largest star formation rate of any low-redshift elliptical galaxy. However, this current rate is considerably less than the average rate at which stars must have formed over the galaxy’s entire lifetime. The specific star formation rate of the BCG (the ratio of current SFR to the mass of galaxy) is \( \sim 2.3 \times 10^{-11} \, yr^{-1} \) and is therefore similar to many elliptical galaxies which are detected in the UV. Had the current rate of star formation lasted since \( z = 1 \), the galaxy would have increased its mass by less than 15%. If this phase of star formation should only lasted a few tens of Myrs, the galaxy would return to its pre-starburst colours in a short time after the end of the burst. Thus, while the ongoing star formation clearly influences the galaxy’s colours and ionization state of the gas, its impact on the mass of the galaxy is not significant.
5.2 Heating of the nebula
The optical emission line ratios indicate stellar photoionization dominates in the centre of the galaxy and along the star forming filaments. The strong blue continuum and shallow D4000 in these regions indicate the presence of ionizing O and B stars.
However, the BPT diagnostic diagrams show stellar photoionization cannot be the exclusive ionization source and at least one additional mechanism needs to be present.
Both Hα and [Oii] line luminosities estimate higher star formation rates relative to the FUV derived rate. The number of ionizing O and B stars in the BCG, inferred from SED fitting, is barely sufficient to produce all the Hα flux emitted. This also implies that some of the line emission is ionized by an additional source. Thus the Hα or [Oii] flux should only be used to calculate an upper limit of the star formation rate. We have also shown that dust extinction can be large in BCGs, hence the observed line flux will be greatly reduced, and star formation rates derived from single emission lines should be considered highly uncertain.
Line ratio maps are a good means to detect where the dominant ionization source changes. The best ratio for this work is [Oii]/Hβ because the [Nii] and Hα lines lie in a wavelength region that is badly affected by fringing, so the ratio of these lines is unreliable. The [Oii]/Hβ map clearly marks the change in dominant ionization source in the South, which coincides with a second X-ray peak. None of the line ratio maps show a correlation with the map of recent star formation (Fig.5c), providing further evidence that stellar photoionization is not the only ionizing source at work. This result does not come as a surprise and it has also been observed in other BCGs (e.g. Sabra et al. 2000; Hatch et al. 2007). Ferland et al. (2009) have proposed an alternative particle heating method which reproduces the line ratios.
5.3 AGN Feedback
The mass deposition rate allowed by the X-ray spectra is less than 10 percent of the mass deposition rate of 1400-1900 M⊙ yr⁻¹ implied by the total X-ray luminosity emitted from the cluster core. Since we have an indication that 10 percent of the ICM is cooling and forming stars, it follows that a heating source must supply 90 percent of the luminosity emitted from within the cooling radius, which is 2.7 × 10⁴⁵ erg s⁻¹. If this is supplied by AGN feedback from the central black hole, the mass accretion rate must be 0.5 M⊙ yr⁻¹, assuming at least a 10 percent output efficiency.
The K-band magnitude of a galaxy correlates with the mass of the supermassive black hole (SMBH) at its centre (e.g. Marconi & Hunt 2003; Graham 2007):
\[ \log \left( \frac{M_{\text{BH}}}{M_\odot} \right) = -0.33 \left[ M_K + 24 \right] + 8.33, \tag{7} \]
where \( M_{\text{BH}} \) is the mass of the black hole and \( M_K \) is the K-band magnitude of the BCG. With \( M_K = -27.0 \) (2MASS), it follows the BCG has a SMBH of mass \( \sim 2 \times 10^9 \) M⊙.
To estimate the power supplied by an AGN, we first consider Bondi accretion from the hot gas (Bondi 1952). The Bondi accretion radius is \( \sim 12 \) pc. As for all SMBH except perhaps the one in M87, this is far beyond the resolution limit of our instruments and we can only speculate about the physical conditions close to the black hole. The black hole will accrete mass at a rate \( M_{\text{Bondi}} = \pi \lambda c_\sigma r_A^2 \), where \( \lambda \) depends on the adiabatic index (here \( \gamma = 5/3 \), \( \lambda = 0.25 \); Bondi 1952), \( c_\sigma \) is the sound speed, \( \rho \) is the gas density and \( r_A \) is the accretion radius. For the central parameters determined from the X-ray data, namely \( kT = 5.6 \) keV and \( n_e = 0.13 \) cm⁻³, we obtain \( M_{\text{Bondi}} \approx 5 \times 10^{-4} \) M⊙ yr⁻¹.
However, this temperature and density represent an average over a large extraction region used for the X-ray spectral analysis, therefore the true temperature and density near the black hole could be very different. If we assume that the thermal gas pressure at the centre is no less than the thermal pressure of the inner region measured spectrally to be at a temperature of 5.6 keV, then \( r_\sigma \) scales as \( 1/\Gamma \) (in practice the extra weight will make it go steeper than this). Therefore, \( M_{\text{Bondi}} \propto 1/\Gamma^{2.5} \). So to increase \( M_{\text{Bondi}} \) by 10³, to 0.5 M⊙ yr⁻¹, requires only a drop in temperature by a factor of 16 which means to 350 eV. Many nearby cool core clusters show a cool component at \( \sim 0.5 \) keV (e.g., Sanders et al. 2010) and we cannot rule out such a low temperature in the centre of RXCJ1504.1-0248.
One could also try to infer the AGN power output over the past 10⁸ years from the radio power at 1.4 GHz using the relation of Birzan et al. (2008):
\[ \log P_{\text{jet}} = \left( 0.35 \right) \log P_{\text{1400}} + 1.9. \tag{8} \]
In the relation above, \( P_{\text{1400}} \) represents the bolometric radio luminosity for the total source at 1.4 GHz in units of 10²⁴ W Hz⁻¹, and \( P_{\text{jet}} \) is the jet/cavity power in units of 10⁴² erg s⁻¹.
The BCG contains a radio source with a brightness of 62 mJy at 1.4 GHz (Bauer et al. 2000). At a luminosity distance of 1075.5 Mpc, this implies a radio power of 8.6 × 10²⁴ W Hz⁻¹, thus a jet power of 1.5 × 10⁴⁴ erg s⁻¹. The corresponding black hole accretion rate at 10 percent efficiency is 0.03 M⊙ yr⁻¹ and an order of magnitude smaller than the power required to heat the ICM. However, the relation found by Birzan has a scatter of 0.85 dex (factor of \( \sim 7 \)) at 1400 MHz, so that our object lies a little over 1σ away from the mean (\( \sim 1.4\sigma \) in log).
In conclusion, we cannot rule out that heating is supplied by a central AGN that accretes at a rate of \( \sim 0.5 \) M⊙ yr⁻¹.
If the accretion rate has been as high as 0.5 M⊙ yr⁻¹ since redshift 1, then the black hole has grown by an amount of \( \sim 3 \times 10^9 \) M⊙, which is larger than the total estimated mass of \( \sim 2 \times 10^9 \) M⊙. This implies that the current accretion rate is atypically high, which suggests that the present phase in the feedback cycle with the high star-formation and accretion rate must be relatively short lived. Furthermore, BCGs may harbor supermassive black holes with masses a few times higher than those predicted by the \( M_{\text{BH}} = M_K \) relation of Graham (2007), as is the case in Abell 1836 (Dalla Bontà et al. 2009).
5.4 Turbulent energy of the ICM
Assuming the velocity dispersions and the line-of-sight velocities of the cold gas are solely due to turbulence gives an upper limit on the turbulent velocities in the nebula. Despite the large density contrast between the ICM and the optical line-emitting gas, the low filling factor suggests that the nebula gas may be blown about by the hot gas and thus serves as a good tracer of ICM motions (Fabian et al. 2003). The speed of sound in the hot central plasma at a temperature of 5.6 keV is roughly 1190 km s⁻¹ and the characteristic velocity of isotropic turbulence is \( \sigma_{\text{turb}} = \sqrt{2 \left( v_A^2 + v_s^2 \right)} \), where \( v_A \) is the dispersion of the velocity shear of the cold gas, measured from the \( \sigma \)-width of the line-of-sight velocity histogram, \( v_s \approx 117 \) km s⁻¹, and \( v_s \) is the average velocity dispersion, \( v_s \approx 200 \) km s⁻¹. \( \sigma_{\text{turb}} \) provides an upper limit on the velocity dispersion of the hot gas, \( \sigma_{\text{turb}} \approx 325 \) km s⁻¹ and therefore the turbulence in the central ICM has a typical Mach number of less than 0.27. We thus obtain an upper limit on the ratio of turbulent to thermal energy in the ICM, given by \( (\gamma/2) M^2 \) with \( \gamma = 5/3 \) the adiabatic index of monatomic ideal gas, of roughly 6%, in agreement with upper limits derived from other methods (Churazov et al. 2008; Werner et al. 2009).
6 CONCLUSIONS
RXCJ1504.1-0248 is one of the most extreme cooling flows ever discovered. The observations presented here offer the first multi-wavelength view of the cluster core, examining both the hot and warm gas as well as the stars of the BCG.
The core of RXCJ1504.1-0248 has an X-ray luminosity of at least \( 3 \times 10^{45} \) erg s⁻¹, approximately 10 percent of which is supplied by gas cooling and forming stars. Accretion onto a supermassive black hole of \( 2 \times 10^9 \) M⊙ is the likely mechanism needed to account for most of the remaining \( 2.7 \times 10^{45} \) erg s⁻¹ heating. The corresponding Bondi accretion rate at 10 percent radiative efficiency is 0.5 M⊙ yr⁻¹. Using the correlation between 1.4 GHz radio luminosity and jet power (Birzan et al. 2008), we infer a jet power of \( 1.5 \times 10^{44} \) erg s⁻¹, still insufficient to heat the ICM.
The intracluster medium mass deposition rate is in
agreement with the star formation rate of the BCG derived
through FUV observations ($\sim 140 M_\odot$ yr$^{-1}$). The IFU images show that most of the star formation occurs in the
galaxy core and in a 42 kpc-long filament that stretches out
from the core to the SW. These regions have blue continuum
colours, and shallow 4000 Å breaks consistent with recent
star formation. Additionally, the $[O\text{II}]/H\alpha$ line ratio
indicates stellar photoionization is the dominant source of
ionization along the filament. The X-ray emission also extends
along the filament indicating a direct connection between
the intracluster medium and the recent star formation.
The line-emitting nebula that surrounds the BCG is the
most luminous observed to date, with extinction-corrected
$[O\text{II}]$ and H$\alpha$ luminosities of $4.7 \times 10^{43}$ erg s$^{-1}$ and $3.4 \times 10^{43}$ erg s$^{-1}$ respectively. The number of ionizing O and B
The nebula kinematics are ordered and the nebula comprises
of a number of kinematically distinct regions. The ve-
clocities are generally low ($< 250$ km s$^{-1}$), and there is no
evidence for rotation or free-falling gas infall. The 42 kpc
filament has a velocity shear of $\sim 400$ km s$^{-1}$ resulting in
a dynamical time of 10$^8$ years, thus the nebula must be
long-lived. The total velocity dispersion of the nebula is
$v_\sigma \sim 200$ km s$^{-1}$ and the dispersion of the velocity shear
of the cold gas is $v_\Delta \sim 117$ km s$^{-1}$. These set an upper limit of
$\sim 325$ km s$^{-1}$ on the mean turbulent velocity in the ICM,
meaning that the ratio of turbulent to thermal energy of the
intracluster medium is limited to less than 6 percent.
Our results suggest mass transfers from the intracluster
medium to the brightest cluster galaxy at a rate of approxi-
mately $10^4 M_\odot$ yr$^{-1}$. The cooling of the intracluster medium results in a gas reservoir in the brightest cluster galaxy. The rate at which gas condenses into this reservoir is similar to the rate at which this reservoir forms stars, so the reservoir
should be approximately stationary. Our observations can-
not reveal the coolest gas components of this reservoir, but
we find that the $10^4$ K gas is partly heated by a source other
than the forming stars.
ACKNOWLEDGMENTS
We thank the referee for many helpful comments. The VLT-
VIMOS Integral Field Unit data presented in this paper were
reduced using the VIMOS Interactive Pipeline and Graphical
Interface (VIPGI) designed by the VIMOS Consortium. We thank Bianca Garilli and Luigi Paioro for their assist-
ance in using VIPGI. This work is partly based on obser-
vations obtained with XMM-Newton, an ESA science mis-
ion with instruments and contributions directly funded by
ESA member states and the USA (NASA). NAH acknowled-
ges funding from the Royal Netherlands Academy of Arts
and Sciences. HB acknowledges support through the Cluster
of Excellence “Origin and Structure of the Universe”, funded
by the Excellence Initiative of the German Federal Gover-
ment as EXC project 153. AS and MB acknowledge support
by the DFG Schwerpunkt programme SP1177. NW is sup-
ported by the National Aeronautics and Space Adminis-
tration through Einstein Postdoctoral Fellowship Award
Number PF8–90056 issued by the Chandra X-ray Observatory
Center, which is operated by the Smithsonian Astrophysical
Observatory for and on behalf of the National Aeronautics
and Space Administration under contract NAS8–03060.
REFERENCES
Allen S. W., 1995, MNRAS, 276, 947
Baldwin J. A., Phillips M. M., Terlevich R., 1981, PASP, 93, 5
Bauer F. E., Condon J. J., Thuan T. X., Broderick J. J., 2000, ApJS, 129, 547
Börhringer H., Belsote E., Kennea J., Matsushita K., Molendi S., Worrall D. M., Mushotzky R. F., Ehle M.,
Guainazzi M., Sakellion I., Stewart G., Vestrand W. T., Dos Santos, 2001, A&A, 365, L181
Börhringer H., Burwitz V., Zhang Y.-Y., Schuecker P., Nowak, 2005, ApJ, 633, 148
Bondi H., 1952, MNRAS, 112, 195
Boulanger F., Abergel A., Bernard J.-P., Burton W. B., Desert F.-X., Hartmann D., Lagache G., Puget J.-L.,
Hartmann D., 1996, A&A, 312, 256
Brzuza A. G., 1983, ApJS, 53, 497
Buat V., Boselli A., Gavazzi G., Bonfanti C., 2002, A&A, 383, 801
Calzetti D., Armus L., Bohlin R. C., Kinney A. L., Koornneef J., Storchi-Bergmann T., 2000, ApJ, 533, 682
Cappellari M., Bacon R., Bureau M., Damen M. C., Davies R. L., de Zeeuw P. T., Emellem E., Falcón-Barroso J.,
Krajnović D., Kuntschner H., McDermid R. M., Peletier R. F., Sarzi M., van den Bosch R. C. E., van de Ven G.,
2006, MNRAS, 366, 1126
Cappellari M., Copin Y., 2003, MNRAS, 342, 345
Cavagnolo K. W., Donahue M., Voit G. M., Sun M., 2008, ApJ, 683, L107
Churazov E., Forman W., Vikhlinin A., Tremaine S., Ger-
hard O., Jones C., 2008, MNRAS, 388, 1062
Covone G., Kneib J.-P., Soucail G., Richard J., Jullo E.,
Ebeling H., 2006, A&A, 456, 409
Cowley L. D., Binney J., 1977, ApJ, 215, 723
Crawford C. S., Allen S. W., Ebeling H., Edge A. C., Fabian A. C., 1999, MNRAS, 306, 857
Dalla Bontà E., Ferrarese L., Corsini E. M., Miralda-
Escude J., Coccato L., Sarzi M., Pizzella A., Beifiori A.,
2009, ApJ, 690, 537
Dickey J. M., Lockman F. J., 1990, ARA&A, 28, 215
Diehl S., Statler T. S., 2006, MNRAS, 368, 497
Edge A. C., 2001, MNRAS, 328, 762
Fabian A. C., Nulsen P. E. J., 1977, MNRAS, 180, 479
Fabian A. C., Sanders C. S., Crawford C. S., Conselice C. J.,
Gallagher J. S., Wyse R. F. G., 2003, MNRAS, 344, L48
Ferland G. J., Fabian A. C., Hatch N. A., Johnstone R. M.,
Porter R. L., van Hoof P. A. M., Williams R. J. R., 2009, MNRAS, 392, 1475
González-Riestra R., 2004, http://xmm.vilspa.esa.es/docs/documents/CAL-TN-
0058-1-0.ps.gz
| 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3550,
1
],
[
3550,
9284,
2
],
[
9284,
14899,
3
],
[
14899,
21091,
4
],
[
21091,
25303,
5
],
[
25303,
26448,
6
],
[
26448,
29273,
7
],
[
29273,
36004,
8
],
[
36004,
42136,
9
],
[
42136,
46249,
10
],
[
46249,
51664,
11
],
[
51664,
55859,
12
],
[
55859,
62848,
13
],
[
62848,
68522,
14
],
[
68522,
68522,
15
]
]
} |
d8bb25933b885c9a6107ffcf3cabd3183dd30b3d | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 4,
"total-fallback-pages": 0,
"total-input-tokens": 11363,
"total-output-tokens": 3152,
"fieldofstudy": [
"Engineering"
]
} | Cogeneration Plant Optimization
N W Mitiukov1,2, S V Spiridonov3,4, and G Z Samigullina3
1Udmurt Federal Research Center of the Ural Branch of the RAS, 426067, Izhevsk, Russia
2International Network Center for Fundamental and Applied Research, 20036 Washington (DC), USA
3Udmurt State University, 426034, Izhevsk, Russia
4Chuvash State University named after I.N. Ulyanov, 428015, Cheboksary, Russia
E-mail: [email protected]
Abstract. As the analysis of publications on cogeneration shows, most authors present it as a way to increase the efficiency of existing power generation plants by utilizing the generated heat. It is shown in the work that cogeneration is a complex system that allows maneuvering the proportion between the received thermal and electrical energy. In this regard, it seems promising to develop a cogeneration system in which this proportion could be regulated. There have not been such systems in the published patents yet. Regulation of the proportion between heat and electricity can be carried out in accordance with the changing tariffs for these two types of energy, as well as depending on production tasks. Existing cogeneration systems are usually adaptations of existing power plants optimized for optimal electrical energy consumption. Cogeneration systems should be optimized for the maximum total consumption of heat and electricity at a minimum cost. Apparently, in the conditions of existing calculation methods, this problem has not been solved yet. Therefore, the optimal values for cogeneration should be sought outside the currently accepted design parameters and operating modes of power plants.
1. Introduction
Recently, cogeneration systems have been considered as an essential element of saving energy resources. Cogeneration is a technology that represents a single process of producing heat and electricity.
In the work R. R. Absalyamova with co-authors examines the issues of obtaining cogeneration effect for serial gas turbine, operating in conditions of the Far North [1]. A. E. Yutuaev with co-authors suggests using cogeneration to utilize associated gas in the development of coal deposits [2]. At the same time, the authors allow this effect to be obtained both during the operation of piston installations and turbine ones. Obviously, there are prospects for the introduction of cogeneration in small enterprises producing polymer containers, since these productions are characterized by a large consumption of electrical and thermal energy [5].
As a rule, the economic effect of cogeneration is obtained as a result of reducing losses during transportation of thermal energy. For a similar process, electrical energy does not have such a problem, its losses are significantly less. In this regard, low-power turbines and piston machines of low power, which provide energy to relatively small industrial and residential areas, are of great
importance. In this case, the loss of transportation of thermal energy is minimal – it is consumed where it was produced. In this regard, for example, V. N. Dorofeev sees great prospects from the use of cogeneration when using mini-thermal power plants for autonomous power supply of residential buildings or small industries [3]. It is obvious that cogeneration also has great prospects in agriculture [6], since such enterprises are usually far from major energy producers.
However, the smaller the power plant, the longer the payback period from the application of the cogeneration effect. Therefore, at present, low-power plants practically do not use it [4]. This is the main paradox of cogeneration – it is not used where it could give the maximum effect.
2. Cogeneration as a way to maneuver energy flows
Previously, we proposed a new method to assess the efficiency of implementation a cogeneration plant based on changing tariffs for different types of energy [7]. Its essence lies in the fact that the existing methods of justifying investments in the fuel and energy sector of a small enterprise, which are laborious and time-consuming, nevertheless do not take into account the specifics of operating cogeneration plants.
In order to understand the effect of introducing cogeneration in a small enterprise, one needs to know the average annual and monthly consumption of different types of energy. In general, we can talk about cogeneration, trigeneration, etc., but in relation to cogeneration, this is thermal and electrical energy.
The average monthly calculation of thermal consumption, if it is not involved in technological processes, can be carried out not during 12 months, but only during 8 months, since in the summer thermal energy is most often not required. This, of course, reduces the efficiency of use and increases the payback period.
When calculating the implementation, it is usually forgotten that the cogeneration plant is an energy converter, and in fact it can regulate the balance of the received electrical and thermal energy in a certain range. And then to assess the efficiency of implementation and assess future profits, it is necessary to know the existing tariffs. These may be tariffs \( T \) for:
- diesel fuel \( T_1 \),
- electricity \( T_2 \),
- water for technological needs \( T_3 \),
- heat (as for water heated to a certain temperature, for example, to 95°C) \( T_4 \),
- tariffs for emissions of harmful substances \( T_5 \).
And then the possible profit from the implementation of the cogeneration plant \( P \) will be determined according to an obvious formula, depending on the volume of resource consumption or energy production \( V_i \):
\[
P = \Sigma (V_i T_i).
\]
The "plus" is placed if the resource is produced, and "minus" is put if it is consumed.
Taking into account seasonal fluctuations or fluctuations in tariffs in the long term, you can choose an individual cycle of the cogeneration plant, changing the proportion between the electrical and thermal energy.
Unfortunately, the patent search revealed the absence of any systems that could regulate the proportion between the generated electrical and thermal energy, which makes it difficult to maneuver freely when using cogeneration.
3. Modernization of existing equipment
In addition to introducing new equipment, a cogeneration plant can also be obtained by reequipping an existing one. It offers great prospects for small enterprises, since such modernization will be much cheaper in any case.
In our opinion, there are two possible ways to modernize, for example, gas-piston installations. First, an additional cogeneration effect can be obtained by replacing the classic radiator. In the basic
configuration, the air blown through the radiator grille is heated and goes out. When modernizing, it should be replaced with a conventional liquid-to-liquid heat exchanger and used for heating the coolant.
The second way is less obvious, but it can also have an effect. This is the removal of heat from exhaust gases. In this case, a heat exchanger is also placed on the output collector, but it is of the gas-liquid type. This way is more difficult, since the exhaust gases have high acidity, and as a result, the heat exchanger must be performed in an acid-resistant version.
The prospects of work in each of these two ways can be assessed by the thermal balance of the engine. Thus, a diesel engine without turbocharging transfers approximately 35% of the supplied heat into mechanical energy. 34% goes into the cooling system, 27% - with exhaust gases. The rest is unaccounted for losses. For a turbocharged engine, this ratio changes by 40%, 29%, and 30%, respectively [9]. It would seem that a large percentage of heat which goes out with exhaust gases makes it promising to remove heat there. But it should be remembered that at the outlet of the heat exchanger, the gas temperature should not allow condensation of water vapor, which will have a low pH during condensation, and as a result, high aggressiveness. This significantly reduces the possibility of heat removal. And thus, the radiator modernization will be the most effective.
It is also possible to use water from the circuit of the engine itself as a coolant. It can save heat during conversion, but it is unlikely to contribute to technological efficiency – the liquid from both circuits may have different requirements.
4. Cogeneration plant optimization during design
As the review of the designs of currently used cogeneration plants showed [10, 11], these are mostly conventional power units that have additional cogeneration units. However, this approach, although it makes it possible to create a cogeneration plant with minimal technological costs, is hardly advisable.
The fact is that the criterion for the quality of the power plant is the maximum of electricity produced with a minimum of fuel costs. And this pattern is not observed for cogeneration. Thus, it is appropriate to optimize the plant taking into account the maximum amount of heat and electricity with a minimum of fuel costs.
At Izhevsk State Technical University, a student S. A. Pershin tried to find the optimal value of the sum of thermal and mechanical energy on the example of a QSX15G8 diesel on the basis of the calculation algorithm used in the design of diesel engines [8]. It turned out that with an increase in the diameter of the piston for all permissible revolutions (he varied them in the range from 1500 to 1800 rpm), the heat transferred to the coolant was constantly growing. This is quite a predictable result, because in this case, the overall engine power is growing as well. He also received quite predictable results in terms of the number of revolutions and the cetane number of fuel used.
As for varying the stroke of the piston, the results showed that with its increase over the entire range of shaft revolutions, the heat removed increases, but the value of the first derivative decreases. It can be predicted that with a further increase in the stroke of the piston, the heat removed could reach an extreme, but its finding, if it really exists, is already beyond the applicability of the method used. In this connection, it can be assumed that for optimal plant cogeneration, it makes sense to use cylinders with an increased stroke, but not with the stroke accepted in the engine industry.
Thus, it should be admitted that the optimal values of the cogeneration plant may lie outside the currently accepted ranges of optimal power plants. Because of this, it may be necessary to reconsider the calculation methods, which have a large number of empirical coefficients that are not suitable for the range of the optimal cogeneration plant.
5. Conclusions
A cogeneration plant is not just a system for utilizing emitted heat. This is a complex system that allows maneuvering the proportion between the received thermal and electrical energy. In this regard, it is promising to develop a cogeneration system that could change this proportion. There have not been such systems in the published patents yet. Regulation of the proportion between heat and
electricity can be carried out in accordance with the changing tariffs for these two types of energy, as well as depending on production tasks.
Existing cogeneration systems are usually adaptations of existing power plants optimized for optimal electrical energy consumption. At the same time, cogeneration systems should be optimized for the maximum consumption of heat and electricity at a minimum cost. Under the existing calculation methods, this problem does not seem to have a solution. Therefore, the optimal values for cogeneration should be sought outside the currently accepted design parameters and operating modes of power plants.
6. References
[1] Absalyamov R R, Kuzichkin N V, Lisitsyn N V 2012 Izvestiya Sankt-Peterburgskogo gosudarstvennogo tehnologicheskogo instituta 17(43) pp 97-102
[2] Yutuaev A E, Belyaev V V, Agafonov V V 2013 Gorny informacionno-analitichesky bulletin 6 pp 69-74
[3] Dorofeev V N 2011 Santehnika, Otoplenie, Kondicionirovanie 9(117) pp 88-93
[4] Borovikov V M, Borodina O A 2010 Akademia energetiki 5(37) pp 92-71
[5] Tarasyuk K 2014 Santehnika, Otoplenie, Kondicionirovanie 8(152) p 89
[6] Voitenok I A, Barbashin S E 2018 Inzhinernye sistemy i sooruzhenia 3(32) pp 18-24
[7] Belosludtsev I S, Mitiukov N W 2013 Vestnik Izhevskogo gosudarstvennogo tehnicheskogo universiteta 3 pp 75-76
[8] Pershin S A 2014 Novyi universitet. Ser. Tehnicheskie nauki 5-6 (27-28) pp 82-95
[9] Melisarov V M at al 2011 Teplovoi raschet i teplovoi balans dizelnogo dvigatelya bez nadduv i s nadduvom (Tambov)
[10] Nurmuhamatov T F 2018 European Journal of Renewable Energy T 1 3 pp 3-8 DOI: 10.13187/ejre.2018.1.3
[11] Belosludtsev I S 2011 Vestnik Kamskogo instituta gumanitarnyh i inzhenerynyh tehnologiy 5(18) pp 54-62
Acknowledgment
The article was prepared with the support of the Integrated Program for Basic Research Ural Branch of Russian Academy of Science, № 0427-2019-0019. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2903,
1
],
[
2903,
6636,
2
],
[
6636,
11073,
3
],
[
11073,
12983,
4
]
]
} |
b824c9a48294327d8f49831ae27038fd915b3443 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 13,
"total-fallback-pages": 0,
"total-input-tokens": 30218,
"total-output-tokens": 9653,
"fieldofstudy": [
"Physics"
]
} | Memory Effect and Triplet Pairing Generation in the Superconducting Exchange Biased Co/CoOx/Cu41Ni59/Nb/Cu41Ni59 Layered Heterostructure
V.I. Zdravkov1,2, D. Lenk1, R. Morari1, A. Ullrich1, G. Obermeier1, C. Müller1, H.-A. Krug von Nidda1, A.S. Sidorenko2, S. Horn1, R. Tidecks1, and L.R. Tagirov1,3
1Institut für Physik, Universität Augsburg, D-86159 Augsburg, Germany
2D. Ghitsu Institute of Electronic Engineering and Nanotechnologies ASM, MD2028 Kishinev, Moldova
3Solid State Physics Department, Kazan Federal University, 420008 Kazan, Russia
We fabricated a nanolayered hybrid superconductor-ferromagnet spin-valve structure, the resistive state of which depends on the preceding magnetic field polarity. The effect is based on a strong exchange bias (about -2 kOe) on a diluted ferromagnetic copper-nickel alloy and generation of a long range odd in frequency triplet pairing component. The difference of high and low resistance states at zero magnetic field is 90% of the normal state resistance for a transport current of 250 μA and still around 42% for 10 μA. Both logic states of the structure do not require biasing fields or currents in the idle mode.
Superimposing two antagonistic phenomena, superconductivity (S) and ferromagnetism (F), on the nanoscale offers rich basic physics1-3 and provides several opportunities to design superconducting devices having unique features.4-7 The S/F interaction can in general be described in terms of the proximity effect with a mutual penetration of charge carriers, electrons or Cooper pairs, or stray field mediated correlations. Superconducting spin-valves (SSVs)8-10, intended to switch between two states with different superconducting transition temperatures, \( T_c \), show extremely large magnetoresistance and are under wide experimental and theoretical consideration since the last decade.11-18 In both, F/S/F and S/F/F designs of the SSV, \( T_c \) is manipulated by altering the magnetic configuration of the F-layers. A positive difference in \( T_c \) between anti-parallel (AP) and parallel (P) configuration, \( \Delta T_c^{AP-P} \), is described in terms of the S/F proximity effect,8-16 while stray-field mediated mutual correlations of micromagnetic structures in the F-layers is a plausible explanation of the negative \( \Delta T_c^{AP-P} \) in F/S/F spin-valves17-19.
Unconventional odd-triplet pairing, recently considered for S/F proximity systems,2 deepens the understanding and extends the functionality of the superconducting spin–valves,20-23 introducing a coupling by long-range spin-polarized Cooper pairs. As a result, \( \Delta T_c^{AP-P} \) in S/F/F-type SSVs can be either positive or negative within the proximity coupling model. Moreover, \( T_c \) can have an absolute minimum at non-collinear alignments of the F-layer magnetic moments, resulting in the triplet switching mode.20
Control of magnetic configurations in spin-valves is often provided by bringing one of the F-layers in contact with an antiferromagnetic (AF) layer. The interfacial exchange coupling induces a unidirectional magnetic anisotropy, the exchange bias effect, which gives rise to a horizontal shift of the hysteresis loop, coercivity enhancement, asymmetric hysteresis loops, and training effects24-28.
The exchange bias phenomenon is widely explored in magnetic field sensors25-27; however, even now it is not thoroughly understood and hardly predictable for an arbitrary AF-F couple of
materials. In particular, a realization of a spin-valve device making use of weak disordered ferromagnetic alloys still remains an unresolved problem (see, e.g. Ref. [28] and citations therein). In this Letter we report on a strong exchange biasing for the Cu$_{41}$Ni$_{59}$ layer adjacent to the Co/CoO$_x$ interface in a Co/CoO$_x$/Cu$_{41}$Ni$_{59}$/Nb/Cu$_{41}$Ni$_{59}$ SSV structure. The magnetoresistive switching properties obtained make the heterostructure suitable for superconducting spintronics applications.
A set of samples was produced in one run by magnetron sputtering, mainly at room temperature. The sketch of the resulting stack is shown in Fig. 1(a). First, a metallic Co layer was deposited on a commercial (111) silicon substrate, covered by a Si buffer layer before. Next, reactive oxygen gas was mixed to argon to deposit a CoO$_x$ oxide layer. Subsequently, a Cu$_{40}$Ni$_{60}$ target was RF sputtered at 200 °C at a rate of 3 nm/sec, resulting regularly in a Cu$_{41}$Ni$_{59}$ composition of the alloy in the film, as checked by the Rutherford backscattering spectrometry (RBS) technique and scanning Auger spectroscopy (see details in Refs. 29-31). To get a set of samples with different thicknesses of the Cu$_{41}$Ni$_{59}$ layer, the wedge technique$^{29-31}$ was applied. Thus, we obtained copper-nickel layer thicknesses as follows: maximum for Sample I, minimum or vanishing for Sample IV, and intermediate for adjacent Samples II and III. A flat superconducting Nb layer was prepared applying the “spray” technique$^{29-31}$. To control precisely the film growth rate we monotonously moved the target during the DC sputtering process along the substrate. Thus, we achieved an effective growth rate of about 1.3 nm/sec for the Nb film, while the rate of the sputtering process was adjusted to 4 nm/sec, to reduce contaminations gotten into the Nb film. In this way we obtained a smooth Nb film of constant thickness of $d_{\text{Nb}} \approx 12$ nm. Finally, the stack was finished by depositing a second wedge-shaped copper-nickel layer and capped with 12-14 nm of silicon to protect it against oxidation.
To obtain the thicknesses of the layers, we used cross-sectional Transmission Electron Microscopy (TEM) measurements. For Sample I (see Fig. 1(a)) the thicknesses were determined as about 4, 14, 25, 13, and 22 nm for Co, CoO$_x$, CuNi-Bottom, Nb, and CuNi-Top layers, respectively, whereas from the TEM image of Sample II we got about 5, 19, 9, 12, and 10 nm. We evaluated the thicknesses of the layers for Samples III and IV as 5, 19, 8, 12, 9 nm and 4, 14, $\leq 1$, 11.5, $\leq 1$ nm for Co, CoO$_x$, CuNi-Bottom, Nb, and CuNi-Top, respectively, by extrapolation, applying, in addition, our experience in evaluating the wedge profile$^{29-31}$.
To explore the magnetic configurations of the Co/CoO$_x$/Cu$_{41}$Ni$_{59}$/Nb/Cu$_{41}$Ni$_{59}$ system, several hysteresis loops were subsequently measured by a Superconducting Quantum Interference Device (SQUID) magnetometer after cooling the samples from above the CoO Néel temperature (291 K) to 10 K in a field of 10 kOe, applied parallel to the heterostructure plane. First, the applied magnetic field was swept from the saturated state achieved in the field cooling direction, towards negative fields until saturation of the layers in the opposite direction is reached (“backward branch” (BB) of the $m(H)$ hysteresis loops). Then, the magnetic field was swept from negative fields to positive ones until saturation of all ferromagnetic layers was achieved (“forward branch” (FB)). The resulting dependences of the magnetic moment on the magnetic field, $m(H)$, as well as their derivatives, $\partial m/\partial H$ (i.e., magnetic susceptibilities), for Sample III are shown in Fig. 2. For the first cycle see Fig. 2(a), for the repeated cycle (the second hysteresis loop and its derivative) see Fig. 2(b).
