Parallel single-cell and bulk transcriptome analyses reveal key features of the gastric tumor microenvironment

Abstract

Background: The tumor microenvironment (TME) has been shown to strongly influence treatment outcome for cancer patients in various indications and to influence the overall survival. However, the cells forming the TME in gastric cancer have not been extensively characterized.

Results: We combine bulk and single-cell RNA sequencing from tumors and matched normal tissue of 24 treatment-naïve GC patients to better understand which cell types and transcriptional programs are associated with malignant transformation of the stomach. Clustering 96,623 cells of non-epithelial origin reveals 81 well-defined TME cell types. We find that activated fibroblasts and endothelial cells are most prominently overrepresented in tumors. Intercellular network reconstruction and survival analysis of an independent cohort imply the importance of these cell types together with immunosuppressive myeloid cell subsets and regulatory T cells in establishing an immunosuppressive microenvironment that correlates with worsened prognosis and lack of response in anti-PD1-treated patients. In contrast, we find a subset of IFNγ activated T cells and HLA-II expressing macrophages that are linked to treatment response and increased overall survival.

Conclusions: Our gastric cancer single-cell TME compendium together with the matched bulk transcriptome data provides a unique resource for the identification of new potential biomarkers for patient stratification. This study helps further to elucidate the mechanism of gastric cancer and provides insights for therapy.

Keywords: Gastric cancer; Single-cell RNA transcriptomics; Tumor microenvironment.

Fig. 1

Cellular landscape of EPCAM-negative cells in non-malignant and malignant gastric patient samples. A Study overview: matched malignant and non-malignant stomach samples were obtained from a total of 24 patients. Samples were processed in parallel as bulk specimen through RNA and exome-sequencing and single cells through droplet RNA sequencing after depleting of EPCAM-positive cells. B UMAP of 96,623 cells, color coded for major cell type. C Dotplot showing the scaled average expression together with the percentage of expression of marker genes per major cell type

Fig. 2

Transcriptional reprogramming in cancer-associated fibroblasts. A UMAP of 16,492 fibroblast cells color coded for tissue (top) and cluster annotation (bottom). B Log ratio of average fraction per fibroblast cluster in tumor to normal tissue (n = 20) (top). Wilcoxon rank-sum test with holm correction, *: p < 0.05, **: p < 0.01, ***: p < 0.001. Dotplot showing the scaled average expression and the percentage of expression of top markers per fibroblast cluster (bottom). C Scaled average expression of fibroblast implicated gene clusters (Additional file 1: Fig. S4). D Top marker genes for F12-ANGPT2 and F13-CTHRC1 connected to gene cluster U6 and U7 respectively. E Overall survival of gastric cancer patients in TCGA, groups split by the top 20 marker gene signature of F13-CTHRC1. Gene signatures U1, U6, and U7 reflect gene clusters that were upregulated in gastric tumor samples while gene signatures D9 and D11 reflect gene clusters that were downregulated in gastric tumor samples

Fig. 3

Hybrid endothelial cell states in gastric cancer. A UMAP of 3684 endothelial cells color coded for tissue (left) and cluster annotation (right). B Log ratio of average fraction per endothelial cell cluster in tumor to normal tissue (n = 20) (top). Wilcoxon rank-sum test with holm correction, **: p < 0.01. Dotplot of top markers per cluster showing the scaled average expression and the percentage of expression (bottom). C Heatmap of differentially expressed genes between endothelial cells of non-malignant and malignant biopsies contained in indicated gene ontology terms. Shown is the scaled average expression and the percentage of expression. D, E Overall survival of gastric cancer patients in TCGA, groups split by SERPINE1 levels and top 20 marker genes of EN10 cluster, respectively.

