Participant Demographics, Diagnostic Group, and Global Brain Measures
There were no significant between-group differences in participants’ age. However, groups differed significantly in gender distribution (χ²(3)=2.26, p=0.016), with a lower percentage of females in the idiopathic ASD group relative to the other subgroups, and in full-scale IQ (F(3)=21.01, p<0.001), with TD controls scoring higher than all other groups, and individuals with 22q11.2DS having a lower IQ than individuals with idiopathic ASD. Further, we found a significant effect of group for total brain volume (F(3)=10.56, p<0.001) and total SA (F(3)=12.41, p<0.001), with both 22q11.2DS groups having a significantly lower total volume and area compared to both idiopathic ASD and TD controls (p<0.05 for all pair-wise comparisons). Last, there was a significant effect of group for mean CT (F(3)=3.74, p<0.05) across the cortex, with idiopathic ASD individuals having a trend towards reduced CT compared to 22q11.nonASD individuals (p=0.088), while no other pair-wise comparison was significant (see Table 1 and Supplementary Tables 1 for all pair-wise comparisons). We thus co-varied for gender, full-scale IQ, and respective total brain measure in all subsequent analyses.
Results of the categorical fixed-effects analyses
Main Effect of 22q11.2DS on rCV, SA, and CT
Significant neuroanatomical differences between 22q11.2 deletion carriers (i.e. 22q11.2DS with and without ASD) and non-carriers (i.e. idiopathic ASD and TD controls) were observed in several large clusters distributed across the cortex. More specifically, rCV was increased in 22q11.2DS in the bilateral superior frontal cortex, the lateral and medial orbitofrontal cortex, the pre- and postcentral gyrus, the insula, and the supramarginal gyrus, with increases being driven by a commensurate increase in SA. Increased rCV in 22q11.2DS was also observed in the left middle temporal gyrus, while increased SA was further observed in the left superior temporal gyrus and the left posterior cingulate cortex (PCC). In contrast, rCV was decreased in 22q11.2DS in a large cluster centered on the bilateral medial occipital and temporal lobes, as well as in the bilateral anterior cingulate cortex, and the pre- and postcentral gyrus, accompanied by commensurate decreases in SA. Further decreases in SA were observed in the bilateral dorsal anterior cingulate area and inferior temporal gyri. Last, we identified increased CT in 22q11.2DS in some scattered regions, including the bilateral lateral occipital cortex, the right postcentral gyrus, and the left supramarginal gyrus, whereas decreases in CT were observed in the bilateral superior temporal lobes, the parahippocampal gyri, and the posterior cingulate cortex (see Fig. 1, Supplementary Fig. S2, and Supplementary Table S2).
Main Effect of ASD on rCV, SA, and CT
For the main effect of ASD, we established that individuals with ASD symptomatology (i.e. individuals with idiopathic ASD and 22q11.ASD) were neuroanatomically distinct from those without (i.e. compared to TD controls and 22q11.nonASD), with significantly increased rCV in the left insula and left superior temporal gyrus, accompanied by a more widespread increase in SA, also spanning the fusiform, parahippocampal, lingual, and supramarginal gyri. rCV was further increased in the right inferior parietal cortex in ASD. In contrast, decreases in rCV in ASD were observed in the left entorhinal cortex, accompanied by a commensurate decrease in SA, that was more pronounced and also implicated the left fusiform gyrus. For measures of CT, individuals with ASD showed significant increases in the right isthmus cingulate cortex and the right superior temporal gyrus (see Fig. 1, Supplementary Fig. S2, and Supplementary Table S3).
Significant Interactions between 22q11.2DS and ASD
In addition to the main effects, we observed significant interactions between 22q11.2DS and ASD for measures of rCV and SA. These were located in the left dorsolateral prefrontal cortex (DLPFC) for both, rCV and SA, as well as in the right precentral gyrus for rCV only, and in the left PCC for SA only (see Fig. 1, Supplementary Fig. S2, and Supplementary Table S4). In significant clusters, ASD was associated with increased rCV and/or SA in 22q11.2DS (i.e. 22q11.ASD > 22q11.nonASD), but reduced rCV and/or SA in individuals without the microdeletion (i.e. idiopathic ASD < TD controls). In the DLPFC and PCC, individuals with 22q11.nonASD were the most affected on the neuroanatomical level (i.e. had the most reduced rCV and/or SA relative to all other groups), while both ASD groups were comparable in terms of their mean rCV and/or SA (22q11.nonASD 22q11.ASD = ASD TD controls). In the precentral cluster exclusively, 22q11.ASD individuals had the largest mean rCV compared to all other groups, with the mean of 22q11.nonASD individuals being between idiopathic ASD and TD controls (see Supplementary Fig. S3). As the 22q11.ASD and idiopathic ASD groups differed in symptom severity in the repetitive behavior domain, we also performed the analysis covarying for the SRS Restricted Interests and Repetitive Behavior subscale. However, the patterns of significant 22q11.2DS-by-ASD interactions remained unchanged overall (see Supplementary Fig. S4).
