EZH2 targeting reduces medulloblastoma growth through epigenetic reactivation of the BAI1/p53 tumor suppressor pathway

Abstract
Medulloblastoma (MB) is a malignant pediatric brain tumor for which new therapies are urgently needed. We demonstrate that treatment with EPZ-6438 (Tazemetostat), an enhancer of zeste homolog 2 (EZH2) inhibitor approved for clinical trials, blocks MB cell growth in vitro and in vivo, and prolongs survival in orthotopic xenograft models. We show that the therapeutic effect is dependent on epigenetic reactivation of adhesion G-protein-coupled receptor B1 (BAI1/ADGRB1), a tumor suppressor that controls p53 stability by blocking Mdm2. Histone 3 trimethylated on lysine 27 (H3K27me3), a marker of silent chromatin conformation is present at the ADGRB1 promoter, and inhibition of EZH2, the catalytic component of the Polycomb Repressive complex 2 (PRC2) that methylates H3K27, switches the gene into an active chromatin status and reactivates BAI1 expression. Mechanistically, targeting EZH2 promotes transition from H3K27me3 to H3K27ac at the promoter, recruits the C/EBPβ (CREB binding protein) and CBP transcription factors and activates ADGRB1 gene transcription. Taken together, our results identify key molecular players that regulate ADGRB1 gene expression in MB, demonstrate that reactivation of BAI1 expression underlies EPZ-6438 antitumorigenic action, and provide preclinical proof- of-principle evidence for targeting EZH2 in patients with MB.

Introduction
Medulloblastoma (MB) is the most malignant brain tumor in children, and is subdivided into four distinct groups (WNT, SHH, Group 3, Group 4). Genomic studies evidenced a high rate of alterations in epigenetic regulators in MB, with group-specific clustering of aberrations [1]. H3K27me3, a marker of suppressive chromatin written by EZH2, the catalytic component of PRC2, is enriched in Groups 3 and 4, and correlates directly with EZH2 over- expression. In WNT and SHH groups, EZH2 expression is increased at the transcriptional level [2], and loss-of- function mutations in the CREBBP gene, which encodes the CBP acetyltransferase that writes H3K27ac marks, are found in WNT-MBs, suggesting that alteration in histone H3 trimethylation by several genetic mechanisms is a general feature across MB groups and an opportunity for therapeutic targeting [3]. Targeting EZH2 has antitumor effects in experimental cancers, and drugs targeting EZH2 (Tazemetostat, GSK2816126 and CPI-1205) are currently in clinical trials for lymphomas and advanced solid cancers [4]. However, a challenge with epigenetic therapy is that broad genome reprogramming is expected to occur, which will be tissue type and context-dependent, and in addition to inducing tumor suppressors, might also awaken cryptic oncogenes.

Therefore, it is important to establish the key genes that drive response to epigenetic therapy so as to select patients who are most likely to benefit, predict routes of therapeutic resistance, and emergence of unwanted side effects. The effectiveness of targeting EZH2 in different MB groups with clinically approved drugs has not been established, and the key target tumor suppressor gene(s) suppressed by H3K27me3 in MB and other cancers remain to be identified. Adhesion G-protein-coupled receptors are the second largest GPCR subfamily in humans. Their long N-termini contain various modular domains involved in cell adhesion, migration, polarity and guidance, all highly relevant pro- cesses in tumor cell biology [5]. ADGRB1 encodes brain- specific angiogenesis inhibitor 1 (BAI1), an orphan adhe- sion GPCR highly expressed in brain that has potent anti- tumorigenic activities [6–9]. BAI1 expression is downregulated in human MB samples and Adgrb1 loss destabilizes p53 and promotes MB development in mice, suggesting it is a new tumor suppressor in MB [10]. However, the mechanisms underlying ADGRB1 transcriptional silencing and potential for epigenetic reactivation towards therapeutic gain remain largely unknown. Here, we investigated the therapeutic impact of targeting H3K27me3 with EPZ-6438, a clinical EZH2 inhibitor, in cultured MB cells and in orthotopic MB xenografts. We examined the importance of epigenetic reactivation of the BAI1/p53 tumor suppressor axis in the antitumor effects and defined the underlying key transcription factors involved.

