Histone Demethylase Jumonji D3 (JMJD3) as a Tumor Suppressor by Regulating p53 Protein Nuclear Stabilization

Histone methylation regulates normal stem cell fate decisions through a coordinated interplay between histone methyltransferases and demethylases at lineage specific genes. Malignant transformation is associated with aberrant accumulation of repressive histone modifications, such as polycomb mediated histone 3 lysine 27 (H3K27me3) resulting in a histone methylation mediated block to differentiation. The relevance, however, of histone demethylases in cancer remains less clear. We report that JMJD3, a H3K27me3 demethylase, is induced during differentiation of glioblastoma stem cells (GSCs), where it promotes a differentiation-like phenotype via chromatin dependent (INK4A/ARF locus activation) and chromatin independent (nuclear p53 protein stabilization) mechanisms. Our findings indicate that deregulation of JMJD3 may contribute to gliomagenesis via inhibition of the p53 pathway resulting in a block to terminal differentiation.


Introduction
Recent evidence suggests that histone methylation is a dynamic process that modulates transcriptional activity in both normal and cancer cells [1][2][3][4][5][6]. Histone methylation is a result of a coordinated interplay between histone methyltransferases and demethylases at lineage specific gene promoters thereby contributing to normal development by regulating cell fate decisions [4,7]. The relevance of the interplay between histone methyltransferases and demethylases during differentiation has been demonstrated in the developing central nervous system [8,9]. Here, it was shown that polycomb protein deletion in embryonic neural stem cells (NSCs) accelerates either neurogenesis or gliogenesis depending on the polycomb component deleted and timing of deletion [8][9][10]. Moreover, others showed that the H3K27 demethylase JMJD3 is required for re-activation of neuronal specific genes during differentiation of embryonic NSCs [4]. In these experiments it was shown that the retinoic acid co-repressor NCOR-2 suppresses JMJD3 expression. Retinoic acid treatment destabilizes nuclear NCOR-2 resulting in de-repression of the JMJD3 promoter and JMJD3 mediated induction of neurogenic differentiation. Moreover, NCOR-2 mediated transcriptional suppression was also previously shown to inhibit astrocytic differentiation [11]. Altogether, these results suggest that a coordinated temporal interplay between EZH2 and JMJD3 modulate stem cell self-renewal and lineage specific gene expression of NSCs. It was been demonstrated that primary human glioblastomas have stem-like cells, termed glioblastoma stem cells (GSCs), that can be enriched for using the neural stem cell marker CD133 and CD15 [12,13]. GSCs isolated from primary glioblastoma patient samples maintain the genotype and phenotype of the original tumour sample from which they were derived [14,15]. GSCs also maintained genomic stability for up to 35 passages in vitro when they are cultured in neural stem cell media containing epidermal growth factor (EGF) and fibroblast growth factor (FGF). Culturing GSCs in serum containing media such as standard immortalized glioma cell line media (DMEM with 10% foetal bovine serum), however, results in genomic instability, aneuploidy and significantly different gene expression profiles compared to the parental glioblastoma tumors from which they were derived [14]. Furthermore, GSCs more accurately recapitulate clinical aspects of human glioblastomas such as white matter migration/invasion, vascular proliferation, necrosis and cellular heterogeneity, features that are absent in immortalized glioma cell lines such as U87 and U251 [14]. Therefore, GSCs are a more representative model of the actual patient disease compared to the long established immortalized glioma cell lines.
In astrocytes, malignant transformation is associated with aberrant accumulation of repressive histone modifications, such as tri-methylated histone 3 lysine 27 (H3K27me3) mediated by histone methyltransferases such as EZH2, resulting in a histone methylation mediated block to differentiation [1,16]. The relevance, however, of histone demethylases in cancer remains less clear. We report that JMJD3, a H3K27me3 specific demethylase, is induced during differentiation of patient-derived glioblastoma stem cells (GSCs), where it promotes a differentiationlike phenotype via a chromatin dependent (INK4A/ARF locus activation) and chromatin independent (direct p53 protein stabilization) mechanism resulting in a p53-mediated cell cycle arrest and differentiation. We demonstrate that a subset of GBMs have somatic mutations of JMJD3 or down regulation of its mRNA expression secondary to DNA hypermethylation of an intragenic regulatory element. JMJD3 re-activation via regulatory element demethylation results in p53-mediated differentiation of GSCs and suppression of tumorgenicity. Our findings demonstrate a tumor suppressor function for a histone demethylase and indicate that deregulation of JMJD3 either through somatic mutations, epigenetic repression of its mRNA expression or NCoR2 mediated repression may contribute to human gliomagenesis via inhibition of the p53 pathway resulting in a block to terminal differentiation.

