Abnormalities in Large-Scale Brain Network Architecture in Autism

Background: Autism is a complex neurological condition characterized by childhood onset of dysfunction in multiple cognitive domains including socio-emotional function, speech and language, and processing of internally- versus externally-directed stimuli. Accumulating evidence suggests that autism is a network-based disease, and that abnormalities in brain network structure underlie the abnormal brain function at the core of the disorder. However, large-scale brain network structure has yet to be fully characterized in autism.

Objectives: Using an emerging technique known as structural covariance MRI (scMRI), this study sought to determine whether specific abnormalities in large-scale brain network organization are associated with autism, and whether network-level abnormalities in brain architecture can be detected with standard clinical MRI.

Methods:   We used scMRI to interrogate network-level differences in gray matter structure within eight canonical large-scale ‘intrinsic connectivity networks’ (ICNs) strongly implicated in autism, in 49 high-functioning autistic subjects and age-, gender-, and IQ-matched controls (mean age 13.4 yrs, range 3.5-22.5 yrs, all male). T1-weighted 3.0 Tesla anatomical MRI scans were realigned, segmented, normalized to a customized template, modulated, and smoothed. To study network structural covariance, we derived 4-mm radius spherical seed regions-of-interest (ROIs) within core hubs of canonical ICNs. Extracted mean ROI gray matter intensities provided covariates-of-interest for whole brain condition (diagnosis)-by-covariate analyses based on the General Linear Model. Resulting seed covariance maps for each age group were thresholded at p <0.001, corrected for family-wise error. Direct between-group comparisons were performed using ROI, diagnosis group, and neuropsychiatric test scores as covariates of interest.

Results: Seed-based scMRI revealed specific perturbations in brain network architecture within distinct ICNs, consistent with phenotypic manifestations of autism. Structural covariance maps in controls were consistent with canonical ICN topologies. Extent and topology of the salience network, involved in social-emotional regulation of environmental stimuli, is markedly underdeveloped in autism. In contrast, the default mode network (DMN) is larger in autism, but demonstrates 'posteriorization'. Moreover, discrete nodes outside of canonical DMN boundaries are present in the autism group, including many regions commonly associated with autism. Other networks demonstrate concurrent over- and under-development, regional decoupling, or remain unaffected.

Conclusions:   Specific abnormalities in large-scale brain network structure underlie autism. Our findings are consistent with a network-based ‘selective vulnerability’ model of autism, provide a plausible substrate for phenotypic features of the disorder, and suggest a unifying interpretation of previous work. Structural brain network abnormalities in autism are quantifiable using standard clinical MRI.