Normal human astrocytes were purchased from ScienCell Research Laboratories (San Diego, CA, USA) and cultured in astrocyte medium (ScienCell Research Laboratories, Carlsbad, CA, USA). The human GBM cell lines LN229 and T98G were purchased from the American Type Culture Collection (Manassas, VA, USA). The human GBM cell lines, U178 and H4, were purchased from iCell Bioscience (Shanghai, China). The human GBM cell lines, U87, U118, and U251, were purchased from the Chinese Academy of Sciences cell bank (Shanghai, China). All GBM cell lines were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM; HyClone, Logan, UT, USA), supplemented with 10% fetal bovine serum (FBS; Gibco, Carlsbad, CA, USA) and 1% penicillin/streptomycin (Gibco) at 37°C with 5% CO₂. Six patient-derived primary glioma stem cells (GSCs) with World Health Organization (WHO) grade II, III, and IV (WHO grade II: GSC2C and GSC2D; III: GSC3C and GSC3D; and IV: GSC4C and GSC4D) were cultured as previously described (23). The detailed clinicopathological information is presented in Supplementary Table 1 . The stemness of GSCs was detected by immunofluorescence staining of CD133 and nestin (Abcam, Cambridge, UK) and the multi-lineage differentiation capacity of GSCs was detected by immunofluorescence staining of GFAP and β III tubulin (Abcam). Human recombinant TGF-β1 (Abcam) was used as a standard at a concentration of 5 ng/mL.

Eighty-seven clinical samples from glioma patients were collected from January 2007 to January 2012 at the First Affiliated Hospital of China Medical University. There were 45 lower-grade gliomas and 42 glioblastomas. During the same period, 18 acute brain injury samples from patients were collected as the control group. Clinical information for these samples is provided in Supplementary Table 2 . This study received the approval of the Ethics Committee of the First Affiliated Hospital of China Medical University, and each patient signed an informed consent form.

The lentivirus transfection was performed as previously described (23). The lentivirus-based vectors for RORA, CASC2, and EIF4A3 overexpression, the RNAi mediated knockdown of RORA, CASC2, EIF4A3, and its negative control were all acquired from Gene-Chem (Shanghai, China). The sequences of all siRNAs are listed in Supplementary Table 3 . The transfection efficacy was detected by RT-qPCR and western blotting.

RT-qPCR was performed as previously described (23). The Mini-BEST Universal RNA Extraction kit (TaKaRa, Kyoto, Japan) was used to extract the total RNA of glioma cells. Then, the Prime-Script RT Master Mix (TaKaRa) was used for first-strand cDNA synthesis. Finally, the RT-qPCR assays were detected using the SYBR Green Master Mix (TaKaRa) via a PCR LightCycler480 (Roche Diagnostics, Basel, Switzerland). Primers used in this study are listed in Supplementary Table 4 .

Western blotting was performed as previously described (23). A total cell protein extraction kit (KeyGen Biotechnology, Nanjing, China) was used to isolate the total proteins of GBM cells or tissues, followed by electrophoresis and transferring to a nitrocellulose membrane, then blocked for 2 h at room temperature with 2% bovine serum albumin (KeyGen Biotechnology). The membranes were incubated overnight at 4°C with the following primary antibodies against: RORA (1:1,000; Abcam), EIF4A3 (1:1,000; Abcam), TGF-β1 (1:1,000; Abcam), p-SMAD2 (1:1,000; Cell Signaling Technology, Danvers, MA, USA), SMAD2 (1:1,000; Cell Signaling Technology), p-SMAD3 (1:1,000; Cell Signaling Technology), SMAD3 (1:1,000; Cell Signaling Technology), Snail (1:1,000; Cell Signaling Technology), Slug (1:1,000; Cell Signaling Technology), E-cadherin (1:2,000; Abcam), vimentin (1:2,000; Abcam), N-cadherin (1:2,000; Abcam), p-AKT (1:1,000; ProteinTech, Chicago, IL, USA), AKT (1:1,000; ProteinTech), p-ERK (1:2,000; ProteinTech), ERK (1:2,000; ProteinTech), p-JNK (1:3,000; ProteinTech), JNK (1:3,000; ProteinTech), p-p38 (1:1,000; ProteinTech), p38 (1:1,000; ProteinTech), and β-actin (1:2,000; ProteinTech). Following a 2 h treatment with secondary antibodies (ProteinTech, Rosemont, IL, USA), all bands were detected using a chemiluminescence ECL kit (Beyotime Biotechnology, Beijing, China) and quantified by ImageJ software (National Institutes of Health, Bethesda, MD, USA). The relative expression was calculated using β-actin as the internal control.

