SRSF3-Regulated RNA Alternative Splicing Promotes Glioblastoma Tumorigenicity by Affecting Multiple Cellular Processes

Abstract

Mis-regulated alternative RNA splicing (AS) contributes to the tumorigenesis and progression of human cancers, including glioblastoma (GBM). Here, we showed that a major splicing factor, serine and arginine rich splicing factor 3 (SRSF3) was frequently upregulated in clinical glioma specimens and that elevated SRSF3 was associated with tumor progression and a poor prognosis for glioma patients. In patient-derived glioma stem-like cells (GSCs), SRSF3 expression promoted cell proliferation, self-renewal, and tumorigenesis. Transcriptomic profiling identified more than 1000 SRSF3-affected AS events, with a preference for exon skipping in genes involved with cell mitosis. Motif analysis identified the sequence of CA(G/C/A)CC(C/A) as a potential exonic splicing enhancer for these SRSF3-regulated exons. To evaluate the biological impact of SRSF3-affected AS events, four candidates were selected whose AS correlated with SRSF3 expression in glioma tissues, and their splicing pattern was modified using a CRISPR/Cas9 approach. Two functionally validated AS candidates were further investigated for the mechanisms underlying their isoform-specific functions. Specifically, following knockout of SRSF3, transcription factor ETS variant 1 (ETV1) gene showed exon skipping at exon 7, while nudE neurodevelopment protein 1 (NDE1) gene showed replacement of terminal exon 9 with a mutually exclusive exon 9’. SRSF3-regulated AS of these two genes markedly increased their oncogenic activity in GSCs. Taken together, our data demonstrate that SRSF3 is a key regulator of AS in GBM and that understanding mechanisms of mis-regulated AS could provide critical insights for developing effective therapeutic strategies against GBMs.

Introduction

Glioblastoma (GBM, grade IV glioma) is the most malignant and common primary brain tumor in adults, and is characterized by rapid growth, dissemination through normal brain, and therapy resistance, all of which contribute to a poor prognosis (1). To develop effective therapeutic strategies against this deadly cancer, large-scale genetic studies, including The Cancer Genome Atlas (TCGA) Network, have identified key alterations in GBM and lower-grade gliomas (LGGs) through analyses of genomic, epigenomic, transcriptomic, and proteomic datasets. As a result, biologically discrete subgroups of gliomas have been defined based on genetic mutations, DNA methylation, and gene expression profiles (2). The multi-dimensional data has also revealed a significant level of intra-tumoral heterogeneity in GBM. These findings, however, have not yet been translated into effective therapeutic treatments for GBM.

Alternative RNA splicing (AS) is an evolutionarily conserved post-transcriptional process that increases RNA and protein diversity in eukaryotes (3). More than 90% of human multi-exons genes undergo alternative splicing (4). The predicted outcomes of AS include 1) production of different protein-coding and non-coding transcripts, 2) introduction of premature stop codons, which results in nonsense-mediated RNA decay (NMD), and 3) variability in mRNA untranslated regions (UTRs) which influences mRNA stability, localization, and translation efficiency (3). In addition to contributing to cell lineage and tissue identity that occur during normal development, AS is involved in numerous pathological processes, including cancer (5). Comprehensive transcriptomic analyses across cancer types have revealed widespread AS alterations in tumors, as compared to their normal tissue counterparts (6,7). Tumor-associated AS can generate cancer-specific splicing isoforms with activities that affect cell proliferation, apoptosis, DNA damage repair, invasion, angiogenesis, metabolism, and therapy response (5,8). Thus, targeting oncogenic AS might provide therapeutic benefit in treating cancer (9).

RNA splicing is carried out by a large ribonucleoprotein complex called the “spliceosome”, which assembles at intron-exon junctions known as “splice sites” and catalyzes the reaction of intron removal (10). The efficient spliceosome recruitment to alternative splice sites is known to involve the binding of auxiliary trans-acting factors to the cis-regulatory elements in proximity to a splice site, likely due to a relatively weaker spliceosome affinity for the alternative splice sites than the constitutive ones (11). Two major families of auxiliary splice-regulating factors are the serine- and arginine-rich proteins (SR proteins), and the heterogeneous nuclear ribonucleoproteins (hnRNPs), that promote and inhibit splicing, respectively (12,13). SR and hnRNP proteins function through association with the short sequence motifs adjacent to splice site that act as exonic/intronic splicing enhancers (ESEs/ISEs) or exonic/intronic splicing silencers (ESSs/ISSs), thereby stimulating or repressing spliceosome assembly (12,13).

The SR protein family consists of 12 members, characterized by the presence of one or two RNA recognition motifs (RRMs) and a serine/arginine-rich domain (12). Genome-wide analysis through crosslinking immunoprecipitation sequencing (CLIP-seq) identified distinct and overlapping binding sites for different SR proteins (14,15), indicating their non-redundant and/or cooperative function. In general, SR proteins bind ESE and enhance the splicing by recruiting the spliceosome. However, some members can act as splicing repressors in a context-dependent manner (15,16). But the mechanisms under which SR proteins exhibit complex functions in splicing has remained incompletely understood. Aberrant expression and/or activation, as well as somatic mutation in SR proteins, have been implicated in the development of various types of cancers (5,17). In glioma, SR splicing factor 1 (SRSF1) is upregulated and promotes cell proliferation and invasion by causing the inclusion of exon 23 and 24 in MYO1B pre-mRNA, generating an oncogenic and membrane-localized isoform (18). Other SR proteins have yet to be investigated in gliomas. In this study, we have begun to address this knowledge gap by examining the expression level and prognostic relevance of SR proteins in clinical glioma samples, which has resulted in the identification of SRSF3 as an important contributor to the molecular and tumor biology of GBM, and to the clinical outcome of GBM patients.