Differentiation of Glioblastoma Stem Cells (GSCs)
After written consent tumor samples were obtained from patients undergoing surgery at the National Institutes of Health (NIH) in accordance with the surgical procedures of the National Cancer Institute's Institutional Review Board that specifically approved this study. Tissue samples were enzymatically dissociated and cultured as previously described. 6-well plates were coated with poly-ornithine (Sigma P4957) for 1 hr at 37uC and washed 3 times with PBS. Cells were plated at 5E 5 cells per well (6-well plate) and treated with 2 mM all trans-retinoic acid, RA (Sigma R2625) for 5 days. Media was changed approximately every 2-3 days to maintain a constant dose of RA. On day 5, RNA was collected using Qiagen RNeasy Kit (74106). Cells were lysed directly on the 6-well plates and total RNA was isolated following the manufacturers protocol. All experiments were done in triplicates. Error bars represent means 6SD.

Quantitative Real Time PCR (RT-qPCR)
500 ng of high-quality RNA from the RA treatments was reversed transcribed using the applied biosystems taqman reverse transcription reagents (N8080234) according to manufacturer's protocol. RT-qPCR was performed using reagents from applied biosystems taqman universal PCR mix (4304437) and 20 nM of FAM-tagged gene specific primers on the 7900HT real time PCR system. Primers from applied biosystems were as follows: human p16/p14 (Primers span INK4A and ARF shared Exon 2-Exon 3; Hs00923894_m1), human and mouse GFAP (Hs 00909236_m1, Mm01253033_m1), human p21 (Hs 00355782_m1), human and mouse JMJD3 (Primers span Exon1-Exon2; Hs 00389738_m1, Mm 01332680_m1), human TP53 (Hs00153349_m1). Relative gene expression was determined from the difference in Ct values between specific genes and endogenous GAPDH control (DCt).
Fold change is represented as 2 2DDCt . All experiments were done in triplicates. Error bars represent means 6SD.

Luciferase Reporter Assay
The human GFAP promoter luciferase was from Albee Messing [17]. The episomal human p21 promoter luciferase reporter _ENREF_41was from addgene (Plasmid 16451) courtesy of Bert Vogelstein [18]. A p53 binding site mutant p21 promoter luciferase reporter was generated by site directed mutagenesis using QuickChange XL site directed mutagenesis kit (Stratagene 200516) with the following primers: 59-TTT CTG GCC GTC AGG AAA ATC TCC CAA GAT TTT G-39 and 59-CAA AAT CTT GGG AGA TTT TCC TGA CGG CCA GAA A-39 targeting the distal 59 TP53 binding site targeting a site previously described to be critical for p53 activation of the p21 promoter [19]. Transfections were done with 1 mg of JMJD3 expression construct, 1 mg of luciferase constructs and 100 ng of Renilla firefly luciferase construct pRL-TK (Promega E2241) for background normalization. Cells were lysed at 24-48 hr for luciferase assay using the promega dual luciferase reporter system (E1910) according to the manufacturer's protocol. All experiments were done in triplicates. Error bars represent means 6SD.

