The U2AF2 /circRNA ARF1/miR-342–3p/ISL2 feedback loop regulates angiogenesis in glioma stem cells

Seventy clinical samples from glioma patients were collected from January 2007 to December 2012 at the First Affiliated Hospital of China Medical University. There were 20 samples of grade II, 25 samples of grade III, and 25 samples of grade IV glioma. During the same period, 10 more acute brain injury patient samples were collected as a control group. Clinical information for these samples is outlined in Table S1. This study was approved by the Ethics Committee of the First Affiliated Hospital of China Medical University, and written informed consent was obtained from each patient.

Human brain microvessel endothelial cells (hBMECs) were purchased from ScienCell Research Laboratories (San Diego, CA, USA). The hBMECs were maintained in endothelial cell medium (ECM; ScienCell Research Laboratories). Six patient-derived primary glioma stem cells from WHO grade II to IV (grade II: GSC205 and GSC207; grade III: GSC306 and GSC307; grade IV: GSC406 and GSC408) were isolated, and neurosphere cultures were obtained as previously described [23]. The detailed clinicopathological information is presented in Table S2. Briefly, freshly resected glioma samples were dissociated into single cells and grown in serum-free DMEM/F12 with 2% B27, 20 ng/mL rh-bFGF, and rh-EGF (Gibco, Gaithersburg, MD, USA). The stem cell markers of GSCs were detected by immunofluorescence using anti-CD133 (Abcam Technology, Cambridge, UK) and nestin antibodies (Abcam). The immunofluorescence staining of glial fibrillary acidic protein (GFAP; Abcam) and β-III tubulin (Abcam) was used to evaluate the multi-lineage differentiation capacity of GSCs.

The preparation of GCM has been previously described [24]. Briefly, we used serum-free DMEM/F12 to wash GSCs three times, followed by culturing the GSCs for 24 h. The medium was then collected and centrifuged at 3000×g for 15 min at 4 °C to remove GSCs and debris. The GCM was prepared and used immediately for the treatment of hBMECs, followed by subsequent experiments, or stored at − 80 °C for no more than 1 week.

The lentivirus-based vectors for ISL2 overexpression, U2AF2 overexpression, cARF1 overexpression, RNAi-mediated knockdown of ISL2, U2AF2 and cARF1, and their negative controls were all constructed by Gene-Chem (GV358, Shanghai, China). The detailed sequence of the lentivirus-based vectors can be obtained on the GeneChem website (http://​www.​genechem.​com.​cn/​index/​supports/​zaiti_​info.​html?​id=​50). The miR-342–3p mimic, inhibitor, and their negative controls were obtained from Thermo Fisher Scientific (Assay ID: MH12328 and MC12328; Thermo Fisher Scientific, Waltham, MA, USA). The sequences of all siRNAs are listed in Table S3. The lentivirus transfection and efficacy measurements were performed as previously described [23].

Real-time PCR was performed as previously described [23]. The Mini-BEST Universal RNA Extraction kit (TaKaRa, Kyoto, Japan) was used to extract the total RNA of GSCs. For circRNA and mRNA, the RNA was reverse transcribed into cDNA using a Prime Script RT Master Mix reagent kit (TaKaRa). The qPCR assays were detected using the SYBR Green Master Mix (TaKaRa) with PCR LightCycler480 (Roche Diagnostics, Basel, Switzerland). Furthermore, RNase R (Epicentre Technologies, Madison, WI, USA) was used to confirm the existence of cARF1 and eliminate the effect of linear ARF1 RNA. The β-actin was used as an endogenous control. For miRNA, cDNA was synthesized using the PrimeScript™ RT reagent kit (TaKaRa, Shiga, Japan). The expression levels of miR-342–3p were detected using the TaqMan Universal Master Mix II (Assay ID: 002260; Applied Biosystems, Foster City, CA, USA). The U6 housekeeping gene was used as an endogenous control (Assay ID: 001973, Applied Biosystems). Primers used in this study are listed in Table S4.

