Inhibition of ID1–BMPR2 Intrinsic Signaling Sensitizes Glioma Stem Cells to Differentiation Therapy

As the most prevalent primary intrinsic brain tumor, glioblastoma (GBM) is a universally lethal cancer with a median survival less than 15 months despite multimodal therapy (1). GBMs contain strikingly heterogeneous cellular populations with diverse transcriptional profiles, morphology, invasive potential, tumorigenicity, and drug sensitivity. Despite unresolved controversies in their origin and identification, glioblastoma stem cells (GSCs) are functionally defined by self-renewal and tumor propagation (2). Because of their contributions to tumor initiation, progression, and recurrence, GSCs are important targets for therapy (3, 4). Like GSCs, neural stem cells (NSCs) self-renew and have long-term proliferative potential, but undergo terminal differentiation to generate different lineages of mature cells, including astrocytes, oligodendrocytes, and neurons for tissue homeostasis (5). GSCs and NSCs share some, but not all, regulatory mechanisms of self-renewal capacity through WNT, SHH, and NOTCH signaling. GSCs undergo forced differentiation through BMP treatment or culture with serum (6, 7). Upon differentiation, GBM cells lose tumorigenicity and become sensitized to chemotherapy and radiotherapy (8). Thus, differentiation therapy has been proposed as a promising strategy for GBM therapy with limited toxicity in adults.

Several modalities based on the induction of terminal differentiation have been proposed as cancer therapies. Retinoic acid, an active form of vitamin A, reduces GSC tumorigenicity and motility by impairing cytokine secretion (9). miR-182 regulates apoptosis, growth, and differentiation in GBM, and administration with miR-182 spherical nucleic acid decreased tumor size and increased animal survival rate (10). Recent small-molecule phenotypic screen discovered a compound that rapidly induced differentiation of GSCs in vitro and in vivo (11). Most notably, clinical trials of BMP ligand infusion have been initiated.

On the basis of their role in normal neural differentiation, BMPs, secreted glycoproteins of the TGFβ superfamily, may be ideal molecular candidates for GBM because they differentiate GSCs and subsequently diminish tumor growth in vivo, with limited toxicity in normal brain (6). BMPs bind to a cognate high-affinity type II receptor, BMPR2, to induce activating phosphorylation and heteromerization with a type I receptor, BMPR1. Activated BMPR1 initiates downstream signaling by phosphorylating the intracellular mediators, receptor-regulated SMADs (R-SMADs: SMAD1, SMAD5, and SMAD8), to promote R-SMAD binding to co-SMAD (SMAD4) and nuclear translocation as a heteromer to bind DNA and regulate transcription of BMP target genes (12), many of which regulate cell fate and differentiation (13). During development, BMPR expression transitions from a fetal form (BMPR1A), which induces neuroepithelial proliferation, to an adult receptor profile (BMPR1B), which mediates astrocytic and neuronal differentiation. Differentiation-inducing properties of BMPs make them promising candidates for GBM therapy, leading to the pharmacologic development of BMP ligands in neuro-oncology (6, 7, 14, 15), but approximately 15% of patients with GBM harbor hypermethylation of BMPR1B promoter (7), which renders them unresponsive toward BMP-induced differentiation. Moreover, GSCs, sitting at the apex of cellular hierarchy in tumor BMP environment, induce autocrine secretion of BMP antagonist Gremlin1, which protects them from differentiation (16). Keeping these reports in mind, we hypothesize that GSCs have evolved mechanisms to evade BMP-induced differentiation, maintaining intratumoral heterogeneity. Further investigation of novel BMP resistance mechanisms should improve outcomes toward BMP-induced differentiation therapy.

GSCs, to evade extracellular BMP signaling, need to activate intrinsic mechanisms independent to their microenvironment. Identification of precise intracellular resistance mechanism blocking microenvironment-specific BMP signaling is important for the targeted differentiation therapy. The inhibitor of differentiation (ID) proteins are basic helix-loop-helix (bHLH) transcriptional factors lacking a DNA-binding domain that perform key functions in normal neurogenesis and GBM tumorigenesis (17). ID proteins maintain stemness traits by interacting with the differentiation-related bHLH transcription factors, such as Mash1 and MyoD, consequently inhibiting differentiation-related genes. Moreover, ID1 can activate noncanonical intracellular WNT and SHH signaling by upregulating Dvl2 and Gli2 proteins through suppression of Cullin3 E3 ubiquitin ligase in a ligand-independent manner for maintenance of GSC stemness (18). Therefore, ID proteins warrant further investigation for their role in suppression of lineage differentiation of GSCs.