Chromatin Immunoprecipitation (ChIP)
Chromatin immunoprecipitation was performed according to SABiosciences Champion CHIP-PCR manufacturer's protocol (SABioscience 334471). Briefly HEK 293 cells grown in DMEM with 10% fetal bovine serum were transfected with either empty vector or JMJD3 using Lipofectamine 2000 (Invitrogen). Cells were cross-linked with 1% formaldehyde and the extracts were enzymatically digested as previously described [20]. Cross-linked material was immunoprecipitated with 4 mg of antibodies overnight at 4uC and followed by incubation with protein A beads (SABiosciences) for 1 hr at 4uC. After several washes, the complexes were eluted and the cross-linking was reversed according to SABiosciences protocol. Immunoprecipitated DNA was recovered by DNA extraction beads (SABiosciences) and analyzed by SYBR green real-time PCR. DCt values were determined by [Ct (target protein)-Ct (IgG negative control)]. Fold changes are represented as 2 2DDCt . The primers used were from SABioscience and corresponded to the p21 transcription start sites (TSS), 1Kb upstream of TSS (21Kb) and the region corresponding to the distal p53 binding site (22 to 23Kb).

Mass Spectrometry Analysis
To detect JMJD3 interacting proteins, empty vector (p3xFLAG-CMV, Sigma) and JMJD3-Flag were transfected into HEK293 cells using Lipofectamine 2000 (Invitrogen). 2 days following transfection, cells were lysed with 1x RIPA lysis buffer (Pierce). Coimmunoprecipitation was performed on approximately 1mg of protein from each sample using 10 mg of anti-flag antibody. The protein complexes were analyzed on a 6% TBE non-denaturing polyacrylamide gel with a PageRuler Plus pre-stained ladder (Thermo scientific) as marker. Gels were fixed in 50% methanol/ distilled water and stained with coomassie blue (GelCode Blue Stain, Pierce) at room temperature for 1 hour. The Coommassie Blue stained gel bands underwent tryptic digestion to extract the peptides for MS analysis [21]. Briefly, each digested peptide sample was loaded on an Agilent 1200 nano-capillary HPLC system (Agilent Technologies) with a 10 cm integrated mRPLCelectrospray ionization (ESI) emitter columns (made in-house), coupled online with a LTQ XP mass spectrometer (Thermo Fisher Scientific) for mRPLC-MS/MS analysis. Peptides were eluted using a linear gradient of 2% mobile phase B (acetonitrile with 0.1% formiac acid) to 42% mobile phase B within 40 min at a constant flow rate of 0.25 mL/min. The seven most intense molecular ions in the MS scan were sequentially selected for collision-induced dissociation (CID) using a normalized collision energy of 35%. The mass spectra were acquired at the mass range of m/z 350-1800. The ion source capillary voltage and temperature were set at 1.7 kV and 200uC, respectively. The MS/MS data were searched against p53 protein database using SEQUEST. The cut-off for legitimate identifications were: charge state dependent cross correlation (X corr ) $2.0 for

Methylation Specific Sequencing
Methylation analysis of the intragenic regulatory area of JMJD3 was performed on genomic DNA extracted from GSC 827 treated 5 mm 5-Azacytidine for 4 days and from untreated control GSC 827. Bisulfite conversion of DNA samples was performed using EpiTect Bisulfite Kit (Qiagen), following manufacturer's instructions. For amplification of the CpG islands, the following primers where used: 39JMJD3_F:TTTGTTTTTTAT-TAATTTGTGTTTTT 39 JMJD3_R: TCCTACAAC-TAAAACTTCCACCTAC. PCR products where then subcloned using the TOPO TA Cloning Kit (Invitrogen, Carlsbad, CA) following manufacturer's instructions. PCR verified 24-32 clones where then sequenced using the vectors T3 priming sites.

JMJD3 Locus Annotation
JMJD3 genomic annotation was done using the The UCSC human genome browser NCB136/Hg18 at (www.genome.ucsc. edu). The JMJD3 regulatory domain was annotated using Open Regulatory Annotation Consortium (www.oreganno.org/ oregano).

