Abstract
ABSTRACTCompared to disorders of similar heritability and contribution of common variants, few genome-wide significant loci have been implicated in autism spectrum disorder (ASD). This undermines the use of polygenic risk scores (PRSs) to investigate the common genetic architecture of ASD. Deconstructing PRS-ASD into its related traits via “developmental deconstruction” could reveal the underlying genetic liabilities of ASD. Using the data of >24k individuals with ASD and >28k of their unaffected family members from the SSC, SPARK, and MSSNG cohorts, we computed the PRSs for ASD and 11 genetically-related traits. We applied an unsupervised learning approach to the ASD-related PRSs to derive “multi-PRSs” that captured their variability in orthogonal dimensions. We found that multi-PRSs captured a similar proportion of genetic risk for ASD in cases versus intrafamilial controls (ORmulti-PRS=1.10, R2=0.501%), compared to PRS-ASD itself (ORPRS-ASD=1.16, R2=0.619%). While multi-PRS dimensions conferred risk for ASD, they had “mirroring” effects on developmental phenotypes among cases with ASD. We posit that this phenomenon may partially account for the paucity of genome-wide significant loci and the clinical heterogeneity of ASD. This approach can serve as a proxy for PRS-ASD in cases where non-overlapping and well-powered GWAS summary statistics are difficult to obtain, or accounting for heterogeneity in a single dimension is preferable. This approach may also capture the overall liability for a condition (i.e.: genetic “P-factor”). Altogether, we present a novel approach to studying the role of inherited, additive, and non-specific genetic risk factors in ASD.
Publisher
Cold Spring Harbor Laboratory