Western blotting was performed as previously described [23]. Briefly, the total proteins of GSCs or tissues were isolated using a total cell protein extraction kit (KeyGen Biotechnology, Nanjing, China). Protein lysates were prepared, subjected to SDS-PAGE, transferred onto polyvinylidene difluoride membranes and blocked with 2% bovine serum albumin (KeyGen Biotechnology). The primary antibodies against ISL2 (1:1000; Abcam), VEGFA (1:1000; Abcam), VEGFR2 (1:1000; Abcam), p-VEGFR2 (1:1000; Abcam), MEK1/2 (1:1000; Abcam), p-MEK1/2 (1:1000; Abcam), ERK1/2 (1:500; Abcam), p-ERK1/2 (1:500; Abcam), and β-actin (1:2000; Proteintech, Rosemont, IL, USA) were incubated at 4 °C overnight. After secondary antibody (Proteintech) incubation, the bands were detected using a chemiluminescence ECL kit (Beyotime Biotechnology, Beijing, China) and quantified by ImageJ software (National Institutes of Health, Bethesda, MD, USA).

IHC was performed and the results were semi-quantified as previously described [23]. Briefly, the paraffin-embedded tissue sections were labeled with primary antibody against ISL2 (1:100; Abcam), U2AF2 (1:100; Abcam), VEGFA (1:100; Abcam), and CD31 (1:100; Abcam). The sections were then treated with an immunohistochemical labeling kit (MaxVision Biotechnology, Fuzhou, China) and photographed with a light microscope (Olympus, Tokyo, Japan). The German immunohistochemical score was used to evaluate the staining intensity and expression levels [25].

Immunofluorescence staining was performed as previously described [24]. Briefly, the GSCs were fixed with 4% paraformaldehyde, permeabilized with 0.5% Triton X-100, blocked with 5% bovine serum albumin, and detected with primary antibodies against CD133, nestin, GFAP, and β-IIItubulin (1:100; Abcam) at 4 °C overnight. The samples were stained by fluorescein isothiocyanate- or rhodamine-conjugated secondary antibodies. Finally, GSCs were counterstained using 4′,6-diamidino-2-phenylindole (Sigma-Aldrich, St. Louis, MO, USA) and were visualized using a laser scanning confocal microscope (Olympus).

The hBMECs were plated in 96-well plates at a density of 1000 cells/well and incubated in GCM for 0, 24, 48, 72, 96, and 120 h. Cell viability was determined using the CellTiter 96® Aqueous Non-Radioactive Cell Proliferation Assay Kit (Promega, Madison, WI, USA) according to the manufacturer’s instructions.

The EDU assay was conducted to examine the proliferation of cells using an EDU assay kit (Beyotime, Biotechnology) according to the manufacturer’s protocol. Briefly, the hBMECs were treated with GCM and seeded into 24-well plates at 1 × 10⁵ cells/well for 24 h, then 10 μM EDU reagent was added to the medium and incubated for 2 h. After being fixed and permeabilized, the hBMECs were counterstained. The percentage of EDU positive cells was calculated using a laser scanning confocal microscope (Olympus).

For the Transwell invasion assay, approximately 1 × 10⁵ hBMECs under different conditions were plated in the upper chamber (Corning, Corning, NY, USA) with a Matrigel filter (BD Biosciences, San Jose, CA, USA) and ECM medium with 10% fetal bovine serum was added to the lower chamber. After incubation for 24 h, the invaded cells were fixed with 4% paraformaldehyde and stained with Crystal Violet (Beyotime, Biotechnology). The stained cells were photographed and counted using a light microscope (Olympus).

The tube formation assay was performed as previously described [24]. Briefly, pre-chilled 96-well plates were coated with 70 μL Matrigel filter reagent (BD Biosciences) per well at 37 °C for 30 min. The hBMECs under different conditions were seeded on the surface of the Matrigel at 2 × 10⁴ cells/well at 37 °C for 4 h. A microscope (Olympus) was used to visualize the images for each well, and Image J software was used to calculate the total number of branches and tubule lengths.

The ELISA was performed using a commercial kit (Cusabio, Stratech, UK) to detect the concentration of VEGFA in the supernatant of the GSCs medium, as previously described [25]. All results were normalized to the protein concentration in the control group.

Luciferase reporter assays were performed as previously described [24]. Briefly, the luciferase reporter plasmids (VEGFA-wt and VEGFA-mt, ISL2–3′-UTR-wt and ISL2–3′-UTR-mt, cARF1-wt and cARF1-mt, and U2AF2-wt and U2AF2-mt) were constructed by Gene-Chem (GV102). The detailed sequence can be obtained on the GeneChem website (http://​www.​genechem.​com.​cn/​index/​supports/​zaiti_​info.​html?​id=​). The luciferase reporter plasmids were co-transfected into GSCs. After 48 h, the luciferase activities were detected using a Dual-Luciferase Reporter Assay System (Promega). Relative luciferase activity was calculated as the ratio of firefly luciferase activity to Renilla luciferase activity.

