Functional Brain Network Mechanism of Hypersensitivity in Chronic Pain

Abstract

Fibromyalgia (FM) is a chronic widespread pain condition characterized by augmented multi-modal sensory sensitivity. Although the mechanisms underlying this sensitivity are thought to involve an imbalance in excitatory and inhibitory activity throughout the brain, the underlying neural network properties associated with hypersensitivity to pain stimuli are largely unknown. In network science, explosive synchronization (ES) was introduced as a mechanism of hypersensitivity in diverse biological and physical systems that display explosive and global propagations with small perturbations. We hypothesized that ES may also be a mechanism of the hypersensitivity in FM brains. To test this hypothesis, we analyzed resting state electroencephalogram (EEG) of 10 FM patients. First, we examined theoretically well-known ES conditions within functional brain networks reconstructed from EEG, then tested whether a brain network model with ES conditions identified in the EEG data is sensitive to an external perturbation. We demonstrate for the first time that the FM brain displays characteristics of ES conditions, and that these factors significantly correlate with chronic pain intensity. The simulation data support the conclusion that networks with ES conditions are more sensitive to perturbation compared to non-ES network. The model and empirical data analysis provide convergent evidence that ES may be a network mechanism of FM hypersensitivity.

Conflict of interest statement

R.E.H. has received research support and consulting fees from Pfizer. D.J.C. has received consulting fees from Pfizer, Cerephex, Tonix, Abbott, Aptinyx, Daiichi Sankyo, Samumed, Zynerba, Astellas Pharma, Williams & Connolly LLP and Therevance and research support from Pfizer, Aptinyx and Cerephex.

Figures

Figure 1

Schematic diagram of study design. The study was composed of two parts, an experimental analysis section and a mathematical modeling section. The experimental phase was carried out by recording 64-channel electroencephalogram (EEG) of fibromyalgia (FM) patients. A functional network was constructed with the weighted phase lag index (WPLI) of the EEG signal, and power spectral density analysis was performed to obtain a node degree and frequency for each EEG channel. We then confirmed the relationship between the three ES conditions and the intensity of chronic pain. For the mathematical phase of the study, we generated a human brain network model based on the Kuramoto model and diffusion tensor imaging (DTI). The frequency configurations with ES and non-ES conditions were considered as the brain network states of high and low pain scores, respectively. These were used to explore the network sensitivity of ES and non-ES networks as tested with a frequency perturbation near the critical point.

Figure 2

Node degree and frequency are positively correlated with chronic pain intensity. (a) The degree-frequency correlation coefficient positively correlates with the pain intensity. Each marker represents an individual. (b) The relationship between node degrees and median frequencies for all EEG channels from two exemplary subjects with low (1) and high (77) pain intensities. The individual with a high pain score displayed a positive relationship between node degree and median frequency of EEG channels, whereas no correlation is observed for the individual with low pain. Each circle and square represents an EEG channel.

Figure 3

Frequency difference and frequency assortativity are correlated with chronic pain intensity. (a) Frequency difference and (b) frequency disassortativity (negative assortativity) among EEG channels correlate with the pain scores. Each marker represents an FM patient.

Figure 4

The relationships of regional node degree (Left) and median frequency (Right) with pain intensity. The correlation coefficients between z-values of node degrees (median frequencies) and pain scores for the ten FM patients and each EEG channel. The colour bar represents Spearman correlation coefficient from −1 to 1.

Figure 5

Network configurations of ES and non-ES conditions. (a) A frequency configuration in the human brain network consisting of 82 brain regions. The asymmetric frequency distributions of the top 15 hub nodes are denoted with different colors. Each circle corresponds to a brain region, and the dark/gray lines are the connections among the regions. 82 regions are clustered into 10 larger brain regions, denoted in Supplementary Table S1. (b) The relationships between node degrees and frequencies for the ES and non-ES conditions. The ES condition has a V-shape relationship, in which the hub nodes have large and small frequencies suppressing synchronization until a critical point (the key mechanism of ES), whereas the non-ES condition has a random relationship. (c) The brain network model with ES conditions has positive degree-frequency correlation coefficient, larger frequency difference, and negative frequency assortativity compared to non-ES condition. Red solid lines throughout the panels are medians of 100 frequency configurations.

Figure 6

The comparison of the human brain networks of ES and non-ES conditions after frequency perturbations in hubs nodes (insula, precuneus, superior frontal cortices, parietal cortices with thalamus). (a) Median synchronization level Rlinked vs. coupling strength λ of networks for ES and non-ES conditions before and after the perturbation. 100 different frequency configurations were implemented for the median Rlinked. The ES condition shows a steeper synchronization, and the perturbation induces larger synchronization. (b) Brain network susceptibility Cp vs. coupling strength λ for ES and non-ES conditions before and after the perturbation. ES conditions have larger Cp than non-ES conditions. The perturbation induces larger synchronization, which is measured as reduced Cp and with larger alternation of Cp in ES. The thick lines and shaded region indicate median Cp and standard error over 100 configurations. (c) The brain network containing the ES conditions shows significantly larger network sensitivity Δ(Cp) (p < 0.05), despite the large variances. The red line denotes the median Δ(Cp) over 100 configurations. (d) The variance of Δ(Cp) indicates the dependency of Δ(Cp) on the network configurations within each ES and non-ES condition. The blue (yellow) bar illustrates the number of network configurations with the same Δ(Cp). The blue (yellow) thick line indicates a fitted line for the distributions of Δ(Cp).