Uterine corpus endometrial carcinoma (UCEC) is the fourth most common malignancy among women in the United States, with a trend of increasing morbidity and mortality worldwide (Cancer Genome Atlas Research et al., 2013). According to the GLOBOCAN, there would be an estimated 382,069 (2.1% of the total cancer cases) new cases, with 89,929 (0.9% of total cancer deaths) deaths in 2018 (Bray et al., 2018). Due to the advance in techniques for early diagnosis, the patients diagnosed at an early stage account for approximately 75% of UCEC, and most of the patients in stage I can be nearly cured by surgery, with a 5-years overall survival rate of 47–69%. In comparison, advanced-stage patients (stage III or IV) have a poor prognosis due to high risk for recurrence and limited therapeutic strategies, with a 5-years overall survival rate of 15–17% (Lee et al., 2017). Therefore, attention should be paid to discovering novel and reliable biomarkers for prognosis prediction, which can also work as a sensitive classifier for a specific subset of UCEC patients who will benefit from immunotherapy and have improved survival.

Recently, increasing attention has been paid into the exploration of the relationship between cancer cell metabolic plasticity and migration and metastasis (Mosier et al., 2021). Cancer cells with a more invasive and distal metastasis phenotype are found to be strongly correlated with the upregulated expression of peroxisome proliferator-associated receptor gamma and coactivator 1-alpha, which are hallmarks of active mitochondrial biogenesis and oxidative phosphorylation (OXPHOS) (LeBleu et al., 2014). OXPHOS is supposed to be a potential biomarker of tumor progression (Commander et al., 2020). In addition, emerging evidence shows that OXPHOS inhibition can be a valuable target in cancer treatment (Ashton et al., 2018; Molina et al., 2018; Xu et al., 2020a; Cardenas et al., 2020). OXPHOS provides the required energy for cancer cells to strive, and cancer stem cells with primary or acquired resistance against chemotherapy or tyrosine kinase inhibitors are characterized with markedly enhanced OXPHOS dependency (Sica et al., 2020). Metformin and thiazolidinediones can inhibit the mitochondrial electron transport chain, which provides energy for ovarian cancer growth. Thus, interfering with the process of OXPHOS may be considered as a new target for cancer therapy (Nayak et al., 2018). With the development of molecular medicine and next-generation sequencing technology, OXPHOS has been employed in the cancer risk prediction model in lung adenocarcinoma (Xu et al., 2020b). OXPHOS has also attached great importance to UCEC. As a regulator participating in oxidative phosphorylation and glycolysis, PKM2 can function as a biomarker for malignant and premalignant endometrial lesions, and the presence of PKM2^high tumor cells in UCEC tissue also indicates a poor prognosis (Lai et al., 2019). To the best of our knowledge, OXPHOS has not been included in the UCEC prognostic model in previous studies, which deserves to be explored for subgrouping of the patients with distinct prognosis for personal treatment and identifying potential novel therapeutic targets.

Here, we identified prognosis-associated OXPHOS-related differentially expressed genes (DEGs) from TCGA–UCEC cohort (Tomczak et al., 2015). Based on the candidate genes, we constructed an OXPHOS-related prognostic signature model using the train dataset and verified it using the entire dataset. To further investigate the underlying mechanism of the model and its correlation with immune microenvironment and energy metabolism, then between high- and low-risk groups, we compared tumor-infiltrating immune cells (TIICs), ESTIMATE score, tumor purity, immunophenoscores (IPSs), the expression of mismatch repair (MMR) proteins, mRNA expression-based stemness index (mRNAsi), microsatellite instability expression (MSI), the expression of markers of N⁶-methyladenosine (m⁶A) mRNA methylation regulators, tumor mutation burden (TMB) and the response to immune checkpoint blockade (ICB) and chemotherapy agents, which may provide evidence for clinical application of the signature for immunotherapy and chemotherapy option (Figure 1). The findings in this study not only provided a useful predictive model for UCEC but also offered evidence for mutual impacts of OXPHOS, immune microenvironment, and tumor malignant biology.

