A Study of Cortical Functional Network Hierarchy: A Novel Perspective on Atypical Connectivity of Autism

Background:

Together with its parallel and modular architecture, network hierarchy has been widely recognized as key principle of human brain organization. This overarching system is thought to facilitate abstract, high-order cognitive functions by helping segregate information that reflects sensory and motor interactions with the immediate environment from more self-generated operations emerging in transmodal, integrative cortices. As ASD is linked to deficits both in sensory processing and high-level cognitive functions such as theory of mind, analysis of macroscale brain network hierarchy may provide a parsimonious account to consolidate its diverse symptoms.

Objectives:

I will present novel findings from resting-state functional MRI (rs-fMRI) connectome analysis, harnessing recent analytical approaches to characterize hierarchical segregation between higher order transmodal areas and lower order systems in a multi-site cohort of individuals with autism and typically-developing controls

Methods:

We studied 103 males with ASD and 108 neurotypical males from three sites of the ABIDE dataset. Rs-fMRI time series were mapped to cortical surfaces in subject-specific space, followed by time-series cross-correlation of all surface-points. Nonlinear connectome compression techniques identified principal gradients of spatial variations in connectivity across the cortical mantle in individual subjects. Following Procrustes alignment, we used surface-based linear models to compare gradient scores in ASD to controls, controlling for site, age, and multiple comparisons. We furthermore carried out a systematic step-wise functional connectivity (SFC) analyses initiated in primary sensory seeds (primary visual, auditory, somatosensory cortex). Supervised pattern learning leveraged hierarchy features to predict symptom severity as indexed by the Autism Diagnostic Observation Schedule (ADOS) within a 5-fold cross-validation setting.

Results:

Compared to typically-developing controls, ASD showed contracted rs-fMRI connectome gradients, indicative of atypical hierarchical segregation, between higher-order transmodal areas and lower order systems. While SFC patterns in controls recapitulated gradient-based findings, showing a selective convergence of sensory-initiated connectivity in default mode core hubs, ASD presented with distorted patterns and a failure to converge in the default mode network. Effects were reproducible across included sites, in an independent validation dataset (n=110; 53 ASD, 57 neurotypical individuals), and with respect to several rs-fMRI processing choices. Using supervised pattern learning, we could furthermore show that imaging features capturing hierarchical network imbalances predict symptom severity in individuals with ASD, supporting the behavioral relevance of our findings.

Conclusions:

Our study shows that ASD is characterized by perturbations in macroscale cortical hierarchy, which may consolidate its seemingly paradoxical combination of low-level and higher-order symptoms.