The correlation of EZH2 expression with the progression and prognosis of hepatocellular carcinoma

EZH2 gene expression levels in tumour and normal tissues was obtained from the Oncomine database (http://​www.​oncomine.​org), a web-based data mining platform for collecting, analysing, and offering tumour microarray information [32].

Tumour Immune Estimation Resource (TIMER), a free database containing 32 TCGA-derived cancers involving 10,897 samples, can assess inner immune infiltrate levels (http://​cistrome.​org/​TIMER/​). We evaluated the association of EZH2 levels with six immune cell types within LIHC using the TIMER database [33‐36]. Similarly, we investigated the association of EZH2 expression with tumour purity.

The OncoLnc database contains 21 pieces of TCGA cancer survival information (http://​www.​oncolnc.​org/​). In addition, the current study analysed EZH2’s prognostic value in 21 cancers. In accordance with the obtained results, the EZH2 expression level was remarkably related to survival in eight cancers.

GEPIA2 (The Gene Expression Profiling Interactive Analysis 2) database is a comprehensive analytical tool that analyses customizable functions, such as the interaction function and genes’ prognostic significance in cancer and noncarcinoma samples [37] (http://​gepia2.​cancer-pku.​cn/​). The current work used GEPIA to detect EZH2 mRNA levels in LIHC and its prognostic value and analysed gene expression correlations.

K-M Plotter can be adopted for evaluating the connection between gene expression and 21 cancer prognoses (http://​kmplot.​com/​). In the current study, we adopted K-M Plotter to detect the connection between EZH2 levels and HRs, P values (upon log-rank test), OS, and RFS in LIHC patients. More than 50,000 samples from the database were evaluated using gene array and RNA sequencing [38].

TISIDB is a public database used for analysing the interactions between the immune system and cancers (http://​cis.​hku.​hk/​TISIDB). It combines different cancer immunology data sources [39]. Based on the TISIDB database, we explored the Spearman correlation between EZH2 expression level and tumour-infiltrating cell level and subtype.

The HPA database includes the gene expression profiles and pathology, which can be adopted to validate prognostic gene immunohistochemistry (IHC) [40] (https://​www.​proteinatlas.​org/​). IHC images from HPA can be directly accessed at EZH2 in tumour tissue (https://​www.​proteinatlas.​org/​ENSG00000148773-EZH2/​pathology/​liver+cancer#img) and EZH2 in normal tissue (https://​www.​proteinatlas.​org/​ENSG00000148773-EZH2/​tissue/​liver).

LinkedOmics is an open platform that includes multiomics data for 32 cancers derived from TCGA (http://​www.​linkedomics.​org/​) [41]. We screened for EZH2-related differentially expressed genes in the TCGA LIHC cohort (n = 371) using LinkFinder's LinkedOmics, and their correlations were analysed using Pearson's correlation. The LinkInterpreter module was employed for network and pathway analyses of DEGs. Kyoto Encyclopedia of Genes and Genomes is a database (https://​www.​kegg.​jp/​) that relates genomic information to higher-order functional information for analysing gene function [42‐44]. KEGG pathway and GO analyses were performed using GSEA tools.

The current study used Oncomine database to analyse the difference in EZH2 expression between carcinomas and normal samples. Our analysis revealed that EZH2 is highly expressed in breast, bladder, head and neck, sarcoma, pancreatic, cervical, liver, and other cancers compared with in normal tissues (Fig. 1A). Additionally, EZH2 downregulation was detected in prostate cancer, myeloma, melanoma, kidney cancer, and leukaemia cancers. Additional file 2: Table S1 offers more detailed results of EZH2 expression across cancers. TIMER was used to detect RNA sequencing data in TCGA to evaluate EZH2 expression across cancers. Differential EZH2 expression between cancer and healthy samples is displayed in Fig. 1B. EZH2 expression in SKCM was significantly lower than that in healthy samples. However, EZH2 expression was upregulated within the BRCA, BLCA, COAD, CHOL, HNSC, ESCA, KIRP, KIRC, KICH, LIHC, PRAD, LUSC, LUAD, STAD, READ, UCEC, and THCA.

