Integrated bioinformatics analysis for the screening of hub genes and therapeutic drugs in ovarian cancer

Ovarian cancer (OC) ranks fifth as a cause of gynecological cancer-associated deaths globally [1], with an estimated 238,700 new cases and 151,900 deaths in 2012 [2]. There are many risk factors associated with OC, such as a family history of breast cancer or OC, excess body weight, smoking, earlier menstruation or later menopause, and not giving birth [2]. Due to the vague symptoms, more than 70% of OC cases are diagnosed at an advanced stage [3]. In most countries, the 5-year survival rate of OC is usually lower than 40% [4]. The poor prognosis and high mortality can be mainly attributed to the lack of early and effective detection methods [5]. Therefore, more efforts need to be invested towards the identification and understanding of novel biomarkers and specific targets of OC, which is considered the key to developing more effective diagnostic and therapeutic strategies.

Recently, gene profiles and gene chips have been used to screen differentially expressed genes (DEGs) [6]. Reanalyzing these data can provide new insights into current research on OC [7]. However, current studies on biomarkers may be insufficient, and the DEG results may be inconsistent because of the complex tumor heterogeneity and complicated molecular regulatory mechanism of OC. Furthermore, chemotherapy resistance remains a major challenge and causes most failures of treatment and the short overall survival (OS) and progression free survival (PFS) of OC patients. Many bioinformatical studies on OC have been proven to be effective and reliable [7, 8], which means integrated bioinformatics analysis could assist with exploring the biomarkers and the mechanisms underlying the tumorigenesis and progression of cancer.

In this study, four original gene expression profiles (GSE54388, GSE69428, GSE36668, and GSE40595) were downloaded from the Gene Expression Omnibus (GEO) database. DEGs were first screened based on the above four datasets. Subsequently, integrated bioinformatics analyses including Gene Ontology (GO) term analysis, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis, protein-protein interaction (PPI) construction, and the identification of hub genes were performed. Furthermore, survival analysis and validation of the hub genes were also performed. Finally, the miRNA-hub gene network was constructed, and targeted small molecular drugs for OC were identified.

Although great progress on surgical and medical therapy has been made for OC, the overall mortality of OC remains the fifth cause of death among malignant gynecologic tumors. The causes of death from OC are mainly attributed to a lack of detection methods at an early stage, a high tendency for metastasis, and chemotherapy resistance [12, 13]. Therefore, exploring reliable biomarkers and precise molecular mechanisms for the early diagnosis, treatment, and prognosis of OC is urgent and necessary. In recent years, with the rapid development of bioinformatics, an increasing amount of microarray and sequencing data have provided a convenient and comprehensive platform to explore general genetic alterations, identify DEGs, and elucidate molecular mechanisms for the diagnosis, therapy, and prognosis of tumors [14].

In this study, four GEO datasets, namely, GSE54388, GSE69428, GSE36668, and GSE40595, were selected to screen DEGs. Applying the GEO2R tool and FunRich software, the intersection of 171 DEGs was identified, including 114 upregulated DEGs and 57 downregulated DEGs. Then, GO and KEGG pathway analyses of the DEGs were performed by DAVID. The KEGG pathway results showed that the upregulated DEGs were mainly associated with metabolic pathways, the biosynthesis of antibiotics, the biosynthesis of amino acids, the cell cycle, and HTLV-I infection, whereas the downregulated DEGs were not enriched. These results also provided significant clues to studying molecular interactions in the progression of OC. Indeed, many studies have indicated that metabolic pathways and the cell cycle are highly associated with the tumorigenesis and progression of OC [15‐17]. For example, autophagy is well known to affect various processes, such as survival under harsh metabolic conditions, and the actual level of autophagy in OC cells could also be affected by cancer-associated fibroblasts [15]. Previously, many studies have reported that antibiotics, such as minocycline, salinomycin, monensin, and pegylated liposomal doxorubicin, could be effective in treating OC [18‐21]. Furthermore, Plewa S et al. once quantitated 42 serum-free amino acids and distinguished amino acid metabolic pathways related to OC growth and development [22]. However, the association between HTLV-I and OC remains unclear. In this study, the upregulated DEGs were mainly involved in biological functions such as cell division, the nucleus, and protein binding. The downregulated DEGs were mainly enriched in biological functions such as the negative regulation of transcription from the RNA polymerase II promoter, the protein complex, the apicolateral plasma membrane, and glycosaminoglycan binding.