Mouse Ink4a/Arf^−/− astrocytes were isolated from the cerebral cortices of 5-day-old Ink4a/Arf knockout mice (19) and were maintained in DMEM supplemented with 10% of FBS (Biotechnics Research). GSC8 and 528 cells obtained from Dr. Ichiro Nakano (University of Alabama at Birmingham) were cultured in neurobasal medium (Invitrogen) supplemented with modified B27, EGF (20 ng/mL; R&D Systems), and basic FGF (20 ng/mL; R&D Systems; ref. 20).

To evaluate tumor neurosphere-forming ability, mouse Ink4a/Arf^−/− astrocytes, GSC8, and 528 cells were plated at 200 cells per well in 24-well plates and cultured under stem cell culture conditions. The neurospheres were counted after 12 days (21). For in vitro limiting dilution assays (LDA), the cells were plated at decreasing numbers per well (100, 50, 25, and 5 for mouse Ink4a/Arf^−/− astrocytes; 20, 10, 5, and 1 for GSC8 cells) in 96-well plates, and the numbers of wells with neurospheres were calculated after 12 days. The tumor neurosphere-forming ability in in vitro LDAs was evaluated using the extreme limiting dilution analysis function (http://bioinf.wehi.edu.au/software/elda/).

Cells were treated with BMP4 (recombinant human BMP4 protein; 50 and 100 ng/mL; R&D Systems; ref. 6), GANT61 (Gli antagonist; 5.5 and 10 μmol/L for GSC8 and Ink4a/Arf^−/− astrocytes, respectively; Tocris Bioscience; ref. 22), JW67 (β-catenin inhibitor; 16.1 and 4.1 μmol/L for GSC8 and Ink4a/Arf^−/− astrocytes, respectively; Tocris Bioscience; ref. 23), and/or JQ1 (c-Myc inhibitor; 2.5 μmol/L for Ink4a/Arf^−/− astrocytes; Sigma-Aldrich; ref. 24) for 12 days to monitor the tumor neurosphere-forming ability.

293FT cells (Life Technologies) were cotransfected with a lentiviral construct of pLL-CMV-GFP, pLL-CMV-ID1-GFP, pLenti-CMV/TO-GFP-DEST, pLenti-CMV/TO-GFP-BMPR2-DEST, pCDH-CMV-dsRed, pCDH-CMV-MYC-dsRed, pLKO.1-puro, or pLKO.1-shMYC-puro vectors at 4 μg; pMDLg/pRRE at 2 μg (Addgene; #12251, MA, USA); pMD2.G at 1.2 μg (Addgene; #12259); and pRSV-Rev at 1 μg (Addgene; #12253) using 24 μL of the PolyExpress reagent (EG-1072, Excellgen). The medium containing the lentivirus was harvested 48 hours after transfection. Lentiviral particles were concentrated and purified using a Lenti-X TM concentrator (Takara). Ink4a/Arf^−/− astrocytes and GSC8 cells were infected with the lentivirus in the presence of 10 μg/mL polybrene (Sigma-Aldrich). Ink4a/Arf^−/− astrocytes were directly transfected with β-catenin^S37A, GLI2, or control vectors (pCDNA3.1-puro) using microporation. The cells were transduced with antisense miR-17 and miR-20a (50 nmol/L; Sigma-Aldrich; SIC001; ref. 25) and nontargeting oligonucleotides using FuGENE HD (Promega). The nucleotide sequences used for short hairpin RNAs (shRNA) and anti-miRNAs are shown in Supplementary Table S1.

Protein samples (30–100 μg) were resolved by SDS-PAGE in 10% gel (NuPAGE Bis-Tris gel, Invitrogen) and transferred to polyvinylidene fluoride membranes (Millipore). The membranes were blocked with 5% nonfat milk and incubated with specific antibodies against target proteins. The antibodies used for Western blotting are shown in Supplementary Table S2.

Patient-derived glioblastoma tissue samples were incubated with primary BMPR2 and Nestin antibodies overnight at 4°C. Nuclei were stained with 4′,6-diamidino-2-phenylindole (1 μg/mL; Sigma-Aldrich) for 5 minutes. Fluorescence images were obtained using a confocal laser scanning LSM 5 Pascal microscope (Carl Zeiss). Quantification of immunofluorescent staining was conducted with MetaMorph software (Molecular Devices). For IHC, cell populations positive for Myc, Nestin, Sox2, S100β, Tuj1, or O4 were quantified by counting the positively stained cells (100 nuclei) in five randomly chosen high-power fields of view. Antibody information is shown in Supplementary Table S2.

