Molecular phenotyping of breast cancer cell lines at the single-cell level for automated cancer diagnosis and prediction of drug sensitivity

Brest Cancer (BC) patient stratification is mainly driven by receptor status and histological grading and subtyping, with about twenty percent of patients for which absence of any actionable biomarkers results in no clear therapeutic intervention to apply. Here, we evaluated the potentiality of single-cell transcriptomics for automated diagnosis and drug treatment of breast cancer. We transcriptionally profiled 35,276 individual cells from 32 BC cell-lines covering all main BC subtypes to yield a Breast Cancer Single Cell Atlas. We show that single cell transcriptomics can successfully detect clinically relevant BC biomarkers and that atlas can be used to automatically predict cancer subtype and composition from a patients tumour biopsy. We found that BC cell lines arbour a high degree of heterogeneity in the expression of clinically relevant BC biomarkers and that such heterogeneity enables cells with differential drug sensitivity to co-exist even within a genomically stable isogenic cell line. Finally, we developed a novel bioinformatics approach named DREEP (DRug Estimation from Expression Profiles) to automatically predict responses to more than 450 anticancer agents starting from single-cell transcriptional profiles. We validated DREEP both in-silico and in-vitro by selectively inhibiting the growth of the HER2-deficient subpopulation in the MDAMB361 cell line. Our work shows transcriptional heterogeneity is common, dynamic and plays a relevant role in determining drug sensitivity. Moreover, our Breast Cancer Single Cell Atlas and DREEP approach are a unique resource for the BC research community and to advance the use of single-cell sequencing in the clinics.