The current study analysed EZH2’s effect on cancers prognosis from three databases. Table 1 presents the correlation of EZH2 levels with cancers survival from the OncoLnc database. We assessed the correlation between EZH2 expression level and prognosis by using GEPIA2. Figure 2A presents the influence of EZH2 levels on survival in cancer. Poor prognosis was associated with high EZH2 expression levels in ACC (OS: (P = 6.3e-04, HR = 2); PFS: P = 2.8e-04, HR = 1.8); LGG (OS: P = 1.5e-04, HR = 2); PFS: P = 2.8e-04, HR = 1.8); LIHC (OS: P = 5.6e-5, HR = 2.1); PFS: P = 1.0e-04, HR = 1.8); KICH (OS: P = 0.01, HR = 9.8); and KIRC (OS: P = 0.039, HR = 1.4) (Fig. 2B–F). Then, we assessed EZH2-related survival (OS and RFS) by using Kaplan–Meier Plotter. Similarly, we discovered that EZH2 is a detrimental prognostic factor for LIHC (OS: P = 2.1e-06, HR = 2.35 [1.63–3.38]; PFS: P = 1.3e-05, HR = 2.08 [1.48–2.91]), LUAD (OS: P = 0.038, HR = 1.4 [1.02–1.94]; PFS: P = 0.03, HR = 1.69 [1.05–2.71]), PAAD (OS: P = 1.7e-03, HR = 1.97 [1.28–3.04]; PFS: P = 0.024, HR = 4.62 [1.08–19.84]), KIRP (OS: P = 1.2e-03, HR = 2.72 [1.45–5.1]; and PFS: P = 2.3e-03, HR = 3.19 [1.45–6.99]) (Fig. 2G–J).

Cancer survival and lymph node metastasis (LNM) might be predicted according to tumour-infiltrating immune cell (TIIC) levels [45‐47]. The current study analysed the connection between EZH2 levels and immunomodulators and lymphocytes within LIHC based on the TISIDB database (Fig. 3). Figure 3A illustrates the connection between EZH2 expression and 28 TIICs across cancers. The EZH2 level revealed a positive relationship with type 2 T helper cells (Th2; Spearman: ρ = 0.26, P = 3.17e − 07) and activated CD4 + T cells (Act-cd4; Spearman: ρ = 0.533, P < 2.2e − 16) (Fig. 3B). Immunomodulators are immune inhibitors, immunostimulators, or major histocompatibility complex molecules (MHC). Figure 3C presents the analyses of the correlation between EZH2 expression and 24 types of human tumour immunoinhibitors. EZH2 expression was positively correlated with LAG3 (Spearman: ρ = 0.166, P = 1.32e − 03) and CTLA4 (Spearman: ρ = 0.182, P = 4.1e − 04) (Fig. 3D). Figure 3E reveals the correlation between EZH2 levels and 45 types of immunostimulators. In LIHC, EZH2 expression was positively correlated with MICB (Spearman: ρ = 0.403, P < 2.2e − 16) and CD276 (Spearman: ρ = 0.257, P = 5.19e − 07) (Fig. 3F). As shown in Fig. 3G, EZH2 levels were correlated with 21 MHC molecules across cancers. EZH2 was positively related to TAP1 and HLA-E in LIHC (Fig. 3H). Subsequently, we explored the relationship between EZH2 expression and immune infiltration in 39 tumour types by using the TIMER database (Figure. S1). Cancer survival and LNM were predicted according to TIIC levels [45‐47]. EZH2 expression revealed a positive relationship with infiltrating levels of CD8 + T cells in 20 cancer cases, those of macrophages for 12 cancers, those of CD4 + T cells for 18 cancers, those of DCs for 19 cancers, and those of neutrophils for 24 cancers. The results showed that in LIHC, EZH2 expression levels correlated positively with the infiltration degrees of CD4 + T cells (R = 0.378, P = 3.84e-13), CD8 + T cells (R = 0.284, P = 9.30e-08), DCs (R = 0.453, P = 1.38e-18), neutrophils (R = 0.374, P = 7.02e-13), macrophages (R = 0.436, P = 3.22e-17), and B cells (R = 0.474, P = 1.24e-20) (Fig. 4A). The current study suggested that EZH2 stimulated TIIC infiltration within LIHC, especially B cells, neutrophils, CD4 + T cells, DCs, and CD8 + T cells.