To analyze the proportions of cells positive for Nestin, Sox2, S100β, Tuj1, and O4, the dissociated cells were fixed with 4% paraformaldehyde and subsequently permeabilized with 0.1% saponin. These cells were then incubated with anti-Nestin, anti-Sox2, anti-S100β, anti-Tuj1, and anti-O4 antibodies, followed by incubation with the appropriate biotin-conjugated secondary antibody and PE-conjugated avidin (BD Pharmingen). The fluorescence intensities were measured by flow cytometry (FACSCalibur). Antibody information is shown in Supplementary Table S2.

qRT-PCR was conducted on an iCycler IQ real-time detection system (Bio-Rad Laboratories) using the IQ Supermix with SYBR-Green (Takara). The PCR primer sequences are presented in Supplementary Table S3. miRNA expression levels were determined using the TaqMan microRNA Assay Kit (Applied Biosystems).

Reporter gene activity was determined in cells transfected with the following plasmids using the Dual-Glo Luciferase Assay System (Promega): pGL3-Gli-BS (containing 8 repeat Gli-binding sites), pTOP-FLASH (containing multiple TCF/LEF-binding sites), pGL3-BMPR2-promoter, pGL3-WT-BMPR2 3′-UTR (containing wild-type BMPR2 3′-UTR), or pGL3-MT-BMPR2 3′-UTR (containing a mutation of miR-17 and miR-20a–binding sequences in BMPR2 3′-UTR). Firefly luciferase activity was normalized to Renilla luciferase activity. The primers used to clone the promoter or 3′-UTR are shown in Supplementary Table S4.

All animal research was conducted in accordance with protocols approved by the Institutional Animal Care and Use Committee of Korea University (Seoul, Korea) and performed in accordance with governmental and institutional guidelines and regulations. We used only 5-week-old female Balb/c nude mice. A model of an orthotopically transplanted brain tumor was established by injecting 10³Ink4a/Arf^−/− astrocyte-ID1 or GSC8 cells combined with BMP4, JW67, and/or GANT61. Ink4a/Arf^−/− astrocyte-ID1 cell–driven brain tumor–bearing mice were administrated intraperitoneally with BMP4 and JQ1 alone or in combination along with elacridar (MedChem Express) for penetration of the blood–brain barrier (26). The cells were injected stereotactically into the right striata of the mouse brain (coordinates: anterior-posterior, +2; medial-lateral, +1; dorsal-ventral, −3 mm from the bregma).

A treatment trial was performed with stereotactic injection of the WNT, SHH inhibitor and BMP inducer (n = 8, respectively), or of the DMSO (mock-treated group, n = 8) in the brain of Balb/c nude mice. Animal sample size was estimated by the statistical method (27). We randomized the mice to the treatment groups.

For this purpose, we used three prediction software packages: miRanda (28), TargetScan (29), and PicTar (30).

Chromatin immunoprecipitation sequencing (ChIP-Seq) data were collected from two published resources (31, 32). Histone H3 lysine 27 acetylation (H3K27ac) and Myc status were verified with genomic sequences of miR-17 in different cell lines on IGV (v 2.3.80) browser. DNA-binding motif analysis of H3K27ac-enriched enhancer region sequences was performed using the MEME Suite 4.11.2 programs (33).

A microarray database (GSE40614; http://www.ncbi.nlm.nih.gov/geo/), The Cancer Genome Atlas (TCGA) database (34), and other databases [Gleize (35), Gravendeel (36), and Pola.network (37)] were used to analyze correlations between mRNA expression of ID family members and the enrichment score of lineage differentiation signatures in glioma specimens. The enrichment score was measured by single sample gene set enrichment analysis. TCGA (38), Sun (39), Murat (40), Gill (41), Grzmil (42), Freije (43), and Lee (20) datasets were used to analyze fold change of mRNA expression levels of BMP-mediated differentiation signaling genes. Furthermore, the GSE40614 microarray database was utilized to analyze a statistical fold change of cancer gene neighborhoods (CGN) gene set signatures (44) between the ID1 knockdown group and control group.

Student t test was used to analyze statistical significance of differences between the paired groups. One-way ANOVA was applied to testing of statistical significance of differences among multiple groups (more than two groups). Overall survival of patients with glioblastoma was presented using Kaplan–Meier curves, and significance was determined by log-rank analysis. The data are expressed as means and 95% confidence intervals. All statistical tests were two-sided, and data with P < 0.05 or P < 0.01 were assumed to be statistically significant.

