Hypoxia-Related lncRNA Prognostic Model of Ovarian Cancer Based on Big Data Analysis

Background Hypoxia is regarded as a key factor in promoting the occurrence and development of ovarian cancer. In ovarian cancer, hypoxia promotes cell proliferation, epithelial to mesenchymal transformation, invasion, and metastasis. Long non-coding RNAs (lncRNAs) are extensively involved in the regulation of many cellular mechanisms, i.e., gene expression, cell growth, and cell cycle. Materials and Methods In our study, a hypoxia-related lncRNA prediction model was established by applying LASSO-penalized Cox regression analysis in public databases. Patients with ovarian cancer were divided into two groups based on the median risk score. The survival rate was analyzed in the Cancer Genome Atlas (TCGA) and International Cancer Genome Consortium (ICGC) datasets, and the mechanisms were investigated. Results Through the prognostic analysis of DElncRNAs (differentially expressed long non-coding RNAs), a total of 5 lncRNAs were found to be closely associated with OS (overall survival) in ovarian cancer patients. It was evaluated through Kaplan–Meier analysis that low-risk patients can live longer than high-risk patients (TCGA: p = 1.302e − 04; ICGC: 1.501e − 03). The distribution of risk scores and OS status revealed that higher risk score will lead to lower OS. It was evaluated that low-risk group had higher immune score (p = 0.0064) and lower stromal score (p = 0.00023). Conclusion It was concluded that a hypoxia-related lncRNA model can be used to predict the prognosis of ovarian cancer. Our designed model is more accurate in terms of age, grade, and stage when predicting the overall survival of the patients of ovarian cancer.


Introduction
Ovarian cancer is a type of malignant tumor that cannot be easily detected in the early stage and has a poor prognosis. Tere are many risk factors associated with its occurrence and development, i.e., family history of ovarian or breast cancer, obesity (BMI of 30 and above), genetic mutations, delayed menopause, fertility treatments, polycystic ovary syndrome, and smoking. Te mortality rate of this type of tumor is ranked ffth among other female malignant tumors [1]. Due to the concealment of the incidence of ovarian cancer, more than half of the patients lost the opportunity for early diagnosis, which seriously afects its prognosis [2]. Surgery plus chemotherapy is a classic treatment for ovarian cancer. First-line maintenance therapy, including bevacizumab or PARP inhibitors, can prolong progression-free survival (PFS), which is diferent from OS [3,4]. Terefore, it is necessary to explore more treatments to prolong the lifespan of ovarian cancer patients. At present, immunotherapy is the new major therapeutic tool of ovarian cancer. However, the efect of single immunotherapy for ovarian cancer is not obvious [5,6].
When malignant tumors increase in their size, tumors gradually form a hypoxia environment, due to which the cancer cells undergo some adaptive changes, such as proliferation and angiogenesis [7]. Te direct reaction of molecules to reverse hypoxia is to stabilize the HIFs. Otherwise, HIF can enhance cell viability and increase angiogenesis and cell invasion. HIF can help in the survival of cancer cells that can undergo apoptosis [8,9]. Hypoxia can also alter the immune microenvironment of malignant tumors [10]. In ovarian cancer, hypoxia promotes cell proliferation [11], epithelial to mesenchymal transformation, invasion, and metastasis [12]. Te aforementioned phenomenon may lead to higher mortality of patients [13].
lncRNAs consist of more than 200 nucleotides and can interact with various kinds of biomolecules such as DNA, RNA, and proteins, which have attracted increasing attention. lncRNAs are extensively involved in the regulation of many important cellular processes, such as gene expression, dosage compensation, and regulation of the cell growth and cell cycle [14]. lncRNAs may play their role in the nucleus as well as in the cytoplasm. Te lncRNAs can act as modulators to afect the protein-coding gene expression by regulating the transcriptional and post-transcriptional processes. As there is diference in expression of IncRNAs in various tissues, some of them have been identifed for their implication in the pathogenesis of the diseases, such as tumor, neurological, cardiovascular, and orthopedic disease. Recent evidence suggests that lncRNAs precipitated the malignant phenotype of cancer through genomic or transcriptional changes. In addition to these changes, changing the immune environment may also promote the malignant phenotype [15,16]. However, a study on hypoxia-related lncRNA in ovarian cancer is still need of the hour. Terefore, a hypoxiarelated lncRNA model was established in our study. Tis model can be applied to pre-calculate the prognosis of ovarian cancer. More importantly, the immune status can be predicted by this model, which can act as a guiding tool for better clinical treatment.

