Human brain structural and functional organizations have features of complex networks, such as modularity, scale-invariance, small-world network topology, and the presence of highly connected hubs. Recent methodological advances in sociology, mathematics and statistical physics have provided a variety of graph metrics for the description of complex network architectures. In parallel, recent neuroimaging evidence suggests that the brain architecture of people with autism is characterized by aberrant (both hyper- and hypo-) connectivity. However, to our knowledge, no study to date has directly examined the network organization of the autistic brain within the context of complex network science.
To characterize the complex network properties of the intrinsic functional organization of the brain in people with autism spectrum conditions (ASC), and to compare them to those of the neurotypical brain networks.
30 right-handed adult males aged 18-45 with a clinical and ADI-R confirmed diagnosis of an ASC (ASC group), and 33 age-, sex-, handedness- and IQ-matched neurotypical adults (NT group) were scanned in a 3T MRI scanner by echo planar imaging in an eye-closed, awake, non-task resting state. Following motion and slice timing correction, the 4D images were linearly registered to a study-specific template (averaged by the high-resolution structural images of the whole 63 participants). The cortical and subcortical areas were parcellated into 118 regions (i.e., network “nodes”) according to the Harvard-Oxford cortical and subcortical atlas. Averaged timeseries for each node were extracted and filtered by a wavelet decomposition to obtain the scale 4 low-frequency (0.024 - 0.048 Hz) correlation coefficients. Connectivity strength was first calculated for each node. After thresholding in the cost-efficient small-world regime (proportion of all possible connections 9-48%), network properties were described at both global and nodal levels using global efficiency, clustering coefficient, local efficiency, hierarchy, assortativity, and nodal degree and betweenness-centrality. Group comparisons were performed using permutation testing.
Brain networks for both groups showed small-world topology. At the global level, there were no group differences in network metrics except that the ASC group showed a trend of lower mean clustering coefficient (1-tailed p=0.067) and local efficiency (p=0.086). However at the nodal level, bilateral medial orbito-frontal and several temporal regions showed lower connectivity strength in the ASC group. The ASC networks had lower global efficiency and nodal degree in left posterior parahippocampal and bilateral anterior temporal areas, but higher in right basal ganglia (particularly nucleus accumbens) and posterior supramarginal gyrus. ASC networks also had lower clustering coefficient and local efficiency at bilateral postcentral and right anterior middle temporal gyri. Moreover, although both networks characterized several overlapping hub regions (e.g. right precuneus and posterior parahippocampal gyrus), the ASC networks had significantly lower betweenness-centrality at left posterior parahippocampal and middle cingulate cortices.
The autistic brain showed a comparable small-world network organization to the neurotypical brain networks, while having slightly more random network characteristics. Whereas most global network properties were conserved across both groups, regional properties showed significant group differences localized to various temporal, subcortical and mid-line regions.