Muscleblind-like proteins (MBNL) belong to a family of tissue-specific regulators of RNA metabolism that control premessenger RNA splicing. Inactivation of MBNL causes an adult-to-fetal alternative splicing transition, resulting in the development of myotonic dystrophy. We have previously shown that the aggressive brain cancer, glioblastoma (GBM), maintains stem-like features (glioma stem cell, GSC) through hypoxia-induced responses. Accordingly, we hypothesize here that hypoxia-induced responses in GBM might also include MBNL-based alternative splicing to promote tumor progression. When cultured in hypoxia condition, GSCs rapidly exported muscleblind-like-1 (MBNL1) out of the nucleus, resulting in significant inhibition of MBNL1 activity. Notably, hypoxia-regulated inhibition of MBNL1 also resulted in evidence of adult-to-fetal alternative splicing transitions. Forced expression of a constitutively active isoform of MBNL1 inhibited GSC self-renewal and tumor initiation in orthotopic transplantation models. Induced expression of MBNL1 in established orthotopic tumors dramatically inhibited tumor progression, resulting in significantly prolonged survival. This study reveals that MBNL1 plays an essential role in GBM stemness and tumor progression, where hypoxic responses within the tumor inhibit MBNL1 activity, promoting stem-like phenotypes and tumor growth. Reversing these effects on MBNL1 may therefore, yield potent tumor suppressor activities, uncovering new therapeutic opportunities to counter this disease.
This study describes an unexpected mechanism by which RNA-binding protein, MBNL1, activity is inhibited in hypoxia by a simple isoform switch to regulate glioma stem cell self-renewal, tumorigenicity, and progression.
Glioblastoma (GBM) is among the least curable cancers because of distinct subpopulations of proliferative and invasive cells that coordinately drive tumor growth, progression, and recurrence after therapy (1–3). The apparent genetic heterogeneity in GBM and other solid cancers, even within a single patient's cancer, demands the identification and targeting of signaling nodes that are critical to multiple oncogenic pathways. Necrotic foci with surrounding hypoxic cellular pseudopalisades and microvascular hyperplasia are histologic features found in GBM (4). Over the past several years, we have demonstrated that hypoxia promotes the expansion of aggressive subpopulations of cells, often referred to as glioma stem cells (GSC; ref. 5). However, the mechanisms by which hypoxia regulates GSCs are not entirely clear.
Oxygen levels play an essential role in regulating stem cells in multiple tissues, including the central nervous system (reviewed in ref. 6). Growth under low oxygen concentrations is known to maintain pluripotency and inhibit the differentiation of embryonic stem cells (ESC; ref. 7). In the rodent brain, hypoxia regulates self-renewal, survival, and differentiation of stem and progenitor cells (8, 9). In humans, hypoxia promotes the expansion of progenitor cells (10). We have recently shown that low oxygen levels, commonly found in human tumors, promote the expansion of the cancer stem cell pool, as well as those grown in vitro for more extended periods (5). Along these lines, we and others have demonstrated the pivotal roles of hypoxia-inducible factors (HIF)-1α and -2α in orchestrating the transcriptional and metabolic adaptations to hypoxia, and in inducing stem cell phenotypes in the hypoxic microenvironment (refs. 5, 11–16; also reviewed in ref. 17). Thus, while the contributions of HIFs and hypoxia to promote self-renewal and expansion of GSCs are established, the identity of additional regulators of GSCs biology remains to be elucidated.
Precursor mRNA (pre-mRNA) splicing is a fundamental process in the regulation of eukaryotic gene expression. The mammalian nervous system makes extensive use of splicing regulation to generate specialized protein isoforms that affect all aspects of neuronal development and function (18–21). While splicing defects are increasingly implicated in neurologic diseases and several types of cancer such as myelodysplastic syndrome and acute myeloid leukemia, a clear role for specific alternative splicing events in GBM and their regulators remains to be established (22). Alternative splicing patterns are regulated by specialized RNA binding proteins that alter spliceosome assembly at specific splice sites (21, 23, 24). The muscleblind-like (MBNL) family of RNA binding proteins have been studied extensively in the context of the neuromuscular disorder, myotonic dystrophy, where inactivation of MBNL proteins results in a shift in splicing from an adult- to fetal-like patterns (25, 26).