Results
First, we examined whether pharmacological inhibition of EZH2 had antitumor effects on MB cells in culture and found that treatment with EPZ-6438 greatly inhibited MB cell growth in vitro (Fig. 1a). To determine whether reduced cell growth was related to cell proliferation, we performed cell cycle analysis and found a G0/G1 block (Fig. 1b and Supplementary Fig. 1). As p53 is a major regulator of cell cycle progression, we repeated the experiment in cells sta- bly expressing TP53-shRNA and observed neutralization of EPZ-6438’s inhibitory effect on MB cell growth (Fig. 1b, c). Since BAI1 can stabilize p53 by blocking Mdm2 and is silenced in MB cells [10], we further tested whether the growth inhibitory effect of EPZ-6438 is dependent upon reactivation of BAI1 tumor suppression activity and found ADGRB1-shRNAs indeed abrogated the antiproliferative effect (Fig. 1d). Taken together, these results demonstratethat EPZ-6438 inhibits in vitro MB cancer cell growth in a BAI1/p53-dependent manner.We then explored the feasibility of targeting EZH2 toward therapeutic gain in vivo, using three genetically diverse orthotopic human MB xenograft models. MB cells (D556, D425, and ONS-76) were implanted in the cerebellum, and 2 (D556 and D425) to 4 (ONS-76) weeks later, mice were treated orally with EPZ-6438 (350 mg/kg). EPZ-6438 has poor brain uptake as it is a substrate for drug efflux pumps [11], so Elacridar (100 mg/kg), a P-glycoprotein inhibitor was administered prior to EPZ-6438 delivery.

Kaplan−Meier curves showed that treatment of EPZ-6438 significantly increased median survival in all three models, from 22.5 to 29 days (p = 0.04) in D556 xenografts, from 48 to 58.5 days in ONS-76 xenografts (p = 0.0008) and from 20.5 to 27 days in D425 xenografts (p = 0.006) (Fig. 2a). EPZ-6438-treated tumors were smaller, had reduced expression of proliferation marker Ki67, and had focal areas of apoptosis (Fig. 2b and Supplementary Fig. 2) and showed augmented nuclear H3K27ac/H3K27me3 staining ratio (Fig. 2b). The tumors also displayed enhanced BAI1 staining, so we determined whether EPZ-6438-mediated enhanced survival was BAI1- dependent by repeating the experiment with D556 and ONS-76 cells stably expressing ADGRB1-shRNAs. ADGRB1 knockdown completely abrogated the survival advantage conferred by EPZ-6438 (p = 0.62 and 0.31, respectively, log- rank test) in both xenograft models (Fig. 2c), demonstrating that EPZ-6438’s antitumor effects in MB are BAI1- dependent.To determine whether EZH2-mediated H3K27me3 plays a direct role in ADGRB1 gene silencing, we first examined whether the H3K27me3 mark is enriched at the ADGRB1 promoter. Chromatin immunoprecipitation (ChIP) assays showed abundant promoter binding of H3K27me3 in ADGRB1-silent MB cell lines (D556 and ONS-76), accompanied by moderate binding of EZH2 (Fig. 3a). Knockdown of endogenous EZH2 protein with two inde- pendent siRNAs or overexpression of KDM6A, the H3K27me3 demethylase, led to robust reactivation of ADGRB1 mRNA expression in both cell lines (Fig. 3b, c).

Treatment with pharmacological inhibitors of EZH2 (EPZ- 6438 and GSK126) induced a global shift from suppressive to active chromatin in the cells, as evidenced by a transitionD556 and ONS-76 cells with stable p53 knockdown has no effect on cell growth. p = 0.76 and 0.32, respectively, two-way ANOVA, n = 3. d Stable knockdown of BAI1 in both cell lines blocks EPZ-6438’s beneficial effect. p = 0.74 and 0.34, respectively, two-way ANOVA, n = 3. (−), DMSO (0.1%) vehicle controlfrom trimethylation to acetylation at the H3K27 mark (Fig. 3d, left). ChIP assays confirmed the same changes at the ADGRB1 promoter (Fig. 3d, middle), and this chromatin remodeling led to abundant reactivation of ADGRB1 expression (Fig. 3d, right). Taken together, these results suggest that targeting H3K27me3 is an effective way to reactivate ADGRB1 gene expression.We next investigated what transcriptional factors are involved in EPZ-6438-mediated ADGRB1 gene reactiva- tion. We focused on EPZ-6438 as it is currently in phase II clinical trials. The factors controlling transcription of theH3K27me3 and H3K27ac in tumors from D556-xenografts by IHC. Scale bar 1 mm (in H&E images) and 50 μm (in IHC images). c Kaplan−Meier survival curves show effect of EPZ-6438 on the sur- vival of mice harboring intracranial xenografts of D556 and ONS-76 cells with lentiviral shRNA-mediated abrogation of BAI1 expression(p = 0.62 and 0.31, respectively, log-rank test)greatly blocked EPZ-6438-mediated reactivation of ADGRB1 expression, while p300 knockdown had no effect (Fig. 4b).

