Triangulating cross-carcinoma GWAS delineates shared programs and disease-specific drivers

Abstract Carcinomas, arising from epithelial tissues and accounting for >90% of cancers, share molecular programs while exhibiting site-specific biology, yet the genetic partitioning of common versus distinct components remains unclear. We harmonized genome-wide association studies (GWAS) summary statistics for 429,158 European-ancestry cases across nine carcinoma types and triangulated evidence at the genome-wide, regional, and locus level to delineate shared and cancer-specific risk. We demonstrate that cross-carcinoma overlap is likely to be systematically underestimated, as loci within the same genomic regions can harbor discordant effects. To address this, we constructed the cross-carcinoma hierarchical latent factor model and, in follow-up multivariate GWAS with multi-omics data integration, identified 170 previously unreported pleiotropic loci and prioritized 167 high-confidence effector genes. This framework partitions general and cancer-specific genetic liability, revealing pleiotropy obscured by conventional analyses. Subsequent multi-omics gene prioritization implicated convergent epithelial growth and differentiation programs, nominating tractable targets for biomarker development, prevention, and mechanism-informed therapies.