Abstract
PURPOSE: Metabolomics is a discovery tool for novel associations of metabolites with disease. Here, we interrogated the metabolome of human breast tumors to describe metabolites whose accumulation affects tumor biology. EXPERIMENTAL DESIGN: We applied large-scale metabolomics followed by absolute quantification and machine learning-based feature selection using LASSO to identify metabolites that show a robust association with tumor biology and disease outcome. Key observations were validated with the analysis of an independent dataset and cell culture experiments. RESULTS: LASSO-based feature selection revealed an association of tumor glycochenodeoxycholate levels with improved breast cancer survival, which was confirmed using a Cox proportional hazards model. Absolute quantification of four bile acids, including glycochenodeoxycholate and microbiome-derived deoxycholate, corroborated the accumulation of bile acids in breast tumors. Levels of glycochenodeoxycholate and other bile acids showed an inverse association with the proliferation score in tumors and the expression of cell-cycle and G(2)-M checkpoint genes, which was corroborated with cell culture experiments. Moreover, tumor levels of these bile acids markedly correlated with metabolites in the steroid metabolism pathway and increased expression of key genes in this pathway, suggesting that bile acids may interfere with hormonal pathways in the breast. Finally, a proteome analysis identified the complement and coagulation cascade as being upregulated in glycochenodeoxycholate-high tumors. CONCLUSIONS: We describe the unexpected accumulation of liver- and microbiome-derived bile acids in breast tumors. Tumors with increased bile acids show decreased proliferation, thus fall into a good prognosis category, and exhibit significant changes in steroid metabolism.