Anatomical imbalance between cortical networks in autism

Influential psychological models of autism spectrum disorder (ASD) have proposed that this prevalent developmental disorder results from impairment of global (integrative) information processing and overload of local (sensory) information. However, little neuroanatomical evidence consistent with this account has been reported. Here, we examined relative grey matter volumes (rGMVs) between three cortical networks, how they changed with age, and their relationship with core symptomatology. Using public neuroimaging data of high-functioning ASD males and age-/sex-/IQ-matched controls, we first identified age-associated atypical increases in rGMVs of the regions of two sensory systems (auditory and visual networks), and an age-related aberrant decrease in rGMV of a task-control system (fronto-parietal network, FPN) in ASD children. While the enlarged rGMV of the auditory network in ASD adults was associated with the severity of autistic socio-communicational core symptom, that of the visual network was instead correlated with the severity of restricted and repetitive behaviours in ASD. Notably, the atypically decreased rGMV of FPN predicted both of the two core symptoms. These findings suggest that disproportionate undergrowth of a task-control system (FPN) may be a common anatomical basis for the two ASD core symptoms, and relative overgrowth of the two different sensory systems selectively compounds the distinct symptoms.

The correlations seen between rGMVs and severity of socio-communicational impairments (Fig. 4) were qualitatively reproduced even when the correlations were estimated separately using ADIR Social and ADIR Communication scores (Supplementary Table 4).
Supplementary Fig. 4 Results of cortical thickness analysis.
Given that smaller CT values are relevant to more optimised and finer information processing 1,2 , results of this cortical thickness analysis were qualitatively consistent with the original observations based on GMV.
a. Relative CT in auditory and visual network regions were conversely correlated with age in autistic children, whereas they showed positive correlations in TD children. The opposite pattern was observed in FPN regions. Although some of these correlations did not survive statistical corrections for multiple comparisons, these results suggest over-optimisation of information processing in auditory and visual networks in autistic individuals, and relatively less-optimisation of FPN functions. b. Difference in relative CT between ASD and TD adults was consistent with such age-related changes seen in the data from children. In auditory and visual network regions, the relative CT values were significantly smaller in the ASD group than in the TD group, whereas the value of FPN regions was larger in the ASD group. * indicates P Bonferroni < 0.05 in two-sample t-tests.
c. The atypical values of relative CT were correlated with autistic core symptoms. In auditory network regions, atypically smaller values of relative CT (i.e., over-optimisation of its function) indicated more severe socio-communicational symptom, whereas those in visual network regions were specifically related to severity of RRB symptom. In contrast, larger values of relative CT in FPN regions (i.e., less optimised information processing) were associated with severity of both core symptoms.

Cortical thickness analysis
To validate our rGMV findings, we conducted the same analysis using preprocessed voxelwise cortical thickness data recorded from the same individuals in Table 1. The preprocessed data were obtained from the ABIDE section in the Preprocessed Connectomes Project (http://preprocessed-connectomes-project.github.io/abide/). Among the 269 individuals in Table 1, cortical thickness data of four ASD children and two TD children were not found in the repository, and therefore, we analysed data of remaining 85 ASD children, 34 ASD adults, 94 TD children, and 50 TD adults. There was no significant difference in demographic data between ASD and TD groups (P > 0.1). According to the data repository, a cortical thickness value was calculated for each grey matter voxel by applying an automated preprocessing and analysis pipeline implemented in Advanced Normalization Tool 3 to a T1-weighted image of each individual.
To extract an average cortical thickness value for different anatomical structures, we first randomly divided a conventional grey matter parcellation map (here, AAL parcellation) to 1024 segments with similar numbers of continuous voxels using a random parcellation algorithm 4,5 . Next, we classified the grey matter segments into nine large-scale cortical brain networks (Fig. 1a): each segment was given one network label when ≥50% of the segment overlaps a specific network area which was defined as a collection of multiple 4mm-radius spheres around the network-specific coordinates 6,7 . Segments showing no sufficient overlap with any network were excluded in the following analyses.
By applying this grey-matter parcellation mask and network labelling to the voxel-wise cortical thickness data, we calculated the ratios of average cortical thickness values between the nine networks in the same manner used for calculation of rGMVs. We then performed the same analysis for these relative cortical thickness values as we did for the rGMVs.