ChIP assays were performed using the ChIP Assay Kit (Beyotime Biotechnology) according to the manufacturer’s instructions. The chromatin complexes were immunoprecipitated using anti-ISL2 antibody or normal rabbit IgG, and the purified DNA samples were analyzed by qPCR. The primers for ChIP qPCR are listed in Table S4.

The RIP assay was performed using the EZ-magna RIP RNA-binding Protein Immunoprecipitation kit (Millipore, Darmstadt, Germany) according to the manufacturer’s protocols. GSCs under different conditions were lysed in RIP buffer including magnetic beads conjugated with negative control IgG, anti-AgO2, or anti-U2AF2 antibodies (Millipore). After incubation with proteinase K, the immunoprecipitated RNAs were isolated. Finally, qRT-PCR was used to examine the precipitants.

The Pierce Magnetic RNA Protein pull-down Kit (Thermo Fisher Scientific) was used to detect the interaction between cARF1 and U2AF2 according to the manufacturer’s suggestions. Briefly, biotinylated RNA probes were used to label purified RNA, and then the positive control (input), negative control (antisense RNA), and biotinylated RNA were mixed and co-incubated with GSCs proteins at room temperature. The RNA-protein complex was added with magnetic beads to prepare a probe-magnetic bead complex. After being washed and boiled, the complexes were detected by western blotting, using β-actin as a control.

GSCs were cultured in the medium containing 2 μg/ml actinomycin D (Act D, NobleRyder, China) to block the de novo RNA synthesis. Then total RNA was collected at indicated times and cARF1 expression was detected by qRT-PCR. The half-life of cARF1 was determined as the time required to reach 50% of the RNA levels before actinomycin D treatment.

Xenograft experiments were performed as previously described [24]. Under specific pathogenic conditions, 6-week-old female BALB/c nude mice (Beijing Vital River Laboratory Animal Technology, Beijing, China) were raised at the Laboratory Animal Center of China Medical University. GSCs under different conditions were injected (5 × 10⁴ cells per mouse) orthotopically into the mouse brains, 2 mm lateral and 2 mm anterior to the bregma using a stereotaxic instrument (n = 5, per group). The tumor volume was measured according to the following formula: V = (D × d²)/2, where D was the longest diameter and d was the shortest diameter of the tumor. All animal experiments were performed in accordance with the Animal Care Committee of China Medical University.

The data of mRNA expression, WHO grades, isocitrate dehydrogenase (IDH) status (IDH 1/2) of ISL2 and U2AF2, the survival times, and status of glioma patients were obtained from the Chinese Glioma Genome Atlas (CGGA, http://​www.​cgga.​org.​cn) using the mRNA seq-693 dataset and The Cancer Genome Atlas (TCGA, http://​cancergenome.​nih.​gov) in the HG-U133A platform. Gene set enrichment analysis (GSEA, http://​www.​broadinstitute.​org/​gsea/​index.​jsp) was used to analyze enrichment of a biological process or signal pathway with high versus low ISL2 expressions. Four online databases, Starbase (http://​starbase.​sysu.​edu.​cn), TargetScan (www. targetscan.​org), microRNA (http://​www.​microrna.​org/​microrna/​home.​do), and miRDB (http://​mirdb.​org) were used to predict possible miRNAs targeting ISL2. Starbase and circBase (http://​www.​circbase.​org/​) databases were used to predict potential circRNAs as sponges of miRNA. The Starbase database was also used to predict the proteins binding to circRNAs.

Results are reported as the mean ± SD of at least three independent experiments. The chi-square test, two-tailed Student’s t-test, and one-way analysis of variance were used to compare the statistical significance among different groups. Pearson’s correlation analysis was used to assess the correlation between two groups. The survival difference was evaluated using a log-rank test and Kaplan-Meier analysis. SPSS statistical software for Windows, version 23.0 (IBM, Armonk, N. Y, USA) was performed for statistical analysis, and two-tailed P values < 0.05 were considered significant.

