Sex-Informed Neuroanatomy of Autism: Evaluating the Quantitative and Qualitative Models of Sex-Moderation in Cortical Anatomy Using Data from the Pond Network

Background:

We proposed two testable models on sex-moderation that underlie the ‘female-protective-effect’ mechanisms contributing to autism etiologies, as a key first step to clarify the sex-differential liability related to the brain. The ‘quantitative sex-moderating model’ (QUAN-model) posits that mechanisms leading to autism are comparable but differ quantitatively across sexes, that females require more brain changes to develop autism. The ‘qualitative sex-moderating model’ (QUAL-model) posits that mechanisms leading to autism operate through different neurobiological/brain changes across sexes.

Objectives:

Testing the QUAN-model and QUAL-model against a null-model of no sex-moderation:
(1) The QUAN-model predicts that across the brain, autism-related changes in females involve similar regions but with larger effect sizes, and/or involve overlapping and additional regions, than those in males.
(2) The QUAL-model predicts that across the brain, autism-related brain changes differ by sex, involving different brain regions and/or with different patterns of changes.

Methods:

Sample includes 729 participants scanned at the Hospital for Sick Children, Toronto and via the Province of Ontario Neurodevelopmental Disorders (POND) network, using the same scanner: 249 autistic males (age mean=13.3 years, FIQ mean=96.9), 60 autistic females (age mean=14.2, FIQ mean=95.7), 210 neurotypical males (age mean=14.0, FIQ mean=112.7), 210 neurotypical females (age mean=14.2, FIQ mean=113.0). T1-weighted scans were processed with CIVET 2.1. Cortical thickness, area, volume and curvature were estimated across 75,284 vertices. To account for heterogeneity of covariates, data were pre-processed with nonparametric matching – neurotypical controls were matched to autism subjects with full-matching by weighted Euclidean matching for age and whole brain volume (weights 1 and 0.5 respectively), but IQ was not used in matching due to high co-occurrence between autism and intellectual disability (ID). For each metric, a whole-cortex-based model selection determined the best model by Akaike’s Information Criterion.
Testing QUAN-model vs. null by Sign-Concordance: By setting ‘sex’ as male=0 and female=1, autism-effect in males is given as β_diagnosis and autism-effect in females β_diagnosis+β_{sex*diagnosis}. The QUAN-model predicts that autism-effect is stronger in females than in males, therefore β_diagnosis and β_{sex*diagnosis} have the same sign (sign-concordance). This was tested against a randomization distribution of test-statistics from the model with sex-labels permuted.
Testing QUAL-model vs. null by Spatial-Correlation: Evidence for the QUAL-model will be if, across vertices, there is less spatial correlation between autism-effects in females vs. males than by chance. This was tested with the same permutation approach.

Results:

Model selection: Thickness, area, volume and smoothed absolute mean curvature were best modeled without sex-covariate interaction terms. Thickness and volume were best modelled by brain volume cubed and a quadratic age term; Area and thickness were best modelled by linear age, and brain volume to the 2/3 power.
QUAN-model: Increased sign-concordance was observed for smoothed absolute mean curvature (but not other metrics) at effect thresholds greater than 50% of maximum observed effect, supporting the QUAN-model; this result was insensitive to whether IQ was included as a covariate.
QUAL-model: All measures were within randomization intervals providing no evidence for the QUAL-model.

Conclusions:

These results suggest possible quantitative sex-moderation of cortical curvature in autism. How sex-moderation is associated with autism-ID co-occurrence requires further investigation.