Fig. 4

Pro- and anti-inflammatory macrophages are negatively correlated and highly diverse between gastric cancer patients. A UMAP of 10,646 myeloid cells color coded for tissue (bottom) and cluster annotation (top). B Bar plot of a proinflammatory (IL1B, CCL20, S100A8, S100A9) and anti-inflammatory (CD163, MAF, SEPP1, APOE) gene score per cell cluster (top). Dotplot of top markers per myeloid cluster showing the scaled average expression and the percentage of expression (bottom). C Log ratio of the average fraction per myeloid cluster annotation in tumor to normal tissue (n = 20). Wilcoxon rank-sum test with Holm correction, *: p < 0.05, **: p < 0.01, ***: p < 0.001. D Bar plot of the proinflammatory and anti-inflammatory gene signature from B in scRNA-seq data of our cohort (n = 20) and the TCGA-STAD cohort (n = 407). E Overall survival of gastric cancer patients in TCGA, groups split by expression level of the proinflammatory (left) and anti-inflammatory (right) gene signatures from B

Fig. 5

Immunosuppresive T cell dynamics in gastric cancer. A,B UMAP of 12,537 CD4⁺ T cells (A) and 28,772 CD8⁺ T cells (B) color coded for cluster annotation. C Pairwise analysis of T cell cluster fraction per patient in normal and tumor tissue, showing the most up- and downregulated T cells. Paired Wilcoxon rank-sum test, Holm-adjusted p-values per cluster shown. D Average expression of T cell implicated gene cluster U4 visualized on CD4⁺ T cell UMAP (Additional file 1: Fig. S3). E Heatmap of genes from gene cluster U4, color as scaled average expression

Fig. 6

Cell communication inferred from single-cell transcriptome profiles in gastric cancer show a central role of F13-Activated-CTHRC1 fibroblasts. A Closeness centrality ranking of all cell subtypes in the inferred cluster-wise cell communication network, top 10 subtypes are shown. B Communication strength of F13-Activated-CTHRC1 fibroblasts with other cell subtypes, only significant communication (connectivity > 0.05) are drawn. C Significant ligand-receptor pairs in F13-Myeloid communication highlight tumor-enriched activation of integrin-Akt pathway in myeloid cells by F13-CTHRC1 cells. D Significant ligand-receptor pairs in F13-Endothelial communication show a tumor-restricted program for intercellular communication. The X-axis represent ligand and receptor pairs, with the first gene expressed on F13-Act-CTHRC1 cells and second gene expressed on the interacting cell types denoted in the Y-axis. Red: upregulated gene in tumor bulk RNA-seq. Blue: downregulated in tumor bulk (none present). E Communication strength of F13-CTHRC1 with other cell subtypes, red line denotes the cluster-wise connectivity cutoff of 0.05 (see “Methods”)

Fig. 7

Deconvolution of bulk RNA-seq from an independent gastric cancer cohort reveal prognostic impact of cell subtypes. A–C Kaplan–Meier curve of TCGA Gastric Cancer patients with high or low F13-Activated-CTHRC1 (A), EN10-Activated-SERPINE1 (B), and M16-cDC-CLEC9A (C) scores. D Significant ligand-receptor pairs in M16-cDC-CLEC9A communication suggest CLEC9A + cDC1 supports antitumor immunity through XCL1-XCR1 signaling. The X-axis represents ligand and receptor pairs. The Y-axis represents cell subtypes interacting with M16-cDC-CLEC9A. E Kaplan–Meier curve of patients with high F13 score and low M16 score achieve best prognosis

Fig. 8

Cellular origin of response and non-response genes to cancer immunotherapy. A Differential expression analysis between responders (complete and partial responders, n = 12) and non-responders (stable and progressive disease, n = 33) performed on bulk RNA-seq from a total of 45 gastric cancer patients that underwent immunotherapy. Genes for downstream analysis (red) with adjusted p-value < 0.1 and log2 fold change > 0.25. B Up- and downregulated genes (A) were clustered based on their expression in minor cell types of the single-cell RNA-seq data (Additional file 1: Fig. S3). Here, a heatmap of the scaled average expression from the upregulated gene clusters is shown per major cell types of the single-cell RNA-seq on gastric cancer. C,D Annotation and UCell score of myeloid cells with gene cluster R8 (C) and fibroblasts with gene clusters NR10-13 (D) visualized on UMAP. E ROC curves of gene signature scores corresponding with specific cell subtypes on response data from cohort in A