Materials and Methods

Cell culture

Human HEK293T cells (ATCC) and glioma U87 cells (ATCC) were cultured in DMEM (Thermo Fisher Scientific, 11995–065) supplemented with 10% FBS (Thermo Fisher Scientific, 10437028) and 1% penicillin and streptomycin (Thermo Fisher Scientific, 15140122). Normal human astrocytes (NHA, Clonetics, CC-2565) and immortalized NHA-ET (NHAs introduced with human papillomavirus 16 E6/E7 and human TERT) (19) were cultured in the Astrocyte Basal Medium (Lonza, CC-3187) supplemented with Astrocyte Growth Medium BulletKit (Lonza, CC-4123). Patient-derived GSCs, GSC83, GSC528 and GSC23, were previously characterized (20,21). GSC1485 were derived from a GBM sample as previously described (21) with surgical resections at Northwestern Memorial Hospital through the Northwestern Nervous System Tumor Bank (NSTB). All GSCs were cultured in DMEM/F12 medium (Thermo Fisher Scientific, 11320–033) containing 2% B27 supplement (Thermo Fisher Scientific, 17504–044), 1x antibiotic-antimycotic (Thermo Fisher Scientific, 15240062), 5 μg/ml heparin (Sigma-Aldrich, 9041–08-1), 20 ng/ml EGF (Peprotech, 100–15R), and 20 ng/ml bFGF (Peprotech, 100–18B). All the glioma cells and GSCs were authenticated by short tandem repeat analysis at IDEXX BioAnalytics in Columbia, MO, USA or Texas Tech University Health Sciences Center in Lubbock, TX, USA, and corresponding institutions described in the references cited above. All cell lines were tested negative for Mycoplasma using VenorGeM Mycoplasma Detection Kit (Sigma-Aldrich, MP0025). The latest authentication and Mycoplasma testing were in July 2019. All cell lines were cultured less than 20 passages prior to use.

Ethics and clinical glioma tumor specimens

Human Subjects Research protocols were approved by the Institutional Review Board (IRB) at Northwestern University in accordance with guidelines by Declaration of Helsinki, NIH, and institutional Ethics Committee. Fresh and snap-frozen tissue fragments were obtained from surgical samples of glioma patients with surgical resections at Northwestern Memorial Hospital through the Northwestern Nervous System Tumor Bank (NSTB). All patients provided informed written consent. In addition, normal brain tissues were obtained from the NeuroBioBank at NIH (https://neurobiobank.nih.gov/).

Animal Studies

Athymic mice (Ncr nu/nu) at 6–8 weeks of age were purchased from Taconic Farms. All experiments using animals were conducted under the Institutional Animal Care and Use Committee (IACUC)-approved protocols at Northwestern University in accordance with NIH and institutional guidelines.

Glioma stem-like cell (GSC) suspension (1 × 10⁴ cells for GSC83, and 5 ×10⁵ cells for GSC528) was intracranially injected into the brain of individual mice (five-six mice per group) as described previously (22). Bioluminescence imaging (BLI) was conducted to monitor in vivo tumor growth using the IVIS Lumina imaging station (Caliper Life Sciences). Tumor-bearing mice were injected with 300 mg/kg of D-luciferin (Gold Biotechnology) before isoflurane anesthesia. Radiance (photons per second per square centimeter per steradian, p/sec/cm²/sr) was measured 12 minutes after substrate injection using Living Image 4.3.1 software (Caliper Life Sciences). For the survival analysis, mice were maintained until pathological symptoms developed resulting from tumor burden and combined with BLI signal intensity indicative of tumor burden.

CRISPR/Cas9-mediated SRSF3 knockout, exon skipping, and CRISPR/Cas13d-mediated ETV1 knockdown

To knock out (KO) SRSF3, the target sequences of gRNA were designed using the SYNTHEGO online tool (https://design.synthego.com). The synthesized forward and reverse primers including a 20-bp target sequence and a BsmBI sticky end were annealed and inserted into a lentiCRISPRv2GFP vector digested with BsmBI.

To induce skipping of targeted exons, gRNAs were designed at the 5’ or 3’-splice sites of targeted exons, based on two criteria: 1) Cas9-mediated cleavage site (∼3–4 nt upstream of the PAM sequence) less than 5 nt away from the splice site; 2) minimal off-target prediction by the SYNTHEGO online tool.

To knockdown (KD) ETV1 using CRISPR/Cas13, the target sequence of gRNA was designed using InvivoGen’s siRNA Wizard online tools (www.invivogen.com/sirnawizard). The sequence of ETV1 exon 7 was used as an input. The synthesized forward and reverse primers including a 22-bp target sequence and a BsmBI sticky end were annealed and inserted into a lenti-Cas13d-gRNA-GFP vector. All the target sequences of gRNAs are listed in Supplementary Table S1.

Lentivirus production and infection of GSCs was performed as previously described (22,23). Briefly, lentiviral vectors expressing gRNAs for SRSF3-KO, exon skipping, or ETV1-KD were transfected into HEK 293T cells using Lipofectamine 2000. After incubation for 48–72 hours, the supernatants containing lentivirus were harvested and used to infect GSCs for 6–8 hrs. After five days, various GSC cells were harvested to determine the KO, exon skipping, or knockdown efficiency by using Sanger sequence, immunoblot (IB) or RT-PCR assays. The Sanger sequence results were analyzed using the SYNTHEGO online tool. Heterogenous cell populations were used for experiments rather than single clone to mitigate clonal variations.

RNA Isolation and quantitative (q)RT-PCR

Total RNA was isolated using a Qiagen RNeasy Mini kit and was reverse-transcribed using the iScript cDNA Synthesis Kit (BioRad) according to the manufacturer’s instructions. qRT-PCR was performed using the Power SYBR Green Master Mix (Life Technologies) on an Applied Biosystems StepOne Plus Real-Time Thermal Cycling Block with three replicates per group. Relative gene expression was determined by normalizing the expression of each target gene to ACTB. Results were analyzed using the 2^-(∆∆Ct) method. To validate isoform switches by RT-PCR, primers were designed flanking the alternative spliced exons or introns. PCR products were separated by 1% agarose gel, and the DNA product was quantified using the Image Studio software (LI-COR Biosciences). The ratio of each isoform was normalized to the sum of the different isoforms. PCR primers are listed in Supplementary Table S1.

