Development and Validation of a Robust Immune-Related Prognostic Signature for Gastric Cancer

Abstract

Background: An increasing number of reports have found that immune-related genes (IRGs) have a significant impact on the prognosis of a variety of cancers, but the prognostic value of IRGs in gastric cancer (GC) has not been fully elucidated.

Methods: Univariate Cox regression analysis was adopted for the identification of prognostic IRGs in three independent cohorts (GSE62254, n = 300; GSE15459, n = 191; and GSE26901, n = 109). After obtaining the intersecting prognostic genes, the three independent cohorts were merged into a training cohort (n = 600) to establish a prognostic model. The risk score was determined using multivariate Cox and LASSO regression analyses. Patients were classified into low-risk and high-risk groups according to the median risk score. The risk score performance was validated externally in the three independent cohorts (GSE26253, n = 432; GSE84437, n = 431; and TCGA, n = 336). Immune cell infiltration (ICI) was quantified by the CIBERSORT method.

Results: A risk score comprising nine genes showed high accuracy for the prediction of the overall survival (OS) of patients with GC in the training cohort (AUC > 0.7). The risk of death was found to have a positive correlation with the risk score. The univariate and multivariate Cox regression analyses revealed that the risk score was an independent indicator of the prognosis of patients with GC (p < 0.001). External validation confirmed the universal applicability of the risk score. The low-risk group presented a lower infiltration level of M2 macrophages than the high-risk group (p < 0.001), and the prognosis of patients with GC with a higher infiltration level of M2 macrophages was poor (p = 0.011). According to clinical correlation analysis, compared with patients with the diffuse and mixed type of GC, those with the Lauren classification intestinal GC type had a significantly lower risk score (p = 0.00085). The patients' risk score increased with the progression of the clinicopathological stage.

Conclusion: In this study, we constructed and validated a robust prognostic signature for GC, which may help improve the prognostic assessment system and treatment strategy for GC.

Figure 1

Gene cluster analysis. (a) Venn plot of the 45 intersecting prognosis-related genes. (b) Heatmap of gene clusters. (c) Kaplan–Meier survival analysis regarding gene clusters and OS in the training cohort.

Figure 2

Construction of a nine-gene prognostic signature. (a, b) Kaplan–Meier survival analysis and time-dependent ROC analysis regarding risk score and OS in the training cohort. (c, d) Prognostic signature visualized as a nomogram and tested by the calibration curve.

Figure 3

Heatmap, risk score distribution, and survival status of patients in the training cohort.

Figure 4

Internal validation of the prognostic signature. (a–c) Kaplan–Meier survival analysis and the time-dependent ROC analysis of the risk score and gene cluster for predicting OS in the GSE62254 cohort. (d–f) Kaplan–Meier survival analysis and the time-dependent ROC analysis of the risk score and gene cluster for predicting OS in the GSE15459 cohort. (g–i) Kaplan–Meier survival analysis and the time-dependent ROC analysis of the risk score and gene cluster for predicting OS in the GSE26901 cohort.

Figure 5

External validation of the prognostic signature. (a–c) Kaplan–Meier survival analysis of the prognostic signature for predicting OS in the GSE84437, GSE26253, and TCGA cohorts. (d–f) The risk score distribution and the survival status of patients in the GSE84437, GSE26253, and TCGA cohorts.

Figure 6

The immune cell infiltration landscape of all included samples. (a, b) Heatmap and the boxplot showing the difference in immune cell infiltration in different risk groups. (c) Kaplan–Meier survival analysis regarding immune cell infiltration and OS.

Figure 7

The correlation analysis between the risk score and clinical characteristics. (a) The Kaplan–Meier survival analysis regarding Lauren classification and OS. (b) Barplot of proportions of different Lauren classification subtypes in high- and low-risk groups. (c) Boxplot of the risk score difference in different Lauren classification subtypes. (d) Kaplan–Meier survival analysis regarding stage and OS. (e) Barplot of proportions of different stages in high- and low-risk groups. (f) Boxplot of the risk score difference in different stages.

Figure 8

Gene set enrichment analysis for immune-related pathways in different risk groups.

Figure 9

The presentation of ESTIMATE calculation results.

Figure 10

The Kaplan–Meier survival analysis of Stromal Score, Immune Score, and ESTIMATE Score in the six independent cohorts.

Figure 11

The association between risk score and the Stromal Score, Immune Score, and ESTIMATE Score.