The EPHX2 mRNA and protein expression levels in pan-cancer
We first analyzed EPHX2 mRNA levels in Oncomine database to systematically interrogate the relative EPHX2 expression in different cancer types. The results indicated that EPHX2 showed generally lower expression in bladder, breast, cervical, colorectal, esophageal, head and neck, kidney, liver, lung, ovarian, and pancreatic cancers, as well as melanoma and sarcoma compared with the normal groups. Meanwhile, substantially upregulated EPHX2 expression was demonstrated in only one lung cancer dataset, one myeloma dataset and two lymphoma datasets (Fig. 1A). The details of EPHX2 expression in the above cancers were summarized in Supplementary Table 1.
To further assess the EPHX2 expression of tumor and normal tissues in human cancers, we applied the TIMER2 approach to analyse EPHX2 expression using the RNA-seq data of multiple malignancies in TCGA. As shown in Fig. 1B, EPHX2 expression was significantly down-regulated in multiple cancer types, such as BLCA, BRCA, CHOL, COAD, HNSC, KICH, KIRC, KIRP, LIHC, LUAD, LUSC, PRAD, READ, STAD, THCA and UCEC. Compared with normal tissue samples from GTEX, further validation results indicated that the expression levels of EPHX2 in ACC, BRCA, CESC, GBM, LGG, OV, PAAD, TGCT and USC tumor tissues were also profoundly decreased, and the expression was significantly up-regulated only in THYM (Fig. 1C). For normal and paired tumor tissues, EPHX2 was similarly low expression in tumor tissues of BLCA, BRCA, CHOL, COAD, HNSC, KICH, KIRC, KIRP, LIHC, LUAD, LUSC, READ, THCA and UCEC (Supplementary Figure 1).
Tellingly, at the protein level, CPTAC data also presented lower expression of EPHX2 protein in breast cancer, ovarian cancer, colon cancer, clear cell RCC, UCEC and lung adenocarcinoma primary tissues compared with normal tissues (Fig. 1D, P < 0.001). Moreover, we also sought to assess EPHX2 protein expression in pan-cancer levels using the Human Protein Atlas, which demonstrated several cases of prostate and hepatocellular carcinomas as well as a few cases of renal cancer showed moderate cytoplasmic positivity and remaining cancer tissues were negative or weakly stained (Supplementary Figure 2).
Furthermore, we investigated the correlation between EPHX2 expression and the tumor pathological stages via the GEPIA2 approach as well, and observed that the EPHX2 expression was significantly associated with tumor stage in six cancers, including COAD, KICH, KIRC, KIRP, LIHC and PAAD (Fig. 1E, all P < 0.05), but not others (Supplementary Figure 3).
The potential prognostic value of EPHX2 in various cancers
Next, we identified the prognostic value of EPHX2 via pan-cancer analysis in different databases. As shown in Fig. 2, EPHX2 expression was significantly associated with the prognosis of 9 cancers in the Cox proportional-hazards models, including ACC, CESC, KIRC, KIRP, LGG, LIHC, MESO, PAAD and UVM. Compared with the low EPHX2 expression group, beyond LGG (OS, overall survival: HR, hazard ratio = 1.44, P = 0.037, DSS, disease specific survival: HR = 1.52, P = 0.023), EPHX2 acted positively in the remaining eight tumour types, incl. ACC (OS: HR = 0.39, P = 0.019, DSS: HR = 0.37, P = 0.019), CESC (OS: HR = 0.55, P = 0.014), KIRC (OS: HR = 0.52, P < 0.001, DSS: HR = 0.43, P < 0.001), KIRP (DSS: HR = 0.41, P = 0.028), LIHC(OS: HR = 0.62, P = 0.007), MESO (OS: HR = 0.45, P = 0.001, DSS: HR = 0.50, P = 0.023), PAAD (OS: HR = 0.64, P = 0.034, DSS: HR = 0.60, P = 0.033) and UVM (OS: HR = 0.28, P = 0.008, DSS: HR = 0.27, P = 0.01). Simultaneously, no significant impact of EPHX2 expression on cancer prognosis was observed in other cancer types. Collectively, these results demonstrated that high-expression of EPHX2 was associated with longer survival time in ACC, CESC, KIRC, KIRP, LIHC, MESO, PAAD and UVM patients, while shorter survival time in LGG patients.
In light of these findings, we next sought to evaluate the prognostic potential of EPHX2 in different tumors using the Kaplan-Meier Plotter tool and data derived from the combination of the EGA, GEO and TCGA databases. Fig. 3 revealed that, in addition to its protective in CESC (OS: HR = 0.42, P = 0.00025), KIRC (OS: HR = 0.44, P = 4.5e-06), KIRP (OS: HR = 0.42, P = 0.003), LIHC (OS: HR = 0.57, P = 0.0013) and PAAD (OS: HR = 0.53, P = 0.0041), the EPHX2 gene also acted protectively against HNSC (OS: HR = 0.61, P = 0.0048) and LUAD (OS: HR = 0.59, P = 0.0034).
