Identification of Hypoxia Signature to Assess the Tumor Immune Microenvironment and Predict Prognosis in Patients with Ovarian Cancer

Background The 5-year overall survival rate of ovarian cancer (OC) patients is less than 40%. Hypoxia promotes the proliferation of OC cells and leads to the decline of cell immunity. It is crucial to find potential predictors or risk model related to OC prognosis. This study aimed at establishing the hypoxia-associated gene signature to assess tumor immune microenvironment and predicting the prognosis of OC. Methods The gene expression data of 378 OC patients and 370 OC patients were downloaded from datasets. The hypoxia risk model was constructed to reflect the immune microenvironment in OC and predict prognosis. Results 8 genes (AKAP12, ALDOC, ANGPTL4, CITED2, ISG20, PPP1R15A, PRDX5, and TGFBI) were included in the hypoxic gene signature. Patients in the high hypoxia risk group showed worse survival. Hypoxia signature significantly related to clinical features and may serve as an independent prognostic factor for OC patients. 2 types of immune cells, plasmacytoid dendritic cell and regulatory T cell, showed a significant infiltration in the tissues of the high hypoxia risk group patients. Most of the immunosuppressive genes (such as ARG1, CD160, CD244, CXCL12, DNMT1, and HAVCR1) and immune checkpoints (such as CD80, CTLA4, and CD274) were upregulated in the high hypoxia risk group. Gene sets related to the high hypoxia risk group were associated with signaling pathways of cell cycle, MAPK, mTOR, PI3K-Akt, VEGF, and AMPK. Conclusion The hypoxia risk model could serve as an independent prognostic indicator and reflect overall immune response intensity in the OC microenvironment.


Introduction
Ovarian cancer (OC) is characterized by relatively high incidence, high mortality rate, and poor prognosis [1,2]. Poor differentiation of tumor, higher stage of disease, the presence of residual disease after cytoreductive surgery, older age, smoking, excessive weight, and lack of physical activity are associated with the poor prognosis of OC [3][4][5][6][7]. While the majority of patients initially respond well to chemotherapy, some patients relapse and become chemoresistant [8,9]. erefore, the identification of potential predictors that improve the prognosis for women diagnosed with OC may have clinical importance.
Under hypoxic conditions, tumor cells adapt by generating energy in oxygen-independent ways by inducing the expression of genes involved in tumor progression [10]. Additionally, hypoxia can increase the resistance to radiotherapy and chemotherapy and lead to the decline of cell immunity [11][12][13]. Moreover, hypoxic environment is significantly related to the poor prognosis in patients with OC [14]. Up to now, the detailed mechanisms by which hypoxia regulates the status of OC cells resulting in physiological changes remain unclear. Hence, exploring the effect of hypoxia on OC may offer opportunities for potential therapeutic purposes. Nowadays, cancer immunotherapy can target the cells of the immune system [15]. A detailed understanding of the interactions between cancer, hypoxia, and the immune system may be vital for the recognition of potential new immunotherapeutic strategies and targets for OC patients. In this study, we tried to use gene expression data to develop a hypoxia risk model to predict the immune microenvironment in OC patients.

Dataset Sources and Preprocessing.
Firstly, the RNA sequencing data for gene expression (FPKM value) and clinical information were downloaded from UCSC Xena in e Cancer Genome Atlas (TCGA) dataset as a training cohort. e FPKM value was then converted to a transcript of millions per kilobase (TPM) value. Secondly, the gene expression data were downloaded from GSE17260 and GSE32062 datasets in the Gene Expression Omnibus (GEO) database as a validation set. Patient characteristics of the above three cohorts are shown in Table 1. e average value of multiple probes corresponding to the same gene was used as its expression quantity to obtain the gene expression matrix file. e "SVA" software package in R language was used for batch normalization of expression data to obtain a standardized gene expression matrix file. e detailed information of the above datasets is shown in Table 2. A list of hypoxia-related genes was obtained from Hallmark gene sets [16] of the Molecular Signatures Database. Totally, 191 genes were included, all of which responded to low oxygen levels.

Construction and Verification of Hypoxic Gene
Signature. e analysis method was referred to the previous literature [17]. In the training cohort, the prognosis-related hypoxic genes were identified by univariate Cox regression analysis and lasso regression analysis. p < 0.01 was considered statistically significant. e hypoxic gene signature for predicting the prognosis of OC patients was constructed through Cox and lasso regression analyses by using the "glmnet" software package in R language. In the analysis, the lasso penalty was applied. At the same time, shrinkage and variable selection were taken into account. e composition of the final gene signature was selected to generate the risk score based on the following formula: (1) e cases were divided into two groups (high risk and low risk) based on the risk score median. In addition, the same formula was used to calculate the risk score in the validation set.

