Nonlinear Modeling Reveals Novel Associations Between Genetically Predicted Protein Levels and Pancreatic Cancer Risk

Pancreatic ductal adenocarcinoma (PDAC) represents a highly fatal malignancy with a huge public health burden. There is a critical need to better understand its etiology for developing innovative strategies for effective prevention and treatment. Leveraging genetic variants as instrumental variables, Mendelian randomization and proteome-wide association study have identified dozens of protein biomarkers associated with PDAC risk, yet potential nonlinear associations have largely been underexplored. In this study, we applied a nonlinear modeling approach, combining two-stage sliced inverse regression (2SIR) with nonlinear transformations via adjusted inverse regression (AIR), to investigate associations between genetically predicted protein concentrations in plasma and PC risk, by integrating blood proteome and genome data from the INTERVAL study (n = 3301), and a large genome-wide association study of PC risk (8275 cases and 6723 controls). We identified 25 genetically predicted proteins associated with PDAC risk after multiple comparison correction, including 22 that had been previously reported using linear modeling methods, and an additional three novel proteins (APOF, CCL15, and CHIT1). Importantly, there has been some level of evidence in the literature supporting potentially important roles of some of these novel proteins in PDAC development. Our study underscores the importance of accounting for nonlinear relationships in uncovering novel proteins associated with PDAC risk. If validated in further studies, our findings could improve the understanding of PDAC pathogenesis and inform future therapeutic and risk assessment strategies to reduce the burden from this deadly cancer.