An apparent strong exchange bias is evident from the figure, as well as the training-effect\textsuperscript{27}, \textit{i.e.} a decrease of the hysteresis loop asymmetry, coercivity and squareness by further magnetic field cyclings (Fig. 2(a) to Fig. 2(b)). The bottom Cu\textsubscript{41}Ni\textsubscript{59} layer, adjacent to the CoO\textsubscript{x} layer, shows a very strong exchange bias $H_{EB} \approx -2$ kOe, where $H_{EB} = (H_{c}^{BB} + H_{c}^{PF}) / 2$. It is clearly distinguishable from the peaks of the $m(H)$ derivative labeled $H_c^2$ in Fig. 2(b), and in this magnitude never reported before for copper-nickel ferromagnetic alloys\textsuperscript{11,28}. Particular magnetic configurations that can be imagined using the Stoner-Wohlfarth model (see, \textit{e.g.}, Section 3.2 in Ref. [24]) of the magnetic moment reversal are depicted in Fig. 2 by the bold arrows. It could be concluded from Fig. 2(a), an antiparallel (AP) alignment of the magnetic moments of the top and bottom Cu\textsubscript{41}Ni\textsubscript{59} layer, which is necessary to observe the direct spin-valve effect (yielding $\Delta T_c^{AP-P} > 0$), is achieved over the wide range, -1.2 to -4.0 kOe, of magnetic fields.
To explore the superconducting spin-valve effects we measured the magnetoresistance, $R(H)$, at fixed temperatures in the range of the superconducting transition, which is most sensitive to the magnetic configurations in the system. The results are presented in Fig. 3. The standard DC four-terminal method was used, applying a sensing current of 10 $\mu$A. The polarity of the current was alternated during the resistance measurements to eliminate possible thermoelectric voltages.
The $R(H)$ measurements start with a BB sweep from the positively saturated state (Fig. 3(a)), which is the starting point of the magnetic hysteresis measurements in Fig. 2(a). The top Cu\textsubscript{41}Ni\textsubscript{59} layer reverses its magnetization and enters a single-domain state at about -1.2 kOe. The bottom Cu\textsubscript{41}Ni\textsubscript{59} layer is expected to save its single domain state up to about -4 kOe due to strong exchange bias. Surprisingly, no indications of the expected direct superconducting spin-
valve effect, $\Delta T_c^{\text{AB}} > 0$, was detected. Instead, a tiny inverse spin-valve effect (reduction of $T_c$ by less than 1 mK) can be deduced from the $R(H)$ data in the range -1.2 to -4 kOe (see dashed line in Fig. 3(a)). Moreover, a clear abrupt resistance change is also visible upon the Co plus Cu$_{41}$Ni$_{59}$-bottom layers reversal at -4.4 kOe (compare Fig. 2(a) with Fig. 3(a)).
Fig. 2 (a) The magnetic hysteresis loop of a Co/CoO$_x$/Cu$_{41}$Ni$_{59}$/Nb/Cu$_{41}$Ni$_{59}$ nanolayered heterostructure (Sample III) measured in a magnetic field parallel to the layers just after cooling down in a magnetic field of 10 kOe. Thin arrows refer to the derivative of the magnetic moment against the field: red solid lines are for the BB, and black dotted lines for the FB. Bold arrows, drawn according the Stoner-Wohlfart model, indicate the orientation of the magnetization of layers (1) to (3), from bottom to top; (b) The second hysteresis loop measured after the previous one and showing the training effect. The derivatives clearly show a splitting of the reversal fields for the BB, and their partial overlapping for the FB. Here, $x$ denotes a certain thickness of the bottom Cu$_{41}$Ni$_{59}$ layer (depending on $H$ and the magnetic history, probably increasing by training effects), the magnetization of which follows the top Cu$_{41}$Ni$_{59}$ layer in the proposed model.
The absence of the standard (direct) superconducting spin-valve effect can be explained if no antiparallel alignment of the top and bottom Cu$_{41}$Ni$_{59}$ layers occurs. A seeming contradiction with the presence of the nearly flat region in the BB of the hysteresis in Fig 2(a), after the top Cu$_{41}$Ni$_{59}$ reversal, could be resolved if we assume the bottom Cu$_{41}$Ni$_{59}$ layer is an exchange spring. A Nb side portion $x$ of the thickness of the bottom Cu$_{41}$Ni$_{59}$ layer rotates its magnetization at almost the same field as the top, soft Cu$_{41}$Ni$_{59}$ layer does, while the exchange biased interface to CoO$_x$ and the rest of the layer keeps the initial direction. There is a region of gradual transition between the oppositely magnetized sub-layers, which can be treated as an exchange spring or a domain wall, depending on its extent. One can expect that the non-uniform distribution of the magnetic
moments in this case may generate triplet pairing components\(^2,3,24-36\), which suppress superconductivity acting hereby against the expected direct spin-valve effect\(^9,10\). A decomposition of the net hysteresis loop of Fig. 2(b) on its components shows that the response of the soft Cu\(_{41}\)Ni\(_{59}\) magnetic moment is \(\sim 1.7\) times larger than the response of the hard Cu\(_{41}\)Ni\(_{59}\) magnetic moment, whereas from the ratio of the Cu\(_{41}\)Ni\(_{59}\) layers thicknesses a value of \(\sim 1.13\) is expected (9 nm for the top CuNi layer against 8 nm for the bottom one, see Fig. 1(b)). Thus, this observation favors the spring-magnet interpretation. The signal from the Co layer can be confidently separated from the above signals ascribed to the Cu\(_{41}\)Ni\(_{59}\) material.
Fig. 3. (a) to (d): \(R(H)\) measured at fixed temperatures within the superconducting transition for vanishing field. (e) and (f): Superconducting transitions at \(H=0\) measured after different sweeps of the magnetic field. Dashed line in (a): BB sweep \(R(H)\) data at positive \(H\) reflected across \(H=0\).
Upon a further increase of the field magnitude in the negative direction, the torque increases until the exchange biased region of the Cu\(_{41}\)Ni\(_{59}\) film abruptly reverses at about -4.4 kOe simultaneously with the Co layer magnetic moment (trace the \(m(H)\) derivatives in Fig. 2(a)). In this interpretation, the resistance change in Fig. 3(a) at the field of reversal (see the zoomed feature in the insert) is a resistance drop (i.e. an increase of \(T_c\)) due to a vanishing of the triplet component generated by the exchange spring.
The FB of the hysteresis loop, being visually quite smooth, produces a more tricky behavior of \(R(H)\) near magnetization reversal, likely because the Cu\(_{41}\)Ni\(_{59}\) layer (3) passes a multidomain state nearby the reversal (Fig. 2). The upturn feature on \(R(H)\) between 0.0 and 1.0 kOe could be treated as a superconducting \(T_c\) suppression by the combined action of the triplet pairing generation by noncollinear magnetic configurations in the system\(^20,21\) (yielding a peak of resistance near the perpendicular alignment of magnetizations), and the mechanism of correlated magnetic microstructures, proposed and elaborated in Ref. \(19\) for F/S/F spin-valve cores. Note here that due to the training effect\(^24,27\), the reversal fields for the Co and bottom
Cu$_{41}$Ni$_{59}$ layers are already split (see Fig. 2(a), especially the derivative of the FB hysteresis loop), and both Cu$_{41}$Ni$_{59}$ layers are in the multidomain state.
The kink in the resistance at -4.4 kOe of the BB branch could possibly also be interpreted as the vanishing of a stray field induced increase of $R(H)$ due to an inhomogeneous magnetization indicated by the only nearly flat (inclination increasing in trained state) $m(H)$ range between -1.2 and -4 kOe$^{17-19,37}$. Magnetostatic coupling between magnetic microstructures enhances the alignment$^{38}$ and intensifies the stray field penetrating the superconductor$^{19}$. Also domain wall (DW) induced superconductivity is possible if the width of the domain wall is small and the saturation magnetization is large$^{39,40}$. However, our previous investigations on Nb/Cu$_{41}$Ni$_{59}$ layered systems$^{22,41}$ indicate that the influence of stray fields of Cu$_{41}$Ni$_{59}$ layers alone are too weak to result in peak structures.
The $R(H)$ dependence for the trained state after ten full sweeps is given in Fig. 3(b) for three temperatures. The coercive fields $H^\text{BB}_c$ and $H^\text{FB}_c$ are split for both branches (as already in the second hysteresis loop, see paired peaks of the derivative in Fig. 2(b)), what results in basically similar, but a bit broadened $R(H)$ features, described for the first loop, and the same physics behind. There is a relatively weak magnetic noncollinearity within the bottom Cu$_{41}$Ni$_{59}$ layer, occurring in the range from about 0 to -2.5 kOe and -4 to -0.5 kOe for the BB and FB, respectively, expressed by a small angle between magnetization arrows in the schematic shown in Fig. 2(b) for layers (2) and (3), which is probably responsible for a suppression of the direct SSV in the BB. At the borders of these ranges there are field regions of strong noncollinearity within the bottom Cu$_{41}$Ni$_{59}$ layer, inherent to the reversal regions of BB and FB branches, generating the upturn features in the $R(H)$ curves. The significant difference is that the resistance at zero magnetic field strongly depends on the pre-history of that state (see Fig. 3(c), representing an earlier cycle at $T = 1.84$ K, at which also the measurement (1) in Fig. 3(b) has been performed). Indeed, around $H = 0$ in the FB (and $H = -3.5$ kOe in the BB) the magnetization directions of all three layers are strongly noncollinear (see Fig. 2(b)), and, especially important, those of layers (3) and (2) rotate against each other for increasing (and, respectively, decreasing) field, achieving nearly perpendicular alignment. Since layer (3) contains a certain part $x$ of the bottom layer, strongly noncollinear magnetization configurations occur inside this layer and, thus, a strong triplet component generation is expected$^{20}$. The triplet component suppresses $T_c$ leading to an increase of the magnetoresistance$^{32}$. The difference in magnetoresistances in Fig. 3(c) corresponds to a difference in the superconducting $T_c$ of 21 mK, measured directly from the resistive transition as shown in Fig. 3(e), at zero field.
To check if the effect arises from stray fields of the cobalt layer (here, a possible DW width is similar as in Cu$_{41}$Ni$_{59}$ layers$^{42,43}$, but the higher saturation magnetization favors cobalt to produce stray field peaks on magnetoresistance$^{39,41}$), we measured Sample IV (shown in Fig. 3(d)), for which the thickness of the copper-nickel layers is very small (below 1 nm) or vanishing. In accordance with the random anisotropy model$^{44}$ the coercivity of Cu$_{41}$Ni$_{59}$ layers reduces$^{45}$, and the coercive loop becomes narrow. So the stray field effect, which is maximal at the coercivity fields, shrinks to a narrow interval of fields around $H = 0$. However, the $R(H)$ splitting of the BB and FB extends over 6 kOe, $i.e.$ it has another origin. It correlates with the Co hysteresis loop. We attribute this to the stray fields arising from disordered spins at the Co/CoO$_x$ and CoO$_x$/Cu$_{41}$Ni$_{59}$ interface in the trained state$^{24}$. The Co layer of 4 nm is expected not to produce substantial stray fields because of Neel type domain wall structure at that film thickness. The triplet pairing generation for such extremely thin copper-nickel layers is also expected to be negligible$^{20}$. Thus,
the features in $R(H)$ shown in Fig. 3(d) represent the background contribution of the Co/CoO$_x$ stray fields. The difference in $T_c$ measured at $H=0$ for the BB and FB is about 4 mK (see Fig. 3(f)).
To check the transport properties of the device structure we measured $R(H)$ for currents in the range 1 - 250 μA. Up to about 200 μA no significant changes were observed. At 200 μA and above a drastic increase of the difference $R_{FB}-R_{BB}$ at zero field was registered for Sample III (see Fig. 4). While the resistance $R_{BB}$ approaches the normal state value $R_N$ in the FB, the resistance $R_{FB}$ nearly vanishes, in the BB, resulting in a ratio $(R_{FB}-R_{BB})/R_N$ of about 90%. The reason is a sharpening of the superconducting transition, with the width $\delta T_c = T(0.9 R_N)-T(0.1 R_N)$ decreasing from $\delta T_c = 50$ mK to 30 mK for 10 μA and 200 μA, respectively (see Figs. 3(e) and 4(insert), $T_c$ is lowered by 70 mK). A plausible explanation could be an avalanche like flux flow at high Lorentz force, due to the increased current$^{46}$, or an instability of the flux line motion in the resistive mixed state, caused by an escape of nonequilibrium quasiparticles from the vortex cores$^{47,48}$. This effect has never been observed in Sample IV (vanishing Cu$_{41}$Ni$_{59}$ layers thickness) for any current up to the critical one for any measurement temperature, although the expected critical velocity for the instability is expected to be smaller for Nb than for Nb/Cu$_{41}$Ni$_{59}$$^{48,49}$. This suggests, that the triplet pairing, generated by noncollinear magnetizations in the Cu$_{41}$Ni$_{59}$ layers, is involved in the phenomenon.
A plausible explanation could be an avalanche like flux flow at high Lorentz force, due to the increased current$^{46}$, or an instability of the flux line motion in the resistive mixed state, caused by an escape of nonequilibrium quasiparticles from the vortex cores$^{47,48}$. This effect has never been observed in Sample IV (vanishing Cu$_{41}$Ni$_{59}$ layers thickness) for any current up to the critical one for any measurement temperature, although the expected critical velocity for the instability is expected to be smaller for Nb than for Nb/Cu$_{41}$Ni$_{59}$$^{48,49}$. This suggests, that the triplet pairing, generated by noncollinear magnetizations in the Cu$_{41}$Ni$_{59}$ layers, is involved in the phenomenon.
In summary, a strong exchange biasing of about 2 kOe for a diluted ferromagnetic copper-nickel alloy was obtained at the interface with a Co/CoO$_x$ bilayer. Combined with an F/S/F SSV core, the Co/CoO$_x$/Cu$_{41}$Ni$_{59}$/Nb/Cu$_{41}$Ni$_{59}$ spin valve exhibits a magnetic memory effect, which depends on the preceding field polarity. High and low resistance states do not require bias fields or currents to keep them in an idle mode. While the writing of a state requires only field sweeps up to magnetic saturation, for reading only a current at $H = 0$ is necessary. We propose to ascribe the difference in resistances to the training effect and the generation of an odd in frequency triplet pairing component at noncollinear alignment of the magnetizations in the system around $H = 0$. Moreover, e.g. stray field effects of the micromagnetic structures of the layers may contribute to the effect. Since the mechanisms described above provide different critical temperatures for the two logical states, one does not need to realize the hardly achievable antiparallel magnetic alignment required for the direct SSV effect.
Acknowledgments
The authors are grateful to S. Heidemeyer, B. Knoblich and W. Reiber for assistance in the TEM sample preparation, and to D. Vieweg for assistance in magnetic measurements. The work was supported by the Deutsche Forschungsgemeinschaft (DFG) under the grant No GZ: HO 955/6-2. In part the Russian Fund for Basic Research (RFBR) supported the project under the grant 11-02-00848-a (LRT). The magnetic investigations (H.-A. K.v.N.) were partially supported by the Deutsche Forschungsgemeinschaft (DFG) within the Transregional Collaborative Research Center TRR 80 "From Electronic Correlations to Functionality" (Augsburg, Munich).
References
1. A. I. Buzdin, Rev. Mod. Phys. 77, 935 (2005).
2. F. S. Bergeret, A. F. Volkov, and K. B. Efetov, Rev. Mod. Phys. 77, 1321 (2005).
3. M. Eschrig, Phys. Today 64, 43 (2011).
4. V. V. Ryazanov, V. A. Oboznov, A. Yu. Rusanov, A. V. Veretennikov, A. A. Golubov, and J. Aarts, Phys. Rev. Lett. 86, 2427 (2001).
5. A. K. Feofanov, V. A. Oboznov, V. V. Bol’ginov, J. Lisenfeld, S. Poletto, V. V. Ryazanov, A. N. Rossolenko, M. Khabipov, D. Balashov, A. B. Zorin, P. N. Dmitriev, V. P. Koshelets, and A. V. Ustinov, Nature Physics 6, 593 (2010).
6. T. I. Larkin, V. V. Bol’ginov, V. S. Stolyarov, V. V. Ryazanov, I. V. Vernik, S. K. Tolpygo, and O A. Mukhanov, Appl. Phys. Lett. 100, 222601 (2012).
7. J. D. Sau, R. M. Lutchyn, S. Tewari, S. Das Darma, Phys. Rev. Lett. 104, 040502 (2010).
8. S. Oh, D. Youm, and M. R. Beasley, Appl. Phys. Lett. 71, 2376 (1997).
9. L. R. Tagirov, Phys. Rev. Lett. 83, 2058 (1999).
10. A. I. Buzdin, A. V. Vedyayev, and N. V. Ryazhanova, Europhys. Lett. 48, 686 (1999).
11. J. Y. Gu, C.-Y. You, J. S. Jiang, J. Pearson, Ya. B. Bazaliy, and S. D. Bader, Phys. Rev. Lett. 89, 267001 (2002).
12. I. C. Moraru, W. P. Pratt, Jr., and N. O. Birge, Phys. Rev. Lett. 96, 037004 (2006).
13. G. Nowak, H. Zabel, K. Westerholt, I. Garifullin, M. Marcellini, A. Liebig, and B. Hjörvarsson, Phys. Rev. B 78, 134520 (2008).
14. Y. Luo and K. Samwer, Europhys. Lett. 91, 37003 (2010).
15. P. V. Leksin, N. N. Garif’yanov, I. A. Garifullin, J. Schumann, H. Vinzelberg, V. Kataev, R. Klingeler, O. G. Schmidt, and B. Büchner, Appl. Phys. Lett. 97, 102505 (2010).
16. G. Novak, K. Westerholt, and H. Zabel, Supercond. Sci. Technol. 26, 025004 (2013).
17. A. Yu. Rusanov, S. Habraken, and J. Aarts, Phys. Rev. B 73, 060505(R) (2006).
18. D. Stamopoulos, E. Manios, and M. Pissas, Supercond. Sci. Technol. 20, 1205 (2007).
19. J. Zhu, X. Cheng, C. Boone, and I. N. Krivorotov, Phys. Rev. Lett. 103, 027004 (2009).
20. Ya. V. Fominov, A. A. Golubov, T. Yu. Karminskaya, M. Yu. Kupriyanov, R. G. Deminov, and L. R. Tagirov, Pis’ma v ZhETF 91, 329 (2010) [JETP Lett. 91, 308 (2010)].
21. J. Zhu, I. N. Krivorotov, K. Halterman, and O. T. Valls, Phys. Rev. Lett. 105, 207002 (2010).
22. V. I. Zdravkov, J. Kehrle, G. Obermeier, D. Lenk, H.-A. Krug von Nidda, C. Müller, M. Yu. Kupriyanov, A. S. Sidorenko, S. Horn, R. Tidecks, L. R. Tagirov, Phys. Rev. B 87, 144507 (2013).
23. P. V. Leksin, N. N. Garif’yanov, I. A. Garifullin, Ya. V. Fominov, J. Schumann, Y. Krupskaya, V. Kataev, O. G. Schmidt, and B. Büchner, Phys. Rev. Lett. 109, 057005 (2012).
24. F. Radu and H. Zabel, in Magnetic Heterostructures, STMP 227, H. Zabel and S. D. Bader, Eds. (Springer, Berlin-Heidelberg, 2008) Ch.3.
25. J. Nogués, J. Sort, V. Langlais, V. Skumryev, S. Suriñach, J.S. Muñoz, M.D. Baró, Phys. Reports 422, 65 (2005).
26. J. Nogues and I. K. Schuller, J. Magn. Magn. Mater. 192, 203 (1999).
27. S. Brems, D. Buntinx, K. Temst, Ch. Van Haesendonck, F. Radu, and H. Zabel, Phys. Rev. Lett. 95, 157202 (2005).
28. M. G. Blamire, M. Ali, C.-W. Leung, C. H. Marrows, and B. J. Hickey, Phys. Rev. Lett. 98, 217202 (2007).
29. V. I. Zdravkov, J. Kehrle, G. Obermeier, S. Gsell, M. Schreck, C. Müller, H.-A. Krug von Nidda, J. Lindner, J. Moosburger-Will, E. Nold, R. Morari, V. V. Ryazanov, A. S. Sidorenko, S. Horn, R. Tidecks, and L. R. Tagirov, Phys. Rev. B 82, 054517 (2010).
30. V. Zdravkov, A. Sidorenko, G. Obermeier, S. Gsell, M. Schreck, C. Müller, S. Horn, R. Tidecks, and L. R. Tagirov, Phys. Rev. Lett. 97, 057004 (2006).
31. V. I. Zdravkov, J. Kehrle, G. Obermeier, A. Ulrich, S. Gsell, M. Schreck, C. Müller, R. Morari, A. S. Sidorenko, L. R. Tagirov, R. Tidecks, and S. Horn, Supercond. Sci. Technol. 24, 095004 (2011).
32. E. E. Fullerton, J.S. Jiang, S.D. Bader, Journ. Magn. Magn. Mater. 200, 392 (1999).
33. J. M. D. Coey, Magnetism and Magnetic Materials (Cambridge University Press, New York, 2010) p.279.
34. I. Sosnin, H. Cho, V. T. Petrashev, and A. F. Volkov, Phys. Rev. Lett. 96, 157002 (2006).
35. J. Y. Gu, J. Kusnadi, and C.-Y. You, Phys. Rev. B 81, 214435 (2010).
36. C-T. Wu, O. T. Valls, and K. Halterman, Phys. Rev. B 86, 184517 (2012).
37. A. K. Suszka, F. S. Bergeret, and A. Berger, Phys. Rev. B 85, 024529 (2012).
38. S. Middelhoek, J. Appl. Phys. 37, 1276 (1966).
39. I. S. Burmistrov and N. M. Chchelkatchev, Phys. Rev. B 72, 144520 (2005).
40. J. Aarts, C. Attanasio, C. Bell, C. Cirillo, M. Flokstra, and J. M. van der Knap, in Nanoscience and Engineering in Superconductivity, NanoScience and Technology, V. Moschchalkov, R. Wöerdenweber, and W. Lang, Eds. (Springer, Heidelberg-Dordrecht-London-New York, 2010) Ch. 13.
41. J. Kehrle, V. I. Zdravkov, G. Obermeier, J. Garcia-Garcia, A. Ullrich, C. Müller, R. Morari, A. S. Sidorenko, S. Horn, L. R. Tagirov, and R. Tidecks, Ann. der Physik (Berlin) 524, 37 (2012).
42. I. S. Veshchunov, V. A. Oboznov, A. N. Rossolenko, A. S. Prokofiev, L. Ya. Vinnikov, A. Yu. Rusanov, and D. V. Matveev, Pis’ma v ZhETF 88, 873 (2008) [JETP Lett. 88, 758 (2008)].
43. C. Bell, S. Tursucu, and J. Aarts, Phys. Rev. B 74, 214520 (2006).
44. G. Herzer, in NATO Science Series II: Mathematics, Physics and Chemistry, Vol. 184, B. Idzikowski, P. Svec, and M. Miglierini, Eds. (Kluwer Academic, Dodrecht, 2005) p. 15.
45. A. Ruotolo, C. Bell, C. W. Leung, and M. G. Blamire, J. Appl. Phys. 96, 512 (2004).
46. Tae-Jong Hwang and D. H. Kim, Appl. Phys. Lett. 101, 072601 (2012).
47. S. G. Doettinger, R. P. Huebener, R. Gerdemann, A. Kühle, S. Anders, T. G. Träuble, and J. C. Villégier, Phys. Rev. Lett. 73, 1691 (1994).
48. C. Attanasio and C. Cirillo, J. Phys. Condens. Matter 24, 083201 (2012).
49. E. A. Ilyina, C. Cirillo, and C. Attanasio, Eur. Phys. J. B 83, 53 (2011)
| 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3464,
1
],
[
3464,
7375,
2
],
[
7375,
9611,
3
],
[
9611,
11947,
4
],
[
11947,
14398,
5
],
[
14398,
18794,
6
],
[
18794,
22338,
7
],
[
22338,
25549,
8
],
[
25549,
28725,
9
],
[
28725,
28725,
10
],
[
28725,
28725,
11
],
[
28725,
28725,
12
],
[
28725,
28725,
13
]
]
} |
c3771139ccc762c3abdf4f20c981db542f7dc152 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 6,
"total-fallback-pages": 0,
"total-input-tokens": 21399,
"total-output-tokens": 6878,
"fieldofstudy": [
"Physics"
]
} | Controlling the Minimal Feature Sizes in Adjoint Optimization of Nanophotonic Devices Using B-spline Surfaces
Erfan Khoram1, Xiaoping Qian2, Ming Yuan3, Zongfu Yu1
Adjoint optimization is an effective method in the inverse design of nanophotonic devices. In order to ensure the manufacturability, one would like to have control over the minimal feature sizes. Here we propose utilizing a level-set method based on b-spline surfaces in order to control the feature sizes. This approach is first used to design a wavelength demultiplexer. It is also used to implement a nanophotonic structure for artificial neural computing. In both cases, we show that the minimal feature sizes can be easily parameterized and controlled.
INTRODUCTION
Adjoint method has become a very useful tool in the inverse design of photonics\cite{1,2} and among other optimization methods such as genetic algorithms\cite{10}, swarm optimization\cite{11}, or nonlinear optimization\cite{12}, adjoint method is particularly efficient as it computes the gradient of a cost function w.r.t. a large number of design parameters.
The inverse design process mostly involves the optimization of the dielectric constant distribution $\varepsilon(\mathbf{r})$. In a gradient-based approach, $\varepsilon$ is treated as a continuous variable. However, in practice, the permittivity can only take discrete values determined by the types of materials used. To bridge this gap, one can apply the method of level-set or density-based topology optimization. However, even with level-set, or density-based topology optimization to address the issue of discrete material parameters, we still end up with structural features that might be too small for today’s nanofabrication. Additional constraints need to be applied to control the size of minimum features. For example Ref\cite{10} defines the medium as a collection of pixels with a fixed size larger than the critical dimension of fabrication; while \cite{11} uses the method proposed in \cite{20}: a spatial low pass filter which controls the feature sizes. Ref \cite{21} uses the convolution between the gray-scale medium and a disk that defines the minimum feature size. Refs \cite{11,23} use radial basis functions to control the feature size. In this work, we are going to use b-splines for topology optimization, which has been a highly successful and proven approach for size control in structural shape optimization in the field of mechanics \cite{22}.
B-SPLINES
Splines are functions generally used for interpolating data and designing curves and surfaces in computer aided designs and computer graphics. These functions are generated by a linear combination of piece-wise polynomial functions. Their strength is in the fact that they are capable of accurate interpolation of data with low degree polynomial components, that would otherwise require a high degree non-local polynomial. The degree of the spline is the degree of its highest order polynomial.
One way of representing a spline is by defining a set of basis functions, called basis-spline functions or b-spline functions. In order to define any one of these basis functions, the domain over which the spline is going to be defined, has to be segmented with a set of knots $U = [u_0, ..., u_m]$. In this work we only work with knots that appear only once in this set, and are equidistant from one another. Another property that is required to define a b-spline function is its degree ($k$). With both knots and degree defined, basis-spline functions can be expressed using the Cox-de Boor recursion formula \cite{23,24}.
$$
N_{(u)}^{i,0} = \begin{cases}
1 & u_i < u < u_{i+1} \\
0 & \text{otherwise}
\end{cases}
$$
$$
N_{(u)}^{i,k} = \frac{u - u_i}{u_{i+k} - u_i} N_{(u)}^{i,k-1} + \frac{u_{i+k+1} - u}{u_{i+k+1} - u_{i+1}} N_{(u)}^{i+1,k-1}
$$
In the equation above $N_{(x)}^{i,k}$ shows a b-spline of degree $k$. As it can be seen, a b-spline of degree zero is only non-zero on one of the segments of the domain, while higher degrees span more segments of the domain. As a matter of fact, a b-spline of degree $k$ defined over a domain segmented by $m + 1$ knots spans $k + 1$ one of these intervals ($m + 1 - k$ such b-splines are defined over the domain). This can be seen in Fig.1 where a domain has been segmented by 4 knots.
With the basis spline functions, a curve can be generated using a linear combination of these basis functions. In Eq.2, the $P_{(i)}$’s are the set of coefficients that determine the contribution of each b-spline component to the curve.
\*
\* 1Department of Electrical and Computer Engineering, University of Wisconsin Madison-Madison, WI53706, USA
2Department of Mechanical Engineering, University of Wisconsin Madison-Madison, WI53706, USA
3Department of Statistics, Columbia University, New York, NY10027, USA
B-splines can be used for a density based approach or a level-set one. Next we use 2 dimensional (2D) photonic design to illustrate this method in the context of the level-set method. In 2D, we use basis spline functions to construct a surface $\phi(x,y)$ using the following equation:
$$
\phi(x,y) = \sum_{i=0}^{n_x} \sum_{j=0}^{n_y} N_{i,k_x}^x N_{j,k_y}^y P_{i,j}
$$
Similar to Eq.2, $N_{i,k_x}^x$ and $N_{j,k_y}^y$ are basis-spline functions of degree $k_x$ and $k_y$ respectively, and $P_{i,j}$ are the set of coefficients. If we were to limit the value of the coefficients between 0 and 1, the generated surface could be used in a density-based topology optimization. Here, we focus on the level-set method.
In order to use the level-set method for topology optimization, we first need to define a surface $\phi(x,y)$ over the area that the optimized device will occupy (The value of each point on this surface can vary across a limited range around zero). Then the medium can be shaped according to Eq.1 based on this surface.
$$
\varepsilon(r) = \begin{cases}
\varepsilon_1 & \phi(r) < 0 \\
\varepsilon_2 & \phi(r) > 0
\end{cases}
$$
Now, optimizing the medium with a level-set method consists of evolving the boundaries between the two materials (all the points where $\phi(r) = 0$) in a manner that improves the performance of the device. With a b-spline surface for this task, we have direct control on this surface, and consequently on the boundary between the constituent materials. Having defined the level-set surface as a b-spline surface, we can now directly optimize a cost function $J$ with respect to the control parameters of the b-spline surface.
$$
\frac{\partial J}{\partial P_{i,j}} = \int_s \left( \frac{\partial J}{\partial \varepsilon(\phi(x,y))} \frac{\partial \varepsilon(\phi(x,y))}{\partial \phi(x,y)} \frac{\partial \phi(x,y)}{\partial (i,j)} \right) \\
\Rightarrow \frac{\partial J}{\partial P_{i,j}} = \int_s \left( \frac{\partial J}{\partial \varepsilon(\phi(x,y))} \frac{\partial \varepsilon(\phi(x,y))}{\partial \phi(x,y)} (N_{i,k_x}^x N_{j,k_y}^y) \right)
$$
In Eq.5, the first term can be calculated with the adjoint method, the second term (a Dirac delta function which is nonzero only where $\phi(r)$ passes zero), can be computed with marching squares algorithm [23], and the final term is a simple matrix multiplication.
Once the gradient of the cost function is calculated, gradient descent can be used to update the b-spline coefficients. However, in order to ensure the stability of the process, in each update iteration the coefficients are normalized by the highest value among them [20]. The mentioned steps can be summarized as shown in Eq.6, where $P'$ is the matrix of all the b-spline coefficients at
---
**Formulation**
B-splines can be used for a density based approach or a level-set one. Next we use 2 dimensional (2D) photonic design to illustrate this method in the context of the level-set method. In 2D, we use basis spline functions to construct a surface $\phi(x,y)$ using the following equation:
$$
\phi(x,y) = \sum_{i=0}^{n_x} \sum_{j=0}^{n_y} N_{i,k_x}^x N_{j,k_y}^y P_{i,j}
$$
Similar to Eq.2, $N_{i,k_x}^x$ and $N_{j,k_y}^y$ are basis-spline functions of degree $k_x$ and $k_y$ respectively, and $P_{i,j}$ are the set of coefficients. If we were to limit the value of the coefficients between 0 and 1, the generated surface could be used in a density-based topology optimization. Here, we focus on the level-set method.
In order to use the level-set method for topology optimization, we first need to define a surface $\phi(x,y)$ over the area that the optimized device will occupy (The value of each point on this surface can vary across a limited range around zero). Then the medium can be shaped according to Eq.1 based on this surface.
$$
\varepsilon(r) = \begin{cases}
\varepsilon_1 & \phi(r) < 0 \\
\varepsilon_2 & \phi(r) > 0
\end{cases}
$$
Now, optimizing the medium with a level-set method consists of evolving the boundaries between the two materials (all the points where $\phi(r) = 0$) in a manner that improves the performance of the device. With a b-spline surface for this task, we have direct control on this surface, and consequently on the boundary between the constituent materials. Having defined the level-set surface as a b-spline surface, we can now directly optimize a cost function $J$ with respect to the control parameters of the b-spline surface.
$$
\frac{\partial J}{\partial P_{i,j}} = \int_s \left( \frac{\partial J}{\partial \varepsilon(\phi(x,y))} \frac{\partial \varepsilon(\phi(x,y))}{\partial \phi(x,y)} \frac{\partial \phi(x,y)}{\partial (i,j)} \right) \\
\Rightarrow \frac{\partial J}{\partial P_{i,j}} = \int_s \left( \frac{\partial J}{\partial \varepsilon(\phi(x,y))} \frac{\partial \varepsilon(\phi(x,y))}{\partial \phi(x,y)} (N_{i,k_x}^x N_{j,k_y}^y) \right)
$$
In Eq.5, the first term can be calculated with the adjoint method, the second term (a Dirac delta function which is nonzero only where $\phi(r)$ passes zero), can be computed with marching squares algorithm [23], and the final term is a simple matrix multiplication.
Once the gradient of the cost function is calculated, gradient descent can be used to update the b-spline coefficients. However, in order to ensure the stability of the process, in each update iteration the coefficients are normalized by the highest value among them [20]. The mentioned steps can be summarized as shown in Eq.6, where $P'$ is the matrix of all the b-spline coefficients at
iteration $t$ and $||P^t||_{\infty}$ is the maximum absolute value in this matrix.
\[ P^{t+1} = P^t - \gamma \frac{\partial J}{\partial P}_{|P=P^t} \]
\[ P^{t+1} = \frac{||P^{t+1}||_{\infty}}{||P^{t+1}||_{\infty}} \] \hspace{1cm} \text{(6)}
**APPLICATION**
In order to demonstrate the application of this approach, we utilized it to design two nanophotonic devices: a wavelength demultiplexer [4] and a nanophotonic structure for artificial neural computing [27].