Intracranial Tumor Cell Injection into SCID Mice
A plasmid, hg01508s1, containing the human JMJD3 cDNA (KIAA0346) was obtained from the cDNA bank section of Kazusa DNA Research Institute, Japan. The JMJD3 open reading frame (ORF) was PCR-amplified and cloned into a pLenti4/TO/V5-DEST Gateway vector (Invitrogen). Viruses were generated and purified using Invitrogen's viraPower lentiviral expression system. An intracranial orthotropic model was utilized for evaluation of gliobastoma stem cell tumorgenicity as previously described [20]. Briefly, empty vector and JMJD3 infected GSC 827 were dissociated and resuspended in 2 ml of HBSS. 1E 5 GSCs, as determined by trypan blue dye exclusion, were injected stereotactically into the lateral ventricles of neonatal SCID mice. This study was carried out in strict accordance with the recommendations from the National Cancer Institute Animal Care and Use Committee who approved the study. Animals were euthanized by carbon dioxide inhalation. All efforts were made to minimize suffering.

JMJD3 is Induced during Differentiation of GSCs
It has been shown that retinoic acid (RA) induces partial cell cycle arrest and differentiation of primary glioblastoma stem cells (GSCs) thereby inhibiting their tumorigenic potential [14,23]. Consistent with these reports, RA promoted glial differentiation of GSC 827 as demonstrated by induction of glial fibrillary acid protein (GFAP) and astrocyte-like morphologic change ( Figure 1A). As seen in normal mouse neural stem cells (MsNSCs) [4], we find that RA induces JMJD3 in GSCs but not in the long-established U87MG glioma cell line ( Figure 1B). This suggests that induction of JMJD3 may not occur in more differentiated cell types but only in cells with stem cell-like properties. We investigated this possibility by comparing RA induction of JMJD3 in the CD15/ SSEA-1 + subpopulation of primary GSC 827 based on our prior report demonstrating that CD15/SSEA-1 + enriches for a population of tumor cells with stem cell-like properties [13]. We purified CD15/SSEA-1 + and CD15/SSEA-1 2 from GSC 827 by magnetic activated cell sorting (MACS) (Table S1) and treated with RA. RA induced JMJD3 and GFAP in the CD15/SSEA-1 + subpopulation, whereas RA did not cause induction of JMJD3 or differentiation in the CD15/SSEA-1 2 population ( Figure S1). These data suggest that JMJD3 may be involved in the induction of differentiation programs in tumor stem cells in much the same way it activates such programs in normal stem cells.
In wild-type INK4A/ARF and wild-type TP53 GSC 923, JMJD3 induced p16/p14 and p21 mRNA expression (Figure 2A) consistent with re-activation of the INK4A/ARF locus via chromatin remodeling as shown in other cell types 6,7 . In INK4A/ARF null GSC 1228 and wild-type INK4A/ARF HEK 293 cells (both with wild-type TP53), however, JMJD3 also induced p21 mRNA but without p16/p14 or p53 induction ( Figure 2B, C) suggesting that JMJD3 mediated p21 induction in these cell types may be independent of both INK4A/ARF and TP53 transcriptional activation. We also find that p21 is induced during RA and serum mediated differentiation of GSCs ( Figure S3), while exogenous p21 overexpression in GSCs is associated with GFAP expression ( Figure  S3). Moreover, JMJD3 mediated p21 protein expression is associated with GFAP protein induction in GSCs only ( Figure 2D). Therefore, our results suggest that induction of JMJD3 during differentiation of GSCs may promote cell cycle arrest and partial glial differentiation through p21 activation, independently of INK4A/ARF or TP53 transcriptional control.
JMJD3 Induction of p21 is p53-dependent We next investigated how JMJD3 could activate p21 independently of transcriptional up-regulation of p14 (ARF) or p53 (TP53). JMJD3 overexpression in HEK 293 cells did not affect total p53 levels (consistent with a lack of INK4A/ARF or TP53 transcriptional activation) but promoted its nuclear accumulation ( Figure 3A, B). ChIP analysis of the endogenous human p21 promoter ( Figure 3C) shows that JMJD3 overexpression results in accumulation of RNA polymerase II ( Figure 3D) and p53 ( Figure 3E), but not JMJD3 at the p21 promoter ( Figure 3F). Moreover, mutation of the p53 binding site on an episomal nonchromatin based human p21 promoter luciferase reporter significantly diminished p21 promoter activation by JMJD3 in HEK 293, mouse neural stem cells and GSC 923 ( Figure 4A-D). These data indicate that increased p53 nuclear localization and subsequent accumulation at the p21 promoter mediates chromatin independent JMJD3 induction of p21.
We then explored a potential non-chromatin modifying mechanism responsible for JMJD3-mediated p53 pathway activation. Recent evidence suggests that histone lysine modifiers may target non-histone substrates. For example, histone methyltransferases G9a, SET7 and SET9 induce the methylation of many non-histone substrates [25] including p53. Additionally, the H3K4me3 demethylase LSD1 was shown to demethylate lysine 370 on the p53 c-terminus, resulting in repression of p53 stability and function [26]. Co-immunoprecipitation experiments demonstrate that JMJD3 interacts with endogenous p53 in both HEK  Figure 5C). Further studies revealed that the c-terminal jumonji domain (JmjC) of JMJD3 is required for this interaction as a JmjC-deletion mutant JMJD3 does not interact with p53. Given its role as a histone lysine demethylase, we asked if JMJD3 induces p21 via direct lysine demethylation of p53. A previous publication demonstrated that the interaction between JMJD3 and p53 results in demethylation of lysine residues on p53 [27]. Specific lysine residues, however, were not assessed or characterized as potential JMJD3 targets on p53. Thus, given the impact of JMJD3 over-expression on differentiation in INK4A/ARF null but wild-type TP53 GSCs, we sought to characterize specific lysine targets of JMJD3 on p53 via methylation profiling with mass spectrometry (MS). MS, however, was unable to detect baseline methylation on p53 protein in control samples. Next, we increased protein amounts as well as enriched for p53 protein from nuclear and cytoplasmic extracts through immunoprecipitation, however, baseline methylation on p53 protein could not be established. We were, therefore, unable to confirm or repudiate previous published evidence of lysine demethylation following JMJD3-p53 interaction [27].
Given a global decrease in lysine methylation on p53 following JMJD3 interaction [27], we looked for putative methylated lysines on p53, that following modification not only promote the stability of p53 but also activate the p21 promoter. It was previously reported that mono-methyl p53 lysine 372 (p53K372me1) is an active form of p53 previously shown to mediate p21 activation [28]. Here, we find that JMJD3 overexpression promoted nuclear retention of a p53K372me1, more than doubling the ratio of nuclear to cytoplasmic p53K372me1 ( Figure 5D). Given that our MS experiments did not provide direct evidence of active p53 demethylation, the mechanism for enhanced nuclear localization of p53K372me1 remains unresolved. Therefore, we propose that p53 lysine 372 demethylation may be one of several mechanisms contributing to p53 nuclear retention during JMJD3-mediated differentiation of glioblastoma stem cells.