IHC was performed as previously described (23). First, the tumor tissues were embedded in paraffin, sliced into 4 mm sections, and labeled with the following primary antibodies against RORA (1:100; Abcam), TGF-β1 (1:100; Abcam), E-cadherin (1:100; Abcam), and vimentin (1:100; Abcam). Then, the slices were stained with an immunohistochemical labeling kit (MaxVision Biotechnology, Fuzhou, China) and imaged using light microscopy (Olympus, Tokyo, Japan). Finally, the staining intensities and the expression levels were evaluated according to the German immunohistochemical score (10).

Immunofluorescence staining was performed as previously described (23). First, the GBM cells were fixed with 4% paraformaldehyde (Solarbio, Beijing, China) for 10 min, permeabilized with 0.5% Triton X-100 (Solarbio) for 20 min, blocked with 5% bovine serum albumin (Solarbio) for 1 h, and probed with primary antibodies to CD133, nestin, GFAP, βIII-tubulin, E-cadherin, and vimentin (1:100; Abcam) at 4°C overnight. Then, all GBM cell samples were treated with fluorescein isothiocyanate or rhodamine-conjugated secondary antibodies. Subsequently, the cells were counterstained with 4ʹ,6-diamidino-2-phenylindole (Sigma-Aldrich, Shanghai, China). Finally, the staining was visualized using a laser scanning confocal microscope (Olympus).

For the migration assay, the cells were resuspended in serum-free medium (HyClone) at a density of 2 × 10⁵ cells/mL, and then 100 μL of cell suspension was seeded into the upper chamber (Costar, Corning, NY, USA), and 600 μL DMEM/high glucose medium (HyClone) with 10% FBS was added to the lower chamber. After incubation at 37°C for 24 h, the cells were fixed with 4% paraformaldehyde (Solarbio) for 10 min and stained with 1% Crystal Violet solution (Solarbio) for 20 min in room temperature. Finally, the cell numbers were counted by calculating the average of five random fields using an inverted microscope (Olympus). For the Transwell assay, the 8 μm pore size polycarbonate membrane was covered with 80 μL of 50 ng/μL Matrigel solution (BD, Franklin Lakes, NJ, USA). The other steps were the same as the migration assay.

According to the manufacturer’s instructions, the RIP assay was performed via the Imprint RNA Immunoprecipitation Kit (Sigma, USA). All GSCs lysates under different conditions were incubated with RIP buffer, including magnetic beads conjugated with the negative control IgG, anti-RORA, or anti-EIF4A3 antibodies (Millipore, UK). The immunoprecipitated RNAs were acquired after incubated with Proteinase K buffer (Omega, Shanghai, China). Finally, the precipitants were detected using RT-qPCR.

The RNA pull-down assays were performed as previously described (24). The Pierce Magnetic RNA Protein pull-down Kit (Thermo Fisher Scientific) was used according to the manufacturer’s instructions. Briefly, the biotinylated RNA probes were used to label the purified RNA, and then the biotinylated RNA, the negative control (antisense RNA), and the positive control (input) were mixed and co-incubated with proteins of GSCs. To prepare a probe-magnetic bead complex, the RNA-protein complex was added with magnetic beads. After being washed and boiled, the complexes were detected by western blotting, and β-actin was used for the control.

All nascent RNAs were detected using the Click-iT nascent RNA capture kit (Thermo Fisher Scientific, USA) according to the manufacturer’s instructions. Briefly, all nascent RNAs were treated with 5-ethynyl uridine (EU), then streptavidin magnetic beads were used to capture the EU-nascent RNAs and finally detected via RT-qPCR.

Actinomycin D (ActD; NobleRyder, China) was added into the cell culture medium to restrain the de novo synthesis of RNA. Then, at different time points, total RNA was isolated and detected by RT-qPCR. Finally, at a certain time point, the half-life of RNA was confirmed by its level decreasing to 50%.