Alternative mRNA splicing can substantially alter the functions of the proteins encoded by the mRNA (27). The MBNL family of sequence-specific pre-mRNA splicing factors bind RNA through pairs of highly conserved zinc finger binding domains that recognize YGCY (where Y = C or U) and similar motifs (28–32). MBNL proteins are predominantly expressed in skeletal muscle, neuronal tissues, thymus, liver, and kidney and are essential for terminal differentiation of myocytes and neurons (33). In the brain, muscleblind-like-1 (MBNL1) levels are significantly higher in astrocytes as compared with neural stem cells (34). MBNL1 is also involved in pluripotent stem cell differentiation, thereby linking isoform expression and the pluripotent and differentiation states (35). Importantly, Mbnl1 transcripts themselves undergo extensive alternative splicing, generating numerous protein isoforms. The inclusion of the highly conserved exon 5 is essential for nuclear localization and splicing activity of the MBNL1 protein (36, 37). The knockdown of Mbnl1 in cultured murine fetal liver progenitors blocks erythroid differentiation (38). Inactivation of the MBNL1 protein is critical in the etiology of myotonic dystrophy, resulting in cataract formation, abnormal muscle relaxation, heart and nerve dysfunction, and other pathologies (25, 39). Importantly, MBNL1 inactivation results in an adult-to-fetal alternative splicing shift in numerous MBNL1 target genes.
The function and expression of MBNL1 in gliomas, and specifically in GSCs, are currently unknown. So is the role of the hypoxic GBM microenvironment in regulating MBNL1 activity. Accordingly, this study explores the impact of hypoxia on MBNL1 activity in GSCs. In doing so, we found that while MBNL1 is expressed in all gliomas, MBNL1 activity is inhibited in hypoxia, mimicking myotonic dystrophy–like adult-to-fetal alternative splicing switching in multiple MBNL1 target genes. These included the fetal isoform of the GSC marker, integrin alpha 6 (ITGA6), and CD47, a novel macrophage immune checkpoint protein that plays a broad role in cancer immune evasion across multiple cancer types, suggesting the involvement of MBNL1 in the regulation of GSC maintenance and immune evasion. From a therapeutic standpoint, we show that the active form of MBNL1 inhibits GSC self-renewal in vitro and tumorigenic potential in vivo, and that inducing MBNL1 activity in established tumors can significantly prolong survival of animals bearing human GSC–derived orthotopic xenografts. Our study not only reveals unexpected mechanisms by which MBNL1 activity is regulated by the hypoxic microenvironment, but also demonstrates how hypoxia regulates cancer stem cell identity, GSC self-renewal, and tumorigenic potential by a simple isoform switch.
Materials and Methods
GBM neurosphere lines and hypoxic conditions
HSR-GBM1 and HSR-040821 and HSR-040622 were a kind gift from Dr. Angelo Vescovi (University of Milano-Bicocca, Milan Italy), and were established from freshly resected GBM tumors and passaged as described previously (40). T387, T3691, and T3832 were a kind gift from Dr. Jeremy N Rich (University of California, San Diego, San Diego, CA). HSR-GBM1 and HSR040821 and HSR-040622 are EGFR-WT and IDH1-WT. HSR-GBM1 is P53-WT, while HSR-040821 carries an S278P point mutation in the P53 gene. The PTEN gene is wild-type (WT) in these three lines. On the basis of transcriptional subtype classification, T387 is mesenchymal (MES; ref. 41) and T3691 and T3832 are proneural (PN; refs. 42, 43). Similar averaged expression of classifier genes in HSR-GBM1, HSR040622, and HSR040821 neurosphere lines led to an inconclusive determination of subtype classification for these lines. Mycoplasma testing was performed regularly, and the cultures were found to be negative. A hypoxic chamber maintained at 37°C, 1% O2, 5% CO2, and 94% N2 (X3 Biospherix) was used to conduct in vitro hypoxic experiments. All hypoxic experiments were conducted on cells that were plated and allowed to recover overnight before hypoxia induction.
MBNL1 depletion and overexpression
GBM neurospheres were transduced with one of two lentiviruses expressing short hairpin RNA (shRNA) constructs, TRCN0000429155 and TRCN0000427657 (Sigma). We refer to these shRNA constructs as sh1 and sh2, respectively. Forty-eight hours later, cells were selected, in bulk, with 5 μg/mL Puromycin (Thermo Fisher Scientific) for 3–7 days. In some experiments, the average of two independent cultures, each expressing one of the two shRNAs, is presented.
