39. UNCOVERING PATHOGENIC SPLICING VARIANTS IN AUTISM SPECTRUM DISORDER THROUGH GENOMIC AND TRANSCRIPTOMIC INTEGRATION

Background Autism Spectrum Disorder (ASD) is a highly heritable neurodevelopmental condition, yet most cases remain genetically unresolved. Rare variants that disrupt splicing in ASD may have profound consequences because alternative splicing gives the brain the highest isoform diversity of any human tissue. However, such variants are predominantly studied using Whole Genome Sequencing (WGS) alone, without integrating RNA sequencing (RNA-seq), and analyses are often restricted to canonical splice sites (CSS) to identify canonical splicing variants (CSVs). Variants occurring outside of CSS, known as non-canonical splicing variants (NCSVs), can also impact splicing but are largely unexplored in ASD. Additionally, the effects arising from both kinds of variants on transcript architecture remain unknown. Methods We utilized matched genomic and transcriptomic data from 3,979 individuals (2,124 affected; 1,855 unaffected) from the Simons Simplex Collection (SSC) to study splicing disruption in ASD. First, we isolated de novo and rare inherited variants (population frequency = 0%) from these individuals, classifying them into CSVs and NCSVs. We then used SpliceAI (score > = 0.2) to prioritize a high-confidence subset predicted to disrupt splicing. Independently, we ran FRASER2 on the transcriptomic data to identify samples with outlier splicing events. Finally, we integrated these findings using Sashimi plots to determine the contribution of the WGS-identified variants to the observed splicing outliers. Results Burden analysis of the WGS data revealed that both de novo and rare inherited CSVs were enriched in ASD-associated genes among individuals with ASD (p < 0.05). While this result was partially anticipated due to discovery bias, the observed enrichment of NCSVs is a novel finding. Within high-confidence ASD genes, de novo NCSVs were approximately 12-fold more likely to be found in ASD cases compared to unaffected controls (p = 0.006). The effect size for rare inherited NCSVs was smaller (OR = 1.39) but remained statistically significant (p = 0.02). Next, we assessed the transcriptomic data to determine the structural impact of these genomically predicted variants. Of the 1,528 variants tested, 102 demonstrated splicing disruption. Among these disruptive variants, CSVs were 3-fold more likely to be present in individuals with ASD (p = 0.0037), compared to a 2-fold enrichment for disruptive NCSVs (p = 0.0086). Importantly, despite a lower relative enrichment, a greater absolute proportion of the affected cohort harboured a confirmed splice-disrupting NCSV (2.17%, n = 46) than a CSV (1.32%, n = 28). Discussion Overall, our study illustrates the power of integrating genomic and transcriptomic data to characterize variants previously overlooked when using WGS alone. Importantly, it highlights the necessity of this multi-omic approach, as relying solely on genomic prediction tools like SpliceAI often drastically overestimates the true number of splice-disrupting variants. We demonstrated that variants beyond CSS can have detrimental effects and should be considered as a source of potential pathogenicity in other conditions. In the context of ASD, we found that NCSVs affect a larger absolute proportion of individuals than CSVs, leading to the discovery of 46 novel splice-disrupting NCSVs. This work has increased our rare-variant diagnostic yield in ASD from approximately 15% to 17%, representing a substantial improvement within the field of ASD genetics.