EIF4G2 was upregulated in GC
Firstly, we analyzed the expression levels of EIF4G2 in multiple tumors and normal tissues based on the TIMER2.0 database. As shown in Figure 1A, compared with normal controls, EIF4G2 was markedly upregulated in a variety of cancers, including esophageal carcinoma (ESCA), cholangiocarcinoma (CHOL), glioblastoma multiforme (GBM), head and neck squamous cell carcinoma (HNSC), liver hepatocellular carcinoma (LIHC), lung squamous cell carcinoma (LUSC) and stomach adenocarcinoma (STAD). Then, the GEPIA database was used to validate the expression of EIF4G2. The results showed that EIF4G2 expression was significantly elevated in GC (Figure 1B, P < 0.01). Subsequently, the association between EIF4G2 expression and clinicopathological features in GC patients was investigated using the UALCAN platform. As presented in Figure 1C-1G, we found that based on the analysis of sample types, patients’ gender, nodal metastasis status, individual cancer stages and tumor grades, the expression of EIF4G2 in GC tissues was significantly higher than in normal tissues. All the above suggested that EIF4G2 was abnormally over-expressed in GC and could possibly serve as a biomarker of GC.
High expression of EIF4G2 indicated poor prognosis in GC
Next, survival analysis was performed to predict whether the expression of EIF4G2 affected GC patients’ prognoses. As depicted in Figure 2A, higher mRNA levels of EIF4G2 in GC were significantly associated with shorter overall survival (OS) in GEPIA. Then, the Kaplan–Meier plotter database was used to evaluate the prognosis associated with EIF4G2 expression in patients with GC, as well as the prognosis in different pathological subtypes. Figure 2B-2D shows that high expression of EIF4G2 was significantly correlated with poor OS, post progression survival (PPS) and first progression (FP) in GC. Exploiting the RNA-Seq data, we further verified the effect of EIF4G2 on the survival of GC patients. The results showed that high expression of EIF4G2 was negatively correlated with OS in patients with grade 3, stage 2 and stage 4 cancers (Figures 2E–2H, P < 0.05). Finally, in view of its prognostic value in GC, we generated ROC curves to further analyze the diagnostic value of EIF4G2 in GC. As shown in Figure 2I, the area under the curve (AUC) value was 0.844, indicating that EIF4G2 had good diagnostic ability to distinguish GC from normal controls. These results indicate that high expression of EIF4G2 is a biomarker of poor prognosis in GC, and that EIF4G2 may serve as a diagnostic biomarker for GC.
EIF4G2 co-expression network in GC
To investigate the mechanism of action for EIF4G2, the co-expression network of EIF4G2 was constructed using the LinkedOmics database. A volcano plot indicated that 11954 genes (dark red dots) were positively correlated with EIF4G2 expression, and 8271 genes (dark green dots) were negatively correlated (Figure 3A). The 50 genes with the strongest positive and negative correlations are presented in Figure 3B-3C. Gene set enrichment analysis was then applied to analyze the GO terms and KEGG pathways of the genes co-expressed with EIF4G2. The results showed that at the GO_BP (biological process) level, these genes were mainly enriched in cargo loading into vesicle (Figure 3D). GO_ CC (cellular component) was mainly involved in endoplasmic reticulum exit site, coated membrane and chromosomal region, among others (Figure 3E). GO_MF (molecular function) was mainly related to ubiquitinyl hydrolase activity, helicase activity, structural constituent of nuclear pore and regulatory RNA binding, among others (Figure 3F). KEGG pathway analysis indicated that the genes joined mainly in ubiquitin-mediated proteolysis, circadian rhythm, inositol phosphate metabolism, TGF-beta signaling pathway and phosphatidylinositol signaling system (Figure 3G).
Correlation analysis between EIF4G2 expression and immune cell infiltration
Using TIMER 2.0, we analyzed the correlation between EIF4G2 expression and the six types of TIICs. As shown in Figure 4A, EIF4G2 expression was significantly and positively associated with CD8+ T cells (r = 0.257, P = 3.74e-07), neutrophils (r = 0.283, P = 1.99e-08), macrophages (r = 0.288, P = 1.19e-08) and DCs (r = 0.138, P = 7.00e-03) in GC. However, the results showed a negative correlation with infiltrating levels of B cells (r = −0.114, P = 2.68e-02) and no correlation with CD4+ T cells (r = 0.026, P = 6.10e-01). Then, we further evaluated the correlation between EIF4G2 expression and 28 types of TILs in the TISIDB database. Figure 4B shows the relationship between expression of EIF4G2 and 28 types of TILs across human cancers. As presented in Figure 4C-4F, the expression of EIF4G2 was correlated with abundance of Type 2 T helper cells (Th2; r = 0.196, P = 6.17e-05), activated CD4 T cells (Act _CD4; r = 0.163, P = 0.00086), effector memory CD4 T cells (Tem _CD4; r = 0.164, P = 0.000832), and immature DCs (iDCs; r = 0.122, P = 0.013). These data indicated that EIF4G2 may play a specific role in immune infiltration in GC.