The **wavelength demultiplexer** is a three port device that guides the light coming from the input port to one of the two output ports based on its wavelength. In our design, this device is of size $4 \mu m \times 2 \mu m$ designed to demultiplex 1500 nm and 1550 nm wavelengths. We optimize three versions of this device with the intention of showcasing how the critical dimensions of the device can be tuned by the basis spline functions.
The b-spline surface is defined on the same grid as the one used for the electromagnetic simulation and over the area to be optimized. We can use either the distribution of the knots on the grid, or the degree of the b-splines to control the critical dimension. In this work we opt for the former. We define our basis functions as cubic b-splines, and use three different knot distributions to realize three version of the devices with different feature sizes. The first design has the knots placed at every second point on the simulation grid, the second design at every fifth point, and the third design at every ninth point. This effectively provides us with 4559, 629, and 152 b-spline coefficients respectively. Once the b-spline surface is set up, we use Eq.4 to define the medium, where $\varepsilon_1$ is set to the permittivity of SiO$_2$ and $\varepsilon_2$ is set to that of Si. Finally, a cost function is necessary to optimize the structure. The cost function we define here tries to maximize the the transmission of the time-averaged energy flux to the output port corresponding to the input wavelength.
The devices were optimized with respect to the cost function mentioned above using Eq.4 and Eq.5. The results for the optimization of these three instances have been depicted in Fig.3. As it was mentioned earlier, since placing the knots more sparsely results in an increase in the area that each basis-spline function affects, the critical dimension of the device increases. This can clearly be seen in Fig.3. As we move from Fig.3(a) to Fig.3(c) the feature sizes distinctly expand. However, even for the case of the sparsely distributed knots, the occasional nuances in the b-spline surface can cause the appearance of small features in the device. This problem is mitigated by applying erosion and dilation processes on the device in each update iteration to eliminate these occasional smaller features. We should point out that this control over the feature sizes of the device comes at a price: there is a trade off between the performance of the device and the critical dimensions of the structure. As we use larger critical dimensions, the degree of freedom, i.e. the number of b-Spline functions used, decreases, and the optimization is further constrained, consequently the performance generally degrades. This is also evident in Fig.3, where the transmission drops moving from the first device to the last one.
**FIG. 3.** The wavelength demultiplexers acting on 1500 nm and 1550 nm wavelengths. These devices are designed with P matrices (matrices of the b-spline coefficients) with sizes (a) $97 \times 47$, (b) $37 \times 17$, and (c) $19 \times 8$. As shown here, the higher the number of coefficients (corresponding to packing the b-spline knots closer together), the smaller the details of the designed device, and the higher the confinement of the field to the device. (d) shows the transmission for each of these three instances. Transmission for 1500 nm is shown with continuous lines, while the dashed lines represent the transmission for 1550 nm. The degrading effect of reducing the number of b-spline coefficients on transmission can be clearly seen here.
Nanophotonic artificial neural computing This device can accomplish the same task of neural inference as in [27]. The objective is to recognize the value of a handwritten digit 0-9. The input light from the handwritten digit is focused by a nanophotonic structure to different locations according to the value of the digit. The medium then has to map all the different writing styles of the same digit to the specific location corresponding to that digit; furthermore, the medium must perform well on handwritten digits that it has not yet seen.
The setup for this task is as follows: the image of the handwritten digit is first vectorized and mapped to an array of light sources placed on the left side of the nanophotonic structure. Then a set of ten receivers corresponding to the each of the ten classes 0-9 are placed on the right side of the medium. Inside the device, the input light scatters and reflects off of the numerous material interfaces, and then focuses on a spatial location on the output side. On the output side, the receivers measure the light intensity at their respective locations and the class with the maximum measured value is selected as the correct class for the input image.
The device is made possible by a much more extensive optimization process than that for the wavelength demultiplexer. This optimization process is equivalent to the training process in digital neural networks. We define a cost function and use adjoint method to calculate the gradient. At the time of training, we measure the intensity of light at the location of all the receivers. Then we use a cross-entropy cost function between the normalized output array and a 1×10 one-hot vector (a vector that is all zeros except at the index of the correct class), to calculate the cost value for that specific instance. Details of the working principle can be found in Ref[27]. In this work we implement the level-set function based on a b-spline surface to control the minimal critical dimension. In our previous work [27], the level-set method was implemented based on a surface set as a signed distance function. Here, the initial medium is created by generating a random set of values for the b-spline coefficients, and then the medium is created based on the generated surface with the surface evolving afterwards by updating the b-spline coefficients. On the other hand, in Ref [27] the medium was generated by randomly distributing inclusions made up of a second material inside the host medium, and then a surface $\phi(r)$ is generated based on this initial medium (similar to the previous section, the surface relates to the medium with Eq.4). The structure then evolves by updating this signed-distance surface at each training iteration using a modified version of the following equation.
$$\partial_t \phi + v(x, y)|\nabla \phi| = 0$$ \hspace{1cm} (7)
In this equation $v(x, y)$ is the velocity with which each point on the zero crossing curves of the level-set function moves normal to each of those curves. This velocity is set equal to the gradient acquired with the adjoint method. This implementation of the level-set method for the image classification task turned out to be quite dependant on the initial distribution of the inclusions, and therefore we had to initialize the medium with many small inclusions to get a good performance from the device. Doing so resulted in small feature sizes in the device which make the fabrication of the device rather difficult. However, with the b-spline approach to level-set function this problem does not happen as the optimization process is less dependant on the initial shape of the medium.
We optimize two versions of this device, one with the signed distance function approach, and the other with the b-spline approach. Both devices were optimized for the wavelength 1 μm with the size 42 μm × 30 μm and made up of SiO₂ as the host material and air as the inclusions. For the signed-distance implementation 2000 inclusions of size 5 μm × 5 μm are spread throughout the device for initialization, and for the b-spline implementation the knots are distributed at every tenth point on the simulation grid in both axial dimensions. The accuracy on the test set for the two devices turned out to be 76.8% and 77.2%. The performance of the b-spline approach is slightly better in this case which is not of great consequence as the gap between the two values can be reduced with better initialization of the signed-distance approach. However, the important point that is apparent in Fig.4 is the difference between the shape of the two structures. As it can be seen, the first device shown in Fig.4(a) has much smaller features whereas the second device depicted in Fig.4(b) has a much larger critical dimension. This makes the second device much easier to fabricate, and thus the b-spline approach much more desirable. Finally, Fig.4(c) and Fig.4(d) show the device in action. As it can be seen, light from different class of images is focused at distinct locations corresponding to that class. Moreover, different image samples of the same digit produce different field distributions; however, these different field distributions result in the same output class.
**CONCLUSION**
In this work, we adopt a proven technique to size control that is used in the field of mechanics to the inverse design of photonics. It use B-spline basis to describe the distribution of the dielectric constant. The minimal feature size is controlled by the knots distribution and the degree. This method is very easy to implement and interpret. We demonstrated the proposed method for a wavelength demultiplexer, and a nanophotonic structure for artificial neural computing. The first class of nanophotonic devices we implemented was used to demonstrate how to handle the trade off between the complexity and their performance. For the second class, two variations of the level-set methods were used to achieve the same goal. The comparison between the two cases shows how having a well-defined set of control parameters can help us in designing a nanophotonic device in one optimization stage, without the need to have a good initial guess of the variables we are trying to optimize.
I. ACKNOWLEDGEMENTS
The authors thank W. Shin for his help on improving the computational speed of the implementation of this method. This work was partially supported by the National Science Foundation under Grant No. 1561917. It was also partially supported by the Defense Advanced Research Projects Agency (DARPA), under agreement HR00111820046. The views, opinions and/or findings expressed are those of the author and should not be interpreted as representing the official views or policies of the Department of Defense or the U.S. Government.
[1] C. M. Lalau-Keraly, S. Bhargava, O. D. Miller, and E. Yablonovitch, Optics express 21, 21693 (2013).
[2] F. Callewaert, V. Velev, P. Kumar, A. Sahakian, and K. Aydin, Scientific reports 8, 1358 (2018).
[3] A. Y. Piggott, J. Petykiewicz, L. Su, and J. Vučković, Scientific Reports 7, 1786 (2017).
[4] A. Y. Piggott, J. Lu, K. G. Lagoudakis, J. Petykiewicz, T. M. Babinec, and J. Vučković, Nature Photonics 9, 374 (2015).
[5] W. Frei, D. Tortorelli, and H. Johnson, Optics letters 32, 77 (2007).
[6] L. Su, R. Trivedi, N. V. Sapra, A. Y. Piggott, D. Vercauyse, and J. Vučković, Optics express 26, 4023 (2018).
[7] R. Pestourie, C. Pérez-Arancibia, Z. Lin, W. Shin, F. Capasso, and S. G. Johnson, Optics express 26, 33732 (2018).
[8] L. F. Frellsen, Y. Ding, O. Sigmund, and L. H. Frandsen, Optics express 24, 16866 (2016).
[9] J. S. Jensen and O. Sigmund, Laser & Photonics Reviews 5, 308 (2011).
[10] A. Michaels and E. Yablonovitch, Optics express 26, 4766 (2018).
[11] T. W. Hughes, M. Minkov, I. A. Williamson, and S. Fan, ACS Photonics 5, 4781 (2018).
[12] L. H. Frandsen and O. Sigmund, in Photonic and Phononic Properties of Engineered Nanostructures VI, Vol. 9756 (International Society for Optics and Photonics, 2016) p. 97560Y.
[13] W. R. Frei, H. Johnson, and K. D. Choquette, Journal of Applied Physics 103, 033102 (2008).
[14] S. Molesky, Z. Lin, A. Y. Piggott, W. Jin, J. Vuckovic, and A. W. Rodriguez, Nature Photonics 12, 659 (2018).
[15] D. Vercauyse, N. V. Sapra, L. Su, R. Trivedi, and J. Vučković, Scientific reports 9, 8999 (2019).
[16] O. D. Miller and E. Yablonovitch, “Inverse optical design,” in Encyclopedia of Applied and Computational Mathematics, edited by B. Engquist (Springer Berlin Heidelberg, Berlin, Heidelberg, 2015) pp. 729–732.
[17] B. J. Offrein, G.-L. Bona, R. Germann, I. Massarek, D. Erni, et al., Journal of Lightwave Technology 16, 1680 (1998).
[18] J. C. Mak, C. Sideris, J. Jeong, A. Hajimiri, and J. K. Poon, Optics letters 41, 3868 (2016).
[19] B. Shen, P. Wang, R. Polson, and R. Menon, Nature Photonics 9, 378 (2015).
[20] M. Zhou, B. S. Lazarov, F. Wang, and O. Sigmund, Computer Methods in Applied Mechanics and Engineering 293, 266 (2015).
[21] L. Su, A. Y. Piggott, N. V. Sapra, J. Petykiewicz, and J. Vuckovic, ACS Photonics 5, 301 (2017).
[22] X. Qian, Computer Methods in Applied Mechanics and Engineering 265, 15 (2013).
[23] C. De Boor, Journal of Approximation theory 6, 50 (1972).
[24] M. G. Cox, IMA Journal of Applied Mathematics 10, 134 (1972).
[25] D. J. Kroon, “Isocontour Webpage,” (2011), https://www.mathworks.com/matlabcentral/fileexchange/30525-isocontour.
[26] O. Bernard, D. Friboulet, P. Thévenaz, and M. Unser, IEEE Transactions on Image Processing 18, 1179 (2009).
[27] E. Khoram, A. Chen, D. Liu, L. Ying, Q. Wang, M. Yuan, and Z. Yu, Photonics Research 7, 823 (2019). | 2025-03-04T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
4838,
1
],
[
4838,
10389,
2
],
[
10389,
14502,
3
],
[
14502,
18917,
4
],
[
18917,
24031,
5
],
[
24031,
24888,
6
]
]
} |
a0d21f560509be7587d893a26063e108c65a9126 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 13,
"total-fallback-pages": 0,
"total-input-tokens": 40928,
"total-output-tokens": 11864,
"fieldofstudy": [
"Physics",
"Materials Science"
]
} | Effect of Sputter Deposition on the Adhesion and Failure Behavior between Cu Film and Glassy Calcium Aluminosilicate: A Molecular Dynamics Study
Hyunhang Park and Sunghoon Lee *
Corning Technology Center Korea, Corning Precision Materials Co., Ltd., 212 Tangeong-ro, Asan 31454, Chungcheongnam-do, Korea; [email protected]
* Correspondence: [email protected]
Abstract: Understanding the physical vapor deposition (PVD) process of metallic coatings on an inorganic substrate is essential for the packaging and semiconductor industry. In this work, we investigate a Copper (Cu) film deposition on a glassy Calcium Aluminosilicate (CAS) by PVD and its dependence on the incident energy. Molecular dynamics simulation is adopted to mimic the deposition process, and pure Cu film is grown on top of CAS surface forming intermixing region (IR) of Cu oxide. In the initial stage of deposition, incident Cu atoms are diffused into CAS bulk and aggregated at the surface which leads to the formation of IR. When the high incident energy, 2 eV, is applied, 20% more Cu atoms are observed at the interface compared to the low incident energy, 0.2 eV, due to enhanced lateral diffusion. As the Cu film grows, the amorphous thin Cu layer of 1 nm is temporarily formed on top of CAS, and crystallization with face-centered cubic from amorphous structure follows regardless of incident energy, and surface roughness is observed to be low for high incident energy cases. Deformation and failure behavior of Cu-CAS bilayer by pulling is investigated by steered molecular dynamics technique. The adhesive failure mode is observed, which implies the bilayer experiences a failure at the interface, and a 7% higher adhesion force is predicted for the high incident energy case. To find an origin of adhesion enhancement, the distribution of Cu atoms on the fractured CAS surface is analyzed, and it turns out that 6.3% more Cu atoms remain on the surface, which can be regarded as a source for the high adhesion force. Our findings hopefully give the insight to understand deposition and failure mechanisms between heterogeneous materials and are also helping to further improve Cu adhesion in sputter experiments.
Keywords: Cu film; aluminosilicate glass; sputter; molecular dynamics; metal-inorganic interface; pulling force; adhesive failure
1. Introduction
Cu metallization on non-conducting materials such as glasses or ceramic substrates has been one of the great interests in microelectromechanical systems (MEMS), packaging for semiconductors, and display industries [1–4]. The adhesion between Cu film and the oxide glasses (e.g., silicate) is too weak to satisfy the criteria for commercial use of the product [5–7]. Note, Cu and glasses are inherently different kinds of materials from each other in terms of molecular interactions, in other words, the bonding natures of Cu and glasses are metallic and ionic, respectively. According to molecular orbital theory, unlike some late transition metal elements such as Ti or Zr, the outermost electron shell of the Cu element is not fully occupied [8]. Therefore, even if Cu makes bonding with electronegative elements in the glasses such as O, overlaps of electron clouds between them are not dense, which leads to a low bond dissociation energy between Cu and O [9].
To overcome this low adhesion strength due to the intrinsic property of Cu, there have been many persistent engineering efforts to improve the adhesion strength of Cu film on the glasses via various deposition techniques such as electroless plating, chemical
vapor deposition, and sputtering [5,9–12]. Among them, the sputter deposition technique has many advantages in obtaining uniform coating and good adhesion with the metal substrates as well as simplicity in the process development. However, practical adhesion of sputter-deposited Cu film on the glass substrate measured by experiments includes some inaccuracies depending on the kinds of adhesion measurement techniques such as pulling, peeling, or bending tests [5,13–16]. In addition, surface interactions during sputtering and underlying mechanisms for the adhesion and failure phenomena of the deposited film are still lacking.
Molecular dynamics (MD) simulation can be considered as a proper tool for the study of the effect of sputter on the adhesion between Cu film and glass. Firstly, MD simulation can track real-time movements and collisions of all atoms during the process, thereby help to understand the surface interactions between energetic incident atoms and the substrate. Indeed, deposition simulations for many kinds of inorganic films such as C, Ni/Cu/Ni, Al$_2$O$_3$, ZnO, or MgAl$_2$O$_4$ have been studied to reveal the mechanism of bond formation, atomic intermixing at the interface, and stress development [17–21]. Secondly, by mimicking experimental adhesion tests, one can directly calculate quantitative adhesion strength considering all kinds of molecular interactions between Cu film and glass substrate and analyze the failure behavior of the Cu-glass bilayer in nanoscale. This unique method which is called steered molecular dynamics (SMD) makes it possible to monitor the temporal evolution of free energy and applied force of the bilayer system while normal or shear stress is applied. Calculation of energy difference between separate state and adhered state of two layers, the conventional approach, does not guarantee a precise adhesion strength since it cannot consider a conformational change of the materials during the pulling path, especially in case of the interface where different kinds of materials form some mixture. On the other hand, the SMD method is expected to well describe adhesion between heterogeneous materials, as proved in the field of protein folding and deformation behavior of organic-inorganic interfaces [22,23].
In this work, a metal-inorganic bilayer formed by the Cu sputter deposition process on the glassy calcium aluminosilicate (CAS) is investigated by means of molecular dynamics simulations. CAS substrate is initially prepared, then Cu atoms are deposited on the substrate surface to form Cu-CAS bilayer, in which there are two distinct regions as an intermixing region (IR) around the top of CAS and the metallic Cu film on CAS up to 7 nm thickness. Structural features of both IR and Cu film regions are analyzed in detail. To estimate the effect of incident energy of Cu atoms on the adhesion, we compare levels of adhesion forces between high, 2 eV, and low, 0.2 eV, incident energy ($E_{\text{inc}}$) cases. Cu-CAS bilayer is pulled in the z direction by the SMD technique until it is completely separated into two parts. During the pulling test, the temporal evolution of the pulling force is tracked, and adhesion force is calculated along with the failure mode analysis. From this work, the critical factor for the adhesion is suggested, and the relevant mechanism is introduced.
2. Computational Methods
2.1. Materials and Building of CAS Substrate
Metallic Cu and oxide glass are considered as two ingredient materials that make an interface. CAS which belong to silicate glasses containing Al and Ca elements are used as a substrate. It is known that they are widely applied in commercial display glasses as well as electronic devices because of reinforced strength with good machinability [24,25]. The composition of CAS used in this simulation is 5CaO-20Al$_2$O$_3$-75SiO$_2$ in wt %, following the fact that Al composition is much higher than Ca composition in typical display glass. A thin Cu film is formed onto the glass substrate by a deposition process. Since every single Cu atom will be directly deposited in the simulation, any information such as preferred orientations of the Cu film does not need to be assumed. For the construction of CAS substrate, nano-sized glass blocks for the surface composition of CAS are constructed according to a consistent protocol of the melt-quench method. Initially, about 20,000 atoms are randomly located in a rectangular simulation box where periodic boundary conditions
are applied for x, y, and z directions. The size of the box is $6.2 \times 6.2 \times 6.2 \text{ nm}^3$. Then the box is heated from 300 K to 3600 K for 100 ps, melted at 3600 K for 200 ps, quenched with a cooling rate of 5 K/ps for 660 ps, and finally relaxed at room temperature for 200 ps [26–28]. NPT dynamics were used for the heating procedure while melting and quenching procedures were performed with NVT dynamics. Another NPT dynamics simulation was then performed for the gradual release of residual stress in the system. The density of relaxed bulk CAS was 2.54 g/cm$^3$, which is similar to the experimental density of 2.4 g/cm$^3$ within 10% error [29]. Then, the open boundary condition was applied in z direction to create the surface. The lower surface of 0.5 nm thick is fixed in its initial position to play a role of bulk inside while the upper surface is open as an empty space where a set of Cu atoms will be introduced afterward.
2.2. Simulation of the Film Growth by PVD
Figure 1 shows the overall modeling procedure for the deposition of Cu atoms and the pulling process. The deposition simulation is performed in the manner that Cu atoms are continuously dropped downward one by one onto the CAS substrate. Initially, the substrate is located at the bottom of the simulation box. When the simulation begins, a single Cu atom is placed at 5 nm above the substrate surface with the random position in x and y dimensions. The introduced atom is ‘shoot’ pointing downward normal to the substrate with a given velocity. Then it collides with the substrate and forms bonding with atoms of the substrate surface. This protocol is repeated until Cu atoms and substrate atoms interact with each other to generate an interlayer, and roughly 18,000 Cu atoms are deposited to form a 7 nm thickness of film from the top of CAS substrate during 100 ns. Basically, the NVE ensemble is applied to the whole system for energy conservation during collisions between the incident atoms and substrate. However, that protocol also leads to divergence of system temperature due to the continuous supply of incident atoms with high kinetic energies into the system. Thus, an NVT ensemble with 300 K of system temperature is applied to a set of atoms in the lower part of the substrate with 2.5 nm thickness, which means it plays the role of the heat bath. Additionally, the incidence of an atom is controlled to occur every 5 ps. It is known that this time period gives enough ‘cooling’ time so that the thermal fluctuation of the surface is diminished out through the substrate [20,30].
Figure 1. Scheme of modeling procedure for Cu sputter on glassy CAS and pulling process.
Since the current simulation focuses on the surface interaction between incident atoms and substrate, any other effects of process variables such as Ar pressure or Cathode power at the sputtering target are not considered. Two kinds of incident energies 2 eV and 0.2 eV are tested, which shall be referred to as high and low $E_{\text{inc}}$ cases, respectively, throughout the simulation. Value of 2 eV has been used as a typical $E_{\text{inc}}$ level of the sputtered particles in the simulations for metal sputter [18,30,31], whereas 0.2 eV is low enough to even represent the energy level of the heated particles in the evaporation process. For a clear comparison of the effect of $E_{\text{inc}}$ difference, the flux for both cases was fixed to be the same. The incident angle was set to be normal to the substrate throughout the simulation.
The whole simulation procedures described in Figure 1 were performed using the LAMMPS simulation package [32]. To describe both ionic and metallic interactions, two kinds of interatomic forcefields were used. The one is a non-bonded type Morse potential developed by Pedone et al. which has been proved to be good for a description of silicate glass [26–28], and the other is an embedded atom method (EAM) potential for Cu element optimized for the study of metal properties [33]. EAM potential parameters between different cations such as Cu, Al, Ca, and Si was not considered in this simulation since mainly required interactions relevant to adhesion and failure behavior are ionic interaction in IR and metallic interaction between Cu only in Cu film side. Both potentials were applied simultaneously during the whole deposition process. It is noted that electric neutral charges were applied to incident Cu particles during metal film growth since EAM potential does not require a fixed value for charge whereas Cu in IR was remained positively charged. We believe this is the most reasonable way to apply a fixed-charge potential to heterogeneous metal-inorganic bilayer since the Cu atom has a neutral charge when ejected from a sputtering target, but will be immediately polarized through hybridization between O 2p and Cu 3d electron state as soon as it makes a close contact to oxide network within its ionic radius [34]. The time step and cutoff for van der Waals interactions for simulation were set to be 1 fs and 12 Å, respectively. The Particle-Particle Particle-Mesh (PPPM) solver is used for the summation of long-range Coulomb interactions with the precision of $10^{-4}$.
2.3. SMD Simulation and Adhesion Calculation
To calculate adhesion force and monitor the failure behavior at the interface between sputter-deposited Cu and CAS substrate, we applied the SMD technique to mimic the experimental pulling test. SMD technique has been utilized to estimate the interfacial adhesion and analyze conformational changes of the polymer systems [22,35–38] and bio-materials [23,39]. In this technique, a fictitious atom is connected to the center of mass (COM) of the system of interest by a virtual spring with a spring constant $k$ at the outside the system, and the atom pulls it along the normal direction to the CAS substrate with a constant velocity $v$. During the pulling process, the total force and the potential energy are calculated as following equations:
$$U_{\text{spring}} = \frac{1}{2}k[vt - (R(t) - R_0) \cdot \mathbf{n}]^2$$ \hspace{1cm} (1)
$$F_{\text{spring}} = -\nabla U_{\text{spring}}$$ \hspace{1cm} (2)
where $R(t)$ is the current position of the COM of the system, $R_0$ is the initial center of mass of the system, and $n$ is a unit vector along the direction in which the spring is pulled. Then the total work done is calculated as:
$$W = \int_{r=R_0}^{r=R_f} \nabla U_{\text{spring}} \cdot d\mathbf{r}$$ \hspace{1cm} (3)
where $R_f$ is the final position of the COM of the system. It is well known that the ensemble average of total work done can be regarded as the potential of mean force (PMF) using Jarzynski equality \cite{40, 41}:
$$\langle \exp(-\beta W) \rangle_{\text{ensemble}} = \exp(-\beta U_{\text{PMF}})$$
(4)
where $\beta = \frac{1}{k_B T}$, $k_B$ is the Boltzmann constant and $T$ is the temperature. Cu film and the CAS substrate are pulled apart along positive and negative $z$ direction respectively, with a constant velocity of 10 m/s. The value of the pulling velocity was determined in such a way that adhesion value approached a saturation point for velocity values lower than 10 m/s.
3. Results and Discussion
3.1. Formation of Intermixing Region (IR)
Figure 2 briefly shows snapshots of the local surface structure of CAS substrate during the incidence of Cu atoms onto it. It is well known that the surface of silicate glass including alkali or alkali-earth elements has high-membered ring structures and defective sites such as tri-clustered oxygen and under-coordinated Si or Al \cite{42}. Figure 2a shows a local network structure that is comprised of two 4-membered rings and one 7-membered ring. The ring structures are denoted by a grey dotted line, and only Si-O and Al-O bonds are visualized because Si and Al are network formers of CAS glass. Two 4-membered rings share one corner with one tri-clustered oxygen. Figure 2b exhibits a first incident Cu atom which is adsorbed in the middle of a 7-membered ring. It is notable that its collision with high kinetic $E_{\text{inc}}$ of 2 eV did not result in the breakage of the silica network itself which means the ring structure is energetically stable. Interestingly, as shown in Figure 2c, we observe that when another incident Cu atom attacks a corner that 4-membered and 7-membered ring shares as denoted in the green line, the Al-O bond is broken and a 9-membered ring is formed. It implies same $E_{\text{inc}}$ may induce a change of ring structure when the collision occurs around the weak side. Figure 2c also shows that newly incident Cu atom aggregates to first incident Cu due to their metallic character within the 9-membered ring. Since the newly formed 9-membered ring is energetically not stable due to the 3-coordinated Si atom in its right corner, an aggregated Cu pair diffuses into the CAS to find its local energy minima, and the 4- & 7-membered ring structures are restored as shown in Figure 2d. In this way, the incident Cu atoms continuously transfer additional energy to the surface, where they can penetrate deeper into the CAS bulk region or form aggregates or clusters to cover the surface which leads to the formation of IR. In addition, thermal heating is also observed due to the collision of atoms, and temperature increases up to 500 K during 30 ns.
**Figure 2.** Incidence of Cu atoms to the local network on CAS surface: (a) Local silica structure with two 4-membered rings and one 7-membered ring (grey dotted line), and one tri-clustered oxygen (b) Incident Cu is adsorbed into the 7-membered ring (dark shaded region) (c) One of the bonds in the 4-membered ring is broken (green oval) due to incidence of another Cu atom (d) Aggregated Cu pair rotates to find its relaxed location in the 7-membered ring, then Al-O bond is recreated (cyan oval) and the 4-membered ring is restored.
Figure 3 presents two snapshots and cross-sections of the CAS substrate depending on $E_{\text{inc}}$ at the same time frame during the stage of IR formation. This stage may be roughly referred to as when Cu film almost fully covers the CAS surface. We observe that some Cu atoms weakly adsorbed on the surface form island-type aggregates. Since it is hard to find a difference in surface coverages of Cu between high and low $E_{\text{inc}}$ cases, the position of incident Cu along $z$ direction is analyzed. Figure 4 shows the distribution of Cu atoms near the CAS surface at the time frame of snapshots in Figure 3. To analyze the effects of high and low $E_{\text{inc}}$ cases on the distribution of Cu, we divide $z$ coordinate into three distinct regions of 48–55 Å, 55–61 Å, and 61–66 Å which will be referred to as region I, II, and III, respectively. It is observed that the black line is located over the red line in region I and III, which indicates that more Cu atoms exist in both regions when we apply high $E_{\text{inc}}$. In region III, it is expected that they form denser ionic bonds with silicate networks, and in region I, more Cu atoms deeply diffuse into CAS bulk as described in the inset. Since the total number of incident Cu atoms are the same for both cases, in region II, one can conjecture more Cu atoms are located for the low $E_{\text{inc}}$ case (red line in the figure) than the high $E_{\text{inc}}$ case. It is not surprising that the distribution of Cu shows a difference depending on the regions because incident Cu atoms for high $E_{\text{inc}}$ case will be more actively moved to both the lateral direction and vertical direction. Table 1 quantitatively compares the number of Cu atoms depending on the regions between high and low $E_{\text{inc}}$ cases. In regions I and III, we find that the high $E_{\text{inc}}$ case exhibits 48 and 20% more Cu atoms than the low $E_{\text{inc}}$ case, respectively. We can summarize the results as follows: In the high $E_{\text{inc}}$ case, due to the high kinetic energy of incident Cu atoms, they actively diffuse to both lateral and vertical directions, and thus they are strongly aggregated near the surface and penetrate more deeply into the CAS bulk. On the other hand, in the low $E_{\text{inc}}$ case, much of Cu atoms are relaxed around their initial incident positions and are diffused into the bulk region before they meet to be aggregated.
Table 1. Comparison of a number of Cu atoms in IR between high and low $E_{\text{inc}}$ cases for three regions along the $z$-direction. Numbers in red font mean that there are more Cu for that $E_{\text{inc}}$ case.
| Regions in $z$ Coordinate | Number of Cu for High $E_{\text{inc}}$ Case | Number of Cu for Low $E_{\text{inc}}$ Case |
|---------------------------|------------------------------------------|------------------------------------------|
| Whole region (48–66 Å) | 769 | 769 |
| Region I (48–55 Å) | 34 | 23 |
| Region II (55–61 Å) | 503 | 552 |
| Region III (61–66 Å) | 232 | 194 |
**Figure 3.** Comparison of IR on CAS substrate between (a) high $E_{\text{inc}}$ and (b) low $E_{\text{inc}}$ cases. Atom sizes of glass elements were reduced in the top view for emphasizing the adsorption of Cu atoms.
Figure 3. Comparison of IR on CAS substrate between (a) high E\textsubscript{inc} and (b) low E\textsubscript{inc} cases. Atom sizes of glass elements were reduced in the top view for emphasizing the adsorption of Cu atoms.
Figure 4. Comparison of $z$ position of Cu distribution in CAS between high and low E\textsubscript{inc} cases during the stage of IR formation. Region I and III show that there are more Cu for the high E\textsubscript{inc} case, whereas region II shows that there are more Cu for the low E\textsubscript{inc} case. The difference in Cu numbers between high and low E\textsubscript{inc} cases is magnified in the inset.
Table 1. Comparison of a number of Cu atoms in IR between high and low E\textsubscript{inc} cases for three regions along the $z$-direction. Numbers in red font mean that there are more Cu for that E\textsubscript{inc} case.
| Regions in $z$ Coordinate | Number of Cu for High E\textsubscript{inc} Case | Number of Cu for Low E\textsubscript{inc} Case |
|---------------------------|-----------------------------------------------|-----------------------------------------------|
| Whole region (48–66 Å) | 769 | 769 |
| Region I (48–55 Å) | 34 | 23 |
| Region II (55–61 Å) | 503 | 552 |
| Region III (61–66 Å) | 232 | 194 |
3.2. Cu Film Growth on IR
After the stage of IR formation, Cu atoms are continuously deposited to form Cu film. As described in the previous subsection, some Cu atoms gather together by diffusion on the surface to form island-type aggregates. It becomes a seed for nucleation of Cu film, and growth of Cu film is initiated following Volmer–Weber mode of thin metal film growth [43]. Figure 5a describes crystallization of Cu film structure with their thickness change from 1 nm to 7 nm by common neighbor analysis (CNA) during deposition process [44]. We observe the film initially takes an amorphous structure following the shape of CAS surface during 1 nm growth. This amorphization can be attributed to huge lattice strains and distortion applied to a bottom thin layer of Cu film due to direct bonding with amorphous CAS [45]. It is also consistent with the fact that an amorphous layer appears for both E\textsubscript{inc} cases, which suggests that potential energy by lattice strain is much higher than the level of the incident kinetic energy of Cu atoms. On the other hand, crystallinity shows up as the Cu film grows more than 1 nm thickness. This crystalline structure is maintained until the whole deposition process is finished, and CNA indicates the crystalline structure of Cu mainly belongs to the fcc crystal structure. X-ray diffraction analysis is also performed to check the crystallographic information of the film as shown in Figure 5b. When Cu film is 1 nm thick, the intensity is very broadly dissipated over the range of 2 theta degree which implies the overall structure is still amorphous. However, after deposition ends, new peaks are developed at the values of 43.4°, 50.2°, and 73.7° whose corresponding surface orientations are (111), (200), and (220) of fcc, respectively. Since the main peak among them is (111), one can say that the Cu film grown in this simulation is a single crystal, and its preferred orientation along the z-axis is (111) [46].
Figure 5. (a) Crystallization of Cu film during deposition process (substrate part was not shown for clear comparison). In the 1 nm thickness, the film shows an amorphous structure (left), but the film is crystallized when it grows as 7 nm thickness (right) (b) X-ray diffraction (XRD) patterns of Cu film grown in the simulation.
Figure 6 shows typical surface morphologies of Cu film after deposition for high and low $E_{\text{inc}}$ cases. Comparison between Figure 6a,b gives a striking difference in surface morphologies of Cu film between high and low $E_{\text{inc}}$ cases. One readily finds that the surface of Cu film for the high $E_{\text{inc}}$ case is almost flat. On the other hand, the low $E_{\text{inc}}$ case yields much of the peaks and valleys with some pores inside the film, which results in high surface roughness. This difference is not surprising because incident particles with high $E_{\text{inc}}$ not only help rearrangement of surface silica networks by the collision but also themselves actively hop over the surface to create a flat terraced structure. This mechanism is called atomic peening [47]. Meanwhile, if incident particles have low energies then it is not easy for them to arrive at valley points, and thus peak points have more chances to be deposited than valley points which are referred to as the shadowing effect [48].
Figure 6. Surface morphology of Cu films grown onto CAS substrate: (a) high $E_{\text{inc}}$ and (b) low $E_{\text{inc}}$ case.
