Identification of differentially expressed genes and construction of lncRNA-miRNA-mRNA network diagram
LncRNAs, miRNAs, and mRNAs were anomalously expressed in gliomas compared to normal brain tissues. To explore its potential biological functions, a total of 3914 differential genes were identified in the TCGA database, including 407 up-regulated lncRNAs, 577 down-regulated lncRNAs, 1340 up-regulated mRNAs, and 1590 down-regulated mRNAs (Fig. 1A, B). We used the R package "survival" to perform KM survival analyses on differential genes with P < 0.001 as the screening condition using TCGA data, and identified three lncRNAs (AC080013.1, LINC01574, and LEF1-AS1) with significant survival effects on glioma patients (Fig. 1C, Supplementary Fig. 1A, B). To increase the accuracy, we used the online site of GEPIA for survival validation of these 3 genes and found that LINC01574 and LEF1-AS1 had more impact on patient survival duration, with lncRNA-AC080013.1 exerting no impact on that, however (Fig. 1D, Supplementary Fig. 1C, D). The expression profile microarray of GSE41032 detecting the expression of miRNAs between glioma stem cells (GSC) and normal neural ones (NSC), was performed using GEO2R (|log2FC| ≥ 1, P ≤ 0.05). The results showed that 31 elevated differential expression miRNAs (DEmiRNAs) and 69 decreased ones in the GSE41032 microarray (Fig. 1E). We predicted 134 miRNAs targeting LINC01574 and LEF1-AS1 using the miRcode online website. Subsequently, the differential miRNAs in GSE41032 were intersected with miRNAs targeting lncRNA. A total of 4 DEmiRNAs were identified, namely miR-27a-3p, miR-670-5p, miR-128-3p, and miR-138-5p (Fig. 1F). Of particular note, LEF1-AS1 has a targeting relationship with miR-27a-3p, miR-670-5p, and miR-128-3p, and LINC01574 with miR-138-5p. Subsequently, we used three online databases, miRTarBase, TargetScan, and miRDB, to predict the target proteins for miR-27a-3p, miR-670-5p, miR-128-3p, and miR-138-5p, and a total of 226 mRNAs targeting the above four miRNAs were obtained. There were only 24 mRNAs after taking intersection with differential mRNAs in TCGA (Fig. 1G). GO analyses were conducted using the Metascape online database and found these 24 mRNAs focused on the regulation of the G1/S cell cycle, regenerating glial cells, regulation of protein complex assembly, and actin structural organization (Fig. 1H). Then, we constructed a network of lncRNA-miRNA-mRNA with targeting relationships and analyzed it using Cytoscape software (Fig. 1I). Meanwhile, the top 10 degrees of lncRNAs, miRNAs, and mRNAs were identified using Cytoscape's plug-in Cytohubba (Fig. 1J). The aforementioned 10 genes were selected as the key genes for this study.
Expression of key genes in glioma and normal samples
To determine the differences in expression of key differential genes in glioma tissues versus normal ones, we used the GEPIA database to query their expression. The results showed that lncRNA-LEF1-AS1 was up-regulated in GBM tumor tissues (Fig. 2A). Four mRNAs, PDIA5 (Fig. 2B), WEE1 (Fig. 2C), CSRP2 (Fig. 2D), and ABCA1 (Fig. 2E) were up-regulated and one mRNA (RELN) was down-regulated (Fig. 2F). In addition, we used GSE41032 to derive the expression of miR-27a-3p, miR-670-5p, and miR-128-3p which has a targeting relationship with LEF1-AS1 in normal neural stem cells versus glioma ones. Low levels of miR-128-3p (Fig. 2G) and miR-670-5p (Fig. 2H) were found in glioma stem cells (Tumor) (P < 0.05), while the expression of miR-27a-3p was not statistically significant (Supplementary Fig. 1E). To establish an important ceRNA with significant prognostic value in GBM, we selected miR-128-3p, possessing the most obvious difference in GSE41032, for the subsequent study. Therefore, according to the ceRNA principle, we selected the LEF1-AS1/miR-128-3p/WEE1, PDIA5, CSRP2, ABCA1 axes. Then, we analyzed the levels of lncRNAs, miRNAs, mRNAs in different glioma grades of GBM. The results showed that LEF1-AS1(Fig. 2I), along with the mRNAs of CSRP2 (Fig. 2K), PDIA5 (Fig. 2L), and WEE1 (Fig. 2M) increases with glioma grade, while miR-128-3p (Fig. 2J) decreases with that. And ABCA1 expression was less correlated with glioma grade (Fig. 2N).
