An integrated breast cancer transcriptomic atlas defines clinical subtypes and informs cellular origin

Efforts to analyze breast cancer transcriptomes at scale are hindered by cross-study technical variability. To overcome this challenge, we adapt single-cell RNA sequencing analytical strategies to harmonize 7,612 bulk breast cancer transcriptomes from 22 datasets, generating a high-resolution atlas that resolves both established and unrecognized malignant transcriptional states. This framework recapitulates intrinsic subtypes and microenvironmental programs while identifying stem-like, basal-luminal progenitor (BLP), mucinous, and neuroendocrine-differentiated (NED) subtypes with distinct clinical behaviors. The NED subtype exhibits luminal characteristics, neuroendocrine differentiation, and poor prognosis after chemo-endocrine therapy. The human epithelial growth factor receptor-2-enriched (HER2E) kinase inhibitor-sensitive (HER2E^KIS) subtype identifies patients who responded to neoadjuvant HER2-targeted kinase inhibition. Transcriptomic positioning of BLP tumors suggests bipotent or myoepithelial progenitor association, unfavorable prognosis after chemotherapy, and potential benefit from VEGF and receptor tyrosine kinase inhibitors. Together, these data indicate that this atlas refines breast cancer taxonomy, provides insight into tumor cell states and origins, and establishes a framework for patient stratification and therapeutic discovery.