Dissecting the metabolic landscape of breast cancer subtypes via elastic net modeling and examining its immune correlates

Abstract Objectives Breast cancer is a heterogeneous disease, and the estrogen receptor (ER) status is a key factor in disease classification and treatment planning. While metabolomic profiling has revealed subtype-specific differences, cross-study comparisons have been limited, posing challenges for data extrapolation. This study aims to investigate metabolites that differentiate ER-positive and ER-negative tumors via integrative analyses of multi-omics data. Methods We jointly analyzed two untargeted metabolomics datasets via elastic net modeling using consistent analysis pipelines tuned for low sample sizes, namely multiple bootstrapping and stability selection. Significant metabolite predictors from two studies were cross-examined to reveal distinctions and commonalities. We also performed differential gene expression analysis using RNA sequencing data from matching samples to link metabolic patterns with transcriptomic signatures and intratumoral immune cell signatures. Results This study identified unique metabolite signatures in distinct datasets and a limited overlap of discriminating metabolites that can be broadly generalizable for subtyping. Nevertheless, several glycolysis and fatty acid metabolism intermediates exhibited variation depending on the tumor ER status. Consistently, genes related to fatty acid metabolism and glycolysis were enriched in ER-positive and ER-negative tumors respectively. Furthermore, we used multiple immune cell deconvolution algorithms to correlate various immune cell types with the metabolite levels within the tumor microenvironment. Conclusions Together, these findings highlight the metabolic and immunological diversity of breast cancer and establish a reproducible machine-learning framework for integrating multi-omics data to interrogate tumor complexity.