Luciferase reporter assays were performed as described previously (25). Firstly, the EIF4A3 and TGF-β1 reporter plasmids were constructed by Gene-Chem. The predicted 3’-UTR sequences of EIF4A3 and TGF-β1 and their corresponding mutant sequence were cloned into pGL3 Dual-Luciferase Vector to construct the luciferase reporter vectors (EIF4A3-Wt or EIF4A3-Mut, TGF-β1-Wt or TGF-β1-Mut). Then, the GSCs were seeded into 96-well plates at a density of 5 × 10³ cells/well, transfected with EIF4A3-Wt or EIF4A3-Mut, TGF-β1-Wt or TGF-β1-Mut reporter plasmids, and incubated for 48 h. Finally, the luciferase activities were detected using a Dual-Luciferase Reporter Assay System (Promega, Madison, WI, USA).

ELISAs were performed as previously described (23). The human TGF-β1 Quantikine ELISA Kit (R&D Systems, Minneapolis, MN, USA) was used to detect the concentrations of TGF-β1 in the media supernatants of the GBM cells and GSCs. All ELISA readings were normalized to the protein concentration in the control groups.

Xenograft experiments were conducted according to the Animal Care Committee of China Medical University. Six-week-old male BALB/c nude mice (Beijing Vital River Laboratory Animal Technology, Beijing, China) were divided into six groups: control, RORA-OE, CASC2-KD1, CASC2-KD1+RORA-OE, EIF4A3-OE, and EIF4A3-OE+RORA-OE. Each group with five mice was bred in the Laboratory Animal Center of China Medical University under specific pathogen-free conditions. The GSC4D treated with different conditions was orthotopically injected into the mouse brain at 2 mm lateral and 2 mm anterior to the bregma using a stereotaxic apparatus (5 × 10⁴ cells each mouse). Each group was observed daily for distress or death signs. The mice were sacrificed, the tumors isolated, and the tumor volume was calculated according to the formula: V = (D × d²)/2, (D was the longest diameter and d was the shortest diameter). The overall survival times of mice were detected through Kaplan-Meier survival analysis.

The data on RORA mRNA expressions of glioma patients were obtained from the Chinese Glioma Genome Atlas (CGGA, http://www.cgga.org.cn) using the RNA-seq platform and The Cancer Genome Atlas (TCGA, http://cancergenome.nih.gov) in HG-U133A platforms. GSEA (http://www.broadinstitute.org/gsea/index.jsp) was used to detect the enrichment of signaling pathways between the high and low RORA expression groups. Starbase (http://starbase.sysu.edu.cn) was used to predict the relative lncRNAs by examining the lncRNAs and RORA 3’-UTR using bioinformatics algorithms.

All experiments were repeated at least three times, and the results are expressed as the mean ± SD using SPSS statistical software for Windows, version 22.0 (SPSS, Chicago, IL, USA). Comparisons of two independent groups were determined using the chi-square test and two-tailed Student’s t-test. The statistical significance among three or more groups was evaluated by one-way analysis of variance. Pearson’s correlation analysis was used to detect the correlation between the two groups. Kaplan-Meier analysis and the log-rank test were conducted to analyze the survival rates of each group. Two-tailed P values < 0.05 were considered significant.

We first detected the expressions of RORA in 87 clinical glioma specimens and 18 normal brain tissues using RT-qPCR, western blotting, and immunohistochemical staining ( Figures 1A–C ). These results showed that RORA expressed lowest in GBM, followed with lower-grade glioma, and highest expression in normal brain tissues. Besides, we further conducted a Kaplan-Meier survival analysis to evaluate whether the expressions of RORA affected the prognoses of glioma patients ( Figures 1D–F ). The results showed that RORA significantly impacted survival in GBM patients. However, RORA had no significant effect on the prognoses of lower-grade glioma patients.

We successfully cultured six glioma stem cell lines derived from GBM patients. Afterward, the six GSCs populations were designated GSC2C and GSC2D (WHO grade II), GSC3C and GSC3D (WHO grade III), and GSC4C and GSC4D (WHO grade IV). All GSCs were confirmed using immunofluorescence staining of stem cell markers, CD133 and nestin ( Figure S1A ), and differentiation markers, GFAP and β-III tubulin ( Figure S1B ). Both RT-qPCR and western blotting were performed to detect the expressions of RORA in GSCs and common GBM cell lines and normal human astrocytes. The results showed that RORA expression in GSC2C was the highest and in GSC4D were the lowest. We also found that the expressions of RORA in common GBM cell lines were lower than those in normal human astrocytes ( Figures 1G–J ).

