Differences in subcortico-cortical interactions identified from connectome and microcircuit models in autism

Abstract

The pathophysiology of autism has been suggested to involve a combination of both macroscale connectome miswiring and microcircuit anomalies. Here, we combine connectome-wide manifold learning with biophysical simulation models to understand associations between global network perturbations and microcircuit dysfunctions in autism. We studied neuroimaging and phenotypic data in 47 individuals with autism and 37 typically developing controls obtained from the Autism Brain Imaging Data Exchange initiative. Our analysis establishes significant differences in structural connectome organization in individuals with autism relative to controls, with strong between-group effects in low-level somatosensory regions and moderate effects in high-level association cortices. Computational models reveal that the degree of macroscale anomalies is related to atypical increases of recurrent excitation/inhibition, as well as subcortical inputs into cortical microcircuits, especially in sensory and motor areas. Transcriptomic association analysis based on postmortem datasets identifies genes expressed in cortical and thalamic areas from childhood to young adulthood. Finally, supervised machine learning finds that the macroscale perturbations are associated with symptom severity scores on the Autism Diagnostic Observation Schedule. Together, our analyses suggest that atypical subcortico-cortical interactions are associated with both microcircuit and macroscale connectome differences in autism.

Inter-individual body mass variations relate to fractionated functional brain hierarchies

“…In addition to MRI-based analyses of regional morphology and microstructure, we performed a transcriptomic association analysis based on post-mortem gene expression maps provided by the Allen Brain Atlas. Although such transcriptomic associations were established through a different and small dataset that is not necessarily representative of the HCP sample, equivalent approaches have been increasingly adopted in neuroimaging research to identify genes whose expression patterns covary with macroscopic findings 45 , 50 , 62 – 66 . In our work, spatial association analyses pointed to specific gene sets, which were cross-referenced with previously reported genome-wide association studies.…”

“…The key to manifold learning is the ability to compress high-dimensional connectomes into a series of lower-dimensional eigenvectors (i.e., gradients) that visualize spatial trends in inter-regional connectivity variations 39 , simplifying connectivity analysis and visualization. Eigenvectors estimated from resting-state functional MRI (rs-fMRI), myelin-sensitive imaging, and diffusion MRI can serve as axes of the brain’s intrinsic coordinate system 39 – 45 . These eigenvectors have been shown to follow established models of neural hierarchy and laminar differentiation 46 .…”

“…Coupled with gene set enrichment analyses 57 – 61 , these approaches can discover molecular, developmental, and disease-related processes, and provide additional context for MRI findings. Recent studies utilized transcriptomic decoding to explore the underpinnings of brain imaging findings in both healthy and diseased cohorts 45 , 50 , 62 – 66 .…”

Subgroups of Eating Behavior Traits Independent of Obesity Defined Using Functional Connectivity and Feature Representation Learning

“…Machine learning is a powerful tool for analyzing neuroimaging data, and manifold learning techniques are increasingly being used to describe macroscale functional connectome organization along the cortex [27][28][29]52]. We generated a series of low-dimensional eigenvectors, and the spatial patterns agreed with those of existing studies based on the Human Connectome Project data [22,31,52].…”

Adolescent development of multiscale structural wiring and functional interactions in the human connectome

“…It is controlled by two parameters α and t, where α controls the influence of the density of sampling points on the manifold (α = 0, maximal influence; α = 1, no influence) and t controls the scale of eigenvalues of the diffusion operator. We set α = 0.5 and t = 0 to retain the global relations between data points in the embedded space, following prior applications [17,20,28,41,46,98,99]. Cortical regions with more similar inter-regional patterns are more proximal in this new structural manifold.…”