This study was approved by the Ethics Committee of the Fifth Affiliated Hospital of Guangxi Medical University. Written consent was obtained from all patients enrolled in the study, or from their legal guardians. A total of 94 patients (45 males and 49 females) were recruited between January 2005 and December 2013 at the Renji Hospital. All recruited patients had a histologically confirmed diagnosis of GBM, which was independently reviewed by two pathologists and classified based on WHO criteria for tumor grading. Samples of cancerous tissue (n=5) and normal tissues (n=5) were collected from patients.

GBM data in The Cancer Genome Atlas (TCGA) (https://www.cancer.gov/) were downloaded from the UCSC Xena browser (http://xena.ucsc.edu/public), including gene expression profiles based on Affymetrix Human Genome U133a array platform (Affymetrix; Thermo Fisher Scientific, Waltham, MA, USA) and clinical information. In addition, two glioma data sets (mRNAseq_693 and mRNAseq_325) were obtained from the Chinese Glioma Genome Atlas (http://cgga.org.cn/index.jsp).

The expression of RBM8A was analyzed using western blots, which were performed using a rabbit antibody against human RBM8A (14958-1-AP, Proteintech, IL, USA; 1:1000) and mouse antibody against beta-actin (Millipore, MA, USA; 1:10,000) (32). The tissues were lysed with an RIPA buffer [50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1% Triton X-100, 0.5% Na-deoxycholate] containing protease inhibitors (CompleteMini, Roche, Basel, Switzerland). Samples of the lysates (20-30 μg) were separated on 8-12% SDS-PAGE gels and transferred to polyvinylidene fluoride membranes, which were incubated overnight with primary antibodies at 4°C, then with horseradish peroxidase-conjugated secondary antibody. An ECL kit (catalog no. PI32209, Pierce, IL, USA) was used to detect the bound antibodies.

Notch activation was assessed using the CBF1 luciferase reporter system (Dual-Glo Luciferase Assay System, Promega, USA). The Renilla luciferase plasmid pRLTK (Promega), which controls for transfection efficiency, was cotransfected with CBF1-Luciferase reporter plasmid using Lipofectamine 2000 (Thermo Fisher Scientific). Cells were harvested after 24 h in culture, and luciferase activity was determined using the Luciferase Assay System (Promega) and a microplate luminometer (Berthold, Bad Wildbad, Germany).

Immunohistochemistry was performed on the paired samples from 94 patients as described (33). Tissue microarray blocks were constructed, and serial sections (4 μm) were obtained from each block. The first slide was stained with hematoxylin and eosin (H&E) to confirm the diagnosis, and subsequent slides were stained appropriately for further analysis.

Tissue microarray slides were deparaffinized, rehydrated, blocked and stained with primary antibody against human RBM8A (14958-1-AP, Proteintech; 1:100). The sections were then stained using a highly sensitive streptavidin-biotin-peroxidase detection system and counterstained with hematoxylin. Negative controls were processed in parallel using pre-immune immunoglobulin instead of primary antibody.

Sections of cancerous tissue from each patient were independently evaluated by two pathologists blinded to clinicopathological information. Immunoreactive staining was quantified in terms of the percentage of positive cells, with a score of 0 meaning that 0% of tumor cells showed positive staining; 1, up to 10% of cells were positive; 2, 11-50%; 3, 51-75%; 4, 75-100%. The intensity of staining was quantified using a four-point scale, where 0 meant negative; 1, weak; 2, moderate; and 3, strong. The two scores for each section were then multiplied together to give a total score ranging from 0 to 12. All patients were classified based on RBM8A expression as negative (total score 0), low (score 1-4), moderate (score 5-8), or high (score 9-12).

The human GBM cell lines U87-MG, U251-MG, A172, and T98G were purchased from ATCC and cultured in Dulbecco’s modified Eagle medium (Invitrogen, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (FBS, Invitrogen), and maintained at 37°C in a humidified atmosphere of 5% carbon dioxide.

Lentiviral expression plasmids encoding small hairpin RNA (shRNA) targeting human RBM8A as well as encoding green fluorescent protein (GFP) were constructed by Hanyin (Shanghai, China). Three plasmids were constructed, each with a different RBM8A-shRNA: 5’-AGAGCATTCACAAACTGAA-3’ (RBM8A-KD1), 5’-CATCAGCGTTGACTGGTGT-3’ (RBM8A-KD2), and 5’-GCAACAGGTCTAGGGTTAAGG-3’ (RBM8A-KD3). As a negative control, lentiviral expression plasmid encoding only GFP (GFP-lentivirus) was also prepared. The knockdown efficiency of cells infected with the RBM8A-shRNA-encoding virus was confirmed after 48 h using quantitative real-time polymerase chain reaction (qRT-PCR) and western blots.

