Biobank-scale Bayesian TWAS reveals splicing-mediated mechanisms of complex disease

Abstract

Alternative splicing is a key mechanism by which genetic variation contributes to human phenotypic diversity and disease risk, yet its incorporation into large-scale genetic studies remains limited. Here we utilize blood RNA-seq from 4,732 European-ancestry individuals in the INTERVAL cohort to construct genetic scores for junction-based splicing phenotypes, 13,851 of which showed R2 > 0.01 in withheld individuals. These models were used to predict splicing in the UK Biobank White-British subset (UKBB; n=408,590) and All of Us individuals of diverse ancestries (AoU; n=201,958), enabling splicing transcriptome-wide association studies (sTWAS) across 1,129 harmonized disease phenotypes. We identified 4,966 splicing-disease associations, of which 67.3% were shared across cohorts. To account for confounding by co-regulation and linkage disequilibrium, we adapted a Bayesian joint fine-mapping framework to prioritize 1,277 likely causal splicing-disease associations across 494 genes, revealing mechanisms underlying immune, vascular, and metabolic traits. Fine-mapped associations included exon-skipping events in GSDMB and splicing in untranslated regions in STAT6 and IL18RAP for asthma, unannotated splicing events in PIEZO1 and PPP3R1 for vascular traits, and alternative exons in BAK1 and TNFRSF14 for celiac disease. Our study demonstrates that common genetic variation influencing splicing substantially shapes complex trait architecture and provides a scalable, statistically rigorous framework to uncover transcriptional disease mechanisms.