1304Pan-cancer analysis of pre-diagnostic blood metabolite concentrations in the EPIC study

Abstract Background Metabolomics studies in cancer epidemiology have mostly focused on single metabolite-cancer site associations. Pan-cancer analyses may larger statistical power when identifying metabolites showing consistent associations across sites, while allowing the identification of site-specific Methods Data from seven cancer-specific case-control nested within European Prospective Investigation into Cancer and Nutrition Cohort (EPIC) were pooled, resulting a total sample 7,957 pairs eight types (breast, colorectal, endometrial, gallbladder, kidney, localized prostate advanced cancer, hepatocellular carcinoma). A 117 pre-diagnostic blood measured. After clustering most highly correlated ones together, we studied association between 50 features (metabolites or clusters metabolites) risk multivariate penalized conditional logistic regression models controlled for body mass index using data shared lasso. Results We identified: (i) 8 with sites: e.g., glutamine C4-acylcarnitine, one cluster lysophosphatidylcholines phosphatidylcholines inversely associated C10-acylcarnitine, valine proline showed positive associations; (ii) 11 heterogeneous arginine was positively colorectal only, sphingomyelins carcinoma endometrial cancer. Conclusions Our pan-cancer analysis notably identified different cancer-types. Key messages results could lead to common pathways types.