To explore the relationship between the ID family and lineage differentiation, we first examined the correlation between four ID family members (ID1–4) and lineage differentiation signatures including astrocytic, oligodendrocytic, and neuronal lineages (45) in multiple available gene expression patient datasets, including Gleize, Gravendeel, Pola.network, and TCGA GBM-low grade glioma (GBM-LGG) datasets (34–37). Among all the IDs, only ID1 consistently showed a negative correlation with each lineage differentiation signature in at least three glioma gene expression datasets (Fig. 1A–C; Supplementary Fig. S1A–S1C). Furthermore, this negative correlation between ID1 and differentiation signatures was confirmed in microarray data from two patient-derived GSC models subjected to ID1 knockdown, and we found ID1 knockdown significantly enriched astrocytic and neuronal lineage differentiation signatures in both GSCs (Fig. 1D; Supplementary Fig. S1D). To ascertain the molecular pathways altered by ID1 knockdown in GSCs, we analyzed cancer gene neighborhood (CGN) gene signatures of two patient-derived GSC models subjected to ID1 knockdown and found that 10 and 12 CGN gene signatures in GSC2 (Fig. 1E) and GSC8 cells (Fig. 1F), respectively, were enriched by more than eightfold increases upon ID1 knockdown. Of these signatures, four overlapped between the two GSC lines, with only BMPR2 as a known gene associated with the lineage differentiation pathway (Fig. 1G). In agreement with this finding, BMPR2 was one of the most markedly upregulated genes in ID1-knocked down GSCs (Fig. 1H). BMPR2 mRNA expression was consistently downregulated in GBM specimens compared with nontumorigenic brain or low-grade glioma specimens in multiple available gene expression patient datasets, including TCGA (38), Sun (39), Murat (40), Gill (41), Grzmil (42), Freije (43), and Lee (20) GBM or GBM-LGG datasets (Fig. 1I). In GBM tissue, Nestin (a neuronal precursor cell marker) and BMPR2 proteins were separately expressed in individual cells, and 48% of the cells were BMPR2-positive and constituted the bulk of the tumor (Fig. 1J; Supplementary Fig. S1E), indicating that BMPR2 expression is distinct from neuronal progenitors, confirming the molecular and cellular heterogeneity among GBM tissue. We further analyzed the TCGA dataset for the effect of BMPR2 expression on overall survival of glioma patients, and high transcript levels of BMPR2 significantly correlated with better prognosis in those patients (Fig. 1K). Taken together, these transcriptional profiling data suggest that ID1 functions as a negative regulator of lineage differentiation through BMPR2, which may act as a central player in the process with its expression, conferring hierarchy in the cellular tumor tissue.

To understand the possible correlation between ID1 and the BMP signaling pathway, we transfected ID1 into mouse astrocyte cells deficient in Ink4a/Arf tumor suppressor genes (Ink4a/Arf^−/− astrocytes; Fig. 2A) and into patient-derived GBM cells (GSC8 cells; Fig. 2B). ID1 decreased BMPR2 expression and phosphorylation of its downstream signaling molecules, SMAD1, -5, and -8, in both Ink4a/Arf^−/− astrocytes and GSC8 cells (Fig. 2A and B). As tumorsphere formation is associated with GSC self-renewal and poor prognosis of GBM (46), we examined sphere formation of GSCs after BMP4 treatment. Although BMP4 treatment decreased tumorsphere formation of both GSC8-control and GSC8-ID1 cells, ectopic ID1 expression markedly attenuated the BMP4-induced suppressive effect on sphere formation (Fig. 2C). BMP4 treatment inhibited sphere formation by 75% and 100% by 50 and 100 ng/mL in GSC8-control cells, and by 25% and 80% by 50 and 100 ng/mL in GSC8-ID1 cells (Supplementary Fig. S2A). FACS analysis of cellular differentiation markers demonstrated that BMP4 treatment (50 ng/mL) resulted in a dramatic reduction in the fractions of precursor cell marker positive cells (Nestin⁺ and Sox2⁺) and a significant increase in S100β⁺ (astrocyte marker), Tuj1⁺ (proneuronal cell marker), and O4⁺ (oligodendrocyte marker) cell populations among GSC8 cells. Reciprocally, ID1 overexpression attenuated the decrease in the number of cells positive for stem cell markers and decreased the populations of cells positive for differentiation markers (Fig. 2D). These findings indicate that ID1 may play a crucial role in the maintenance of GSC traits by inhibiting BMP-mediated lineage differentiation.

