Identification of immune cells and mRNA associated with prognosis of gastric cancer

We download gene expression data, somatic mutation data and clinical data of stomach adenocarcinoma (STAD) from the cancer genome atlas (TCGA) database by TCGAbiolinks and maftools packages in R (3.5.1, [11, 12]). In order to verify the results of the study in the TCGA data, gene expression profile and clinical data in GSE84437 were downloaded from the Gene Expression Omnibus (GEO) database. In the TCGA dataset, samples with death reason of other malignancy, other and non-malignant disease, sample type is not “Primary Tumor”, and samples with incomplete overall survival information were excluded, total 360 samples were included finally. The GEO dataset included 433 gastric cancer tissues. For further investigate the underlying mechanisms in digestive system tumors, gene transcripts per million (TPM) data of Pan-cancer in TCGA and normal tissues in genotype tissue expression (GTEx) database were downloaded from the UCSC Xena database, which processed by the TOIL process, free of computational batch effects. All analyses and plots are done by R (3.6.0).

Gastric cancer patients died of non-malignant disease and other malignancies were excluded, and samples with complete survival data were included. According to Gabriela Bindea et al., we obtained the marker genes of 24 immune cells, including aDC, B cells, CD8 T cells, Cytotoxic cells, DC, Eosinophils, iDC, Macrophages, Mast cells, Neutrophils, NK CD56 bright cells, NK CD56 dim cells, NK cells, pDC, T cells, T helper cells, Tcm, Tem, TFH, Tgd, Th1 cells, Th17 cells, Th2 cells and TReg [10] . Then, based on the gene expression data and marker genes, infiltrating immune cells were quantified by ssGSEA.

The association between immune cells and overall survival was carried out using univariate Cox regression, immune cells with statistically significant(P < 0.05) in both groups be considered as effects of prognosis. For further evaluate the impact of the immune cells with statistically significant, patients were divided into 2 groups according to the method of best separation in “survminer” R package, then, overall survival were analyzed by “survival” R package. Kaplan Meier-plotter (KM plotter, http://​kmplot.​com/​analysis/​) could assess the effect of hub genes on survival [13] . The hazard ratio (HR) with 95% confidence intervals and log rank P value were calculated and displayed on the plot.

Gastric cancer is one of digestive system tumor, we further compare the differences of immune cells and hub gene expression between digestive tumors and normal tissue. Gene transcripts per million (TPM) of digestive system normal and tumors tissues were downloaded from UCSC Xena (https://​xenabrowser.​net/​datapages/​), Normal tissue data is from Genotype tissue expression (GTEx) database, tumor tissue data is from TCGA database, and infiltrating immune cells were quantified by ssGSEA.

Gene counts of TCGA-STAD were downloaded by “TCGAbiolinks” R package, patients were divided into 2 groups according to the expression of hub genes by method of best separation. Then, differentially expressed genes (DEGs) screened between the high and low group, gene set enrichment analysis (GSEA) [14] and enrichment analysis were performed with “clusterProfiler” package in R [15]. We use “GOSemSim” package to calculate the similarity between Gene Ontology (GO) terms and then plot it with “ggtree” package.

After quantification of infiltrating immune cells, univariate Cox regression was used to screen immune cells that affect prognosis. Results of TCGA and GEO datasets were shown in Fig. 1a. Infiltration of Th2 cells, T helper cells and Mast cells (P < 0.05) related to the survival of patients with gastric cancer in two datasets, and the three immune cells showed the same effect. Th2 cells and T helper cells were protective factors, and Mast cells were risk factors. The survival plot based on the best separation of high and low infiltration of each immune cell in TCGA and GEO datasets. As shown in Fig. 1b, patients with higher each protective immune cells showed a significantly higher overall survival rate, patients with higher risk immune cell showed a significantly lower overall survival rate.