### 3.3. Deformation and Failure of Cu-CAS Bilayer
In this subsection, pulling simulation is performed to Cu-CAS bilayer along $z$ direction. As mentioned in Section 2.3, a set of Cu atoms in a rectangular region whose thickness is half of the Cu film are pulled upward by a virtual spring. As pulling simulation begins, tensile stress is applied to the bilayer and induces severe strain of the bilayer along $z$ direction. This strain is represented by the increased center of mass (COM) of pulled Cu atoms, which is called pulling distance. One can monitor the continuous change of force during pulling until the bilayer experience a failure. Figure 7a shows typical temporal curves of pulling force as a function of pulling distance. As pulling distance increases, the pulling force also increases which implies that the bilayer withstands its molecular interactions against the external harmonic force. However, it draws a maximum at around 9 Å of pulling distance, then rapidly diminishes to zero. This abrupt change corresponds to a failure of bilayer at that distance, and the maximum value of pulling force can be regarded as an adhesion force. It is notable that the difference in the shape of the force curve between high and low $E_{\text{inc}}$ cases gives a clue for structural features of the Cu-CAS interface. Around the point of maximum pulling force, it seems that the high $E_{\text{inc}}$ case undergoes more severe strain hardening, while the low $E_{\text{inc}}$ case shows more ductile behavior which is inferred from the low slope. Figure 7b compares maximum pulling forces between high and low $E_{\text{inc}}$ cases, averaged over 5 samples for each. The calculation result exhibits that the high $E_{\text{inc}}$ case shows a 7% of higher adhesion force than the low $E_{\text{inc}}$ case.
To clearly analyze how $E_{\text{inc}}$ affects adhesion force, we firstly focus on the region of failure and morphology of the fractured surface. One can imagine three possible locations of failure, which are inside the CAS, the interface between CAS and amorphous Cu layer, and inside Cu film. Among them, we observe the bilayer experiences a failure at the interface between CAS and amorphous Cu layer for both high and low $E_{\text{inc}}$ cases as shown in Figure 8. In other words, it undergoes an adhesive failure irrespective of the incident energy and implies the interface is the weakest part of the bilayer system. This is not surprising because the other two locations are essentially bulk regions of glass and Cu film, and their cohesive strengths are expected to be high compare to the interface. Furthermore, the top view in Figure 8 indicates that Cu is a main element on the fractured surface. As mentioned in the previous subsection, Cu atoms act as a nucleation site at the interface, and thus the number of Cu atoms on the CAS surface would be a major contributor to the adhesion strength.
For quantitative analysis, the number of atoms, specifically, Cu and O are counted along the z direction and presented in Figure 9. As already indicated in Figure 4 and Table 1, there were 20% more Cu atoms on the CAS surface for the high E\textsubscript{inc} case compared to that for the low E\textsubscript{inc} case during the stage of IR formation. A similar trend is observed on the fractured surface when comparing Figure 9a,b. The green area in the graph represents additional Cu atoms at the surface after the formation of Cu oxide assuming that amount of interstitial Cu in the oxide is negligible, and thus it can be regarded as exposed Cu atoms on the fractured surface. As shown in Figure 9c, there are 6.3% more Cu atoms in the green area for the high E\textsubscript{inc} case than that for the low E\textsubscript{inc} case. These additional Cu atoms mainly form metallic bonding with Cu film, and thus finally lead to 7% higher adhesion for the high E\textsubscript{inc} case. Based on the analysis so far, we can propose a mechanism of the whole procedure of Cu deposition, and clarify the origin of the difference in adhesion.
In the stage of IR formation, due to active lateral and vertical diffusion, around 20% more Cu atoms with high E\textsubscript{inc} were distributed to the surface than those with low E\textsubscript{inc}. Although some of the Cu atoms may diffuse into Cu film and glass during the whole deposition process, the growth of thick Cu film onto it effectively conserves the shape and composition of IR, and thus some more Cu atoms could still remain at the Cu-CAS interface as an amorphous thin layer. These excessive Cu atoms finally resulted in higher adhesion than the low E\textsubscript{inc} case. Even though the difference in adhesion value between high and low E\textsubscript{inc} cases looks to be small, we emphasize that our results showed control of a single sputter process parameter can bring into an increase of adhesion with an understanding of the underlying mechanism by analysis of its failure behavior. Based on this understanding, it is expected that changes of other parameters such as raise of incident ion flux can, even more, increase the adhesion in the experiment. Also, if one aims to further increase adhesion, we can attempt to control the bonding strength of an amorphous thin layer of Cu which is another factor for adhesion force irrespective of E\textsubscript{inc}. For example, an increase in the amount of oxygen at the CAS surface by surface treatment with oxygen plasma will lead to enhancement of bonding strength of the amorphous Cu layer.
**Figure 8.** Adhesive failure of Cu-CAS bilayer and top view of the surface after the failure for (a) high E\textsubscript{inc} and (b) low E\textsubscript{inc} cases.
Figure 9. Count of O and Cu atoms along z coordinate around the fractured CAS surface for (a) high E\text{inc} case and (b) low E\text{inc} case. Shaded regions represent the number of Cu atoms that are expected to form metallic bonds with Cu film, thus contributing to adhesion strength. (c) Comparison of a number of Cu atoms in the green regions for high and low E\text{inc} cases.
4. Conclusions
To find the effect of incident energy of sputter deposition, adhesion and failure behavior of Cu-CAS bilayer is investigated by molecular dynamics simulations. High and low E\text{inc} of 2 eV and 0.2 eV, respectively, are compared throughout the deposition process and pulling test. In the stage of IR formation, it is observed that incident Cu atoms are diffused into bulk CAS and are aggregated near the surface. It turns out that 20% more Cu atoms for the high E\text{inc} case are distributed at the surface than that for the low E\text{inc} case due to active lateral diffusion with their high kinetic energies. During the stage of Cu film growth, incident Cu atoms are nucleated to form a metallic Cu film with crystallization of fcc type. Irrelevant to incident energy, an amorphous thin layer of Cu is initially grown as 1 nm thickness due to high lattice strains by the bonding with CAS while crystallization of Cu film is observed when the film is deposited more than 1 nm height. High and low E\text{inc} cases disclose a clear difference in their surface morphologies, namely, low and high surface roughness of the film, respectively. Pulling test with SMD technique reveals that failure occurs at the interface between Cu film and CAS for both high and low E\text{inc} cases, while the former gives 7% higher adhesion force than the latter. Analysis for z positions of O and Cu atoms at the fractured CAS surface shows that there are 6.3% more Cu atoms at the surface for the high E\text{inc} case compared to the low E\text{inc} case. It is suggested that additional Cu atoms remained after participating in the formation of Cu oxide from metallic bonding with Cu film, which finally leads to higher adhesion. Our work ultimately implies that an effect of even a single process variable such as incident energy can result in the non-negligible difference in surface structure and failure behavior. We believe our finding will be helpful to understand the deposition mechanism as well as structure-property relations for the interface between heterogeneous materials.
Author Contributions: H.P., writing—original draft and formal analysis; S.L., conceptualization, writing—original draft, and supervision. Both authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Acknowledgments: We would like to acknowledge Mathew E. McKenzie for English language editing as well as fruitful comments and discussions. We also would like to thank Hong Yoon for supporting our work.
Conflicts of Interest: The authors declare no conflict of interest.
References
1. Tilli, M.; Mottoka, T.; Airaksinen, V.-M.; Fransilla, S.; Paulasto-Krockel, M.; Lindroos, V. Handbook of Silicon Based MEMS Materials and Technologies, 2nd ed.; Elsevier: London, UK, 2015.
2. Lu, D.; Wong, C.P. Materials for Advanced Packaging; Springer: New York, NY, USA, 2008.
3. Datta, M. Microelectronic Packaging Trends and the Role of Nanotechnology; Springer: New York, NY, USA, 2010.
4. Ohno, T. Thin-Film PVD: Materials, Processes, and Equipment. In Flat Panel Display Manufacturing; Souk, J.M.S., Luo, F.-C., Bita, I., Eds.; John Wiley and Sons: Hoboken, NJ, USA, 2018.
5. Mittal, K.L. Adhesion Measurement of Films and Coatings, 2nd ed.; CRC Press: Boca Raton, FL, USA, 2014; Volume 1.
6. Teraoka, A.; Watanabe, M.; Nabetani, Y.; Kondoh, E. Room-Temperature Formation of a ZnO-Based Adhesion Layer for Nanoprecision Cu/Glass Metallization. Jpn. J. Appl. Phys. 2013, 52, 05FB04. [CrossRef]
7. Benjamin, P.; Weaver, C. The Adhesion of Evaporated Metal Films on Glass. Proc. R. Soc. Lond. A 1962, 261, 516–531.
8. Mingos, D.M.P.; Day, P.; Dahl, J.P. Molecular Electronic Structures of Transition Metal Complexes; Plenum Press: New York, NY, USA, 1991.
9. Haynes, W.M. CTC Handbook of Chemistry and Physics, 95th ed.; CRC Press: Boca Raton, FL, USA, 2014.
10. Lee, L.H. Fundamentals of Adhesion; Plenum Press: New York, NY, USA, 1991.
11. Paunovic, M.; Schlesinger, M. Fundamentals of Electrochemical Deposition, 2nd ed.; John Wiley & Sons Inc.: Hoboken, NJ, USA, 2006.
12. Wasa, K.; Kanno, I.; Kotera, H. Handbook of Sputter Deposition Technology: Fundamentals and Applications for Functional Thin Films, Nanomaterials, and MEMS, 2nd ed.; Elsevier Inc.: Amsterdam, The Netherlands, 2012.
13. Pizzi, A.; Mittal, K.L. Handbook of Adhesive Technology; Marcel Dekker Inc.: New York, NY, USA, 2003.
14. Piggot, M. Why Interface Testing by Single-fiber Methods can be Misleading. Compos. Sci. Technol. 1997, 57, 965–974. [CrossRef]
15. Favre, J.P.; Desarmot, G.; Sudre, O.; Vassel, A. Were McGarry or Shiriajeva Right to Measure Glass. J. Reinf. Plast. Compos. 1997, 4, 313–326. [CrossRef]
16. Hampe, A.; Marotzke, C. The Energy Release Rate of the Fiber/polymer Matrix Interface: Measurement. J. Reinf. Plast. Compos. 1997, 16, 341–352. [CrossRef]
17. Marks, N.A.; McKenzie, D.R.; Pailthorpe, B.A. Molecular-dynamics Study of Compressive Stress Generation. Phys. Rev. B. 1996, 53, 4117–4124. [CrossRef] [PubMed]
18. Zhou, X.W.; Wadley, N.G. Atomistic Simulations of the Vapor Deposition of Ni/Cu/Ni Multilayers: The Effects of Adatom Incident Energy. J. Appl. Phys. 1998, 84, 3201–3215. [CrossRef]
19. Houska, J. Pathway for a Low-temperature Deposition of α-Al₂O₃: A Molecular Dynamics Study. Surf. Coat. Technol. 2013, 235, 333–341. [CrossRef]
20. Hong, Z.-H.; Fang, T.-H.; Hwang, S.-F. Atomic-level Stress and Induced Growth of Wurtzite Zinc Oxide Using Molecular Dynamics Simulation. J. Phys. D Appl. Phys. 2011, 44, 505301. [CrossRef]
21. Georgieva, V.; Saraiva, M.; Jehanathan, N.; Levelev, O.I.; Depla, D.; Bogaerts, A. Sputter-deposited Mg-Al-O Thin Films: Linking Molecular Dynamics Simulations to Experiments. J. Phys. D Appl. Phys. 2009, 42, 065107. [CrossRef] [PubMed]
22. Min, K.; Kim, Y.; Goyal, S.; Lee, S.H.; Savoy, E.; Rammohan, A.R.; Mauro, J.C.; Kim, H. Interfacial Adhesion Behavior of Polyimides on Silica Glass: A Molecular Dynamics Study. Polymer 2016, 98, 1–10. [CrossRef]
23. Davis, D.A.; Hamilton, A.; Yang, J.; Cremar, L.D.; Goyal, S.; Kim, Y.; Goyal, S.; Savoy, E.; Rammohan, A.R.; Mauro, J.C.; Kim, H.; Vassel, A. Were McGarry or Shiriajeva Right to Measure Glass. Surf. Coat. Technol. 2013, 235, 333–341. [CrossRef]
24. Varshneya, A.K. Fundamentals of Inorganic Glasses, 2nd ed.; Society of Glass Technology: Sheffield, UK, 2013; Volume 1.
25. Musgrave, J.D.; Hu, J.; Calves, L. Springer Handbook of Glass; Springer Nature: Sheffield, UK, 2019.
26. Pedone, A. Properties Calculations of Silica-Based Glasses by Atomic Simulation Techniques: A Review. J. Phys. Chem. C 2009, 113, 20773–20784. [CrossRef]
27. Pedone, A.; Malavasi, G.; Menziani, M.C.; Segre, U.; Cormack, A.N. Molecular Dynamics Studies of Stress-Strain Behavior of Silica Glass under a Tensile Load. Chem. Mater. 2008, 20, 4356–4366. [CrossRef]
28. Pedone, A.; Malavasi, G.; Cormack, A.N.; Segre, U.; Menziani, M.C. Insight into Elastic Properties of Binary Alkali Silicate Glasses: Prediction and Interpretation through Atomistic Simulation Techniques. Chem. Mater. 2007, 19, 3144–3154. [CrossRef]
29. Bansal, N.P.; Doremus, R.H. Handbook of Glass Properties; Academic Press: Orlando, FL, USA, 1986.
30. Xie, L.; Brault, P.; Bauchire, J.-M.; Thomann, A.-L.; Bedra, L. Molecular Dynamics Simulations of Clusters and Thin Film Growth in the Context of Plasma Sputtering Deposition. J. Phys. D Appl. Phys. 2014, 47, 224004. [CrossRef]
31. Wadley, H.N.G.; Zou, W.; Zhou, X.W.; Groves, J.F. Multiscale Simulations of the RF Diode Sputtering of Copper. Mat. Res. Soc. Symp. Proc. 1999, 536, 323–328. [CrossRef]
32. Plimpton, S. Parallel Algorithms for Short-Range Molecular Dynamics. J. Comput. Phys. 1995, 117, 1–19. [CrossRef]
33. Sheng, H.W.; Kramer, M.J.; Cadieu, A.; Fujita, T.; Chen, M.W. Highly Optimized EAM Potentials for 14 fcc Metals. Phys. Rev. B 2011, 83, 134118. [CrossRef]
34. Zhang, S.-T.; Li, C.-M.; Yan, H.; Evans, D.G.; Duan, X. Density Functional Theory Study on the Metal-Support Interaction between Ru Cluster and Anatase TiO₂(110) Surface. J. Phys. Chem. C 2014, 118, 3514–3522. [CrossRef]
35. Goyal, S.; Park, H.-H.; Lee, S.H.; Savoy, E.; McKenzie, M.E.; Rammohan, A.R.; Mauro, J.C.; Kim, H.; Min, K.; Cho, E. Characterizing the Fundamental Adhesion of Polyimide Monomers on Crystalline and Glassy Silica Surfaces: A Molecular Dynamics Study. J. Phys. Chem. C 2016, 120, 25631–25639. [CrossRef]
36. Lee, S.H.; Stewart, R.J.; Park, H.; Goyal, S.; Botu, V.; Kim, H.; Min, K.; Cho, E.; Rammohan, A.R.; Mauro, J.C. Effect of Nanoscale Roughness on Adhesion between Glassy Silica and Polyimides: A Molecular Dynamics Study. *J. Phys. Chem. C* 2017, 121, 24648–24656. [CrossRef]
37. Min, K.; Rammohan, A.R.; Lee, H.S.; Shin, J.; Lee, S.H.; Goyal, S.; Park, H.; Mauro, J.C.; Stewart, R.; Botu, V.; et al. Computational Approaches for Investigating Interfacial Adhesion Phenomena of Polyimide on Silica Glass. *Sci. Rep.* 2017, 7, 10475. [CrossRef]
38. Min, K.; Rammohan, A.R.; Lee, S.H.; Goyal, S.; Park, H.; Stewart, R.; He, X.; Cho, E. Grafting Functional Groups in Polymeric Binder toward Enhancing Structural Integrity of LixSiO2 Anode during Electrochemical Cycling. *J. Phys. Chem. C* 2018, 122, 17190–17198. [CrossRef]
39. Chabria, M.; Gertig, S.; Smith, M.L.; Vogel, V. Stretching Fibronectin Fibres Disrupts Binding of Bacterial Adhesins by Physically Destroying an Epitope. *Nat. Commun.* 2010, 1, 135. [CrossRef]
40. Jarzynski, C. Nonequilibrium Equality for Free Energy Differences. *Phys. Rev. Lett.* 1997, 78, 2690. [CrossRef]
41. Park, S.; Schulten, K. Calculating Potentials of Mean Force from Steered Molecular Dynamics Simulations. *J. Chem. Phys.* 2004, 120, 5946. [CrossRef]
42. Agnello, G.; Wang, L.; Smith, N.; Manley, R.; Cormack, A. Structure of CAS Glass Surfaces and Eletrostatic Contact Charging Behavior: A Joint Simulation and Experimental Investigation. *Int. J. Appl. Glass Sci.* 2021, 12, 111–123. [CrossRef]
43. Herman, M.A.; Richter, W.; Sitter, H. *Epitaxy: Physical Principles and Technical Implementation*; Springer: Berlin, Germany, 2004.
44. Stukowski, A.; Bulatov, V.V.; Arsenlis, A. Automated Identification and Indexing of Dislocations in Crystal Interfaces. *Modeling Simul. Mater. Sci. Eng.* 2012, 20, 085007. [CrossRef]
45. Zhang, L.; Zhang, H.; Ren, X.; Eckert, J.; Wang, Y.; Zhu, Z.; Gemming, T.; Pauly, S. Amorphous Martensite in \( \beta \)-Ti Alloys. *Nat. Commun.* 2018, 9, 506. [CrossRef] [PubMed]
46. Salavati-Niasari, M.; Davar, F. Synthesis of Copper and Copper(I) Oxide Nanoparticles by Thermal Decomposition of a New Precursor. *Mater. Lett.* 2009, 63, 441–443. [CrossRef]
47. Windischmann, H. Intrinsic Stress in Sputter-deposited Thin Films. *Crit. Rev. Solid State* 1992, 17, 547–596. [CrossRef]
48. Thornton, J.A.; Hoffman, D.W. Stress Related Effects in Thin Films. *Thin Solid Film.* 1989, 171, 5–31. [CrossRef] | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
3585,
1
],
[
3585,
8096,
2
],
[
8096,
10778,
3
],
[
10778,
14599,
4
],
[
14599,
18012,
5
],
[
18012,
21590,
6
],
[
21590,
25195,
7
],
[
25195,
26692,
8
],
[
26692,
29816,
9
],
[
29816,
32611,
10
],
[
32611,
35639,
11
],
[
35639,
41501,
12
],
[
41501,
43985,
13
]
]
} |
d769ceea8f0bbf04a123d4c717421238288ae652 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 26,
"total-fallback-pages": 0,
"total-input-tokens": 92677,
"total-output-tokens": 24445,
"fieldofstudy": [
"Biology"
]
} | Homology sensing via non-linear amplification of sequence-dependent pausing by RecQ helicase
Yeonee Seol††, Gábor M Harami‡‡, Mihály Kovács‡,‡*, Keir C Neuman*†
Abstract RecQ helicases promote genomic stability through their unique ability to suppress illegitimate recombination and resolve recombination intermediates. These DNA structure-specific activities of RecQ helicases are mediated by the helicase-and-RNAseD like C-terminal (HRDC) domain, via unknown mechanisms. Here, employing single-molecule magnetic tweezers and rapid kinetic approaches we establish that the HRDC domain stabilizes intrinsic, sequence-dependent, pauses of the core helicase (lacking the HRDC) in a DNA geometry-dependent manner. We elucidate the core unwinding mechanism in which the unwinding rate depends on the stability of the duplex DNA leading to transient sequence-dependent pauses. We further demonstrate a non-linear amplification of these transient pauses by the controlled binding of the HRDC domain. The resulting DNA sequence- and geometry-dependent pausing may underlie a homology sensing mechanism that allows rapid disruption of unstable (illegitimate) and stabilization of stable (legitimate) DNA strand invasions, which suggests an intrinsic mechanism of recombination quality control by RecQ helicases.
Introduction RecQ helicases are a family of DNA helicases that play essential roles in maintaining genomic integrity through extensive involvement in DNA recombination, replication, and repair pathways (Bachrati and Hickson, 2003; Bennett and Keck, 2004; Chu and Hickson, 2009). Escherichia coli RecQ (Ec RecQ) helicase is the founding member of the family (Nakayama et al., 1984) and plays roles in both suppressing illegitimate recombination and facilitating various steps of DNA recombinational repair (Hanada et al., 1997; Ryder et al., 1994; León-Ortiz et al., 2018). RecQ helicases are highly conserved from bacteria to humans and eukaryotic RecQ helicases have been shown to play similar pro- and anti-recombination functions. Most unicellular organisms, such as E. coli and yeast, express a single RecQ homolog, whereas multi-cellular organisms often possess multiple RecQ helicases specialized to different roles in genome maintenance processes.
The fundamental conserved activity of RecQ helicases is the ATP-dependent unwinding of double-stranded DNA (Nakayama et al., 1984). All RecQ members possess two evolutionarily conserved RecA-like helicase domains with an ATP binding and hydrolysis site located in a cleft between them (Bennett and Keck, 2004; Chu and Hickson, 2009; Bachrati and Hickson, 2008). Similar to other superfamily (SF) one and SF2 helicases, RecQ members also contain N- and C-terminal accessory domains that provide additional or specialized functionalities (Fairman-Williams et al., 2017).
The RecQ C-terminal domain (RQC) comprises zinc binding and winged-helix (WH) sub-domains associated with protein structural integrity and duplex DNA binding, respectively. Although less conserved, many RecQ-family members, including Ec RecQ and multiple human RecQ homologs, possess an accessory single-stranded (ss) DNA-binding module termed the helicase-and-RNaseD-C-terminal (HRDC) domain (Bernstein and Keck, 2005; Vindigni and Hickson, 2009). The HRDC, while generally dispensable for helicase activity, is critical for certain recombination intermediate processing steps, such as disruption of displacement strand (D-loop) invasion and double Holliday junction resolution (Rezazadeh, 2012; Singh et al., 2012; Chatterjee et al., 2014; Harami et al., 2017). Biochemical studies have established that full length RecQ has a higher ssDNA binding affinity than RecQ constructs lacking the HRDC, which is consistent with the findings that the interaction between the HRDC and ssDNA contributes to DNA substrate specificity of RecQ helicases (Bernstein and Keck, 2005; Vindigni and Hickson, 2009).
Recently, we provided evidence that HRDC interactions contribute to DNA substrate-geometry dependent binding orientation and unwinding by RecQ, and demonstrated that these HRDC-mediated interactions play a role in suppressing illegitimate recombination in E. coli (Harami et al., 2017). Whereas these findings indicate that the HRDC strongly favors binding of RecQ to D-loop structures in an orientation that promotes disruption of the invading DNA strand (Harami et al., 2017), it is not clear how RecQ can subsequently discriminate between homologous and non-homologous strand invasions; once correctly oriented on the D-loop, RecQ can unwind any invading strand and indiscriminately disrupt all D-loop formations. In this work we identify a potential solution to this quandary, suggested by the observation that HRDC-dependent pausing during hairpin DNA unwinding is not random but occurs repeatedly at distinct positions on the DNA hairpin. We reason that if the frequency or duration of the unwinding pauses is related to the degree of DNA homology, then the more than 10-fold decrease in average unwinding rate due to pausing can provide a mechanism of homology sensing. Thus, if pausing is correlated with homology, then the resulting modulation of...
the average unwinding rate of an oriented RecQ helicase will result in discrimination of legitimate versus illegitimate D-loops.
To test this theory, we set out to determine the origin of HRDC-mediated pausing by investigating the unwinding mechanism of *E. coli* RecQ and HRDC-induced pausing using single-molecule magnetic tweezers (MT)-based assays and rapid transient kinetic assays. We found that long-lived HRDC-induced pauses of wild type RecQ (RecQ<sup>wt</sup>) and shorter-lived pauses of RecQ core domain (HRDC deletion mutant; RecQ-dH) are sequence dependent and both correlate with DNA duplex stability. Sequence-dependent pausing is a direct consequence of the unique DNA unwinding mechanism: RecQ unwinds one base-pair per ATP hydrolysis cycle but releases the nascent ssDNA only after unwinding ~5 bp. The translocation kinetics arising from this 5 bp kinetic step depend on the duplex stability, which results in sequence-dependent pausing of the core RecQ that is further stabilized by the HRDC binding to the displaced ssDNA. Kinetic modeling indicates that the affinity of the HRDC for ssDNA is enhanced at pause sites, rather than remaining constant. The HRDC thus acts as a non-linear amplifier of the transient sequence-dependent pauses of the core enzyme. Our study demonstrates that the coupling between the core unwinding mechanism and the HRDC-ssDNA interactions dramatically alter the mode of unwinding in a sequence dependent manner, and, in conjunction with previous work, potentially implicates a mechanistic basis for recombination quality control provided by RecQ helicases.
**Results**
**RecQ pause positions are strongly correlated with DNA sequence**
Single-molecule measurements of RecQ helicase unwinding activity were performed with 174- or 584-base pair DNA hairpins using an MT apparatus (Figure 1A). DNA hairpin substrates were attached to the flow-cell surface and to a 1- or 2.8 μm magnetic bead via a 1.1 kbp double-stranded DNA handle and 60-nucleotides of single-stranded poly-dT, respectively (Figure 1A). Measurements with the DNA hairpin were performed at a constant force of 8 pN under which the hairpin did not open spontaneously. In the presence of RecQ helicase (20–100 pM), unwinding activity was monitored in real-time by tracking the three-dimensional position of a tethered bead at 60 or 200 Hz. Trajectories of the bead extension as a function of time were analyzed by fitting with a T-test based step finding algorithm to obtain the mean unwinding rate, the ‘step’ unwinding rate between pauses, the pause positions, and the pause durations (Harami et al., 2017; Seol et al., 2016).
As described previously (Harami et al., 2017), frequent pausing and strand-switching by WT RecQ (RecQ<sup>WT</sup>) is caused by the HRDC as the HRDC deletion mutant (RecQ-dH) shows significantly less pausing during DNA hairpin unwinding (Figure 1B). Pausing is attributed to transient binding of the HRDC domain to the displaced single-stranded DNA behind RecQ. Since both the displaced and the translocation strands of ssDNA are under tension in the hairpin substrate, binding of the HRDC to the displaced strand will prevent forward motion of the helicase. HRDC binding to either duplex DNA ahead of the helicase, or to the translocation strand of ssDNA behind the helicase, are ruled out by the lack of pauses during the unwinding of a ‘gapped’ DNA substrate in which the displaced strand is not constrained (Harami et al., 2017). Given the mechanical origin of the pausing associated with transient binding of the HRDC, the pause positions would be expected to be random, dependent on the stochastic kinetics of the interaction between HRDC and the displaced ssDNA.
Interestingly, the dwell-time histogram as a function of position for RecQ<sup>WT</sup> unwinding traces exhibits peaks at distinct positions along the hairpin (Figure 2A; top). The peaks in the dwell-time histogram of unwinding traces arise from long and/or frequent pauses at specific positions during DNA hairpin unwinding by RecQ helicase (Figure 2—figure supplement 1). To identify the sequence context of the pauses, the extension change associated with DNA hairpin opening was converted to base-pairs via the worm-like chain (WLC) model of DNA (Manosas et al., 2010). Each unwound base pair resulted in the increase of the molecular extension by two ssDNA nucleotides, which at an applied force of 8 pN corresponds to ~0.8 nm assuming a 1 nm persistence length and a 0.65 nm inter-phosphate distance. With this conversion factor, the extension change for the fully open hairpin was 174 bp, consistent with the actual DNA hairpin size (174 bp). To determine if pausing is related to DNA base-pair energy, we compared the unwinding dwell time histogram
Figure 1. DNA hairpin unwinding activity of RecQ helicase is modulated by the HRDC domain. (A) Cartoon representation of the experimental scheme (not to scale). The 3' biotinylated end of the single-stranded poly-dT segment (blue) is attached to a streptavidin-coated 1- or 2.8 μm magnetic bead (brown sphere), whereas the 5' digoxigenin-labeled double-stranded handle (black line) is attached via anti-digoxigenin (red square) to the surface of the flow-cell. Small magnets above the flow-cell apply a constant upwards force on the magnetic bead. RecQ (purple and green RecA-like domains, yellow zinc binding and winged helix domain, orange HRDC domain) binds at the base of the hairpin (blue helix) and unwinds it, which results in the increase in the extension of the bead. (B) Individual unwinding events of RecQ<sup>WT</sup> and RecQ-dH. Unwound DNA indicates the amount of DNA hairpin opened by RecQ in base pairs. The ends of unwinding events are indicated by a yellow pointer. Pause locations identified from T-test fitting are indicated as solid red lines. Additional RecQ-dH unwinding traces are displayed to show the range of average unwinding rates (gray lines; note that only the region from the beginning to the maximum unwound positions are plotted).
DOI: https://doi.org/10.7554/eLife.45909.003
The following source data is available for figure 1:
Source data 1. Source data for Figure 1.
DOI: https://doi.org/10.7554/eLife.45909.004
(Figure 2A; top) with the DNA base-pair stability calculated by performing a running average (6 bp window) of the exponential of the DNA base-pair energy for the 174 bp DNA hairpin sequence based on the nearest neighbor base-pair energy model (Patten et al., 1984; SantaLucia, 1998; Huguet et al., 2010). We found that the peak locations of pausing and duplex stability were highly correlated (Figure 2A; bottom). The exact locations of peaks were identified by globally fitting the dwell-time histogram and the exponential of the average DNA melting energy with the sums of Gaussian distributions (Figure 2—figure supplement 2). The relationship between pausing during unwinding and the peaks in the dwell-time histogram is explained in Figure 2—figure supplement 2B (top). Consistent with this observation, pause positions from the dwell time histogram of RecQ<sup>WT</sup> were linearly correlated with peak positions from the DNA base-pair energy profile (Figure 2B; top) with a slope of 0.99 ± 0.03, linear correlation coefficient (Pearson’s r) of 0.97, and $\chi^2 = 0.85$, indicating a strong linear correlation. The sequence around the peak positions (±4 bp) contained a high percentage of GC (~70%), consistent with the finding that the pause positions are related to the duplex stability of the DNA. This finding raises the question of how HRDC-dependent pausing is correlated with DNA base-pair melting despite the fact that the HRDC itself does not play a role in unwinding DNA or exhibit sequence-specific ssDNA binding. We hypothesized that the HRDC may amplify or stabilize transient pauses associated with RecQ core domain (RecQ-dH) encountering regions of increased duplex stability (high GC content).
Sequence-dependent pausing originates from sequence-dependent unwinding kinetics
To test this hypothesis, we determined if the transient pausing positions of RecQ-dH correlated with the peaks in the DNA base-pair stability curve (Figure 2B). The pause positions for RecQ-dH were obtained from dwell time histograms following the same procedure used for RecQ<sup>WT</sup> (Figure 2—figure supplement 2C) and plotted as a function of the peak positions of DNA base-pair stability (Figure 2B). The pause positions of RecQ-dH were linearly correlated with the duplex stability peaks, returning a slope of 0.96 ± 0.06, Pearson’s r = 0.95, and \( \chi^2 = 1.1 \). Moreover, the pause positions of RecQ-dH are statistically identical to those of RecQ<sup>WT</sup>, confirming that HRDC-dependent pausing likely originates from stabilization of sequence-dependent unwinding kinetics of RecQ helicase. Sequence-dependent pausing by RecQ-dH reveals important mechanistic insights into the unwinding and translocation mechanism. If the enzyme unwinds one base pair per each kinetic step, the largest energy difference for a single base-pair opening (G/C vs A/T) is ~2.0 \( k_B T \) so the pause duration ratio of G/C to A/T will be a maximum of ~7 fold. However, the roughly 20-fold difference in the time the enzyme requires to unwind DNA at the longest pause duration sites in comparison to the average unwinding rate, suggests that more than a single base pair is being opened by the enzyme during each kinetic step. Following this simple analysis, we suggest that pausing is governed by a combination of the DNA base-pair stability and the number of base pairs melted by the helicase during each kinetic step. During processive unwinding, this melting step is the rate limiting step that determines the unwinding rate and pause durations.
Simulation of unwinding mechanism of RecQ reveals multi-base pair kinetic step
To distinguish among possible models for the unwinding mechanism of RecQ-dH based on its pausing behavior, we simulated unwinding trajectories comprising a series of pauses and translocations (Figure 3A).
Based on previous studies of helicases (Manosas et al., 2010; Cheng et al., 2007; Neuman et al., 2005; Cheng et al., 2011; Lin et al., 2017; Myong et al., 2007), we considered two scenarios for RecQ-dH unwinding with an n-bp kinetic step size: either the enzyme unwind n base-pairs simultaneously then rapidly translocates along the unwound DNA (simultaneous melting
model), or it sequentially unwinds \( n \) base-pairs then releases the newly melted ssDNA (delayed release model) (Figure 3A). We exclusively simulated RecQ-dH unwinding and pausing kinetics rather than RecQ\(^{WT}\) due to the significantly more complex behavior of the RecQ\(^{WT}\) unwinding trajectories (Figure 1B).
In the simultaneous melting model, the pause duration, \( \tau \) is related to the sum of \( n \) base-pair energies at the position of the \( i \)th kinetic step,
\[
\tau_p(i) = \lambda_{\text{RecQ}} \exp \left( \sum_{s=1}^{n} G_{1\text{bp}}((i-1) + s) \right)
\]
(1)
Figure 3. Kinetic modeling of the kinetic step-size. (A) Example simulated 174 bp DNA hairpin unwinding traces (1 bp melt and step (orange line), 1 bp melt and 5 bp step (red line), and 2 bp melt and step (green line)) overlaid with an example RecQ-dH unwinding trace (black line). Unwinding traces were simulated using Equations 1 and 2. The overall unwinding events are composed of pauses of lifetime (\( \tau_p \)) due to melting of the base-pairs, followed by a rapid translocation step in time (\( \tau_t \)). \( \tau_p \) was calculated based on the sequence stability using nearest neighbor energy parameters (Patten et al., 1984; SantaLucia, 1998; Huguet et al., 2010). The total duration was adjusted to match the mean unwinding rate of RecQ-dH. (B) Pause times plotted as a function of the unwound hairpin for the three example models (with the same marker and line colors) in panel (A). Pause times and positions were obtained by analyzing simulated unwinding traces (100 traces for each condition) using \( T \)-test analysis and averaging pause times over a 5 bp window. The experimental pause lifetimes of RecQ-dH are shown in the black solid line. (C) Reduced \( \chi^2 (\chi_n^2) \) measure of the correspondence between measured and simulated pause durations as a function of pause position plotted as a function of the kinetic step size for three kinetic stepping models (see main text): 1 bp melt and step (orange filled circles), 1 bp melt and \( n \) bp step (red filled circles), and \( n \) bp melt and step (green filled circles). \( \chi_n^2 \) for 1 bp melt and step is significantly larger than the minima of the other two models. The \( \chi_n^2 \) is minimized for \( n = 2 \) bp for the \( n \)-bp melt and step model whereas \( \chi_n^2 \) is minimized for \( n = 5 \) bp for the 1 bp melt and \( n \) bp step model. (D) \( \text{Na}^+ \) dependent unwinding rates of RecQ-dH (black filled circles and dashed line) and predictions of the two kinetic models with the kinetic step-size, \( n \), that minimizes \( \chi_n^2 \) for each model: 1 bp melt and 5 bp step (red filled circles and dashed line), and 2 bp melt and step (green filled circles and dashed line). The error bars correspond to the standard error of the mean (SEM).
