Abstract
Background Air pollution exposure has been identified as a pathogenic factor of lung cancer, whereas the metabolic profile disturbance involved and its underlying role remain unclear while attract much attention. Methods Metabolomic profiling in plasma was conducted among 205,974 participants in the UK Biobank. Particulate matter (PM) with aerodynamic diameter ≤ 10 μm (PM10), PM2.5, PM2.5-10, nitrogen dioxide (NO2), and nitrogen oxides (NOx) were assessed by land-use regression models. Mediation roles of metabolic features involved in air pollution and incident lung cancer, and performance of the lung cancer prediction model incorporating crucial metabolite features identified by least absolute shrinkage and selection operator (LASSO) and multivariate Cox regression, were evaluated. Results During a median follow-up period of 13.1 years, 1,536 incident lung cancer cases were recorded. Among the 143 metabolite features, 66 overlapped in PM2.5, NO2, or NOx exposure-associated incident lung cancer after multivariate adjustment (false discovery rate P < 0.05). The highest mediation proportions were observed for Albumin (percentage mediated: 4.02 %), Phospholipids in Medium Very-Low-Density Lipoproteins (M-VLDL) (6.38 %), and M-VLDL (6.42 %) in incident lung cancer from PM2.5, NO2, and NOx exposure, respectively. LASSO and multivariate Cox regression identified 15 metabolite features associated with lung cancer, and inclusion of these metabolite features significantly improved the prediction of lung cancer (C statistic: 0.851; Net reclassification improvement index: 0.144; Integrated discrimination improvement index: 0.005). Discussion Disturbance and mediation role of circulating metabolic features in air pollution exposure and incident lung cancer were identified, and metabolite profiling may well improve early prediction of lung cancer.