Identification of iron metabolism-related predictive markers of endometriosis and endometriosis-relevant ovarian cancer

In this study, the genes related to iron metabolism, endometriosis, and ovarian cancer were evaluated, and a total of 7 hub genes were obtained and their biological effects were explored. Based on logistic regression, we further selected 3 hub genes to construct a diagnostic model, and the results revealed good diagnostic predictive values for both endometriosis and ovarian cancer. Moreover, immune infiltration analysis was performed to further confirm the correlations for the 3 hub genes. As the current clinical treatments are insufficient to cure endometriosis and ovarian cancer, new treatment options are urgently required. This study was based on bioinformatics; however, combined with the conclusions of previous studies, these findings can offer new directions for treatment of these 2 diseases.

Endometriosis is a common disease in women of childbearing age that refers to the appearance, growth, and infiltration of the endometrioid tissue in the endometrium and other parts of the uterus. Endometriosis is estimated to affect 6% to 10% of women of reproductive age and can be detected in up to 50% of women afflicted with infertility. [ 1 , 2 ] The origin of an ectopic endometrium has not been elucidated to date. The main causative factors proposed include a countercurrent of menstrual blood flow, coelomic metaplasia, and induction. Ectopic endometrium development involves the interaction of endocrine, immune, microbial, proinflammatory, and proangiogenic processes as well as environmental factors. [ 3 ] These factors in turn cause patients to develop clinical symptoms such as lower abdominal pain and dysmenorrhea, infertility, dyspareunia, and abnormal menstruation. These symptoms cause further difficulties with work and problems in completing daily tasks, affecting subjective wellbeing and health-related quality of life. [ 4 , 5 ] Endometriosis is generally considered a benign disease but can feature malignant cell behavior such as invasiveness, adhesiveness, and metastasis. Iron is the most common type of transition metal in cells and plays a role in a variety of basic enzymatic reactions and oxygen-carrying proteins. The association between endometriosis and iron homeostasis disorder is becoming increasingly clear, mainly reflected in the extensive ferritin staining of local macrophages, significant increase in erythrocytes, and increase in transferrin receptor (TFR) gene expression. [ 6 ] High levels of iron can cause inflammation, oxidative stress, and lipid peroxidation, and the iron-binding protein hemoglobin is an additional possible hazardous factor. [ 7 , 8 ] In most cases, endometriosis occurs in the ovaries. The histological degree of endometriosis may be classified as typical or atypical. Recent epidemiological evidence suggests that endometriosis may cause the development of malignant tumors. Thus, atypical ovarian endometriosis may be a precursor lesion for epithelial ovarian cancer, mainly endometrioid and clear-cell carcinomas, especially in patients with long-term ectopia. Oxidative stress is the main pathogenic mechanism caused by local iron overload. [ 9 – 11 ] Oxidative stress leads to local peritoneal mesothelium destruction, further resulting in ectopic endometrial cell adhesion. Epithelial ovarian cancer is one of the deadliest gynecological malignancies globally. A majority of patients with these types of tumors are asymptomatic, and thus, the tumor is only diagnosed once at an already advanced stage. Therefore, there is an urgent need to explore relevant biomarkers for this disease to overcome its late diagnosis, unsatisfactory treatment outcome, poor prognosis, and low survival rate. [ 12 ] To this end, the aim of this study was to explore the correlations among iron metabolism, endometriosis, and ovarian cancer. Based on abundant public resources and bioinformatic methods, 7 hub genes related to iron metabolism, endometriosis, and ovarian cancer were identified by analysis of differentially expressed genes (DEGs) in a transcriptome database for endometriosis and ovarian cancer, and 3 hub genes were screened out by a step-based regression model. The diagnostic performance of the identified genes was validated in independent datasets. Gene ontology (GO) and gene set enrichment analysis (GSEA) were further performed to investigate the enriched pathways and functions of the hub genes, and immune infiltration analysis was used to explore the correlation between different immune factors and the identified hub genes. Finally, the hub genes were searched in the DrugBank database, to identify candidate drugs for the pharmacological treatment of ovarian cancer-related to endometriosis based on iron metabolism. The overall workflow of this study is shown in Figure 1 . Overall analysis workflow. DE = differentially expressed, GO = gene ontology, GSEA = gene set enrichment analysis, ROC = receiver operating characteristic.