Immunoblotting (IB)

Cells were lysed in RIPA buffer (50 mM Tris-HCl, 150 mM NaCl, 2 mM EDTA, 1% NP-40, pH 7.4) containing 1x protease inhibitor cocktail and 1x phosphatase inhibitor cocktail (Sigma-Aldrich & Roche). To extract protein from tumor and normal brain tissues, the flow-through from the RNeasy spin column during RNA isolation was preserved and the proteins were precipitated using cold acetone according to the manufacturer’s instructions. Protein samples were separated by SDS-PAGE and then transferred onto PVDF membranes. After blocking with 5% non-fat milk in TBS-T for 1 hr, membranes were incubated with indicated antibodies at desired dilutions overnight at 4°C. Following washing with TBS-T, the blot was incubated with corresponding HRP-conjugated secondary antibodies (DAKO, anti-rabbit immunoglobulins, P0217; anti-mouse immunoglobulins, P0260). Blots were developed with enhanced chemiluminescence (ECL, Amersham Bioscience) reaction according to manufacturer’s instructions. Signal was quantified using the Li-Cor Image Studio software (Version 5.2.5).

The antibodies used for IB in this study are as follows: anti-SRSF3 (Bioworld, BS2559); anti-α/β-tubulin (Cell Signaling Technology, 2148); anti-GAPDH (Santa Cruz, sc-47724); anti-Flag (Sigma-Aldrich, F1804); anti-ETV1 (Sigma-Aldrich, SAB2104467); anti-phosphoserine/threonine (BD Biosciences, 612549); anti-GFP (Molecular Probes, A-11122); anti-PARP (Cell Signaling Technology, 9542); anti-cleaved Caspase-3 (Cell Signaling Technology, 9661); anti-Caspase-3 (Enzo Life Sciences, ADI-AAP-113).

Immunofluorescence (IF)

For α-tubulin staining, GSC83 cells were plated on Lab-Tek chamber slide (Thermo Fisher Scientific) with 10% growth factor-reduced Matrigel Matrix (BD Biosciences) and incubated at 37 °C overnight. Cells were fixed with 4% paraformaldehyde (Thermo Fisher Scientific), permeabilized with 0.1% Triton X-100 (Sigma-Aldrich) in PBS, blocked with 1% bovine serum albumin (Sigma-Aldrich) in PBS and incubated with an anti-α-tubulin antibody (Sigma-Aldrich, T9026, 1:500 dilution) at 4°C overnight. Cells were then washed thrice with PBS and incubated with a goat anti-mouse Alexa 594 antibody (Thermo Fisher Scientific, A11005, 1:500 dilution) at room temperature for 1 hr. DNA was stained with DAPI (Vector Laboratories) for 10 min. Coverslips were mounted with 50% glycerol in PBS and images were acquired with a Nikon A1R confocal microscope and processed with NIS-Elements software and Adobe Photoshop CC 2019. 15–25 mitotic cells per group were randomly selected to evaluate spindle formation and chromosome congression.

In vitro cell proliferation assays

Cell proliferation analyses were performed as previously described (22,23). Briefly, dissociated GSCs were seeded in 96-well plates at density of 1000 cells per well, 3–4 replicates per group. Cell viability was measured using CellTiter-Glo Luminescent Cell Viability Assay Kit (Promega) at day 0, 2, 4, 6 according to the manufacturer’s instructions. Luminescence was measured using SpectraMax M3 Multi-Mode Microplate Reader (Molecular Devices) and normalized to their levels at day zero.

In vitro limiting dilution assays

GSCs were dissociated into single-cell suspensions and seeded into 96-well plates at density of 1, 2, 4, 8, 16, or 32 cells per well. Cells were incubated at 37 °C for 1 to 2 weeks. At the time of quantification, each well was examined for formation of tumor spheres. Stem cell frequency was calculated using Extreme Limiting Dilution Analysis online tool (http://bioinf.wehi.edu.au/software/elda/).

Statistical Analysis

Statistical analyses were carried out using GraphPad Prism version 7. All grouped data are presented as mean ± SD. For comparing two groups, the two-tailed Student’s t-tests were used unless otherwise stated. The Mann-Whitney test was performed to determine whether SR proteins were differentially expressed among normal brain, LGG, and GBM samples. For survival analysis, groups were divided based on the cutoff points determined by the Cutoff Finder online tool (http://molpath.charite.de/cutoff/index.jsp) . Kaplan-Meier analysis was performed and then compared using a log-rank test.

Description of the experimental procedures related to plasmids, reagents, Ki-67 staining, TUNEL assay, RNA-seq and bioinformatics analyses are detailed in the Supplementary Materials and Methods section.

Results

Elevated expression of SRSF3 is associated with progression and poor prognosis in human gliomas

We evaluated SR family (SRSF1–12) gene expression by RNA sequencing (RNA-seq) in 85 GBMs and 18 LGGs obtained from Northwestern NSTB, as well as in 15 normal brain samples from the NIH NeuroBioBank. As shown in Figure 1A, SRSF2, 3, 7, 9, 10, and 11 showed significantly higher levels of expression in GBMs when compared with LGG and normal brain tissues. Among these upregulated SRSFs, SRSF2, 3, and 7 expression levels were most strongly associated with patient outcome (Fig. 1B and Supplementary Fig. S1A). Since SRSF2 has been studied extensively in relation to its oncogenic activity as well as downstream AS targets (15,24). Therefore, we selected SRSF3 and SRSF7 to study their roles in glioma tumorigenicity. Moreover, we decided to focus subsequent molecular and tumor biology studies on SRSF3, which has not been previously investigated in glioma and not include our investigations of SRSF7 in this report.

Figure 1.