Though the overall survival state of UCEC patients has improved a lot due to the development of early diagnostic techniques (Braun et al., 2016), there are still 15% of patients with UCEC diagnosed at an advanced stage, thus, suffering from tumor recurrence owing to a limited response to unsuitable therapies (Li and Wan, 2020). Therefore, developing valid and reliable biomarkers for survival prediction and treatment selection to avoid overtreatment in patients who will not relapse and propose individual adjuvant treatment to patients who will recur counts (Lee et al., 2017), and efforts have been put into developing a robust prognostic signature based on DEGs for identification of biologic subsets to guide treatment strategies for UCEC patients, such as a cell cycle-related prognostic signature (Liu et al., 2020a), an autophagy-related prognostic signature (Wang et al., 2020b), and so on. However, there is no consensus on predictive biomarkers for scientifically validated therapy.

In recent years, emerging evidence has proved that OXPHOS could function as a new target in cancer therapy (Ashton et al., 2018). However, OXPHOS has not been involved in developing a prognostic signature for UCEC, thus far, and the potential underlying mechanism and possible clinical application of OXPHOS-related signature in cancer are worth exploring.

Based on the 289 OXPHOS-related gene matrix obtained from the research of Zihao Xu, we used differential expression analysis and identified 66 OXPHOS-related DEGs, including 54 upregulated genes and 12 downregulated genes; most of them have been reported to be associated with cancer. For instance, EPAS1 could enhance the effect of paclitaxel on breast cancer cells by inhibiting growth and promoting apoptosis of MCF-7/TAX cells (Song et al., 2020). Furthermore, functional annotation of the OXPHOS-related DEGs also indicated that the candidate genes were significantly related to ATP metabolic process, purine ribonucleoside triphosphate metabolic process, and the essential pathways reported in many cancers (Matsuda et al., 2017; Fallah and Rini, 2019), including retrograde endocannabinoid signaling and HIF-1 signaling pathways.

Then, we randomly selected samples from the entire set in half into the train set and constructed the model based on it. After univariate Cox analysis and LASSO Cox regression analysis, finally, a prognostic model composed of four OXPHOS-related genes (ATP5IF1, COX6B1, FOXP3, and NDUFB11) was built. Except for ATP5IF1, other candidate genes included in the prognostic signature have been proven to play a vital role in cancer development. COX6B1, as one of the core proteins involved in the molecular basis of the below-background radiation stress response, might inhibit the proliferation of laryngeal squamous cell carcinoma cells (Liu et al., 2020b). FOXP3 can activate the Wnt/beta-catenin signaling pathway, thus, inducing epithelial–mesenchymal transition and promoting the progression of non-small cell lung cancer (Yang et al., 2017). In addition, the regulation of the expression of the NDUFB11 gene can regulate the programmed cell death process, so it promises to be a new target for cancer therapy (Panelli et al., 2013).

Furthermore, Kaplan–Meier survival curves in the train and entire set indicated that the patients with high risk score tend to have a worse prognosis. Besides, a survival ROC curve analysis demonstrated the predictive value of the OXPHOS-related prognostic signature. In addition, by conjoint univariate and multivariate analyses, the independence of the OXPHOS-related signature for prognosis prediction was verified. By combining the clinical features, we performed subgroup survival analysis in different risk groups and discovered that high risk was related to poor prognosis in all subgroups. Primarily, we observed that the risk score was highly correlated to tumor stage, tumor grade, histological type, and patient age. What is more, multi-ROC curve analyses and the evaluation of the nomogram provided evidence for the potential clinical application of the signature in prognosis prediction.

