Abstract
Breast cancer brain metastasis (BCBM) is a major cause of poor prognosis in breast cancer, driven by complex molecular mechanisms. Innovative diagnostic and therapeutic strategies are urgently needed. We integrated single-cell RNA sequencing, multiomics profiling from the TCGA database, and machine learning to explore the molecular features of BCBM. Cell composition, gene expression, and subtypes were characterized. Prognostic models were developed, and potential drug targets were computationally identified through analysis of differentially expressed genes and molecular interactions. Experimental validation of these targets was performed using orthotopic implantation of MDA-MB-231-Luc cells in nude mice. scRNA-seq revealed 10 cell types and 1,479 differentially expressed genes, highlighting significant differences between the primary brain cancer and BCBM. Multiomics clustering defined two distinct subtypes (CS1 and CS2) with differential prognosis. A CoxBoost + RSF model identified hub genes (BTG2, PSMB8, SRGN, HLA-DPB1) and demonstrated high predictive accuracy for 3-, 5-, and 10-year survival (AUCs: 0.813, 0.788, and 0.776, respectively). Drug sensitivity analysis highlighted five candidate agents, with molecular docking confirming strong binding affinity to targeted proteins. In vivo experiments confirmed that PSMB8 and HLA-DPB1 promoted brain metastasis, while BTG2 and SRGN suppressed it. High-risk patients exhibited elevated monocyte proportions, which were involved in intercellular interactions. This study delineates the molecular landscape of BCBM, establishes robust prognostic models, and identifies promising therapeutic targets, offering a framework for precision diagnosis and individualized treatment. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1007/s10238-025-02003-4.