Data curation: Xu Wang, Lixiang Zhou. Resources: Guangming Wang. Software: Lixiang Zhou. Visualization: Xu Wang, Lixiang Zhou. Writing – original draft: Xu Wang. Writing – review & editing: Guangming Wang, Zhaomei Dong.

We used “endometriosis” and “ovarian cancer” as keywords to search the gene expression omnibus database. The GSE51981 dataset on the GPL570 platform was selected for endometriosis analysis, which included 71 endometriosis and 77 normal endometrial samples. Next, the GSE26712 dataset on the GPL571 platform was selected as an independent validation dataset for receiver operating characteristic (ROC) curve analysis of endometriosis, which included 16 endometriosis and 6 normal endometrial samples. The GSE26712 dataset on the GPL96 platform was selected for ovarian cancer analysis, which included 185 ovarian cancer and 10 normal samples. Finally, the GSE14407 dataset on the GPL570 platform was selected for ROC analysis to validate the results for ovarian cancer, which included 12 ovarian cancer and 12 normal samples. The “limma” package from Bioconductor in R (version 4.2.1; https://cran.r-project.org/ ) was used to analyze the differential expression of genes between the endometriosis and ovarian cancer datasets. Genes with |log2 fold-change (FC)| > 1.5 and a P value  1.5 were defined as up-regulated genes, and those with log2 FC≤−1.5 were classified as down-regulated genes. Volcano plots and heat maps were generated using the R packages “ggplot2” and “pheatmap.” Seventeen datasets related to iron metabolism were found from the molecular signatures database (mSigDB), and 341 genes remained following duplicate gene removal (see Table S1, Supplemental Digital Content, http://links.lww.com/MD/I761 ). The “Venn Diagram” package in R software was used to determine the intersection of iron metabolism-related genes, DEGs in endometriosis, and DEGs in ovarian cancer, and violin plots were used to represent the differential expression levels of hub genes. The “Cluster Profiler” package of R software was used for GO functional enrichment analysis of the hub genes; the minimum gene set was 5, and the maximum gene set was 5000, with a screening condition of P  < .05. GSEA was performed with GSEA software (version 3.0; http://www.gsea-msigdb.org/gsea/ ), and the samples were divided into high- and low-expression groups according to the median expression levels of the different hub genes (≥50% and ≤50%, respectively). We then downloaded a subset of Kyoto encyclopedia of genes and genomes pathways from the mSigDB to assess relevant pathways and molecular mechanisms, and the screening condition was set at P  < .05. Logistic regression is a form of probabilistic nonlinear regression, which is a commonly used multivariate statistical analysis method to study the association between dichotomous variables or multi-categorical variables and some influencing factors. We used a stepwise regression model to screen out the hub genes with the most significant associations. In the endometriosis samples, 0 represented the normal endometrial sample, and 1 represented the endometriosis samples. Similarly, in the ovarian cancers, 0 represented the normal samples, and 1 represented the ovarian cancer samples. The diagnostic value of the hub genes was evaluated using the endometriosis samples of the GSE26712 and GSE14407 datasets on the GPL571 and GPL570 platforms, respectively. CIBERSORT was used to evaluate the standardized data of the GSE51981 and GSE26712 datasets to evaluate the abundance of immune cell infiltration, and the immune cell expression matrix was established and screened according to the threshold of P  < .05. The results are presented as multiple groups in stacked-bar and box graphs. The Drug Bank database provides comprehensive drug target (e.g., sequence, structure, and pathway) information in addition to detailed drug (e.g., chemical, pharmacological, and pharmaceutical) data based on both bioinformatics and cheminformatics. The corresponding drugs for the identified hub genes were searched in the Drug Bank database.

