Massively parallel characterization of non-coding de novo mutations in autism spectrum disorder

Autism spectrum disorder (ASD) is a neurodevelopmental disorder where de novo mutations play a significant role. Although coding mutations in ASD have been extensively characterized, the impact of non-coding de novo mutations (ncDNMs) remains less understood. Here, we integrate cortex cell-specific cis-regulatory element annotations, a deep learning-based variant prediction model, and massively parallel reporter assays to systematically evaluate the functional impact of 227,878 ncDNMs from Simons Simplex Collection (SSC) and Autism Speaks MSSNG resource (MSSNG) cohorts. Our analysis identifies 238 ncDNMs with confirmed functional regulatory effects, including 137 down-regulated regulatory mutations (DrMuts) and 101 up-regulated regulatory mutations (UrMuts). Subsequent association analyses reveal that only DrMuts regulating loss-of-function (LoF) intolerant genes rather than other ncDNMs are significantly associated with the risk of ASD (Odds ratio = 4.34; P = 0.001). A total of 42 potential ASD-risk DrMuts across 41 candidate ASD-susceptibility genes are identified, including 12 recognized and 29 unreported genes. Interestingly, these noncoding disruptive mutations tend to be observed in genes extremely intolerant to LoF mutations. Our study introduces an optimized approach for elucidating the functional roles of ncDNMs, thereby expanding the spectrum of pathogenic variants and deepening our understanding of the complex molecular mechanisms underlying ASD.