To characterize the expression and functions of ISL2 in gliomas, we first searched its expression in CGGA datasets. Compared to WHO grade II and grade III glioma, ISL2 expression was higher in glioblastoma (GBM; WHO grade IV) (Fig. 1a, b). ISL2 was also highly enriched in the IDH wildtype glioma, and was associated with decreased survival rates among different WHO grade glioma in the CGGA datasets (Fig. 1c-g). These results were also validated in TCGA datasets (Figure S1a-g). We then characterized the expressions of ISL2 in our 70 glioma patients, 10 normal brain tissues, and patient-derived primary GSCs. Compared to the normal brain tissues, all qPCR, western blotting, and immunohistochemistry results showed that ISL2 expression was higher in glioma tissues, and was especially increased in higher glioma WHO grades (Fig. 1h–j). Kaplan-Meier survival analyses showed that the median survival times of lower grade glioma (LGG) patients, GBM patients, or total glioma patients with higher ISL2 expressions were all shorter than in patients with lower ISL2 expression levels (Fig. 1k-m). Finally, both Cox univariate and multivariate analyses showed that ISL2 expression, WHO grades, and IDH status were independent prognostic factors of glioma patients (Table 1).

We successfully isolated six GSCs from glioma patients with different WHO pathological diagnoses. The original patient tumors were stained by hematoxylin and eosin (Figure S2a). The enrichment of stem cell markers, CD133 and nestin, were confirmed by immunofluorescence (Figure S2b). The multilineage differentiation capacity of GSCs and differentiation markers, GFAP and β-III tubulin, were also confirmed (Figure S2c). Both qPCR and western blotting showed that ISL2 was expressed highest in WHO grade IV GSCs (GSC406 and GSC408), followed with WHO grade III GSCs (GSC306 and GSC307), and lowest in WHO grade II GSCs (GSC205 and GSC207) (Figure S2d, e). Moreover, we found that ISL2 expression was higher in each GSC, when compared with the non-GSCs (Figure S2f, g). Together, these results suggested that ISL2 was elevated in glioma and was associated with poor patient survival.

To determine the possible effect of ISL2 on glioma, we performed gene set enrichment analysis (GSEA) of ISL2 expression based on TCGA and CGGA datasets. The results showed there was a positive association with “GO_POSITIVE_REGULATION_OF_VASCULAR_ENDOTHELIAL_ GROWTH_FACTOR_PRODUCTION” signatures in ISL2 high expression glioma. (Fig. S1h, i). According to the expression of ISL2 in GSCs shown in Figure S2d-e, GSC406 with the highest expression was treated for ISL2 silencing, while GSC205 with the lowest expression was treated for ISL2 overexpression. Lentiviral-based transfection and the effects on ISL2 silencing or overexpression were validated in Figure S3a, b. All qPCR, western blotting, and ELISA assays showed that the expression and secretion of VEGFA decreased after ISL2 silencing of GSC406, while it increased in ISL2 overexpressing GSC205 (Fig. 2a-c). As a transcription factor, we found two binding sites for ISL2 in the promoter of VEGFA when using the Jaspar database (Fig. 2d, e). A luciferase plasmid with the top 2000 nucleotides of the promoter domain of the VEGFA gene (pGl3-wt) and a luciferase plasmid with mutant sequences in both binding sites of the promoter domain (pGL3-mt) were generated (Fig. 2e). Luciferase reporter assays showed that ISL2 enhanced the luciferase activity of pGL3-wt, but not that of pGL3-mt (Fig. 2f). ChIP assays also revealed that the enrichment of VEGFA was decreased in ISL2 silencing GSC406 and increased in ISL2 overexpressing GSC205 (Fig. 2g). Together, these results showed that ISL2 transcriptionally regulated VEGFA expression in GSCs.

We evaluated the effects of ISL2-regulated GCM on the proliferation of hBMECs using MTS and EDU assays. The results showed that treatment with the conditioned medium from ISL2-silenced GSC406 decreased cell viability and the rates of EDU-positive hBMECs, while ISL2-overexpressed GSC205-GCM increased the cell viability and the rates of EDU-positive hBMECs (Figure S4a–d). Transwell assays showed that treatment with ISL2-silenced GSC406-GCM decreased the invading cell numbers of hBMECs, whereas treatment with ISL2-overexpressed GSC205-GCM increased its invasive cell numbers (Figure S4e, f). Moreover, tube formation assays showed that ISL2-silenced GSC406-GCM treatment decreased the number of branches and tubule lengths of hBMECs, while the opposite results were obtained after treatment with ISL2-overexpressed GSC205-GCM (Figure S4g-i).

