Classification of colorectal cancer into clinically relevant subtypes based on genes and mesenchymal cells

Most studies on subtype identification of colorectal cancer (CRC) were based on expressions of either genes or immune cells. However, few studies have hitherto used the combination of genes with immune and stroma cells for subtype identification.

Dataset GSE17536 was obtained from the Gene Expression Omnibus (GEO) database. The xCell algorithm was used to estimate the composition and density of 64 cell types, including immune and stroma cell types. Clustering analysis was then conducted on the top 3000 most variable genes from a total of 20,174 genes for CRC subtype identification. We employed the ensemble method of Similarity network fusion and 112 Consensus Clustering (SNF-CC) for cancer subtype identification. Reactome pathway analysis was conducted to identify the impact of the representative genes on prognosis. The results were validated in independent gene expression data from dataset GSE17537.

In this study, we identified 3 clinically relevant subtypes and their representative genes, immune and stroma cells. Moreover, we confirmed the correlation of these subtypes with their clinical characteristics. The representative genes of the subtype with poor prognosis correlated with extracellular matrix structural constituent, while the subtype with good prognosis correlated with Toll-like receptor signaling pathway or chemokine signaling pathway. However, different subtypes were associated with distinct cell subtypes; the subtype with poor prognosis had a high abundance of fibroblasts and endothelial cells; the subtype with median prognosis had a higher abundance of immune cells, such as CD4 + T-cell, Th2 cells and aDC; the subtype with good prognosis had a higher abundance of NKT.

This study highlights the utility of immune and innate cells, especially during gene analysis, to provide the theoretical basis for personalized treatment in colorectal cancer patients.