Data Acquisition.
TCGA database (TCGA-OV, 379 samples) (https://tcga-data.nci.nih.gov/tcga/) and the ICGC portal (OV-AU, 81 samples) (https://dcc.icgc.org) were used to extract the RNA sequence data. At the same time, the corresponding clinical features were also downloaded. One of the inclusion criteria of the study was that the patients must survive more than 30 days. Te gene expression profles of normal ovarian tissue as a controlled study were downloaded from the GTEx database (88 samples).

Obtention of Genes and lncRNAs Tat Are Related to
Hypoxia. We downloaded 200 genes related to hypoxia (Table S1 of GSEA). In TCGA database, using Pearson correlation (|R| > 0.4, p < 0.001), 1330 lncRNAs that have a relationship with hypoxia were selected. Ten, the difference analysis was performed by a limma R package. To screen the hypoxia-related diferentially expressed lncRNAs (DElncRNAs), we set the standard as (FDR) < 0.05 and |log2FC| ≥ 1.

Risk Scoring of Candidate Genes for Hypoxia-Related lncRNAs.
To identify candidate genes for hypoxia-related lncRNAs from TCGA cohort, we analyzed OS by univariate Cox analysis. Firstly, we built the prognostic model using LASSO-penalized Cox regression analysis. Te risk score was calculated using the following formula.

(1)
Ovarian cancer patients were then categorized into two groups on the basis of median of TCGA cohort risk scores and named as high or low risk. PCA was performed with the "stats" R packages to explore the distribution of the groups. OS of the two groups was analyzed by Kaplan-Meier analysis. ROC curves with "survival ROC" R package were plotted for 1/3/5 years. Cox regression was utilized to predict the independent values. Te above analyses were carried out simultaneously in TCGA and ICGC datasets. After that, the nomogram including risk, grade, stage, and age was set up by the "rms" R package. Finally, we plotted the correction curve to evaluate the diference between the predicted values and actual values.

GSEA.
GSEA between the two groups was performed in the gene set with the parameter kegg.v7.4.symbols.gmt. To detect the signifcantly enriched pathways, the criterion was p < 0.05 and FDR < 0.05.

Immunity Analysis.
Te following methods were used to calculate the immune penetration status between TCGA project samples, including XCell, TIMER, QUANTISEQ, EPIC, CiberSort-ABS, and CiberSort. All these methods are in silico techniques that were used to integrate the advantages of gene enrichment. Tese are deconvolution techniques to reanalyze the data in comprehensive way [17,18]. Meanwhile, we compared the TME scores between the two risk groups by the "ggpubr" R package.

Division of Clusters by Risk Model.
Using the "Consensus Cluster Plus" R package, two molecular subgroups were grouped based on the prognostic model in ovarian cancer patients. Kaplan-Meier survival analysis, PCA, and tSNE were performed. In addition, analysis of immune-related indexes including immune infltration cells and TME scores was carried out between the two molecular subgroups.

Hypoxia-Related lncRNAs in Ovarian Cancer.
In ovarian cancer, we identifed 1330 hypoxia-related lncRNAs. Te network of the hypoxia-related genes and lncRNAs is presented in Figure 1(a). Tere were a total of 145 hypoxiarelated DElncRNAs, of which 111 lncRNAs were downregulated, and 34 were upregulated (Table S2). Trough the prognostic analysis of DElncRNAs, a total of 5 lncRNAs were found closely associated with OS of ovarian cancer patients, serving as candidate lncRNAs for modeling ( Figure 1(b)).