Since there was significantly low RORA expression in GBMs, we predicted that RORA might be associated with the aggressiveness of GBMs. U87 cells and GSC2C with the highest RORA expression were used for RORA knockdown, while T98G cells and GSC4D were chosen for RORA overexpression ( Figures 1G–J ). Both RT-qPCR and western blotting were performed to validate RORA overexpression ( Figures 2A, B ). We further detected the migration, invasion, and EMT in RORA-overexpressed T98G cells and GSC4D by migration assays, transwell assays, phase contrast microscope, immunofluorescence staining, and western blotting. All results showed that the migration rates, invasion rates, the relative number of mesenchymal-like cells, expression levels of vimentin and N-cadherin significantly decreased, and the relative number of epithelioid-like cells and E-cadherin expression increased ( Figures 2C–G ). However, we obtained the opposite results in RORA-silenced U87 cells and GSC2C ( Figures S2A–G ).

To identify the specific signaling pathway involved in RORA-regulated migration, invasion, and EMT of GBM, we conducted a GSEA based on the CGGA and TCGA databases and found that RORA expression was associated with the TGF-β signaling pathway ( Figures 3A, B ). The correlations between RORA and TGF-β1 expressions in 87 cases of clinical glioma specimens were then detected by RT-qPCR, which showed significant negative correlations in lower-grade gliomas and GBMs ( Figures 3C, D ). Then, we detected the expressions of TGF-β1 after RORA knockdown or overexpression by RT-qPCR, western blotting, and ELISA assays. All results showed that the expressions of TGF-β1 were upregulated in RORA-silenced U87 cells and GSC2C, while the opposite results were found in RORA-overexpressed T98G cells and GSC4D ( Figures 3E, F, H, I, L, M ). We further detected the downstream proteins of the TGF-β1/Smad signaling pathway by western blotting and found that the expression of p-SMAD2, p-SMAD3, Snail, and Slug were significantly upregulated in RORA-silenced U87 cells and GSC2C ( Figure 3L ). We obtained the opposite results in RORA-overexpressed T98G cells and GSC4D ( Figure 3M ). Besides, we also performed western blotting to detect the non-canonical signaling of TGF-β1, including MAPK pathways, JNK, p38 and PI3K cascade. The results showed there was no change in the expression levels of p-ERK, p-JNK, p-p38 and p-AKT after RORA overexpression ( Figure S3 ). Because RORA is a transcription factor, we investigated whether RORA transcriptionally regulated TGF-β1 expression ( Figure 3G ). We performed luciferase reporter assays and found that the relative luciferase activities of TGF-β1 significantly increased in RORA-silenced U87 cells and GSC2C; however, the opposite results were obtained in RORA-overexpressed T98G cells and GSC4D ( Figures 3J, K ). Taken together, RORA negatively affected the TGF-β1/Smad signaling pathway.

To confirm whether RORA inhibited the migration, invasion, and EMT of GBM via negatively affecting the TGF-β1/Smad signaling pathway, the RORA-overexpressed GSC4D was treated with human recombinant TGF-β1. Both the migration assays and the transwell assays found that RORA overexpression-induced migration and invasion inhibition was reversed after TGF-β1 treatment ( Figures S4A, B ). We then used phase-contrast microscopy to observe the cell morphology and found that RORA-overexpressed GSC4D showed more mesenchymal-like morphology after TGF-β1 treatment. However, the epithelioid-like morphology decreased after TGF-β1 treatment ( Figure S4C ). Subsequently, we detected E-cadherin, vimentin, and N-cadherin by immunofluorescence staining and western blotting ( Figures S4D, E ). The results showed that E-cadherin expression decreased and vimentin and N-cadherin expression increased in RORA-overexpressed GSC4D after TGF-β1 treatment. Besides, the expression of p-SMAD2, p-SMAD3, Snail, and Slug were also upregulated in RORA-overexpressed GSC4D after TGF-β1 treatment. In summary, RORA inhibited the migration, invasion, and EMT of GBM via inhibiting the TGF-β1/Smad signaling pathway. Moreover, we also explored whether RORA inhibited the proliferation of GBM via TGF-β1/Smad signaling pathway. Both MTS and EDU assays confirmed that RORA can inhibit the proliferation of GBM and these inhibiting effects were reversed after TGF-β1 treatment ( Figures S5A, B ).

