Abstract
Background: Breast cancer heterogeneity necessitates robust prognostic biomarkers and therapeutic targets. This study aimed to identify key molecular drivers through integrative multi-omics approaches and validate their clinical relevance. Methods: We combined differential expression analysis, weighted gene co-expression network analysis (WGCNA), and machine learning (StepCox-Random Survival Forest [RSF]) to screen prognostic signatures across TCGA, GEO (GSE42568, GSE9893, GSE7390), and METABRIC datasets. Immune microenvironment characterization utilized ESTIMATE, CIBERSORT, and functional enrichment analyses. Mechanistic validation included single-cell RNA sequencing, in vitro/in vivo experiments, and clinical cohort profiling. Results: WGCNA identified 102 hub genes linked to breast cancer progression. Machine learning optimization yielded a 3-gene signature (EGR3, RECQL4, MMP1) with superior prognostic stratification. Multi-cohort validation confirmed signature robustness. The C2 subtype, defined by high-risk scores, exhibited an immunosuppressive microenvironment with elevated PD-L1/LAG3/TIGIT and M2 macrophage enrichment. EGR3 emerged as a pivotal tumor suppressor: its expression inversely correlated with tumor stage and positively associated with CD8+ T cell infiltration. EGR3-high patients showed prolonged survival and enhanced immunotherapy response. Functional studies demonstrated EGR3 overexpression suppressed tumor growth and activated CD8+ T cells. Conclusion: Our integrative framework established a machine learning-optimized 3-gene prognostic model with cross-platform reliability. EGR3 was validated as a dual-function regulator of tumor suppression and immunomodulation, offering a novel therapeutic target for breast cancer, particularly in immunologically "cold" triple-negative subtypes.