According to the screening criteria, a total of 1483 DEGs were identified in endometriosis dataset GSE51981 . The DEGs are shown in the volcano plot with the top 40 DEGs displayed as a heatmap in Figure 2 . In addition, there were 3042 DEGs identified in ovarian cancer dataset GSE26712 (Fig. 3 ). Genes related to iron metabolism (n = 341) were obtained from mSigDB, and 7 hub genes were obtained via the intersection of the 3 datasets (Fig. 4 ). Figure 5 shows a violin plot displaying the expression levels of the 7 hub genes (Fig. 5 ). Differentially expressed genes (DEGs) between endometriosis and normal endometrial tissues. (A) Volcano plot; up-regulated genes are indicated in red and down-regulated genes are indicated in blue. (B) Heatmap of the top 40 DEGs. Red and blue indicate differential gene expression in grouped samples. Red indicates that the expression value is high, and blue indicates that the expression value is low. Differentially expressed genes (DEGs) between ovarian tumor and normal samples. (A) Volcano plot; up-regulated genes are indicated in red and down-regulated genes are indicated in blue. (B) Heatmap of the top 40 DEGs. Red and blue indicate differential gene expression in grouped samples. Red indicates that the expression value is high, and blue indicates that the expression value is low. Seven hub genes obtained from endometriosis differentially expressed genes (DEGs), ovarian cancer DEGs, and iron metabolism-related genes. DE, differentially expressed. Expression levels of the 7 hub genes differentially expressed in endometriosis and ovarian cancer. (A) Endometriosis dataset GSE51981 . (B) Ovarian cancer dataset GSE51981 . Stepwise regression analysis for the GSE26712 dataset selected 3 of the 7 hub genes: NCOA4, ETFDH , and TYW1 . These 3 hub genes were verified in endometriosis dataset GSE25628 and ovarian cancer dataset GSE14407 , with area under ROC curve values of 0.88 and 0.90, respectively (Fig. 6 ). These 3 hub genes were found to have high diagnostic values in both endometriosis and ovarian cancer. Receiver operating characteristic (ROC) curves of 3 hub genes in endometriosis and ovarian cancer. (A) Endometriosis samples in the GSE26712 dataset. (B) Ovarian cancer samples in the GSE14407 dataset. GO enrichment analysis revealed that the 3 hub genes are involved in iron metabolism-related processes, including iron ion binding, 4-iron-4-sulfur cluster binding, and iron-sulfur cluster binding. Oxidation-reduction related processes were also identified, including oxidoreductase activity and oxidation-reduction process. In addition, coenzyme, metal cluster, and transition metal ion binding were enriched biological processes (Fig. 7 ). GSEA revealed that 2 of the 3 genes are associated with the ribosome, progesterone-mediated oocyte maturation, oocyte meiosis, cell cycle, and nucleotide excision repair in the endometriosis dataset (Fig. 8 ). In the ovarian cancer dataset, these genes were associated with maturity-onset diabetes in the young and Parkinson disease, along with various pathways of metabolism, including ubiquitin-mediated proteolysis, taste transduction, RNA-degradation, riboflavin-metabolism, fatty acid metabolism, pantothenate, and CoA biosynthesis, butanoate metabolism, nicotinate and nicotinamide metabolism, and tryptophan metabolism (Fig. 9 ). Gene ontology (GO) enrichment analysis of hub genes. (A) Circle diagram of GO enrichment and (B) bubble diagram of GO enrichment. Gene set enrichment analysis (GSEA) in endometriosis of dataset GSE51981 . (A) MSMO1 , (B) NCOA4 , (C) PHYH , (D) RRM2 , (E) TYW1 , (F) ETFDH , and (G) HBA1. Gene set enrichment analysis (GSEA) in ovarian cancer of dataset GSE26712 . (A) MSMO1 , (B) NCOA4 , (C) PHYH , (D) RRM2 , (E) TYW1 , (F) ETFDH , and (G) HBA1. The proportions of 22 immune cell types in 148 samples from endometriosis dataset GSE51981 were estimated using the CIBERSORT algorithm. The results showed that the T cell follicular helper cells and monocytes of the endometriosis group were significantly higher than those of the normal group, whereas the proportions of T-gamma delta cells and macrophages of the normal group were significantly