DOI: https://doi.org/10.7554/eLife.45909.009
The following source data and figure supplement are available for figure 3:
Source data 1. Source data for Figure 3 and Figure 3—figure supplement 1.
DOI: https://doi.org/10.7554/eLife.45909.011
Figure supplement 1. Comparison of simulated and experimental traces.
DOI: https://doi.org/10.7554/eLife.45909.010
Here $G_{1bp}$ is the free energy required for 1 base-pair melting at the $i^{th}$ position calculated using the nearest-neighbor energy parameters (Patten et al., 1984; SantaLucia, 1998; Huguet et al., 2010), $s$ is a step index ranging from 1 to $n$, and $A_{RecQ}$ is a pre-factor used to adjust the simulation to give the same average unwinding (46 nt/s) and translocation rate (~100 nt/s) as the RecQ-dH construct (Manosas et al., 2010). In the delayed release model, $\tau_p(i)$ is the sum of the pause times associated with melting each of $n$ base pairs at the $i^{th}$ kinetic step,
$$\tau_p(i) = \sum_{j=1}^{n} A_{RecQ} \exp(G_{1bp}(i-1)+s)$$
(2)
Stochastic simulations of both models were run with different step-sizes, $n$. For each value of $n$, the pre-factor $A_{RecQ}$ was adjusted to match the measured average rate of RecQ-dH, and the single-strand DNA translocation rate was 100 bp/s (Manosas et al., 2010; Bagchi et al., 2018). Simulated unwinding traces were generated for different kinetic step-sizes for the two different models (100 traces per each condition, example traces are shown in Figure 3—figure supplement 1). Simulated traces were analyzed with a T-test based step finding algorithm with the same parameters used for experimental data analysis. Pause durations were binned over 5 bp intervals for simulation and experimental traces and the mean pause duration for each bin was calculated (Figure 3B). Simulation results were compared with experimental data for RecQ-dH by calculating the reduced $\chi^2$ ($\chi^2$) between the simulated and experimental traces (Figure 3C). For the delayed release model, $\chi^2$ reached a minimum around 5 bp ($\chi^2 = 0.9 \times 10^{-6}$), lower than the minimum for simultaneous melting model that reached a minimum at 2 bp ($\chi^2 = 1.4 \times 10^{-6}$). This suggests that a delayed release scenario may describe the unwinding mechanism of core RecQ. To confirm this finding, we investigated how the RecQ-dH unwinding rate was affected by Na$^+$ concentration and compared the results with the two unwinding models. As DNA base-pair melting energy increases with Na$^+$ concentration (SantaLucia, 1998; Huguet et al., 2010), the average unwinding rate predicted by the simultaneous melting model should decrease more rapidly than that predicted by the delayed release model (Figure 3D). We varied the Na$^+$ concentration from 25 to 500 mM while maintaining Mg$^{2+}$ at 5 mM under otherwise identical buffer conditions. The unwinding rate of RecQ-dH decreased with increasing Na$^+$ concentration. The relative decrease in unwinding rate was much better described by the delayed release model with a 5 bp kinetic step, than the simultaneous 2 bp DNA melting model (Figure 3D). The small deviation between the delayed release model and the measured Na$^+$ concentration dependence of the unwinding rate suggests that although the duplex unwinding remains the rate-limiting step, the Na$^+$ concentration effects other aspects of unwinding such as protein-DNA interactions, which are beyond the scope of the simple model. Thus, to further test and confirm the delayed release unwinding model, we performed two additional experiments as explained below.
### Unwinding kinetics of forked DNA substrates in ensemble rapid kinetic experiments support the delayed release model
The significant DNA sequence dependence of the RecQ-catalyzed DNA unwinding rate and pausing characteristics detected in MT single-molecule experiments should be reflected in ensemble unwinding kinetic measurements, which are suitable for the determination of the kinetic step size and the macroscopic dsDNA unwinding rate (Lucius et al., 2003). In these experiments unwinding kinetics are monitored via the appearance of fully unwound reaction products. Thus, ensemble unwinding experiments complement MT experiments, in which individual unwinding steps are monitored. Importantly, these techniques together should allow determination of the microscopic unwinding mechanism of RecQ helicase constructs based on the proposed base-pair energy dependent unwinding models.
To test this idea, we performed single-turnover unwinding kinetic experiments in which we rapidly mixed complexes of RecQWT or RecQ-dH with forked DNA substrates of varying GC content with ATP and excess unlabeled ssDNA traps in a quenched-flow instrument and monitored the time course of fluorescently-labeled ssDNA generation via gel electrophoresis of reaction products (Figure 4A). Forked DNA substrates used in the experiments comprised two 21-nt ssDNA arms and a 33 bp dsDNA segment containing 12 (gc36), 16 (gc48) or 26 GC (gc79) bps (sequences described in Supplementary file 1Table S1). Unwinding traces comprised a short (~0.1 s) initial lag, followed by a biphasic appearance of the labeled ssDNA reaction product (Figure 4B). The rapid rise
Figure 4. Single-turnover ensemble kinetic experiments. (A) Electrophoretogram of a single-turnover unwinding experiment. Preincubation of fluorescein-labeled forked duplex DNA (30 nM, gc36) with RecQ-dH (100 nM) was followed by rapid mixing with ATP (3 mM) plus excess ssDNA trap strand (3 μM) (final post-mixing concentrations). Reactions were stopped by the addition of EDTA (40 mM) and loading dye at different time points (0–150 s, cf. panel B) using a quenched-flow instrument or by manual mixing. Amounts of DNA species (forked duplex and ssDNA, depicted by cartoons) labeled with fluorescein (asterisk) were detected by a fluorescence imager. “–” denotes a 150 s control reaction in which ATP was absent. (B) Single-turnover experiments were repeated for RecQ-dH and RecQ WT. The fraction of DNA unwound (%) was recorded as a function of time. (C) Rate constants for the Reviewed panel B reactions. The rate constant $k_{\text{bind}}$ is the rate constant for the binding of the ssDNA trap strand. $k_{f}$ is the rate constant for the formation of the forked duplex, $k_{(i)}$ is the rate constant for the formation of the single-stranded DNA. (D) The Table presented the kinetic step size (bp) for the n-step melting experiments. The Table presented the kinetic step size (bp) for the n-step melting experiments. (E) The Table presented the kinetic step size (bp) for the n-step melting experiments.
turnover unwinding kinetics of forked DNA substrates with GC contents of 36% (light gray), 48% (gray) and 79% (black) of RecQ-dH and RecQ\textsuperscript{WT}. Error bars represent SEM calculated from three experiments. Solid lines show fits based on the n-step model at n = 5 for both helicase constructs (see scheme on panel C), simultaneous model (Equation 1 and panel C) at n = 4 for RecQ-dH and n \leq 5 for RecQ\textsuperscript{WT}, and delayed release model (Equation. and panel C) at n = 4 for RecQ-dH and n = 5 for RecQ\textsuperscript{WT}. (C) Common scheme for the modified n-step and derived simultaneous melting and delayed release models. In the models, unwinding starts from the ssDNA-dsDNA junction. Of all DNA-RecQ complexes (D.R), only a fraction (f \textsubscript{D-R}) rebind and start a new unwinding run (D.R). (D-E) Determined \chi\textsuperscript{2} values from fitting the (D) n-step model for RecQ-dH (filled circles) and RecQ\textsuperscript{WT} (open circles) or (E) fitting the simultaneous melting (green) and delayed release models (red) for the indicated helicase construct. Other determined parameters are listed in Supplementary file 1 Table S2.
DOI: https://doi.org/10.7554/eLife.45909.012
The following source data is available for figure 4:
Source data 1. Source data for Figure 4.
DOI: https://doi.org/10.7554/eLife.45909.013
originated from single unwinding runs of initially DNA-bound helicase molecules. The slow rise phase originates from premature dissociation, followed by slow rebinding, of the enzyme to the DNA substrate (hindered but not totally inhibited by the ssDNA trap strand) that eventually led to full unwinding of the DNA fork (Harami et al., 2017). To obtain parameters of unwinding, data were analyzed with a modified version of a previously described n-step kinetic model (Lucius et al., 2003). In its simplest form the model assumes that DNA unwinding occurs as a result of n consecutive rate limiting steps that have a uniform rate constant. This model is generally suitable for the determination of the macroscopic dsDNA unwinding rate, the kinetic step size and the number of intermediates in the unwinding reaction (Lucius et al., 2003).
Using a modified version of the n-step model (Figure 4C), global fitting of the unwinding kinetics of the gc36, gc48 and gc79 substrates using an integer series of n ranging from 1 to 7 revealed smallest \chi\textsuperscript{2} values for an apparent kinetic step size of 5 bp for both RecQ and RecQ\textsuperscript{WT} (Figure 4B–D) with all DNA substrates, similar to that suggested by our MT results (Figure 3A–B) and by previous findings (Lin et al., 2017; Harami et al., 2015). However, the n-step model does not consider the sequence dependence of the rates of elementary unwinding steps, precluding the distinction between different microscopic mechanisms producing the same kinetic step size. Therefore, we used the same physical framework as described for the MT experiments (Equations 1 and 2) and performed global kinetic fitting to all transients of a given helicase construct (RecQ\textsuperscript{WT} or RecQ-dH) unwinding the different forked DNA substrates, based on the DNA sequence-dependent simultaneous melting and delayed release unwinding models (Figure 4B–C). For both models, fitting was done using an integer series of n ranging from 1 to 7. In agreement with the results of the MT analysis (Figure 3C), the smallest \chi\textsuperscript{2} value was obtained for the delayed release model with a kinetic step size of 4 bp for RecQ-dH and 5 bp for RecQ\textsuperscript{WT} (Figure 4B and C, other parameters are listed in Supplementary file 1 Table S2).
**Direct measurement of 5 bp kinetic step size and time-dependent release of ssDNA**
If RecQ-dH takes a certain kinetic step size, it could in principle be directly observed in the single-molecule unwinding traces. However, the enzyme unwinds DNA too rapidly at high ATP concentrations for steps to be routinely and accurately detected, given the spatial resolution limits of the measurement. Under our experimental conditions, the average baseline noise was ~14 nm at 200 Hz data collection rate. Thus, in order to observe, for example, a 4 bp step (i.e. a 3 nm change in DNA extension), the average pause duration should be >300 ms or the unwinding rate should be less than 13 bp/s (~3 fold slower than 42 bp/s). We tried three different conditions to decrease the unwinding rate of RecQ: lowering the ATP concentration (Figure 5—figure supplement 1) and including non-hydrolysable ATP analogues, ATP\textsubscript{γ}S or AMP-PNP, in the assay (Figure 5—figure
supplement 2). We found that decreasing the ATP concentration (sufficiently lowering the unwinding rate) resulted in frequent and extensive enzyme backsliding (observable as rapid partial rzippering of the hairpin during an unwinding event), which complicates kinetic step size measurements (Figure 5—figure supplement 1). AMP-PNP showed extremely slow dissociation kinetics from RecQ-dH that were inappropriate for unwinding assays (Figure 5—figure supplement 2). On the other hand, ATPγS, showed a comparable binding affinity to ATP with a significantly shorter binding time (~1 s) than AMP-PNP (Figure 5—figure supplement 2). In addition, ATPγS binding transiently locks RecQ in the strong DNA-binding ATP bound state without backsliding, leading to long duration pauses that effectively increased the spatial resolution by permitting longer averaging times (Figure 5A). We measured the unwinding activity of RecQ-dH at different fractions of ATPγS (0.05–0.5 mM) while keeping the total combined concentration of ATP and ATPγS constant at 1 mM. The unwinding rate decreased with increasing ATPγS fraction (Figure 5A). We reason that when the concentration of ATPγS is such that it is bound at least once per kinetic step, then the predominant
Figure 5. 5 bp kinetic step-size and tight mechano-chemical coupling. (A) Example traces of hairpin unwinding by RecQ-dH with increasing concentration of ATPγS while maintaining the combined concentration of ATP and ATPγS at 1 mM. (B) Step-size distributions were obtained by analyzing unwinding traces collected at each ATPγS concentration with either T-test or Kerssemakers (KM) step analysis algorithms (see main text). (C) The mean step-size obtained by fitting the distributions in panel (B) with Gaussians plotted as a function of ATPγS concentration. The average step sizes from both the T-test and Kerssemakers (KM) analysis converge to 5 bp with increasing ATPγS concentration. Inset: An example trace with 5 bp steps (red line) and the T-test fit (blue line). (D) Global fitting of the mean pause duration (green solid triangles) and average unwinding rate (red solid circles) as a function of ATPγS concentration, using Equations 4 and 5, reveal a tight mechano-chemical coupling ratio of \( C = 1.0 \pm 0.2 \) bp/ATP.
DOI: https://doi.org/10.7554/eLife.45909.014
The following source data and figure supplements are available for figure 5:
Source data 1. Source data for Figure 5 and Figure 5—figure supplements 1–4.
DOI: https://doi.org/10.7554/eLife.45909.019
Figure supplement 1. ATP dependence of unwinding, stepping, and backsliding kinetics of RecQ-dH.
DOI: https://doi.org/10.7554/eLife.45909.015
Figure supplement 2. Comparison of RecQ-dH unwinding the 584 bp DNA hairpin in the presence of AMPPNP or ATPγS and ATP.
DOI: https://doi.org/10.7554/eLife.45909.016
Figure supplement 3. Example trace of RecQ-dH at 500 µM ATPγS showing a 2.5 bp step.
DOI: https://doi.org/10.7554/eLife.45909.017
Figure supplement 4. RecQ-dH inefficiently hydrolyzes ATPγS.
DOI: https://doi.org/10.7554/eLife.45909.018
physical step-size measured in the hairpin unwinding trajectories will correspond to the kinetic step-size. Step-sizes were estimated with two different step finding algorithms: a step finding program originally developed by Kerssemakers and coworkers (Kerssemakers et al., 2006) and the T-test based step finding analysis (Seol et al., 2016). To determine the average kinetic step-size for each condition, the estimated step-sizes were histogrammed and fit with Gaussian distributions (Figure 5B and C). We found that the estimated step size of RecQ-dH from both step-finding algorithms were comparable, converging from ~8 bp at a low ATPγS fraction to 5 bp at higher ATPγS fractions, suggesting that the average kinetic step size of RecQ-dH is 5 bp (T-test: 5.3 ± 0.1 (center); 3.0 ± 0.6 (Standard Deviation); Kerssemakers: 5.2 ± 0.1 (center); 2.1 ± 0.1 (Standard Deviation), errors correspond to the standard deviations from Gaussian fitting). The broad step-size distribution could reflect the stochastic nature of ssDNA release by RecQ. Also, it is likely that the two ssDNA strands are released asynchronously by RecQ. In line with this, we occasionally observed a 2.5 bp kinetic step at 500 μM ATPγS, and the step-size distribution at lower ATPγS fractions included peaks at 7.5, 10, and 12.5 bp, consistent with a fundamental step-size of 2.5 bp corresponding to the release of one ssDNA strand of 5 nt (Figure 5—figure supplement 3).
Tight mechano-chemical coupling of ATP-dependent unwinding by RecQ
The prolonged pause state due to ATPγS binding instead of ATP at the cleft between two RecA domains of RecQ enabled us to probe the mechano-chemical coupling, C, of RecQ helicase that is a measure of the number of ATP hydrolyzed per kinetic step. For C = m/n (m ATP hydrolysis per n kinetic step size), the average number of bound ATPγS, l, can be estimated based on the binomial probability distribution.
\[ l = \sum_{i=0}^{m} \begin{pmatrix} m \\ i \end{pmatrix} \frac{m!}{(m-i)!} P^i (1-P)^{m-i} \]
(2)
\[ P = \frac{k_{ATP}}{k_{ATP} + k_{ATPγS}} \]
(3)
P is the probability of ATPγS binding per each cycle, \( k_{ATP} \) is the ATP on-rate, \( k_{ATPγS} \) is the ATPγS on-rate, and [ATPγS] and [ATP] are the concentrations of ATPγS and ATP, respectively. The hydrolysis rate of ATPγS by RecQ in the presence of excess dTAS was measured by monitoring thio phosphate production (Saran et al., 2006) and estimated to be < 0.2/s (Appendix 1 and Figure 5—figure supplement 1). This is significantly slower than the measured pause exit rate (2.8 ± 0.1/s) at 500 μM ATPγS, suggesting that \( k_{off} \) is much faster than the rate of ATPγS hydrolysis. Thus, we could simplify the mean pause duration per kinetic step, \( \tau \) and the mean unwinding rate, \( v \) as,
\[ \tau = \frac{1}{k_{off}} + \frac{1}{k_{step}} \]
(4)
\[ v = \frac{n}{\tau} \]
(5)
where \( k_{step} \) is the mean kinetic stepping rate without ATPγS. We obtained the average pause durations for different fractions of ATPγS from 5 to 50 % and globally fitted the pause durations and the average unwinding rates as a function of ATPγS concentration with Eq. 4 and 5 respectively (Figure 5D). From this global fitting, we found that \( C = 1.0 ± 0.2 \) bp/ATP, \( k_{ATP}/k_{ATPγS} = 1.2 ± 0.2 \), and \( 1/k_{off} = 0.4 ± 0.1 \) s suggesting a tight mechano-chemical coupling in agreement with previous ensemble measurements (Harami et al., 2015; Sarlós et al., 2012). We note that rebinding of ATPγS was not taken into account for simplicity in Equations 2 and 3, which is reasonable as ATPγS concentration is lower than ATP except for 50% ATPγS, and because the relative on rate \( k_{on}[ATPγS]/k_{on}[ATP] \) vs \( k_{off}[ATPγS]/k_{off}[ATP] \) of ATPγS is lower than that of ATP. To ensure that this simplification is reasonable, we simulated how many ATPγS molecules are bound instead of ATP per base pair based on the fitting parameter, \( k_{ATP}/k_{ATPγS} = 1.2 \) at 50% ATPγS. We found that the average is less than one ATPγS per base pair at this condition indicating that repetitive ATPγS binding at the same site is rare.
Multi-step HRDC dependent pausing kinetics results in a non-linear amplification of intrinsic sequence-dependent pausing
The sequence-dependent unwinding mechanism of RecQ consisting of a 5 bp kinetic step results in transient pauses that are further stabilized by the HRDC, which results in the long-lived sequence-dependent pausing of RecQ WT (Figure 2). In addition to the sequence dependence, HRDC-dependent pausing exhibits two interesting features: occasional repetitive rezipping and unwinding (shuttling) around the pause position and significantly prolonged pausing durations for certain pausing positions (Figure 2 and Figure 2—figure supplement 1). It appears that HRDC-binding triggers this shuttling behavior in which 5–10 bp are repetitively unwound and rezipped at the relatively long-lived (>0.14 s) intrinsic pause positions. Shuttling activity repeats until RecQ passes the sequence-dependent roadblock. This complex shuttling behavior was significantly enhanced at those regions where long pauses of RecQ-dH are clustered, such as at 55, 90, and 120 bps (Figure 2—figure supplement 1) resulting in the apparent high dwell probabilities at these sites (Figure 2A). Since these positions also exhibit the highest base-pair stabilities (Figure 2A), the average dwell time of RecQ WT is strongly correlated with the base-pair stability. Indeed, the average dwell times for RecQ WT are highly non-linearly correlated with the base-pair stability (Figure 6A). In contrast, the dwell-times for RecQ-dH scale linearly and much less dramatically with the base-pair stability (Figure 6A), indicating that the HRDC-stabilized pausing can be described as a non-linear amplifier of the intrinsic sequence-dependent unwinding kinetics. We found that a simple kinetic competition model in which HRDC binding is in kinetic competition with the forward motion of the helicase (Appendix 1) cannot reproduce the dramatic changes in pause probability observed for RecQ WT hairpin unwinding (Figure 6—figure supplement 1). In line with this, the pausing duration distribution for RecQ WT is better described by a double rather than single exponential distribution (Figure 6—figure supplement 1B). Both pause lifetimes (1.8 ± 0.3 s and 0.4 ± 0.1 s) are longer than the average pause duration (0.14 ± 0.03 s, see SI) for the core RecQ (RecQ-dH), indicating that there are multiple HRDC-dependent pause states.
To test the proposal that base-pair stability-dependent RecQ WT pausing underlies a potential mechanism of homology sensing, we investigated the effect of introducing single mismatches at high probability pause sites. We tested hairpin substrates containing 1, 2, or three single mismatches: (i) a mismatch introduced at 90 bp, (ii) mismatches introduced at 90 bp and 104 bp and (iii) mismatches introduced at 90 bp, 104 bp, and 124 bp. All mismatches were generated by changing G to T on the displaced strand (detailed sequence information is in Appendix 1). We found that pausing of RecQ WT around the 90 bp unwound hairpin position was significantly reduced compared to intact 174 bp hairpin when a mismatch was present at 90 bp and additional mismatch at 124 bp further suppressed pausing around 120 bp (Figure 6B and Figure 6—figure supplement 2). The effect of mismatches on pausing of RecQ WT can be clearly demonstrated by comparing the dwell-time histograms of three DNA substrates (Figure 6B). The prominent peaks in the dwell time histogram of the intact hairpin DNA were diminished one by one with the introduction of mismatches at the corresponding positions confirming the correlation between pausing and homology (Figure 6B and Figure 6—figure supplement 2).
Discussion
RecQ helicases are well-established as critical enzymes that contribute to genome stability through their multiple roles in DNA repair and genome integrity. Whereas the mechanistic basis of many of the specialized activities of RecQ helicases have been established, the mechanism through which RecQ helicases can distinguish legitimate from illegitimate homologous recombination intermediates has not been established. We previously proposed a mechanistic model in which the HRDC domain of RecQ orients the enzyme to preferentially disrupt the strand invasion (or D-loop structure) corresponding to the earliest HR intermediate. In this model, discrimination between legitimate and illegitimate HR was achieved by modulating the degree of HRDC-induced pausing through an unknown mechanism. Here we demonstrate that HRCD-induced pausing is sequence-dependent and establish the mechanistic basis for this behavior. The pronounced pausing of WT RecQ at GC-rich sequences results from two amplification steps that convert the ~k_B T energy differences between GC and AT base pairs to a robust readout of sequence stability (Figure 6A). The first amplification step arises from the basic mechanochemistry of DNA unwinding that involves a ~ 5 bp kinetic step in which ~ 5
bp of DNA are unwound in 5 rounds of ATP hydrolysis followed by release of the two DNA strands. By coupling the forward motion of the helicase to the unwinding of 5 bp rather than a single bp, individual dwell times can vary ~5 fold more in relation to the underlying sequence than they would for single-bp steps. The second amplification step involves the non-linear amplification of the intrinsic difference in unwinding rate in proportion to GC content through the binding and stabilization of short sequence-dependent pauses by the HRDC. Together these two rounds of linear and non-linear amplification result in strong sequence dependent pausing of WT RecQ on hairpin substrates that results in a greater than 10-fold difference in average unwinding rate of legitimate versus illegitimate paired sequences. A recent single molecule study of *E. coli* RecQ<sup>WT</sup> indicated a switching behavior between a fast unwinding mode, similar to that of RecQ-dH, and a slower unwinding mode that is...
similar but not identical to what we observed for RecQ\textsuperscript{WT} unwinding in the present study (Bagchi et al., 2018). However, we did not observe switching of RecQ\textsuperscript{WT} unwinding (frequent and prolonged pausing) to RecQ-dH-like unwinding (transient pausing). It is possible that the switching to the latter mode is caused by sequestration of the HRDC from ssDNA, which may be dependent on physicochemical conditions such as higher temperature and/or low salt concentrations.
Our study reveals how the unwinding mechanism of the core RecQ helicase, for example RecQ-dH, directly impacts HRDC-dependent pausing and the subsequent control of biological functions mediated by HRDC-dependent helicase activities. We were able to elucidate the coupling between ATP hydrolysis and the unusual kinetics of DNA unwinding by varying the ATP and ATP$_\text{S}$ concentrations. We find that RecQ hydrolyzes 5 ATP molecules during a 5 bp kinetic unwinding step that concludes with asynchronous release of two five nt ssDNA segments on average (Figures 3–5). Furthermore, ATP binding likely stimulates RecQ binding to and melting of DNA duplex prior to hydrolysis (Figure 5—figure supplement 1).
**Sequence-dependent unwinding mechanism**
The fundamental activity of helicases is the unwinding of duplex nucleic acids. In general, the unwinding mechanism has been classified as either passive or active depending on the degree to which the enzyme ‘actively’ destabilizes the duplex rather than ‘passively’ waiting for a thermal fluctuation to expose ssDNA (Betterton and Jülicher, 2005; Lohman et al., 2008). For a purely passive helicase, the enzyme does not provide external work to destabilize duplex DNA and translocates only when ssDNA is exposed by thermal fluctuations. On the other hand, an active helicase is actively involved in disrupting the DNA duplex, and in principle, is less sensitive to base-pair energy or sequence. In previous single molecule experiments, E. coli RecQ helicase was identified as an active helicase based on the minimal force and DNA sequence dependence of duplex unwinding (Manosas et al., 2010). In that study, following theoretical work by Betterton et al (Betterton and Jülicher, 2005), RecQ unwinding was modeled as one base-pair melting followed by 1–2 bases translocation. However, we found that the pause durations were generally longer than would be expected for melting of 1 base pair when we compared our results with simulations. We considered two different scenarios: RecQ either destabilizes multiple base-pairs (≥2 bp) during each kinetic step similar to NS3 helicase (Cheng et al., 2007) or delays releasing of multiple unwound base-pairs similar to speculative models suggested in previous studies (Cheng et al., 2011; Lin et al., 2017; Myong et al., 2007; Ma et al., 2018). However, the minimal dependence of the unwinding rate on Na$^+$ concentration in addition to the sequence-dependent pauses cannot be explained by multi-base-pair melting. Rather, we found that an alternative scenario in which RecQ delays the release of nascent single-strand DNA (delayed release) was a better fit to the pause duration and Na$^+$-dependent unwinding rate data, though the associated kinetic step size (number of bp unwound prior to release) was not uniquely constrained by the pause duration or Na$^+$-dependent unwinding rate measurements (Figure 3). This finding is consistent with previous studies revealing ‘asynchronous’ release of nascent ssDNA (Lin et al., 2017; Ma et al., 2018). Nonetheless, the mechanism of delayed release of newly melted nucleotides remains unclear. Previous results suggest that a putative electrostatic interaction between newly melted ssDNA and RecQ sequesters several nucleotides of ssDNA. We consider a similar possibility in which RecQ releases the nascent ssDNA only when the accumulated torsion or tension on bound ssDNA is high enough to disrupt the interaction (Myong et al., 2007).
**RecQ takes 5-base kinetic steps and unwinds one base-pair per ATP hydrolysis**
We further refined the delayed release model by directly measuring a 5 bp kinetic step size for DNA unwinding using ATP$_\text{S}$, which sufficiently slows down the unwinding rate without inducing the frequent back-sliding observed at reduced ATP concentrations (Figure 5 and Figure 5—figure supplement 1). Recent single molecule fluorescent studies showed 2–4 bp kinetic step (Lin et al., 2017; Ma et al., 2018). This smaller and more random nature of the kinetic step size is likely due to the low ATP concentration (2–5 μM), at which ATP binding likely becomes the dominant rate-limiting step slower than or on the same order as the intrinsic off-rate of the nascent DNA. Consistent with
this model, the study found a correlation between the ATP concentration and the measured kinetic step size.
The mechano-chemical coupling and kinetic analysis of ssDNA translocation of RecQ have been studied in detail (Sarloš et al., 2012; Rad and Kowalczykowski, 2012). Our unwinding kinetic step is consistent with a recent a study in which a five nucleotide kinetic step for RecQ translocation was estimated (Rad and Kowalczykowski, 2012). Other helicases display multi base-pair kinetic unwinding steps under sufficient ATP concentrations (above \( K_M \sim 20 \mu M \)) (Lohman et al., 2008). The mechano-chemical coupling is a measure of how many chemical cycles an enzyme completes to take one mechanical step. In the case of RecQ or other helicases, it corresponds to how many ATP molecules are consumed per one base translocation (or base-pair unwound for unwinding). For translocation, RecQ shows a tight coupling close to one nucleotide step per ATP hydrolysis (Sarloš et al., 2012). Our study reveals that the mechano-chemical coupling for unwinding maintains one base-pair melting per ATP hydrolysis (Figure 4C), which is also supported by the results of a recent single-molecule fluorescence study of RecQ unwinding (Lin et al., 2017). The proposed kinetic model based on our ATP dependent kinetic analysis (Appendix 1; Figure 5—figure supplement 1) suggests that DNA melting precedes ATP hydrolysis. In this model, ATP binding stabilizes the DNA-RecQ interaction and facilitates DNA melting presumably coupled to an ATP binding-dependent conformational change of RecQ such as rotation of the helicase domains relative to one another, which explains more frequent backsliding under lower ATP concentration (Bernstein et al., 2003; Manthei et al., 2015; Pike et al., 2009). Recent structural results suggest that RecQ binding may melt two base-pairs of DNA before ATP binding (Manthei et al., 2015). This may occur at the initial binding of RecQ (or rebinding) as the initiation of unwinding, but not the unwinding rate, is highly dependent on Na\(^+\) concentration.
Whereas we establish that pausing arises from the stability of DNA duplex, recent work by Voter et al. suggests an alternative mechanism for sequence-dependent pausing. In their work, they identify a ‘Guanine binding pocket’ located in the helicase domain that specifically interacts with guanine bases to destabilize G-quadruplex structures. It is possible that these interactions could also slow down the unwinding rate at clusters of guanine bases in the translocation strand by inducing short pauses (Voter et al., 2018). However, the translocation sequence at the strong pause locations of our DNA hairpin is mixture of G and C bases, suggesting that the pauses we observed originate from the duplex stability. Nevertheless, we cannot entirely rule out the possibility that these specific guanine interactions contribute slightly to the pausing of RecQ core over and above the dominant effect of DNA duplex stability.
**HRDC amplifies weak sequence-dependent pauses during unwinding of RecQ core in a DNA substrate geometry-dependent manner**
One of the essential aspects of RecQ is that it processes diverse, non-canonical, DNA substrates in which the HRDC plays an important role in modulating substrate-specific unwinding of RecQ. It has been shown that the HRDC regulates the binding orientation of RecQ core to promote disruption of D-loop structures, early homologous recombination intermediates (Harami et al., 2017). However, it was not clear how it can regulate unwinding of RecQ to selectively disrupt illegitimate or non-homologous invading DNA strands since the HRDC presumably cannot directly sense DNA sequence homology (Harami et al., 2017). Our present study reveals that the HRDC-ssDNA interactions are strongly coupled to DNA sequence-dependent pausing of the RecQ helicase core: ssDNA binding by the HRDC is not random but occurs at DNA sequences where the helicase core pauses due to the high duplex stability (Figure 2). On the other hand, either a low homology (base-pair mismatches) or low duplex stability (low GC regions) strongly reduces RecQ pausing (Figures 2 and 6B and Figure 6—figure supplement 2). Importantly, this feature can support discrimination between legitimate and illegitimate recombination events by RecQ helicases, in accordance with the increased illegitimate recombination frequencies detected in vivo upon compromising RecQ HRDC function (Harami et al., 2017; Wang et al., 2016). Recombination events proceed through the formation of a displacement loop (D-loop) flanked by genomic DNA, which, due to the limited mobility of these large DNA domains, mimics the hairpin geometry of the magnetic tweezers experiments in which the displaced DNA strand is constrained (Figure 6C and D). Previously we showed that the HRDC both targets RecQ to D-loop intermediates and orients the enzyme in a configuration favoring D-loop disruption (Harami et al., 2017). The results obtained here provide a mechanistic basis for
the subsequent discrimination between legitimate and illegitimate recombination based on the length and stability of the D-loop structure. RecQ-catalyzed unwinding of long and stable D-loops will be frequently interrupted by HRDC-stabilized pauses that drastically decrease the average unwinding rate. This slow average unwinding rate potentially permits the initiation of down-stream recombination processes associated with DNA synthesis resulting in extension and further stabilization of the D-loop. Conversely, RecQ unwinding of short and/or unstable D-loops will proceed rapidly (60–80 bp/s) resulting in the efficient disruption of the D-loop before it can be further extended. Our study reveals that the strategic location of the HRDC relative to the core domain, combined with sequence-dependent DNA unwinding, enable RecQ helicase to control pausing and shuttling in a substrate-dependent manner and expand its biological activity beyond simple duplex DNA unwinding. Whereas this study focused exclusively on *E. coli* RecQ, the homology sensing mechanism we propose is potentially applicable to the suppression of illegitimate, or so called ‘homeologous recombination’ by BLM (Wang et al., 2016).
Another biological role of HRDC domain-mediated pausing and shuttling could be linked to the role of RecQ helicases in G-quadruplex secondary DNA structure processing. G-quadruplex structures were shown to act as replication road blocks and these regions were shown to be recombinational hot spots (van Wietmarschen et al., 2018; Rhodes and Lipps, 2015). RecQ helicases can efficiently unwind G-quadruplex structures, possibly to aid DNA replication, suppress genome instability and to influence transcription of various genes (Voter et al., 2018; Mendoza et al., 2016). Prolonged shuttling at G-quadruplex sites could ensure that these secondary structures remain unfolded until further steps of replication or DNA repair can proceed. In line with this idea, the HRDC domain of human BLM helicase was shown to be essential for efficient, repetitive unwinding of G-quadruplexes (Chatterjee et al., 2014).
In this study, we focused on elucidating the sequence-dependent unwinding and pausing mechanism of RecQ helicase in vitro with purified proteins. Whereas our results indicate a possible mechanism for homology sensing by RecQ helicases, the translocation and pausing kinetics on which the model is based could be modulated in vivo due to the interactions with other DNA binding and processing enzymes. For example, single-strand binding protein (SSB) would likely compete with the HRDC for ssDNA binding. However, we recently demonstrated that SSB is displaced by RecQ despite the much higher apparent binding affinity of SSB for ssDNA (Mills et al., 2017). Furthermore, the high local concentration of the HRDC, which is tethered to the RecQ core by a flexible linker, likely results in the HRDC out-competing other ssDNA binding proteins for the newly melted ssDNA. Nonetheless, as is often the case, RecQ helicases play diverse roles in DNA processing through the interaction with other proteins, thus, future experiments in the presence of other proteins that interact with RecQ in vivo including, SSB, RecJ, RecA, and Topoisomerase III are warranted to test our homology model in a context that more closely approximates physiological conditions.