To test whether ID1 suppresses BMP-induced differentiation by inhibiting BMPR2 expression, we transduced BMPR2 into ID1-overexpressing Ink4a/Arf^−/− astrocytes and GSC8 cells. Western blot analysis showed that SMAD1, -5, and -8 phosphorylation in ID1-overexpressing cells was increased by BMPR2 restoration (Supplementary Fig. S2B and S2C). ID1-overexpressing Ink4a/Arf^−/− astrocytes and GSC8 cells formed larger and more numerous tumorspheres as compared with respective controls, whereas BMPR2 restoration decreased sphere size and number to control levels (Fig. 2E). Furthermore, in vitro LDAs revealed that the frequency of GSCs capable of forming tumorspheres significantly increased among ID1-overexpressing Ink4a/Arf^−/− astrocytes and GSC8 cells compared with their respective controls, but BMPR2 restoration in these cells caused marked reduction (Fig. 2F and G). A pairwise χ² test also indicated a significant effect of ID1-driven inhibition of BMPR2 expression on the acquisition and maintenance of GSC traits (Supplementary Fig. S2D and S2E). ID1 overexpression induced expression of the Nestin and Sox2 stem cell markers and suppressed the differentiation markers S100β and Tuj1 in Ink4a/Arf^−/− astrocytes and GSC8 cells, measured by qRT-PCR. In contrast, BMPR2 restoration reversed these effects in ID1-overexpressing cells (Fig. 2H and I). All these in vitro effects were confirmed in other patient-derived GSCs, 528 cells (Supplementary Fig. S2F–S2K). Thus, suppression of BMPR2 expression is necessary for ID1-mediated maintenance of GSCs.

To investigate the mechanism behind ID1-mediated inhibition of BMPR2 expression, we examined both transcriptional and posttranscriptional regulation of BMPR2 using a BMPR2-promoter-luciferase and BMPR2-3′-untranslated region (UTR)-luciferase construct, respectively, upon dose-dependent overexpression of ID1. We observed a minor but significant decrease in BMPR2 promoter-driven luciferase activity only at high level expression of the ID1 (Fig. 3A). In contrast, BMPR2-3′-UTR-luciferase activity significantly decreased even with low level of ID1 and showed a dose-dependent decrease upon ID1 expression (Fig. 3B). These results indicated that ID1 primarily inhibits BMPR2 levels via 3′-UTR–mediated posttranscriptional regulation of BMPR2 transcripts. As miRNAs control protein levels through mRNA degradation or translation blockade by targeting the 3′-UTR of mRNAs (47), we performed an in silico screening for miRNAs capable of targeting 3′-UTR of BMPR2 mRNA using PicTar, TargetScan, and miRanda prediction software packages. All these programs indicated that BMPR2-3′-UTR contains one evolutionarily conserved binding site for nine miRs: miR-17, miR-19a, miR-19b, miR-20a, miR-20b, miR-93, miR-106b, miR-130a, and miR-301 (Fig. 3C and D). Among these nine miRNAs, only miR-17 and miR-20 were markedly upregulated by ID1 overexpression in Ink4a/Arf^−/− astrocytes and GSC8 cells (Fig. 3E and F). To examine whether ID1 mediates its inhibitory effects upon BMPR2 expression via miR-17 and miR-20, we ectopically transduced the antisense oligos of miR-17 and miR-20a in Ink4a/Arf^−/− astrocytes-ID1 and GSC8-ID1 cells. Antisense-mediated inhibition of miR-17 and miR-20 was sufficient to increase BMPR2 expression and phosphorylation of SMAD1, -5, and -8 proteins in these cells (Fig. 3G and H). Confirming a regulatory role of these miRNAs, antisense miR-17 and miR-20a oligonucleotides restored BMPR2 mRNA levels of ID1-overexpressing cells to the levels similar to those in control cells (Supplementary Fig. S3A and S3B). Taken together, these results indicate that ID1 suppresses BMPR2 expression, at least in part, through miR-17 and miR-20a.

To further examine how ID1 upregulates miR-17 and miR-20a, we analyzed the status of chromatin modification in genomic locus encoding miR-17 using a published high-throughput sequencing of chromatin immunoprecipitation (ChIP-Seq) in GBM (31). Histone H3 lysine 27 acetylation (H3K27ac), a marker for activated promoters and enhancers (48), was enriched on the promoter region of the MIR17HG cluster encoding both miR-17 and miR-20a only in GSC-like tumor propagating cells compared with differentiated glioma cells (Fig. 4A). Moreover, the MYC-binding element is one of most conserved cis-acting elements within the MIR17HG enhancer region (Fig. 4B). MYC and H3K27ac ChIP-Seq analysis (32) indicated enriched MYC binding on the promoter region of miR-17 (Fig. 4C). Indeed, we found that histone H3 lysine 9 acetylation (H3K9ac), another marker for activated promoters and enhancers, and MYC are more enriched at the promoter region of MIR17HG cluster in ID1-overexpressing Ink4a/Arf^−/− astrocytes and GSC8 cells than control cells (Fig. 4D and E). Thus, these results suggest that miR-17 and miR-20a may be transcriptional targets of c-Myc.

