Fig. 3

Multimodal approaches to mechanism and statistical definition of ASD subpopulations: a white-matter fiber tracking of the thalamocortical auditory radiations, defined using high angular resolution diffusion imaging (HARDI). b Sagittal and axial depiction of the voxel placement for spectrally edited MEGAPRESS MRS, yielding GABA estimates (c). Modeling M50 latency (from Fig. 1) using the acoustic radiations FA along with age in d allowed prediction of M50 latency in TD children (accounting for 52% of the variance) with e white-matter FA a significant predictor of M50 latency (p < 0.0001). The model, although still significant (but with different coefficients), did not perform as well in the ASD cohort, likely due to the heterogeneity of the ASD cohort. However, and in fact addressing the heterogeneity of ASD (f), a subpopulation of extreme M50 latency outliers was identified, as depicted in the histograms of residual latency (deviation from the TD model). This subpopulation was characterized by having significantly lower GABA than the ASD children that conformed to the TD model [1]