Testing the Gene Expression Classification of the EMT Spectrum

Abstract

The epithelial-mesenchymal transition (EMT) plays a central role in cancer metastasis and drug resistance - two persistent clinical challenges. Epithelial cells can undergo a partial or full EMT, attaining either a hybrid epithelial/mesenchymal (E/M) or mesenchymal phenotype, respectively. Recent studies have emphasized that hybrid E/M cells may be more aggressive than their mesenchymal counterparts. However, mechanisms driving hybrid E/M phenotypes remain largely elusive. Here, to better characterize the hybrid E/M phenotype(s) and tumor aggressiveness, we integrate two computational methods - (a) RACIPE - to identify the robust gene expression patterns emerging from the dynamics of a given gene regulatory network, and (b) EMT scoring metric - to calculate the probability that a given gene expression profile displays a hybrid E/M phenotype. We apply the EMT scoring metric to RACIPE-generated gene expression data generated from a core EMT regulatory network and classify the gene expression profiles into relevant categories (epithelial, hybrid E/M, mesenchymal). This categorization is broadly consistent with hierarchical clustering readouts of RACIPE-generated gene expression data. We show that the EMT scoring metric can be used to distinguish between samples composed of exclusively hybrid E/M cells and those containing mixtures of epithelial and mesenchymal subpopulations using the RACIPE-generated gene expression data.