Dynamics of Segregation and Integration in Directional Brain Networks: Illustration in Soldiers With PTSD and Neurotrauma

FIGURE 1. Illustration of our hypothesis showing decreasing temporal variability (implying rigidity) of segregation/integration, and either increasing or decreasing segregation/integration (implying inflation or deflation, respectively) as we move from Control to PTSD to PCS + PTSD. Font and circle sizes are symbolic of the increasing/decreasing trend, with smaller circles/text representing deflation and rigidity, and larger circles/text representing inflation and flexibility. (A) Hypothesis-1: some network properties would be disrupted only in PTSD (significant for Control vs. PTSD and Control vs PCS + PTSD comparisons, but not PTSD vs. PCS + PTSD comparison). (B) Hypothesis-2: some network properties would be significantly different between all three groups. Note that inflation and deflation generally correspond to elevation or suppression of network properties, respectively, and not just connection strengths of individual paths. However, in the special case when local network properties of the paths are considered, inflation and deflation are referred to the network properties, as well as connection strengths of the paths under consideration.

FIGURE 2. Using two time series from experimental fMRI data, we illustrate the importance of performing hemodynamic deconvolution for effective connectivity modeling. The latent neural signals are convolved with the hemodynamic response function (HRF) to provide the BOLD fMRI time series. Within-subject spatial HRF variability across the brain could often result in a scenario wherein (A) the latent neural signals have true low directional connectivity [quantified using Granger causality (GC) from blue to pink signal] while the BOLD fMRI time series show high GC wherein the pink time series seems to follow after the blue time series, and (B) the latent neural variables have true high directional connectivity while the BOLD fMRI time series show low GC. In the former case, while the neural signals nearly overlap, the delay in the HRFs causes an observable delay in the BOLD time series, resulting in high GC from the blue to the pink BOLD time series. In the latter case, the delay noticeable in the neural signals (pink signal leads blue) is negated by the delay in the HRFs (blue signal leads pink), resulting in nearly overlapping BOLD time series and a low GC value.

FIGURE 3. Illustration of static and dynamic effective connectivity (SEC and DEC) from a neuroimaging standpoint using two experimental fMRI time series. In (A), the two time series seem highly correlated and nearly overlapping. However, the variations in the pink time series do not appear to happen after (or before) the variations in blue time series (top-left figure). This poor causal relationship results in a low SEC value (= 0.07). Correspondingly, DEC values hover around the zero-value (bottom-left figure) since a causal relationship does not seem to emerge for most part of time, except for a brief span (marked by the arrow) when there is a visible causal relationship. In (B), the pink time series seems to constantly follow after the blue time series (top-right figure), indicating that the pink signal’s associated brain region activates (and deactivates) immediately after the blue signal’s region activates (and deactivates), thus a causal relationship and a high SEC value (= 0.95). DEC provides additional insight (bottom-right figure), wherein steady causality is maintained almost for the entire duration except for a brief span (marked by the arrow), wherein DEC dips because of…

FIGURE 4. Schematic of the complete processing pipeline executed in this study. Results corresponding to different analyses are summarized, which can be viewed in more detail by referring to the corresponding figures.

FIGURE 5. Group differences (p-values) for the two global measures (transitivity and global efficiency) obtained from both SEC and DEC matrices. We observe that whole-brain differences were driven by PTSD, while mTBI likely did not cause whole-brain-level changes.

FIGURE 6. Brain regions with altered functional segregation and/or integration across groups. (A) Regions-of-interest (ROIs) associated with significantly disrupted functional segregation. (B) ROIs associated with significantly disrupted functional integration. The ROIs were defined by the Craddock-200 atlas (Craddock et al., 2012). MFG, middle frontal gyrus; TPJ, temporo-parietal junction.

FIGURE 7. Functional segregation/integration (metastability) results. (A) The network of integration was broken down into two sub-networks. (A) Shows the first of the two sub-networks, the fronto-visual sub-network that exhibited lower strength of integration and lower temporal variation of integration, which was significant for control vs. PTSD and control vs. PCS + PTSD comparisons (but not PTSD vs. PCS + PTSD comparison). The yellow node had altered segregation between all three groups, while the red nodes were different except for the PTSD vs. PCS + PTSD comparison. This sub-network likely represents reduced prefrontal inhibition of visual memory processing and retrieval. (B) Second of the two sub-networks, the parietal inflation sub-network that exhibited altered strength of integration and lower temporal variation of integration. Yellow paths were significantly different for all group-wise comparisons. Green paths were altered except for the PTSD vs. PCS + PTSD comparison. This sub-network showed parietal-inflation caused by subcortical and visual network disruptions, which were in-turn driven by the left middle frontal gyrus (MFG). (C) The entire network of disruption found in the…

FIGURE 8. Schematic of the entire network: Yellow nodes/paths were significantly different for all three group-wise comparisons. Green paths (and red nodes) were significantly different except for the PTSD vs. PCS + PTSD comparison. Thick lines correspond to connections between major sub-networks while thin lines correspond to connections within sub-networks. The prefrontal sub-network consisted of MFG and medial frontal, the parietal sub-network consisted of TPJ and precuneus, the visual sub-network consisted of lingual and primary visual areas while the emotion-memory sub-network consisted of sub-cortical regions such as amygdala and hippocampus and cortical regions such as the insula. Disrupted left-MFG causes deflation of emotion-memory regions and visual memory-related regions, culminating in parietal-inflation causing heightened symptoms often observed in PTSD and PCS.

FIGURE 9. Association and prediction results. (A) Partial least squares regression maps independent (all included network measures combined) and dependent (all neurobehaviors combined) variables into a latent space to find an aggregate relationship between them. The regression displayed in the figure is performed in the latent space, which contains categorical variables representing all included network measures of the four integration paths (MFG → Insula, Insula → Amygdala, Amygdala → Hippocampus, Hippocampus → Precuneus) and two functional segregation nodes (MFG, Insula) (both of which fit our hypothesis), and the neurobehaviors in latent space. Their association in the latent space could be considered as the net association of all these network measures with all the neurobehaviors. (A) Shows the linear fit in latent space (R2 = 0.56, R = 0.75, p = 3.5 × 10–32). (B) Machine learning classification result, which was obtained through recursive cluster elimination based support vector machine (RCE-SVM) classifier, to classify between control, PTSD and PCS + PTSD groups. (B) Shows worst-case classification accuracies obtained using recursively reducing number of discriminative features…

FIGURE 10. Flowchart illustrating an integrated model of connectivity and network-level aberrations in PTSD and mTBI. Paths with thin gray lines correspond to lower strength of network properties and connectivity (SEC) and lower temporal variability of network properties (i.e., rigidity) and connectivity (vDEC) in the clinical groups compared to healthy controls, indicative of breakdown in prefrontal top-down modulation. Paths with thick brown lines correspond to higher strength and lower temporal variability (i.e., rigidity) of network properties and connectivity, indicative of inflation in subcortical limbic and parietal memory-related regions. Yellow paths were significantly different between all the three groups, while the green paths were significantly different only for the control vs. PTSD and control vs. PCS + PTSD comparisons.