CBP can bind BRG1 (Brahma-related gene 1), one of two catalytic ATPase subunits of mammalian SWI/ SNF remodeling complexes [15]. SWI/SNF complexes interact with transcription factors, coactivators/repressors, and can mobilize nucleosomes at promoters to regulate gene expression [16]. However, in our system, knockdown of BRG1 expression through SMARCA4 siRNA had no effect on EPZ-6438-induced ADGRB1 expression (Fig. 4b).Altogether, these results suggest that C/EBPβ and CBP are required for ADGRB1 reactivation, while p300 and BRG1are dispensable for this process.To further test whether C/EBPβ and CBP can directly activate ADGRB1 transcription, we used expression vectors forexpression was determined by RT-PCR. d Left, EZH2 inhibitors induce a global decrease of H3K27me3 expression, accompanied by an increase of the H3K27ac marker. Middle, ChIP assay shows treatment with the EZH2 inhibitor EPZ-6438 in D556 and ONS-76 cells depletes H3K27me3 from the ADGRB1 promoter. Right, EZH2 inhibitors reactivate ADGRB1 expression in MB cells by RT-PCR analysis. (−), DMSO (0.1%) control. All blots show representative images from three independent experiments with similar resultsC/EBPβ or CBP and cotransfected them with a luciferase reporter vector under the control of the ADGRB1 promoter (−400 to +100 of TSS) (Fig. 4a) in HEK293 cells. Either C/ EBPβ or CBP could induce a 5~7-fold activation of luciferase activity, confirming that these two factors can directly activateADGRB1 transcription (Fig. 4c). Consistently, ChIP experi- ments showed EPZ-6438 treatment promoted abundant bind- ing of both transcription factors to the endogenous ADGRB1 promoter (Fig. 4d). Taken together, these results indicate that epigenetic reactivation of ADGRB1 expression by EZH2inhibition involves recruitment of C/EBPβ and CBP to theADGRB1 promoter in MB cells.

Discussion
Targeting aberrant epigenetic programming in medullo- blastoma holds promise for new therapy. However, a challenge is to determine the extent of therapy-induced reprogramming and the key genes that drive response, as this will determine which tumors will respond and how resistance might emerge. We showed that treatment with EPZ-6438 (Tazemetostat), an EZH2 inhibitor approved for clinical trials extends survival in both SHH and Group 3 MB xenograft models, and this is associated with both a reduction in tumor cell proliferation and increase in apop- tosis. Remarkably, while this epigenetic therapy likely reprograms multiple genes across the MB genome, itseffects are entirely dependent upon reactivation of the BAI1-p53 tumor suppressor signaling axis.Evaluation of EZH2 targeting drugs in cancer is ongoing [17, 18], but the critical effectors of epigenetic reprogram- ming remain unknown. Finding them is important to predict which patients will respond and which pathways will lead to resistance and tumor recurrence. Our work shows that BAI1 is the main effector of EZH2 targeting in MB, and that the therapeutic effect of EPZ-6438 is lost in its absence. EPZ- 6438 therapeutic effect is also dependent upon p53, sug- gesting that BAI1’s protective effect on p53 underlies the tumor suppressive effects [10]. Thus, our findings have direct clinical applicability; we can predict that MB having overexpression of Mdm2 or mutated p53 will not respond to EZH2 inhibition and that inactivation of the p53 pathway will lead to therapeutic resistance.

Consistently, recent results showed that deletion of EZH2 before or after tumor formation accelerated tumor growth in a MYC-driven Group-3 MB model with TP53 deletion [19]. This shows that tumors mutated for p53 are resistant to EZH2-mediated antitumor effects, and that this may even unveil cryptic oncogenes.Targeting EZH2 shifts the ADGRB1 promoter from a transcription suppressive to active state, as evidenced by depletion of H3K27me3 and enrichment of H3K27ac. Muta- tions in the CREBBP gene, which encodes CBP, are frequent in WNT-MB, and lead to truncated CBP proteins [3] that lack the histone acetyltransferase (HAT) domain responsible forHuman MB cell lines ONS-76 (SHH Group; TP53 WT), ONS-76-shTP53, D556-Med (Group 3; TP53 WT), D556-shTP53, D425 (Group 3; TP53 WT), were authenticated by STR profiling, tested free of mycoplasma and cultured as described [10]. Experiments were repeated three times and representative data are shown. For cell viability experi-ments, MB cells (25 × 103 cells cultured on six-well plates in triplicate) were treated with 10 μM EZH2 inhibitor EPZ- 6438 (Chemietek, Indianapolis, IN) for 5 days, live cellswere counted by trypan blue exclusion using a hemocyt- ometer. For cell cycle analysis, cells were fixed in 70% ethanol/PBS at −20 °C overnight. Propidium iodide (PI) was used to stain DNA. Cell cycle profile data were acquired with a FACS Canto flow cytometer (Becton Dickinson) and analyzed with FlowJo 9 (Tree Star, Inc.). For knockdown experiments, cells were transiently trans- fected by two different gene-specific siRNA pools with TransIT-X2 transfection reagent (Mirus, Madison, WI). EZH2-siRNAs were from Cell Signaling Technology (6509) and OriGene (SR320099).

