Brain connectivity is extensively used to assess interactions between different regions. However, our understanding of the brain is organized around properties of regions and not the connections between them. Given that connectivity contains invaluable information not available through fMRI activation studies, attaining region-specific information from connectivity data could progress our understanding of the neural circuitry and associated brain processes. We also recognize that large effects in any system are often caused by few sources, and that identifying and fixing such source(s) could naturally rectify rest of the network abnormalities in a given disorder. Hence, we developed a novel framework to identify such affected pathological foci from directional brain networks.

In this study, we used effective connectivity (EC) modeling [1] to infer directional interactions. We constructed networks using strength and temporal variability of EC, obtained from resting-state fMRI. These whole-brain connectivities were then used in a probabilistic framework to identify affected disorder foci. We tested our framework on data obtained from Soldiers with posttraumatic stress disorder (PTSD) and mild traumatic brain injury (mTBI). We hypothesized that PTSD with and without mTBI are characterized by certain affected foci, which are associated with connections having altered strength and lower variability of directional brain connectivity.

Eighty-seven U.S. Army Soldiers were recruited for the study, with 17 having PTSD, 42 having both PTSD and mTBI (PTSD+mTBI) and 28 matched combat controls. Resting-state fMRI was acquired in a 3T Verio Siemens scanner, with TR=600ms, TE=30ms, voxel-size=3×3×5mm3, 1000 volumes and 2 sessions. Standard pre-processing was performed with realignment, normalization to MNI-space, detrending and regressing out white-matter, CSF, six-head motion parameters and global-mean signal. Hemodynamic deconvolution was performed [2] on voxel-wise timeseries to obtain underlying latent neuronal variables. Mean timeseries from functionally homogenous brain regions were obtained (cc200-template [3]).

Static EC (SEC) was evaluated (whole-brain) using Granger causality [1]. The time-varying version of EC (dynamic EC [DEC]) was evaluated by employing Kalman-filter based time-varying Granger causality [4]. Variance of DEC was computed as the measure of variability of connectivity. We used a Bayesian probabilistic model to identify disorder foci from connectivity data [5], which assumes that salient regions are associated with large number of abnormal connections. This method has been demonstrated earlier with static functional connectivity (FC) [5], but it made certain assumptions which were suited for FC. Here we demonstrate the utility of this method with static as well as dynamic EC, albeit with certain modifications in the model formulation given that EC matrices are not symmetric, unlike FC, and that the distributions of FC and EC data are dissimilar.

One thousand iterations were performed and statistical significance of foci were determined. The method also gave affected connectivities associated with the foci. Significantly altered foci and their associated altered connections were obtained (p<0.05, FDR-corrected, controlled for age, race, education and head motion).

We identified three foci (left middle frontal gyrus [MFG], left anterior insula and right hippocampal formation) affected in PTSD and PTSD+mTBI, and they were connected to/from other affected brain regions (see Fig.1). Fig.2 shows networks associated with each focus. Further clarity was obtained by reorganizing them into three functionally separable networks (Fig.3): (i) frontal top-down under-modulation network steered by MFG, causing disinhibition of insula, (ii) insula-amygdala-hippocampal loop which goes into a positive-feedback overdrive due to frontal disinhibition, and (iii) hippocampal memory-retrieval network which results in overdrive of association areas involving memory processing/retrieval. Taken collectively, we identified the MFG to be the pivotal source of disruption in Soldiers with PTSD and PTSD+mTBI (Fig.4), which was further affecting other emotion and memory processes, potentially exacerbating symptoms. The other two foci also play a key role in mediating disruption with the insula involved in subjective cognitive-emotional processing, and hippocampus involved in declarative memories. In concert, this network provides a mechanistic explanation of emotion dysregulation and subsequent lack of control over traumatic memories, contributing to hyperarousal, flashbacks and other symptoms observed in soldiers with PTSD.

Earlier studies have repeatedly identified these and other regions [6, 7, 8] to be involved in both PTSD and mTBI, but a precise understanding of the sources of disruption and their subsequent causal relationships has not emerged from them. Employing a novel framework involving foci identification and EC networks, we identified the regional foci associated with the disorders and elucidated their causal relationships. Our characterization fits well with behavioral manifestations of PTSD and PTSD+mTBI, thus illustrating the utility and fidelity of our approach. Future works could take advantage of this approach in identifying sources of disruption/alteration in various psychiatric illnesses and cognitive domains.