Multi-regional transcriptomics reveals robust consensus subtypes beyond tumor heterogeneity of breast cancer

BACKGROUND: Commercial molecular classifiers such as Prediction Analysis of Microarray 50 (PAM50), Oncotype DX, and MammaPrint have revolutionized breast cancer management. However, their clinical reliability is undermined by tumor heterogeneity. METHODS: We quantified the discordance in clinical molecular typing across multi-regional samples. Gene set enrichment analysis (GSEA) was employed to explore potential sources of intratumoral heterogeneity (ITH). Gene-level heterogeneity was evaluated by quantifying expression variation within versus between tumors. A robust subtyping framework was constructed via Non-negative Matrix Factorization (NMF) based on a screened low-ITH gene set. RESULTS: ITH was observed in over 50% of tumors, triggering significant classification discordance across commercial tests. GSEA revealed that transcriptomic heterogeneity might be driven by stromal and matrix-related pathways. The low-ITH-based NMF framework established an optimal three-subtype classification that exhibited superior prognostic stratification and maintained high stability across metastatic sites. CONCLUSIONS: We established a low-ITH subtyping framework that overcomes the limitations of single-biopsy-based molecular typing. Our findings offer a novel strategy to enhance the precision of breast cancer management.