We refer to the active MBNL1 isoform that includes exon 5, only as MBNL1. The inactive isoform, lacking exon 5, is referred to as MBNL1Δ5. Lentivirus vector encoding for human MBNL1 under the constitutive cytomegalovirus (CMV) promoter was created by subcloning full-length MBNL1 cDNA from pCMV-SPORT6/MBNL1 (TrueClone, OriGene) into pLenti6 using the Gateway Cloning System (Invitrogen). The construction of the conditional MBNL1-expressing vector, pLV-GFP:T2A: Bsd-TRE>hMBNL1[NM_207293.1]:3xGGGGS:mCherry, was contracted to VectorBuilder Inc. All final constructs were sequenced either by the manufacturer or us. Generation of MBNL1Δ5-expressing lentivirus vector was achieved by site-directed mutagenesis, eliminating the coding sequence for exon 5 from the full-length cDNA in the pLV vector indicated above. Expression levels of MBNL1 and MBNL1Δ5 mRNA and protein were determined by qPCR and Western blot analysis, respectively.
RNA isolation and qRT-PCR analysis
RNA was isolated from tissues and cells using the Quick-RNA Isolation Kit (Zymo Research). cDNA was generated by reverse transcription with a high-capacity cDNA reverse transcription kit with an RNase Inhibitor (Thermo Fisher Scientific). For qPCR assays, gene expression analysis was performed using 10 ng of cDNA and AmpliTaq Gold DNA Polymerase with SYBR Green (Thermo Fisher Scientific). mRNA levels were normalized using hypoxanthine phosphoribosyltransferase 1 (HPRT). End-point RT-PCR was performed using the Bio-Rad C1000 Touch PCR Cycler and the following the manufacturer's conditions. The first cycle of 8 minutes at 94°C was followed by 35 cycles of 30 seconds at 94°C, 30 seconds at 55°C, and 30 seconds at 72°C. The reaction was ended with the extension step of 10 minutes at 72°C. Visualization and analysis of amplified products were done using a Gel Doc XR System w/Universal Hood II (Bio-Rad) or an iBright FL1500 Imaging System (Invitrogen). For the MBNL1 target genes, the amplified products were loaded on 2% agarose gel. Primer sequences are available upon request.
GBM neurospheres were cultured in normoxia or hypoxia for 48 hours. Neurospheres were fixed in 4% paraformaldehyde in PBS and incubated at room temperature for 5 minutes with occasional gentle mixing. Following fixation, 200 μL aliquots were “cytospun” against Premium Superfrost Microscope Slides (Thermo Fisher Scientific, 12-544-7) for 15 minutes at 1,200 rpm. Flattened neurospheres were washed once with PBS containing 1 mol/L glycine and two times with PBS. Following a 2-minute permeabilization step with 0.1% Triton X-100 (in PBS) and incubation in PBS containing 0.1% BSA, MBNL1 was revealed using primary mAb, MB1a(4A8), and secondary monkey-anti-mouse conjugated to Alexa Fluor 488 (Invitrogen), and nuclei were stained using Hoechst 33342 (Sigma-Aldrich). Cells were washed in PBS and water, mounted with ProLong Diamond Antifade Mountant, and observed with a Nikon W1 Spinning Disk Confocal on Nikon Ti2 Inverted Microscope. Images were captured with a Hamamatsu sCMOS Camera (Hamamatsu).
MB1a was deposited in the Developmental Studies Hybridoma Bank (DSHB) by Morris, G.E. [DSHB Hybridoma Product MB1a (4A8)]. Mouse MB1a hybridoma cells were purchased from DSHB and cultured in Iscove's Modified Dulbecco's Medium (IMDM, Sigma, catalog no., I3390) supplemented with 10% FBS, 0.1% penicillin, and 0.1% gentamicin. For mAb production, 1 × 106 cells, suspended in 5 mL IMDM without serum, were inoculated into the cell compartment of a CELLine (catalog no., CL350) bioreactor. IMDM growth medium (340 mL) containing 10% FBS was added in the nutrient compartment. Supernatant was harvested once per week, where, in each cycle, one-fifth of the cells were inoculated back into the cell compartment, and the nutrient compartment was emptied, by aspiration, and filled with 340 mL of fresh growth medium. This cycle was repeated for seven times. The monoclonal MB1a (4A8) antibody was filtered through a 0.45-μm syringe filter, and aliquots were mixed with an equal volume of glycerol for long-term storage at −20°C. Each batch of antibody was tested by Western blotting to confirm the presence of MBNL1 and MBNL1Δ5 isoforms in lysates of the 913 neurosphere line.