higher than those of the endometriosis group (Fig. 10 ). The proportions of immune cells were also calculated in 195 samples of ovarian cancer dataset GSE26712 . The ovarian cancer group had significantly higher activated levels of plasma cells, T follicular helper cells, M0 macrophages, dendritic cells, and mast cells than the normal group. In contrast, monocytes, resting dendritic, and mast cells, eosinophils, and neutrophils in the normal group were found at significantly higher proportions than those in the ovarian cancer group (Fig. 11 ). Immune infiltration analysis of endometriosis dataset GSE51981 . (A) Relative percentages of 22 immune cell types in each sample are shown in stacked plots. (B) Box plot of differences in immune cell infiltration. Immune infiltration analysis of ovarian cancer dataset GSE51981 . (A) Relative percentages of 22 immune cell types in each sample are shown in stacked plots. (B) Box plot of differences in immune infiltration. A total of 29 drugs targeting the 7 hub genes were identified from the Drug Bank database (Table 1 ). NADH (DB00157) is specific to the MSMO1 gene. Furthermore, the human recombinant antihemophilic factor (DB00025), ascorbic acid (DB00126), lonoctocog alfa (DB13998), and moroctocog alfa (DB13999) were specific to the PHYH gene. Imexon (DB05003), motexafin gadolinium (DB05428), GTI 2040 (DB05801), cladribine (DB00242), and gallium nitrate (DB05260) were specific to the RRM2 gene. Metformin (DB00331) is specific to the ETFDH gene. Iron dextran (DB00893), sodium ferric gluconate complex (DB09517), nitrous acid (DB09112), iron (DB01592), zinc (DB01593), ferric pyrophosphate (DB09147), ferric pyrophosphate citrate (DB13995), pentaerythritol tetranitrate (DB06154), zinc acetate (DB14487), ferrous gluconate (DB14488), ferrous succinate (DB14489), ferrous ascorbate (DB14490), and ferrous fumarate (DB14491) were specific to the HBA1 gene. No specific drugs were identified to be associated with the NCOA4 and TYW1 genes. Twenty-nine drugs targeting 7 hub genes.

A large amount of free iron is commonly detected in the peritoneal fluid of patients with endometriosis, as is oxidative deoxyribonucleic acid (DNA) damage in the local microenvironment of the lesions, which may be responsible for cancer development. [ 13 ] Studies examining the role of iron overload in endometriosis revealed that iron can act as a catalyst to oxidative stress, producing the strongest reactive oxygen species – hydroxyl radicals [ 14 ] – which affect macrophage scavenging and in turn stimulate the related nuclear factor kappa B signaling pathway and associated inflammatory responses. [ 15 ] Nuclear factor kappa B promotes the expression of multiple genes encoding proinflammatory cytokines, growth and angiogenesis factors, adhesion molecules, and inducible enzymes such as nitric oxide synthase and cyclooxygenase, [ 16 , 17 ] thus supporting the inflammatory response, anti-apoptosis, and proliferation of endometrial lesions, [ 18 ] as well as other diseases such as atherosclerosis, neurodegenerative diseases, reproductive system diseases, and cancer. Iron is one of the most important metal ions in the human body, playing a role in various diseases and participating in DNA synthesis and repair as well as a variety of enzymatic reactions for cell survival, oxygen transport, and energy metabolism. [ 19 ] Cellular uptake of iron occurs via TFR, which is found in 2 main forms: TFR1 is regulated by iron levels, and TFR2 is regulated by the cell cycle. [ 20 , 21 ] Tumor cells contain more TFR than normal cells, which is related to chronic lymphocytic leukemia, [ 22 ] hepatocellular carcinoma, [ 23 ] gastric cancer, [ 24 ] pancreatic cancer, [ 25 ] breast cancer, [ 26 ] and ovarian cancer. [ 27 ] Although oxidative stress is considered carcinogenic, the specific mechanism of ovarian cancer is unclear. Furthermore, oxidation-reduction imbalance and DNA damage repair exhibit an invalid cause-and-effect relationship [ 28 ] ; hence, iron overload-induced oxidative stress may be a cancer-causing factor, [ 29 ] with associations found for several cancer types, including ovarian clear-cell carcinoma and