Correlation between EIF4G2 expression and immune cell markers in GC
Next, we further explored the relationship between EIF4G2 expression and TIIC markers in GC using the TIMER database. As shown in Table 1, we found that EIF4G2 was positively correlated with B cell markers (CD38), CD4 T Cell markers (CD4), M1 macrophage markers (NOS2, IRF5, PTGS2), M2 macrophage markers (CD163, VSIG4, MS4A4A, ARG1, MRC1), neutrophil markers (CEACAM8, ITGAM, CCR7, MPO), DC markers (NRP1, ITGAX, CD141), monocyte markers (CSF1R, CD86), natural killer cell markers (KIR2DS4, KIR3DL2, KIR3DL1, KIR2DL4, KIR2DL3, KIR2DL1), T cell markers (CD2), T cell exhaustion markers (CTLA4, HAVCR2), tumor-associated macrophage (TAM) markers (IL10), T follicular helper (Tfh) markers (BCL6, IL21), Th1 markers (TBX21, STAT4, IFNG), Th2 markers (STAT6, STAT5A), Th17 markers (STAT3) and T regulatory (Treg) markers (FOXP3, CCR8, STAT5B, TGFB1).The results showed that EIF4G2 could be involved in the regulation of tumor immune infiltration in GC.
Prediction and analysis of upstream miRNAs of EIF4G2
It has been widely acknowledged that non-coding (nc)RNAs participate in the regulation of gene expression. To ascertain whether EIF4G2 was modulated by some ncRNAs, we first predicted upstream miRNAs that could potentially bind to EIF4G2. Considering the underlying mechanisms of miRNAs in the regulation of target gene expression, we predicted a negative correlation between miRNAs and EIF4G2. Finally, we found 15 negatively correlated miRNAs (Table 2, P < 0.05) by starBase. Then, the expression level of these 15 miRNAs in GC and normal samples were examined. The miRNAs with low and statistically significant expression levels in GC patients were considered for analysis. Eventually, hsa-miR-26a-5p was identified. As shown in Figure 5A, EIF4G2 was negatively correlated with hsa-miR-26a-5p, and only hsa-miR-26a-5p was significantly downregulated in GC (Figure 5B, P < 0.001). Subsequently, the prognostic value of hsa-miR-26a-5p in GC was investigated. As described in Figure 5C–5F, upregulation of hsa-miR-26a-5p was positively correlated with OS and with favorable OS in patients with stage 2, stage 4 and grade 3 cancers. These findings all suggested that hsa-miR-26a-5p may be the most likely regulatory miRNA of EIF4G2 in GC.
Prediction and analysis of upstream lncRNAs of EIF4G2
Next, the upstream lncRNAs of hsa-miR-26a-5p were predicted using the starBase database. According to the competing endogenous RNA (ceRNA) hypothesis, lncRNAs can increase mRNA expression by competitively binding to shared inhibitory miRNAs. Therefore, there should be negative correlation between lncRNAs and miRNAs and a positive correlation between lncRNAs and mRNAs. A total of nine possible lncRNAs were predicted (Table 3-4, P < 0.05). Then, the expression levels and prognostic values of these lncRNAs in GC were assessed. The lncRNAs with high and statistically significant expression levels and unfavorable prognosis in GC patients were further analyzed. Ultimately, only TUG1 met the requirements. Figure 6A–6B shows the negative correlation between hsa-miR-26a-5p and TUG1 and the positive correlation between TUG1 and EIF4G2. As shown in Figure 6C-6D, TUG1 was significantly upregulated in GC compared with normal controls. As described in Figure 6E-6I, overexpressed TUG1 indicated poor OS or shorter disease-specific survival (DSS) of GC patients with different clinicopathological parameters. Taking expression analysis, survival analysis and correlation analysis into consideration, TUG1 appears to be the most likely upstream lncRNA regulating the hsa-miR-26a-5p/EIF4G2 axis in GC.