Increasing evidence has revealed that lncRNAs in the cytoplasm may conduct as decoys for miRNAs or proteins (26). The expression level of lncRNA CASC2 is downregulated in glioma, which is similar to our previous research that CASC2 serves as a tumor suppressor. Moreover, we have certified that RORA mRNA and protein were also downregulated in glioblastoma. Therefore, we further studied the relationship between CASC2 and RORA in GBM. It was predicted that CASC2 could bind to RORA proteins according to catRAPID ( Figure 4A ). Then RIP assays were performed and indicated an obvious enrichment of CASC2 co-precipitated within RORA immunocomplex. The relative enrichment of CASC2 in the anti-RORA group was obviously decreased after RORA knockdown but increased after RORA overexpression. However, the relative enrichment of CASC2 in the IgG-treated group showed no significant changes ( Figures 4B, C ). Moreover, RNA pull-down assays also showed that CASC2 could bind to RORA protein in both GSC2C and GSC4D ( Figures 4D, E ). Besides, we performed RT-qPCR to validate the efficiency of CASC2 knockdown or overexpression ( Figure 4F ). Our study also certified that CASC2 could upregulate the expression of RORA protein according to the western blotting ( Figures S6A, B ). Surprisingly, we found that RORA mRNA expression had no significant change after CASC2 overexpression or knockdown via RT-qPCR ( Figures S6C, D ).

Next, we tried to explore the possible molecular function between CASC2 and RORA protein. Since the protein levels in tumor tissues could be regulated by transcription, mRNA translation, protein stability, or proteasome-mediated degradation (27), we speculated that CASC2 might upregulate RORA protein expression by maintaining the protein stability. Notably, the reduced protein level of RORA by CASC2 knockdown was apparently recovered after MG-132 treatment in GSC2C, while the opposite results were obtained in CASC2-overexpressed GSC4D ( Figure 4G ). Moreover, cycloheximide (CHX) chase assay demonstrated that RORA in CASC2-silenced GSC2C showed shorter half-life, while the half-life of RORA was much longer in CASC2-overexpressed GSC4D than that in controls ( Figures 4H, J ), suggesting that the regulation of RORA protein by CASC2 might through inhibiting proteasome degradation. Taken together, these results suggested that CASC2 could bind to RORA proteins and actively regulate the expression of RORA proteins.

Since lncRNA can regulate the distribution of targeted genes in the nucleus and cytoplasm, the distribution of RORA in GSCs was observed via immunofluorescence staining ( Figure 4I ). The results showed that RORA levels presented in the nucleus and cytoplasm were obviously decreased in CASC2-silenced GSC2C, while an increased RORA level was presented after CASC2 overexpression in GSC4D. Then western blotting analysis of nuclear and cytosolic fractions was performed, and similar results were also obtained ( Figures 4K, L ). CASC2 overexpression can upregulate RORA expression in nuclear and cytosolic, while the opposite results were obtained after CASC2 knockdown. Together, these results suggested that CASC2 could promote the nuclear translocation of RORA.

Previous studies have confirmed that CASC2 functions as a suppressor for glioma and could inhibit glioma cell proliferation (28, 29). To further confirm that CASC2 inhibited the migration, invasion, and EMT of GBM, we first detected the migration, invasion, and EMT in the CASC2-overexpressed GSC4D. All results showed that the migration rates, invasion rates, the relative number of mesenchymal-like cells, the expression levels of vimentin and N-cadherin significantly decreased, and the relative number of epithelioid-like cells and E-cadherin expression were increased in CASC2-overexpressed GSC4D, while the inhibitory effects of CASC2 overexpression were reversed after RORA knockdown via rescue experiments ( Figures 5A–D ). In summary, CASC2 inhibited the migration, invasion, and EMT of GBM, and the inhibiting effects were restrained following RORA knockdown.

