Refined cellular activity expression signatures provide a targeted framework to quantify phenotypic intra-tumor heterogeneity in single-cell data

Abstract

Single cell RNA-seq (scRNA-seq) now allows deeper insight into cellular biology at both the individual and population level. Measuring cell-to-cell variations in the population enables quantification of phenotypic heterogeneity in populations in which cell states and identities deviate from healthy transcriptomic profiles. Cellular activities quantifiable using gene set enrichment analyses can provide useful grounds to quantify phenotypic heterogeneity, but the specificity and adequacy of existing molecular signatures for scRNA-seq data is still insufficient. Here we induced 6 activities in vitro, for which we refined existing expression signatures to enhance specificity and detection in scRNA-seq data: epithelial-mesenchymal transition (EMT), DNA repair, responses to interferons and {gamma} (IFN and IFN{gamma}, respectively), glycolysis, oxidative phosphorylation (OxPhos). We report new signatures, with much lower redundancy between IFN and IFN{gamma}, and glycolysis and OxPhos signatures, achieving average AUCs of 0.85 across bootstrapped datasets for each activity. We could use these signatures to quantify phenotypic intra-tumor heterogeneity (ITH) in 20 patient samples and 14 cell lines, observing high correlation with diversity indices in classified healthy cells (p<0.001). Focusing on cancer cells only, we furthermore report higher phenotypic ITH in patients than in cell lines (p<0.001), and in basal tumors (p=0.028).