To further clarify the regulatory relationship of mRNA and prognostic immune cells, methods of Pearson correlation coefficient and Spearman’s rank correlation coefficient were used to calculate the correlation between mRNA and prognostic immune cells. By two methods, genes under the threshold of P < 0.01 and correlation> 0.3 were selected. In TCGA group, 4844 genes which associated with Mast cells, 2160 genes which associated with T helper cells and 2984 genes which associated with Th2 cells were screened out, in GEO group, 2128 genes which associated with Mast cells, 372 genes which associated with T helper cells and 1590 genes which associated with Th2 cells were screened out. SUPV3L1 and SLC22A17 (Fig. 1c), they are considered as the hub genes that associated with infiltration of three prognostic immune cells. The survival plot based on the best separation of high and low expression of hub genes in TCGA and GEO datasets. Thus, SUPV3L1 expression was used to divide patients into SUPV3L1^high (172 samples) and SUPV3L1^low(188 samples) groups and SUPV3L1 expression was used to divide patients into SLC22A17^high (243 samples) and SLC22A17^low(117 samples) groups. As shown in ​Fig. 2a, patients with higher expression of SUPV3L1 showed significantly higher overall survival rate, patients with higher expression of SLC22A17 showed significantly lower overall survival rate. Prognostic value of SUPV3L1 and SLC22A17 in gastric cancer patients were reconfirmed by Kaplan Meier-plotter (KM plotter, http://​kmplot.​com/​analysis/​). It was found that expression of SLC22A17 (HR = 1.58 (1.18–2.12), P = 0.0022) was associated with worse overall survival (OS) for gastric cancer patients,expression of SUPV3L1 (HR = 0.6 (0.45–0.8) P = 0.00034) was associated with good overall survival (OS) for gastric cancer patients (Fig. 2b).

To show the correlation of hub genes and prognostic immune cells, we calculated the correlation by methods of Pearson correlation coefficient and plotted (Fig. 2c). SUPV3L1 was positively correlated with Th2 cells and T helper cells, and negatively correlated with Mast cells, SLC22A17 was exactly the opposite.

Tumor mutational burden of gastric cancer in TCGA were downloaded by “TCGAbiolinks” R packages. According to hub genes expression and infiltration of prognostic immune cells, 322 samples were divided into groups of high and low. There was a significant difference (P < 0.05) in tumor mutational burden between patients in the high and low group, high expression of SUPV3L1 and infiltration of Th2 cells and T helper cells and low expression of SLC22A17 and infiltration of Mast cells coupled with a high mutational burden (Fig. 2d). It may indicate that a high mutational load coupled with high infiltration of Th2 cells and T helper cells and low infiltration of Mast cells, TMB-H (high tumor mutation load) patients produce more new antigens, and tumors are attacked by a large number of Th2 cells and T helper cells.

Digestive system tumor compared with paracancer, there was a significant difference (P < 0.05) in the expression of SUPV3L1 and SLC22A17 (Fig. 2e). Through comparing with infiltration of prognostic immune cells in the tumor and normal tissues, we have summarized that Mast cells and Th2 cells in the digestive system normal tissues and tumor were different. Compared to the normal tissues, the population of Th2 cells in tumor had more, but Mast cells and T helper cells in the tumor were less than normal tissues, where liver seems different, Mast cells in normal liver tissue nearly no, and liver tumor tissue has a small amount of Mast cells infiltration (Fig. 3).

According to the groups of expression of SUPV3L1 and SLC22A17, difference analysis was used to investigate the biological role of SUPV3L1 and SLC22A17. To obtain further insight into the function of the hub gene, GSEA was conducted by “clusterProfiler” R package. Six representative pathways about SUPV3L1 were “Dilated cardiomyopathy (DCM)”, “ECM-receptor interaction”, “Glycosaminoglycan biosynthesis-chondroitin sulfate/dermatan sulfate”, “Malaria”, “Protein digestion and absorption” and “Regulation of lipolysis in adipocytes”, Six representative pathways about SLC22A17 were “Apelin signaling pathway”, “Cushing syndrome”, “MAPK signaling pathway”, “Oxytocin signaling pathway”, “PI3K-Akt signaling pathway” and “Platelet activation” (Fig. 4). 3 up-regulated and 236 down-regulated genes (| log2foldchange | > 2.5, P < 0.01) were identified significantly associated with SLC22A17 expression (Fig. 5), 125 up-regulated and 30 down-regulated genes (| log2foldchange | > 1.8, P < 0.01) were identified significantly associated with SUPV3L1 expression (Fig. 5). In order to explore biological relevance of differential genes, significantly differentially expressed genes were enriched using “clusterProfiler” R package for Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses. Biological processes were grouped according to functional theme, it suggested to focus on “transport”, “regulation”, “contraction” and “circulatory system and pain” (Fig. 6a), KEGG analyses showed that hub genes may be related to “insulin secretion”, “cGMP-PKG signaling pathway”, “cAMP signaling pathway”, “calcium signaling pathway” and “neuroactive ligand-receptor interaction” (Fig. 6b).