Recombinant RBM8A-shRNA lentivirus and negative control (NC) lentivirus were prepared. To obtain stable cell lines showing RBM8A knockdown (RBM8A-KD) and stable NC cell lines, U87-MG and U251-MG cells were seeded in 6-well plates at a density of 2 × 105 cells per well. On the following day, the cells were infected with the virus at a multiplicity of infection of 1 using Polybrene (8 μg/ml; Sigma-Aldrich, IL, USA). GFP expression was confirmed under a fluorescence microscope approximately 72 h after viral infection, and the culture medium was replaced with a selection medium containing puromycin (4 μg/ml). Cells were cultured for a minimum of 14 days. The puromycin-resistant cells were amplified in a medium containing 2 μg/ml puromycin for 7-9 days, then cultured in medium without puromycin.

An expression plasmid encoding FLAG-tagged RBM8A was engineering using the pMSCV-IRES-GFP vector (Hanyin), and this plasmid or the corresponding empty vector as NC were transfected into 293T cells. Recombinant retrovirus from these cells was used to infect A172 and T98G cells at a multiplicity of infection of 1, giving RBM8A-overexpressing (RBM8A-OE) cells or NC cells.

Cell proliferation over a period of five days was examined using the Cell Counting Kit-8 assay (Beyotime, Shanghai, China). Cells (n = 1000) were seeded into 96-well plates. At different time points during 5-day treatment with the γ-secretase inhibitor DAPT, CCK-8 solution (10 μl) was added to each well, and cells were incubated for 2 h at 37°C. Absorbance at 450 nm was measured using a microtiter plate reader (33). This experiment was conducted in triplicate.

A proliferation assay based on 5-ethynyl-2′-deoxyuridine (EdU) was performed according to the manufacturer’s instructions (Invitrogen).

Cells were plated in the upper chamber of Transwell assay inserts (Millipore, Billerica, MA, USA) with or without a Matrigel-coated membrane containing 8-μm pores in serum-free RPMI 1640 medium (200 μl). The inserts were then placed in the bottom chamber of a 24-well plate containing RPMI 1640 with 10% FBS as a chemoattractant. The top layer of the insert was scrubbed after 24 h with a sterile cotton swab to remove any remaining cells. The invading cells on the bottom layer were stained with 0.1% crystal violet, examined, counted, and imaged using digital microscopy. The numbers of cells in five random fields of each chamber were counted and averaged. These assays were conducted in triplicate.

This study followed institutional guidelines as well as the guidelines put forth by the US National Institutes of Health on animal welfare and experimentation. The animal experiments were approved by the Institutional Animal Care and Use Committee of Guangxi Medical University. RBM8A-KD or NC U87-MG cells were intracranially implanted into the corpus striatum of anesthetized 6-week-old athymic nude mice using a stereotactic frame for small animals (David Kopf Instruments, Tujunga, California, USA). Tumor size was monitored by magnetic resonance imaging (MRI).

Functional enrichment analysis was performed using the Clusterprofiler package in R (34). In addition, GSEA was carried out using GSEA software, which can be found on the official website (http://software.broadinstitute.org/gsea/index.jsp). Gene sets c5.bp.v7.1.symbols.gmt and c2.cp.kegg.v7.1.symbols.gmt in the MsigDB database were used as the reference gene sets (35). A nominal value of p < 0.05 was considered statistically significant.

Based on the STRING database (36), a comprehensive regulatory network involving RBM8A in GBM was constructed. We downloaded the PDB files of RBM8A from the Protein Database (https://www.rcsb.org) (37). The three-dimensional structure of CBF1 was predicted using RNA Composer (http://rnacomposer.cs.put.poznan.pl). Molecular docking was performed using Hex8.0.0 software (38), and the results were visualized with Pymol software (39).

We extracted the genes involved in the comprehensive network regulated by high RBM8A expression, and calculated the gene set variation analysis (GSVA) scores of these genes using the GSVA package (40). In addition, the receiver operating characteristic (ROC) curve for the GSVA score was analyzed using the pROC package (41), and survival analysis based on the GSVA score was performed using the “survival” package in R (https://CRAN.R-project.org/package=survival).

Patient survival was calculated from the date of surgery until the date of death or last follow-up. Survival curves for patients whose tumors contained different levels of RBM8A were plotted using the Kaplan-Meier method and compared using the log-rank test. Median survival times and hazard ratios (HRs) were presented with 95% confidence intervals (CIs). All statistical analyses were performed using the R package and SPSS for Windows 17.0 (SPSS, Chicago, IL, USA). Differences were considered significant when the two-tailed P value was less than 0.05.

First, the expression of RBM8A in GBM was analyzed using the TCGA database. Compared with control samples, RBM8A was significantly higher in GBM, suggesting that RBM8A may play an important role in GBM (Figure 1A). Even more interesting is that GBM patients with high expression of RBM8A in TCGA had poor overall survival, based on the optimal gene expression grouping threshold (Figure 1B). RBM8A expression in GBM specimens from the 94 patients was examined using immunostaining (Figure 1C). The mean age of the 94 included patients at diagnosis was 49.38 ± 15.87 years (range, 13-85 years). RBM8A localized primarily to the nucleus and we found that a large proportion of patients with low-grade tumors (78.57%) had low RBM8A expression, while those with high-grade tumors (66.25%) had high RBM8A expression (Table 1). Follow-up data lasting a mean of 14.74 ± 13.13 months (range, 0.03-59 months) were available for 76 patients, who were therefore included in the survival analyses.