Taken together, the abovementioned results suggested that ISL2 overexpression in GSCs promoted the proliferation, invasion, and angiogenesis of hBMECs. Therefore, human recombinant VEGFA or VEGFA-neutralizing antibody were combined with treatment of ISL2-silenced GSC406-GCM or ISL2-overexpressed GSC205-GCM, respectively. Both MTS and EDU assays showed that the cell viability and the rates of EDU-positive hBMECs were increased after additional human recombinant VEGFA treatment, when compared with ISL2-silenced GSC406-GCM treatment alone, while the cell viability and the rates of EDU-positive hBMECs were decreased after additional VEGFA-neutralizing antibody treatment, when compared with ISL2-overexpressed GSC205-GCM treatment alone (Fig. 2h, i). Similar results were also obtained using Transwell and tube formation assays. After additional human recombinant VEGFA treatment, all invading cell numbers, number of branches, and tubule lengths of hBMECs were increased when compared with ISL2-silenced GSC406-GCM treatment alone, whereas the opposite results were obtained after additional treatment with VEGFA-neutralizing antibody treatment (Fig. 2j, k). We further characterized the possible downstream ISL2-regulated VEGFA treatment on hBMECs. Western blotting showed that ISL2-overexpressed GSC205-GCM treatment upregulated the expression of p-VEGFR2, p-MEK1/2, and p-ERK1/2 of hBMECs (Fig. 2l), while the opposite results were obtained after treatment with ISL2-silenced GSC406-GCM (Fig. 2m). Together, the results showed that ISL2 promoted angiogenesis of hBMECs in GSCs via VEGFA-mediated ERK signaling.

To explore which miRNA negatively regulated the expression of ISL2, we searched four datasets including microRNA, miRDB, TargetScan, and Starbase to identify possible miRNAs. The results showed that miR-342–3p was the only intersection among these four datasets that bound to the 3′-UTR of ISL2 (Fig. 3a, b, Table S5). The binding site of miR-342–3p on the ISL2 3’UTR was predicted via Starbase, which had the highest “AgoExpNum”, “CleaveExpNum” and “Pan-Cancer” scores. We therefore designed luciferase reporter assays and found that miR-342–3p mimic treatment decreased the luciferase activity of the luciferase reporter plasmid with the wildtype ISL2 mRNA 3′-UTR in GSC406 (Fig. 3d), while the luciferase activity of wildtype ISL2 mRNA 3′-UTR was increased after miR-342–3p inhibitor treatment in GSC205 (Fig. 3g). We then detected the expression of miR-342–3p in our clinical glioma specimens and found its expression was negatively correlated with ISL2 expression in each WHO grade of glioma (Fig. 3c). Both qPCR and western blotting showed the expression of ISL2 was significantly decreased after miR-342–3p mimic treatment in GSC406, while it was upregulated after miR-342–3p inhibitor treatment in GSC205 (Fig. 3e, f). Based on these results, miR-342–3p was a possible upstream regulatory factor, which negatively regulated ISL2 expression by binding with the ISL2 3′-UTR.

We detected the possible functions of miR-342–3p expression of GSCs in hBMECs. The MTS, EDU, Transwell, and tube formation assay results showed the proliferation, invasion, and angiogenesis of hBMECs were decreased after miR-342–3p mimic-transfected GSC406-GCM treatment (Fig. 3h-k). The qPCR and ELISA assays also showed that the expression and secretion of VEGFA were decreased in GSC406 after transfection with the miR-342–3p mimic (Fig. 3l, m). However, the opposite results were obtained after miR-342–3p inhibitor-transfected GSC205-GCM treatment (Fig. 3h-m). To further determine whether miR-342–3p inhibited these biological functions by downregulating ISL2 expression, rescued experiments were performed with additional transfection with ISL2 overexpression or knockdown on the basis of the miR-342–3p mimic or inhibitor treatment, respectively. Compared with miR-342–3p mimic-transfected GSC406-GCM treatment alone, additional ISL2 overexpression-transfected GSC406-GCM increased the cell viability and the rates of EDU-positive hBMECs, the invading cell numbers of hBMECs, and the number of branches and tubule lengths of hBMECs as measured by MTS, EDU, Transwell, and tube formation assays, respectively (Fig. 3h-k). The qPCR and ELISA also showed VEGFA expression in GSC406 was increased after additional transfection with ISL2 overexpression (Fig. 3l-m). However, opposite results were also obtained after additional ISL2-silenced transfection when compared with the miR-342–3p inhibitor-transfected GSC205-GCM treatment alone (Fig. 3h-m). Together, these data suggested that miR-342–3p suppressed the proliferation, invasion, and angiogenesis of hBMECs by inhibiting ISL2 and VEGFA expression in GSCs.

