基于NMR的代谢组学与UK Biobank队列中肺癌风险的关联

Background Lung cancer remains the most common type of cancer and the leading cause of cancer-related death globally. Nuclear magnetic resonance (NMR) techniques enable detailed profiling of lipoprotein subclasses by concentrations and compositions, offering potential opportunity for lung cancer risk screening. Methods We included 273,375 participants (2164 developed lung cancer) from the UK Biobank, with 251 metabolites (170 absolute and 81 ratio metabolites) measured at baseline. Cox proportional hazards models were used to estimate the individual hazard ratio (HR) between each metabolite and lung cancer risk, adjusting for core and extended covariates. P-values were corrected for multiple testing. Subgroup analyses were conducted by sex, age group, and smoking status. Principal component analyses (PCA) were performed separately on absolute and ratio metabolites. PCs explaining 90% of the variance in each set were incorporated into Cox models to assess patterns in metabolomic profiles on lung cancer risk. A LASSO penalty was applied to identify key metabolites for lung cancer risk and their associations were obtained using multivariable Cox regression. Results A total of 163 metabolites (98 absolute and 65 ratios metabolites) were found to be significant. Among absolute metabolites, glycoprotein acetyls showed the strongest association with increased risk [HR per SD = 1.19, 95% CI = 1.14-1.25], whereas the degree of unsaturation in fatty acids showed the strongest protective association [HR per SD = 0.84, 95% CI = 0.80-0.88]. Across all lipoprotein compositions, only triglycerides consistently showed increased risk in the main and subclasses of lipoproteins [HR range: 1.00-1.07], with 5 associations reaching statistical significance. Similar patterns were observed for ratio metabolites, except for phospholipid percentages, which showed increased risk. These patterns remained largely unchanged after adjustment for extended covariates, except for triglycerides, whose associations were attenuated and even reversed in some cases. In subgroup analyses, the largest differences were observed for triglycerides, where stronger positive associations were seen among females, former smokers, and people aged <60 years. In absolute metabolites, PC1 (VLDL-driven), PC2 and PC3 (HDL-driven) showed increased risk; PC8 and PC9 (glycolysis- and fatty acids-driven) showed decreased risk. In ratio metabolites, PC1, PC4, and PC6 (LDL and VLDL-driven) showed increased risk, PC5, PC7 and PC9 (fatty acids-driven) showed decreased risk. The LASSO-Cox models selected 15 metabolites and the direction and magnitude of these metabolites were largely consistent with those observed in the individual CoxPH models. Conclusion Findings from this study highlighted the complexity of metabolic patterns and their role in lung cancer risk, warranting further investigation through pathway analyses.