Furthermore, a PPI network with 168 nodes and 450 edges was constructed based on the DEGs, and 10 hub genes with the highest degrees of connectivity in the PPI network were identified. Subsequently, survival analysis of these hub genes demonstrated that 8 upregulated hub genes (CCNB2, TYMS, KIF11, KIF4A, BIRC5, BUB1B, FOXM1, and CDC20) were significantly associated with a poorer PFS in OC patients at an early stage (stages 1 and 2). Then, the GEPIA2 database was used to further verify the expression levels of these hub genes in tumor and normal tissues, and the results were in accordance with those from the GEO database. The expression levels of hub genes in different stage OC patients were also conducted. Considering the significant variations in the expression levels of KIF4A and TYMS in early and advanced stage OC patients (Fig. 7), we hypothesized that this result may explain why KIF4A and TYMS were related to poorer PFS in early stage (stages 1 and 2) OC and SOC patients (Fig. 4g, k, Additional file 9: Figure S5G, & K), whereas the 2 hub genes described above were associated with better PFS in advanced stage (stages 3 and 4) OC and SOC patients (Additional file 12: Figure S8 and Additional file 13: Figure S9). Thus, we further hypothesized whether KIF4A and TYMS might be considered as the tumor marker in the diagnosis of early OC. It is well known that tumor invasion and metastasis are always accompanied by complicated molecular regulatory mechanisms. In this study, the expression levels of all hub genes were dysregulated in OC tissues compared with normal tissues, meaning that these genes may be crucial to tumorigenesis and progression in OC. Although many studies have studied these hub genes connected with the diagnosis, treatment, and prognosis of various carcinomas, the molecular mechanisms underlying the tumorigenesis and progression of OC have not been fully understood.

First, kinesin family member 4A (KIF4A) is a microtubule-based motor protein that has been reported to be associated with various cancers, such as oral squamous cell carcinoma, cervical cancer, breast cancer, prostate cancer, colorectal cancer, and lung cancer [23‐28]. However, there have been no studies conducted on KIF4A in regard to OC. In this study, the downregulated expression of KIF4A was correlated with poorer PFS in OC patients, especially in SOC patients at an early stage (Fig. 4g, Additional file 6: Figure S4G). We also found 11 potential miRNAs (hsa-miR-183-5p, hsa-miR-223-3p, hsa-miR-136-5p, hsa-miR-150-5p, hsa-miR-376c-3p, hsa-miR-375, hsa-miR-335-5p, hsa-miR-494-3p, hsa-miR-411-5p, hsa-miR-421, and hsa-miR-543) that KIF4A might target (Fig. 8).

The overexpression of many cyclins has been reported in many tumors [29]; cyclin B2 (CCNB2) is located on the Golgi apparatus during the whole cell cycle and may cause a compromised G2/M transition in Hec1-depleted mouse oocytes [30, 31]. However, the association between CCNB2 and OC is not clearly understood. Cell division cycle 20 (CDC20) is an activator of the ligase anaphase-promoting complex/C (APC/C) that has been identified as a key candidate gene in OC patients based on many bioinformatics studies [32, 33]. High expression levels of CDC20 have been reported to be correlated with poor prognosis in a variety of cancers [34‐36]. However, its specific molecular mechanisms underlying OC have not been fully explained. In this study, the survival analysis of CDC20 in OC patients at an early stage suggested that higher CDC20 expression could be related to poorer prognosis (Fig. 4e).

