Abstract PS2-12-03: A Comprehensive Transcriptional Atlas of Breast Cancer Reveals Clinically Relevant Molecular Subtypes

Abstract Background: Breast cancer comprises molecularly diverse subtypes associated with prognosis and treatment response. Currently, these subtypes are characterized by the transcriptional analyses of individual, large-scale datasets. A more thorough understanding of breast cancer transcriptional diversity has been hindered by the challenge of performing integrated analysis across multiple independent RNA-seq datasets without compromising their intrinsic biology. Methods: We applied single-cell RNA sequencing methods to integrate bulk RNA-seq data from 20 independent datasets, encompassing 6,661 breast tumors, 368 normal breast tissues, 708 cell lines, and 161 patient-derived xenografts/organoids. Unsupervised clustering of bulk transcriptomes identified molecular subtypes, which were annotated using clinical and molecular features, including histology and PAM50 classification. Multivariate analyses evaluated prognosis and treatment response across subtypes. Gene signatures were developed to validate subtype-specific features and predict therapeutic responses across multiple independent datasets. Results: We constructed a comprehensive, integrated transcriptomic map and identified 14 biologically distinct subtypes, including three novel subtypes; mucinous (MUC), neuroendocrine (NE) and mammary stem-like (STM), TNBC subtypes (luminal androgen receptor [LAR], basal-like 2 [BL2]) and intrinsic subtypes (LumA, LumB, LumB-A, basal, HER2 and normal-like/claudin-low). MUC and NE displayed increased transcript and protein levels of mucin and neuroendocrine markers. Breast cancer cell lines, organoids and PDXs shared similar molecular subtypes with corresponding tumors, including the newly identified NE DU4475 (SYP+) and MUC MDAMB134 (INSM1+) cell lines which express NE protein markers. Additionally, we identified three non-tumor subtypes, immune dominate (ID), fibroblast (FB) or normal (N) in which transcriptomes were driven by adjacent normal cells. Multivariate analysis, adjusted for age, stage, TMB, tumor size, histology showed significant differences in overall survival compared to LumA (reference): LumB (HR=1.76, CI:1.27-2.44, p<0.001), basal (HR=2.96, CI:2.21-3.97, p<0.001), HER2 (HR=1.83, CI:1.3-2.56, p<0.001), BL2 (HR=2.3, CI:1.42-3.73, p<0.001), LAR (HR=1.91, CI:1.17-3.12, p=0.009), and STM (HR=4.83, CI:2.97-7.85, p<0.001). Furthermore, molecular subtypes showed differing therapeutic responses. LumB-A, MUC, and NE patients benefited more from combined chemo-endocrine therapy than endocrine therapy alone. ER+ tumors outside the luminal clusters were frequently ER-low and had worse outcomes with endocrine therapy alone and significantly benefited from added chemotherapy. ER+HER2+ patients within luminal clusters did not respond to HER2-targeted therapy. The STM cluster contained a mixture of ER+, HER2+ and TNBC tumors that were enriched in metaplastic spindle morphology and stem markers and displayed the worst outcomes, suggesting current treatment strategies are ineffective for this subtype. To validate our findings, we developed an integrated breast transcriptome classification (iBTC) signature that reproduced prognosis and classification in multiple datasets, predicting therapy response to immunotherapy and HER2-targeted therapy. Conclusions: Our comprehensive integrated analysis of RNA-seq data refines the current understanding of molecular subtyping, particularly the relationship between molecular signatures and histology-based classifications, provides insight into cell origin and identifies matching preclinical models to study alternative treatment strategies. These findings provide a rationale for revisiting clinical management for patients with rare subtypes and histology-molecular subtype mismatched tumors. Citation Format: X. Sun, P. I. Gonzalez-Ericsson, M. E. Sanders, H. An, H. Jin, Q. Sheng, J. M. Balko, J. A. Pietenpol, B. D. Lehmann. A Comprehensive Transcriptional Atlas of Breast Cancer Reveals Clinically Relevant Molecular Subtypes [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS2-12-03.