Increasing evidence has implied that circRNAs have many microRNA response elements (MREs), which can affect the expression and biological functions of miRNAs via competing (ceRNAs) or molecular sponges [26]. We therefore searched Starbase and found that cARF1 was the appropriate circRNA that harbors one conjectural binding site of miR-342–3p (Fig. 4a). A schematic representation showing that cARF1 was generated from the ARF1 gene, located at chr1, is shown in Fig. 4b. RNase R is usually used to confirm the circular form of RNAs because of its ability to degrade linear RNAs with short 3′ tails, while it does not degrade circular RNAs. Figure 4c shows that the expression of ARF1 was decreased after RNase R treatment, while there was no change in cARF1 expression. To confirm the possibility that miR-342–3p directly bound to cARF1, we constructed full-length cARF1 sequences (cARF1-wt) and cARF1 sequences with mutant binding sites (cARF1-mt) (Fig. 4a), followed by luciferase reporter assays. The results showed that the miR-342–3p mimic significantly decreased the activity of cARF1-wt vector and miR-342–3p inhibitor increased the activity of the cARF1-wt vector, while there was no change in the activity of the cARF1-mt vector group (Fig. 4d). Because miRNAs bind to MREs via the RNA-induced silencing complex (RISC), and Argo-naute 2 (AGO2) protein is the key component of RISC [17], we performed an anti-AGO2 RIP assay to determine whether miR-342–3p and cARF1 were co-enriched in the RISC. Figure 4e shows that both cARF1 and miR-342–3p were efficiently pulled down by anti-AGO2 antibody, when compared with IgG. Moreover, significant enrichment of both cARF1 and miR-342–3p were also observed after miR-342–3p mimic treatment, when compared with the miR-342–3p negative control (Fig. 4e). The qPCR also showed that the expression of cARF1 was decreased after miR-342–3p mimic treatment in GSC406, while it increased after miR-342–3p inhibitor treatment in GSC205 (Fig. 4f). In addition, lentiviral-based transfection and the effects on cARF1 knockdown or overexpression were confirmed via qPCR (Figure S3c). Overexpression or knockdown of cARF1 led to downregulation or upregulation of miR-342–3p, respectively (Fig. 4g). In summary, these results showed that cARF1 acted as a miR-342–3p sponge in GSCs.

We examined the expression of cARF1 in our 70 glioma patients and 10 normal brain tissue samples. The qPCR results showed that cARF1 was more highly expressed in higher WHO grade glioma, while it was expressed lowest in normal brain tissues (Fig. 4h). We also performed Pearson’s correlation analyses and found that there was a negative correlation between cARF1 and miR-342–3p expressions in each WHO grade glioma and overall in all glioma samples (Fig. 4i). Kaplan-Meier survival analyses showed that glioma patients with higher cARF1 expression showed a shorter median survival time than lower expression patients (Fig. 4j).

We detected the possible functions of cARF1 on hBMECs using the abovementioned methods. MTS, EDU, Transwell, and tube formation assays results showed that the proliferation, invasion, and angiogenesis of hBMECs were decreased after cARF1-silenced GSC406-GCM treatment (Fig. 4k-n). The qPCR and ELISA assays showed that the expression and secretion of VEGFA were also decreased in GSC406 after cARF1 knockdown (Fig. 4o, p), while cARF1-overexpressed GSC205-GCM treatment promoted the proliferation, invasion, and angiogenesis of hBMECs and expression and secretion of VEGFA in GSC205 (Fig. 4k-p). To further determine whether miR-342–3p reversed all these biological functions, rescue experiments were performed with additional treatment of the miR-342–3p mimic or inhibitor on the basis of cARF1 overexpression or knockdown, respectively. Comparing cARF1-silenced GSC406-GCM treatment alone, additional treatment of the miR-342–3p inhibitor promoted the proliferation, invasion, and angiogenesis of hBMECs, and expression and secretion of VEGFA of GSC406 (Fig. 4k-p). However, opposite results were also obtained after additional miR-342–3p mimic treatment, when compared with cARF1-overexpressed GSC205-GCM treatment alone (Fig. 4k-p). These results suggested that higher cARF1 expression in GSCs also promoted the proliferation, invasion, and angiogenesis of hBMECs, while miR-342–3p reversed these functions of cARF1 in GSCs.

