Developing White Matter Microstructure Networks in Autism Spectrum Disorders

Background: Brain imaging findings in children with autism spectrum disorder (ASD) suggest the disorder is associated with altered brain development and disrupted structural and functional brain “connectivity.” Thus, while white matter microstructure is fundamental to such brain connectivity, developmental disruptions to networks of white matter may provide a core neurobiological feature of ASD. Diffusion tensor imaging (DTI) is an integral neuroimaging technique to assess the white matter microstructure and has been influential to the study of white matter alterations in ASD. However, while vast white matter differences have been reported to exist in ASD, these differences are often reported at the regional level. Thus, information can only be gleaned in regards to the evaluated region of interest. As distinct white matter regions work harmoniously with others, a more informative approach may be to study the patterns and characteristics of white matter networks in ASD.

Objectives: We utilized a data-driven technique to identify underlying white matter networks in a large sample of ASD individuals and subsequently characterized the developmental trajectories of these white matter networks.

Methods: Participants for this study consisted of 100 males with ASD between 3 and 39 years of age. DTI data were acquired from each participant, images were corrected for distortion and head motion and maps of fractional anisotropy (FA), were calculated. Diffusion tensors were spatially aligned to a study-specific template and the resulting transformations were used to bring each individual’s FA map into spatial alignment. Spatially aligned FA maps were subsequently smoothed with a 5mm full-width-at-half-maximum kernel and concatenated into a single 4D image file. Spatial independent component analysis (sICA) was then performed using the MELODIC tool (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/MELODIC), providing spatially independent networks. Trajectories of these networks were examined using mixed-effects models using R.

Results: A total of 57 spatially independent networks were found to exist, though 11 of these were observed to be associated with artifacts (errors resulting from image misregistration or motion). The remaining 46 components were visually assessed to correspond to biologically plausible bilateral and unilateral white matter regions spanning both individual and multiple white matter tracts. Mixed effects modeling of FA from these networks show that FA significantly (p<0.05, Bonferroni corrected) changes across the examined age range, while the timing of age-related changes are distinct among the different networks.

Conclusions: Our results suggest that measures derived from diffusion imaging combined with data-driven analysis techniques, such as sICA, may be used to parcellate the brain into biologically meaningful networks. Future analyses will examine whether the developmental trajectories of these white matter networks relate to the autism severity measures. We will additionally compare the developmental patterns from these distinct networks to those estimated from individuals with typical development.