Identification and external validation of a prognostic signature associated with DNA repair genes in gastric cancer

Abstract

The aim of this study was to construct and validate a DNA repair-related gene signature for evaluating the overall survival (OS) of patients with gastric cancer (GC). Differentially expressed DNA repair genes between GC and normal gastric tissue samples obtained from the TCGA database were identified. Univariate Cox analysis was used to screen survival-related genes and multivariate Cox analysis was applied to construct a DNA repair-related gene signature. An integrated bioinformatics approach was performed to evaluate its diagnostic and prognostic value. The prognostic model and the expression levels of signature genes were validated using an independent external validation cohort. Two genes (CHAF1A, RMI1) were identified to establish the prognostic signature and patients ware stratified into high- and low-risk groups. Patients in high-risk group presented significant shorter survival time than patients in the low-risk group in both cohorts, which were verified by the ROC curves. Multivariate analysis showed that the prognostic signature was an independent predictor for patients with GC after adjustment for other known clinical parameters. A nomogram incorporating the signature and known clinical factors yielded better performance and net benefits in calibration plot and decision curve analyses. Further, the logistic regression classifier based on the two genes presented an excellent diagnostic power in differentiating early HCC and normal tissues with AUCs higher than 0.9. Moreover, Gene Set Enrichment Analysis revealed that diverse cancer-related pathways significantly clustered in the high-risk and low-risk groups. Immune cell infiltration analysis revealed that CHAF1A and RMI1 were correlated with several types of immune cell subtypes. A prognostic signature using CHAF1A and RMI1 was developed that effectively predicted different OS rates among patients with GC. This risk model provides new clinical evidence for the diagnostic accuracy and survival prediction of GC.

Figure 1

Prognostic DNA repair genes identification and signature construction in the TCGA cohort. (A) The volcano plot of the differentially expressed genes between GC and normal samples; (B) Univariate Cox regression analysis identifying prognostic variables with HR with 95% CI and P values; (C) Prognostic signature construction using multivariate Cox regression analysis.

Figure 2

Kaplan–Meier survival analysis of the signature risk score between the high- and low-risk groups. Survival differences in the TCGA cohort (A), and the GSE66229 validation cohort (B).

Figure 3

Prognostic value of the two genes signature for prediction of overall survival of patients with GC. (A) ROC curve analysis for predicting survival in patients with GC according to the risk score in the TCGA cohort; (B) From top to bottom are the risk score, patients’ survival status distribution, and the expression heat map of two genes in the low- and high-risk groups in the TCGA cohort; (C) ROC curve analysis for predicting survival in patients with GC according to the risk score in the GSE66229 cohort; (D) From top to bottom are the risk score, patients’ survival status distribution, and the expression heat map of two genes in the low- and high-risk groups in the GSE66229 cohort. A heat map was generated using the “pheatmap” package (version 1.0.12; https://cran.r-project.org/web/packages/pheatmap/index.html) of the R software (version 3.6.3).

Figure 4

Nomogram construction based on the DNA repair gene signature. (A) Nomogram predicting overall survival probability for patients with GC; Assign the points of each variable of the patient by drawing a vertical line from that variable to the points scale, next, sum all the points, and draw a vertical line from the total points’ scale to the 1-, 3-, and 5-year OS to obtain the probability of death. (B) Calibration plots for the nomogram; Nomogram-predicted OS is plotted on the x-axis, and actual OS is plotted on the y-axis. A plot along the 45° line would present a perfect calibration model in which the predicted probabilities are identical to the actual outcomes. (C) decision curve analyses comparing nomogram and AJCC stage; the net benefit was plotted versus the threshold probability.

Figure 5

Validation of expression pattern of two identified genes in the validation cohort and the diagnostic performance of signature genes in distinguishing GC from normal samples. The expression changes of CHAF1A (A) and RMI1 (B) in the GSE66229 cohort; The ROC curves of the two genes-based diagnostic classifier in the TCGA cohort (C) and the independent GSE66229 cohort (D); ROC curves of the diagnostic classifier for stage I patients with GC in the TCGA cohort (E) and the GSE66229 cohort (F).

Figure 6

Distribution and visualization of immune cell infiltration in patients with GC and the correlation between two DNA repair genes. Summary of estimated compositions of 22 immune cell subtypes from the CIBERSORT algorithm in GC patients (A); Comparison of 22 immune cell subtypes between low- and high-risk samples (B). The correlation between CHAF1A (C) and RMI1 (D) and infiltrating immune cells in patients with GC.

Figure 7

GSEA illustrated the significantly altered biological processes in high-risk group and low-risk group in the TCGA cohort.