Proteomic signatures of smoking and their associations with risk of incident diseases and mortality in diverse populations

Abstract Smoking is the most important behavioural determinant of morbidity and mortality. Here we used a machine learning model to assess plasma levels of 2,917 proteins in the UK Biobank (n=43,914) and generated a proteomic Smoking INdex (pSIN). pSIN comprised 51 proteins which accurately discriminated current versus never smokers (AUC=0.95; CI:0.94-0.95) and the accuracy of the pSIN was further validated in participants from the China Kadoorie Biobank (n=3,977; AUC=0.91; CI:0.89-0.92). pSIN was significantly associated with the risk of all-cause mortality and incidence of 18 major chronic diseases (including diseases of the heart, kidney and lungs, neurodegeneration and cancers). Among current and previous smokers, pSIN was associated with risks of death and 11 diseases, independent of self-reported smoking history and major socio-economic and lifestyle factors. Beyond smoking behaviour, we demonstrated that pSIN is modified by genetic and exposome factors. Genome-wide association analysis identified 125 genes associated with pSIN, including ALPP, CST5, and IL12B. Exposome-wide analysis identified maternal smoking, diet, physical activity, and air pollution as the most significant exposures associated with pSIN beyond self-reported smoking behaviour. Importantly, pSIN was able to identify a subgroup of previous smokers who, even after decades of smoking cessation, showed similar cumulative incidence rates of asthma, chronic kidney disease, chronic liver disease, and congestive cardiac failure as the current smokers. These findings demonstrate that plasma proteomics can be used to effectively capture smoking patterns and risk of smoking-related morbidity and mortality, and that proteomics-based estimation of smoking history provides more nuanced information about differences between individuals in the dynamics of biological response to smoking.