Integrated single-cell potency and expression landscape in mammary epithelium reveals novel bipotent-like cells associated with breast cancer risk

Abstract

The identification of progenitor and stem like cells in epithelial tissues, as well as those that may serve as the cell of origin for epithelial cancers, is an outstanding challenge. Here we present a novel algorithm, called LandSCENT, which constructs a 3-dimensional integrated landscape of cell-states, encompassing cell-potency and expression subtypes, to facilitate the identification of progenitor and stem-like cells. Application to thousands of single-cell RNA-Seq profiles from the normal mammary epithelium reveals a rare 5% subpopulation of highly potent single-cells. The integrated landscape naturally predicts that these cells define a bi-potent-like state, a result not obtainable via standard methods or without invoking prior assumptions. The bi-potent-like cells are overrepresented within the basal compartment but also overlap with an immature luminal phenotype. We characterize the transcriptome of these cells and show that is enriched for a mammary stem-cell module. We further identify YBX1, a regulator of breast cancer risk identified from GWAS, as the key transcription factor defining this candidate bi-potent cellular phenotype. We validate the putative bi-potency of YBX1-marked cells using independent FACS-sorted bulk expression data. In addition, YBX1 is overexpressed in basal breast cancer and correlates with clinical outcome. In summary, we here provide a novel computational framework which may serve to identify and prioritize candidate normal or cancer progenitor/stem-like single-cell phenotypes, for subsequent functional studies.