Epigenetic Regulation of Survivin by Bmi1 Is Cell Type Specific During Corticogenesis and in Gliomas§

Authors


  • Author contributions: S.A. and A.G.: collection and/or assembly of data, data analysis and interpretation, and manuscript writing; D.L. and E.S.: data analysis and interpretation; H.B., D.C., and Z.R.: collection and/or assembly of data; J.G. and M.S.-R.: provision of study material or patients; N.V.H. and S.B.: provision of study material or patients and data analysis and interpretation; S.M.: conception and design, data analysis and interpretation, financial support, and manuscript writing. S.A. and A.G. contributed equally to this article.

  • Disclosure of potential conflicts of interest is found at the end of this article.

  • §

    First published online in STEM CELLSEXPRESS November 6, 2012.

Abstract

Polycomb group proteins are essential regulators of stem cell function during embryonic development and in adult tissue homeostasis. Bmi1, a key component of the Polycomb Repressive Complex 1, is highly expressed in undifferentiated neural stem cells (NSC) as well as in several human cancers including high-grade gliomas—highly aggressive brain tumors. Using a conditional gene activation approach in mice, we show that overexpression of Bmi1 induces repressive epigenetic regulation of the promoter of Survivin, a well-characterized antiapoptotic protein. This phenomenon is cell type-specific and it leads to apoptotic death of progenitor cells exclusively upon commitment toward a neuronal fate. Moreover, we show that this is triggered by increased oxidative stress-induced DNA damage. In contrast, undifferentiated NSC as well as glioma-initiating cells display an open chromatin configuration at the Survivin promoter and do not undergo apoptotic death. These findings raise the possibility that normal and neoplastic stem cells depend on the same mechanism for surviving the hyperproliferative state induced by increased Bmi1 expression. STEM Cells2013;31:190–202

INTRODUCTION

During embryonic development, epigenetic mechanisms control transcriptome complexity of stem cells acting via chromatin modification, thereby significantly contributing to fate specification of their progeny. Polycomb group (PcG) proteins are repressive chromatin modifiers required for regulating stem cell function. They form multiprotein chromatin-associated complexes that play an essential role as transcriptional repressors through post-translational modifications of histones. The two best characterized PcG complexes are Polycomb repressive complex (PRC) 1 and PRC2. The core of the PRC2 comprises three proteins, Suz12, Eed, and Ezh2, the latter being the catalytic subunit that modifies histone H3 by trimethylation of lysine 27 (H3K27me3), a typical epigenetic silencing mark. PRC1 depends upon PRC2 for recruitment to PcG target genes and is responsible for monoubiquitylation of histone H2A at lysine 119, an activity dependent on the E3 ubiquitin ligase activity of Ring1B and enhanced by Bmi1. This sequence of events (recently reviewed in [1]) induces chromatin compaction and represses the transcription of a number of key genes, such as the Ink4a locus encoding the cell cycle regulators p16Ink4a and p19Arf [2]. However, to allow commitment and differentiation into different lineages, the regulatory system must retain plasticity to rapidly activate the expression of specific genes in response to developmental and environmental cues. This is mainly achieved through demethylation of H3K27Me3 to H3K27Me1 by H3K27 demethylases, namely UTX and JmjD3 [3–5].

The roles of several PcG genes during the development of the mammalian central nervous system and in the maintenance of neural stem cells (NSC) have been recently characterized. Conditional ablation of Ezh2 during embryonic cortical development in the mouse results in reduction of trimethylation of H3K27 and leads to a shift in the balance between self-renewal and differentiation toward the latter. It also causes narrowing of the neurogenic period [6] or impaired transition from a neurogenic to an astrogenic phase [7] depending on the developmental time point at which Ezh2 was ablated. Conversely, conditional inactivation of Ring1B has been reported to significantly enhance neuronal differentiation during late embryonic development due to a prolonged neurogenic phase, as determined by extended expression of Neurogenin1 [7]. Loss of function studies in the mouse has shown that Bmi1 is essential for regulation of cell cycle entry of neural progenitors and for maintenance of self-renewal of NSC [2, 8, 9]. Taken together, these studies suggest a distinct temporal pattern of PcG-mediated gene repression which plays an essential role in fate specification.