Finally, we evaluated the relationship between EPHX2 expression and prognosis of each cancer by PrognoScan database, which mainly extracted data from the GEO database. Detailed outcomes were summarized in Fig. 4. Our results indicated that EPHX2 played an oncogenic role in Blood cancer (OS, EFS, event free survival), and it played a protective role in Colorectal cancer (DFS, disease specific survival), Eye cancer (DMFS, distant metastasis free survival), Lung cancer (OS), Prostate cancer (OS) and Soft tissue cancer (DRFS, distant recurrence free survival). It was also worth noting that the precise role of EPHX2 in Breast cancer was controversial (such as OS of GSE9893, OS of E-TABM-158, RFS, relapse free survival of E-TABM-158 and DSS of E-TABM-158: HR > 1, Cox P < 0.05, remarkably differed from the other datasets in Breast cancer: all HR <1, Cox P <0.05).
Taking all the above results into consideration, both survival analysis methods, Cox proportional-hazards model and Kaplan-Meier plotter survival analysis, all demonstrated the significant prognostic power of EPHX2 in CESC, KIRC, KIRP, LIHC and PAAD tumors for OS or DSS. At the same time, some contradictory data related to EPHX2 expression was observed in some cancers (Table 1). These contradictory results might be due to different data collection methods, sample sizes and based on hypothetical mechanisms of different biological characteristics.
The potential association between EPHX2 expression and immune related factors
Determining the interaction of the host immune system with tumors was essential for discovering new prognostic biomarkers, decreasing drug resistance and developing novel targeted therapies[14]. It has been widely accepted that the immune infiltration in the tumor microenvironment might have a great impact on patient survival. To further explore potential correlation between EPHX2 expression and immune score, patients were divided into high and low EPHX2 expression groups using the median EPHX2 expression as the cutoff value. The results showed that EPHX2 expression was significant negatively related to immune score in BRCA, KICH, KIRC, KIRP, LIHC, MESO, OV, PRAD, SARC, TGCT, THCA and UVM, whereas positively related to immune score in DLBC, LAML, LGG, LUSC, SKCM and THYM (Fig. 5A). These results suggested that EPHX2 expression was related to abnormal immune infiltration in some cancer types and might impact the progression of cancer by affecting the regulation of immune infiltration.
In terms of immune cells infiltration, CIBERSORT algorithm was applied to analyse the relationships between EPHX2 expression and immune cell infiltrates in the tumor microenvironment. Our data also presented that the level of immune cell infiltration was significantly correlated with EPHX2 expression in most cancer types (Fig. 5B). Interestingly, EPHX2 expression showed significant negative correlation with M0 macrophage, M1 macrophage, M2 macrophage, activated NK cell and activated CD4+ memory T cell and positive correlation with naive CD4+ T cell, regulatory T cell and follicular helper T cell in THYM. Whereas in TGCT, EPHX2 expression showed significant positive correlation with M2 macrophage and negative correlation with follicular helper T cell. Furthermore, the associations between EPHX2 expression and immunomodulators were investigated (P < 0.01 and |R| > 0.5). As shown in Fig. 6, 24 immune inhibitors were analyzed. The EPHX2 expression showed a negative correlation with IL10 and TGFB1 in SARC, TGFB1 in MESO, and LAG3 in UVM. The correlation analysis of 45 immune stimulators (Fig. 7) revealed that EPHX2 expression was negatively correlated with ULBP1 and TNFRSF8 in UVM, CD276 in PAAD and CD276 in SARC. Strikingly, as shown in Fig. 8, a significant negative correlation was found between the expression of EPHX2 and B2M, HLA-DOB and TAP1 in UVM separately.
Given the strong correlation of EPHX2 with PAAD, SARC, TGCT, THYM, and UVM, GSEA was performed to investigate potential several immune pathways related to EPHX2 signaling involved in these cancers (Fig. 9). Together these data indicated that EPHX2 expression was positively correlated with primary immunodeficiency pathway in PAAD. Contrasting results were found for cytokine-cytokine receptor interaction pathway in THYM and for chemokine signaling pathway and cytokine-cytokine receptor interaction pathway in UVM. Additionally, the correlation between EPHX2 and immune checkpoint blockade (TMB and MSI) was explored further. We observed that EPHX2 expression tended to have positive associations with TMB in LAML and KIRP, while inverse correlations were observed in SKCM, KIRC, PCPG, OV, BLCA, LUAD, PRAD, SARC, PAAD, BRCA, ACC, THYM and DLBC (Fig. 10A). For MSI, negative correlations with COAD, LAML and SARC were determined (Fig. 10B).
The immunotherapy response prediction of EPHX2
Cancer immunotherapy, based on harnessing patients' own immune system to fight cancer, is a promising class of treatments after surgery, radiation, and chemotherapy[15]. ICB has revolutionized the cancer therapy. Here, based on the expression profile data of TCGA, the ICB responses of high and low EPHX2 expression groups were predicted by TIDE algorithm. In 20 TCGA tumor types (ACC, BRCA, CESC, COAD, ESCA, HNSC, KIRC, LGG, LIHC, LUAD, MESO, PAAD, PRAD, READ, SARC, THCA, THYM, UCEC, LAML and TGCT), patients with high EPHX2 expression achieved a lower prediction of TIDE score compared with the low EPHX2 group (Wilcox.Tests, P < 0.05, Fig. 11A-R). Yet, opposite results were obtained in LAML and TGCT (Wilcox.Tests, P < 0.01, Fig. 11S-T). There were no significant differences between groups in TCGA other tumor types. Since patients with higher TIDE scores were significantly more likely to have a higher opportunity of antitumor immune escape and showed a lower response rate of ICB treatment[16], implying that patients with low-risk score appeared to be more sensitive to treated with ICB.