For whole-cell extracts, cells were lysed in buffer [50 mmol/L Tris-HCl, pH 8, 1% (w/v) NP-40, 1% (w/v) SDS, 0.5% (w/v) Na-deoxycholate (DOC), and 2 mM EDTA] supplemented with “Protease Inhibitor Cocktail” (Roche), incubated for 30 minutes at 4°C, centrifuged for 30 minutes at 4°C 16,000 × g, and the protein amounts were quantified using Pierce BCA Protein Assay Kit. Next, cell lysates were sonicated and centrifuged at 4°C 16,000 × g. Supernatants were resolved by SDS-PAGE and transferred to nitrocellulose membranes using an iBlot 2 Gel Transfer Device (Invitrogen). Membranes were blocked in PBS containing 5% (w/v) BSA and 0.05% (v/v) Tween-20 (PBST) for 1 hour and incubated with primary antibodies at 4°C overnight. After washing with PBST, membranes were incubated with either anti-mouse or anti-rabbit secondary antibodies conjugated to horseradish peroxidase for 1 hour at room temperature. Signals were detected with an iBright FL1500 Imaging System (Thermo Fisher Scientific), anti-MBNL1 (1:1000, see above), anti-β-actin (1:20,000, AC-15 clone, Sigma), and KPL Peroxidase-labeled antibody to mouse IgG (1:7,500, 074-1806, SeraCare.com). MBNL1 IHC staining was performed using the mAb at 1:200 concentration using a Ventana BenchMark ULTRA System. Immunostaining was optimized on deidentified normal human brain and GBM specimens.
Bioinformatics and gene set enrichment analysis
Overall survival analysis of patients whose tumors were defined as MES (n = 58) was performed on the Betastis portal: http://www.betastasis.com/glioma/tcga_gbm/. The portal censored three samples because of insufficient expression data.
We performed gene set enrichment analysis (GSEA) according to (44). Normalized gene-level FPKM values for all samples were downloaded from the Ivy Glioblastoma Atlas Project (Ivy GAP) portal (https://glioblastoma.alleninstitute.org/static/download.html). We compared all 19 leading edge (LE) and 40 pseudopallisading cells around centers of necrosis (PAN) samples available against the “cancer hallmarks” gene sets.
Most experiments were performed three or more times. Where appropriate, we included a bar graph representation of the average of independent experiments.
Finally, all animal studies were approved by Case Western Reserve University Institutional Animal Care and Use Committee (protocol no., 2018-0051).
MBNL1 expression is highest in GBM defined as MES, inhibited in hypoxic elements of the tumor and within the MES subgroup, and correlates with better overall patient survival
MBNL1 is a crucial protein involved in the etiology of the RNA disease, myotonic dystrophy. Its primary normal function is to modulate splicing during embryonic development (25, 32, 35, 38, 45–47). MBNL1 protein levels are low in pluripotent ESCs and dramatically increased in their differentiated progenies. This increase is accompanied by an MBNL1-regulated switch from a fetal-to-adult splicing pattern (15–17). A recent gene expression study of 1,348 RNA binding protein genes in 11 solid tumor types, together with splicing changes, has implicated MBNL1 as a significant driver of splicing patterns in these tumors (48). To explore a potential role for MBNL1 in GBM, we queried The Cancer Genome Atlas (TCGA) database for MBNL1 expression in the three, transcriptionally defined, GBM subtypes: classical (CL), PN, and MES. This analysis showed that MBNL1 is expressed in all GBMs. In CL and PN GBMs, MBNL1 levels were 17%–22% over healthy brain controls. However, this finding was found not to reach statistical significance.
In contrast, we found that MES tumors express, on average, 50% higher levels of MBNL1 mRNA, and this was found to be highly significant (Fig. 1A). Next, we queried the Ivy GAP RNA-sequencing (RNA-seq) database to identify differences in MBNL1 expression in different anatomic structures of GBM defined by reference histology. To this end, we focused on the LE, which is generally accepted to contain primarily normoxic cells, and on PAN, which we have previously suggested to contain the hypoxic cancer stem cell niche (17). GSEA (44) comparing PAN with LE further confirmed that the hypoxia gene set is significantly enriched in PAN as compared with LE (Fig. 1B). A heatmap of the entire hypoxia gene set (Fig. 1B, right) shows that the vast majority of the genes in this gene set are overexpressed in PAN. Also, and consistent with our recent publication showing that mTOR signaling is induced in hypoxic GBM neurospheres (16), we found the mTOR gene set to be similarly enriched in PAN (Fig. 1B). MBNL1 mRNA levels were significantly downregulated in PAN as compared with the LE (Fig. 1C). To determine whether reduced MBNL1 expression in hypoxia is a property of a specific subtype or shared among the subtypes, we further queried MBNL1 expression in LE and PAN by subtype. We found that significant decreases in MBNL1 mRNA levels were primarily the property of MES (Fig. 1D) and CL (Supplementary Fig. S1) tumors. Taken together, these data show that MBNL1 is expressed in all GBMs and that in the majority of GBMs, MBNL1 mRNA expression is lower in hypoxic elements of the tumor. To determine the potential clinical implications of reduced MBNL1 expression in this area, we queried TCGA database focusing on patient survival by GBM subtype. The analysis revealed that reduced MBNL1 expression is associated with worse patient survival (8.7 vs. 14.3 months), and this was found to be the exclusive property of tumors that are defined as MES subtype (Fig. 1E).