Survival Analysis.
Overall survival (OS) was compared between the high hypoxia risk group and the low hypoxia risk group via Kaplan-Meier analysis. e multivariate Cox regression analysis was used to determine risk score as an independent risk factor for OS in OC. e receiver operating characteristic (ROC) curve was generated to validate the accuracy of the risk model in predicting the patients' OS via the "timeROC" software package in R language.

Gene Set Enrichment Analysis (GSEA).
Underlying mechanisms were studied through gene set enrichment analysis (GSEA) [18] with the Java program in the TCGA dataset. e adjusted p value was calculated by using the method of Benjamini-Hochberg false discovery rate (FDR). e FDR value of ≤0.05 of the enriched gene set was considered to be statistically significant.

Estimation of Tumor Immune Microenvironment (TIME) Cell Infiltration.
e single-sample gene set enrichment analysis (ssGSEA) algorithm was used to quantify the relative abundance of TIME cell infiltration in each OC sample. e gene set for marking each TIME infiltration immune cell type was obtained from the previous study [19,20]. e enrichment scores (calculated by ssGSEA) were used to represent the relative abundance of each TIME infiltrating cell in each sample. e immune score, matrix score, purity of tumor, and ESTIMATE score were calculated [21]. e violin plot and boxplot were used to compare the levels of immune cell infiltration and immune score between the high and low hypoxia risk groups.

Expression Analysis of Genes at Negatively
Regulated the Cancer Immune Cycle. Cancer immune cycle describes a cycle of processes involved in the eradication of cancer by the immune system [22]. In this study, the gene signature was downloaded from the Tracking Tumor Immunophenotype website [23] to study the gene expression that negatively regulated the cancer immune cycle between the high and low hypoxia risk groups.

Analysis of Immune Checkpoint and Somatic Mutation.
To further clarify the potential association between TIME hypoxia and clinical immunotherapy, the expression of 6 immune checkpoints including PDCD1 (PD1), CD274 (PDL1), PDCD1LG2 (PDL2), CTLA4, CD80, and CD86 were analyzed in the high hypoxia risk group and low hypoxia risk group. In addition, somatic mutation analysis was performed to identify mutation status in the high and low hypoxia risk groups. e somatic mutation data of OC patients were downloaded from the TCGA dataset. e numbers of variant types and classifications were visualized with Oncoplot.

In Vitro Expression Analysis of Hypoxic Gene Signature.
In order to study the expression of hypoxic gene signature at the mRNA level, the RT-qPCR was performed in tissue samples. Totally, 5 patients with OC were enrolled. e Total RNA of the samples was extracted using the TRIzol ® reagent. en, RT-qPCR was performed in an ABI 7300 RT-qPCR system with SuperReal PreMix Plus. Relative gene expressions were analyzed by the 2 -△△ct method and represented as fold change (compared with healthy control). Fold change >1 and fold change <1 represented upregulation and downregulation, respectively. GAPDH and ACTB were used for internal reference.
2.9. Statistical Analysis. All statistics were performed using the R software. Wilcoxon test was used to identify differentially expressed genes and infiltrating immune cells. In addition, the Wilcoxon test was used to screen for differentially expressed infiltrating immune cells and to analyze statistical differences in the expression of risk scores in different clinical features. Kaplan-Meier curves were plotted, and a log-rank test was used to check the significant difference in OS between the high and low hypoxia risk groups. e t-test was used to test the significant difference in hypoxic gene expression between the high and low hypoxia risk groups. e value of p less than 0.05 was set as statistically significant.

Prognostic Value of the Hypoxia Risk Signature in OC.
e risk scores of the training and validation cohorts were calculated using the coefficients obtained by the lasso algorithm. Patients were divided into high and low hypoxia risk groups. e distribution of the risk scores, OS, OS status, and expression profiles of the 8 hypoxic gene signature in the training cohort and validation cohort is displayed in Figures 2(a) and 2(b). Heat map results showed that 6 hypoxic genes, including AKAP12, ALDOC, ANGPTL4, CITED2, PPP1R15A, and TGFBI, were highly expressed in the high hypoxia risk group, which indicated that patients in the group tended to develop hypoxic microenvironments. e mortality was significantly higher in the high hypoxia risk group. e OS of the high hypoxia risk group was shorter in the training and validation cohorts (Figures 2(c) and 2(d)).