**Materials and methods**
| Key resources table |
|---------------------|
| **Reagent type (species) or resource** | **Designation** | **Source/reference** | **Identifiers** | **Additional information** |
| Strain, strain background (*Escherichia coli*) | ER2566 | New England Biolabs | NCB Cat. #: E4130 | |
| Recombinant DNA | Modified pTXB vector | PMID: 26067769 | | Transformation and expression of RecQ constructs |
| Recombinant DNA | pKZ1 | PMID: 28069956 | | Template for hairpin DNA substrate |
| Antibody | Anti-digoxigenin (Sheep Polyclonal) | Roche | | 113308901, RRID:AB_514496 |
| Commercial assay or kit | IMPACT purification system | New England Biolabs | | NCB Cat. #: E6901S |
*Continued on next page*
## DNA substrate preparation
### DNA hairpin substrates
Generation of the 174 bp DNA hairpin was previously described in detail ([Harami et al., 2017](https://doi.org/10.7554/eLife.45909)). The 584 bp DNA hairpin was prepared by ligation of a 500 bp DNA hairpin with ~1.0 kb DNA handle. The 1.0 kb DNA handle was generated first by PCR of pKZ1, which contains two Bbvcl sites spaced by 37 bp, between 4550 and 258 using one primer (258 position) containing a BsaI digestion site and the other primer (4550 position) labeled with 5´-digoxigenin. PCR products were digested by BsaI and gapped with Nt. Bbvcl following the same method to generate the handle of the 174 bp DNA hairpin. 3´ biotin-labeled poly dT with a 33 bp region complementary to the gapped region of the 1 kb DNA handle was ligated to the 37-nt gapped region of 1 kb DNA handle. The 500 bp DNA hairpin was generated by PCR of Lambda DNA (NEB) between 23104 and 23608 and both ends were digested by BsaI. The final product was made by ligation of the 1 kb DNA handle with 3´ biotin-labeled poly dT, 500 bp DNA hairpin, and 12 bp DNA with a loop of 4 dT nucleotides to form the hairpin from the PCR product.
174 bp DNA hairpin with 1, 2, or 3 specific mismatch mutations (90 bp; 90 and 104 bps; 90, 104, 124 bps on the displacing strand) were generated by first cutting the PCR product for the 174 bp hairpin with Nhel, yielding a 5´CTAG overhang. In order to prevent 100 bp fragments from Nhel digestion to religate back to the DNA handle, the digested DNA band was extracted from an agarose gel. The two complementary oligos (88 nucleotides; Appendix 1) were annealed by incubation at 94°C for 5 min and then subsequent cooling to 4°C at a rate of −1°C/s. The final product was made by ligation of the Nhel-digested PCR product with 3´ biotin-labeled poly dT, 84 bp annealed DNA with differential four nt-overhangs, and 12 bp DNA with a loop of 4 dT nucleotides to form the hairpin.
### Enzyme preparation
The production of RecQ<sup>WT</sup> and RecQ-dH were previously described in detail ([Seol et al., 2016](https://doi.org/10.7554/eLife.45909)).
Ensemble kinetic measurements
Forked DNA substrates (described in Supplementary file 1 Table S1) were generated and single-turnover unwinding experiments were performed as in ref (Harami et al., 2015). Global fitting kinetic analysis was performed using KinTek Global Kinetic Explorer 4.0.
Single-molecule measurements and data analysis
The magnetic tweezers and the experimental set-up were previously described (Seol and Neuman, 2011). A mixture of DNA hairpin (3 pmol) and anti-digoxigenin (0.5 μg) was incubated in phosphate buffered solution (PBS, pH 7.5) for 20 min and introduced into the sample chamber, which was incubated overnight at 4°C. The sample chamber was washed with 1 ml of wash buffer (WB, 1X PBS, 0.02 % v/v Tween-20, and 0.3 % w/v BSA) to remove unbound DNA molecules and 40 μl of magnetic beads (MyOne, Invitrogen) were introduced to form DNA hairpin tethers. Correct DNA hairpins were identified by the sharp DNA extension change upon DNA hairpin unfolding under high force (~19 pN). Upon finding a proper DNA substrate, the chamber was washed with 200 μl of RecQ buffer (30 mM Tris pH 8, 50 mM NaCl, 5 mM MgCl₂, 0.3 % w/v BSA, 0.04 % v/v Tween-20, 1 mM DTT, and 1 mM ATP). After washing, RecQ was added at a concentration of 20–100 pM in 200 μl RecQ buffer. DNA unwinding measurements were done by tracking a DNA tethered magnetic bead in real-time with custom written routines in Labview. During the measurement, a 1 μm polystyrene stuck bead was tracked to correct sample cell drift by adjusting the sample cell position using 3-D piezo stage (Physik Instrumente) to compensate for the drift. The unwinding traces were analyzed with a custom-written T-test based algorithm in Igor Pro 6 (Wavemetrics) and the Kerssemakers step finding program in MatLab (Seol et al., 2016; Kerssemakers et al., 2006; Carter and Cross, 2005).
Acknowledgements
We thank Yasuharu Takagi, Sarah Heissler and Jim Sellers for assistance with ATP hydrolysis assays and Dr. Jonathan Silver for comments. This research was supported by the Human Frontiers Science Program (RGY0072/2010) and the Intramural Research Program of the National Heart, Lung, and Blood Institute, National Institutes of Health (to KCN); the ‘Momentum’ Program of the Hungarian Academy of Sciences (LP2011-006/2011), ELTE KMOP-4.2.1/B-10-2011-0002, NKFIH K-116072, NKFIH ERC_HU 117680, and NKFIH K-123989 grants (to MK); and a Premium Postdoctoral Fellowship of the Hungarian Academy of Sciences (to GMH).
Additional information
| Funding | Grant reference number | Author |
|-----------------------------|------------------------|-------------------------|
| Human Frontier Science Program | RGY0072/2010 | Yeonee Seol |
| | | Gábor M Harami |
| | | Mihály Kovács |
| | | Keir C Neuman |
| National Institutes of Health | HL001056-12 | Yeonee Seol |
| | | Keir C Neuman |
| Hungarian Academy of Sciences | LP2011-006/2011 | Gábor M Harami |
| | | Mihály Kovács |
| Eötvös Lorand University | ELTE KMOP-4.2.1/B-10-2011-0002 | Mihály Kovács |
| Nemzeti Kutatási és Technológiai Hivatal | NKFIH K-116072 | Mihály Kovács |
| Nemzeti Kutatási és Technológiai Hivatal | NKFIH ERC_HU 117680 | Mihály Kovács |
| Nemzeti Kutatási és Technológiai Hivatal | NKFIH K-123989 | Mihály Kovács |
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
Author contributions
Yeonee Seol, Conceptualization, Resources, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing; Gábor M Harami, Conceptualization, Data curation, Software, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review and editing; Mihály Kovács, Conceptualization, Formal analysis, Supervision, Funding acquisition, Investigation, Methodology, Writing—original draft, Project administration, Writing—review and editing; Keir C Neuman, Conceptualization, Software, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Methodology, Writing—original draft, Project administration, Writing—review and editing
Author ORCIDs
Mihály Kovács https://orcid.org/0000-0002-1200-4741
Keir C Neuman https://orcid.org/0000-0002-0863-5671
Decision letter and Author response
Decision letter https://doi.org/10.7554/eLife.45909.029
Author response https://doi.org/10.7554/eLife.45909.030
Additional files
Supplementary files
• Supplementary file 1. Supplementary Tables including DNA sequences and fitting parameters. Table S1: Sequence (5’ to 3’) of ssDNA strands composing gc36, gc46 and gc79 forked dsDNA substrates (complementary regions are bold; Flu represents 3’ fluorescein labeling). Table S2: Parameters determined from global fitting ensemble unwinding data with the n-step and delayed release models. a Parameters determined from fitting ensemble unwinding data with the n-step model with n = 5 (cf. Figure 4B). b Parameters determined from fitting ensemble unwinding data with the delayed release model with n = 5 for RecQWT and n = 4 for RecQ-dH (cf. Figure 4B). Table S3 Parameters from fitting of Poff, toff, and v as a function of ATP with three schemes. Table S4 Number of events in the analysis: a Number of events (Na’ mM), b Number of events (ATPγS μM), c Number of events (ATP μM) KSvK.
DOI: https://doi.org/10.7554/eLife.45909.024
• Transparent reporting form
DOI: https://doi.org/10.7554/eLife.45909.025
Data availability
The single molecule experimental data analysis codes in this study were previously published and referenced in the manuscript. The Kerssemakers step-finder routine (Kerssemakers et al. (2006) Nature 442:709-712) is available from the authors. Alternative step-finding routines (Wiggins (2015) Biophys J 109:346-354; Hill et al. (2018) J Chem Phys 148:123317) are available online (at http://mtshasta.phys.washington.edu/website/steppi/ or https://github.com/duderstadt-lab/Julia_KCP_Notebooks). Source data for all of the figures and graphs are provided in the main and supplemental data.
References
Bachrati CZ, Hickson ID. 2003. RecQ helicases: suppressors of tumorigenesis and premature aging. Biochemical Journal 374:577–606. DOI: https://doi.org/10.1042/bj20030491, PMID: 12803543
Bachrati CZ, Hickson ID. 2008. RecQ helicases: guardian angels of the DNA replication fork. Chromosoma 117: 219–233. DOI: https://doi.org/10.1007/s00412-007-0142-4, PMID: 18188578
Bagchi D, Manosas M, Zhang W, Manthei KA, Hodeib S, Ducas B, Keck JL, Croquette V. 2018. Single molecule kinetics uncover roles for E. coli RecQ DNA helicase domains and interaction with SSB. Nucleic Acids Research 46:8500–8515. DOI: https://doi.org/10.1093/nar/gky647, PMID: 30053104
Bennett RJ, Keck JL. 2004. Structure and function of RecQ DNA helicases. Critical Reviews in Biochemistry and Molecular Biology 39:79–97. DOI: https://doi.org/10.1080/10409230490460756, PMID: 15217989
Bernstein DA, Zittel MC, Keck JL. 2003. High-resolution structure of the E. coli RecQ helicase catalytic core. The EMBO Journal 22:4910–4921. DOI: https://doi.org/10.1093/embojd/22.19.4910, PMID: 14517231
Bernstein DA, Keck JL. 2005. Conflating substrate specificity to DNA helicases: role of the RecQ HRDC domain. Structure 13:1173–1182. DOI: https://doi.org/10.1016/j.str.2005.04.018, PMID: 16084389
Betterton MD, Julicher F. 2005. Opening of nucleic-acid double strands by helicases: Active versus passive opening. Physical Review E 71:011904. DOI: https://doi.org/10.1103/PhysRevE.71.011904
Carter NJ, Cross RA. 2005. Mechanics of the kinesin step. Nature 435:308–312. DOI: https://doi.org/10.1038/ nature03528, PMID: 15902239
Chatterjee S, Zagelbaum J, Savitsky P, Sturzenegger A, Huttner D, Janscak P, Hickson ID, Gileadi O, Rothenberg E. 2014. Mechanistic insight into the interaction of BLM helicase with intra-strand G-quadruplex structures. Nature Communications 5:5556. DOI: https://doi.org/10.1038/ncomms6556, PMID: 25418155
Cheng W, Dumont S, Tinoco I, Bustamante C. 2007. NS3 helicase actively separates RNA strands and senses sequence barriers ahead of the opening fork. PNAS 104:13954–13959. DOI: https://doi.org/10.1073/pnas.0702351104, PMID: 17709749
Cheng W, Arunajadai SG, Moffitt JR, Tinoco I, Bustamante C. 2011. Single-base pair unwinding and asynchronous RNA release by the hepatitis C virus NS3 helicase. Science 333:1746–1749. DOI: https://doi.org/10.1126/science.120623, PMID: 21940894
Chu WK, Hickson ID. 2009. RecQ helicases: multifunctional genome caretakers. Nature Reviews Cancer 9:644–654. DOI: https://doi.org/10.1038/nrc2682, PMID: 19657341
Fairman-Williams ME, Guenther UP, Jankowsky E. 2010. SF1 and SF2 helicases: family matters. Current Opinion in Structural Biology 20:313–324. DOI: https://doi.org/10.1016/j.sbi.2010.03.011, PMID: 20456941
Hanada K, Ukita T, Kohno Y, Saito K, Kato J, Ikeda H. 1997. RecQ DNA helicase is a suppressor of illegitimate recombination in Escherichia coli. PNAS 94:3860–3865. DOI: https://doi.org/10.1073/pnas.94.8.3860, PMID: 9108069
Harami GM, Nagy NT, Martina M, Neuman KC, Kovács M. 2015. The HRDC domain of E. coli RecQ helicase controls single-stranded DNA translocation and double-stranded DNA unwinding rates without affecting mechanoenzymatic coupling. Scientific Reports 5:11091. DOI: https://doi.org/10.1038/srep11091, PMID: 26067769
Harami GM, Seol Y, In J, Ferencziévá V, Martina M, Gyimesi M, Sarlós K, Kovács ZJ, Nagy NT, Sun Y, Vellai T, Neuman KC, Kovács M. 2017. Shuttling along DNA and directed processing of D-loops by RecQ helicase support quality control of homologous recombination. PNAS 114:E466–E475. DOI: https://doi.org/10.1073/pnas.1615439114, PMID: 28069956
Herbert KM, La Porta A, Wong BJ, Mooney RA, Neuman KC, Landick R, Block SM. 2006. Sequence-Resolved detection of pausing by single RNA polymerase molecules. Cell 125:1083–1094. DOI: https://doi.org/10.1016/j.cell.2006.04.032
Huguet JM, Bizarro CV, Forns N, Smith SB, Bustamante C, Ritort F. 2010. Single-molecule derivation of salt dependent base-pair free energies in DNA. PNAS 107:15431–15436. DOI: https://doi.org/10.1073/pnas.1001454107, PMID: 20716688
Kerssemakers JW, Munteau EL, Laan L, Noetzel TL, Janson ME, Dogterom M. 2006. Assembly dynamics of microtubules at molecular resolution. Nature 442:709–712. DOI: https://doi.org/10.1038/nature04928, PMID: 16799566
León-Ortiz AM, Panier S, Sarek G, Vannier JB, Patel H, Campbell PJ, Boulton SJ. 2018. A distinct class of genome rearrangements driven by heterologous recombination. Molecular Cell 69:292–305. DOI: https://doi.org/10.1016/j.molcel.2017.12.014, PMID: 29351848
Lin W, Ma J, Nong D, Xu C, Zhang B, Li J, Jia Q, Dou S, Ye F, Xi X, Lu Y, Li M. 2017. Helicase Stepping Investigated with One-Nucleotide Resolution Fluorescence Resonance Energy Transfer. Physical Review Letters 119:138102. DOI: https://doi.org/10.1103/PhysRevLett.119.138102
Lohman TM, Tomko EJ, Wu CG. 2008. Non-hexameric DNA helicases and translocases: mechanisms and regulation. Nature Reviews Molecular Cell Biology 9:391–401. DOI: https://doi.org/10.1038/nrm2394, PMID: 18414490
Lucius AL, Maluf NK, Fischer CJ, Lohman TM. 2003. General methods for analysis of sequential “n-step” kinetic mechanisms: application to single turnover kinetics of helicase-catalyzed DNA unwinding. Biophysical Journal 85:2224–2239. DOI: https://doi.org/10.1016/S0006-4956(03)74648-7, PMID: 14507688
Ma JB, Jia Q, Xu CH, Li JH, Huang XY, Ma DF, Li M, Xi XG, Lu Y. 2018. Asynchrony of Base-Pair breaking and nucleotide releasing of helicases in DNA unwinding. The Journal of Physical Chemistry. B 122:5790–5796. DOI: https://doi.org/10.1021/acs.jpcb.8b01470, PMID: 29733603
Manosas M, Xi XG, Bansimon D, Croquette V. 2010. Active and passive mechanisms of helicases. Nucleic Acids Research 38:5518–5526. DOI: https://doi.org/10.1093/nar/gkq273, PMID: 20423906
Manthei KA, Hill MC, Burke JE, Butcher SE, Keck JL. 2015. Structural mechanisms of DNA binding and unwinding in bacterial RecQ helicases. PNAS 112:4292–4297. DOI: https://doi.org/10.1073/pnas.1416746112, PMID: 25831501
et al. eLife 2019;8:e45909. DOI: https://doi.org/10.7554/eLife.45909
Appendix 1
DOI: https://doi.org/10.7554/eLife.45909.026
Kinetic analysis of ATP-dependent unwinding and backsliding reveals that DNA melting precedes ATP hydrolysis
Usage of limiting ATP concentrations in hairpin unwinding experiments revealed a kinetic state of RecQ-dH that branched off the unwinding pathway, which enabled us to formulate a kinetic scheme relating ATP binding and hydrolysis by RecQ to its unwinding and stepping activities.
We considered three potential schemes for the coupling between the ATP binding and hydrolysis cycle and enzyme translocation. Each scheme consists of a different arrangement of four steps: DNA melting, ATP binding, hydrolysis, and RecQ translocation along ssDNA. In scheme A, RecQ binding triggers DNA melting that allows ATP binding and hydrolysis, followed by translocation of RecQ. In scheme B, DNA melting and immediate translocation of RecQ are driven by ATP hydrolysis. In scheme C, DNA melting occurs after ATP binding followed by ATP hydrolysis and translocation of RecQ.
Upon decreasing the ATP concentration from 1 mM to 5 μM, we observed sudden decreases in DNA hairpin extension, corresponding to rapid re-zipping of the hairpin that terminated in a pause followed by the resumption of unwinding. The probability and duration of these backsliding and pausing events increased with decreasing ATP concentration, consistent with an off-pathway weakly DNA-bound RecQ state in kinetic competition with ATP binding. We obtained the off-pathway probability (P_{off}) by counting the number of sudden extension decrease events (Figure 5—figure supplement 1A) normalized by the number of unwound base-pairs for each trace. To obtain the off-state duration (t_{off}), we measured individual pause durations between sudden DNA extension decrease (re-zipping) events and the resumption of DNA unwinding (Figure 5—figure supplement 1A). For each kinetic scheme, P_{off} can be calculated by summing P_{off} for all possible pathways while the off-state duration (t_{off}) can be calculated using a recursive relation (Shaevitz et al., 2005). As some of the kinetic rates are common in P_{off}, t_{off}, and the unwinding velocity, v, we globally fitted the experimentally measured P_{off}, t_{off}, and v as a function of ATP concentration with the analytical expressions for P_{off} and t_{off} for each scheme and v (see below) resulting in k_{on}, k_{off}, K_{M} (=k_{ra}/k_{a}), k_{p}, n and k_{cat}. The kinetic parameters are defined as follows:
| Parameter | Description |
|-----------|-------------|
| D_{n} | DNA hairpin unwound by n base pairs |
| R | RecQ enzyme |
| D+R | Weakly bound DNA-RecQ state |
| D*R | Tightly bound DNA-RecQ state, competent for unwinding |
| D_{n}* | DNA unwound by n base pairs with one additional melted base-pair |
| k_{a} | ATP binding rate to RecQ during DNA unwinding |
| k_{a} | ATP dissociation rate from RecQ during DNA unwinding |
| k_{h} | ATP hydrolysis rate during DNA unwinding |
| k_{on} | Rate from off-pathway weak DNA binding state to tight DNA binding state |
| k_{off} | Rate from tight DNA binding state to weak, off-pathway, binding state |
| K_{M} | ATP dissociation constant (k_{ra}/k_{a}) |
| k_{cat} | Maximum unwinding rate of RecQ |
| n | Coupling between unwinding rate and ATP hydrolysis rate |
continued on next page
In order to decrease the free parameters, we set \( k_a \) and \( k_h \) and \( k_m \) to be \( 2 (\text{ s}^{-1} \mu\text{M}^{-1}) \), 200 (\text{ s}^{-1} \)), and 121 (\text{ s}^{-1} \)), respectively, based on previous measurements (Sarloš et al., 2012) and the average unwinding time from the experimental measurements and used \( K_M k_a / C_{031}^2 \) instead of \( k_a / C_{031}^2 \) in the fitting. The Chi-squared value (\( \chi^2 \)) from the global fitting with scheme A is the minimum among the three as \( \chi^2 \) values for schemes A, B, and C are 82, 604, and 9420, respectively. This suggests that ATP binding and DNA melting of RecQ may precede ATP hydrolysis, as encompassed in scheme A.
### Three kinetic schemes of RecQ unwinding and translocation for single-molecule measurement analysis
For all schemes, \( R \) is RecQ, \( D \) is DNA, \( n \) indicates the current DNA base-pair position where RecQ binds and performs its catalytic activity and \( D_n^* \) represents a DNA state in which the \( n+1 \)st DNA base-pair is melted prior to translocation, which results in \( D_{n+1} \).
#### A ATP binding and DNA melting
\[
D + R \\
\stackrel{k_{on}}{\rightleftharpoons} \stackrel{k_{off}}{\rightleftharpoons} D_n \cdot R \\
\stackrel{k_m}{\rightleftharpoons} \stackrel{k_p}{\rightleftharpoons} D_n \cdot R \cdot \text{ATP} \stackrel{k_p}{\rightleftharpoons} D_n \cdot R \cdot \text{ADP} \cdot \text{Pi} \stackrel{k_{off}}{\rightleftharpoons} D_{n+1} \cdot R \\
\]
\[P_{off} = \frac{k_{off}}{k_{off} + k_m} \left[ 1 + \frac{k_m k_p (k_{on} + k_h)}{(k_{off} + k_m) k_a [\text{ATP}] k_h + (k_m + k_h) k_m k_p} \right] \]
\[\tau_{off} = \frac{1}{k_a [\text{ATP}]} + \frac{1}{k_m [k_a [\text{ATP}]] + 1} + \frac{1}{k_h} \left[ 1 + \frac{k_p (k_{on} + k_{off}) (k_{on} + k_h)}{k_h k_m k_a [\text{ATP}]} \right] \]
#### B DNA melting and ATP binding
\[
D + R \\
\stackrel{k_{on}}{\rightleftharpoons} \stackrel{k_{off}}{\rightleftharpoons} D_n \cdot R \\
\stackrel{k_m}{\rightleftharpoons} \stackrel{k_p}{\rightleftharpoons} D_n \cdot R \cdot \text{ATP} \stackrel{k_p}{\rightleftharpoons} D_n \cdot R \cdot \text{ADP} \cdot \text{Pi} \stackrel{k_{off}}{\rightleftharpoons} D_{n+1} \cdot R \\
\]
\[P_{off} \text{ and } \tau_{off} \text{ for scheme B are:} \]
C ATP binding and hydrolysis follow by DNA melting
\[
P_{\text{eff}} = \frac{k_{\text{off}}}{k_{\text{off}} + k_{\text{ATP}}} \left[ 1 + \frac{k_{\text{ATP}}k_{\text{ATP}}(k_{\text{p}} + k_{\text{h}})}{(k_{\text{off}} + k_{\text{ATP}})k_{\text{h}}(k_{\text{p}} + k_{\text{h}})} \right]
\]
\[
\tau_{\text{eff}} = \frac{1}{k_{\text{on}}} \left( \frac{1}{k_{\text{on}}} + \frac{1}{k_{\text{ATP}}} \right) + \frac{1}{k_{\text{ATP}}} \left[ 1 + \frac{k_{\text{p}}(k_{\text{on}} + k_{\text{off}})}{k_{\text{h}}k_{\text{on}}k_{\text{in}}} \right]
\]
The dependence of the unwinding velocity, \(v\), on ATP concentration with a Hill coefficient \(r\):
\[
v = \frac{k_{\text{off}}}{1 + \left( \frac{K_{M}}{[\text{ATP}]^r} \right)}
\]
**Adenosine 5'-O-(3-thiotriphosphate) (ATP\gamma S) hydrolysis measurement**
ATP\gamma S hydrolysis by RecQ-dH was estimated by measuring the level of thiophosphate in the reaction mixture using Malachite Green (BioAssay System; POMG-25H). First, the thiophosphate standard was generated by measuring the OD at 620 nm of 8 different thiophosphate concentrations (0, 1, 2, 4, 8, 10, 20, and 40 \(\mu M\)) in RecQ activity buffer. Using this standard, the amount of thiophosphate production from ATP\gamma S hydrolysis by RecQ was estimated by measuring OD at 620 nm of the reaction mixtures containing 200 nM dT\(_{54}\) DNA, 1 mM ATP\gamma S and 200 nM RecQ at 0, 30 min, 1, 2, and 3 hr reaction times. Concurrently, the OD of the same reaction mixture without RecQ was also measured to correct any effects of auto-hydrolysis of ATP\gamma S.
**HRDC-dependent pausing analysis and simulation**
In a simple kinetic scenario, HRDC-dependent pausing can be considered as an off-pathway state with a single rate-limiting step, \(k_{\text{HP}}\) (the rate of entering an HRDC-stabilized pause state) that is in kinetic competition with the forward unwinding rate, \(k_{\text{step}}\). The rate of escaping from the pause, that is the inverse lifetime of the HRDC-stabilized pause state or unbinding rate is \(k_{\text{-HP}}\) (Herbert et al., 2006). As there are no potential long pauses before or after the peak at 40 bp (Figure 2), we can estimate the probability and kinetics of HRDC-dependent pausing related to the sequence-dependent pausing. Although the average pause duration at 40 bp for RecQ\(^{WT}\) (0.16 ± 0.03 s) is comparable to that of RecQ-dH (0.14 ± 0.03 s), the long pauses (>0.5 s) were observed only in the traces of RecQ\(^{WT}\) but not RecQ-dH hairpin unwinding suggesting that they represent HRDC binding (Figure 6—figure...
At the 40 bp hairpin position, the step rate, \( k_{\text{step}} \) (inverse of pause duration for RecQ-dH) is \( \sim 7 \text{ s}^{-1} \) and the HRDC-stabilized pause efficiency is 0.25, estimated from the fraction of the RecQ\textsuperscript{WT} pause duration distribution that exceeds the exponential fit with \( k_{\text{step}} \) (Figure 6—figure supplement 1). The estimated \( k_{\text{HP}} \) is then 2.4 \text{s}^{-1} and the estimated \( k_{\text{-HP}} \) is 0.89 \text{s}^{-1}. Based on the sequence-dependent stepping rate \( k_{\text{step}} \) and the calculated \( k_{\text{HP}} \) and \( k_{\text{-HP}} \) rates, HRDC-dependent pauses were calculated over the entire 174 bp hairpin. We found that the calculated pause durations and pause locations based on these simple assumptions fail to reproduce the RecQ\textsuperscript{WT} pause behavior (Figure 6—figure supplement 1). In general, the pause locations are less localized, with HRDC-stabilized pauses occurring randomly throughout the hairpin sequence, and the average lifetimes of the specific pauses are shorter, in the simulations as compared to the RecQ\textsuperscript{WT} data. This suggests that the pathway into and out-of the HRDC-stabilized pause may not be a simple single-step process, but rather a multi-phasic kinetic step. In support of this possibility, we found that the pausing duration distribution is better described by double-exponential (\( \chi^2 = 2.1 \)) than single-exponential (\( \chi^2 = 3.2 \)) (Figure 6—figure supplement 1B).
**DNA hairpin sequence information**
The sequence information indicates the 3´ to 5´ translocating strand (green region) for all DNA hairpins.
174 base-pair DNA hairpin (translocation strand)
| 3´GTCGAAGGCTGACGTCGGACTCGCTCCGACTCCACTAGGGCTGGG |
| :----------------- |
| TAAACGACAGGTCGTCAGT |
| ACGATCGGTATACCGACGGCGCCGTGGTCCCGGACGACACTACTACTAC |
| TACCGACAGGCCGTACCATAGAGGAAGAATTTCtggtgtaccgtcatcctt |
174 base-pair DNA hairpin containing mismatches in the displacing strand
Two mismatches: G(90) \( \rightarrow \) T; [Original sequence (position in the hairpin) \( \rightarrow \) mutated sequence]
| 5´CAGCTTCCGACTGCAGCCTGACGCCAGGGCTGATGATGATA |
| ACCGTCATGCTAGCCATATGCTGCGCCGTGCAACCCAGGGCTGCTGATGATGATA |
| TGATGATGCTGCCATATGCTGATGATGATGATA |
Two mismatches: G(90) \( \rightarrow \) T; G(104) \( \rightarrow \) T; [Original sequence (position in the hairpin) \( \rightarrow \) mutated sequence]
| 5´CAGCTTCCGACTGCAGCCTGACGCCAGGGCTGATGATGATA |
| ACCGTCATGCTAGCCATATGCTGCGCCGTGCAACCCAGGGCTGCTGATGATGATA |
| TGATGATGCTGCCATATGCTGATGATGATGATA |
Three mismatches: G(90) \( \rightarrow \) T; G(104) \( \rightarrow \) T; G(124) \( \rightarrow \) T; [Original sequence (position in the hairpin) \( \rightarrow \) mutated sequence]
| 5´CAGCTTCCGACTGCAGCCTGACGCCAGGGCTGATGATGATA |
| ACCGTCATGCTAGCCATATGCTGCGCCGTGCAACCCAGGGCTGCTGATGATGATA |
| TGATGATGCTGCCATATGCTGATGATGATGATA |
Seol et al. eLife 2019;8:e45909. DOI: https://doi.org/10.7554/eLife.45909
Oligos that were used to generate 174 bp with mismatches
Translocation strand
- 5’TGGTCTTTAAGAAGGAGATATACCATGGGCAGCAGCCATCATCATCA
- TCACAGCAGCGGCCTGGTGCGCCCGGCCAGCCATATGG
Displacing strand
- 5’CTAGCCATATGgCTGCCGCGCTGCACCAGGCCGCTTCTGTGATGATGATGATGTA
- TGGCTG
- CTGCCCATGGTATATCTCCTCTTTAAAC (one mismatch)
- 5’CTAGCCATATGgCTGCCGCGCTGCACCAGGCCGCTTCTGTGATGATGATGATGTA
- TGGCTG
- CTGCCCATGGTATATCTCCTCTTTAAAC (two mismatches)
- 5’CTAGCCATATGgCTGCCGCGCTGCACCAGGCCGCTTCTGTGATGATGATGATGTA
- TGGCTG
- CTGCCCATGGTATATCTCCTCTTTAAAC (three mismatches)
584 base-pair DNA hairpin
3’ cagcttgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagctgcagcagtgtt
Seol et al. eLife 2019;8:e45909. DOI: https://doi.org/10.7554/eLife.45909 | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
2832,
1
],
[
2832,
5191,
2
],
[
5191,
9969,
3
],
[
9969,
13142,
4
],
[
13142,
15629,
5
],
[
15629,
18851,
6
],
[
18851,
23728,
7
],
[
23728,
25137,
8
],
[
25137,
29788,
9
],
[
29788,
32859,
10
],
[
32859,
36997,
11
],
[
36997,
41958,
12
],
[
41958,
42961,
13
],
[
42961,
47718,
14
],
[
47718,
52765,
15
],
[
52765,
56854,
16
],
[
56854,
58975,
17
],
[
58975,
62640,
18
],
[
62640,
65860,
19
],
[
65860,
71518,
20
],
[
71518,
71587,
21
],
[
71587,
74983,
22
],
[
74983,
77268,
23
],
[
77268,
79845,
24
],
[
79845,
82832,
25
],
[
82832,
83959,
26
]
]
} |
11184e45678c5a551cbbd5869e46e119688bfcb6 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 4,
"total-fallback-pages": 0,
"total-input-tokens": 14866,
"total-output-tokens": 7150,
"fieldofstudy": [
"Physics"
]
} | LITHIUM ABUNDANCES IN RED GIANTS OF M4: EVIDENCE FOR ASYMPTOTIC GIANT BRANCH STAR POLLUTION IN GLOBULAR CLUSTERS?
Valentina D’Orazi1 and Anna F. Marino2
1 INAF–Osservatorio Astronomico di Padova, vicolo dell’Osservatorio 5, I-35122, Padova, Italy; [email protected]
2 Dipartimento di Astronomia, Università di Padova, vicolo dell’Osservatorio 3, I-35122, Padova, Italy; [email protected]
Received 2010 March 31; accepted 2010 May 17; published 2010 June 1
ABSTRACT
The determination of Li and proton-capture element abundances in globular cluster (GC) giants allows us to constrain several key questions on the multiple population scenarios in GCs, from formation and early evolution to pollution and dilution mechanisms. In this Letter, we present our results on Li abundances for a large sample of giants in the intermediate-metallicity GC NGC 6121 (M4), for which Na and O have been already determined by Marino et al. The stars analyzed are both below and above the red giant branch bump luminosity. We found that the first and second generation stars share the same Li content, suggesting that a Li production must have occurred. This provides strong observational evidence supporting the scenario in which asymptotic giant branch stars are GC polluters.
Key words: globular clusters: individual (NGC 6121) – stars: abundances – stars: individual (Population II)
Online-only material: color figures, machine-readable table
1. INTRODUCTION
The presence of multiple populations as a characterizing property of globular clusters (GCs) is widely accepted nowadays. Since the first photometric and spectroscopic studies, which revealed multiple sequences (e.g., Dickens & Woolley 1967; Lee et al. 1999, 2009; Bedin et al. 2004) and/or light element (anti)correlations (see Gratton et al. 2004), it became evident that GC stars are neither coeval nor chemically homogeneous. Hence, GCs host at least two stellar generations. Thanks to the advent of 8–10 m class telescopes, which also allowed targeting fainter main-sequence (MS) stars, several abundance studies have shown that chemical (anti)correlations are also present in unevolved (MS/turnoff (TO)) or scarcely evolved (SGB) stars (Gratton et al. 2001; Carretta et al. 2004; Pasquini et al. 2005). This evidence implies that a previous generation of stars has activated CNO, NeNa, and MgAl cycles in their interiors in order to deplete O and Mg and enhance Na and Al, respectively. The origin/nature of such stars is still debated with two coexistent scenarios: (1) intermediate-mass asymptotic giant branch (AGB) stars in hot bottom burning phase (Ventura & D’Antona 2009) and (2) fast rotating massive stars (FRMSs; Decressin et al. 2007).
Our group has recently carried out an extensive survey, focusing on the determination of proton-capture elements in 19 GCs (Carretta et al. 2009c), with the main objective of discovering and understanding the nature (and the extent) of the chemical (anti)correlations and their link with global cluster parameters (horizontal branch (HB), morphology, metallicity, etc.). The large sample of stars (≈1200 GC members), analyzed in a very homogeneous and accurate way, revealed that while the Na–O anticorrelation is present in all the GCs (i.e., the second generation is not a “perturbation”), the shape of the Na–O distribution varies from cluster to cluster (Carretta et al. 2009c). On the other hand, the analysis of UVES spectra for ≈200 stars by Carretta et al. (2009b) has shown that the Mg–Al anticorrelation is not present in all GCs. Both of these indications suggest that the typical polluter masses change from cluster to cluster: this variation is apparently driven by a combination of cluster luminosity and metallicity.
