Connectivity and Network Efficiency in ASD

Background: There is now a consensus that in autism spectrum disorder (ASD) there is reduced long-distance connectivity (Just et al 2007).  The status of short-distance connectivity in ASD is less clear.  It has been hypothesized that long-distance under-connectivity is matched with an increase in short-distance connectivity (Belmonte et al 2004), but there is a paucity of evidence to support that speculation.  Further, there is a lack of understanding of how abnormalities in connectivity interact with each other within the overall network.

Objectives:  We investigate connectivity in ASD, for all sets of cortico-cortical connections, and the structure and efficiency of the overall networks, to determine whether altered connectivity in long- and short-distance connectivity are compensatory at the network level. 

Methods: T1-weighted (t1w) and diffusion-weighted (dw) scans were collected from 22 male adults with ASD (34.14 ± 10.67 years) and 22 male controls (31.68 ± 8.75 years). CIVET, a fully automated structural image analysis pipeline, was used to construct white-matter surfaces for each subject, and to segment subcortical regions (Collins et al 1995, Kim et al 2005). Seed masks, stop masks, and target masks were then constructed for use with FSL’s probtrackx. Seed masks were white-matter; stop masks were the boundary of white-matter; and target masks were the white-matter surface. The dw scans were corrected for distortions using fieldmaps, and affine registered to the t1w volumes using FSL’s flirt. The dw volumes were then preprocessed with FSL’s bedpostx, and probabilistic tractography was seeded from 10000 random locations within each voxel of the seed masks; once to generate the number of tracts connecting voxels in the target mask, and a second time to generate the length of those connections. The AAL atlas was then overlayed on the target masks, and the connectivity between cortical regions calculated, as well as the mean length of those connections.  Measures of local, global, and small-world efficiency, as defined by Latora and Marchiori (2001, 2003), were computed for each subject, and group differences in efficiency were assessed, as well as group differences in connectivity between all pairs of AAL regions.

Results: Group comparisons of connectivity for all pairs of AAL regions showed a mix of under- and over-connectivity in ASD, but predominately under-connectivity, both in long- and short-distance connections.  The ASD group showed strongly reduced local efficiency, which was broadly distributed, an increase in global efficiency, and a non-significant reduction in small-world efficiency.

Conclusions: Connectivity abnormalities in ASD represent a shift from an organization with strong local connectivity and sufficient long-distance connectivity to support efficient communication between these local modules, to an organization which compensates for reduced long-distance connectivity by sacrificing local structure.  This is not a shift toward greater short-distance connectivity, but rather toward a randomly connected network.