RBPs are a special protein that can bind to lncRNA and regulate its processing, transport, localization, and stability (30). We found that EIF4A3 was the most probable RBP to interact with CASC2 according to the Starbase database. Firstly, we performed RT-qPCR and western blotting to validate EIF4A3 silencing or overexpression ( Figures 6A–D ). Then RIP assay showed that the enrichment of CASC2 was significantly increased in the anti-EIF4A3 group compared with the negative control anti-IgG group. The relative enrichment of CASC2 in the anti-EIF4A3 group was obviously increased after EIF4A3 knockdown but decreased after EIF4A3 overexpression. However, the relative enrichment of CASC2 in the IgG-treated group showed no significant changes ( Figures 6E, F ). Moreover, RNA pull-down assays also showed that the CASC2-wt probe pulled down EIF4A3 in GSC2C and GSC4D, rather than the CASC2-mt probe ( Figures 6G, H ).

We also performed RT-qPCR and found that the expression of CASC2 increased in GSC4D after EIF4A3 knockdown while decreased in GSC2C after EIF4A3 overexpression ( Figure 6I ). To further elucidate the mechanism of EIF4A3 downregulated CASC2 expression, we next analyzed the transcription of nascent CASC2 in EIF4A3-silenced GSC4D and EIF4A3-overexpressed GSC2C by RT-qPCR. All these results showed that EIF4A3 could not regulate the transcription of nascent CASC2 ( Figure 6J ). However, RNA stability measurement showed that the half-life of CASC2 was significantly prolonged after EIF4A3 knockdown, while shortened after EIF4A3 overexpression ( Figure 6K ). Taken together, these results suggested that EIF4A3 could bind to CASC2 and downregulated the expression of CASC2 via inducing its cleavage.

We further explored the interaction between RORA and EIF4A3. Both RT-qPCR and western blotting showed that the expressions of EIF4A3 were increased after RORA knockdown while decreased after RORA overexpression ( Figures 6L, M ). We next performed luciferase reporter assays and found that the relative luciferase activities of EIF4A3 significantly increased in RORA-silenced GSC2C. However, the opposite results were obtained in RORA-overexpressed GSC4D ( Figure 6N ). All these results suggested that RORA transcriptionally inhibited the expression of EIF4A3.

To confirm the common effects of EIF4A3 and CASC2 on the migration, invasion, and EMT of GBM, we detected the migration, invasion, and EMT in EIF4A3-overexpressed GSC2C. All results showed that the migration rates, invasion rates, the relative number of mesenchymal-like cells, expression levels of vimentin and N-cadherin were significantly increased, and the relative number of epithelioid-like cells and E-cadherin expression were decreased in EIF4A3-overexpressed GSC2C, while the promoting effects of EIF4A3 overexpression were reversed after CASC2 overexpression ( Figures S7A–D ). In summary, EIF4A3 promoted the migration, invasion, and EMT of GBM, and the promoting effects were reversed following CASC2 overexpression. Besides, we detected the expression of TGF-β1 after CASC2 or EIF4A3 knockdown or overexpression by qPCR and western blotting. The results showed that CASC2 negatively regulated TGF-β1 expression ( Figures S8A–D ), while EIF4A3 positively regulated the expression of TGF-β1 ( Figures S8E–H ).

To evaluate the effects of RORA on GBM tumorigenesis, we further constructed orthotopic xenograft models. The results showed that tumor volumes were decreased in the RORA-OE group, the CASC2-KD1+RORA-OE group, and the EIF4A3-OE+RORA-OE group when compared with the control group, while the tumor volumes were significantly increased in the CASC2-KD1 group and the EIF4A3-OE group compared with the control group ( Figures 7A, B ). Besides, IHC assays showed that in the RORA-OE group, the CASC2-KD1+RORA-OE group, and the EIF4A3-OE+RORA-OE group, the staining intensity and expression levels of RORA and E-cadherin were increased compared with the control group, while the expression levels of TGF-β1 and vimentin were decreased. However, the opposite results were obtained in the CASC2-KD1 group and the EIF4A3-OE group ( Figure 7C ). Moreover, Kaplan-Meier survival analysis showed that the survival times in the RORA-OE group, the CASC2-KD1+RORA-OE group, and the EIF4A3-OE+RORA-OE group were longer than the control group, while the opposite results were obtained in the CASC2-KD1 group and the EIF4A3-OE group ( Figure 7D ). To illustrate our findings, the schematic diagram in Figure 7E shows that the EIF4A3/CASC2/RORA feedback loop regulates the tumorigenesis, migration, invasion, and EMT of GBM through negatively affecting the TGF-β1/Smad signaling pathway.