Decoupling epithelial-to-mesenchymal transitions from stromal profiles by integrative analysis

Introduction

Epithelial-to-mesenchymal transition (EMT) is the most commonly cited mechanism for cancer metastasis, but it is difficult to distinguish from expression profiles of normal stromal cells in the tumour microenvironment. In this study we compared expression of mesenchymal signature genes in cancer cells and stromal cells using single cell RNA-seq data for several tumour types. We then developed a method to deconvolve the mesenchymal signature in bulk expression profiles into stromal and cancer-cell-specific EMT components. We applied this method to bulk RNA-seq data for hundreds of samples from many cancer types, and examined the common properties of the resulting EMT signatures and their association with clinical features.

Materials and Methods

We used single cell RNA-seq data from several published studies to examine expression of mesenchymal signature genes in different cell types. By aggregating samples of single cells from these datasets, we constructed simulations of bulk expression profiles, which we used to test our deconvolution method. We applied this method to bulk RNA-seq data from The Cancer Genome Atlas (TCGA), and tested the resulting EMT signatures for association with prognostic features using the TCGA clinical annotations.

Results and Discussion

We showed that many classical EMT marker genes are more strongly associated with fibroblasts than with cancer cells, indicating that their expression levels in bulk profiles primarily reflects stromal content. Other genes, including several laminins and integrins, proved to be more reliable indicators of EMT in cancer cells. The EMT signatures correlated with metastasis in only a few cancer types, and in some cases they showed association with other clinical features, such as therapy resistance.

Conclusion

Through our pan-cancer deconvolution analysis of bulk expression data, we showed that classical EMT marker genes often primarily reflect stromal content, while our inferred cancer-cell-specific EMT signatures usually do not correlate with metastasis. This study demonstrated the importance of distinguishing ‘true’ EMT from stromal contributions in order to elucidate the therapeutic relevance of EMT in cancer.