Since miR-342–3p can inhibit ISL2 expression via binding to its 3′-UTR and cARF1 acts as a sponge of miR-342–3p, we further determined whether cARF1 regulated the expression of ISL2 via the miR-342–3p-mediated ceRNA mechanism in GSCs. Both western blotting and qPCR showed that ISL2 expression was overexpressed after cARF1 overexpression in GSC205 and decreased after cARF1 knockdown in GSC406 (Fig. 5a, b). We also performed rescue experiments using additional treatment of the miR-342–3p mimic or inhibitor. Western blotting and qPCR also showed the expression of ISL2 was decreased after miR-342–3p mimic treatment in cARF1-overexpressed GSC406, while there was overexpression after miR-342–3p inhibitor treatment of cARF1-knockdown GSC406 (Fig. 5c-e). We also performed Pearson’s correlation analyses between cARF1 and ISL2 mRNA expressions among our glioma specimens, and found that there were strong positive correlations in each WHO grade glioma and overall in all glioma samples (Fig. 5f). Rescue experiments were further performed with additional treatment of ISL2 knockdown or overexpression on the basis of cARF1 overexpression or knockdown, respectively. Comparing the cARF1-silenced GSC406-GCM treatment alone, the MTS, EDU, Transwell, and tube formation assays results showed that the proliferation, invasion, and angiogenesis of hBMECs were increased after ISL2 overexpression combined with cARF1-silenced GSC406-GCM treatment (Fig. 5h-j). The qPCR and ELISA assays also showed that VEGFA expression and secretion were increased in GSC406 after cARF1 knockdown combined with ISL2 overexpression (Fig. 5k, l). However, the opposite results were also obtained after ISL2 knockdown combined with cARF1 overexpression in GSC205 (Fig. 5h-l). Taken together, these results suggest that cARF1 regulates ISL2 expression via a miR-342–3p-mediated ceRNA mechanism, and promotes gliomas angiogenesis by upregulating ISL2 expression in GSCs.

Previous studies have reported that RBPs interact and regulate the expression of RNAs and contribute to the malignant behaviors of tumors [27]. We searched Starbase and found that U2AF2 was the most probable RBP with the highest “Clip Exp Num” which could interact with cARF1. We further determined whether U2AF2 regulated the expression of cARF1 or its linear ARF1. Lentiviral-based transfection and the effects on U2AF2 knockdown or overexpression were validated as shown in Figure S3d, e using qPCR and western blotting. The qPCR results showed that U2AF2 overexpression upregulated cARF1 expression, and that knockdown downregulated cARF1 expression, while there was no change in linear ARF1 expression (Fig. 6a, b). RNA pull-down assays were then performed to show that biotinylated cARF1-wt pulled-down U2AF2 in GSC406 and GSC205, while cARF1-mt could not (Fig. 6c). We further performed the RIP assay to determine whether U2AF2 bound to cARF1. The relative enrichment of cARF1 in the anti-U2AF2 group was significantly increased when compared to that in the IgG treated group (Fig. 6d). U2AF2 knockdown decreased the enrichment of cARF1 in GSC406, while U2AF2 overexpression further increased the enrichment of cARF1 in GSC205 (Fig. 6d). Moreover, RNA stability measurement showed the half-life of cARF1 was obviously shortened after U2AF2 knockdown compared with negative control group (Fig. 6e). Together, these results suggested that as a type of RBP, U2AF2 directly promoted the stability and expression of cARF1.

To characterize the expression and functions of U2AF2 in glioma, we searched its expression in both TCGA and CGGA datasets. The results showed that U2AF2 was expressed highest in WHO grade IV gliomas and GBM, and lowest in WHO grade II and LGG (Fig. 7a, b). U2AF2 was also highly enriched in IDH wildtype gliomas, and was associated with decreased survival rates in both TCGA and CGGA datasets (Fig. 7a, b). We further examined its expression in our 70 glioma patients and 10 normal brain tissues. All qPCR, western blotting, and immunohistochemistry results confirmed higher U2AF2 expressions in glioma tissues than in normal brain tissues, with the highest expression in WHO grade IV gliomas (Fig. 7c, d, f). Kaplan-Meier survival analyses also showed that the median survival time of higher U2AF2 expression patients was shorter than those patients with lower U2AF2 expression levels (Fig. 7e). Together, these results suggested that U2AF2 was also more highly expressed in gliomas tissues, and was correlated with poor patient survival.