To ascertain the possible relationship between MYC and the ID1–BMPR2 axis, we assayed c-Myc expression levels and stem phenotypes in our models. Both ID1-overexpressing Ink4a/Arf^−/− astrocytes and GSC8 cells upregulated MYC protein (Fig. 4F and G). MYC overexpression in Ink4a/Arf^−/− astrocytes or GSC8 cells decreased BMPR2 expression and phosphorylation of SMAD1, -5, and -8 proteins as much as ID1 overexpression. In contrast, MYC knockdown in ID1-overexpressing cells restored BMPR2 expression and phosphorylation of SMAD1, -5, and -8 proteins to the levels of control cells (Fig. 4F and G). These results support MYC upregulation as a mechanism in ID1 suppression of BMP signaling.

We next tested whether ID1 upregulates miR-17 and miR-20a via MYC. MYC and ID1 overexpression in Ink4a/Arf^−/− astrocytes and GSC8 cells induced miR-17 and miR-20a expression, while MYC knockdown in ID1-overexpressing cells repressed expression of these miRNAs (Fig. 4H and I), indicating that MYC mediates ID1 upregulation of miR-17 and miR-20a. To further support the phenotypic consequences of MYC and ID1 on GSC stemness, overexpression of either MYC or ID1 increased sphere formation of our glioma models, whereas MYC knockdown in ID1-overexpressing cells diminished sphere formation to control levels (Supplementary Fig. S4A and S4B). A pairwise χ² test also supported MYC- and ID1-mediated acquisition and maintenance of GSC traits (Supplementary Fig. S4C and S4D). Thus, ID1-mediated GSC stemness phenotype may be supported by MYC upregulation.

Next, we investigated how ID1 promotes MYC expression in GSCs. MYC is a common target of WNT and SHH signaling pathways (49), and ID1 activates the WNT and SHH signaling pathways through upregulation of Dvl2 and Gli2, respectively (18). Therefore, we determined expression levels of downstream target genes of WNT and SHH signaling in ID1-overexpressing Ink4a/Arf^−/− astrocytes and GSC8 cells. ID1 overexpression significantly increased mRNA levels of several target genes of WNT (COX2, AXIN2, MYCN, LGR5, CCND1, and MYC) and SHH (BMI1, HES1, GLI1, PTCH1, CCND1, and MYC; Supplementary Fig. S4E and S4F). As CCND1 and MYC were upregulated targets common to both WNT and SHH signaling, we examined whether increased MYC expression by WNT or SHH signaling activation would inhibit BMP differentiation signaling. Transduction of Gli2 and constitutively active β-catenin (β-catenin^S37A) into Ink4a/Arf^−/− astrocytes upregulated MYC and decreased BMPR2 and phosphorylation of SMAD1, -5, and -8 (Fig. 4J). Furthermore, ID1, β-catenin^S37A, Gli2, and c-Myc significantly suppressed luciferase activity regulated by wild-type 3′-UTR of BMPR2, but not by the mutant 3′-UTR of BMPR2 that lacks miR-17 and miR-20a target sites (Fig. 4K). Taken together, ID1 inhibits BMPR differentiation signaling through activation of the MYC–miR-17 and miR-20a regulatory axis via WNT and SHH signaling.

Informed by our mechanistic understanding of ID1 in GSCs, we investigated whether the MYC–miR-17/miR-20a axis can be therapeutically targeted to promote GSC differentiation. To this end, we examined the effects of WNT and SHH signaling inhibitors, JW67 and GANT61 (using concentrations that inhibited the WNT or SHH signaling reporter activities by half, respectively; Supplementary Fig. S5A and S5B), on BMPR and MYC expression. Combinatorial treatment with both JW67 and GANT61 dramatically elevated BMPR2 mRNA levels suppressed by ID1 overexpression in both Ink4a/Arf^−/− astrocytes and GSC8 cells, compared with each compound alone (Fig. 5A and C). In sharp contrast, treatment with JW67 and/or GANT61 significantly inhibited ID1-induced c-Myc mRNA levels in Ink4a/Arf^−/− astrocytes and GSC8 cells (Fig. 5B and D). Moreover, JW67 or GANT61 treatment markedly diminished miR-17 and miR-20a levels in Ink4a/Arf^−/− astrocytes-ID1 and GSC8-ID1 cells (Fig. 5E and F).