The results of GSEA indicated that the representative pathways in the low-risk group were mainly related to immunization. What is more, the immune score, stromal score, and ESTIMATE score of the low-risk group were higher, and the tumor purity was lower, which implied that patients with a low risk score appeared to be more immunogenic and might preferentially benefit from immunotherapy. To penetrate the immune mechanism of the signature, we compared TIICs between the two different risk groups and analyzed the relationship between the infiltrating immune cells and risk score. We found that plasma cells, CD8⁺ T cells, activated memory CD4⁺ T cells, and regulatory T cells (Tregs) were remarkably elevated in the low-risk group, while M0 macrophages, M2 macrophages, and activated mast cells were significantly elevated in the high-risk group. Immune cell profiles in the tumor microenvironment (TME) of UCEC tissues could be utilized to predict the survival of patients with UCEC (Li and Wan, 2020). Under the influence of environmental stimuli, tumor-associated macrophages (TAMs) can differentiate into two types of macrophage with opposite functions, in which M1 macrophages are protective immune cells that can remove pathogens and malignant cells. In contrast, M2 macrophages are dangerous factors, which can promote angiogenesis and the growth of tumor. High M2 macrophage infiltration in most tumor types is correlated with an adverse prognosis (Zhao et al., 2017; Wang et al., 2018), which is consistent with our results. A meta-analysis proved that UCEC patients with a high CD8⁺ T-cell density had a favorable prognosis (Guo et al., 2020). Increased expression of activated CD4⁺ memory T cells had been confirmed to be beneficial for prognosis in bladder cancer (Li et al., 2020), and the previous study of our team also provided evidence for the opinion (Liu et al., 2020c). The regulatory T cell (Treg) could suppress the antitumor immune response, thus, becoming a target of immunotherapy (Tanaka and Sakaguchi, 2017; Antomarchi et al., 2019), which is different from our results. However, another study showed that FOXP3 lymphocytic infiltration had no significant impact on the survival state of UCEC patients (Giatromanolaki et al., 2008). Our results suggested that CD8⁺ T cells, activated memory CD4⁺ T cells, regulatory T cells (Tregs), M0 macrophages, and M2 macrophages were crucial to the development of UCEC, which had the potential to be targeted for immunotherapies. The immune cell subpopulation enrichment analyses, as well as the ssGSEA-related functions or pathway analyses, showed that the scores of all the items except for aDCs and type I IFN response were higher in the low-risk group, which further proved that UCEC patients with low risk scores appeared to be more immunogenic.

Furthermore, patients with low risk scores had significantly higher scores of IPS, IPS-PD1-PD-L1-PD-L2, IPS-CTLA4, and IPS-PD1-PD-L1-PD-L2-CTLA4; thereby, they tend to respond more sensitively to immune checkpoint inhibitors. The Food and Drug Administration (FDA) has approved a series of monoclonal antibodies (mAbs) targeting PD-1, PD-L1, and CTLA-4 in several malignancy therapies (Arora et al., 2018). Besides, several clinical trials demonstrated the clinical value of anti-PD-1 antibody, pembrolizumab, which can inhibit tumor immune escaping in endometrial cancer. It has been approved as an alternative therapy for patients in the advanced stage (Musacchio et al., 2020). Also, the expression of CD27, CTLA4, ICOS, PD-L2, TIGIT, PD-1, LAG3, TIM-3, CD86, PD-L1, CD70, and CD270 were also remarkably higher in the low-risk group, which deserved to be explored as a new target for immunotherapy. TIGIT could enhance T-cell function by inhibiting PVRIG, so TIGIT-PVRIG pathways play a vital role in human cancers (Whelan et al., 2019). LAG-3 has been identified on various tumor cells, including UCEC, as a critical inducer for the malignancy progression (Friedman et al., 2020). There is also increasing evidence for TIM-3 as an emerging target for malignancy immunotherapy (Moore et al., 2019). The difference found in the comparison of the response to immune checkpoint blockade (ICB) therapy between different risk groups showed the same trend with the above findings. We also compared the risk score of the patients showing a different response to ICB, and found that the responders tended to possess a low risk score. These results demonstrated that the OXPHOS-related prognostic signature had the potential to be the classifier for a specific subset of UCEC patients, who would have a favorable effect on ICB therapy. Compared with UCEC patients in the advanced stage, patients with primary tumor tend to have more somatic loss of mismatch repair (MMR) protein expression; thus, the loss of MMR should be considered as guidance for immunotherapy (Ta et al., 2018). We compared the MSI status of different risk groups and found that the expression of dominant MMR proteins (MLH1, MSH2, MSH6, PMS2, and EPCAM) and mRNAsi in the low-risk group was downregulated, which illustrated that the patients with low risk score were burdened with high-MSI and showed fewer stemness features associated with oncogenic dedifferentiation. Haoya Xu et al. found that cell stemness can predict patient prognosis and constructed an mRNAsi-related prognostic signature in endometrial cancer (Xu et al., 2021). Accumulating evidence revealed that m⁶A mRNA methylation participates in the pathogenesis and progression of cancers through molecular mechanisms, such as inhibiting the antitumor response of CD8⁺ T cells (Chen et al., 2019; Sun et al., 2019). Patients with a low risk score had decreased gene expression of the m⁶A mRNA methylation, thus, showing a favorable response to immunotherapy.