SRSF3 is upregulated in gliomas and associates with glioma tumor progression and prognosis. A, Heatmap of SR family gene expression in normal brain, LGG, and GBM specimens and statistical analysis of SR gene expression related to tumor grade, IDH1 mutation, and overall survival (OS) of GBM patients based on our RNA-seq data. B, Kaplan-Meier survival analysis for SRSF3 expression in patients with GBM from NU datasets. C, IB analysis for SRSF3 in representative normal brain, LGG and GBM specimens of NU glioma cohort. α/β-tubulin was a loading control. Data are representative of two independent experiments with similar results. D, Quantification of IB data for SRSF3 expression in NU datasets. E and F, Expression of SRSF3 in normal brains and gliomas from TCGA (E) and CGGA datasets (F). G and H, Kaplan-Meier survival analyses for SRSF3 expression. G, TCGA data and H, CGGA data. *, p < 0.05; **, p < 0.01; ***, p < 0.001; ns, no significance.

We confirmed the elevated SRSF3 expression revealed from RNA-seq analysis by IB analysis for SRSF3 protein in tissue samples from the glioma cohort indicated above. We found that SRSF3 protein expression was significantly higher in glioma tumor tissues compared to normal brain tissues and correlated with WHO tumor grades (Fig. 1C and 1D). Of note, in normal brain tissues, some samples showed a lower SRSF3 protein band in SDS-PAGE gels compared to glioma specimens and GSC83 cells, similar with alkaline phosphatase-treatment samples, indicating hypo-phosphorylation of SRSF3 in some regions of normal brain (Supplementary Fig. S1B) (12). In addition, we detect SRSF3 expression in four GSC lines (GSC83, 528, 23, and 1485), U87 glioma cell line and normal human astrocytes (NHAs). The SRSF3 protein levels are slightly higher in glioma cells than normal astrocytes (Supplementary Fig. S1C).

We queried public TCGA and Chinese Glioma Genome Atlas (CGGA) RNA-seq datasets of gliomas and found that elevated expression of SRSF3 associates with tumor progression and poor prognosis in both databases (Fig. 1E to 1H). After adjusting for clinical variables that are commonly used in GBM diagnosis (including status of IDH1, TP53 mutations, patients’ age and gender), the relationship of GBM prognosis with elevated SRSF3 expression remained significant except in the group of “age < 50” from the CGGA database (Supplementary Fig. S1D and S1E).

SRSF3 expression is important for cell growth, self-renewal, and tumorigenicity of GSCs

To investigate tumor biologic properties affected by SRSF3 expression, we knocked out (KO) SRSF3 in GSC83 and 528 cells through a lentiviral CRISPR/Cas9 system and used the resulted heterogenous cell populations rather than single clone to mitigate clonal variations. The successful KO was confirmed by both the Sanger sequencing and immunoblotting (IB) analysis (Supplementary Fig. S2 and Fig. 2A). In both GSC lines, SRSF3-KO markedly inhibited GSC cell viability (Fig. 2A) and self-renewal (Fig. 2B) in vitro, and suppressed intracranial tumor growth in athymic nude mice, thereby significantly extending animal subject survival (Fig. 2C and 2D). This significant suppression in cell viability upon SRSF3-KO was caused by reduced cell proliferation (Supplementary Fig. S3A) and increased cell apoptosis (Supplementary Fig. S3B and S3C). Next, we expressed a CRISPR/Cas9-resistant SRSF3 (with a synonymous mutation that disrupts the protospacer adjacent motif, which is necessary for CRISPR/Cas9-mediated DNA cleavage) in SRSF3-KO GSC cells (Fig. 2E). Expression of the exogenous SRSF3 proteins restored SRSF3-KO-affected GSC growth, sphere-forming frequency, tumor growth in the brains, and overall survival of engrafted animals (Fig. 2F to 2H).

Figure 2.

SRSF3 promotes proliferation, self-renewal and in vivo tumorigenicity of GSCs. A and B, Effect of SRSF3-KO in GSC83 and GSC528 (upper panel in A, IB) on cell proliferation (lower panel in A) and sphere-forming frequency (B). C and D, Effect of SRSF3-KO on brain tumor xenograft growth of GSC83 (C) and GSC528 (D) (left, representative BLI images; middle, quantification of BLI), and extended survival of tumor-bearing mice (right, Kaplan Meir analysis). E, IB analysis for expression of endogenous SRSF3 and exogenous Flag-tagged SRSF3 in GSC83 cells with indicated modifications. F to H, Effect of SRSF3 re-expression in SRSF3-KO GSC83 cells on proliferation (F), sphere-forming frequency (G), brain tumor xenograft growth (H) (left, BLI images; middle, quantification of BLI), and survival of tumor-bearing mice (right, Kaplan Meir analysis).

Data are representative of two to three independent experiments with similar results. *, p < 0.05; **, p < 0.01; ***, p < 0.001.

To investigate whether the role of SRSF3 in glioma tumorigenesis was a common phenomenon, we knocked out SRSF3 in two additional GSC lines, GSC23 and 1485, U87 glioma cells, normal human astrocyte NHA and immortalized NHA-ET (NHAs introduced with human papillomavirus 16 E6/E7 and human TERT) (19). KO of SRSF3 markedly suppressed cell growth of GSCs, U87, and immortalized NHA-ET but had modest effect on NHA cell viability in vitro (Supplementary Fig. S3D to S3G). Interestingly, NHA-ET cells expressed SRSF3 proteins at a much higher level when compared with that in NHA cells (Supplementary Fig. S3G), suggesting that the immortalization of NHA cells upregulated the expression of SRSF3.

