Plasma Proteomic Determinants of Common Causes of Mortality

Abstract In complex physiological systems, causal associations between protein concentration and outcomes can be difficult to determine, as proteins can be both effectors of outcomes and responders. This problem becomes especially difficult when the outcome of interest is death because this is often preceded by the failure of multiple physiological systems. Our aim in this study was to determine the extent to which variation in plasma protein abundance can predict all-cause and disease-specific mortality, and to assess the potential causal nature of any observed associations. To identify biomarkers for mortality, we analyzed the largest plasma proteome dataset compiled to date: 1,459 proteins measured in 54,306 individuals with over 12 years of prospective follow-up data from the UK Biobank (during which 5,032 mortality events occurred). We carried out a multivariate stability selection analysis of all biomarkers and identified 14 proteins that were robustly predictive of all-cause mortality. These biomarkers remained significant after accounting for pre-existing conditions and medication usage. Participants in the highest quintile of mortality risk prediction had ~25 times higher number of events as compared to participants in the lowest quintile. Moreover, these proteins are more robustly associated with mortality than established diagnostic biomarkers such as glycated hemoglobin (HbA1c) and cholesterol. To complement our epidemiological analyses, we quantified the genetic determinants of plasma protein abundance and assessed causality via Mendelian randomization and mediation analysis using polygenic scores. Integrating multiple lines of evidence, we prioritized eight plasma protein biomarkers causally associated with human lifespan.