We compared overall survival between the patients showing relatively low and high expression levels. Among the 76 patients with follow-up data, those with low nuclear levels of RBM8A expression in tumor tissues (n = 26) exhibited significantly longer overall survival than those with high nuclear RBM8A expression (n = 50; Figure 1D). The median survival time was 11.0 months (95%CI 6.00-15.99) among patients with low RBM8A expression but 7.0 months (95%CI 5.46-8.54) among those with high RBM8A expression (Table 2). The log-rank test revealed that patients with high RBM8A expression had significantly shorter overall survival time (χ2 = 4.884, P = 0.27; Table 2).

To understand the function of RBM8A in GBM progression, we examined the expression of RBM8A in GBM cell lines using western blotting. RBM8A expression was higher in U87-MG and U251-MG cells (Supplementary Figure S1). After effectively knocking down RBM8A in the U87-MG and U251-MG cells (Figure 2A), we assessed the function of RBM8A in GBM cell proliferation and invasion. CCK8 and EdU assays indicated that the suppression of RBM8A reduced the ability of GBM cells to proliferate (Figures 2B, C). Transwell assays demonstrated that RBM8A knockdown significantly suppressed migration and invasion by GBM cells (Figures 2D, E). These findings suggest that RBM8A expression plays a crucial role in promoting the proliferation and invasive potential of GBM cells.

To verify and extend our results obtained with RBM8A knockdown, we overexpressed RBM8A in A172 and T98G cells (Figure 3A). Overexpression enhanced GBM cell proliferation (Figures 3B, C), as well as migration and invasion (Figures 3D, E). These findings provide evidence that elevated RBM8A expression promotes GBM cell tumorigenesis and invasion.

To further explore the effect of high expression of RBM8A on GBM, we performed differential expression analysis between the GBM and control samples, as well as between groups expressing high or low RBM8A levels. Genes consistent with high RBM8A expression were identified as differentially expressed genes (DEGs) (Figure 4A). DEGs may be affected by the high expression of RBM8A and may be involved in the ability of RBM8A to promote tumor growth and invasion. These DEGs were involved mainly in the cell cycle, Notch signaling pathway and Hippo signaling pathway (Figure 4B).

Based on the STRING database, we identified the pathway genes interacting with Notch1 and STAT3. RBM8A may regulate Notch1 and STAT3, and then regulate pathway genes to affect the occurrence and development of GBM (Figure 4C). Molecular docking led to negative docking energies (Figure 4D), suggesting that RBM8A can bind the genes whose transcription is regulated by C promoter-binding factor 1 (CBF1).

We verified these bioinformatics analyses by Western blotting for levels of Notch, phospho-STAT3, and phospho-H3 in cells in which RBM8A was knocked down. We observed decreased levels of all three proteins (Figure 4E), while RBM8A overexpression had the opposite effects (Figure 4F). CBF1 interacts with the Notch1 receptor to activate the Notch signaling pathway (42). Consistent with this, we found that overexpression of RBM8A showed significantly stronger activation of the CBF1 luciferase reporter in T98G and A172 cells than in the controls. In contrast, knocking down RBM8A in U87 and U251 cells weakened activation of the CBF1 luciferase reporter (Figure 4G). These results suggest that RBM8A is sufficient to activate the Notch signaling pathway. Therefore, we propose that RBM8A promotes the proliferation and migration of GBM cells by activating the Notch/STAT3 pathway (Figure 4H).

We integrated GBM data from TCGA and CGGA. Then, we extracted genes involved in a mechanism through which RBM8A activated the Notch/STAT3 pathway to promote proliferation and migration of GBM cells. We calculated a GSVA score to explore whether these genes might have diagnostic or prognostic potential in GBM. Survival analysis showed that patients with high GSVA scores had poor overall survival (Supplementary Figure S2A). In addition, the GSVA score showed good diagnostic accuracy (Supplementary Figure S2B).

Suppression of the Notch pathway in A172 or T98G cells using the γ-secretase inhibitor DAPT significantly reversed RBM8A-enhanced proliferation and invasion (Figures 5A–C), and this reversal was associated with down-regulation of p-STAT3 and Notch1 protein (Figure 5D). These results provide support for the idea that the Notch/STAT3 pathway mediates the pro-oncogenic function of RBM8A in glioblastoma cells.

To examine how these effects of RBM8A on GBM cells in vitro may translate to clinical phenotypes, we injected U87-MG cells stably expressing RBM8A-KD1 shRNA or no shRNA intracranially into female nude mice and monitored GBM progression. By six weeks, mice inoculated with NC cells developed larger tumors than those inoculated with RBM8A-KD1 cells (Figures 6A–C). Interestingly, RBM8A knockdown also dramatically reduced the levels of Notch1 protein in tissues (Figure 6D). Tumor growth was much slower in the RBM8A-KD group than in the control group, with average tumor diameter 0.25 ± 0.15 mm³ in RBM8A-KD animals and 11.92 ± 4.98 mm³ in NC animals (p=001; n=5 per group; Supplementary Figure S3). These results are consistent with the in vitro evidence that RBM8A contributes to GBM progression.