Evaluation of Prognostic Role of Hypoxia-Related lncRNAs by Risk Model.
We constructed a risk model including 5 lncRNAs (DNM3OS, AC073046.1, AC083799.1, C6orf223, and HCP5) in TCGA cohort. Te formula we used to calculate the risk score is risk score � DNM3OS × 0.152 + AC073046.1 × 0.126-AC083799.1 × 0.184-C6orf223 × 0.097-HCP5 × 0.082. After calculations, we evaluated revisions to the risk models established in TCGA and ICGC databases. Kaplan-Meier analysis showed that low-risk patients can live longer than high-risk patients (TCGA: p � 1.302e − 04; ICGC: 1.501e − 03) (Figures 2(a) and 2(b)). At the same time, we analyzed the OS time of patients in diferent clinical groups in TCGA dataset. Tere was no signifcant difference in OS time when patients were diagnosed with stage I-II or grade 1-2 (p � 0.765, 0.651). But, signifcant diferences were observed in OS time for the stage III-IV group (p < 0.001), G3 group (p < 0.001), under 60 years (p � 0.033), and over 60 years In ROC curve analysis, the one-year AUC in TCGA cohort was 0.652, while the 3-and 5-year AUCs were 0.641 and 0.613, respectively (Figures 3(a) and 3(c)). Te value of AUC in the ICGC cohort was calculated as 0.707, 0.626, and 0.626, respectively (Figures 3(b) and 3(d)). We also found that the model was more accurate in terms of age, grade, and stage when predicting the OS. Te distribution of risk scores and OS status indicated that a higher risk score will lead to a lower OS (Figures 3(e)-3(h)).
When univariate Cox regression analysis was performed, core risk showed its association to OS. Te HR value of TCGA cohort was 2.714 (95% CI � 1.652-4.458, p > 0.001). For the ICGC cohort, the HR value was 3.248 (95% CI � 1.189-8.869), and the p value was 0.022. (Figures 4(a) and 4(b)). Diferent from univariate analysis, multivariate analysis demonstrated an independent role of the risk score in predicting OS in the both cohorts. For TCGA cohort, the HR value was 2.574, 95% CI was 1.560 Finally, we used other factors including risk, grade, age, and stage to predict 1-/3-/5-year OS ( Figure 5(a)). Calibration chart was used to judge whether the result of the nomogram is accurate ( Figure 5(b)).

Cancer-Related Pathways Are Enriched.
In the high-risk group, the pathways enriched in cancer-related pathways were adherens junction, TGF-beta, Wnt, Notch, GnRH signaling pathway, and glycerophospholipid metabolism. In the low-risk group, the pathways were enriched in oxidative phosphorylation, antigen processing, antigen presentation, graft-versus-host disease, metabolism related to glutathione, allograft rejection, and protein export ( Figure 6).

Immune Scoring of Risks Groups Evaluated by GSEA.
We further analyzed the immune status of the two groups with diferent risks in TCGA database. More M1 macrophages, myeloid dendritic, activated NK, and CD8 + T cells were detected in the low-risk group, while fewer cancer-associated fbroblasts (CAFs) and neutrophils were detected in the low-risk group (Figure 7(a), Supplementary 1). Te low-risk group had a higher immune score (p � 0.0064) and a lower stromal score (p � 0.00023). However, the ESTIMATE score did not show signifcant diferences between the two risk groups (Figures 7(b)-7(d)). Journal of Oncology 5

Analyses Related to Molecular
Subtype. Based on this hypoxia-related lncRNA model, we redivided ovarian cancer patients into two clusters (Figure 8(a)). Te OS of cluster1 patients was shorter than cluster2 patients (Figure 8(b)).
Te heatmap of immune cells was drawn by using the following methods, including TIMER, CIBERSORT, CIBERSORT-ABS, QUANTISEQ, MCPCOUNTER, XCELL, and EPIC. All the graphs are presented in Figure 10.

