Machine learning-enabled spatial multi-omics uncovers lactate-driven targets and tumor microenvironmental reprogramming in cancer.

Abstract

Lactate accumulation is a central feature of tumor metabolic reprogramming, yet its spatial and cell-type-specific effects in cancer, such as lung adenocarcinoma (LUAD), remain poorly defined. We integrated single-cell transcriptomics, spatial transcriptomics, spatial metabolomics, and immunofluorescence with TCGA survival data and machine-learning models. High-lactate tumors exhibited increased epithelial and fibroblast abundances, whereas T/NK cells and monocytes/macrophages were enriched in low-lactate samples. Spatial metabolomics revealed cell-type-restricted lactate and pyruvate distributions, with endothelial cells showing minimal lactate accumulation. Endothelial subclusters in high-lactate tissues displayed angiogenic and stress-response signatures and were strongly associated with poor prognosis. Multiple machine-learning frameworks-including random forest, elastic-net regression, SVM, ANN, and decision-tree models-consistently identified endothelial and fibroblast programs as key determinants of high-lactate states and adverse clinical outcomes. Collectively, our multi-omics spatial profiling demonstrates that lactate reshapes the LUAD microenvironment by driving angiogenesis, immune suppression, and prognostic stratification, highlighting lactate-centered pathways as promising therapeutic targets.