Abstract
Alternative splicing shapes isoform diversity and gene dosage but how genetic variation impacts splicing in brain disease is still not fully characterized. We assembled BigBrain, a multi-ancestry resource of 10,725 bulk RNA-seq profiles with matched genotypes from 4,656 individuals across 43 tissue-cohort pairs, and mapped 68,358 cis-sQTLs affecting 10,966 genes using mixed-model meta-analysis. Using SuSiE, we finemapped over half of these sQTLs into 95% credible sets, frequently to a single variant near splice sites. We further annotated variants predicted to alter dosage through frameshifts or nonsense-mediated decay or disrupt protein domains. Colocalization with seven neurodegenerative and psychiatric GWAS highlighted 97 loci where alternative splicing appears to mediate genetic risk. Among sQTL-eQTL pairs with colocalization probability ≥ 0.8 (posterior probability of a shared causal variant), half shared credible-set variants, showing that splicing can complement or act independently of expression. Mechanistic examples include CAMLG (Parkinson's), ZDHHC2 (Schizophrenia), and CLU (Alzheimer's).