JMJD3 (KDM6B) Locus Harbor Aberrations in GBM
To assess the potential relevance of JMJD3 in human gliomagenesis, we looked for naturally occurring alterations within the JMJD3 locus that could affect JMJD3 activity or expression in GBMs. First, we queried whole exome sequence data from The Cancer Genome Atlas (TCGA) [29] as well as sequence data from primary GSCs derived in the Neuro-Oncology Branch at the National Institutes of Health. We found at least 3 different missense mutations and insertions within the TCGA database, (Table S3). Likewise, 3 of our 5 whole exome sequenced GSCs: GSC 308, GSC 1228 and GSC 827, also harbor missense mutations within the JMJD3 locus (Table S3).
To look for epigenetic changes in primary GBMs that could affect JMJD3 expression and function, whole genome DNA methylation arrays were performed on a series of parental glioblastoma samples (GBM-P, n = 7) which demonstrated that compared to non-tumor brain tissue, all GBM samples show relative hypermethylation of a CpG site (Illumina cg09911083) within the JMJD3 intragenic regulatory element ( Figure 6A; Chr.17p13.1:7,695,644-7,696,309; UCSC NCBI36/Hg18 [30][31][32]). By contrast, the JMJD3 promoter shows no significant hypermethylation in non-tumor brain tissue ( Figure 6A). For example, primary GBM 827 (GBM 827-P) demonstrates highly significant levels of methylation at the JMJD3 regulatory element relative to non-tumor brain tissue and other tumor samples ( Figure 6A; red arrow). JMJD3 expression is also relatively suppressed in GBM 827-P compared to non-tumor brain and other tumors GBM 206-P and GBM 923-P ( Figure S4) while western blot analysis showed lower JMJD3 protein expression in GSC 827 compared to other GSCs ( Figure S4). These results suggest that methylation of the JMJD3 regulatory element may affect JMJD3 expression in GBM 827.

