Abstract
Prostate cancer exhibits a strong familial association, and its heritability indicates a significant contribution from germline variants. While genome-wide association studies (GWAS) have identified common germline variants associated with prostate cancer risk, translating these statistical associations into functional mechanisms has remained a long-standing challenge. Consequently, most of our understanding of the genetic basis of prostate cancer stems from extensive studies of somatic mutations, leaving the germline genetic architecture largely unresolved. Because most germline variants lie in the noncoding genome and complex human diseases are predominantly driven by regulatory mutations, we herein asked which prostate cell types mediate the functional effects of germline variants, and thus represent the most genetically vulnerable populations. We generated paired epigenomic and transcriptomic profiles from reference human prostate tissues. Integrating these single-cell data with large-scale GWAS data identified a terminally differentiated luminal epithelial subtype that mediates the strongest germline risk in prostate cancer. We subsequently developed a deep learning model to score ~17 million GWAS variants based on their predicted impact on altering local chromatin accessibility in this vulnerable luminal epithelial subtype, and identified high-confidence candidate loci where high-risk germline variants likely alter promoter accessibility in prostate cancer. The implicated genes were involved in several pathways in tumorigenesis, displayed strong dosage sensitivity, and converged on the androgen receptor (AR)-mediated regulon, a mechanism also observed for somatic mutations. Overall, by unveiling cell types and candidate loci that mediate germline risk, our study defines the cell-type-specific germline architecture in prostate cancer and provides a comprehensive framework for understanding cancer heritability.