Subsequently, an association between EZH2 levels with immune infiltration was analysed in 39 cancers derived from the TIMER database. EZH2 expression revealed a positive relationship with the degree of CD8 + T cell infiltration in 20 cancers, macrophages in 12 cancers, CD4 + T cells in 18 cancers, DCs in 19 cancers, and neutrophils in 24 cancers (Additional file 1: Figure S1). Figure 4A showed that EZH2 expression levels correlated positively with the infiltration degrees of CD4 + T cells (R = 0.378, P = 3.84e-13), CD8 + T cells (R = 0.284, P = 9.30e-08), DCs (R = 0.453, P = 1.38e-18), neutrophils (R = 0.374, P = 7.02e-13), macrophages (R = 0.436, P = 3.22e-17), and B cells (R = 0.474, P = 1.24e-20) in LIHC. Moreover, in some cancer types, including THYM, BRCA, HNSC, and KIRC, the immune infiltration degrees were markedly associated with EZH2 (Additional file 1: Figure S1). The results confirmed that EZH2 expression was related to immune markers and the levels of different T cells, TAMs, M1/M2 macrophages, monocytes, and DCs in LIHC. The scatter plot in Fig. 4B–E shows the correlation of EZH2 levels with macrophage phenotype markers (MS4A4A, CD63 for M2 macrophages; IFR5, COX2 for M1 macrophages; IL10, CD68 for TAMs) and monocytes (such as CSF1R and CD86) within LIHC. The TIMER database confirmed that EZH2 expression was markedly related to marker expression in TAMs, M2 macrophages, M1 macrophages, and monocytes (Fig. 4B–E). To investigate the relationship between EZH2 and TIIC levels in LIHC, the TIMER and GEPIA databases were adopted to analyse the connection between EZH2 and immune markers (Tables 2, 3). As shown in Table 2, after purity adjustment, EZH2 expression exhibited a positive relationship with marker expression in T cells and immune cells in HCC. EZH2 upregulation in M2 macrophages was associated with CD8 + T-cell and DC infiltration in LIHC. DCs can cause cancer migration by reducing CD8 + T-cell toxicity and elevating the number of Tregs [48]. In addition, the EZH2 level revealed a positive relationship with markers for Tregs and exhausted T cells (CTLA4, TIM-3, LAG3, Act-CD4, CD276, and PD-1). As shown in Table 3, we verified a similar correlation between EZH2 levels and monocytes, M2 macrophages, and TAM markers in the GEPIA database in LIHC. As a result, immune markers of diverse TAMs, T cells, DCs, monocytes, and M1/M2 macrophages were related to the EZH2 level in LIHC.

The GEPIA database demonstrated that the expression level of EZH2 in hepatocellular carcinoma samples increased in relation to healthy controls (Fig. 5A). Furthermore, EZH2 was related to the tumour immune microenvironment (TIME). According to the GEPIA database, compared with different stages of LIHC, the EZH2 expression level was higher in stage III and lower in stage IV (Fig. 5B). Vesteinn Thorsson's study clustered six immune subtypes for cancer and revealed that the immune subtypes of cancer might play a key role in predicting disease outcome [49]. According to our results, we detected that EZH2 expression was the highest in C1 (wound healing) and C2 (IFN-γ), whereas it was the lowest in C3 (inflammation) in TISIDB (Fig. 5D). Moreover, we analysed the correlation between EZH2 expression levels and different molecular subtypes in hepatocellular carcinoma. We discovered EZH2 expression, which was highest in iCluster: 1 type and lowest in iCluster: 2 type, in four molecular types (Fig. 5C). According to the HPA database, the increased staining intensity of the EZH2 protein level was detected in tumour tissues compared with noncarcinoma tissues (Fig. 5E).

The biological effect of EZH2 on LIHC was probably connected with the neighbouring gene expression within LIHC. EZH2 coexpression profiles were checked through the ‘LinkFinder’ module LinkedOmics. The EZH2 coexpression gene levels within 371 LIHC cases were analysed through the LinkedOmics database (Additional file 3: Table S2). We discovered that 12,451 genes were positively related to EZH2, whereas 7,471 were negatively correlated with EZH2 (Fig. 6A). The heatmap presented 50 gene sets positively and negatively correlated with EZH2 (Fig. 6B, C).

GO analysis conducted using GSEA in LinkedOmics revealed that EZH2 coexpressed genes were mainly related to DNA recombination, mitotic cell cycle phase transition, and chromosome segregation (Fig. 6D). Through KEGG pathway analysis, coexpressed genes revealed a major enrichment of microRNAs in cancer, the cell cycle, the spliceosome, and pyrimidine metabolism (Fig. 6E). The top 50 most remarkably positive genes were the high-risk factors for LIHC, and 49 of them had large HRs (Fig. 6F). By contrast, 22 markedly negative genes had low HRs (Fig. 6G).