A novel signature based on CeRNA and immune status predicts prognostic risk and drug sensitivity in gastric cancer patients

Abstract

Background: At present, there is increasing evidence that both competitive endogenous RNAs (ceRNAs) and immune status in the tumor microenvironment (TME) can affect the progression of gastric cancer (GC), and are closely related to the prognosis of patients. However, few studies have linked the two to jointly determine the prognosis of patients with GC. This study aimed to develop a combined prognostic model based on ceRNAs and immune biomarkers.

Methods: First, the gene expression profiles and clinical information were downloaded from TCGA and GEO databases. Then two ceRNA networks were constructed on the basis of circRNA. Afterwards, the key genes were screened by univariate Cox regression analysis and Lasso regression analysis, and the ceRNA-related prognostic model was constructed by multivariate Cox regression analysis. Next, CIBERSORT and ESTIMATE algorithms were utilized to obtain the immune cell infiltration abundance and stromal/immune score in TME. Furthermore, the correlation between ceRNAs and immunity was found out through co-expression analysis, and another immune-related prognosis model was established. Finally, combining these two models, a comprehensive prognostic model was built and visualized with a nomogram.

Results: The (circRNA, lncRNA)-miRNA-mRNA regulatory network of GC was constructed. The predictive power of ceRNA-related and immune-related prognosis models was moderate. Co-expression analysis showed that the ceRNA network was correlated with immunity. The integrated model of combined ceRNAs and immunity in the TCGA training set, the AUC values of 1, 3, and 5-year survival rates were 0.78, 0.76, and 0.78, respectively; in the independent external validation set GSE62254, they were 0.81, 0.79, and 0.78 respectively; in GSE15459, they were 0.84, 0.88 and 0.89 respectively. Besides, the prognostic score of the comprehensive model can predict chemotherapeutic drug resistance. Moreover, we found that plasma variant translocation 1 (PVT1) and infiltrating immune cells (mast cells) are worthy of further investigation as independent prognostic factors.

Conclusions: Two ceRNA regulatory networks were constructed based on circRNA. At the same time, a comprehensive prognosis model was established, which has a high clinical significance for prognosis prediction and chemotherapy drug selection of GC patients.

Keywords: CircRNA; competing endogenous RNA network; gastric cancer; immunocyte infiltration; mast cell; plasma variant translocation 1; prognostic model.

Figure 1

Volcanic maps and heat maps of circRNAs expression. The differential expression heat maps of GSE78092 (A) and GSE83521 (B). The differential expression volcanic maps of GSE78092 (C) and GSE83521 (D). Highly expressed circRNAs in the intersection of GSE78092 and GSE83521 (E). Low expression circRNAs in the intersection of GSE78092 and GSE83521 (F).

Figure 2

Construction of ceRNA prognostic models. circRNA-miRNA-mRNA regulatory network (A). (circRNA, lncRNA)-mRNA regulatory network (B). (circRNA, lncRNA)-miRNA-mRNA regulatory network (C). Red represents up regulation, blue represents down regulation.

Figure 3

Construction and validation of ceRNA related prognostic model. (A, B) Lasso regression analysis was used to simplify the prognostic model. (C) Correlation between ceRNAs and prognosis. (D) K-M survival analysis of prognostic model. (E) ROC curve of prognostic model. (F-H) Survival status of gastric cancer patients with different risk scores.

Figure 4

RCAN2 may adjust TGF- β signaling pathway. (A) RCAN2 was positively correlated with TGFB1-3 and TGFBR1-3 at the expression level. (B) RCAN2 was positively correlated with SMAD1,4,5,7,9 at the expression level.

Figure 5

Correlation between ceRNAs and immune microenvironment of gastric cancer. (A) GSEA analysis of high-risk group divided by ceRNA related prognostic model. (B) Correlation analysis between ceRNAs and infiltrating immune cells. (C) Correlation analysis between ceRNAs and ImmuneScore/StromalScore.

Figure 6

Correlation between risk-score of ceRNA related prognosis model and immune microenvironment. (A1-3) Correlation analysis between risk score and ImmuneScore/StromalScore of gastric cancer. (B1-4) Correlation analysis between risk-score and abundance of invasive immune cells in gastric cancer.

Figure 7

Somatic mutation map of 5 ceRNAs. (A) Mutation frequency and expression of 5 ceRNAs. (B, C) Mutation patterns of VCAN and RCAN2.

Figure 8

Construction and validation of comprehensive prognostic model. (A) Multivariate Cox model was used to construct a comprehensive prognosis model. (B1-2) ROC curve and K-M survival analysis of training set. (C1-2, D1-2) ROC curve and K-M survival analysis of GSE62254 and GSE15459 external validation set.

Figure 9

Nomogram and validation of comprehensive prognostic model. (A) Nomogram of the comprehensive prognostic model. (B1–3) The calibration curves for the nomogram.

Figure 10

(A, B) The MSI levels of patients in high-risk groups were different. (C) Risk grouping is related to clinical indicators of patients. * Indicates P < 0.05. ** Indicates P < 0.01. *** Indicates P < 0.001. Ns indicates not statistically significant.

Figure 11

Prediction of chemotherapeutic drug resistance in gastric cancer by comprehensive prognosis model. (A) Correlation between five RNAs and chemotherapeutic drug resistance in gastric cancer. (B-G) The risk score obtained by the comprehensive model can predict the drug resistance of six common chemotherapeutic drugs.

Figure 12

Protein expression levels and cellular spatial localization of VCAN and RCAN2. (A) Immunohistochemical experiments of VCAN and RCAN2 in gastric cancer and normal tissues. (B) The immunofluorescence experiments of HPA database showed that VCAN mainly exists in the vesicles of U-251 MG cells, and RCAN2 mainly exists in the mitochondria and nucleoplasm of U-2 OS cells.