Strong Power of Hypoxia Risk Signature for Prognosis Assessment in OC.
To evaluate the predictive efficiency of the hypoxia risk signature in the 1-, 3-, and 5-year survival rates, the ROC curve was performed. e AUC was 0.67 at 1 year, 0.64 at 3 years, and 0.71 at 5 years, respectively, indicating a high predictive value (Figure 3(a)). is was further validated in the validation cohort (Figure 3(b)). is indicated that patients with high risk score could develop hypoxia tumor microenvironment. e univariate analysis suggested that high hypoxia risk score was significantly associated with poor OS (Figure 3(c)). Multivariate analysis showed that high hypoxia risk score was significantly related to poorer OS of OC patients (Figure 3(d)). ese were validated in the validation cohort (Figures 3(e) and 3(f )). In addition, the relationship between the hypoxia signature and clinical parameters (such as age, grade, stage, therapy outcome, lymphatic invasion, and venous invasion) is shown in Figure 4. e risk score for grade III/IV was significantly higher than that for grade I/II. e risk score of G3/4 was  International Journal of Endocrinology significantly higher than that of G1/2. In therapy outcome, the risk score of PD/SD was significantly higher than that of CR/PR.

Identification of Hypoxia-Related Signaling Pathways in OC.
GSEA results showed that processes associated with stimulating tumor proliferation and antiapoptosis were significantly enriched in the high hypoxia risk group ( Figure 5), including cell cycle, MAPK signaling pathway, mTOR signaling pathway, PI3K-Akt signaling pathway, VEGF signaling pathway, and AMPK signaling pathway.

Immunity Analysis between High and Low Hypoxia Risk Groups in OC.
e ability to assess hypoxia risk signals in the immune microenvironment was investigated through ssGSEA. Patients at high risk of hypoxia had significantly higher percentages of cells (such as plasmacytoid dendritic cell and regulatory T cell) and significantly lower percentages of activated dendritic cell, type 17 T helper cell, and natural killer cell (Figures 6(a), 6(b)).
is suggested that immunosuppressive cells may drive the immunosuppressive microenvironment. e results of the ESTIMATE algorithm also confirmed that the immune score ( Figure 6(c)) and matrix score (Figure 6(d)) were significantly lower in patients with high hypoxia. e tumor purity was significantly higher in patients with low hypoxia (Figure 6(e)).

A High Risk Score Indicates the Immunosuppressive Microenvironment in OC.
e gene signature that negatively regulated cancer immune cycle was downloaded from the website "Tracking Tumor Immunophenotype." e results showed that most of these genes, such as arginase 1 (ARG1), CD 160 molecule (CD160), CD 244 molecule (CD244), C-X-C motif chemokine ligand 12 (CXCL12), DNA methyltransferase 1 (DNMT1), and hepatitis A virus cellular receptor 1 (HAVCR1), were upregulated in the high hypoxia risk group (Figure 7).

Analysis of Expression Patterns and Mutation Types of Immune Checkpoints in OC.
e expression of immune checkpoints was investigated in the high and low hypoxia risk groups. e results showed that the expression of most immune checkpoints, such as CD80 molecule (CD80), cytotoxic T-lymphocyte-associated protein 4 (CTLA4), and cd274 molecule (CD274), were significantly higher in the high hypoxia risk group (Figure 8(a)), which was validated in the validation cohort (Figure 8(b)). ese results suggested that the immune microenvironment in the high hypoxia risk group was suppressed by upregulating immunosuppressive cytokines and immune checkpoints. In addition, the difference in gene mutation frequency was analyzed between the high and low hypoxia risk groups (Figure 9). e mutation frequency of ALDOC, ANGPTL4, and PPP1R15A was slightly higher in the high hypoxia risk group (1%) compared with that in the low hypoxia risk group (did not show any mutations). is suggested that mutation of these genes may be associated with hypoxia.