Global landscape of SRSF3-affected AS and gene expression in GSCs

To study the mechanism of SRSF3 involved in glioma tumorigenicity, we performed high throughput RNA-seq of WT and KO GSC83 and 528 cells. The MISO pipeline (25) was used to calculate the PSI value (Percentage Spliced In, which represents the mRNA percentage of one indicated isoform) for each AS event. With a cutoff of ∆PSI at 0.2, we identified 2752 and 2548 SRSF3-regulated AS events in GSC83 and 528 cells, respectively: 1143 events were shared by the two cell lines (Fig. 3A and Supplementary Table S2). The overlapping SRSF3-regulated splicing targets were associated with processes important to the biology of GBM including mitotic nuclear division, intracellular receptor signaling, regulation of apoptotic signaling, and cell cycling (Fig. 3B). Among different types of AS events including skipped exon (SE), retained intron (RI), mutually exclusive exon (MXE), and alternative 5’/3’ splice site (A5SS/A3SS), we observed that KO cells preferentially induced exon skipping compared to other AS events (Fig. 3C), suggesting a regulatory role of SRSF3 as a splicing activator for cassette exons. To verify the accuracy of the RNA-seq profiles, we selected 10 SRSF3-affected AS events for validation by RT-PCR. All of the selected AS events were confirmed with appreciable correlations between RNA-seq derived ∆PSI and RT-PCR-quantified ∆PSI (Fig. 3D). Representative examples of five validated AS events are shown in Figure 3E.

Figure 3.

Global landscape of SRSF3-affected alternative splicing (AS) in GSCs. A, Quantification of SRSF3-regulated AS events in each category: SE, skipped exons; RI, retained introns; MXE, mutually exclusive exons; A3SS/A5SS, alternative 3’/5’ splice sites. B, Top 10 significantly enriched GO annotations of genes presenting splicing alterations upon SRSF3-KO. C, Changes in PSI values of SRSF3-regulated AS events in each category in indicated GSCs. D, Correlation between RNA Seq-derived ∆PSI and RT-PCR quantified ∆PSI. E, RT-PCR for SRSF3-regulated AS events in five representative genes in GSC83 and 528 cells. ACTB was used as a control. Data are representative of two independent experiments with similar results. F, Flowchart of the SRSF3-motif discovery and functional prediction for SRSF3-dependent exons. G, Isoform-specific function prediction for SRSF3-dependent exons based on ASpedia database.

To determine the cis-regulatory sequence for SRSF3-associated exon inclusion, we selected 107 SRSF3-dependent alternative exons for which the Percent Spliced In (PSI) values dropped more than 50% after SRSF3-KO in both GSCs. The regions adjacent to the splice sites of these 107 exons were used to perform de novo discovery of SRSF3 binding motif using the DREME algorithm (26). A discriminative cis-regulatory sequence motif “CA(G/C/A)CC(C/A)” was enriched in the exon region adjacent (<200 bp) to the splice site (Fig. 3F), which is consistent with the A/C-rich SRSF3 binding motifs described by others (16,27). We examined the peptide sequences encoded by the 107 exons using the ASpedia database (28). Eighty-one of them were annotated in this database and encode amino acids that either harbor post-translational modification sites, protein domains, protein interaction sites, or are involved in nonsense-mediated mRNA decay (Fig. 3G). These results indicate that SRSF3-regulated AS is important for the inclusion of functionally important peptide sequences in numerous proteins.

Apart from splicing regulation, we also observed strong effects of SRSF3 on global gene expression. The expression level of 2,591 protein-coding genes were altered after SRSF3-KO (absolute log₂ fold-change > 1 in GSC83 and 528 cells), with 1,684 genes down-regulated and 907 genes up-regulated (Supplementary Table S3). Among the down-regulated genes, biological processes related to cell division, chromosome segregation, DNA replication, and Wnt signaling were markedly enriched (Supplementary Fig. S4A). Up-regulated genes were enriched for processes related to transport, autophagy, and response to endoplasmic reticulum stress (Supplementary Fig. S4B). These significant transcriptomic alterations could either be indirect effects following SRSF3-KO induced proliferation suppression and apoptosis induction or directly regulated by SRSF3 on targeted transcripts.

SRSF3-regulated AS events are involved in glioma tumorigenesis

To determine which SRSF3-regulated AS events are functionally involved in GBM tumorigenesis, we performed an AS profile analysis using both TCGA and CGGA glioma datasets. Fifty-eight SRSF3-affected AS events presented significant correlation between their PSI value and SRSF3 expression (absolute value of Pearson’s correlation coefficients > 0.2) in TCGA and CGGA datasets (Fig. 4A and Supplementary Table S4), indicating their potential involvement in SRSF3-regulated glioma tumorigenesis. To select candidates for further functional study, we ranked these 58 events according to the ∆PSI values (between GSC83-SRSF3-WT and -KO cells) and chose candidates from the top 20 most significant SRSF3-affected events that are SRSF3-promoted alternative exons and also harbor the predicted exonic splicing enhancer “CA(G/C/A)CC(C/A)”. Among the 9 satisfactory candidates, MAP4 is a large protein (1152 amino acids) and has multiple transcript variants, which increase the difficulty of functional study, such as vector construction for gene overexpression. IDH3B and its altered exon are adjacent to another gene NOP56. Therefore, it is complicated to determine the splicing pattern and CRISPR-mediated splicing manipulation might have influence on NOP56 expression. For WRAP53, upon SRSF3 KO, the skipping of exon 8 is accompanied by the inclusion of intron 7, making CRISPR-mediated splicing manipulation difficult to accomplish. For CDCA3 and UNK, in addition to the splicing pattern change, they also showed significant reduction in mRNA expression upon SRSF3-KO in our RNA seq data analysis (more than 6 folds for CDCA3 and more than 3 folds for UNK in GSC83 cells). Therefore, we finally selected the remaining four candidates: cassette exon 7 of transcription factor ETS variant 1, ETV1 (ETV1-E7); mutually exclusive exon 9 of nudE neurodevelopment protein 1, NDE1 (NDE1-E9 & NDE1-E9’); cassette exon 13 of pumilio RNA binding family member 2; PUM2 (PUM2-E13); and cassette exon 6 of survival motor neuron, SMN (SMN-E6). RT-PCR results confirmed that SRSF3-KO caused exon skipping of ETV1-E7, PUM2-E13 and SMN-E6, as well as the substitution of NDE1-E9 with NDE1-E9’ (Fig. 4B). Expression of exogenous SRSF3 rescued the SRSF3 KO-induced splicing alterations in these four candidate genes (Fig. 4B). In addition, analyses of SRSF3 expression in relation to PSI values of these four candidate genes on RNA seq data of TCGA, CGGA, and our own data revealed positive correlations between SRSF3 levels and PSI of these four genes (Supplementary Fig. S5).