Clinical significance of LEF1-AS1 and CSRP2 in patients with glioma
To determine whether the more correlated RNAs were associated with GBM prognosis, OS analyses on GBM patients were performed using the online database CGGA. As a result, one DEmiRNA (Fig. 3A) and four DEmRNAs (PDIA5, WEE1, CSRP2) were found associated with prognosis (P < 0.001) (Fig. 3B-D). These data suggest that LEF1-AS1 probably promotes the expression of WEE1, PDIA5, and CSRP2 by sponging miR-128-3p as a ceRNA. Furthermore, expression correlation analyses showed that the expression of WEE1, PDIA5, and CSRP2 was positively correlated with that of LEF1-AS1. And we selected CSRP2 which has not been studied in glioma for the subsequent study (Fig. 3E-G). Next, the LEF1-AS1/miR-128-3p/CSRP2 axis in the ceRNA network was selected as a potential predictive model. To determine whether the levels of the LEF1-AS1 and CSRP2 are influenced by clinical characteristics, we performed the correlation between levels of LEF1-AS1 as well as CSRP2 and clinical factors. Therefore, univariate and multivariate Cox regression analyses were performed to identify OS-related features. In the univariate Cox regression model for LEF1-AS1, certain prognostic factors (glioma grade, surgical treatment, TMZ treatment, IDH, and 1p19q) were strongly associated with OS in the TCGA cohort of glioma patients, and LEF1-AS1 (hazard ratio [HR] = 1.518, P < 0.001) overexpression was significantly associated with poorer prognosis (Fig. 3H, J). However, in the univariate Cox regression model for CSRP2, certain prognostic factors in the TCGA cohort of glioma patients (glioma grade, surgical treatment, TMZ treatment, IDH, and 1p19q) were strongly associated with OS, and CSRP2 (HR = 1.004, P < 0.001) overexpression were significantly associated with poorer prognosis (Fig. 3J). Meanwhile, multifactorial Cox regression analyses on LEF1-AS1 showed significant association between high expression of LEF1-AS1 and poor prognosis (LEF1-AS1 HR = 1.219, P = 0.006, CSRP2 HR = 1.002, P = 0.005) (Fig. 3I). And multifactorial Cox regression analyses on CSRP2 showed that a high level of CSRP2 was significantly associated with poor prognosis (CSRP2 HR = 1.002, P = 0.005) (Fig. 3K). Using the R language, we derived the risk scores of LEF1-AS1 and CSRP2 based on which the patients were divided into high-risk and low-risk groups, respectively. As shown in Figs. 3L and 3M, patients with glioma died earlier in the high-risk group for LEFA1-AS1 and CSRP2. We also analyzed the disease-specific survival (DSS) of both groups, and glioma patients in the high-risk group presented shorter DSS (Fig. 3N, O). In addition, we also validated the ROC prediction model, and the AUC values of LEF1-AS and CSRP2 were 0.702 and 0.747, respectively, demonstrating the feasibility of our model (Fig. 3P, Q). Therefore, LEF1-AS1 and CSRP2 may be independent prognostic factors for patients with glioma.
Prognostic value of CSRP2 low expression in glioma tissues
Based on the aforementioned, we conclude that CSRP2 plays an important role in the development and progression of glioma. Next, we explored its further potential role in glioma. Immunohistochemical (IHC) staining using the Human Protein Atlas (HPA) database confirmed high expression of CSRP2 in glioma tissues and low in normal ones (Fig. 4A). It has been shown that IDH mutations promote gliomas by disrupting chromosomal topology and permitting aberrant modulation of the interplay inducing oncogene expression. Unlike primary glioblastomas and IDH gliomas which are fast-growing and treatment-resistant, gliomas harboring IDH mutations grow slowly and have a relatively good prognosis. And, oligodendrogliomas with 1p/19q codeletion, the prototypical IDH mutant cancer, display both enhanced radio- and chemo-sensitivity. In particular, lower expression of CSRP2 was found in IDH mutant patients (Fig. 4B) who have better overall time than the wild-type ones (Fig. 4C). Codeletion of chromosome arms 1p and 19q (1p/19q codeletion) highly benefit diagnosis and prognosis in gliomas. And CSRP2 low expression was lower in the 1p/19q codeletion phenotype (Fig. 4D) and the combined deletion of 1p/19q would be more favorable to the prognosis of patients (Fig. 4E). The above results suggest that low expression of CSRP2 may be associated with the development of glioma and patient healing.
Relationship between CSRP2 expression and immune cells in glioma tissues
In some tumors, tumor-infiltrating lymphocytes (TIL) are independent predictors for sentinel lymph node (SLN) status and survival. TIMER was employed to investigate the relationship between CSRP2 expression and immune infiltration levels in glioma tissues. Firstly, analysis of the "Gene" module showed that CSRP2 expression was significantly correlated with tumor purity and positively correlated with the infiltration levels of CD4 + and CD8 + T cells in glioma tissues (Fig. 5A, B). Then, we further evaluated the correlation between CSRP2 expression and immune cell marker genes, and CSRP2 expression was shown positively correlated with immune cell marker genes of CD4 and CD8B (Fig. 5C, D). Meanwhile, the results showed that the expression of CSRP2 was positively correlated with immune checkpoints of EGF and DDX60 (Fig. 5E, F). In conclusion, CSRP2 may be closely related to the development of glioma.
GSEA functional annotation and drug sensitivity
GSEA was applied to investigate the relationship between CSRP2 expression in glioma samples and cancer-related signaling pathways based on the clinically and FDA-validated data from the CellMiner database. The enrichment of glutathione metabolism, oxidative phosphorylation, glycine, serine and threonine metabolism, peroxisome, pyruvate metabolism, and fatty acid metabolism in the group with high expression of CSRP2 compared with the group with low expression of CSRP2 (Fig. 6A). Meanwhile, the CellMiner database was used to predict the potential therapeutic agents for CSRP2. It was found that drugs of Elesclomol (Fig. 6B), Entinostat (Fig. 6C), and Staurosporine (Fig. 6D) have high sensitivity to CSRP2, and it is reasonable to believe that these three drugs could be effective for glioma treatment. In addition, we have identified the molecular structures of Elesclomol (Fig. 6E), Entinostat (Fig. 6F), and Staurosporine (Fig. 6G), contributing to their application in the treatment of glioma.