Next, we examined the effect of WNT and SHH signaling inhibition on GSC stemness and sensitivity to differentiation therapy. Singular therapies with BMP4, JW67, or GANT61 alone reduced sphere formation of Ink4a/Arf^−/− astrocytes-ID1 and GSC8 cells, but greater efficacy was observed in cells treated with all three agents (Fig. 5G and I). Furthermore, FACS analysis revealed that all single or combined treatment conditions of ID1-overexpressing Ink4a/Arf^−/− astrocytes and GSC8 cells with the three therapies reduced Nestin⁺ and SOX2⁺ cell populations (to a lesser extent in ID1-overexpressing Ink4a/Arf^−/− astrocytes), whereas the combined treatment specifically increased S100β⁺, Tuj1⁺, and O4⁺ populations of differentiated cells (Fig. 5H and J; Supplementary Fig. S5C). Moreover, we investigated whether the BET bromodomain inhibitor JQ1 (also known as Myc inhibitor) or well-known Myc inhibitor 10058-F4 could replace the combined treatment with JW67 and GANT61. Treatment with JQ1 or 10058-F4 in ID1-overexpressing Ink4a/Arf^−/− astrocytes cells increased the BMPR2-mediated differentiation signaling pathway (Supplementary Fig. S5D). Combined treatment with JQ1 and BMP4 resulted in a much more significant reduction in tumorsphere-forming ability than either single treatment (Supplementary Fig. S5E). FACS analysis revealed that the Nestin⁺ and SOX2⁺ cell populations markedly reduced after both single and combined treatments, whereas the S100β⁺ and Tuj1⁺ populations of differentiated cells significantly increased after combined treatment (Supplementary Fig. S5F). Thus, these in vitro studies indicate that combinatorial action on all three ID1-driven regulatory nodes of WNT/SHH stemness signaling and BMPR2 differentiation signaling pathways effectively suppresses GSC properties.

To determine the in vivo efficacy of simultaneous inhibition of ID1-dependent GSC stemness and differentiation resistance, we implanted ID1-overexpressing Ink4a/Arf^−/− astrocytes or GSC8 cells orthotopically into the brains of nude mice and then initiated the triple combined treatment or controls. Treatment with either BMP4 alone or combined JW67 + GANT61 significantly improved the survival of tumor-bearing mice compared with the control arm, but the most pronounced extension of mouse survival was observed in the triple combined treatment group (Fig. 6A). In GSC8 cells, combined JW67 + GANT61, but not BMP4, resulted in a modest but statistically significant prolongation in survival of tumor-bearing mice, with additional benefit from triple BMP4 + JW67 + GANT61 treatment (Fig. 6B). When control and BMP4-treated mice bearing ID1-overexpressing Ink4a/Arf^−/− astrocytes cells were sacrificed 25 days after implantation, dually or triply treated mice did not give rise to tumors. Thus, the latter two groups were sacrificed 80 days after implantation when they displayed neurologic signs (Supplementary Fig. S6A). In mice bearing GSC8 tumors, consistent with mice survival results shown in Fig. 6B, brain tumor size of dual and triple treatment groups was markedly decreased, whereas BMP4 alone treatment group did not produce robust effects (Supplementary Fig. S6B). Dual or triple treatments significantly reduced stem cell marker Nestin- or Sox2-positive tumor cells in tumor tissues compared with control and BMP4 alone treatment, whereas triple treatment strongly increased differentiation marker–positive tumor cells (GFAP, Tuj1, or O4) in brain tumor tissues generated by injecting ID1-overexpressing Ink4a/Arf^−/− astrocytes or GSC8 cells (Fig. 6C and D; Supplementary Fig. S6A and S6B). Furthermore, BMP4 and JQ1 alone or in combination were administrated intraperitoneally to the mice at 24 days after tumor cell implantation. Double and single treatment (to a lesser extent) significantly improved the survival of ID1-overexpressing Ink4a/Arf^−/− astrocyte-derived brain tumor–bearing mice compared with the control arm (Supplementary Fig. S7A). Double treatment also decreased the number of cells expressing Myc, Nestin, or Sox2, while increased cells expressing GFAP or Tuj1 (Supplementary Fig. S7B and S7C). These results indicate that reactivation of differentiation signaling by blocking ID1-driven intrinsic stemness signaling is an effective therapeutic strategy for GBM.