RT-qPCR.
e tumor tissues from 5 patients with OC were used to test the expression of hypoxic gene signature. e clinical information of these patients is shown in Table 3.
In addition, the sequence of the primers used for the RT-qPCR is listed in Table 4. e results showed that ALDOC, CITED2, ISG20, and PRDX5 were significantly   International Journal of Endocrinology upregulated, AKAP12, ANGPTL4, and PPP1R15A were remarkably downregulated, and TGFBI was downregulated without significant difference ( Figure 10). is suggested that the expression level of these genes in tumor tissue was different from that under the hypoxic tumor environment.
In OC, regulatory T cells play roles in reducing survival [54]. In the TIME, the presence of immature plasmacytoid dendritic cells is associated with the poor clinical outcome in patients with breast cancer and epithelial OC [55,56], which indicates that plasmacytoid dendritic cells could play roles in establishing the TIME mediated by forkhead box P3 (Foxp3) + regulatory T cells. Activated dendritic cells can boost the patients' immune system to fight against cancer [57,58]. In this study, we found that patients with high hypoxia risk had significantly higher proportions of regulatory T-cell and plasmacytoid dendritic cell phenotypes. Besides, activated dendritic cells were decreased in the high hypoxia group. In addition, the immune score and matrix score were significantly lower in patients with high hypoxia. Our results showed that the hypoxia risk model could predict the immune microenvironment in OC.
Identification of immunosuppressive factors produced within the TIME and the ability to target these factors can enhance antitumor immune responses. e expression of ARG1 in immune cells is a potent suppressor of antitumor T cells [59]. CD160 is significantly upregulated in OC [60]. CD244 is an exhaustion marker of T cells in OC [61]. Hypoxia induces the expression of CXCL12 in primary  (cm)  1  58  156  54  1  2  62  157  59  10  3  70  159  50  8  4  73  168  46  20  5  56  165  53 10 Table 4: e sequence of the primers used for the RT-qPCR.
Primer name Primer sequences (5′ to 3′) human ovarian tumor cells [62]. It is suggested that CXCL12 is an independent predictor of poor survival in OC [63].
Overexpression of DNMT1 contributes to gene promoter hypermethylation and is associated with the malignant potential and poor prognosis of breast cancer [64,65]. HAVCR1 serves as the marker for ovarian clear cell carcinoma susceptibility [66]. In the present study, the above immunosuppressive cytokines were upregulated in the high hypoxia risk group, which further promoted immunosuppression in OC. e immune checkpoint inhibitor-based antibody has improved survival for patients with various cancer types, such as lung cancer, malignant melanoma, and bladder cancer [67]. High levels of CD80 contribute to the maintenance of tolerance and immunosuppression in epithelial OC [68]. e OC microenvironment can induce migration of CTLA4 + regulatory T cells via C-C motif chemokine ligand 22 (CCL22) and C-C motif chemokine receptor 4 (CCR4) [54,69]. CTLA4 immunotherapy has shown an optimistic antitumor effect in OC [70]. PD-L1 interacts with the corresponding receptor, inhibits the antitumor activity of immune cells, and allows cancer cells to escape immune surveillance [71]. Drug-resistant OC cells exhibit repression of immune-stimulatory molecules, with concomitant augmented expression of CD274 [72]. In this study, we found that the above immune checkpoints including CD80, CTLA4, and CD274 were all upregulated in the high hypoxia risk group, which indicated these immune checkpoints play an important immunosuppression action in OC.
Based on functional analysis, we found that gene sets associated with the high hypoxia risk group were involved in signaling pathways of cell cycle, MAPK, mTOR, PI3K-Akt, JAK-STAT, VEGF, TGF-beta, and AMPK. Genes involved in cell cycle play a key role in OC development and prognosis [73]. In in vitro conditions, induction of CD8 + regulatory T cell is critically mediated by the activation of p38MAPK in OC immunotherapy [74]. e mTOR pathway is found to be activated in about half of patients with highgrade serous OC [75]. e PI3K-Akt signaling pathway plays crucial roles in the process of mesenchymal stem cells induced by hypoxia [76]. Activation of the PI3K/Akt/mTOR pathways is found in OC [77]. VEGF, an important angiogenic factor in advanced OC, is related to tumor aggression and the poor prognosis of OC [78,79]. AMPK protects living cells from hypoxia, which results in elevations in the cellular AMP/ATP ratio [80][81][82]. It is found that AMPK subunits are generally upregulated in OC [83]. Our results indicated that the above signaling pathways may play important roles in the low oxygen environment of OC. CA: carcinoma antigen; HE4: rabbit anti-human HE4 antibody.

Conclusions
We developed and validated a risk model based on 8 hypoxia-associated genes, which could serve as an independent prognostic factor for OC patients and reflect the overall intensity of the immune response in the OC microenvironment. Our study may offer a novel understanding of how hypoxia status affects prognosis and the TIME in OC and benefits future hypoxia-targeted therapies. However, there are limitations to our study. More molecular investigation in the experimental model is further needed.

Data Availability
All data are available in the article.

Conflicts of Interest
e authors declare that there are no conflicts of interest regarding the publication of this study.