Integrated Analysis of lncRNA-Associated ceRNA Network Identifies Two lncRNA Signatures as a Prognostic Biomarker in Gastric Cancer

Abstract

Background: Gastric cancer (GC) is a malignant tumour that originates in the gastric mucosal epithelium and is associated with high mortality rates worldwide. Long noncoding RNAs (lncRNAs) have been identified to play an important role in the development of various tumours, including GC. Yet, lncRNA biomarkers in a competing endogenous RNA network (ceRNA network) that are used to predict survival prognosis remain lacking. The aim of this study was to construct a ceRNA network and identify the lncRNA signature as prognostic factors for survival prediction.

Methods: The lncRNAs with overall survival significance were used to construct the ceRNA network. Function enrichment, protein-protein interaction, and cluster analysis were performed for dysregulated mRNAs. Multivariate Cox proportional hazards regression was performed to screen the potential prognostic lncRNAs. RT-qPCR was used to measure the relative expression levels of lncRNAs in cell lines. CCK8 assay was used to assess the proliferation of GC cells transfected with sh-lncRNAs.

Results: Differentially expressed genes were identified including 585 lncRNAs, 144 miRNAs, and 2794 mRNAs. The ceRNA network was constructed using 35 DElncRNAs associated with overall survival of GC patients. Functional analysis revealed that these dysregulated mRNAs were enriched in cancer-related pathways, including TGF-beta, Rap 1, calcium, and the cGMP-PKG signalling pathway. A multivariate Cox regression analysis and cumulative risk score suggested that two of those lncRNAs (LINC01644 and LINC01697) had significant prognostic value. Furthermore, the results indicate that LINC01644 and LINC01697 were upregulated in GC cells. Knockdown of LINC01644 or LINC01697 suppressed the proliferation of GC cells.

Conclusions: The authors identified 2-lncRNA signature in ceRNA regulatory network as prognostic biomarkers for the prediction of GC patient survival and revealed that silencing LINC01644 or LINC01697 inhibited the proliferation of GC cells.

Figure 1

Volcano plots of differentially expressed genes in gastric cancer (GC). (a) Differentially expressed lncRNAs (DElncRNAs). (b) Differentially expressed miRNA (DEmiRNAs). (c) Differentially expressed mRNAs (DEmRNAs). The red dots indicate upregulated genes with FDR < 0.01 and LogFC > 1.5; the blue dots show downregulated genes with FDR < 0.01 and LogFC < −1.5; the grey dots represent genes with no significant difference. FDR: false discovery rate; LogFC: log fold change.

Figure 2

Competing endogenous RNA (ceRNA) network and Kaplan-Meier analysis results of DElncRNAs. (a) The ceRNA network of lncRNA–miRNA–mRNA in GC. Yellow diamonds indicate lncRNA; pink triangles indicate miRNA; blue circles indicate mRNA. (b) Kaplan-Meier survival plot and boxplot of DElncRNAs in ceRNA network including LINC01537, LINC01644, LINC01697, and LINC02268. Log-rank test was used to assess the survival differences and between two groups.

Figure 3

Functional enrichment analysis of DEmRNAs in the ceRNA network. (a) Fifteen most enriched KEGG pathways. Fifteen most enriched GO annotations that consist of three structured ontologies describing biological process (b), molecular function (c), and cellular component (d).

Figure 4

Construction of protein-protein interaction (PPI) network and ceRNA subnetworks. (a) The gastric cancer PPI network was identified for 69/88 DEmRNAs in the ceRNA network. (b) The modules were obtained from the PPI network following ClusterOne algorithm containing 10 modules. Red and blue dots represent up- and downregulated genes, and orange represents mRNAs with no significant difference in expression in the ceRNA network. (c) Interaction network of 32 hub genes. (d) Screening of lncRNA-miRNA-hub gene subnetwork. All shapes in red and blue represent upregulation and downregulation. Yellow diamonds indicate lncRNA; pink triangles indicate miRNA; and blue circles indicate mRNA.

Figure 5

Identification and performance evaluation of the 2-lncRNA signature. (a) Forest plot shows the hazard ratio (HR) and P value for overall survival with clinical information and differentially expressed lncRNAs. (b) The survival differences between the high-risk and low-risk groups in TCGA training set. (c) Time-dependent receiver-operating characteristic curve analysis evaluating predictive accuracy of the 2-lncRNA signature for 3-year overall survival in TCGA training set. (d) Kaplan-Meier curves in the high- and low-risk group in ICGC testing cohort. (e) Time-dependent receiver-operating characteristic curve analysis evaluating the predictive accuracy of the 2-lncRNA signature for 3-year overall survival in the ICGC testing cohort. (f) The AUC values of 2-lncRNA compared with single biomarker in TCGA training set. (g) The AUC values of 2-lncRNA compared with single biomarker in GEO testing set integrated with GSE65801 and GSE84787 datasets for reduction of batch effect (g). (h) The expression levels of LINC01644 and LINC01697 were validated using the adjusted GSE65801 and GSE84787 databases to remove the batch effect.

Figure 6

The effect of lncRNAs on the proliferation of gastric cancer cells. (a) The mRNA expression levels of LINC01644 in gastric cancer cell lines and normal GES-1 cells. (b) The mRNA expression levels of LINC01697 in gastric cancer cell lines and normal GES-1 cells. (c) Evaluation of gene silencing of LINC01644 and LINC01697 by transfection of cells with lentiviral shRNAs. (d) CCK-8 assay shows SGC-7901 cell viability after different transfections.