Modulation of the expression of the PcG proteins also plays a role in cancer. Bmi1 is highly expressed in several types of human tumors [10] including gliomas [11], the most common intrinsic brain tumors. High-grade gliomas (HGG) are characterized by poor prognosis and high mortality due to their highly invasive nature and resistance to conventional anticancer treatments. High Bmi1 expression has also been found in glioblastoma-initiating cells (GBM-IC) [12]. Interestingly, knockdown of Bmi1 in human GBM-IC significantly reduces tumor growth in a xenograft mouse model [13]. In keeping with these data, gliomas originating in mice deficient for p16Ink4a/p19arf show increased tumor latency and a shift toward a lower histological grade when Bmi1 is also ablated [14].

We recently reported that overexpression of Bmi1 in NSC induced increased proliferation of uncommitted progenitors during corticogenesis. However, upon commitment toward a neuronal fate, transgenic neural progenitor cells (NPC) underwent apoptosis which was triggered by the increased and ectopic proliferation of these cells. Reduced expression of Survivin was seen in these cells in the developing neocortex [15].

Survivin is a member of the inhibitor of apoptosis (IAP) protein family. It is ubiquitously expressed during embryonic development but only minimally expressed in adult tissues. It is highly expressed in most human cancers, including HGG [16], where its ability to counteract apoptotic stimuli leads to cell survival and significantly contributes to their resistance to current anticancer therapies, such as radiotherapy and chemotherapy [17].

Here, we investigate the molecular and functional basis of the differential impact of Bmi1 overexpression on the biological properties of NSC and NPC at various developmental stages. We analyze whether epigenetic regulation of the promoter of Survivin determines the outcome of Bmi1-induced hyperproliferative state. Finally, we compare the epigenetic regulation of target genes induced by Bmi1 overexpression in NSC and GBM-IC.

MATERIALS AND METHODS

See supporting information.

RESULTS

Bmi1 Represses Survivin Expression in Neurospheres But Not in Undifferentiated NSC by Increased H3K27me3 and Direct Binding

We have previously shown in a conditional transgenic mouse model (Nestin-Cre;STOP FloxBmi1) that overexpression of Bmi1 in NSC and NPC increases their proliferation in the developing neocortex. In addition, upon commitment toward a neuronal lineage, overexpression of Bmi1 triggered apoptosis in NPC, leading eventually to a reduced overall brain size [15]. These findings could be faithfully reproduced in a neurosphere (NS) assay in vitro [18], where both, increased proliferation of NSC and NPC and induction of apoptosis of neuronal progenitors, were detected. However, when NSC were cultured in monolayer (NSCmon), a condition that allows for preservation of their undifferentiated state [15, 19], increased proliferation but no significant apoptotic death was noted, in keeping with the observation in radial glia cells in vivo. Because downregulation of p16Ink4a and p19Arf was seen in both cellular contexts, we hypothesized that different molecular targets may be involved in mediating this dual role of Bmi1 overexpression on progenitors cells at different commitment stages. In our previous studies, we have described a striking downregulation of the expression of Survivin in NPC overexpressing Bmi1 upon neuronal commitment both in vivo and in NS [15] (supporting information Fig. S1).

To further understand the mechanism underlying this dual role of Bmi1 overexpression in different cellular contexts, we first assessed the subcellular localization of Bmi1 in NS overexpressing Bmi1. The in vitro system was chosen because it closely recapitulates the in vivo phenotype and at the same time allowed us to obtain sufficient material for the analysis. NSC were isolated from Nestin-Cre;STOPFloxBmi1 E16.5 neocortices, cultured under NS-inducing conditions for 2 to 3 passages, and processed for cellular fractionation analysis. We show that, in the context of mild Bmi1 overexpression (Fig. 1A), the majority of the protein is found in the chromatin fraction with a minor nonchromatin nuclear component while barely any protein was detected in the cytoplasm (Fig. 1B and densitometry for the chromatin fraction 1C). These data suggest that overexpression of Bmi1 may still elicit its role as an epigenetic chromatin modifier.

Figure 1.