Control-siRNAs were from Santa Cruz (sc-37007) and OriGene (SR30004). EP300-siRNAs were from Santa Cruz (sc-29431) and Ori- Gene (SR320041). CEBPB-siRNAs were from Santa Cruz (sc-29229) and OriGene (SR300760). SMARCA4-siRNAswere from Santa Cruz (sc-29827) and OriGene (SR321835). CBP-siRNAs were from Santa Cruz (sc-29244) and Ori- Gene (SR300976). For stable ADGRB1-knockdown, ONS-76 cells were infected with ADGRB1-specific shRNA- expressing lentiviral particles (sc-45208-V, Santa Cruz) in the presence of polybrene (5 μg/ml) (Santa Cruz, Dallas,TX). The lentiviruses contain 3−5 expression constructs each encoding target-specific 19–25 nt (plus hairpin) shRNAs. Infected cells were selected with puromycin (1 μg/ml) (Santa Cruz) for 4 weeks.Chromatin immunoprecipitation (ChIP) assaysChIP was performed as described [9] using 100 μg of genomic DNA/sample. Antibodies (2 μg) against the fol-lowing proteins were used: histone H3K27me3 (ab6002, Abcam), histone H3K27ac (#8173, Cell Signaling), EZH2 (ab191250, Abcam), C/EBPβ (sc-7962, Santa Cruz) and CBP (#7389, Cell Signaling). ADGRB1 promoter primers used: 5′-GCT CAC TCT GAC CCT CTG CTC TTTC-3′(forward) and 5′-CGT CCC AGC AGC AGC AGC AGCAG-3′ (reverse).Western blottingWestern blotting was performed as described [9] on 30 μg of protein/sample with antibodies against EZH2 (E7031,Sigma-Aldrich; 1:2000 dilution), actin (sc-1616, Santa Cruz; 1:5000), HA-tag (sc-805, Santa Cruz; 1:2000), his- tone H3K27me3 (ab6002, Abcam; 1:1000), histone H3K27ac (ab4729, Abcam; 1:1000), total histone H3 (#4499, Cell Signaling; 1:1000), p53 (ab28, Abcam;1:2000), C/EBP β (sc-7962, Santa Cruz; 1:2000), p300(#54062, Cell Signaling; 1:500), CBP (#7389, Cell Sig-naling; 1:500), BRG1 (#49360, Cell Signaling; 1:500),cleaved caspase 3 (#9661, Cell Signaling; 1:1000), andcleaved PARP (#5625, Cell Signaling; 1:1000).Total RNA was isolated with Trizol reagent (Thermo Fisher Scientific) and first-strand cDNA prepared using AMVRNA PCR kit (TaKaRa) from 1 μg total RNA.

PCR amplification for ADGRB1 mRNA detection was carried outas described [9].IHC was carried out as described [9]. Briefly, deparaffinized slides were subjected to antigen retrieval by boiling (20 min, 100 °C) in 0.01 M Tris HCL (pH 10) and incubated over- night at 4 °C with anti-Ki67 (ab15580, Abcam; 1:1000), Cleaved caspase 3 (1:1000), histone H3K27me3 (1:1000), histone H3K27ac (1:1000), p53 (1:1000), and BAI1 (NB110-81586, Novus, 1:1000) antibodies. Immunostain- ing was detected with the avidin-biotin complex method, using diaminobenzidine (Abcam) as the chromogen.All protocols for mouse experiments were approved by the Institutional Animal Care and Use Committee (IACUC) at Emory University. Eight-week-old Female outbred athymic nude mice (Hsd:Athymic Nude-Foxn1nu; Harlan/Envigo)marked by tattoos [21] were injected with tumor cells (5 × 105 in 5 μl) into the cerebellum as described [10]. Two weeks later, surviving mice were randomly divided into two groups (8–13 mice per group) to minimize variance. The treatment group received oral delivery of Elacridar andEPZ-6438 (350 mg/kg) 5 days/week as described [17]. Elacridar was orally administered at a dose of 100 mg/kg 4 h before EPZ-6438 treatment. The control group mice received vehicle. No blinding was Tazemetostat performed .Results were analyzed using either two-tailed Student’s t test, two-way analysis of variance (ANOVA) or Log-rank test in GraphPad Prism 5.0 software to assess statistical significance. Sample size was determined based on our prior experience with similar experiments. P values < 0.05 were considered to be statistically significant. All histogram data represent mean ± SEM (standard error of the mean).