Integrative single-cell and spatial transcriptomics with machine learning identify a Luminal-inflam malignant program and reveal an RPN1-PERK UPR vulnerability in triple-negative breast cancer.

Triple-negative breast cancer is marked by extensive cellular heterogeneity and limited availability of actionable targeted treatments, which contributes to an unfavorable prognosis. In this work, single-cell and spatial transcriptomic profiling was integrated with network-based analyses and machine-learning approaches to characterize malignant epithelial programs in TNBC and to pinpoint prognostic biomarkers.Single-cell RNA sequencing identified a malignant epithelial subpopulation, Luminal_inflam, characterized by elevated inferred copy number variation, a terminal pseudotime state, and enrichment of cell cycle-associated transcriptional programs. Cell-cell communication analysis indicated microenvironmental remodeling in TNBC and prioritized a fibroblast-associated S100A4-EGFR axis that may regulate Luminal_inflam-associated gene expression. Gene regulatory network analysis further revealed increased activities of transcription factors including MYBL2, TFDP1, CEBPD, and MBD2. A 12-gene risk signature constructed from Luminal_inflam-associated modules and survival cohorts effectively stratified overall survival and captured differences in immune features and potential drug sensitivities. In our MDA-MB-231 model, RPN1 knockdown was associated with reduced cell viability, which could be partially rescued by 4-PBA. We further found that RPN1 depletion was accompanied by increased intracellular ROS and Ca