Thymidylate synthase (TYMS) is crucial for DNA synthesis in both normal and tumor cells. Biason P et al. calculated six TYMS polymorphisms related to OS in epithelial OC (EOC) through a multivariate analysis and found that the TYMS 1494ins/del, 1053C > T and IVS6-68C > T polymorphisms might be considered prognostic markers for OS in EOC patients [37]. Additionally, Kelemen LE et al. also reported that re495139 in the TYMS-ENOSF1 region was associated with the risk of OC by evaluating large samples of assembled cases [38, 39]. KIF11, a member of the kinesin-like protein family, was once identified as a hub gene in OC by Zhanzhan Xu et al. [8], and they also found that KIF11 was potentially targeted by hsa-miR-424 and hsa-miR-381. A recent study explored whether death receptor 6 can promote OC cell migration by interacting with KIF11 and TNF receptor-associated factor 4 [40]. In this study, KIF11 was also considered a chemotherapy target for OC using cBioPortal (Fig. 5a). Ribonucleotide reductase M2 (RRM2) is the regulatory subunit of ribonucleotide reductase and is not only involved in the tumorigenesis and progression of a variety of carcinomas (such as colon, breast, and pancreatic cancer) [41‐43] but also leads to resistance to cancer chemotherapy [44]. Katherine M Aird et al. found that the knockdown of RRM2 expression could inhibit the growth of human EOC cells by triggering cellular senescence [45], which suggests that RRM2 might be a novel prosenescence therapeutic strategy to improve the OS of EOC patients. Our drug-hub gene interaction result was consistent with that of Katherine M Aird et al. Type 2 topoisomerase alpha (TOP2A) is a gene close to the HER2 gene that encodes a nuclear enzyme that is crucial in the cell cycle. Mano et al. first explored TOP2A amplifications in EOC [46], whereas in our study, we found that the type of alteration for TOP2A in SOC mainly included truncating mutations, missense mutations, and deep deletions (Fig. 5b). Mitotic checkpoint serine/threonine kinase B (BUB1B), the mammalian homolog of yeast Mad 3, is susceptible to cancer and causes chromosome loss and apoptotic cell death in human cancer cells [47, 48]. Many studies have reported that BUB1B is highly associated with advanced stages, serous histology and high grades in EOC, and a weakened spindle checkpoint with the downregulated expression of BUB1B is also related to acquired paclitaxel resistance in OC cells [49, 50]. Forkhead box protein M1 (FOXM1), a transcription factor with a “winged helix” DNA-binding domain, plays a significant role in tumor progression through cell proliferation, tumor invasion, migration, and angiogenesis [51, 52]. Yu Wang et al. reported that FOXM1 promotes the reprogramming of glucose metabolism in EOC by activating GLUT1 and HK2 transcription [53]. Finally, baculoviral inhibitor of apoptosis repeat-containing 5 (BIRC5) is an inhibitor of apoptosis that is always absent in normal tissues and is associated with various cancers, such as breast cancer and OC [54, 55]. Epithelial to mesenchymal transition (EMT) is a biological process that is associated with tumor metastasis and chemoresistance in ovarian tumors [56]. Zhao G et al. reported that higher expression of BIRC5 always promoted EMT in ovarian cancer cells [57]. In this study, both the cBioPortal and DGIdb databases considered BIRC5 to be a target of chemotherapy drugs used to treat OC (Fig. 5a, Table 3).

Furthermore, 10 genes were identified as the hub genes, and their expression levels in OC patients of different stages were all verified according to both the GEO and GEPIA2 databases. A list of 62 drugs with potential therapeutic efficacy against OC were selected (Table 3). For the 10 hub genes, the potential gene targets of these drugs were TYMS and BIRC5. However, the mechanisms of most of these drugs were unknown. The effects of the screened drugs were verified for many other types of cancers such as lung cancer, liver cancer, and colon cancer. Only a few of the TYMS- and BIRC5-targeting drugs have been used for OC treatment, such as fluorouracil/oxaliplatin (5-fluorouracil can induce the upregulation of human TYMS expression through TYMS mRNA transcription) [58], paclitaxel (the inhibition of BIRC5 expression can activate cell responses to paclitaxel treatment) [57], and carboplatin [59]. Additional studies and clinical trials are needed to identify and explore the drugs that are effective for OC treatment in future. Nevertheless, this study used multiple available databases to provide novel insights into the OC pathogenesis and treatment, and the identified conventional drugs might find potential new uses.

Therefore, considering the crucial roles of these 10 hub genes on the basis of this study and other previous studies mentioned above, further studies may be focused on exploring their precise mechanisms in the tumorigenesis and prognosis of OC, especially studying KIF4A, CCNB2, and CDC20. There were still several limitations in this study. One is that the microarray data and other analyses were obtained from many public databases, not generated by authors. The other is that only the TCGA and GEPIA2 databases were used to verify the expression levels of the hub genes, and experiments were not performed by the authors. Furthermore, due to a lack of experimental studies and verifications, we could not further explore how hub gene-miRNAs networks have effects on the diagnosis and therapy of ovarian cancer in depth. Despite these limitations, this study may provide more accurate results based on integrated bioinformatic analysis compared to the single dataset studies.

In summary, a total of 171 DEGs, including 114 upregulated DEGs and 57 downregulated DEGs in OC, were screened through integrated bioinformatic analysis, and 10 hub genes, namely, KIF4A, CDC20, CCNB2, TOP2A, RRM2, TYMS, KIF11, BIRC5, BUB1B, and FOXM1, may play crucial roles in the tumorigenesis and prognosis of OC. Among these hub genes, one gene, KIF4A, has not been previously reported to be related to OC, which indicates that KIF4A might be a potential biomarker for OC diagnosis and prognosis at an early stage. Additionally, the miRNA-hub gene network and potential targeted drugs related to these hub genes were constructed and selected, respectively, and may contribute to studying the mechanisms of OC. In-depth molecular mechanisms of the novel hub genes in OC are needed in future, and relevant experimental models can be constructed on the basis of these genes for the early detection, prognostic judgment, risk assessment, and targeted therapy of OC.