To explore the cause of tumorigenesis, we compared the TMB of each group, and we discovered that the low-risk group had remarkably higher TMB, and the risk score was negatively correlated with TMB. It is reported that patients with high TMB had a more sensitive response to immune checkpoint blockade, such as anti-PD-1 agents (Fancello et al., 2019). A multicohort study demonstrated the value of tissue TMB (tTMB) as a predictive biomarker for response to pembrolizumab monotherapy (Marabelle et al., 2020). In our present study, the most frequent somatic mutations in the low-risk group were PTEN, ARID1A, PIK3CA, TTN, PIK3R1, CTCF, MUC16, ZFHX3, KMT2D, and MUC5B. PTEN functions as a tumor inhibitor by decreasing the activity of phosphoinositide 3-kinase (PI3K) (Bian et al., 2018), and its expression diminished with the progression of UCEC (Xi et al., 2019). It has been proven that ARID1A alterations indicated improved prognosis of tumor patients for compromising MMR and enhancing the infiltration of lymphocytes (Jiang et al., 2020). UCEC patients with high PIK3CA had shorter survival time because PIK3CA could change the tumor immune microenvironment by altering the fraction of tumor-associated neutrophils (Pan et al., 2019). Eighty-five percent of lung squamous cell carcinoma patients with TTN mutant type had missense variations and favorable overall survival (Cheng et al., 2019). In addition, patients in the high TMB group showed better overall survival than in the low TMB group, and the combined Kaplan–Meier analyses proved that risk score was a reliable predictive indicator independent of TMB. Our results were consistent with previous studies, indicating that UCEC patients with low risk score tend to have improved response to immunotherapy and favorable prognosis.

We compared the predicted effectiveness of chemotherapy drugs between different risk groups by using the GDSC dataset. Patients with UCEC in the low-risk group were more sensitive to temsirolimus, roscovitine, AZD6244, PD.0325901, RDEA119, PF.02341066, AKT. inhibitor.VIII, BMS.509744, vinblastine, bryostatin.1, metformin, AZ628, nutlin.3a, bortezomib, bicalutamide, AZD6482, cytarabine, rapamycin, camptothecin, mitomycin. C, S. Trityl.L.cysteine, tipifarnib, parthenolide, sorafenib, and methotrexate, which might be a potential helpful treatment option for UCEC.

These findings indicated that patients with low risk scores possessed higher IPS, higher TMB, and responded more sensitively to ICI and chemotherapy, which partly explained the reason for the improved prognosis of the low-risk patients.

However, there are still some deficiencies with our study. First of all, our data were based entirely on TCGA databases and lacked an external database for validation. Second, our molecular results lack the evidence of the experiments in vitro or in vivo. Third, these OXPHOS-related prognostic genes are supposed to be further examined by cell function assay for biological function and pathway analyses.

In conclusion, an OXPHOS-related and immune microenvironment prognostic signature classifying EC patients into different risk subsets was constructed in our study. This model could predict the prognosis of patients and help to select a specific subset of patients who might benefit from immunotherapy and chemotherapy, thus, improving the overall survival rate. These results may also contribute to the identification of novel immune biomarkers or target therapy in UCEC and give a new insight into the tumorigenesis and progression of UCEC.