Development of a CD8+ T cell associated signature for predicting the prognosis and immunological characteristics of gastric cancer by integrating single-cell and bulk RNA-sequencing

Abstract

The universally poor clinical outcome makes gastric cancer (GC) still a significant public health threat, the main goal of our research is to develop a prognostic signature that can forecast the outcomes and immunological characteristics of GC via integrating single-cell and bulk RNA-sequencing. The CD8+ T cell feature genes were screened out by exploring single-cell RNA-sequencing (scRNA-seq) profiles retrieved from the TISCH2 database. Then, Cox and LASSO regressions were exploited for constructing a prognostic model in TCGA cohort based on these CD8+ T cell feature genes. Survival analysis was conducted to investigate the predictive capability of the signature for the clinical outcome of GC patients in TCGA and GEO cohorts. Additionally, we further examined the correlations between the risk signature and tumor immunotherapeutic response from the perspectives of immune infiltration, tumor mutation burden (TMB), immune checkpoint biomarker (ICB) expression, tumor microenvironment (TME), microsatellite instability (MSI), TIDE, and TCIA scores. In total, 703 CD8+ T cell feature genes were identified, eight of which were selected for constructing a prognostic signature. GC patients who possess high-risk score had significantly poorer survival outcomes than those who possess low-risk score in TCGA and GEO cohorts. Immune infiltration analysis proved that the risk score was negatively connected with the infiltration abundance of CD8+ T cells. Then, our findings demonstrated that GC patients in the high-risk subgroup possess a higher proportion of MSI-L/MSS, lower immune checkpoint biomarker expression, lower TMB, higher TIDE scores and lower TCIA scores compared to those in the low-risk subgroup. What's more, immunotherapy cohort analysis confirmed that patients who possess high-risk score are not sensitive to anti-cancer immunotherapy. Our study developed a reliable prognostic signature for GC that was significantly correlated with the immune landscape and immunotherapeutic responsiveness. The risk signature may guide clinicians to adopt more accurate and personalized treatment strategies for GC patients.

Keywords: Gastric cancer; Immune; Immunotherapy; Prognosis; Single-cell RNA-sequencing.

Figure 1

The flowchart for the entire study.

Figure 2

Identification of CD8+ T cell marker genes. (A) Cell clusters identified with marker genes for each cell type generated by TISCH2 in GSE167297. (B) Cell clusters identified with marker genes for each cell type generated by TISCH2 in GSE134520. (C) The Venn diagram indicated CD8+ T cell marker genes between GSE167297 and GSE134520 datasets. (D) GO analysis of CD8+ T cell marker genes. (E) KEGG analysis of CD8+ T cell marker genes.

Figure 3

Development of CD8+ T cell-associated risk signature. (A) Forest plot of survival related CD8+ T cell marker genes based on univariate Cox regression analysis. (B) Plots of the produced coefficient distributions for the logarithmic series for parameter selection (lambda). (C) The trajectory of each independent variable with lambda. (D) Forest plot of optimal prognostic genes used for the construction of the risk signature based on multivariate Cox regression analysis. (E) Kaplan–Meier survival curve of the risk signature in TCGA cohort. (F) ROC analysis of survival status for the risk signature and eight single CD8+ T cell marker genes. (G) Time-dependent ROC curve of the risk signature in TCGA cohort.

Figure 4

The correlation between the risk signature and clinical parameters. (A) Heat maps of clinical parameters between low- and high-risk groups. *P < 0.05; **P < 0.01. (B) The subgroup survival analysis according to the tumor stage. (C) Univariate Cox analysis of risk score and clinicopathological parameters. (D) Multivariate Cox analysis of risk score and clinicopathological parameters. (E) Multi-index ROC curve of risk score and clinicopathological parameters.

Figure 5

External validation of the risk signature in predicting overall survival of GC based on independent cohorts. (A) The Kaplan–Meier survival analysis of the risk signature in the GSE62254 cohort (n = 300). (B) The univariate and (C) multivariate Cox regression analyses of risk score and clinicopathological parameters in GSE62254 cohort. (D) The Kaplan–Meier survival analysis of the risk signature in the GSE15459 cohort (n = 192). (E) The univariate and (F) multivariate Cox regression analyses of risk score and clinicopathological parameters in GSE15459 cohort.

Figure 6

The development of a prognostic nomogram. (A) Nomogram model integrating the risk score and clinical parameters was constructed. (B) Calibration curve of the nomogram to predict the probability of 1-, 3-, and 5-year survival. (C) Multi-index ROC curve of nomogram model and other parameters. (D) DCA analysis showing the performance of the nomogram for predicting the 1- 3-, and 5-year survival.

Figure 7

Relevance exploration of risk signature with tumor somatic mutation. (A) Difference in TMB between low- and high-risk groups. (B) Spearman correlation analysis between risk score and TMB. (C) The Kaplan–Meier curve showing overall survival after combining the risk score with TMB. (D) Waterfall plots of top 20 mutated genes in the low-risk subgroup. (E) Waterfall plots of top 20 mutated genes in the high-risk subgroup.

Figure 8

The correlation between the risk signature and tumor immune microenvironment. (A) The lollipop plot of the relationship between risk score and TIICs in multiple databases, and the boxplot shows the difference of CD8+ T cell infiltration between different risk groups. (B) The boxplot of the differences in stromal score, immune score and ESTIMATE score between low- and high-risk groups.

Figure 9

Association of risk signature with immunotherapy sensitivity. (A) The differences in IPS between low- and high-risk groups. (B) The differences in TIDE scores between low- and high-risk groups. (C) Expression and prognostic value of PDCD1 in GC. (D) Expression and prognostic value of LAG3 in GC. (E, F) The correlation between the risk score and MSI status. (G) Kaplan–Meier survival analysis of the risk signature in IMvigor210 cohort. (H) The comparison of risk score between SD/PD and CR/PR groups in IMvigor210 cohort. *P < 005; **P < 0.01; ***P < 0.001.