To determine whether U2AF2 promotes glioma angiogenesis, we performed MTS, EDU, Transwell, and tube formation assays and found that the proliferation, invasion, and angiogenesis of hBMECs were decreased after U2AF2-knockdown GSC406-GCM treatment (Fig. 6e-i). The qPCR and ELISA assay results showed the expression and secretion of VEGFA were also decreased in GSC406 after U2AF2 knockdown (Fig. 6j, k), while U2AF2-overexpressed GSC205-GCM treatment promoted the proliferation, invasion, and angiogenesis of hBMECs, and expression and secretion of VEGFA in GSC205 (Fig. 6e-i). Rescue experiments were also performed with additional treatment of cARF1 overexpression or knockdown on the basis of U2AF2 knockdown or overexpression, respectively. Compared to U2AF2-silenced GSC406-GCM treatment alone, the results showed that the proliferation, invasion, and angiogenesis of hBMECs were increased after cARF1 overexpression when combined with U2AF2-silenced GSC406-GCM treatment (Fig. 6j-k). The qPCR and ELISA assays also showed that VEGFA expression and secretion were increased in GSC406 after U2AF2 knockdown when combined with cARF1 overexpression (Fig. 6j, k). However, U2AF2 overexpression combined with cARF1 knockdown in GSC205 showed the opposite results (Fig. 6f-k). Taken together, these results suggested that U2AF2 promoted glioma angiogenesis via upregulating cARF1 expression in GSCs.

Because ISL2 is a transcription factor, we determined whether ISL2 transcriptionally regulated U2AF2 expression in GSCs. The Jaspar database showed that there existed two binding sites for ISL2 in the promoter of U2AF2 (Fig. 7g). Luciferase reporter assays were then performed to show that ISL2 overexpression enhanced the luciferase activity of pGL3-U2AF2-wt, and ISL2 knockdown decreased the luciferase activity of pGL3-U2AF2-wt, but not that of pGL3-U2AF2-mt (Fig. 7h). ChIP assays also showed that the enrichment of U2AF2 was decreased in ISL2-knockdown GSC406 and increased in ISL2-overexpressed GSC205 (Fig. 7i). Finally, western blotting and qPCR showed that ISL2 upregulated the expression of U2AF2 (Fig. 7j, k). Together, these results show that ISL2 transcriptionally regulates U2AF2 expression in GSCs and forms a feedback loop.

Finally, we performed orthotopic xenografts to determine the effects of the U2AF2/cARF1/miR-342–3p/ISL2 axis in glioma tumorigenesis and angiogenesis in vivo. Compared to the control group, the tumor volumes were enlarged in the ISL2 overexpression and U2AF2 overexpression groups, and decreased in the miR-342–3p-mimic and cARF1-knockdown groups (Fig. 8a, b). Moreover, ISL2 overexpression combined with the miR-342–3p mimic group also showed enlarged tumor volumes, while it was decreased in the U2AF2-overexpression combined with the cARF1-knockdown group (Fig. 8a, b). Similar results were obtained using Kaplan-Meier survival analysis, as the ISL2-overexpression and U2AF2-overexpression groups, and ISL2-overexpression combined with the miR-342–3p-mimic group showed shorter median survival times (MST) compared with the normal control group, while the miR-342–3p-mimic, cARF1-knockdown, and U2AF2-overexpression combined with cARF1-knockdown groups showed longer mean survival times (Fig. 8c). Immunohistochemistry was performed to detect the effects of the U2AF2/cARF1/miR-342–3p/ISL2 axis on tumor tissues. The results showed that the ISL2-overexpression, U2AF2-overexpression, and ISL2-overexpression combined with the miR-342–3p-mimic or cARF1-knockdown groups showed higher expression of VEGFA, CD31 and increased microvessel density (MVD) (as indicated by anti-CD31 staining), while the lowest expression was found in the miR-342–3p-mimic, cARF1-knockdown, and U2AF2-overexpression combined with the ARF1-knockdown groups (Fig. 8d-f). Finally, we detected the mRNA expression of CD31 in our glioma specimens, and found that there were positive correlations between CD31 and cARF1, ISL2 and U2AF2, while there was a negative correlation between CD31 and miR-342–3p (Figure S5). We also study the direct function of these candidate genes on the GSC properties. All, MTS, EDU and Transwell assays showed the overexpression of ILS2, cARF1 and U2AF2 can promote the proliferation and metastasis of GSCs (Figure S6). A schematic diagram showing that the U2AF2/cARF1/miR-342–3p/ISL2 feedback loop promotes glioma tumorigenesis and angiogenesis through VEGFA-mediated ERK signaling pathway is presented in Fig. 8g. Taken together, these results showed that the U2AF2/cARF1/miR-342–3p/ISL2 axis regulated glioma tumorigenesis and angiogenesis in nude mice.