JMJD3 Re-activation Induces a p53-mediated Differentiation of GSCs
The publically available UCSC genome browser contains ChiPsequencing data and whole genome methylation profiles from several cell lines including embryonic stem cells and cancer cell lines. [30][31][32][33]. By analyzing data within the UCSC browser genome browser, we found that cg019911083 is part of a CpG island (Chromosome 17p13.1:7,695,644-7,696,309. UCSC NCBI36/Hg18) previously annotated as a 'positive regulatory element' [34,35] (Open Regulatory Annotation OREG0024145. Chromosome 17p13.1:7,695,692-7,696,642. UCSC NCBI36/ Hg18) [30][31][32][33]. This region also demonstrates enhancer-like features based on published regulatory element properties [30,36] such as evolutionary conservation, DNase hypersensitivity, enhancer-promoter associated histone mark (H3K4me1), RNA polymerase II occupancy and binding by transcription factors (TF) reported to activate JMJD3 expression such as NF-kB [3,[30][31][32][33] and the retinoic acid receptor RXR-A [4,[30][31][32][33]. Therefore, to determine if hypermethylation of this intragenic JMJD3 enhancerlike region, is a regulatory mechanism contributing to suppression of JMJD3 expression in gliomas such as GBM 827-P, we treated GSC 827 with 5-Azacytidine and confirmed JMJD3 regulatory element (but not promoter) demethylation through bisulfite sequencing ( Figure 6B). Regulatory element demethylation reactivates JMJD3 mRNA expression ( Figure 6C) and induces differentiation markers b-TUBULIN III ( Figure 6D) and GFAP ( Figure S5) in GSC 827. Furthermore, siRNA mediated knockdown of JMJD3 following DNA demethylation in GSC 827 inhibited this differentiation ( Figure 6D, Figure S5 and Figure S6 for protein knockdown confirmation). These results suggest that DNA hypermethylation of the JMJD3 regulatory element negatively regulates JMJD3 transcription and differentiation of GSC 827. Next, we assessed if RA mediated JMJD3 induction in GBM GSCs is regulatory element methylation-dependent. We find that RA treatment of GBM 827 GSCs induces JMJD3 without changes to the methylation status of the JMJD3 promoter or downstream regulatory element (Data not shown) suggesting that RA induces JMJD3 via a methylation independent mechanism. In normal neural stem cells (NSCs), it was shown that the RA nuclear corepressor SMRT/NCOR-2 controls the differentiation of neural stem cells into neurons [4] and astrocytes [11]. In GSC 827, we find that RA mediated astrocytic differentiation is associated with a decrease in nuclear SMRT/NCOR-2 protein, JMJD3 induction and p53 nuclear accumulation ( Figure S7). These findings suggest that JMJD3 mediated p53 pathway activation may be suppressed in GBMs via multiple mechanisms.
Wild-type JMJD3 over-expression has been previously shown to promotes p53 nuclear accumulation in normal NSCs [27]. To establish whether regulatory element mediated JMJD3 re-activation induces differentiation of GSCs via p53 protein activation, independently of INK4A/ARF, we assessed p53 nuclear localization following 5-azacytidine treatment of INK4A/ARF null (but wild-type TP53) GSC 827. 5-Azacytidine significantly induced p53 nuclear accumulation with morphological evidence of differentiation in GSC 827 ( Figure 6E-F, Figure S8). SiRNA mediated knockdown of JMJD3 following 5-azacytidine treatment of GSC 827, however, significantly inhibited p53 nuclear accumulation and morphologic evidence of differentiation ( Figure 6F and Figure  S8). These results demonstrate that JMJD3 mediates p53-induced differentiation of GSCs independently of the INK4A/ARF locus.