Results of the CCA
Initially, CCA was performed across individuals without the microdeletion. Here, we observed a significant multivariate association between the 19 regional measures of brain anatomy (see Methods) and the five symptom domains of the SRS (RVcoef=0.196, p<0.001). Based on the number of clinical predictors, the CCA yielded five canonical variates (CVs) with correlations of 0.665, 0.632, 0.502, 0.427, and 0.368 for each successive canonical pair, respectively (see Fig. 2A). Collectively, the full model including all CCs was also statistically significant at p<0.1 using Wilk’s l=0.176 (F(95,281)=1.27, p<0.01) and Pillai’s trace=1.41 (F(95,305)=1.26, p<0.07). As Wilk’s l indicates the variance unaccounted for by the model, the R-square type (r2) effect size of the model was 0.823 (i.e. 1-l), indicating that the full model explained about 82.35% of the variance shared between measures of neuroanatomy and clinical symptom profile. Moreover, the total variance in clinical symptoms that could be explained by neuroanatomical variation was 40.38%, to which only the first two CVs contributed significantly (27.81% and 10.1%, respectively; see Fig. 2A). Out of those, only the 1stCV was statistically significant (Bartlett’s c2(95)=117.04, p<0.065) and explained a total of 62.88% of variability within the set of clinical variables on its own (clinical canonical variate adequacy, see Fig. 2B). Thus, given the r2 effects for each CV, only the first canonical variate was considered noteworthy in the context of the present study. This first CV was also sufficient to attain a good discrimination between individuals with and without ASD (see Fig. 2C). Figure 2D shows the canonical loadings (lc) for each neuroanatomical predictor on the cortical surface, which highlights the set of brain regions maximally correlated with the 1st canonical variate. As expected, high positive loadings (i.e. >3) were observed in many regions of the social brain including the right medial orbitofrontal lobe (CT, lC1=0.47), the rostral middle frontal gyrus (CT, lC1=0.51), the left insula (rCV, lC1=0.41), and the right superior temporal gyrus (rCV, lC1=0.38). High negative loadings were observed in the left precuneus (CT, lC1=-0.38), the bilateral superior parietal lobes (CT, bilateral: lC1=-0.38; rCV, left: lC1=-0.30, right: lC1=-0.36), and the right lateral orbitofrontal cortex (CT, lC1=-0.34). Detailed information on canonical loadings (lc) for all CVs can be found in the neuroanatomical canonical loading plot in the Supplementary Fig. S5. In individuals without the microdeletion, clinical variability across the five subdomains of the SRS can therefore be reduced to a single latent trait variable that is (1) highly predictive of group membership (i.e. ASD vs. controls; see Fig. 2C), and that is (2) significantly associated with neuroanatomical variability across multiple morphometric features in a specific set of brain regions (see Fig. 2D).
In 22q11.2DS individuals, the multivariate association between autistic symptoms and neuroanatomical variability in the investigated set of brain regions was considerably reduced relative to non22q11.2DS individuals (RVcoef=0.06, p<0.7), and the full model did not reach statistical significance (Wilk’s l=0.065, F(95,131)=10.4, p<0.5; Pillai’s trace=1.92, F(95,150)=0.98, p<0.6). While canonical correlations were high overall (0.832, 0.704, 0.624, 0.444, 0.379, respectively, see Fig. 3A), only 24.74% of the clinical variance could be explained by the set of neuroanatomical features examined, with anatomical CVs 3 and 5 explaining the largest percentage of clinical variability (8.20% and 7.43%, respectively; see Fig. 3A). To identify the CV in 22q11.2DS that is comparable to CV1 in non22q11.2DS, clinical CVs were sorted based on the percentage of clinical variability explained (i.e. clinical variate adequacies). As shown in Figure 3B, CV5 in the 22q11.2DS group was the clinically most relevant variate, accounting for more than 50% of the clinical variance on its own, followed by CVs 3 and 4, which explained 21% and 17% of the clinical variance, respectively. Moreover, CV5 displayed a clinical loading profile that closely resembled the profile of CV1 in the non22q11.2DS individuals (i.e. high positive loadings across all five SRS subdomains; see Fig. 3B) with a congruence coefficient of 0.99 (high degree of factor similarity). CV5 in the 22q11.2DS group was therefore considered the equivalent of CV1 in the non22q11.2DS individuals and also showed a good discrimination between individuals with and without ASD (see Fig. 3C).
However, while these CVs displayed a high degree of clinical factor similarity across groups, they were mediated by different neuroanatomical substrates between groups (see Fig. 2D, Fig. 3D, and Fig. 4). When comparing the neuroanatomical underpinnings of CV1 in non22q11.2DS and CV5 in 22q11.2DS based on their loadings, we established that there was a low level of congruence overall (congruence coefficient=0.64). Largest differences in CV loadings (i.e. >0.3) were observed for CT in the right lateral and medial orbitofrontal cortex (Dl=-0.56 and 0.46, respectively), in the right rostral middle frontal cortex (Dl=0.38), and in the left precuneus and superior parietal cortex (Dl=-0.35 and -0.30, respectively; see Fig. 4). Thus, while the complex clinical phenotype of ASD may be reduced to a single continuous variable in both 22q11.2DS and non22q11.2DS individuals, this latent clinical factor seems to be mediated by different neuroanatomical substrates.