A hierarchical cancer stem cell (CSC) model, a hypothesis suggesting that the tumor bulk may contain self-renewing tumorigenic cells, existing among nontumorigenic, differentiated tumor cells constituting the majority of the tumor, accounts for phenotypic and functional heterogeneity observed in patient tumors (50). CSCs, present in the same tumor microenvironment along with bulk tumor cells, are able to maintain their identity by activating intrinsic signaling cascades. Identification and effective targeting of these intrinsic mechanisms are key to the differentiation therapy for GBM. In the current study, we demonstrated that the characteristics of GSCs depend on ID1, which plays a pivotal role in maintaining the tumorigenic self-renewal capacity of GSCs by simultaneously promoting stemness and evading differentiation (Fig. 6E).

BMPs have been reported to induce differentiation of GSCs. However, their use as differentiation therapies for GBM remains challenging due to the high dose (100 ng/mL) of BMP4 required and frequent hypermethylation of the BMPR1B promoter (7). Our study establishes an elaborate novel regulatory mechanism through which GSCs evade the BMP-mediated differentiation signaling pathway. Our findings reveal an intricate posttranscriptional regulation of BMPR2 by ID1 through miR-17 and miR-20a. We demonstrate that ID1 regulates the expression of MYC, which in turn transcriptionally activates oncomiRs miR-17 and miR-20a in GSCs. The regulation of expression of BMPR2 by these oncomiRs ensures attenuation of BMP-mediated lineage differentiation. The inactivation of BMP signaling by miR-17 and miR-20a may be a general mechanism of GSC maintenance because these miRNAs are upregulated by ID1 and MYC, which are commonly activated in GBM and directly contribute to the formation of GSCs. Taken together, ID1, via suppressing BMPR2 expression, enables GSCs to escape differentiation in BMP-enriched tumor microenvironment maintaining hierarchy.

Recent evidence suggests that BMP-based differentiation therapies have the potential of overcoming possible resistance. GSCs are known to activate autocrine Gremlin secretion, which shows a competitive binding toward BMPR2, leading to the suppression of differentiation (16). However, analysis of the patient data demonstrated that BMPR2 expression is strongly repressed in GBM tumors, indicating that BMPR2 serves as a strong connecting link between stem and differentiated states. Our study underscores the importance of a two-pronged approach that can be applied to target GSCs. We demonstrate that simultaneous inhibition of ID1-mediated WNT/SHH stemness signaling and activation of BMPR2-mediated differentiation signaling leads to loss of stem-like properties of GSCs and also an enhanced differentiation of these cells in vitro. Thus, combinatorial targeting of core GSC signaling cascades can not only increase sensitivity of GSCs to differentiation therapy, but also reduce the risk of resistance.

This study presents strong rationale for a concurrent intervention actively crippling both GSC stemness signaling and promoting differentiation in contrast to the currently prevalent treatment strategies for targeting stemness alone. By demonstrating a novel GSC-specific intrinsic signaling program, our study presents a strong mechanism that helps in maintaining hierarchy in GBM tumors. Current CSC differentiation therapy is based largely on findings observed in normal stem cell. Further study is required to investigate other stem cell–specific intrinsic cascades, which may help tumors to escape microenvironment-specific differentiation programs. We strongly advocate the requirement for a similar combinatorial treatment approach described herein using pharmacologic inhibitors of intrinsic signaling cascades along with a differentiation inducer for effective targeting of CSCs and eradication of malignant tumors.

No potential conflicts of interest were disclosed.

Conception and design: X. Jin, X. Jin, H. Kim

Development of methodology: X. Jin, X. Jin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): X. Jin, X. Jin, H.-Y. Jeon, E.-J. Kim, J.-K. Kim

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): X. Jin, X. Jin, L.J.Y. Kim, D. Dixit, J. Yin

Writing, review, and/or revision of the manuscript: X. Jin, X. Jin, L.J.Y. Kim, J.N. Rich, H. Kim

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): X. Jin, X. Jin, J. Yin, J.N. Rich

Study supervision: J.N. Rich, H. Kim

Other (graphical abstract): S.Y. Lee

We thank Dr. Eek-Hoon Jho (University of Seoul, Seoul, Republic of Korea) for providing pCS2-MT-Gli2 and pBI-HA-β-catenin^S37A plasmids. This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Ministry of Science, ICT and Future Planning (2011-0017544 & 2015R1A5A1009024), Next-Generation Biogreen21 Program grant (PJ01107701), Korea University grant. J.N. Rich is supported by NIH grants CA154130, CA169117, CA171632, NS087913, NS089272, and the James S. McDonnell foundation. X. Jin (Xun) was supported by grants from the General Program of the National Natural Science Foundation of China (no. 81572891).

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.