Figure 4.

Characterization of SRSF3-regulated AS events involved in glioma tumorigenesis. A, Venn diagram of SRSF3-affected AS events identified from SRSF3-KO RNA-seq data and SRSF3-correlated AS events in both TCGA and CGGA glioma datasets. B, RT-PCR validation of four identified SRSF3-regulated AS events. C-F, CRISPR-mediated exon skipping in four candidate genes and their effects on GSC proliferation. P-F/P-R: forward/reverse primers used in RT-PCR; Mut: CRISPR-mediated exon skipping; E: exon. G and H, Effects of exon skipping in ETV1 (G) and NDE1 (H) on sphere-forming frequency of GSCs. I, Exon skipping in ETV1 and NDE1 reduced the growth of GSC83 brain tumor xenografts of nude mice (left, representative BLI images; middle, quantification of BLI) and extended animal survival (right, Kaplan-Meier analysis).

Data are representative of two to three independent experiments with similar results. *, p < 0.05; **, p < 0.01; ***, p < 0.001; ns, no significance.

Next, we used CRISPR/Cas9 technology to induce indel mutations at the splice sites of targeted exons, resulting in exon skipping of these four candidates. We successfully induced exon skipping of ETV1-E7, PUM2-E13, and SMN-E6 in GSC83 cells (Fig. 4C to 4E). For NDE1, the splicing disruption of E9 by CRISPR/Cas9 did not lead to the inclusion of the alternative E9’ (Fig. 4F), most likely indicating that E9-excluded NDE1 mRNA underwent AS-mediated RNA decay. Among these four candidates, skipping of ETV1-E7 and NDE1-E9 markedly impaired proliferation and sphere formation ability of GSC83 and 528 cells in vitro (Fig. 4C to 4H, Supplementary Fig. S6A and S6B), suppressed GSC83 tumorigenicity in the brains of mice, and extended the overall survival of tumor-bearing animal subjects (Fig. 4I). Additionally, we used CRISPR/Cas13-mediated knockdown (KD) (29) to induce the degradation of ETV1-E7-containing transcripts, which inhibited the proliferation and sphere formation ability of GSC83 cells (Supplementary Fig. S6C). Together, these data support SRSF3-mediated ETV1-E7 and NDE1-E9 inclusions as contributing to the tumor biology of GBM.

Phosphorylation of ETV1-E7-encoded peptide regulates the oncogenic activity of ETV1

ETV1 is an oncogenic member of the E twenty-six (ETS) family of transcription factors that modulate essential biological processes, including cell proliferation, differentiation, migration, and angiogenesis (30). Of note, four serine/threonine (S/T) residues that are encoded by ETV1-E7 (T139, T143, S146) and its upstream sequence (S94) (Fig. 5A) are phosphorylated by MAPK (30). Inhibition of MAPK signaling dephosphorylated these four S/T residues, leading to ubiquitin-mediated degradation of ETV1 protein, suggesting that MAPK-induced phosphorylation at ETV1-E7 stabilizes ETV1 protein (31). Based on this information, we tested the hypothesis that E7 skipping of ETV1 transcript would lead to ETV1 protein degradation, thus diminishing ETV1-mediated oncogenic transcriptional program. Consistent with this hypothesis, we found that the S/T phosphorylation level of ETV1 proteins was significantly decreased in E7-absent ETV1 (ETV1-∆E7) proteins, compared with the full-length ETV1 (ETV1-FL) (Fig. 5B).

Figure 5.

Inclusion of exon 7 promotes oncogenic activity of ETV1. A, ETV1-E7-coded region harbors three amino acid residues for MAPK phosphorylation. B, Immunoprecipitation (IP)-IB analyses for exogenous expression of Flag-ETV1 full-length (FL) or -E7-excluded isoform (∆E7). C, RNA-seq analysis. ETV1-target genes were down-regulated upon SRSF3-KO. D, qRT-PCR for ETV1-target genes in GSC83 with SRSF3-KO, ETV1-E7 skipping (ETV1-Mut) or control. E and F, qRT-PCR of ETV1-target genes in GSC83 cells with ETV1-KD or control (E) and in GSC528 cells with ETV1-E7 skipping or control (F). G, RT-PCR (upper panels) and IB (lower panels) analyses of endogenous and exogenous ETV1 expression in GSC83 cells with indicated modifications. H, qRT-PCR of ETV1-target genes in GSC83 cells with indicated modifications. I, Cell proliferation of GSC83 cells with indicated modifications.

Data are representative of two to three independent experiments with similar results. *, p < 0.05; **, p < 0.01; ***, p < 0.001.

The transcriptional factor ETV1 modulates the transcription of several cancer-related genes (30,32). From the RNA-seq data of SRSF3-KO/WT GSCs, we identified six ETV1-target genes whose mRNA levels were markedly decreased in SRSF3-KO GSCs (Fig. 5C). qRT-PCR results validated the downregulation of ADAM metallopeptidase domain 19 (ADAM19), fermitin family member 1 (FERMT1), and integrin alpha 2 (ITGA2) in SRSF3-KO and ETV1-E7 skipping (ETV1-Mut) GSC83 cells (Fig. 5D). Interestingly, these genes are involved in the processes of integrin activation and cell-extracellular matrix adhesion that are critical in oncogenic signaling, stemness and invasion of cancer cells (33). We validated the downregulation of ADAM19 and ITGA2 in GSC83 cells upon CRISPR/Cas13d-mediated ETV1 KD (Fig. 5E) and in GSC528 cells upon CRISPR/Cas9-mediated ETV1-E7 skipping (Fig. 5F). Exogenous expression of ETV1-FL increased the transcription of ADAM19, FERMT1, and ITGA2 to significantly higher levels than expression of ETV1-∆E7 in ETV1-Mut GSCs (Fig. 5G and 5H), and the re-expression of ETV1-FL, but not ETV1-∆E7, rescued the inhibition of GSC growth caused by CRISPR-mediated ETV1-E7 skipping (Fig. 5I). Taken together, these data demonstrate that the E7-encoded region of ETV1 is critical for its oncogenic transcriptional program.

