Identification of pleiotropic genetic risk variants for pancreatic cancer and diabetes.

692 Background: Pancreatic cancer (PC) is a deadly disease most often caught in late stages. Because there is a lack of efficacious treatment options for advanced stage patients, it is extremely important to identify high-risk individuals and prevent or diagnose the disease at earlier timepoints. While some traits are well-known risk factors for PC - for example, obesity and diabetes – risk models with using existing clinical and demographic features do not effectively stratify the general population. It is difficult to pinpoint which patients with diabetes have a heightened risk for developing pancreatic cancer. The goal of this work is to develop more powerful risk stratification algorithms by leveraging massive germline genetic data and focusing on the underlying genetic determinants of diseases that are correlated with PC. Methods: For this study, we utilized publicly available biomedical databases: UK Biobank and the Million Veterans Program. A novel statistical learning procedure known as empirical Bayes composite null hypothesis testing is utilized. Specifically, we search for pleiotropic variants that are common risk factors for both PC and diabetes, and we use these pleiotropic mutations to create more refined polygenic risk scores for PC. Results: We identified causal variants for PC and diabetes at six loci on five different chromosomes, at locations where previous single-disease studies have already implicated genes associated with PC and diabetes. These regions include CTRB1/CTRB2, SPRED2, and ABO, all of which have previously been implicated in single disease studies of pancreatic cancer and diabetes. We then demonstrated how these pleiotropic variants are more predictive when used in polygenic risk algorithms. Specifically, risk scores created using pleiotropic variants increase prediction performance by statistically significant amounts compared to risk scores using variants identified from single disease studies. Conclusions: To the best of our knowledge, this is the first attempt to identify truly pleiotropic variants for PC and diabetes, as most previous studies for pleiotropy in PC have focused instead on causal variants for PC or another disease through testing a global null hypothesis. Enhanced polygenic risk scores can better stratify individuals with diabetes with propensity for developing pancreatic cancer, in turn improving the counseling, management, and prevention of those at risk for PC.