Cross-ancestry and phenome-wide associations of cancer-specific polygenic risk scores

BackgroundGenome-wide association studies (GWAS) have identified many common variants associated at low effect sizes with various cancers. Summing the effects of these variants into polygenic risk scores (PRS) can improve cancer risk prediction. However, these PRS have largely been analyzed in cohorts restricted to individuals of European descent, and research into potential cross-cancer and cross-phenotype pleiotropic associations of cancer-specific PRS is limited. MethodsUsing logistic regression models, we tested the association of 13 cancer-specific PRS (bladder, colorectal, esophageal, glioma, lung, melanoma, oral cavity, pancreatic, renal, thyroid, breast, prostate, and testicular) with curated phenotypes representing the same 13 cancers and 340 cancer and cardiometabolic phecodes in 560,287 individuals (114,255 African ancestry, AFR and 446,032 European ancestry, EUR) from the Million Veteran Program (MVP), a large, ancestrally-diverse biobank. Logistic regression models were stratified by ancestry and used age, principal components, and cancer-specific PRS as independent variables. ResultsAll 13 cancer-specific PRS were significantly associated (p<0.002) with their respective cancers among EUR individuals (OR 1.05 to 1.70). Among AFR individuals, cancer-specific PRS were significantly associated with their respective cancers for five of 13 cancers (bladder, breast in females, colorectal, prostate, and thyroid, OR 1.19 to 1.48). In cancer-cancer pleiotropy studies, only the renal cancer-specific PRS was significantly associated with skin cancer (OR=1.04, P=4.5x10-06) among EUR individuals. PheWAS demonstrated five positive pleotropic associations with cardiometabolic conditions (thyroid cancer PRS with thyroid goiter, oral cancer PRS with type 1 diabetes, ophthalmic and renal diabetic complications, and hypothyroidism) and two negative associations (oral and lung cancer PRS separately with coronary artery disease). There were no significant cancer-phecode associations among AFR individuals. ConclusionsWe confirm the validity of 13 cancer-specific PRS on predicting their corresponding cancer in MVP, observing stronger associations per cancer among EUR versus AFR individuals. In contrast to PRS of other chronic diseases, our study shows that the majority of cancer-related PRS are highly specific and pleiotropic associations with other cancers and cardiometabolic traits are relatively uncommon.