Isoform-Specific Function of NDE1 in Mitotic Spindle Formation

NDE1 is a dynein adaptor protein that regulates a variety of microtubule-mediated processes including mitosis (34). Two AS isoforms of NDE1 are expressed in the brain that differ in exon 9, NDE1_KMLL and NDE1_SSSC (named for their four C-terminal amino acid residues, Fig. 6A) (35). We found that SRSF3-KO induced a switch from NDE1_KMLL to NDE1-SSSC isoform (Fig. 4B and Fig. 6A). CRISPR-mediated NDE1-E9 skipping resulted in defects of mitotic spindle formation in GSCs, which was rescued by exogenous expression of the NDE1-KMLL isoform but not the -SSSC isoform (Fig. 6B to 6D). Furthermore, the exogenous NDE1-KMLL isoform but not the NDE1-SSSC isoform proteins rescued GSC cell growth caused by CRISPR-mediated E9 skipping (Fig. 6E). These results revealed an isoform-specific function of NDE1 in mitotic spindle formation that is important for tumor cell growth.

Figure 6.

Isoform-specific function of NDE1 in mitotic spindle formation. A, Schematic of NDE1 protein depicting structural features, binding regions for the indicated NDE1-interacting proteins and alternative last E9. DIC, dynein intermediate chain. DHC, dynein heavy chain. LIS1, Lissencephaly-1. The distinctive amino acids in C-terminus of NDE1-SSSC and NDE1-KMLL isoforms were shown below. B, RT-PCR analysis of endogenous and exogenous NDE1 expression in GSC83 cells with indicated modifications. F: forward primer; R1: reverse primer for NDE1-KMLL isoform; R2: reverse primer for NDE1-SSSC isoform. C, Immunofluorescence (IF) images of DNA (DAPI) and α-tubulin in GSC83 cells with indicated modifications. Scale bar, 5 μm. D, Quantification of the spindle formation defects shown in IF images (C). E, Cell proliferation of GSC83 cells with indicated modifications. F, IB analysis of GSC83 cells that overexpressed GFP-NDE1-KMLL or -SSSC exposed to CHX (20 µg/ml) with indicated time. Right panel, quantification of IB data from three independent experiments. G, IB analysis of GSC83 cells that overexpressed GFP-NDE1-KMLL or -SSSC exposed to CHX (20 µg/ml) with or without MG132 (10 µM) for 12 hr. H, NDE1-E9 splicing pattern in TCGA and CGGA datasets. I, Kaplan-Meier analysis of TCGA and CGGA glioma patients in relation to PSI values of NDE1-E9.

Data are representative of two to three independent experiments with similar results. *, p < 0.05; **, p < 0.01; ***, p<0.001.

Next, we determined the subcellular localization and protein stability of the two NDE1 isoforms. Exogenous expression of GFP-tagged NDE1-KMLL or -SSSC isoforms produced similar diffuse cytoplasmic staining, with an accumulation in the form of dot-like structures that represent the centrosomes in interphase, but that are absent in metaphase of GSCs (Supplementary Fig. S7A). This result indicates that the variable C-terminus of NDE1 does not affect its centrosome recruitment.

NDE1 protein expression is cell-cycle-dependent, with expression increases found during M phase, and marked reductions in G1 phase (36), likely due to ubiquitin-mediated degradation (37). Consistent with these findings, marked degradation of the NDE1-KMLL isoform protein was seen after blocking protein synthesis with cycloheximide (CHX), and that was mediated by the ubiquitin-proteasome pathway (Fig. 6F, 6G, and Supplementary Fig. S7B). In contrast, the NDE1-SSSC isoforms were relatively stable for 12 hours in CHX-treated GSCs (Fig. 6F and Supplementary Fig. S7B). The absence of cell cycle-dependent expression might affect the mitotic function of NDE1-SSSC isoform.

Lastly, we investigated the clinical relevance of NDE1-E9 alternative splicing in clinical glioma patients. As shown in Fig. 6H, NDE1-SSSC is the dormant isoform in normal brain, while the percentage of KMLL isoform in glioma samples increased at various degrees, which correlated with tumor grades in both TCGA and CGGA datasets. Patients with a higher proportion of the NDE1-KMLL transcript associated with markedly worse prognosis (Fig. 6I). After adjusting for clinical variables, including status of IDH1, TP53 mutations, patients’ age and gender, the relationship of prognosis with NDE1-E9 splicing remained remarkably significant in both TCGA (Supplementary Fig. S8A) and CGGA (Supplementary Fig. S8B) datasets. These results suggest that SRSF3-regulated NDE1-E9 inclusion contributes to GBM tumorigenicity.

Discussion

Dysregulation of RNA splicing is a molecular characteristic of human cancer (5,38). Our study is the first extensive investigation of SRSF3 and its downstream AS events using GBM as model. Our data indicate an oncogenic role of SRSF3 in the tumor biology of GBM through SRSF3-controlled AS of gene transcripts such as ETV1 and NDE1.

