Intratumor heterogeneity in human basal like breast cancer is revealed by single cell molecular subtyping

Breast cancer is a molecularly heterogeneous disease composed of multiple intrinsic subtypes. Recent studies have highlighted the substantial intratumor heterogeneity of breast cancer, wherein malignant cells of distinct intrinsic subtypes co-exist within the same tumor. However, most existing subtyping methods are designed for bulk transcriptomic data and are therefore limited in their ability to resolve such intratumor heterogeneity at single-cell resolution. We develop UBS93, a computational framework that enables robust molecular subtyping of both bulk tumor samples and individual breast cancer cells. We rigorously validate UBS93 and demonstrate its superior performance relative to existing approaches, particularly in identifying the highly aggressive Claudin-low subtype. Applying UBS93 to single-cell RNA sequencing data from human Basal-like breast cancers, we identify the co-existence of Basal-like and Claudin-low cancer cell populations within the same tumor—a form of intratumor heterogeneity previously observed in mouse models with genetically engineered RAS pathway alterations. Further analyses suggest that Claudin-low cancer cells originate from Basal-like population, with down-regulation of transcription factor ELF3 playing a pivotal role in the Basal-like/Claudin-low transition. Our findings establish UBS93 as a powerful tool for breast cancer subtyping and uncover the intratumor heterogeneity in human Basal-like breast cancer. Breast cancer is not a single disease and different subtypes of cancer cells can exist within the same tumor. Most current methods used to classify breast cancer look at tumors as a whole and cannot detect differences between individual cells. In this study, we developed a computational tool called UBS93 which can classify breast cancer using data from either whole tumor or single cells. Using UBS93, we found that two aggressive subtypes can exist together within the same tumor. We also discovered that one subtype may arise from the other through changes in gene activity. These findings improve our understanding of breast cancer diversity and may help guide more precise diagnosis and treatment in the future. Li, Gao et al. introduce UBS93, a computational framework which enables breast cancer molecular subtyping at both bulk and single-cell resolution. Application of UBS93 to cell line and patient single-cell transcriptomic data reveals the co-existence and lineage relationship of Basal-like and Claudin-low cancer cells, providing insights into tumor evolution and heterogeneity.