Hypoxia inhibits MBNL1 splicing activity
Next, we determined MBNL1 mRNA expression levels in snap-frozen tissues from numerous GBM neurosphere–derived xenograft models by qRT-PCR. We found MBNL1 levels were variable, ranging from 0.39 to 1.8 the level of the house keeping gene, HPRT (Fig. 2A). It has been recently shown that GBM subtype, determined on the basis of “bulk” RNA analyses, is the property of individual cells. Each tumor contains cells, transcriptionally defined as MES, PN, or CL (49). Furthermore, in a more recent study, Neftel and colleagues have shown that transcriptional subtype classification can be defined as a “state” rather than as a fixed identity and that cells may transition between “states” (50). To determine whether MBNL1 expression is associated with a particular state (namely, OPC-like, NPC-like, AC-like, or MES-like), we queried the database described by Neftel and colleagues (https://singlecell.broadinstitute.org/single_cell/study/SCP393/single-cell-rna-seq-of-adult-and-pediatric-glioblastoma) and found that MBNL1 is expressed in all four “states” (Fig. 2B). In cultured neurospheres, MBNL1 protein levels were variable and not significantly affected by hypoxia (Fig. 2C). It is well documented that the splicing activity of MBNL1 is regulated in a negative feedback loop. MBNL1 promotes “self-splicing” of the MBNL1 pre-mRNA to generate a shorter isoform that lacks the highly conserved exon 5 (MBNL1Δ5), encoding for 18 amino acids that are C-terminal of the fourth and final zinc finger RNA binding domain (Fig. 2D). Inhibition of MBNL1 results in the generation of an MBNL1 isoform that includes exon 5 (MBNL1). Therefore, a delicate balance between MBNL1 and MBNL1Δ5 isoforms controls the overall MBNL1 activity (36). We, therefore, used Western blot analysis to quantify relative MBNL1 protein isoform expression levels. In healthy human brain tissues (NB1 and NB2), MBNL1Δ5 was by far the dominant isoform representing 94% and 93% of total MBNL1 protein (Fig. 2E). In 913 xenografts implanted in the flank or orthotopically, we documented a significant amount of the MBNL1 isoform, representing 38% and 25% of total MBNL1 protein, respectively. RT-PCR analysis of MBNL1 mRNA isoforms in GBM neurosphere–initiated xenografts consistently showed both MBNL1 and MBNL1Δ5 isoform transcripts were expressed in vivo. In contrast, and consistent with Fig. 2E, we noted that in healthy brain tissues, MBNL1Δ5 mRNA was expressed almost exclusively (Fig. 2F). To determine whether MBNL1 expression in vivo may be due to hypoxia, we cultured 913 and 821 neurospheres in normoxia or hypoxia followed by Western blotting to analyze MBNL1 protein isoform expression. Indeed, we found that hypoxia promoted a relative increase of MBNL1 over MBNL1Δ5 from 56% to 74% and from 37% to 61%, in 913 and 821 neurospheres, respectively (Fig. 2G). Because MBNL1 protein promotes the splicing of its pre-mRNA (36), these data suggest that hypoxia regulates MBNL1 splicing activity. To test this directly, we examined mRNA splicing patterns of five genes previously shown to be directly regulated by MBNL1 in human ESCs (35). Importantly, the functional consequence of MBNL1 binding to pre-mRNAs is known to result in either inclusion or exclusion of the downstream exon. For example, while MBNL1 binding to the MBNL1 and CD47 pre-mRNAs results in the exclusion of the downstream exon, MBNL1 binding to ITGA6, CTTN, and CLSTN1 pre-mRNAs promotes the inclusion of the downstream exon. We found dramatic switch-like shifts in the splicing patterns of all five genes from the differentiated to undifferentiated, ESC-like, states. Representative data from 913 neurospheres are shown in Fig. 2H and the average of three independent experiments is shown in Fig. 2I. These changes are consistent with loss of MBNL1 activity as they mirror the results obtained when MBNL1 expression was knocked down with shRNAs (Fig. 2I, purple bars; Supplementary Fig. S2A and S2B). Similar results were obtained with 622 and 821 neurosphere lines and are shown in Supplementary Fig. S3. In contrast, an analysis of 08-387, 3691, and 3832 neurosphere lines showed no significant changes in MBNL1 activity. Expression of the MBNL1Δ5 isoform did not affect the splicing of MBNL1 pre-mRNA targets (Fig. 2J), confirming that pre-mRNA splicing is facilitated by the MBNL1 isoform. Together, these results show that while MBNL1 mRNA and protein levels are variable in GBMs, hypoxia inhibits MBNL1 activity in some tumor-derived neurosphere lines to promote the expression of numerous gene isoforms associated with an ESC-like state.
