More and more attentions to the globe burden ovarian cancer have been paid. Despite current advances in surgery and chemotherapy, its poor prognosis remains big challenges. Because of its heterogeneity and the lack of convenient and accurate biomarkers, the current prognostic tools of OV patients has limited clinically predictive abilities.[2, 26] Subtype identification, risk stratification, and characterization of the underlying mechanism are critical for improvement of the existing treatment methods, development of more precise and personalized therapies, and prolongation of survival time. Thus, a predictive model with a broad scope of application is needed for accurate prediction of OS in patients with OV and for guiding clinicians in targeted treatment and better prognosis. With the popular application of large databases, more and more prognostic markers are recognized.[8, 9, 27] Many studies have explored the biological function of GRGs in cancers.[14–16] In addition, considering that an increasing number of studies have discovered prognostic markers of glycolysis-related genes, the establishment of GRG-based risk signature to predict the survival of OV patients is necessary.
In our predictive model, this study consisted of a training set and 3 validation cohorts, which included 813 patients with OV. Nine genes with prognostic value for patients with OV were identified using univariate, multivariate, and LASSO Cox regression analyses. The study results indicate that the nine-GRG signature developed herein significantly correlates with poor prognosis in OV. In addition, this risk signature was still an independent prognostic factor in the multivariate Cox analyses. Results of survival analysis suggested that patients with higher risk score tend to have worse clinical outcomes. The 8-gene model showed better predictive ability than any single gene and clinicopathological factors. Nevertheless, due to the lack of relevant data in TCGA database, one limitation is that our model cannot predict recurrence and distant metastasis of OV patients. The model established in present study de well in the OS prediction. And a novel nomogram was established in our study by combing the prediction model and clinical characteristics. The nomogram used the complementary values of clinical characteristics and the prediction model and provided superior estimation of OS. Moreover, the gene signature could further assessed the survival risk when the patients have different clinical features (age, TNM stage, and histological grade). The risk model have effective prediction power for patients with diverse clinical characteristics, but its predictive power showed limited when patents with low grade which should be explored in depth in the future. This result implies that the clinical application of genetic models is far-reaching and the means for patients to predict prognosis in clinic will become more diverse, which can guide clinicians to treat more accurately.
The 9 GRGs we identified include ISG20, CITED2, PYGB, IRS2, ANGPTL4, TGFBI, LHX9, PC, and DDIT4. Among the 9 biomarker genes identified in the present study, DDIT4 (DNA damage-inducible transcript 4) with high expression level actively responded to hypoxia-inducible factor 1 and act together to regulated the generation of cell reactive oxygen species. As an oncogene, [28, 29] its overexpression correlates with tumor progression and worse outcomes in several human cancers, including OV.[18, 30–32] PYGB (brain-type glycogen phosphorylase) could regulate multiple biological characteristics of cancer cells, such as proliferation, invasion, and apoptosis and metastatic phenotypes of several cancers.[33–38] PYGB regulates the Wnt/β-catenin signaling pathway to achieve cancer-promoting effects in OV, non-small cell lung cancer (NSCLC), and gastric cancer. IRS2 (insulin receptor substrates 2) mediates mitogenic and antiapoptotic signaling of insulin-like growth factor 1 receptor (IGF-IR), insulin receptor (IR), and other oncoproteins[42, 43] and is essential for cancer cell motility and metastasis.[44–46] IRS2 acted as an oncogene in OV and was involved in cell proliferation and ascites migration effect in OV progression.[47, 48] ANGPTL4 (angiopoietin-like 4) has been reported to be involved in ferroptotic cell death and chemoresistance of epithelial OV. Moreover, large amounts of it has been detected in malignant ascites of serous OV patients. High ANGPTL4 levels predict shorter relapse-free survival in serous OV.[50, 51] Studies have found that high promoter hypermethylation of TGFBI (transforming growth factor-beta-inducible gene) is involved in chemotherapy resistance of paclitaxel in OV.[52, 53] A study showed that TGFBI and periostin, predict poor prognosis in serous epithelial OV. PC (pyruvate carboxylase) is a biotin-containing enzyme that converts pyruvate to oxaloacetate and has been implicated in cancer progression. PC is strongly involved in tumorigenesis in several cancers, such as breast cancer, NSCLC, glioblastoma, renal carcinoma, and gallbladder cancer.[55–58] Moreover, PC may mediate the regulation of TNKS (tankyrase) in aerobic glycolysis and may be involved in the TNKS-regulated development of OV, as its oncogenic activity is induced by TNKS activating Wnt/β-catenin/snail signaling. Not much evidence has been accumulated on the following genes from OV basic research: ISG20 (interferon-stimulated gene 20), an RNA exonuclease, stimulates tumor progression in hepatocellular carcinoma, clear cell renal cell carcinoma, and glioma.[61–63] The high expression level of ISG20 has association with the poor clinical outcome of OV patients. CITED2 (Cbp/p300-interacting transactivator 2), a pleiotropic protein, has been reported to participate in several biological functions of cells, which included transcription and differentiation. High CITED2 expression level is correlated with poor patient survival in breast  and prostate cancers. CITED2 participated in the regulation of the cell cycle, promoted cell proliferation and then played active role in progression of lung cancer.[66, 67] and supports gastric cancer cell colony formation and proliferation. CITED2 was involved in resistance to platinum-based chemotherapy in OV. LHX9 (LIM homeobox 9) is a developmentally expressed transcription factor and is strongly expressed in the ovarian surface epithelium. Previous research has the development of childhood malignant gliomas involved LHX9 abnormal methylation and silencing  The relationship of ISG20, CITED2, and LHX9 with OV and its molecular mechanism must be examined in depth in future studies. We integrated the 9 GRGs into a panel and established a novel multigene signature for predicting prognosis in OV. This signature showed a strong predictive ability and acted as an independent prognostic molecular factor in patients with OV.
To our knowledge, our study firstly identified the GRG risk predictive signature using the data from TCGA public database. The nine-GRG risk model showed a promising survival prediction ability for the prognosis of OV. We also analyzed mutations in the 9 selected genes in the cBioPortal database. Despite these promising results, questions remain. First, our study was not a prospective study, and all the patients with OV were identified from public databases. Second, large-scale multicenter cohorts are necessary to verify our findings, and further studies are need to further explored the functional roles of the GRGs involved in the initiation and development of OV. Moreover, the gene signature performed more effectively in high-grade OV patients than in low-grade patients, and the mechanism of this observation is should be fully investigated in the future. To further validate the utility of this risk model, we the works of clinical data and specimens collection have been undertaken.