SRSF3 has been described as a proto-oncogene and is overexpressed in various types of cancers, including breast (39), colon (40), ovarian (41), cervical, and bone cancer (42,43). However, its significance in glioma has not yet been thoroughly investigated. In this study, our gene expression profiling of 12 SR genes revealed SRSF3 as being markedly increased in glioma clinical specimens compared to normal brain tissues, and its elevated expression is associated with glioma histopathologic malignancy and reduced patient survival. Aberrant SRSF3 expression in gliomas does not result from the hypomethylation on the SRSF3 gene promoter, as SRSF3 methylation was similar between normal brain and glioma (http://maplab.imppc.org/wanderer/) (44). Potential reasons for SRSF3 upregulation in gliomas include amplification of chromosome 6p21 (42) and elevated Wnt signaling (45). Additionally, SR proteins are frequently phosphorylated at their C-terminal arginine/serine-rich (RS) domains by SR kinases (SRPKs) and cdc2-like kinases (CLKs), which is critical for the cellular localization and activity of the SR proteins (12). Moreover, SRSF3 appears to be hypo-phosphorylated in cells at steady state (46). Toward this end, we observed that SRSF3 migrated as a lower molecular weight protein in our IB analyses in normal brain tissues compared to glioma specimens, suggesting possible hypo-phosphorylation of SRSF3 in normal brain tissues. Together, our results suggest that active SRSF3 transcription and enhanced phosphorylation of SRSF3 promote oncogenic SRSF3-AS programs that contribute to GBM tumorigenicity.

Recent genome-wide studies have identified downstream AS targets of SRSF3 in various types of cancers (16,39,42,47) and mouse oocytes (48). However, since RNA splicing is often tissue-specific (49), it is plausible that the SRSF3-governed AS program in glioma differs from that of other cancers. In this study, analysis of deep RNA-seq data identified more than 1000 AS alterations upon SRSF3 KO in GSCs, including established SRSF3-target genes such as PKM (40), MAP4K4 (50), and BRD8 (48). Ajiro et al. reported that knocking down of SRSF3 activated a cryptic intron within the exon 4 of SRSF1 in HeLa cells (43). However, analysis of our RNA-seq data showed that SRSF3-KO had no effect on the splicing pattern of SRSF1 mRNA but caused a 2.3-fold decrease in SRSF1 mRNA level in both GSC83 and GSC528 lines. Therefore, it is possible that some of AS events identified from RNA-seq data were resulted from decreased SRSF1 rather than directly by SRSF3-KO. Therefore, we chose the most significantly altered AS events for the motif analysis and candidate selection to mitigate the indirect effects.

We found that SRSF3 primarily contributes to exon inclusion and that SRSF3-regulated alternative exons were enriched in genes involved with cell mitosis. To demonstrate the importance of SRSF3-regulated AS events in glioma tumor biology, we characterized the biological impact of SRSF3-induced AS of two novel target genes, ETV1 and NDE1. SRSF3-induced inclusion of ETV1-E7 results in retention of MAPK phosphorylation sites that are critical for preventing ETV1 from ubiquitin-mediated degradation (30,31). The enhanced ETV1 stability promoted an ETV1-mediated oncogenic transcriptional program in GSCs. Furthermore, SRSF3-induced exon switching promoted expression of an oncogenic NDE1-KMLL isoform that is crucial for mitosis and cell proliferation and also associated with GBM prognosis.

The development of next-generation sequencing has resulted in rapid progress in uncovering the regulatory aspects of AS networks. Although wide-spread AS alteration has been identified across cancers, there have been relatively scarce investigations of the functional consequences of AS alterations and the biological relevance of AS in cancer biology (49,51). Typically, isoform-specific functions are interrogated for their biological functions by exogenous expression of specific isoforms resulting from AS. However, this approach suffers from pathophysiological-irrelevant effects due to aberrant expression level, laborious process, and infeasibility for high-throughput screening. Splice-switching antisense oligonucleotide (SSO), a synthetic oligonucleotide with chemical modification targeting specific splice site(s) or cis-regulatory element(s) offers an alternative to determine the biological functions of AS. However, caveats of SSO that limit its application include low efficiency and off-target effects, reduced long-term effects, and difficulty for high-throughput screening (52). In contrast, CRISPR/Cas9 technology offers a simple, efficient, and inexpensive tool to specifically manipulate the genome (53). In this study, we utilized CRISPR/Cas9 technology to efficiently promote exon skipping by designing gRNA around the splice sites of targeted exons. In using this approach, we identified two novel SRSF3 downstream AS events that influence the biology of GBM. To our knowledge, this is the first study of CRISPR/Cas9-mediated splicing manipulations. The most valuable benefit of this CRISPR/Cas9-mediated RNA splicing editing is its application for high-throughput screening. The establishment of alternative splice site-based gRNA library would significantly facilitate the screening process for functional exons involved in various physiological and pathological processes including cancer. Moreover, with the newly developed RNA-targeting CRISPR/Cas13d system (29), novel types of AS manipulation other than induction of exon skipping could be developed by fusing catalytically “dead” Cas13d protein with the functional domain of specific RNA splicing factors.

There are several limitations to bulk tumor sequencing, especially for splicing analysis. Considering that alternative splicing is a tissue- and cell-type specific process (4), the complex cellular heterogeneity or even contamination by non-tumor stromal cells in bulk tumor tissues largely influence the overall splicing pattern. Therefore, it is difficult to distinguish truly aberrant AS events from those that are more cell type specific. With the development of single cell sequencing technology, deciphering the splicing code at single cell level will help us better understand the function of alternative splicing in tumor development.

Another limitation of our study is that we only testified the function of SRSF3 in glioma cells, but we cannot conclude whether the upregulation of SRSF3 is responsible for the mis-regulated RNA splicing during gliomagenesis. Whether the overexpression of SRSF3 in normal human astrocyte or neural progenitor cells leads to oncogenic AS and tumorigenesis warrants further investigations.

In conclusion, our study indicates SRSF3 as a significant regulator of glioma-associated AS program, implicating SRSF3 as an oncogenic factor that contributes to the tumor biology of GBM. Our study also establishes the feasibility of CRISPR/Cas9-mediated splicing manipulation, providing a novel method to investigate functional roles of mis-regulated RNA AS events in cancer tumorigenesis.