Relevance of JMJD3 Expression in vivo
To determine if the above in vitro effects of JMJD3 on p53 have any clinical relevance, we asked whether JMJD3 expression in parental GBM samples with wild-type TP53 influences p53 activity within those samples. To investigate this hypothesis, we examined upstream promoter regions (2450/+50 of the Transcription Start Site) of genes differentially expressed between GBM TP53 WT samples with high JMJD3 expression versus low JMJD3 expression. Using the PSCAN motif enrichment algorithm, we found that a consensus binding motif for p53 is significantly enriched in these promoters (fdr , = .05) greater than would be expected by chance. This set of differentially expressed genes showed predominant repression of p53 target genes in high JMJD3 expressing GBMs compared with low JMJD3 expressing GBMs ( Figure 7A). These data are consistent with previous reports that demonstrate that following p53 activation as many as 80% of p53 responsive genes are repressed rather than activated [37,38] and suggest that the p53 pathway is more active in high JMJD3 expressing GBMs retaining wild-type TP53 than in GBMs with wild-type TP53 but low JMJD3 expression. Consistent with these findings, mice with intracranial stereotactically injected JMJD3 over-expressing GSC 827 cells lived dramatically longer than mice injected intracranially with parental GSC 827 cells ( Figure 7B) further demonstrating the tumor suppressive effects of JMJD3 in vivo.