A bichromatic fluorescence reporter for MBNL-directed alternative splicing confirms hypoxia-dependent inhibition of MBNL1 splicing activity at the single-cell level
To measure MBNL1 activity at the single-cell level, we utilized a bichromatic fluorescence reporter for MBNL-directed alternative splicing. This reporter contains an artificial exon of 28 nt, responsive to MBNL1, that shifts the reading frame from dsRED (exon skipping) to GFP (exon inclusion) as described previously (51) and schematically illustrated in Fig. 3A. In this case, the expression of dsRED is indicative of MBNL1 activity, while GFP represents inactivity. To validate that the reporter faithfully reports MBNL1 activity in GSC, we first examined the 913 neurosphere line, electroporated with the RG6 reporter, by imaging flow cytometry (Fig. 3B) and documented the presence of sorted dsREDHIGH and dsREDLOW subpopulations. To test whether the dsREDHIGH subpopulation is enriched in active MBNL1 protein, we electroporated the 913 neurosphere line with the RG6 reporter and FACS sorted dsREDHIGH and dsREDLOW cells and analyzed the splicing pattern of MBNL1 target genes (Supplementary Fig. S4A and S4B). This set of experiments confirmed that MBNL1 activity is significantly higher in dsREDHIGH as compared with dsREDLOW cells. To determine whether MBNL1 is required for splicing bias toward dsRED, we electroporated the 913 neurosphere line, integrating control or one of two shMBNL1-expressing lentiviruses (sh1 or sh2) with the RG6 reporter. We found that reduced MBNL1 expression resulted in a significant reduction in the dsREDHIGH population, indicating that MBNL1 is indeed required for splicing of the reporter to generate dsRED (Supplementary Fig. S4C). To measure the effects of hypoxia on MBNL1 activity, we cultured the 913 neurosphere line, electroporated with the RG6 reporter, in normoxia or hypoxia for 24 hours and determined MBNL1 activity by flow cytometry. This analysis showed that when cultured in normoxia condition, the 913 neurospheres had an almost even distribution of dsREDHIGH and dsREDLOW populations (Fig. 3C). In contrast, cells cultured in hypoxia condition showed an 87.5% inhibition of the dsREDHIGH population with a substantial increase from 52% to 94% in the dsREDLOW subpopulation (Fig. 3D). This set of experiments provides further support to the “bulk” analyses described in Fig. 2 and indicates that each GBM cell expresses both MBNL1 isoforms in normoxia, but under hypoxic conditions, MBNL1 activity is rapidly inhibited in most cells.
Hypoxia promotes MBNL1 nuclear export
Given the rapid inactivation of MBNL1 splicing activity we observed in hypoxic GBM neurospheres, we next determined whether this process may involve nuclear export of MBNL1. To this end, we cultured GBM neurospheres in normoxia and hypoxia conditions, followed by immunostaining for MBNL1. Remarkably, we found that MBNL1 localizes to the cytoplasm and nucleus of GBM neurospheres, which were cultured in normoxia. In contrast, MBNL1 is rapidly exported out of the nucleus under hypoxic conditions. These results clearly show that the hypoxia-dependent inactivation of MBNL1 splicing activity involves MBNL1 nuclear export (Fig. 3E and F). Finally, to determine whether MBNL1 nuclear exclusion occurs in vivo, we IHC stained human GBM surgical sections for MBNL1. As shown in Fig. 3G, cells with robust nuclear localization and with predominantly cytoplasmic localization were found in these specimens confirming that nuclear–cytoplasmic transport occurs in vivo.
