Integration of scRNA-Seq and Bulk RNA-Seq to Analyse the Heterogeneity of Ovarian Cancer Immune Cells and Establish a Molecular Risk Model

Background Considerable evidence suggests that the heterogeneity of ovarian cancer (OC) is a major cause of treatment failure. Single-cell RNA sequencing (scRNA-seq) is a powerful tool to analyse the heterogeneity of the tumour at the single-cell level, leading to a better understanding of cell function at the genetic and cellular levels. Methods OC scRNA-seq data were extracted from the Gene Expression Omnibus (GEO) database and the FindCluster () package used for cell cluster analysis. The GSVA package was used for single-sample gene set enrichment analysis (ssGSEA) analysis to obtain a Hallmark gene set score and bulk RNA-seq data were used to analyse the key genes of OC-associated immune cell subsets. CIBERSORT was used to identify immune scores of cells and the “WGCNA” package for the weighted correlation network analysis (WGCNA). KEGG (Kyoto Encyclopedia of Genes and Genomes) and GO (Gene Ontology) analyses of subtype groups were performed by GSEA. Then, univariate Cox and lasso regression were performed to further establish a signature. Finally, qPCR and immunohistochemistry staining were used to evaluate the expression of signature genes in OC. Results Two scRNA-seq (GSE154600 and GES158937) datasets were integrated to obtain 20 cell clusters. T cells or NK cells (cluster 5, 6, 7, 11), B cells (cluster 16, 19, 20) and myeloid cells (cluster 4, 9, 10) were clustered according to immune cell markers. The ssGSEA revealed that M1- and M2-like myeloid cell-related genes were significantly upregulated in P3 and P4 patients in the GSE154600 data. Immune cell analysis in TCGA-OC showed that a high abundance of M1-like tumour-associated macrophages (TAMS) predicts better survival. WGCNA, univariate Cox and lasso Cox regression established a two-gene signature (RiskScore=-0.059*CXCL13-0.034*IL26). Next, the TCGA-test and TCGA-OC were used to test the risk prediction ability of the signature, showing a good effect in the datasets. Moreover, the qPCR and immunohistochemistry staining revealed that the expression of CXCL13 and IL26 was reduced in OC tissues. Conclusion A two-gene signature prognostic stratification system (CXCL13 and IL26) was developed based on the heterogeneity of OC immune cells to accurately evaluate the prognostic risk.