CircRNA can act as a molecular sponge for miRNAs to de-repress all target genes of these miRNAs. Accumulating evidence has shown that this circRNA-miRNA-mRNA network plays an important role in the pathogenesis of gliomas, encompassing a wide range of phenotypes, such as proliferation, migration, and invasion (24–26). Therefore, circRNAs and miRNAs are increasingly regarded as promising therapeutic targets or diagnostic biomarkers. GBM can be divided into different subsets with diverse anticancer responses to various therapies as a group of heterogeneous intracranial neoplasms with distinct histopathological and molecular biological characteristics (27). Thus, there is an urgent need to establish reliable risk stratification methods to classify GBM patients into various risk groups to benefit from various treatment strategies. Many studies have explored the prognostic signatures of gliomas in the context of epigenetic modifications or lncRNAs (28, 29). However, so far, there has been no comprehensive and in-depth study of the molecular signatures of circRNA-related ceRNA networks in gliomas. Therefore, in this study, we constructed a circRNA–miRNA–mRNA network in glioma to help understand its pathogenesis as well as aid in risk stratification and therapeutic decision-making.
In this study, eight DECs (hsa_circ_0001156, hsa_circ_0030788, hsa_circ_0034182, hsa_circ_0000227, hsa_circ_0018086, hsa_circ_0000229, hsa_circ_0036592, and hsa_circ_0002765) were identified as DECs in the first step. To the best of our knowledge, all of them were found to be abnormally expressed in glioma for the first time and have not been studied so far, which makes them potential novel biomarkers or therapeutic targets.
Seven out of eight circRNAs (except for hsa_circ_0001156) were identified as ceRNAs to bind 15 miRNAs (hsa-miR-1200, hsa-miR-1205, hsa-miR-1248, hsa-miR-1303, hsa-miR-3925-5p, hsa-miR-5693, hsa-miR-581, hsa-miR-586, hsa-miR-599, hsa-miR-607, hsa-miR-640, hsa-miR-647, hsa-miR-6867-5p, hsa-miR-767-3p, and hsa-miR-935). As for hsa_circ_0001156, it might still be involved in the pathogenesis of gliomas through functions other than miRNA sponges, such as coding proteins, interacting with RNA binding proteins, or modulating the stability of mRNAs.
As shown in Table 2, among the 15 miRNAs identified, miR-581, miR-586, and miR-1248 were reported to promote tumor progression, while miR-599, miR-607, and miR-767-3p were shown to be tumor suppressors in various cancers. Some studies have shown contradictory results regarding the effects of miR-1303, miR-647, and miR-935 on some tumors. This could partly result from differences in the cell lines used, the phenotypes selected, or the pathways studied. However, further research is needed to resolve these discrepancies. Nevertheless, it is noteworthy that miR-647, miR-767-3p, and miR-935 reportedly suppress the progression of glioma through multiple regulatory axes, including miR-647/HOXA9, miR-935/FZD6, miR-935/HIF1α, or miR-767-3p itself, making them promising biomarkers and therapeutic targets for glioma. In addition, the other six miRNAs without previous studies on tumors (miR-1200, miR-1205, miR-3925-5p, miR-5693, miR-640, and miR-6867-5p) could also be novel fields worth exploring, which could possibly lead to unexpected discoveries.
CircRNAs fulfill their functions by de-repressing the target genes of miRNAs. Therefore, to further explore the effects of circRNAs on glioma, 1076 overlapping target genes were collected and used for functional enrichment analyses. As we all know, the ten major hallmarks of tumors include sustaining proliferative signaling, evading growth suppressors, resisting cell death, enabling replicative immortality, inducing angiogenesis, activating invasion and metastasis, reprogramming of energy metabolism, evading immune destruction, genome instability and mutation, and dysregulating cellular energetics (30), most of which were covered in our enrichment result of biological processes. This indicates that the ceRNA network we built here is extensively involved in the initiation and progression of gliomas.
The pathway enrichment results uncovered the involvement of many essential tumor-related pathways such as the Wnt, TGFβ, cell cycle, p53, Hippo, MAPK, and stemness regulating pathways. Wnt signaling, commonly divided into β-catenin-dependent (canonical) and independent (non-canonical) signaling, is one of the key cascades regulating development. Its role in carcinogenesis has mostly been described in colorectal cancer, along with some other cancer entities (31). The transforming growth factor (TGF)-β signaling pathway is deregulated in many diseases and has dual functions in cancers. It suppresses tumors in healthy cells and early stage cancer cells but promotes tumorigenesis, metastasis, and chemoresistance in late-stage cancer (32). p53 is a tumor suppressor protein that regulates cell growth by promoting apoptosis and DNA repair under stressful conditions (33). The Hippo pathway largely consists of a kinase cascade (MST1/2 and LATS1/2) and downstream transcriptional coactivators (YAP and TAZ), controlling transcriptional programs involved in cell proliferation, survival, mobility, stemness, and differentiation (34). The MAPK/ERK pathway is a chain of proteins that communicate signals from a receptor on the cell surface to the DNA in the nucleus of the cell. Alteration of this pathway is often a necessary step in the development of many cancers (35). Cancer stem cells are capable of sustaining tumors by aiding metastasis, therapy resistance, and tumor microenvironment maintenance, making the stemness regulation key traits and mechanisms for tumor progression (36). All of these pathways, which were under the control of the ceRNA network we constructed in this study, have been shown to participate in the initiation or progression of gliomas (37–42). Altogether, these 1076 target genes, regulated indirectly by the circRNAs identified in the present study, play essential roles in the pathogenesis of gliomas.