endometrioid histologic subtypes. [ 27 , 30 ] However, the majority of previous studies on ovarian cancer studies did not objectively describe whether the patients included in the study had endometriosis, which limits the data available for assessing these relationships. [ 31 , 32 ] Here, by integrating iron-related genes and endometriosis and ovarian cancer public datasets, a total of 7 hub genes were screened, among which MSOM1, NCOA4, PHYH, TWY1 , and ETFDH were down-regulated in both endometriosis and ovarian cancer. RRM2 expression was down-regulated in endometriosis and up-regulated in ovarian cancer, whereas HBA1 was up-regulated in endometriosis and down-regulated in ovarian cancer. MSMO1 is an oxidase-encoding gene that catalyzes the demethylation of c4 methylsterols and stimulates cell hyperproliferation. [ 33 ] Furthermore, MSMO1 is involved in the biosynthesis and metabolism of sterols and cholesterol. [ 34 ] Previous studies reported that MSMO1 expression is up-regulated in cervical cancer [ 35 , 36 ] and down-regulated in hepatocellular carcinoma, [ 37 ] cholesterol synthesis in chondrocytes, [ 38 ] pancreatic cancer, [ 39 ] and iron overload, [ 40 ] which is consistent with our results. NCOA4 encodes a selective receptor that mediates ferritinophagy and the cytosolic iron storage complex [ 41 , 42 ] and plays a selective regulatory role in the case of iron deficiency or iron overload. This gene has also been detected in a variety of diseases related to iron metabolism. [ 43 – 46 ] PHYH is a ubiquitous member of the Fe (II) and 2-oxygenase superfamily [ 47 ] and is implicated in Refsum disease [ 48 ] and renal clear-cell carcinoma. [ 49 ] RRM2 is a subunit of nucleotide reductase that is involved in the regulation and modification of proteins and is a cancer gene regulator. [ 50 ] RRM2 is also related to the malignant transformation of liver cancer, [ 50 ] glioblastoma, [ 51 ] breast cancer, [ 52 ] and endometriosis. [ 53 ] Our results further demonstrated the down-regulation of RRM2 expression in endometriosis and its up-regulation in ovarian cancer. TYW1 is a radical S-adenosyl-L-methionine enzyme that catalyzes reactions associated with ferredoxin and consumes free iron in the cytoplasm by binding to other substances to form iron-sulfur clusters, protecting cells from the effects of high iron levels. [ 54 ] At present, this gene has mainly been studied in the context of acute lymphoblastic leukemia. [ 55 ] ETFDH has been reported to be associated with multiple acyl-CoA dehydrogenase deficiency (MADD, OMIM 231680) or glutaric aciduria type II (GAII). [ 56 , 57 ] HBA1 is responsible for the active transport of oxygen, and loss of HBA1 function often leads to thalassemia. Furthermore, HBA1 has been related to acute lymphoblastic leukemia [ 58 ] ; however, the correlation between endometriosis and ovarian cancer has not been reported until now. GO enrichment analysis of the 7 hub genes showed that these genes are involved in iron metabolism-related processes, including iron ion binding, 4-iron-4-sulfur cluster binding, iron-sulfur cluster binding, and oxidation-reduction related processes such as oxidoreductase activity and the oxidation-reduction process. These processes are related to the occurrence and development of iron overload endometriosis, as one of the pathogenic causes of endometriosis. Additionally, coenzyme, metal cluster, and transition metal ion binding were enriched biological processes. GSEA results revealed that these genes are associated with egg development and cell metabolism in endometriosis, as well as in various metabolic pathways in ovarian cancer, along with diabetes and Parkinson disease. Although the correlations of these genes with endometriosis and ovarian cancer have not been directly demonstrated, we further focused on the associated metabolic pathways. We used a stepwise regression model to screen out 3 hub genes, which were further validated in 2 independent datasets: the endometriosis GSE26712 dataset and the ovarian cancer GSE14407 dataset. The area under the ROC curve was 0.88 and 0.90 for endometriosis and ovarian cancer, respectively, which indicated that the obtained