Discussion
Ovarian cancer is a malignant disease that cannot be cured completely. Surgery, chemotherapy, and targeted therapy are the most commonly used methods for its treatment nowadays, but the prognosis remains poor [19,20]. Many research studies have evaluated that hypoxia-related lncRNAs are involved in the progression of various cancers. Hypoxic regions are commonly found in solid tumors, and the appearance of these regions often harms the progression of tumors and triggers tumor immunosuppression and may afect the therapeutic response. However, the underlying mechanism is not fully understood. In our study, hypoxia-related lncRNA was selected as the standard to subgroup patients with various risks. Patients in diferent groups have diferent prognoses and diferent immune statuses. Tis model can help the clinicians to classify and individualize the treatment of ovarian cancer patients and inspire researchers to gain insight into the important role of hypoxia-related lncRNAs in ovarian cancer.
Hypoxia can change the repair mechanism of DNA [21], promote tumorigenesis [22] and metastasis [23,24], and lead to the development of cancer stem cells [25,26], which are resistant to chemotherapy and radiotherapy [27,28]. Terefore, the relationship between hypoxia and cancer needs further study, including hypoxia-related coding genes and non-coding genes. However, it is reported that DNA damage is not induced by hypoxia; instead this hypoxia leads   to some genomic instabilities [21]. Te modeling of hypoxiarelated lncRNA to stratify patients with malignant tumors has been a concern by some scholars. For example, related research has been carried out on hepatocellular cancer and renal cell cancer [29,30]. In our study, a prognostic model consisting of fve lncRNA was constructed, including DNM3OS, AC073046.1, AC083799.1, C6orf223, and HCP5. DNM3OS was overexpressed in ovarian cancer and facilitated ovarian cancer progression, and its high expression might lead to a poor prognosis [31]; the conclusion is similar to our research. At the same time, the role of DNM3OS in other malignant tumors has been confrmed. Tumorassociated mesenchymal stem cells can target DNM3OS, leading to the progression of hepatocellular cancer [32]. In retinoblastoma, the DNM3OS-miR-134-5p-SMAD6 axis can promote cell proliferation, migration, and the EMT process [33]. DNM3OS can also promote tumor progression in gastric cancer [34] and oral cancer [35]. HCP5 has been studied in several kinds of tumors, including ovarian [36], esophageal [37], gastric [38], and colorectal cancer [39]. It has been demonstrated that HCP5 can target miR-525-5p/ PRC1 signaling pathway and can target the Wnt/betacatenin pathway [36]. Other types of lncRNAs are presented for the frst time, through our study.
In the high-risk group, enriched Wnt, Notch, TGF-beta, and tumor-related pathways were found. It has been regarded as one of the leading factors of high mortality in the high-risk group. Wnt/beta-catenin pathway played an important role in ovarian cancer cells' carcinogenesis, stemness, and resistance ability against chemotherapy [40]. Te hyperactivation of the Wnt signaling pathway mediated some drug resistance in ovarian cancer, such as olaparib [41]. Te synergistic efect of Wnt and Notch signaling pathway can promote the proliferation of cancer cells and further increase the migration of cancer cells [42]. Studies have shown that the Notch pathway is closely related to angiogenesis and chemotherapy resistance of ovarian cancer [43,44]. Te Notch signal pathway was regarded as the characteristic of enriched ovarian cancer stem cells induced by hypoxia [45]. TGFβ pathway also facilitates epithelial-mesenchymal transition (EMT) of epithelial ovarian cancer (EOC) [46].
Hypoxia can promote the development of malignant tumors, including ovarian cancer. At the same time, ascites can also induce hypoxia [47]. One reason is that the function of anti-tumor immune cells is inhibited in a hypoxia environment [10]. Terefore, the immune environment of the two groups was compared. More M1 macrophages, myeloid dendritic cells, activated NK cells, and CD8 + T cells were detected in the low-risk group, while less cancer-associated fbroblasts (CAFs) and neutrophils were detected in the lowrisk group. M1 macrophages have an anti-tumor efect, while M2 macrophages can promote tumor [48]. M2 macrophages were also closely related to the progression of cancer cells [49]. We found higher M1/M2 values in the lowrisk group, which did not difer from previous fndings, and patients with higher M1/M2 values lived longer [50].
Dendritic cells can activate T cells such as CD4+ and CD8+ cells to fght tumors by presenting an antigen [51,52]. Other studies have shown that NK cells had abilities that could lead to ovarian cancer cell death, and they often co-infltrate with CD8 + T cells [53]. CD8 + T cells are quite important in antitumor immunity, and further, we can predict patients' OS [54]. In ovarian cancer, CAF may lead to deterioration and drug resistance of the ovarian cancer [55]. Regarding the former research, ovarian tumors are generally regarded as cold tumors, which pose a challenge to the treatment. Te immune activity of patients in the low-risk group tends to be "hot" and may be sensitive to immunotherapy, which provides new opportunities for patients.
lncRNAs play many roles in cancer diagnosis and treatment, but the most important role of these RNAs is to control gene expression and regulate many important biological processes in the body, such as proliferation, genome stability,  apoptosis, pyrolysis, autophagy, and immunity, and dysregulation of these functions contributes to the progression of many tumors. We can conclude that the model is verifed in the external dataset; the model can be further stratifed through analysis of patients with diferent clinicopathological characteristics, which can provide more accurate guidance for clinical treatment. Te limitation of study is that there is no experimental verifcation of lncRNA in the model. Te reason is that it is impossible to accumulate enough fresh tissue specimens for survival analysis in a short period.

Conclusion
It was concluded the hypoxic microenvironment is closely related to the occurrence and development of ovarian cancer. Our established hypoxia-related lncRNA model can be applied to pre-calculate the prognosis of ovarian cancer. In addition, the immune status can be predicted using this model. Our result indicates that the hypoxia-related lncRNA model can serve as a guiding tool for better clinical treatment of ovarian cancer.

Conflicts of Interest
Te authors declare that they have no conficts of interest.