Thereafter, LASSO regression analysis and multivariate Cox regression analysis were applied to the 1076 genes consecutively. Forty-six LASSO significant genes and 11 independent prognosis-related hub genes were identified (ARHGAP11A, DRP2, HNRNPA3, IGFBP5, IP6K2, KLF10, KPNA4, NRP2, PAIP1, RCN1, and SEMA5A). Among them, ARHGAP11A, DRP2, HNRNPA3, KLF10, PAIP1, and RCN1 have not yet been studied in gliomas. Three mRNAs were identified as oncogenes in gliomas, which was consistent with our multivariate Cox regression analysis result: IGFBP5 can increase cell invasion and inhibit cell proliferation via the EMT and Akt signaling pathways in GBM (43); IP6K2 was reported to promote cell proliferation and inhibit cell apoptosis under the regulation of the LINC00467/miR-339-3p axis (44); and KPNA4 is capable of facilitating epithelial-mesenchymal transition in glioma, which can be suppressed by miR-181b, a tumor-suppressive miRNA (45). Surprisingly, the roles of the other two mRNAs in glioma were shown to be different from our analysis result: NRP2 promoted glioma cell growth, invasion, and angiogenesis (46); and SEMA5A, whose expression is markedly reduced in higher grades of glioma, can impede motility and promote differentiation of human gliomas (47). This discrepancy might result from the fact that we used survival data for analysis, and those genes were studied only in vitro for some specific phenotypes, while the in vivo result of survival might be influenced by multiple other conditions and phenotypes (such as immune response and therapy sensitivity). However, the functions of these genes require further experimental verification. Nevertheless, the K-M curve analysis and ROC curve analysis in both training and external validation datasets proved that these genes together are a competent signature for predicting the prognosis of gliomas.
Recently, circRNAs were reported to play a significant role in multiple immune-related biological processes, including innate and adaptive immune responses, immune cell homeostasis, immune recognition, and anti-tumor immunity (48, 49). Comprehensive recognition of circRNA-mediated immune cell infiltration in glioma can provide novel insights into risk stratification and clinical therapeutic strategies. Hence, we profiled tumor microenvironment immune cell infiltration utilizing 46 prognosis-related genes from the LASSO regression analysis as shown in Fig. 8. The current consensus is that the anti-tumor immune response in glioma is largely suppressed by brain-resident microglial cells and bone marrow-derived macrophages, and is mainly promoted by CD8 + T cells. However, the roles of other immune cells, such as B cells, are still debatable (50, 51). Our results showed that the selected genes were closely related to immune infiltration and had generally opposite correlations with macrophages and CD8 + T cells. Genes that were positively related to macrophages were negatively correlated with CD8 + T cells, and vice versa. This indicated that these genes could play considerable roles in immune infiltration of the tumor microenvironment. However, it should be noted that there is still a lack of systemic immune cell markers for gliomas. Existing immune markers are mostly constructed in other tumors, and some glioma-specific immune cells (such as microglial cells) still lack convincing specific markers. Specific immune cell markers for gliomas are urgently needed for a robust assessment of immune infiltration and understanding of immune response mechanisms in gliomas.
Glioma is one of the most drug-resistant malignancies with frequent recurrence after chemotherapy, making it necessary to explore novel compounds or drugs that may have a therapeutic effect. Here, with the hub genes identified, some potential drugs were acquired from the CMap database. Although the prediction of CMap was mostly based on experiments in prostate cancer and leukemia cell lines, the effects of these drugs have also been verified in multiple other tumors. For instance, fulvestrant is a selective estrogen receptor degrader that has been extensively studied for its therapeutic effects in breast cancer (52). Harmane is a tremorigenic β-carboline capable of inhibiting mitochondrial viability and increasing reactive oxygen species levels. Its semi-synthetic derivative, B-9-3, showed an anti-proliferative effect in lung cancer, breast cancer, and colorectal carcinoma cell lines via induction of apoptosis and inhibition of cell migration (53). Some of these drugs have been proven to interfere with the progression of gliomas. Tanespimycin is a well-characterized HSP90 inhibitor that can inhibit the growth of GBM and synergize with radiation (54). Mifepristone was reported to be a potential therapy for reducing angiogenesis and TMZ resistance in GBM (55). Tretinoin, an all-trans retinoic acid, was shown to significantly induce apoptosis and suppress stemness in GBM (56, 57). Importantly, mifepristone and tretinoin were shown to inhibit cell cycle and DNA repair of glioma according to GSEA analysis. Given that radiation and temozolomide, the major non-surgical treatments for glioma, both work through inducing DNA damage, those novel drugs could be promising supplementary therapeutic treatments which can be applied in combination with radiotherapy or chemotherapy.
Several limitations of this study should be considered. The construction of circRNA/miRNA/mRNA regulatory networks and the prediction of therapeutic drugs largely relied on a series of bioinformatics algorithms and databases, whose authenticity and accuracy still await the verification of numerous experiments. Therefore, we adopted and integrated multiple databases for all predictions in the present study to improve robustness. In addition, the retrospective research design could display some statistical bias and the traditional bulk sequence transcriptome data would lack comprehensive exploration of intra-tumoral heterogeneity. A prospective study design and utilization of single-cell omics techniques will help address this issue and provide more accurate and reliable results in the future. However, based on circRNA/miRNA/mRNA regulatory networks, we established a superior predictive signature to assess the clinical outcomes of patients with GBM and forecasted some promising candidate drugs.