Breast cancer (BC) contributes the highest global cancer mortality in women. BC tumors are highly heterogeneous, so subtyping by cell-surface markers is inadequate. Omics-driven tumor stratification is urgently needed to better understand BC and tailor therapies for personalized medicine. We used unsupervised k-means and partition around medoids (pam) to cluster metabolomics data from two data sets. The first comprised 271 BC tumors (data set 1) that were estrogen receptor (ER) positive (ER+, n = 204) or negative (ER-, n = 67) with 162 identified and validated metabolites. The second data set contained 67 BC samples (data set 2; ER+, n = 33; ER-, n = 34) and 352 known metabolites. Significance Analysis of Microarrays (SAM) was used to identify the most significant metabolites among these clusters, which were then reassigned into new clusters using prediction analysis of microarrays (PAM). Generally, metabolome-defined BC subtypes identified from either data set 1 or data set 2 were different from the well-known receptor- or transcriptome-defined subtypes. Metabolomics-directed clustering of data set 2 identified distinctive BC tumors characterized by metabolome profiles that associated with DNA methylation (p-value = 0.000 048, χ2 test). Pathway analysis of cluster metabolites revealed that nitrogen metabolism and aminoacyl-tRNA biosynthesis were highly related to BC subtyping. The pipeline may be run from GitHub: https://github.com/FADHLyemen/Metabolomics_signature. Our proposed bioinformatics pipeline analyzed metabolomics data from BC tumors, revealing clusters characterized by unique metabolic signatures that may potentially stratify BC patients and tailor precision treatment.
Keywords: breast cancer; estrogen receptor; metabolomics; unsupervised clustering.