A single-cell atlas of metastatic breast cancer charting oncogenic transcriptional programs in malignant cells and the tumor microenvironment

Metastatic breast cancer (MBC) remains a leading cause of cancer-related death among women due, to the ubiquitous evolution of drug-resistance. While recent studies have begun to elucidate the genomics of metastatic breast cancer (MBC), the transcriptional programs that drive the drug-resistant phenotype remain poorly understood.

We prospectively collected biopsies from patients with MBC augmented by detailed clinicopathologic features, including treatment and response characteristics. We profiled 29 biopsies by single-cell RNA seq, as well as bulk RNA-seq and whole-exome sequencing on an additional 207 biopsies (ongoing biopsies collection). We analyzed these data to generate an atlas, delineating the cell-types, cell-states, and transcriptional programs.

We profiled 100,470 single-cell transcriptomes and generated a comprehensive MBC atlas of the tumor and tumor microenvironment (TME). We next inferred cell types and programs associated with clinicopathologic characteristics. For example, we found significant differences in the TME of liver metastases compared to other sites, consistent with immunosuppression in the hepatic space. In particular, liver metastases were depleted in activated B-cells with lower expression of CXC chemokine receptors, and several activation-related chemokines, and their infiltrating T-cells expressed lower levels of effector and cytotoxicity markers (Odds-ratio (OR)=4.07, p-value (p)= 7.29e-07, PMID: 28052254) including CD8, CD3, beta chemokines, perforin, granulysin, and granzymes, and in addition - exceptionally low expression of TCR signaling genes and antigen processing genes including B2M and several HLA genes (OR=32.1, p=7.26e-07, PMID: 28052254).

To increase our power to make clinically relevant associations, we performed a joint analysis of the single-cell and bulk RNA-Seq data, to identify malignant programs related to specific oncogenic mutations, with implications for metastatic and drug-resistance phenotypes. For example, we characterized the oncogenic prog associated with activating estrogen receptor (ESR1) mutation (ESR-mut). As expected, ESR1-mut prog overlaps with many known ER and luminal B markers (OR=5.8, p=2.3e-11, PMID: 11823860). ESR1-mut prog also included specific Interferon-stimulated genes (ISGs) – IFI6, ISG15, IFIT1, STAT1, which are associated with tamoxifen resistance (OR=10.36, p=0.00036, PMID 17016442) and extracellular matrix-mediated regulation of apoptosis (OR=8.24, p=0.0028, PMID 17016442). These ISGs are predictive of poor prognosis among endocrine-treated patients (HR=1.69. p=1e-04, n=929, kmplotter). The ESR1-mut prog also included genes associated with cell-migration (SOX9, AGR2, TXNIP, and several S100 genes). This suggests a role for ESR1 mutation in pathogenicity, beyond ligand-independent activation of ER signaling. We similarly recovered additional mutation-specific oncogenic programs, including for RB1, TP53, GATA3, FOXA1, HER2, and FGFR mutants, forming a compendium of in-vivo oncogenic signatures.

To the best of our knowledge these data represent of the first integration of single-cell and bulk RNA-seq data in MBC, resulting in a comprehensive single-cell-resolution transcriptional atlas, and a catalog of drug-resistance oncogenic programs with implications for immunotherapy and precision-oncology.