In this context, lithium abundances offer a complementary approach to p-capture elements allowing to address several important issues. If no Li is produced by the polluters, the multiple population scenario predicts that Li and O are positively correlated, while Li and Na anticorrelated. Na-poor, O/Li-rich stars are the first population born in the cluster (they share the same chemical composition of field stars at the same metallicity), and Na-rich, O/Li-poor stars constitute the second generation. Within the same hypothesis, Li is an excellent tracer of the dilution process acting within each star: only through Li abundance determinations we can determine the amount of pristine (and of polluted) material present in each star. In particular, we hope to answer two fundamental questions: do 100% polluted stars (Li = 0) exist or does even the most extreme population still contains a certain fraction of primordial matter? Also, is the minimum measurable Li content the same for all GCs or does it vary from cluster to cluster?
On the other hand, Li offers the exciting chance to observationally constrain the nature of the polluters. If the progenitors of second generation stars are FRMSs, they have destroyed their original Li content. On the other hand, if AGB stars are responsible for intracluster pollution, they may have non-negligible Li yield, given the Li production via the “7Be transport” mechanism (Cameron & Fowler 1971). As a consequence, we can reveal whether the AGB stars are responsible for GC pollution through two main observational facts: (1) the presence of very Li-rich stars among GC populations and (2) the lack of Li–Na anticorrelation (or Li–O correlation), with the second generation stars also showing a rather high Li content. In a recent work, we focused on ≈90 TO stars belonging to the metal-rich GC 47 Tuc: in this case, likely because of the high metallicity, a large star-to-star scatter in Li abundances erases any Li–Na anticorrelation, while Li and O appear to be only weakly positively correlated (D’Orazi et al. 2010). The cluster seems to display a different behavior from NGC 6752 and NGC 6397: for the first one, Pasquini et al. (2005) detected a significant Li–Na anticorrelation (and also Li–O correlation and Li–N anticorrelation). Concerning the second cluster, the situation seems more complex since Lind et al. (2009) detected a quite constant value in Li abundances, with only three
stars (out of 100) driving a hint of Li–Na anticorrelation. Enlarging the sample of simultaneous determinations of Li and $p$-capture elements in GCs is hence of paramount importance: in this Letter, we present Li results on the intermediate-metallicity ([Fe/H] = −1.18; Carretta et al. 2009a) cluster NGC 6121 (M4), by analyzing the same sample of a hundred red giant branch (RGB) stars studied by Marino et al. (2008). As found by Marino et al., M4 hosts two distinct populations of stars, mainly characterized by a different sodium content (i.e., the Na-rich and Na-poor groups) and defining different sequences on the color–magnitude diagram $U$ versus $(U − B)$. We derived Li abundances for the stars belonging to the two groups and located both below and above the RGB bump luminosity; this evolutionary stage plays a fundamental role in this context.
Theoretical models (Iben 1967), confirmed by observations of field stars by Gratton et al. (2000), predict a depletion in Li due to the first dredge-up (1DUP) of a factor of ∼20 at the base of the SGB branch. On the lower RGB, below the bump luminosity, the molecular weight gradient associated with the H abundance jump acts as a barrier that prevents further extramixing (Charbonnel et al. 1994); then the Li abundance remains constant until the RGB bump is reached. At this stage, the H shell reaches and cancels this discontinuity, and non-standard mixing processes (non-convective extramixing; see, e.g., Charbonnel & Zahn 2007) cause the total destruction of the remaining Li.
### 2. SAMPLE AND ANALYSIS
Our sample consists of 104 RGB stars, whose spectra were acquired with FLAMES on VLT/UT2 (Pasquini et al. 2002) with the fiber link to the high-resolution spectrograph UVES ($R ∼ 40,000$). A detailed description of observations, target properties, and data reduction and derivation of atmospheric parameters is provided in Marino et al. (2008). Adopting Kurucz (1995) model atmospheres and using the ROSA abundance code (Gratton 1988), we derived Li abundances by means of a spectral synthesis of the Li $i$ resonance doublet at 6708 Å. We changed the CN values for the two different groups of Na-rich and Na-poor stars (threshold value at [Na/Fe] = 0.2 dex) in order to optimize the synthesis best fit and to account for the CN enhancement in the Na-rich population (see Marino et al. 2008).
Abundances for Na $i$ along with stellar parameters and metallicity are the ones presented in Marino et al. (2008). Concerning Li abundance, error estimates have been computed in the same fashion as described in D’Orazi et al. (2010), taking into account both stellar parameter and best-fit uncertainties; for errors in Na (internal and systematic), we refer the reader to Marino et al. (2008). Stellar parameters and abundances are given in Table 1 (completely available in electronic version through CDS).
### 3. RESULTS AND DISCUSSION
In Figure 1, we show the Li abundances as a function of the absolute magnitude $M_V$ for all our sample stars: as expected, Li disappears above the bump luminosity ($M_V = −0.05 ± 0.10$; Ferraro et al. 1999). If we focus on the region below the bump level (at the left side of the dashed line in Figure 1), there is no systematic difference in Li abundances between Na-rich (filled squares) and Na-poor stars (empty symbols). However, when we look at the diagram as a whole we can see a different drop in the Li content with magnitude for the two populations. Specifically, while Li seems to have a gentle decrease with luminosity for the Na-poor stars, the Na-rich group presents a very abrupt decline, i.e., at the bump luminosity Li suddenly disappears. This fact, which reflects different timescales for mixing and hence for Li depletion, suggests a structural difference between Na-rich and Na-poor stars; however, no current theoretical model
\[\text{Table 1} \]
| Star | $T_{\text{eff}}$ (K) | log $g$ | $\xi$ (km s$^{-1}$) | [Na/Fe] | Err Na | log $n$(Li) | Err Li |
|------|---------------------|--------|--------------------|-------|-------|-------------|-------|
| 30345 | 4850 | 2.73 | 1.31 | 0.43 | 0.07 | 1.37 | 0.08 |
| 30452 | 4830 | 2.56 | 1.25 | 0.06 | 0.03 | 1.29 | 0.09 |
| 30719 | 4810 | 2.65 | 1.24 | 0.42 | 0.06 | 1.22 | 0.08 |
| 31306 | 4900 | 2.87 | 1.33 | 0.40 | 0.05 | 1.35 | 0.07 |
This table is available in its entirety in a machine-readable form in the online journal. A portion is shown here for guidance regarding its form and content.
\[\text{Figure 1} \]
Li as a function of absolute magnitudes for Na-rich (filled squares) and Na-poor stars (empty squares). The dashed line marks the bump luminosity as derived by Ferraro et al. (1999), while the solid line is an eye fit to Na-poor population.
(A color version of this figure is available in the online journal.)
---
3 We estimated the e-folding time for Li abundance using the tracks by Bertelli et al. (2008). Na-poor stars reduce to about a factor of 2 their Li content in 0.17 mag, which corresponds to $\approx 10$ Myr (this is indeed the time required by a 0.8 $M_\odot$ star to become brighter by 0.17 mag after it has left the RGB bump). For the Na-poor stars, this time is smaller by at least an order of magnitude, i.e., $\lesssim 2$ Myr.
predicts such a behavior and we cannot provide a satisfactory explanation to date. In this context, we mention that the so-called thermohaline mixing has been proposed responsible for non-canonical mixing acting at the RGB bump (see, e.g., Eggleton et al. 2006; Charbonnel & Zahn 2007). Eggleton and coworkers suggested that the molecular weight inversion created by the $^3\text{He}(^3\text{He,2p})^4\text{He}$ reaction could be the cause of such a mixing: why Na-rich and Na-poor stars should be differentially affected by this kind of mixing is not obvious and our result could be the input for further theoretical and observational investigations in this direction.
By considering only the stars fainter than the bump luminosity, we show Li abundances as a function of Na in Figure 2: as one can see, there is no Li–Na anticorrelation, with second generation stars ([Na/Fe] $> 0.2$ dex) sharing the same Li content of the primordial population. As an example, we show in Figure 3 the spectra, around the Li1 region, for the two most extreme cases in Na abundances. The two stars, with [Na/Fe] = $-0.02$ and [Na/Fe] = $+0.43$, respectively, show identical Li features (note that the stars have very similar parameters, and the same line strength reflects the same Li abundance). The average Li abundances are log $n$(Li) = $1.336 \pm 0.023$ (rms 0.062) and log $n$(Li) = $1.387 \pm 0.038$ (rms 0.136), respectively, for Na-poor and Na-rich stars: although we derive the same Li abundance for the two populations, it is interesting to note that Na-rich stars have a larger scatter in Li with respect to Na-poor ones. A one-tailed Fisher test returns a 5% probability such that a difference can be obtained by chance.
A natural explanation for this similarity in Li content between first and second generation stars (with the last showing a larger scatter) is a Li production. In fact, if a decrease in O of $\sim 50\%$ occurred (as derived, e.g., by Marino et al. 2008 and Carretta et al. 2010), also Li must have been depleted. In a recent work, D’Antona & Ventura (2010) have presented the expected Li production as a function of the polluter mass (AGB stars) for metallicity $Z = 0.001$. Looking at their Figure 5, one can see that a very low mass AGB polluter (i.e., $\approx 4 M_\odot$) can produce a moderate Li content with values very close to the Li plateau (log $n$(Li) $\sim 2.2–2.3$). After considering a depletion of a factor of $\sim 20$ at the 1DUP, this result agrees very well with our values (i.e., log $n$(Li) $\sim 1.3–1.4$). As also briefly explained in Section 1 (and widely discussed in Carretta et al. 2010), there are further indications that only low-mass polluters could have contributed to the observed chemical pattern in M4: (1) an almost “vertical” Na–O anticorrelation, with very small oxygen variation (depletion) and (2) the lack of Mg–Al anticorrelation, which in fact requires high-mass polluters for the activation of higher temperature cycles ($T \approx 65$ MK; Prantzos & Charbonnel 2006). A similar case could have occurred for NGC 6397, where according to Pasquini et al. (2008), two stars differ by $\sim 0.6$ dex in O, but have the same “normal” Li (log $n$(Li) = 2.2). Also, in a recent work Lind et al. (2009), based on a sample of $\sim 100$ MS and early SGB/stars, found no difference in Li abundances between Na-rich and Na-poor stars with only two stars driving a Li–Na anticorrelation. They concluded that Li content is independent of intracluster pollution; however, the Na–O distribution points out to a certain (though small) degree of oxygen depletion and, as a consequence, of Li destruction as well. Hence, if first and second generation stars share the same Li abundances, a Li production should also be required for this cluster. Along with a difference in metallicity of $\sim 0.9$ dex, the two clusters, NGC 6397 and M4, have both quite small integrated magnitudes (i.e., mass) with $M_V = -6.63$.
\footnote{We note that Marino et al. (2008) found evidence for a small increase ($\sim 0.10$ dex) of Al with Na, with Na-rich stars also slightly Mg depleted (see their Table 7). Evidence for a Na–Al correlation was also found by Iovs and coworkers. In any case, the Al variation is small, and no Mg–Al anticorrelation has been observed among M4 stars by Evans et al. (1999), Marino et al. (2008), and Carretta et al. (2009b). Also the Mg isotope ratios, as derived by Yong et al. (2008), show no variation within M4 with values very close to solar ones: the same authors concluded that this is not surprising due to the very little Al variation in this GC.}
Figure 2. Li vs. Na for stars below the luminosity bump. Error bars for Na come from Marino et al. (2008); the uncertainties in Li are due to errors on best fit and effective temperatures.
Figure 3. Comparison of two spectra, near the Li doublet at 6708 Å, for the two extreme cases for Na abundances. Solid and dashed lines are for [Na/Fe] = $-0.02$ and [Na/Fe] = $+0.43$, respectively.
(A color version of this figure is available in the online journal.)
and $M_V = -7.20$, respectively (Harris 1996). The similarity in masses between these two GCs also seems to suggest a similar “typical” polluter for both M4 and NGC 6397, with the requirement to have in both cases neither very high mass polluters (no extended MgAl/NaO anticorrelations, very little He enhancement,\(^5\) and no Li–Na anticorrelation) nor low mass polluters ($\lesssim 4 M_\odot$), otherwise the C+N+O is not constant and/or $\alpha$-process variations should be present (see Ivans 1999; Yong et al. 2008).
The more massive GC NGC 6752 ($M_V = -7.73$) could present a different behavior. We might speculate that only a very low Li production from higher mass polluters of $\approx 5-6 M_\odot$ (see Figure 5 of D’Antona & Ventura 2010) does not erase the Li–Na anticorrelation for this cluster.\(^6\) Note in fact that NGC 6752 presents an extended Na–O anticorrelation and a large variation in Al (i.e., the MgAl chain was active in the polluter stars).
On the other hand, it seems very difficult to discriminate the nature of polluters and their properties for 47 Tuc: maybe this GC was similar to NGC 6752 but the intrinsic scatter in Li abundance, independent of intracluster pollution, washes out the fossil imprint by the previous generation of polluter stars (D’Orazi et al. 2010).
Given the large uncertainties linked to model predictions (cross sections, mass loss law, overshooting, convection treatment, etc.), here our general aim is to provide a “qualitative” comparison between theoretical prescriptions by D’Antona & Ventura (2010) and observational evidence. Also, we stress that our result, based on only a few objects, needs to be confirmed by including a larger number of clusters and of star per cluster. However, we think that our data provide a quite robust observational evidence of AGB stars responsible for GC pollution.
### 4. SUMMARY AND CONCLUSIONS
We report in this Letter Li abundances for a sample of ~100 giants (the same sample already presented in Marino et al. 2008), both below and above the RGB bump luminosity, belonging to the GC NGC 6121 (M4). The main purpose of our work was the study of the correlation (if any) between Li content and elements involved in $p$-capture reactions. The principal results we obtained in our investigations can be summarized as follows.
1. As expected, Li tends to disappear as the stars reach the RGB bump luminosity; however, the Na-rich and Na-poor stars show very different trend of Li with magnitude. Specifically, while the decline of Na-poor stars with $M_V$ is rather smooth, there is a very brusque decrease for Na-rich stars. The so-called thermohaline mixing, which seems responsible for extra-mixing processes at the RGB bump, is not predicted to have different outcomes in Na-poor and Na-rich stars. Further investigations of this aspect are mandatory both from observational and theoretical points of view. New observations, focusing on Li determination along the RGB in several GCs, are necessary to assess if M4 is a “peculiar” case or other GCs share the same behavior; as a consequence model predictions could be revised in this sense.
2. M4 does not show any Li–Na anticorrelation, with first and second generation stars having almost the same Li content. Along with similarities in Li abundances, the larger scatter found in Na-rich stars indicates that a Li production, from the previous generation of polluters, must have happened. This provides support to intermediate-mass AGB stars responsible for intracluster contamination, since FRMSs can only destroy Li.
We warmly thank Raffaele Gratton, Angela Bragaglia, Eugenio Carretta, and Sara Lucatello for very helpful suggestions and useful comments in all the stages of the preparation of this Letter. This work was funded by the Italian MIUR under PRIN 20075TP5K9.
### REFERENCES
Bedin, L. R., et al. 2004, ApJ, 605, 125
Bertelli, G., Girardi, L., Marigo, P., & Nasi, E. 2008, A&A, 484, 815
Cameron, A. G. W., & Fowler, W. A. 1971, ApJ, 164, 111
Carretta, E., Bragaglia, A., Gratton, R., & Lucatello, S. 2009a, A&A, 508, 695
Carretta, E., Bragaglia, A., Gratton, R., & Lucatello, S. 2009b, A&A, 505, 139
Carretta, E., Gratton, R., Bragaglia, A., Bonifacio, P., & Pasquini, L. 2004, A&A, 416, 925
Carretta, E., et al. 2009e, A&A, 505, 117
Carretta, E., et al. 2010, A&A, in press (arXiv:1003.1723)
Charbonnel, C., & Zahn, J.-P. 2007, A&A, 467, 15
Charbonnel, C., et al. 1994, A&A, 283, 155
D’Antona, F., & Ventura, P. 2010, in IAU Symp. 268, Light Elements in the Universe, ed. C. Charbonnel et al. (Cambridge: Cambridge Univ. Press), http://obswww.unige.ch/iau268/Proceedings.htm
Decressin, T., Charbonnel, C., Prantzos, N., & Ekstrom, S. 2007, A&A, 464, 1029
Dickens, R. J., & Woolley, R. v. d. R. 1967, R. Greenwich Obs. Bull., 128, 255
D’Orazi, V., et al. 2010, ApJ, 713, L1
Eggelton, P. P., Dearborn, D. S., & Lattanzio, J. C. 2006, Science, 314, 1580
Ferraro, F. R., et al. 1999, AJ, 118, 1738
Gratton, R. 1988, Rome Obs. Preprint Ser., 29
Gratton, R., Carretta, E., Bragaglia, A., Lucatello, S., & D’Orazi, V. 2010, A&A, in press (arXiv:1004.3862)
Gratton, R., Sneden, C., & Carretta, E. 2004, ARA&A, 42, 385
Gratton, R., Sneden, C., Carretta, E., & Bragaglia, A. 2000, A&A, 354, 169
Gratton, R., et al. 2001, A&A, 369, 87
Harris, W. 1996, AJ, 112, 487
Iben, I. Jr 1967, ApJ, 147, 624
Ivans, I., et al. 1999, AJ, 118, 1273
Kurucz, R. L. 1995, CD-ROM 13 (Cambridge, MA: Smithsonian Astrophysical Observatory)
Lee, J.-W., Kang, Y.-W., Lee, J., & Lee, Y.-W. 2009, Nature, 462, 480
Lee, Y.-W., et al. 1999, Nature, 402, 55
Lind, K., Primas, F., Charbonnel, C., Grundahl, F., & Asplund, M. 2009, A&A, 503, 545
Marino, A. F., et al. 2008, A&A, 490, 625
Pasquini, L., et al. 2002, Messenger, 110, 1
Pasquini, L., et al. 2005, A&A, 441, 549
Pasquini, L., et al. 2008, A&A, 489, 315
Prantzos, N., & Charbonnel, C. 2006, A&A, 458, 135
Ventura, P., & D’Antona, F. 2009, A&A, 499, 835
Yong, D., Karakas, A. I., Lambert, D. L., Chiefari, A., & Limongi, M. 2008, ApJ, 689, 1031
---
5 Although He is the main product of hot H-burning through CNO cycle, the relationship between He and $p$-capture elements is quite complex. For “extreme” GCs, such as NGC 2808 and αCen (with extended NaO–MgAl anticorrelations, multiple MSs, and peculiar HB morphology), the $Y$ content can reach up to $\approx 0.40$. For all the other GCs, even if they show the NaO anticorrelations, large differences in $Y$ are not necessary (see for details Gratton et al. 2010).
6 The trend of Li production with polluter mass is not linear (see Figure 5 of D’Antona & Ventura 2010): the low-mass polluters (3.5–4.5 $M_\odot$) show a Li production close to the Plateau values, while in the higher masses regime (5–7 $M_\odot$) the Li yields become smaller (a “$v$-like” distribution). Moving to very high-mass stars (7–8 $M_\odot$), the Li production reaches extremely high values, up to log $n$(Li) $\approx 4$ dex. | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
6202,
1
],
[
6202,
11452,
2
],
[
11452,
16535,
3
],
[
16535,
23488,
4
]
]
} |
a345bec623599604ed08a62c14fbc0b83bb97d96 | {
"olmocr-version": "0.1.59",
"pdf-total-pages": 12,
"total-fallback-pages": 0,
"total-input-tokens": 42973,
"total-output-tokens": 14599,
"fieldofstudy": [
"Biology"
]
} | We demonstrate that in humans, two metalloproteases, ADAMTS-9 (1935 amino acids) and ADAMTS-20 (1911 amino acids) are orthologs of GON-1, an ADAMTS protease required for gonadal morphogenesis in Caenorhabditis elegans. ADAMTS-9 and ADAMTS-20 have an identical modular structure, are distinct in possessing 15 TSRs and a unique C-terminal domain, and have a similar gene structure, suggesting that they comprise a new subfamily of human ADAMTS proteases. ADAMTS20 is very sparingly expressed, although it is detectable in epithelial cells of the breast and lung. However, ADAMTS9 is expressed in embryonic and adult tissues, and therefore we characterized the ADAMTS-9 protein further. Although the ADAMTS-9 zymogen has many proprotein convertase processing sites, pulse-chase analysis, site-directed mutagenesis, and amino acid sequencing demonstrated that maturation to the active form occurs by selective proprotein convertase (e.g. furin) cleavage of the Arg1557-Phe1568 bond. Although lacking a transmembrane sequence, ADAMTS-9 is retained near the cell surface as well as in the ECM of transiently transfected COS-1 and 293 cells. COS-1 cells transfected with ADAMTS9 (but not vector-transfected cells) proteolytically cleaved bovine versican and aggrecan core proteins at the Glu141-Ala146 bond of versican V1 and the Glu177-Ala172 bond of aggrecan, respectively. In contrast, the ADAMTS-9 catalytic domain alone was neither localized to the cell surface nor able to confer these proteolytic activities on cells, demonstrating that the ancillary domains of ADAMTS-9, including the TSRs, are required both for specific extracellular localization and for its versicanase and aggrecanase activities.
The ADAMTS (A disintegrin-like and metalloprotease (reprolysin type) with thrombospondin type I motif) family consists of secreted zinc metalloproteases with a precisely ordered modular organization that includes at least one thrombospondin type I repeat (TSR)1, 2. Important functions have been established for several members of the family. ADAMTS-4, ADAMTS-5, and (less efficiently) ADAMTS-1 degrade the cartilage proteoglycan aggrecan and are referred to as aggrecanases (3–5). They play a major role in aggrecan loss in arthritis (6, 7). ADAMTS-1 and ADAMTS-4 participate in the turnover of the aggrecan-related proteoglycans versican and brevican in blood vessels (8) and the nervous system, respectively (9). ADAMTS2 mutations cause dermatosparaxis, a recessively inherited disorder characterized by severe skin fragility that results from incomplete proteolytic removal of the procollagen I amino propeptide (N-propeptide) (10). ADAMTS-3 and ADAMTS-14 are procollagen N-propeptidases with probable roles in procollagen II processing in cartilage or procollagen I processing in tissues other than skin, respectively (11, 12). ADAMTS13 mutations lead to inherited thrombocytopenic purpura, a coagulation disorder caused by deficient proteolytic processing of von Willebrand factor (13). ADAMTS1-null mice have abnormal adipogenesis, defective angiogenesis in the adrenal gland, and a defect of ureteric ECM turnover, leading to hydronephrosis (14). ADAMTS2-null mice have fragile skin, and males are infertile (15). Many other ADAMTS enzymes have been discovered through molecular cloning, and their functions are presently unknown. Altogether, 19 human ADAMTS symbols identifying 18 distinct genes and their products have been assigned (note that ADAMTS5 (1) and ADAMTS11 (4) designate the same gene). ADAMTS are also present in invertebrates, which contain fewer ADAMTS genes than mammalian genomes. Caenorhabditis elegans ADAMTS gene, gon-1, has an essential role in reproduction (16). The protease (GON-1) encoded by gon-1 is required for migration of distal tip cells during gonadal morphogenesis. It may have a role in degradation of basement membrane or for processing of extracellular cues required for cell migration (16). GON-1 is the largest of all ADAMTS en-
* This work was supported in part by a grant from the Northeast Ohio Chapter of the Arthritis Foundation, a Yamatouchi USA Foundation Award, and National Institutes of Health (NIH) Grant AR47074 (to S. A.), by NIH Grant HL18645 (to T. W.), and by Canadian Institutes of Health Research Grant MOP-13755 (to R. L.). Tissue samples were provided by the Cooperative Human Tissue Network, which is funded by SCI, NIH. The costs of publication of this article were defrayed in part by the payment of page charges. This article must be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
† A Fonds de la Recherche en Sant du Quebec Senior Scholar.
‡ To whom correspondence should be addressed: Dept. of Biomedical Engineering (ND20), Lerner Research Institute, Cleveland Clinic Foundation, 9500 Euclid Ave., Cleveland, OH 44195; Tel. 216-445-3278; Fax: 216-445-4383; E-mail: [email protected].
The nucleotide sequence(s) reported in this paper has been submitted to the GenBank®/EBI Data Bank with accession number(s) AF488803 (ADAMTS9) and AF488804 (ADAMTS20).
The abbreviations used are: TSR, thrombospondin type I repeat; DMEM, Dulbecco’s modified Eagle’s medium; GAG, glycosaminoglycan; ORF, open reading frame; RT, reverse transcriptase; GAPDH, glyceraldehyde-3-phosphate dehydrogenase.
Gene nomenclature (ADAMTS9 and ADAMTS20) was assigned after consultation with the Human Gene Nomenclature Committee. Adamts9 and Adamts20 are the respective mouse orthologs. The protein products of these genes are designated as ADAMTS-9 and ADAMTS-20. Similar nomenclature is used for other ADAMTS genes and their products. GON-1 refers to the product of the C. elegans gon-1 gene.
zymes described to date and contains 15 TSSRs (16). In addition, it has a presumed globular domain at the C terminus without similarity to known proteins.
Human ADAMTS-9, as previously described (17) contains four TSSRs. Despite being a much smaller enzyme than GON-1, it had greater sequence similarity to it than to any other human ADAMTS (17). Here, we characterize a considerably longer ADAMTS-9 (designated ADAMTS-9B), transferred to subsequently in this paper as ADAMTS-9 that propose the authentic full-length product of ADAMTS9. In addition, we have discovered a novel enzyme, ADAMTS-20, and determined its complete primary sequence. ADAMTS-9 and ADAMTS-20 have an identical domain organization and exon structure and a very similar primary sequence, showing that they comprise a distinct subfamily of GON-1-related ADAMTS proteases in the mammalian genome. We have characterized the zymogen maturation and cellular localization of the more highly expressed of these two proteins, ADAMTS-9, and have investigated its role in proteolysis of the large aggregating proteoglycans versican and aggrecan. Our data demonstrate the critical requirement of the ancillary domains for the pro-peptidase cleavage site. These proteins are therefore of ADAMTS-9, (the signal peptide, prodomain, and catalytic domain), PCR amplification was done using the same forward primer as for the full-length ADAMTS9 cDNA, the reverse primer 5'-AACCTGAGTTTAG-GCAAAGGTTAGGTCG-3' (Xhol site underlined), and fetal heart cDNA (Clontech) as template. The resulting amplicon was cloned in pPFLAG-CMV-5a (Sigma) to generate proteins with in-frame C-terminal FLAG or myc-His tags, respectively, ADAMTS9-GFPFLAG and ADAMTS9-508MYC/HIS. Site-directed mutagenesis of the convertase (e.g. furin) sites (Arg270→Ala, Arg270→Ala, Arg270→Ala) in ADAMTS9-508MYC/HIS was done using the QuikChange site-directed mutagenesis kit (Stratagene).
The insert of the KIAA0688 gene (20) encoding ADAMTS-4 (3) in pBluescript SK (Stratagene) was excised with EcoRI and Xhol and inserted into the corresponding sites of pcDNA3.1MYC/HIS B (Invitrogen) to generate a mammalian expression vector producing untagged ADAMTS-4. The ADAMTS4 and ADAMTS5 ORFs from the convertase-processing site to the stop codon were PCR-amplified and cloned into pPFLAG-CMV-9 (Sigma) for expression in frame with a prepropeptidase leader sequence and three tandem FLAG tags present just downstream of the signal peptide cleavage site. These proteins are therefore secreted with N-terminal FLAG tag 15-FLAG, ADAMTS4-3-FLAG and ADAMTS5-3-FLAG. Site-directed expression plasmids and site-directed mutations were verified by DNA sequencing.
Northern Blot and Quantitative RT-PCR of ADAMTS9 and ADAMTS20 RNA in Situ Hybridization Analysis—Northern blot analysis of mouse and human tissues (Clontech, Palo Alto, CA) was used as the template for rapid amplification of cDNA ends as previously described (1). To confirm that the overlapping cDNA clones obtained represented a contiguous mRNA, the complete ORF was amplified by PCR. The oligonucleotide primers 5'-AACCGCGCCACCACTGATTGTTGATCC-3' (Not I site underlined and start codon italicized) and 5'-CTTCGGAATTAACACTGGCACTTCGACCC-3' (Xhol site underlined and modified stop codon italicized) were used for PCR with human fetal skeletal muscle cDNA as template and Advantage 2 polymerase (Clontech, Palo Alto, CA). The 5.8-kb PCR product was cloned into pGen-T Easy (Promega, Madison, WI) and sequenced completely. cDNA cloning of ADAMTS9 will be reported elsewhere.3 To ask whether there existed additional ADAMTS proteases with a domain organization similar to GON-1 and ADAMTS-9, the human genome sequence (Celera, Rockville, MD) was searched using the amino acid sequence of the unique C-terminal domain of ADAMTS-9, GENSCAN (available on the World Wide Web at genescan.html) to the Genes Database—Multiple tissue northern blots containing 1 μg/lane poly(A+) RNA from mouse embryos and individual adult mouse and human tissues (Clontech, Palo Alto, CA) were hybridized to [α-32P]dCTP-labeled ADAMTS9, ADAMTS20, or ADAMTS9 probes, followed by autoradiographic exposure for 3–7 days. CDNA panels derived from human adult and fetal organs normalized with respect to GAPDH mRNA levels were purchased from Clontech. Real time PCR of these cDNA templates was performed in an ABI Prism 7700 sequence detector using SYBR Green PCR Core Reagents (Applied Biosystems, Foster City, CA), as previously described (12). PCR amplifications were performed in triplicate for all templates, along with parallel measurements of GAPDH cDNA for normalization.
The GAPDH-normalized quantitative data for ADAMTS9 and ADAMTS20 were used to determine the ADAMTS9/ADAMTS20 transcript ratio in all templates examined. The following primers were used for amplification at a concentration of 300 nM each: ADAMTS9 forward, 5'-GGCAAGGGAGGAGCACCTC-3'; ADAMTS9 reverse, 5'-ATCCATC-GAATTCCCTC-3'; ADAMTS20 forward, 5'-GGTGGCAGTTATTGGCAGAAA-3'; ADAMTS20 reverse, 5'-CACGATCCATGCGAACATG-3'. GPHO primers were described previously (12). RT-PCR performed in the absence of template was negative with all primer pairs.
RNA in situ hybridization was performed essentially as previously described (19), using [35S]-labeled antisense and sense RNA probes transcribed from a 600-nt cDNA template encoding the unique domain of ADAMTS-20. Normal human breast and lung tissues, as well as samples of squamous cell carcinoma of breast and adenocarcinoma of lung were obtained under a Cleveland Clinic Foundation Institutional Review Board-approved protocol and fixed in formalin (tissue samples were provided by the Cooperative Human Tissue Network). 5-μm-thick paraffin sections were hybridized to the probes prior to dipping in photographic emulsion (Eastman Kodak Co.) and followed by autoradiographic exposure for 7 days. Nuclei were stained with 4',6-diamidino-2-phenylindole.
ADAMTS9, ADAMTS4, and ADAMTS8 Expression Plasmids—The ADAMTS9 cDNA was excised as a NotI-Xhol fragment and cloned into the NotI and SalI sites of pFLAG-CMV-5a (Sigma) to introduce an in-frame C-terminal FLAG tag (ADAMTS9-FLAG). For expression of ADAMTS9-FLAG, the complete primary sequence was PCR-amplified and cloned into pPFLAG-CMV-9 (Sigma) for expression in frame with a prepropeptidase leader sequence and three tandem FLAG tags present just downstream of the signal peptide cleavage site. These proteins are therefore secreted with N-terminal FLAG tag 15-FLAG, ADAMTS4-3-FLAG and ADAMTS5-3-FLAG. Site-directed expression plasmids and site-directed mutations were verified by DNA sequencing.
ADAMTS-9 Localization in Transfected Cells—COS-1 and 293-HEK cells were maintained and transfected with ADAMTS9-FLAG, ADAMTS9-508MYC/HIS, ADAMTS4-3-FLAG and ADAMTS5-3-FLAG, as described previously (21). Transfected cell lysates and culture medium were harvested separately after 48 h and were separated by reducing SDS-PAGE followed by Western blot analysis using the FLAG M2 monoclonal antibody (Sigma). For immunolocalization of extracellular ADAMTS9-FLAG, ADAMTS9-508MYC/HIS, ADAMTS4-3-FLAG and ADAMTS5-3-FLAG, cells were transfected with anti-FLAG M2 monoclonal antibody 48 h post-transfection without permeabilization as previously described (21). Alternatively, transfected cells were stained following fixation in 4% paraformaldehyde (permeabilization). Nuclei were stained with 4',6-diamidino-2-phenylindole. As controls, COS-1 and 293 cells were transfected with the empty FLAG vector alone, followed by the immunostaining procedure, or the primary antibody was omitted for FLAG staining.
Release of ADAMTS-9 from the cell surface, transfected 293 cells and ECM were harvested by scraping and resuspended in phosphate-buffered saline (10 mM phosphate buffer, pH 7.4, 2.7 mM KCl, 137 mM NaCl). Cells and ECM were gently agitated by end-over-end rotation in PBS alone or in PBS plus 100 mM or 200 mM NaCl at 4 °C for 30 min. ADAMTS-9-508MYC/HIS Purification and Analysis—To obtain stably transfected 293 cells expressing ADAMTS9-508MYC/HIS, selection with G418 (750 μg/ml) was applied after transfection. These cells were maintained in culture medium containing 5% serum and 250 μg/ml G418. Conditioned medium was dialyzed into binding buffer (20 mM sodium phosphate, 500 mM NaCl, pH 7.8, containing 0.03% Brij-35 (Sigma)) prior to binding on a 5-nl Ni2+–Sepharose column (ProBond®).
3 K. A. Jungers and S. S. Apte, unpublished data.
Metalloprotease ADAMTS-9
9505
In vitro, the column was washed with 3 column volumes of binding buffer. A gradient of 0–42.5 mM imidazole in binding buffer was used to remove nonspecifically bound molecules from the column. Stepwise elution was done using one-column volume batches of 0–250 mM imidazole in binding buffer. Elution was monitored by Western blotting using antibody 9E10. The majority of protein was determined to elute at 50 mM imidazole. ADAMTS-9-508FLAG-Transfected cells were used to determine the GAG-bearing region, GAG-α, as a result of alternative splicing (8).
Aggrecan monomer was isolated from bovine articular cartilage as previously described (23). Aggrecan (20 μg) was incubated with transfected cells as described above. Neoeptipe Western blot analysis was performed for versican (above), except that the proteolytic cleavage at the Glu1771-Ala1772 bond of aggrecan was detected using anti-Ala1772-Glu-Gly-Gly (AGEG) antisera (24) (provided by Micky Tortorella).