MBNL1 inhibits GSC self-renewal in vitro
Because MBNL1 activity is required for differentiation of ESCs (35), we next asked whether MBNL1 may inhibit GSC self-renewal, one of the fundamental properties of all stem cells. Because self-renewal and proliferation are linked, we first determined the effect of MBNL1 expression of neurosphere proliferation. Supplementary Fig. S5 shows that conditional expression of MBNL1-mCherry (Supplementary Fig. S5A) results in a small, but significant reduction of cell viability (Supplementary Fig. S5B) and that this translates to a significant decrease in proliferation (Supplementary Fig. S5C). To determine the effect of MBNL1 on the self-renewal capacity of GBM neurospheres in a quantitative fashion, we transiently electroporated plasmids encoding for human MBNL1 or GFP, as control, under the control of a constitutive promoter. Cells were recovered overnight before being challenged in a self-renewal assay in methylcellulose for 10 days, as we described previously (5). We found that transient expression of MBNL1 significantly reduced GSC self-renewal in most neurosphere lines. Specifically, MBNL1 reduced the average sphere diameter of the 913 neurosphere line from 162.9 to 109.9 μm (Student t test, ***, P < 0.0001; Fig. 4A). Similar reductions were documented in 821 (164.1 to 106.0 μm; Student t test, ***, P < 0.0001; Fig. 4B), 3691 (159.8 to 102.3 μm; Student t test, ***, P < 0.0001; Fig. 4C), and 3832 neurosphere lines (154.8 to 95.5 μm; Student t test, ***, P < 0.0001; Fig. 4D). A similar trend, albeit less robust, was documented in 622 and 08-387 neurosphere lines (Supplementary Fig. S6).
MBNL1 inhibits GBM tumor initiation and progression
Next, we tested the effect of MBNL1 on tumor initiation and progression using a constitutive and doxycycline-inducible system approaches, respectively. qRT-PCR analysis confirmed a modest 4-fold increase in MBNL1 mRNA expression levels in 913 neurospheres transduced with pLenti6-MBNL1 compared with vector control (Fig. 5A). Tumor initiation, as detected by bioluminescence imaging, was confirmed in the control group as early as 23 days after injection. In contrast, it took cells transduced with lentivirus expressing MBNL1 roughly 80 days for luciferase activity to cross the radiance threshold of 1 × 106 (Fig. 5B). Representative animals from this experiment are shown in Fig. 5C. This delay in tumor initiation affected median survival (MS). Animals bearing xenografts transduced with pLenti6 lentivirus had a MS of 58 days compared with more than 200 days for MBNL1-expressing tumors (Fig. 5D; log-rank survival analysis, **, P < 0.01).
Next, we sought to determine the effects of MBNL1 on the progression of established tumors. To this end, we transduced the 913 neurosphere line with lentiviruses expressing MBNL1-mCherry under the control of a tetracycline-inducible system (913 MBNL1-mCherry Tet ON). We found that doxycycline-dependent induction of MBNL1-mCherry occurs in a dose-dependent fashion (Fig. 5E). Flow cytometry was used to determine the percentage and magnitude of this induction at the single-cell level. The percentage of mCherry-positive cells increased from 1.1% to roughly 60% in the vehicle and doxycycline-treated 913 MBNL1-mCherry Tet ON neurospheres, respectively. Also, we found roughly a 6-fold increase in the fluorescence intensity, over the background, in neurospheres treated with doxycycline (Fig. 5F and G). Next, we determined the activity of doxycycline-induced MBNL1-mCherry by examining the splicing patterns of MBNL1 target genes (Fig. 5H) and its effects on GSC self-renewal (Supplementary Fig. S7A). Supplementary Fig. S7B shows self-renewal analysis of control cells, expressing rtTA, but not MBNL1 (913 Luciferase Control Tet ON).
913 Luciferase MBNL1-mCherry Tet ON and 913 Luciferase rtTA control cultures were also used to determine the effects of MBNL1 on tumor progression. Following the establishment of large orthotopic xenografts, where radiance exceeded 1 × 106, mice were randomized to control and doxycycline treatment groups. Each group included 4 females and 4 males. We found that doxycycline-induced MBNL1 expression in tumors significantly prolonged the survival of animals (Fig. 5I). At the end of the experiment, 0 of 8 and 7 of 8 animals were alive in the control and doxycycline treatment groups, respectively. In contrast, and consistent with the in vitro self-renewal assays (Supplementary Fig. S7), doxycycline-treated animals bearing tumors integrating only rtTA had similar MS (59.5 vs. 63.5 days) to their control, noninduced, counterparts (Supplementary Fig. S8).
