Abstract
PURPOSE: Metabolomics is the comprehensive global study of metabolites in biological samples. In this retrospective pilot study we explored whether serum metabolomic profile can discriminate the presence of human breast cancer irrespective of the cancer subtype. METHODS: Plasma samples were analyzed from healthy women (n = 20) and patients with breast cancer after diagnosis (n = 91) using a liquid chromatography-mass spectrometry platform. Multivariate statistics and a Random Forest (RF) classifier were used to create a metabolomics panel for the diagnosis of human breast cancer. RESULTS: Metabolomics correctly distinguished between breast cancer patients and healthy control subjects. In the RF supervised class prediction analysis comparing breast cancer and healthy control groups, RF accurately classified 100% both samples of the breast cancer patients and healthy controls. So, the class error for both group in and the out-of-bag error were 0. We also found 1269 metabolites with different concentration in plasma from healthy controls and cancer patients; and basing on exact mass, retention time and isotopic distribution we identified 35 metabolites. These metabolites mostly support cell growth by providing energy and building stones for the synthesis of essential biomolecules, and function as signal transduction molecules. The collective results of RF, significance testing, and false discovery rate analysis identified several metabolites that were strongly associated with breast cancer. CONCLUSIONS: In breast cancer a metabolomics signature of cancer exists and can be detected in patient plasma irrespectively of the breast cancer type.