RESULTS
Cloning of ADAMTS9 and ADAMTS20 cDNAs—Our search for novel ADAMTS proteases identified a human expressed sequence tag (GenBank™ accession number AA205581) encoded by IMAGE clone 646675 from neuroepitheliomorph-derivated NT2 cells treated with retinoic acid. The ORF of this expressed sequence tag was homologous to ADAMTS proteases and encoded four TSRs followed by a C-terminal domain containing 10 cysteines that was similar to the C terminus of a polypeptide predicted by the C. elegans F25H18.3 cosmide (C. elegans protein data base Wormpep, www.sanger.ac.uk/Projects/C_elegans/wormpep) and subsequently identified as GON-1. The novel human ORF was designated ADAMTS-9. Completion of the full-length protein coding sequence to the putative start codon required several rounds of rapid amplification of cDNA ends. Together, the cloned cDNA sequences represent an mRNA of 8 kb (Fig. 1a). The 3’ untranslated region in IMAGE clone 646675 contained a consensus polyadenylation signal (AATAAA) 15 nucleotides upstream of the poly(A) tail. The most 5’ clone obtained (TS9-B10) contained 32 bp of the 5’ untranslated region. The putative signal peptide coding sequence was preceded by a methionine codon within a satisfactory Kozak consensus sequence (A at –3 relative to ATG), but there was no upstream, in-frame stop codon.
The search for ADAMTS-9-related proteins led to identification of a polypeptide (Celera hCP1629711) predicted by exons on human chromosome 12. The complete 5733-nl-long ADAMTS-20 ORF was assembled from overlapping cDNA clones (Fig. 1a). The ADAMTS20 mRNA was found in low quantities, routinely requiring 35 cycles of PCR or nested PCR for visualization of the PCR products on a gel. Because of the rarity of ADAMTS20 transcripts as well as the presence of numerous regions that are difficult to PCR-amplify, we have been so far unable to obtain the complete ORF in a single PCR reaction.
Identical Domain Organization and Similar Primary Structure of ADAMTS-9 and ADAMTS-20—ADAMTS-9 and ADAMTS-20 are similar in length, containing 1935 and 1911 amino acids, respectively (Figs. 1b and 2). Each contains a C-terminal array of 14 TSRs (15 TSRs/enzyme) that is interrupted by short “linker” peptides located between TSR-6 and -7 and TSR-8 and -9 that do not have similar sequences. ADAMTS-9 and ADAMTS-20 are very similar to each other, with 48% identity and 64% similarity. The cysteine signatures of individual modules in ADAMTS-9 and ADAMTS-20 are identical to those of most other ADAMTS enzymes, with the exception of the procollagen aminoproteinidases (ADAMTS-2, ADAMTS-3, ADAMTS-14) and ADAMTS-13, which have distinctive prodomains and catalytic domains (12). Each module in ADAMTS-9 and ADAMTS-20 (with one exception, described below) contains an even number of cysteines, suggesting participation in internal disulfide bonds. There are 126 cysteines in mature ADAMTS-9, predicting 63 intrachain disulfide bonds. ADAMTS-20 has a Cys to Tyr substitution in TSR-13 (Fig. 2). Since the substituted Cys is the fourth of six conserved cysteines in TSRs, TSR-13 in ADAMTS-20 may contain two intrachain disulfide bonds instead of three and have an unattached cysteine.
The predicted molecular mass of the full-length enzymes is
Both enzymes contain consensus sites for ADAMTS-9, which is predicted to have a molecular mass of 185,000. The mass will decrease by 3 kDa following signal peptide processing. In addition, both enzymes have a prodomain that is likely to be proteolytically processed prior to or during secretion. ADAMTS-9 contains five consensus furin cleavage sites in its prodomain, whereas ADAMTS-20 contains three (Figs. 1a and 2). Two sites, those corresponding to Arg74 and Arg287 in ADAMTS-9, are conserved with ADAMTS-20. Following processing at the furin recognition sequence closest to the C terminus, mature ADAMTS-9 is predicted to have a molecular mass of 184,000 and mature ADAMTS-20 a molecular mass of 185,000. Both enzymes contain consensus sites for N-linked glycosylation (Asn-X-Ser/Thr, where X is any amino acid except Pro), 9 in ADAMTS-9 and 15 in ADAMTS-20 (Fig. 1b). Five such sites, including three in the unique C-terminal domain, are conserved in ADAMTS-9 and ADAMTS-20. Because of the high likelihood of utilization of these sites, the molecular mass of ADAMTS-9 and ADAMTS-20 will probably be in excess of that predicted (i.e., >185,000). Although there are a number of Ser-Gly or Gly-Ser motifs in both ADAMTS-9 and ADAMTS-20, most are within presumed disulfide-bonded domains and lack the expected sequence context for xylosyltransferase recognition (25, 26). However, one motif in the middle of the ADAMTS-9 spacer domain with the sequence Glu-Tyr-Ser\[sup]330\]-Gly-Ser\[sup]332\]-Glu-Thr-Ala-Val-Glu lies within a sequence context that is compatible with GAG attachment to Ser\[sup]330\] or Ser\[sup]332\]. A similar sequence is present at this location in ADAMTS-20 (Fig. 2). ADAMTS-9 and ADAMTS-20, respectively, contain three and two Cys-Ser-Val-Thr-Cys-Gly (CSVTCG) motifs that are believed to mediate binding to the cell surface molecule CD36 (27, 28) (Fig. 2). In addition, each enzyme contains two BBXB motifs (where B represents basic amino acid and X represents any amino acid) that have been shown to mediate heparin and sulfatide binding (27, 29) (Fig. 2). Neither enzyme contains an Arg-Gly-Asp motif.
**ADAMTS-9 and ADAMTS-20 Do Not Have Identical Zinc-binding Catalytic Site Motifs**—The ADAMTS-9 catalytic site is identical to that of ADAMTS-1 and ADAMTS-15 and very similar to that of ADAMTS-4 (Fig. 3a). The unique feature of ADAMTS-9, ADAMTS-1, and ADAMTS-15 is the presence of a proline residue preceding the third zinc-coordinating histidine (Fig. 3a). The corresponding amino acid is leucine in ADAMTS-4, the next most closely related enzyme. The ADAMTS-20 zinc-binding site is not identical to that of any other ADAMTS but is most closely related to that of ADAMTS-7 and ADAMTS-12 with 4/12 variant amino acids (Fig. 3a). All of the substitutions in the ADAMTS-20 active site relative to ADAMTS-7 and ADAMTS-12 are conservative ones. Alignment and clustering of the published ADAMTS proteases confirm the unique place of ADAMTS-9 and ADAMTS-20 in the ADAMTS family (Fig. 3b) and indicate that they constitute a distinct subfamily of proteases.
**ADAMTS-9 and ADAMTS-20 Are Related to GON-1**—The domain organization and primary sequence of ADAMTS-9 and ADAMTS-20 have a greater similarity to GON-1 than any other mammalian ADAMTS enzyme (Figs. 1b and 2). ADAMTS-9 and ADAMTS-20 are equally related to GON-1 in paired BLAST comparisons. The percentage identity of ADAMTS-9 protein to GON-1 is 33% (that of ADAMTS-20 is 32%), and the percentage similarity (including conservative substitutions) is 46% for both ADAMTS-9 and ADAMTS-20 relative to GON-1. The zinc-binding active site sequence of GON-1 resembles ADAMTS-9 more closely than ADAMTS-20, with just 2 of 14 variant amino acids (Fig. 3a). The conserved C-terminalmost convertase-processing site is at an identical location in ADAMTS-9, ADAMTS-20, and GON-1. The unique C-terminal domain varies slightly in length but nevertheless is highly similar in the three enzymes, including an identical cysteine signature (Fig. 2). TSR-1 is well conserved in these ADAMTS enzymes, but there is less similarity between TSRs 2–15 of...
Metalloprotease ADAMTS-9
ADAMTS-9 and ADAMTS-20 are linked to the carboxyl-terminal domain of GON-1 (amino acids 384–1954) and the carboxyl domain of GON-1 (amino acids 1943–2165). The entire sequence of GON-1 (GenBank™ accession number NP501792) and its corresponding mRNA are not aligned, because it has more TSRs than ADAMTS-9 and ADAMTS-20. Map of GON-1 shows the N terminus of mature ADAMTS-9. Consensus sequences for metalloprotease ADAMTS-9 were indicated by filled asterisks.
Fig. 2. Alignment of primary sequences of ADAMTS-9, ADAMTS-20, and GON-1 with the TSRs of ADAMTS-9 and ADAMTS-20 are boxed. The exposed amino acids are underlined and numbered consecutively. Two linker peptides between TSR-6 and -7 and between TSR-8 and -9 are indicated by filled circles. Cysteine in the carboxyl-terminal domain (but not elsewhere) is indicated by filled asterisks. A Cys→Tyr change in TSR-13 of ADAMTS-9 is demonstrated.
ADAMTS-20 is indicated by the filled circle. Prospective GAG attachment sites are indicated by open asterisks. Exon junctions are indicated by open circles. Sense probe showed no hybridization (Fig. 4, i).
ADAMTS-9 is located near the cell surface but not in conditioned medium—ADAMTS-9-pH101 was detected in lysates of transiently transfected COS-1 and 293 cells as two major anti-FLAG reactive bands migrating at ~180 and ~250 kDa under reducing conditions (Fig. 5a), although the 250-kDa band was inconsistently seen. In addition, a number of smaller FLAG-tagged bands, presumably derived from the full-length ADAMTS-9 were also seen (Fig. 5a, upper panel). Treatment of ADAMTS-9-expressing cells with an increasing concentration of NaCl demonstrated a concentration-dependent release of ADAMTS-9 from the cells (Fig. 5a, lower panel). Due to the unfavorable effects of supraphysiological salt concentrations on cell viability, concentrations higher than 340 mM were not tested.
To identify the cellular or extracellular location of ADAMTS-9 and contrast it with ADAMTS-4, ADAMTS-5, and the ADAMTS-like protein, punctin (21), transiently transfected COS-1 and 293 cells were immunostained with anti-FLAG M2 antibody. ADAMTS-9-transfected cells produced a stronger immunoreactive band than ADAMTS-9-transfected cells. The ADAMTS-9 catalytic domain without the ancillary domains (encoded by ADAMTS-9-pH101) did not process aggrecan or versican at these sites (data not shown).
Intracellular Maturation of ADAMTS-9 Involves N-Glycosylation of the Prodomain and Furin Processing at the Arg287-Phe288 Bond—The predicted molecular masses of signal peptide processed and ADAMTS-9-pH101 that is processed at the consensus proprotein convertase sites are shown in Fig. 6a. Transient expression of ADAMTS-9-pH101FLAG in 293 cells followed by pulse-chase analysis, immunoprecipitation using anti-FLAG M2 antibody, and fluorography identified three major immunoreactive bands in cell lysates with molecular masses of ~66, 56, and 54 kDa, respectively. The relative intensity of these bands varied with the duration of pulse and chase. After a 15-min pulse and 60-min chase, the amount of the 66-kDa protein seen was significantly greater than that seen after a 15-min chase (Fig. 6b). Conversely, the 54–56-kDa doublet was more prominent after a 15-min chase (Fig. 6b). The 66-kDa band intensified substantially after a 135-min chase with very little of the 54–56-kDa doublet being detectable. When cell lysate and culture medium were immunoprecipitated and immunoblotted with anti-FLAG M2 antibody 48 h following transfection of QBI 293A cells, the cells contained the 66-kDa band and essentially no 54–56-kDa doublet (Fig. 6c). When these cell lysates were treated with PNGase F, this 66-kDa band was reduced to a doublet of ~54–56 kDa (Fig. 6c). Collectively, these observations suggest that the 66-kDa band is derived from a 54–56-kDa precursor by N-linked glycosylation. N-Glycosylation of ADAMTS-9-pH101FLAG was confirmed by culture of stably transfected cells in the presence of the tunicamycin A homolog (data not shown). Under the pulse-chase conditions used, no labeled protein could be immunoprecipitated from the conditioned medium (Fig. 6b), and protein corresponding in size to the active form (28 kDa) was not seen in cell lysate. However, in stably transfected cells (not shown) or immunoprecipitation 48 h after transfection, the mature, tagged protein could be detected in culture medium (Fig. 6c). Deglycosylation did not alter the migration of the secreted mature enzyme (Fig. 6c). N-terminal sequencing of the secreted mature ADAMTS-9-pH101FLAG gave the sequence Phe-Ser-Leu-Tyr-Pro-Arg-Phe.
Furin-deficient CHO-RPE 40 cells did not process ADAMTS-9 (Fig. 7a). Processing was rescued by transfection with furin (Fig. 7a). In QBI 293A cells, the Arg287 → Ala, Arg288 → Ala, or Arg289 → Ala mutants did not affect the appearance of...
the mature protein in the medium, but abrogation of the most C-terminal processing site (Arg$^{287}$ → Ala) resulted in failure of processing to the mature form (Fig. 7b). Expression of the Arg$^{287}$ → Ala mutant resulted in anomalous bands of 40 and 45 kDa in conditioned medium in addition to the mature protein (Fig. 7b). Instead of the mature 28-kDa form, expression of the Arg$^{287}$ → Ala mutant resulted in the appearance of 37- and 42-kDa proteins in culture medium whose identity is not known (Fig. 7b).
### DISCUSSION
**Identification of the Full-length Product Of ADAMTS9**—Although a report of the ADAMTS9 mRNA (GenBank™ accession number AF 261918) and ADAMTS9 chromosomal localization was published (17) while our work was in progress, the novel sequence data we report here extend the predicted C terminus of that protein further to include an additional 10 TSRs and the unique C-terminal domain. Our data suggest that the ADAMTS9 transcript presented here encodes the full-
Metalloprotease ADAMTS-9
length, authentic product of this gene for several reasons. First, the previously described ADAMTS9A cDNA diverges from our ADAMTS9 sequence at an unspliced intron (deduced by comparison of the ADAMTS9A and mRNA sequence with the ADAMTS9 genomic sequence). Another ADAMTS9 product predicted by the sequence of the KIAA1233 gene (GenBank™ accession number AB037733) is incomplete at both the amino- and carboxy termini. Comparison of this sequence with the cDNA sequence reported here and the ADAMTS9 genomic sequence suggests the inclusion of an unspliced intron leading to a premature stop codon. Intron inclusion suggests cloning of partially processed pre-RNA, not authentic mRNA. Second, the ADAMTS9-9A transcript does not contain a consensus polyadenylation sequence upstream of the poly(A) tail, in contrast to the ADAMTS9 transcript reported here. Third, Northern analysis demonstrated that probes from the novel sequences we describe here, as well as a probe from the region shared by all transcripts (data not shown), hybridized to the same major 8-kb band on Northern blots, suggesting that the dominant transcript in most tissues encodes the longer form, ADAMTS-9B.
Previous studies of ADAMTS9 had shown widespread expression in fetal and adult tissues by RT-PCR (18). Our studies using Northern blots and quantitative RT-PCR are in agreement with this and provide additional information about the mRNA size. The 8-kb ADAMTS9 mRNA is compatible with the long ORF we have cloned. The identity of the smaller mRNAs found in kidney and ovary is presently unclear. These could represent alternative splice forms or unrelated transcripts that cross-hybridize with the probe. Since ADAMTS20 mRNA is undetectable on Northern blots, both its size and the existence of alternative forms is unknown. In fact, ADAMTS20 transcripts are extremely rare in all of the tissues we have examined, and there are only two human ADAMTS20 expressed sequence tags (AU132653 and BG212007) reported in GenBank™. Nevertheless, a sensitive RNA in situ hybridization approach did demonstrate low levels of expression in epithelial cells of breast and lung origin. The prevalence and biological significance of this low level expression is unknown. Therefore, at the protein level, detailed characterization of the more abundantly expressed enzyme, ADAMTS-9, was subsequently undertaken.
ADAMTS9 and ADAMTS20 Constitute a Distinct Subfamily of ADAMTS Proteases—ADAMTS proteases can be clustered into subfamilies of closely related enzymes on the basis of their domain organization and primary sequences. The procollagen aminopropeptidase subfamily (ADAMTS-2, -3, and -14) represents the most striking example, and other enzymes such as ADAMTS-7 and -12 and ADAMTS-6 and -10 occur in closely related pairs. The ADAMTS-9 and ADAMTS-20 subfamily is particularly interesting, because it is the first such ADAMTS subfamily with a closely related ortholog in invertebrates, indicating, perhaps, a highly conserved physiological role. However, unlike the other ADAMTS subfamilies, ADAMTS-9 and ADAMTS-20 do not have identical zinc-binding active site sequences. Furthermore, their expression patterns are quite different, suggesting they may have nonredundant biological roles.
The genomic organization of ADAMTS9 and ADAMTS20 bears little resemblance to other genes in the family. ADAMTS-1 is encoded by nine exons, and the prodomain, disintegrin-like domain, and central TSR are each encoded by single exons, whose boundaries coincide with the domain boundaries (30). In ADAMTS-1, a single terminal exon encodes the spacer and two C-terminal TSRs (30). This is clearly not the case with ADAMTS9 and ADAMTS20, where few domains other than the TSRs are encoded by single exons. ADAMTS13 (13) has 29 protein-coding exons whose boundaries are different from ADAMTS-1, -9, and -20. The procollagen aminopropeptidases share a different genomic organization (12). Therefore, gene structure may be conserved in ADAMTS subfamilies, but there is not a characteristic gene structure that is shared by the entire family.
The Cys to Tyr substitution in TSR-13 is not an artifact of cloning, because we found it both in the genomic DNA (in Celera and GenBank™ databases), in the cloned cDNA, and in a small number of normal human alleles in which the corresponding exon was subjected to PCR-direct sequencing (data not shown). It may represent a non-synonymous single nucleotide (4715A→G) polymorphism, since TSR-13 in mouse ADAMTS-20 has the typical six-cysteine signature. The preva-
ence and significance of this amino acid change in humans is not presently known and will be investigated further.
Intracellular Maturation Of ADAMTS-9 Involves Glycosylation of the Prodomain and Processing at a Single Proprotein Convertase-processing Site—Following removal of the signal peptide and entrance into the secretory pathway, ADAMTS proteases, like ADAMs and some MMPs, are processed further by one or more proprotein convertases to remove the prodomain and undergo additional post-translation modification such as glycosylation. Proprotein convertases (e.g. furin) are serine proteases present in the Golgi apparatus or at the cell surface that typically cleave immediately following a consensus recognition sequence rich in basic residues (31). Our studies showed that processing did not occur in the absence of furin but could be rescued by transfection of furin, demonstrating that proprotein convertases were essential for pro-ADAMTS-9 maturation.
Our studies suggest that there is rapid glycosylation of the ADAMTS-9 prodomain following synthesis that is essentially complete in about 2 h. There is no N-glycosylation of the catalytic domain, consistent with the observation that the prodomain contains three consensus N-glycosylation sites, whereas the catalytic domain has none. Our data indicated processing of the Arg287–Phe288 peptide bond, whereas none of the other furin sites appear to be used for enzyme maturation. We should emphasize that the Arg280 mutation would abrogate two furin sites, since this residue serves as the P1 Arg for the Arg-Glu-Lys-Arg287 site as well as the P4 residue for the Arg-Thr-His-Arg283 site. We could detect the 28-kDa mature form intracellularly in the wild-type and Arg33Ala mutant. This then accumulates in the medium following secretion through the constitutive secretory pathway. On the other hand, in the Arg287Ala mutant, the precursor is not processed intracellularly and accumulates in the medium along with other unidentified bands. The N terminus of mature ADAMTS-9 determined by amino acid sequencing was in agreement with the location of the N terminus of mature ADAMTS-1, ADAMTS-4, and ADAMTS-13, suggesting that although more than one processing site may be present, the C-terminal furin-processing site is generally used for production of the mature ADAMTS enzymes.
Western blotting of full-length ADAMTS-9 suggested that it undergoes substantial post-translational modification. In keeping with the number of consensus sites for N-linked glycosylation and the large number of serine and threonine residues, glycosylation of full-length ADAMTS-9 has also been noted (data not shown), as is shown in the prodomain. Expression of full-length ADAMTS-9 demonstrated the existence of a number of smaller FLAG-tagged fragments that were presumably derived from it by proteolysis. Regulated processing has been noted in ADAMTS-1 (32), ADAMTS-4 (33), and ADAMTS-12.
**FIG. 6.** a, scheme of the protein encoded by ADAMTS91–508FLAG. The domains included in the expressed proteins and the locations of N-linked sugar attachment (lollipops) and FLAG tag are shown. Below this are the protein species predicted following signal peptidase cleavage or cleavage at each of five consensus furin cleavage sites. The expected molecular mass of each unmodified protein species is shown at the right. b, pulse-chase analysis of ADAMTS91–508FLAG–transfected QBI 293A cells. Cells were pulsed with radiolabeled amino acids and chased for varying times as indicated. Control cells were transfected with empty expression vector. Cell extracts and media were immunoprecipitated with anti-FLAG M2 monoclonal antibody and detected by fluorography. The arrowhead indicates a doublet at 54–56 kDa, and the arrow indicates a major N-glycosylated band at 66 kDa. C, cell lysates; M, medium. Molecular mass markers are shown at left. c, deglycosylation of ADAMTS91–508FLAG by PNGase F. Transiently transfected QBI 293A cell lysates and culture medium were immunoprecipitated with anti-FLAG M2 48 h after transfection. Western blot analysis was done using anti-FLAG-M2. Ig, the immunoglobulin heavy chain. The arrow indicates the mature form in culture medium, and the arrowheads indicate the intracellular zymogen form.
(34) and is a potentially intriguing phenomenon because the released ancillary domains could have interesting biological functions or modify the function of ADAMTS-9 (33). Proteolytic fragments of the native enzyme will be sought in tissues and cells once specific high affinity antibodies are available.
ADAMTS-9 Is Located near the Cell Surface and Is Involved in Versican and Aggrecan Degradation—Neither ADAMTS-9 nor ADAMTS-20 nor any of the other known ADAMTS proteases has a potential transmembrane sequence or a glycosylphosphatidylinositol signal anchor sequence. Therefore, these are not predicted to be membrane-anchored enzymes. Accordingly, studies with various ADAMTS proteases have shown that they are soluble or associated with the ECM (3, 4, 35). ADAMTS-9 and ADAMTS-4 are therefore the first ADAMTS proteases shown to localize near the cell surface, as demonstrated by immunofluorescence microscopy, although their precise location relative to the cell membrane or the binding mechanism is presently unknown. In contrast, both the localization and appearance of ADAMTS-5 distribution are different. Furthermore, although restricted to the ECM, ADAMTS-5 presents a different distribution than punctin, an ADAMTS-like protein comprising only ancillary domains (21). Punctin localization to the cell substratum (21) and the failure of ADAMTS-91–508 or C-terminally truncated ADAMTS-1 (35) to be located in either the ECM or cell surface strongly validates the role of the ancillary domains in anchoring these enzymes near the cell. ADAMTS-9 has consensus sites for binding to heparin (and therefore to heparan sulfate proteoglycans) and CD36, and these may be candidate cell surface and pericellular ECM ligands. In support of this possibility, ADAMTS-9 was released from cells and ECM by gentle washes with low concentrations of salt.
To identify potential substrates for ADAMTS-9, we relied upon comparison of the ADAMTS active site sequences, the phylogenetic profile of the ADAMTS family, and the previous descriptions of their enzymatic activities. The ADAMTS enzymes (ADAMTS-1, ADAMTS-4, and ADAMTS-5) that process the large aggregating proteoglycans versican, aggrecan, and brevican have very similar (although not identical) active site sequences, but they have different domain structures. Because ADAMTS-9 has an active site sequence identical to that of ADAMTS-1 and similar to that of ADAMTS-4, we considered that it might be a proteoglycan core protein-degrading enzyme. Since ADAMTS-9 was not secreted into the culture medium of cells, we used a cell-based ADAMTS assay. Serum-free culture medium has the appropriate pH and salt concentration for ADAMTS activity and, when supplemented with calcium, provided the reaction conditions necessary for the versicanase and aggrecanase assays.
By analogy with aggrecanase-susceptible sites in aggrecan, Sandy et al. (8) had previously predicted two putative ADAMTS cleavage sites in human versican and had prepared polyclonal antisera recognizing one such predicted neoepitope generated by proteolysis of the V1 Glu441–Ala442 bond (8). Versican V0 and V1 forms differ in the inclusion of the GAG-α region that is present in the V0 form but missing in the V1 form. Accordingly, the peptide bond cleaved has a different location in the two forms (8). Consistent with the mixed population of versican made by smooth muscle cell cultures, two bands (70 and 180 kDa corresponding to G1 versican fragments DPEAAE (V0 form) and DPEAAE (V1 form)) were seen in previous studies of ADAMTS-4 processing of versican (8). Of these, the 70-kDa band was considerably stronger, consistent with there being more of the V1 form in the versican preparation (8). In contrast, neither ADAMTS-4 nor ADAMTS-9 proteolysis gave an anti-DPEAAE reactive band at 180 kDa in our experiments.
A refined comparison of ADAMTS-4 and ADAMTS-9 cannot be done in the cell-based assay, since transfection efficiency, expression levels, secretion, and zymogen processing may be different. Purified ADAMTS-9 is not yet available, and given its complex domain structure and cell surface localization, it may be difficult to obtain. We have purified ADAMTS-91–508, but it does not process versican or aggrecan, demonstrating the essential role of the ancillary domains in substrate recognition and/or binding. Therefore, this form cannot be used in kinetic studies to compare with ADAMTS-4. With these limitations,
The procollagen N-propeptidases have identical domain organizations and identical active site sequences (in fact, 70 amino acids around the zinc-binding site are identical in these enzymes) (12). This identity suggests that the structural requirements for procollagen processing are very stringent, and indeed, the catalytic sites of the procollagen N-propeptidases have a distinctive cysteine signature not found in other ADAMTSs. Similarly, ADAMTS-13, the von Willebrand factor protease, is unlike any other ADAMTS in its domain organization and is clearly the major, if not only, von Willebrand factor-processing enzyme (13), suggesting that the structural requirements for this function are stringent as well. In contrast, the four ADAMTS enzymes that degrade proteoglycan core proteins have neither an identical domain organization nor identical active site sequences. This dissimilarity suggests a relatively relaxed structural requirement for proteoglycan processing and supports the likelihood that ADAMTS enzymes may have activity against proteoglycans. In the future, it will be important to define the relative prevalence of each of the proteoglycan-degrading ADAMTS enzymes in different tissues as well as in diseases such as arthritis and to determine their tissue-specific role by targeted inactivation of the corresponding mouse genes. In future studies, it will also be important to ask whether ADAMTS-20 can process versican and aggrecan and to ask whether ADAMTS-9 and ADAMTS-20 have biological roles similar to GON-1.
Acknowledgments—we are grateful to Dr. Satya Yadav for amino acid sequence analysis; Dr. Takahiro Nagase for providing the KIAA0688 (ADAMTS9) cDNA; Dr. John Sandy and Micky Tortorella for amino acid sequence analysis; Dr. Takahiro Nagase for providing the ADAMTS4 KIAA0688 (13), suggesting that the structural requirements for this function are stringent as well. In contrast, the four ADAMTS enzymes that degrade proteoglycan core proteins have neither an identical domain organization nor identical active site sequences. This dissimilarity suggests a relatively relaxed structural requirement for proteoglycan processing and supports the likelihood that ADAMTS enzymes may have activity against proteoglycans. In the future, it will be important to define the relative prevalence of each of the proteoglycan-degrading ADAMTS enzymes in different tissues as well as in diseases such as arthritis and to determine their tissue-specific role by targeted inactivation of the corresponding mouse genes. In future studies, it will also be important to ask whether ADAMTS-20 can process versican and aggrecan and to ask whether ADAMTS-9 and ADAMTS-20 have biological roles similar to GON-1.
REFERENCES
1. Hurskainen, T. L., Hirohata, S., Seldin, M. F., and Apte, S. S. (1999) J. Biol. Chem. 274, 25555–25563
2. Kuno, K., Okada, Y., Kawashima, H., Nakamura, H., Miyasaka, M., Ohno, H., and Matsushima, K. (2000) FEBS Lett. 478, 241–245
3. Sandy, J. D., Flannery, C. R., Neame, P. J., and Lohmander, L. S. (1992) J. Clin. Invest. 89, 1512–1516
4. Lohmander, L. S., Neame, P. J., and Sandy, J. D. (1995) Arthritis Rheum. 36, 1214–1222
5. Sandy, J. D., Westling, J., Kenagy, R. D., Irula-Arispe, M. L., Verscharen, C., Rodriguez-Mazanque, C. J., Zimmermann, D. R., Lemire, J. M., Fischer, J. W., Wight, T. N., and Clowes, A. W. (2001) J. Biol. Chem. 276, 13372–13378
6. Matthews, R. T., Gary, S. C., Zerillo, C., Pratta, M., Solomon, K., Arner, E. C., and Hochfield, S. (2000) J. Biol. Chem. 275, 22695–22703
7. Colige, A. L., Li, S. W., Schwarze, U., Petty, E., Wertelecki, W., Wilcox, W., Krakow, D., Cohn, D. H., Reardon, W., Byers, P. H., Lapiere, C. M., Prokop, D. J., and Nusgens, B. V. (1999) Ann. Hum. Genet. 65, 368–377
8. Sandy, J. D., Sieron, A. L., Li, S. W., Schwarze, U., Petty, E., Wertelecki, W., Wilcox, W., Krakow, D., Cohn, D. H., Reardon, W., Byers, P. H., Lapiere, C. M., Prokop, D. J., and Nusgens, B. V. (2002) J. Biol. Chem. 277, 5756–5766
9. Fernandez, R. J., Hirohata, S., Engle, J. M., Colige, A., Cohn, D. H., Eyre, D. R., and Apte, S. S. (2001) J. Biol. Chem. 276, 31529–31536
10. Colige, A., Sieron, A. L., Li, S. W., Schwarze, U., Petty, E., Wertelecki, W., Wilcox, W., Krakow, D., Cohn, D. H., Reardon, W., Byers, P. H., Lapiere, C. M., Prokop, D. J., and Nusgens, B. V. (2002) J. Biol. Chem. 277, 5756–5766
11. Leibach, U., Eichele, G., Helms, J. A., and Lu, H. C. (1997) in Cellular Methods in Developmental Toxicology (Daston, G. P., ed) pp. 23–48, CRC Press, Inc., Boca Raton, FL
12. Ishikawa, K., Nagase, T., Suyama, M., Miyajima, N., Tanaka, A., Kotsuji, H., Nomura, N., and Obara, O. (1998) DNA Res. 5, 149–176
13. Hirohata, S., Wang, L. W., Miyagi, M., Yan, L., Seldin, M. F., Keene, D. R., Crabb, J. W., and Apte, S. S. (2002) J. Biol. Chem. 277, 23–22
14. Spence, M. J., Sucic, J. F., Foley, B. T., and Mohringer, T. J. (1995) Somat. Cell. Mol. Genet. 21, 1–18
15. Sagdier, S. W., and Hassall, V. C. (1989) J. Biol. Chem. 264, 77–87
16. Tortorella, M. D., Pratta, M., Liu, R. Q., Austin, J., Ross, O. H., Abades, I., Burn, T., and Arner, E. (2000) J. Biol. Chem. 275, 18566–18572
17. Huber, S., Winterhalter, K. H., and Vaughan, L. (1988) J. Biol. Chem. 263, 752–756
18. Kresse, H., Seidler, D. G., Muller, M., Breuer, E., Haussler, R., Roughley, P. J., and Schonherr, E. (2001) J. Biol. Chem. 276, 13411–13416
19. Bornstein, P., and Sage, E. H. (1994) Methods Enzymol. 245, 62–85
20. Aesch, A. S., Silbiger, S., Heimer, E., and Nachman, R. L. (1992) Biochem. Biophys. Res. Commun. 182, 1208–1217
21. Pajouhof, A., E. Fritschery, S., borlat, F. Shaw, J. P., Vilbois, F., Zwahlen, C., Trkola, A., Marchant, D., Clapham, P. R., and Wells, T. N. (2001) J. Biol. Chem. 1208–120626
22. Kuno, K., Inzau, H., Ohno, S., and Matsushima, K. (1997) Genomics 46, 466–471
23. Bergeron, L., Redde, L., and Day, R. (2000) J. Biol. Chem. 275, 13912–13917
24. Rodriguez-Manzaneque, J. C., Milchanowski, A. B., Dubur, E. K., Leude, R., and Irula-Arispe, M. L. (2000) J. Biol. Chem. 275, 33471–33479
25. Gao, G., Westling, J., Thompson, V. P., Howell, T. D., Gottschall, P. E., and Sandy, J. D. (2002) J. Biol. Chem. 277, 11034–11041
26. Cal, S., Arques, J. M., Fernandez, P. L., and Leiper-Otin, C. (2001) J. Biol. Chem. 276, 17392–17394
27. Kuno, K., and Matsushima, K. (1998) J. Biol. Chem. 273, 13912–13917
28. Leeb, L., Aspberg, A., Keene, D. R., Ots, R. N., Reinhardt, D. P., and Sakai, L. Y. (2002) J. Biol. Chem. 277, 4565–4572
29. Olin, A. I., Morgelin, M., Sasaki, T., Tilp, R., Heinegard, D., and Aspberg, A. (2001) J. Biol. Chem. 276, 1255–1261
30. Aspberg, A., Adams, S., Kastrup, G., Tilp, R., and Heinegard, D. (1999) J. Biol. Chem. 274, 20444–20449
31. Perri, R., Perrin-Baunok, D., Pettway, Z., Bronner-Fraser, M., Morgelin, M., and Kimata, K. (1996) FASEB J. 10, 203–301
Characterization of ADAMTS-9 and ADAMTS-20 as a Distinct ADAMTS Subfamily Related to Caenorhabditis elegans GON-1
Robert P. T. Somerville, Jean-Michel Longpre, Katherine A. Jungers, J. Michael Engle, Monique Ross, Stephen Evanko, Thomas N. Wight, Richard Leduc and Suneel S. Apte
J. Biol. Chem. 2003, 278:9503-9513.
doi: 10.1074/jbc.M211009200 originally published online January 3, 2003
Access the most updated version of this article at doi: 10.1074/jbc.M211009200
Alerts:
• When this article is cited
• When a correction for this article is posted
Click here to choose from all of JBC's e-mail alerts
This article cites 38 references, 21 of which can be accessed free at http://www.jbc.org/content/278/11/9503.full.html#ref-list-1 | 2025-03-05T00:00:00 | olmocr | {
"pdf_page_numbers": [
[
0,
5708,
1
],
[
5708,
14168,
2
],
[
14168,
18423,
3
],
[
18423,
22544,
4
],
[
22544,
23678,
5
],
[
23678,
27420,
6
],
[
27420,
28414,
7
],
[
28414,
32993,
8
],
[
32993,
37260,
9
],
[
37260,
41709,
10
],
[
41709,
48553,
11
],
[
48553,
49292,
12
]
]
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.