hub genes had high diagnostic value and could be further verified by the GO enrichment analysis and GSEA. We evaluated endometriosis dataset GSE51981 and ovarian cancer dataset GSE26712 using the CIBERSORT algorithm and found that endometriosis was associated with T follicular helper cells, monocytes, T-gamma delta cells, and macrophages. Endometriosis has long been considered an autoimmune disease given that the increased number and activation of peritoneal macrophages and the decreased cytotoxicity of T cells and natural killer cells are common alterations in cellular immunity and lead to inadequate clearance of ectopic endometrial cells from the peritoneal cavity. [ 59 ] Although these results conflict with previous research results, [ 60 , 61 ] it is considered that sample error may have occurred. The inflammatory response activated by these immune cells is also associated with the oxidative stress caused by iron overload as described above, further confirming the relevance of iron metabolism in endometriosis. Ovarian cancer was related to plasma cells, T follicular helper cells, M0 macrophages, dendritic cells, activated mast cells, monocytes, resting dendritic and mast cells, eosinophils, and neutrophils, which is consistent with previous findings. These immune cells are involved in oxidative stress processes and in both antitumor and pro-tumor functions. [ 62 ] The main forms of ovarian cancer associated with endometriosis are endometrioid and clear-cell carcinomas. Along with the research on iron overload and associated explorations, this study is the first to explore the combined relationships of iron metabolism, endometriosis, and ovarian and immune cells, which proved their association to a certain extent. A total of 29 drugs related to the 7 hub genes were identified from the Drug Bank database. Regarding the MSMO1 gene, the reduction–oxidation reaction involved in the specific drug NADH may be helpful for Parkinson disease, chronic fatigue syndrome, Alzheimer disease, and cardiovascular disease treatment. [ 63 – 65 ] Specific drugs can be used to treat hemophilia, scurvy, and other blood-related diseases, [ 66 – 68 ] and to treat and control episodes of bleeding as required. Regarding RRM2 , specific drugs can be used to treat melanoma, multiple myeloma, ovarian cancer, leukemia, non-Hodgkin lymphoma, and other diseases. [ 69 , 70 ] The drug metformin (related to ETFDH ) is used to control blood glucose in patients with type 2 diabetes mellitus and can treat insulin resistance in patients with polycystic ovary syndrome. [ 71 ] The specific drugs for HBA1 were mainly related to the treatment of iron deficiency diseases. [ 72 ] For example, these drugs are used to replenish body iron stores in patients with non-dialysis-dependent chronic kidney disease receiving or not receiving erythropoietin, and in patients with hemodialysis-dependent and peritoneal dialysis-dependent chronic kidney disease receiving erythropoietin. Furthermore, these drugs are related to the treatment of zinc deficiency, strengthened immunity, and other disease treatments. [ 73 , 74 ] The drugs related to the hub genes identified in this study were largely associated with iron metabolism and ovarian cancer-related disorders; however, no specific treatment is currently used. Commonly used clinical antitumor drugs exhibit poor selectivity, substantial side effects, and many shortcomings such as drug resistance, which severely limit the curative effects. The associations with iron metabolism suggest a novel therapeutic direction to overcome these limitations. Furthermore, iron metabolism has become a topic of interest in cancer research and demonstrates potential for novel treatment options via associated inducers and inhibitors. [ 75 ] Although women with endometriosis have high iron concentrations, these high iron concentrations do not necessarily lead to cancer. The low proportion of iron-related genes associated with both conditions analyzed together may imply that, although relevant, the iron pathway is not a dominant or causative factor of disease or that this observation is due to the small sample size of the selected database.