The vast majority of human genes are alternatively spliced, generating RNA isoforms that code for functionally distinct proteins (52, 53). Several recent reports highlight genes that are alternatively spliced in response to hypoxia. However, very few studies address the molecular mechanism regulating splicing in hypoxia (reviewed in ref. 54). Initial in silico analyses provided support to the notion that while the splicing regulator, MBNL1, is expressed in all GBMs and in the vast majority of the cells in each tumor, as evidenced by single-cell RNA-seq, generally, reduced expression of MBNL1 correlated with worse overall survival of patients. Interestingly, this association was found to be the exclusive property of tumors transcriptionally defined as MES. Taken together with the knowledge that MBNL1 promotes cellular differentiation during embryonic development (35), we hypothesized that MBNL1 might act as a tumor suppressor and thus, explored the possibility that its activity may also be regulated. Indeed, we show that healthy brain tissues primarily express the short MBNL1Δ5 isoform, representing the “differentiated” state.
In contrast, GBMs express a mix of the “embryonic” and the “differentiated” state isoforms. Importantly, when GBM neurospheres are cultured in hypoxia condition, they mostly express the “embryonic” mRNA isoform of MBNL1, which is indicative of MBNL1 protein inactivation (Fig. 2). This inactivation is readily detectable at the single-cell level (Fig. 3). Here, the bichromatic RG6 MBNL1 splicing minigene proved to be a sensitive measure of MBNL1-dependent splicing. Also, confocal microscopy performed on GBM neurospheres cultured in normoxia and hypoxia provided evidence for hypoxia-dependent MBNL1 inactivation by MBNL1 nuclear export. It would be necessary, in future studies, to elucidate this transport mechanism. These results show that while MBNL1 mRNA and protein are expressed in all GBMs and in most cells within each tumor, the activity of MBNL1 is rapidly inhibited in hypoxia condition and that in tumors, reduced MBNL1 expression is associated with worse patient outcome. Finally, we showed, for the first time, that the MBNL1 protein is localized to the cytoplasm or nucleus in human GBM surgical specimens (Fig. 3G). These data suggest that cells may require to expel MBNL1 out of the nucleus under certain physiologic conditions. We show that hypoxia promotes the export of MBNL1 out of the nucleus, but we would like to speculate that other stress conditions, such as acidic stress or inflammation, may also require cells to export MBNL1 out of the nucleus. These are all areas of interest that we continue to investigate.
If MBNL1 activity is inhibited in GBM to promote a stem cell state, would forced expression of the active isoform of MBNL1 inhibit tumor growth? Also, if MBNL1 expression is forced upon GSCs, will they exhibit reduced clonogenic and tumorigenic capacities? Indeed, we found that lentivirus expression of the prodifferentiation, active isoform of MBNL1 significantly inhibited GSC self-renewal of multiple GBM neurosphere lines (Fig. 4). Importantly, these effects documented even in GBM neurosphere lines were refractory to hypoxia in the context of MBNL1 splicing activity, indicating that most GBMs may share a sensitivity to MBNL1 activity. Furthermore, MBNL1 blocked tumor initiation and progression in orthotopic GSC-derived xenografts, indicating that the prodifferentiation isoform of MBNL1 is tumor suppressive (Fig. 5).
It has been recently reported that in breast cancer, MBNL1 suppresses metastatic colonization and stabilizes metastasis suppressor transcripts (55). While we have not observed apparent effects of MBNL1 expression on cell invasion in vivo, this aspect of MBNL1 activity remains to be explored in future studies. Also, a recent study by Fischer and colleagues described the role of MBNL2 in controlling the transcript abundance of hypoxia response genes in lung and breast carcinomas (56). We could not detect significant changes in MBNL2 mRNA levels in hypoxia in the models we studied. Our study reveals, for the first time, a novel and essential role for MBNL1 in GSC maintenance and links the hypoxic microenvironment to the regulation of alternative RNA splicing. Future studies will be required to identify the signaling pathways that are regulated by MBNL1-directed alternative RNA splicing.
We propose that a signaling pathway, activated in hypoxic cells, inhibits MBNL1 activity, by nuclear export, to induce the stem cell state. While the identity of this inhibitory pathway remains to be uncovered, therapies that are aimed at reversing the inhibition of MBNL1 splicing activity in GBM, and likely other tumors, may result in a new therapeutic avenue for these untreatable cancers.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
D.M. Voss: Data curation, formal analysis, validation, investigation, visualization, methodology. A. Sloan: Investigation. R. Spina: Formal analysis, supervision, validation, investigation. H.M. Ames: Formal analysis, investigation, visualization, methodology. E.E. Bar: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing.
This work was supported by grants from NIH NINDS (R21NS106553) and NCI (R01CA187780).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.