Discussion
There is increasing evidence that a small subpopulation of cells within human glioblastomas (GBMs) is instilled with tumorinitiating and neural stem cell properties (glioma stem cells, or GSCs). The fact that GSCs can be driven toward terminal differentiation in vitro, but rarely in vivo suggests, at least in part, epigenetic regulation of differentiation. Consistent with this supposition, we previously reported that a small subset of GBMs had GSCs that were temporally and developmentally fixed in a pro-proliferative, non-differentiated state through epigenetic silencing of the bone morphogenic protein receptor 1b [20]. More recently, it was demonstrated that IDH1 mutations promoted aberrant histone demethylase activity resulting in a histone methylation (H3K9me3 and H3K27me3) mediated block to differentiation [1,16]. This was in part due to the inhibition of a-ketoglutarate (a-KG) dependent enzymes such as the H3K9me3 demethylase, KDM4C. Amongst the many unresolved questions, however, is the relevance of H3K27me3 demethylases in GBMs and cancer in general.
In normal neural stem cells, the H3K27 specific demethylase JMJD3 was shown to modulate differentiation by affecting p53 nuclear distribution [27]. The implications of these findings in any cancer model, however, remains unclear. Here, we reveal that the H3K27-specific demethylase, JMJD3, promotes differentiation and suppresses proliferation of GSCs, not only through its wellknown chromatin-dependent activation of the INK4A/ARF locus, but also through p53 protein nuclear stabilization. Based on previous studies demonstrating global lysine demethylation on p53 following interaction with JMJD3 in differentiating neural stem cells [27], our results suggest that lysine 372 modifications may represent one of many putative mechanisms resulting in p53 protein stabilization following interaction with JMJD3 resulting in differentiation and suppression of proliferation in glioblastoma stem cells. We also demonstrate that primary-patient-derived GSCs and primary human GBMs counter the tumor suppressor effects of JMJD3 through hypermethylation of an evolutionarily conserved intragenic and enhancer-like DNA regulatory element within the JMJD3 locus or direct somatic mutations of the JMJD3 gene. Pharmacological demethylation of GSC with hypermethylation-silenced JMJD3 or overexpression of a wildtype JMJD3 in JMJD3 mutated GSCs results in p53 pathway activation and GSC differentiation in vitro and hugely extended survival of animals with intracranial GSC xenografts. The clinical relevance of these findings is reflected by our demonstration through genome wide transcriptional motif analyses in a large number of GBM patients that GBMs with high levels of JMJD3 expression and wildtype TP53 show significantly enhanced p53 pathway activation. Furthermore, our findings describe an alternative mechanism (apart from 2-HG and a-KG) for how a DNA hypermethylation phenotype in tumors like GBMs may functionally modulate the activity/expression of the histone demethylases such as JMJD3 in tumors like GBM 827. Moreover, we speculate that DNA methylation may also affect JMJD3 mRNA processing via transcription independent mechanisms (no impact on total mRNA expression levels). It was demonstrated that intragenic DNA hypermethylation may regulate differential exon inclusion during mRNA elongation [39]. Here, DNA methylation impairs CTCF binding to exon 5 and its inclusion into full length CD45 mRNA. The JMJD3 regulatory element spans exon 17, which encodes for part of the JMJD3 catalytic domain called jumjonji C (JmjC; Exons 14-17), a region previously shown to be absolutely required for the demethylase activity of JMJD3 (22 exons) [40]. Whole genome ChIP-sequence data from embryonic stem cells in the UCSC genome browser online also show that CTCF binds the JMJD3 regulatory element, [30][31][32]. Thus, it is possible that intragenic DNA hypermethylation could also impair JMJD3 activity by modulating CTCF-mediated JmjC demethylase domain inclusion into full length JMJD3 mRNA. Therefore, apart from modulating the activity of methylation sensitive enhancers (as in GSC 827), intragenic hypermethylation in some GBMs could also contribute to suppression of JMJD3 enzymatic activity resulting in an aberrant block to terminal differentiation. If true, the aberrant methylation within the JMJD3 loci may prove to be a useful therapeutic target in cancer.
Methylation independent mechanisms also exist for JMJD3 reactivation in GSCs. It has been shown that during neuronal differentiation SMRT/NCOR-2 directly inhibits JMJD3 transcriptional activation by promoter-mediated repression, resulting in a H3K27-mediated repression of differentiation genes [4]. RA however, relieves this block by inducing the destabilization of nuclear SMRT/NCOR-2 resulting in JMJD3 activation and differentiation of NSCs [4]. SMRT/NCOR-2 has also been previously shown to modulate the differentiation state and tumorigenicity of GSCs [23,41]. Here, RA-mediated destabilization of nuclear SMRT/NCOR-2 [23] or serine/threonine protein phosphatase 2A (PP2A)-mediated inhibition [41] results in differentiation and suppresses proliferation of GSCs. Thus, JMJD3 may be re-activated in GSCs via methylation dependent and independent mechanisms. Methylation and demethylation of p53 has been previously shown to modulate p53 stability and function [26][27][28]42,43]. Others have demonstrated that the interplay between methylation and acetylation of the p53 protein activates p53 during DNA damage response [44]. Also, methylation of specific lysines has been shown to affect the ability of p53 to interact with coactivators involved in p53-mediated apoptosis [26]. Therefore, it is speculated that the interplay between methylation, demethylation and acetylation may fine tune p53 activity during tumor initiation and progression [45]. It remains largely unclear, however, how methylation specifically stabilizes or destabilizes p53. To this end, following our results showing the significant impact of a lysine demethylase on p53 stability in cancer, it will be necessary to perform methylation profiling on p53 to definitively establish whether demethylases such as JMJD3 stabilize p53 through direct demethylation and if not, through what other mechanisms.
In conclusion, the relevance of most histone demethylases, including the H3K27me3 demethylase, JMJD3, in cancer remains unclear. Here, we propose that JMJD3 acts as a tumor suppressor gene in glioblastoma multiforme. We find that JMJD3 interacts with p53 protein in GSCs and JMJD3 re-activation in GSCs induces p53 nuclear accumulation in an INK4A/ARF null background resulting in a p53-mediated differentiation of GSCs (See model in figure 8). These findings point to a role for histone demethylases in cancer and suggest that JMJD3 may serve as a tumor suppressor gene and a potential therapeutic target in glioblastoma multiforme.  All experiments were done in triplicates. Error bars represent means 6SD. p = two-tailed Student's t test comparing indicated samples to non-tumor brain tissue, *p,0.1, **p,0.05, ***p,0.01, N.S.-not significant. (JPG) Figure S5 JMJD3 knockdown inhibits differentiation of GSC 827. Effect of 5-Azacytidine on GFAP mRNA expression with and without JMJD3 knockdown. All experiments were done in triplicates. Error bars represent means 6SD. p = two-tailed Student's t test comparing indicated samples, *p,0.1, **p,0.05, ***p,0.01, N.S., not significant.