Bmi1 represses Survivin expression in neurospheres (NS) by increased H3K27me3 and direct binding. Overexpression of Bmi1 in NS isolated from NestinCre;STOPFloxBmi1 as assessed by Western blot (A). In the context of mild Bmi1 overexpression, the majority of the protein is found in the chromatin fraction with a minor nonchromatin nuclear component while barely any protein is detected in the cytoplasm (B and densitometry for the chromatin fraction C). Schematic representation of the promoter region of survivin. PS1, PS2, and PS4 indicate location of primer pairs. (D). qChIP assay of the Survivin promoter region in NS overexpressing Bmi1 as compared to controls. Data represent means ± SD (n = 3). The antibodies used in qChIP are indicated at the top of each panel; IgG was used as a control. (E--G): Schematic representation of the p16INK4a promoter. PSink4a indicates the location of primer pair (H). qChIP assay of the p16INK4a promoter region in NS overexpressing Bmi1 as compared to controls. Data represent means ± SD (n = 3). The antibodies used in qChIP are indicated at the top of the panel; IgG was used as a control. (I, J): The binding activity of each protein is given as percentage of total input. *, p < .05; **, p < .01 with Student's t test, Error bars represent SD, n ≥ 3. Abbreviations: qChip, quantitative chromatin immunoprecipitation; TSS, transcription start site.

Next, we asked whether Bmi1 would directly bind to the Survivin promoter. Eight pairs of primers were designed that span the promoter region of Survivin (approximately 3 kb) and quantitative chromatin immunoprecipitation (qChip) assays were performed on NS overexpressing Bmi1 and controls with an antibody against Bmi1. We found that Bmi1 specifically bound to the Survivin promoter region in NS and the most significant enhancement of Bmi1 binding activity was detected from −996 bp to −700 bp and from −1,730 bp to −1,387 bp, respectively, relative to the transcription start site (+1) of the Survivin promoter (Fig. 1D, 1E).

Then we set out to investigate whether overexpression of Bmi1 could repress the expression of Survivin in NS by post-translational modification of histones. Methylation of histone lysine residues is a critical determinant of active and silent gene expression state. H3K27me3 is an epigenetic silencing mark, whereas H3K4me3 is associated with active transcription. To confirm the essential role of the PRCs in Bmi1-mediated Survivin repression, qChIP assays were performed with antibodies specific for H3K27me3 and H3K4me3. Increased H3K27me3 levels were found from −2,438 bp to −1,993 bp and from −996 bp to −700 bp relative to transcription start site in NS overexpressing Bmi1, whereas either no significant level of H3K4me3 or a reduction in the levels of H3K4me3 was observed at these loci, respectively (Fig. 1F, 1G). When PRC-mediated suppression of p16INK4a was used as control for these experiments, qChIP showed that there is increased Bmi1-binding and H3K27me3 levels at the promoter of p16INK4a in NS overexpressing Bmi1, compared with the control, in keeping with published data [20] (Fig. 1H-1J).

Interestingly, when the H3 methylation pattern at the promoter region of Survivin was investigated in NSCmon, increased trimethylation of H3K4 but not of H3K27 was noted upon Bmi1 overexpression (Fig. 2A, 2B, 2E), a finding accompanied by no evidence of Bmi1 binding to the promoter of Survivin (supporting information Fig. S2). In keeping with these data, immunolabeling of NSCmon revealed increased Survivin expression in cultures overexpressing Bmi1 (Fig. 2C, 2D). However, while a differential pattern of post-translational modification of histones was found on the promoter of Survivin in different cellular contexts, a similar pattern of high H3K27me3 levels was detected on the promoter of p16INK4a both in NSC (Fig. 2F, 2G) and in NS (Fig. 1J).

Figure 2.

Coexpression of Bmi1 and Survivin in neural stem cells (NSC) cultured as monolayer. Schematic representation of the promoter region of Survivin. PS1 and PS4 indicate location of primer pairs. (A). qChIP assay of the Survivin promoter region in NSC overexpressing Bmi1 cultured as monolayer as compared to controls. Data represent means ± SD (n = 3). The antibodies used in qChIP are indicated at the top of each panel; IgG was used as a control (B, E, G). Expression of Survivin is increased in NSC overexpressing Bmi1 cultured as undifferentiated monolayer. Scale bar is 125 μm (C, D). Schematic representation of the p16INK4a promoter. PSink4a indicates the location of primer pair (F). qChIP assay of the p16INK4a promoter region in NSC overexpressing Bmi1 cultured as monolayer as compared to controls. Data represent means ± SD (n = 3). The antibody used in qChIP is indicated at the top of the panel; IgG was used as a control (G). The binding activity of each protein is given as percentage of total input. qRT-PCR analysis of the expression of JmjD3 and Ezh2 in neurospheres overexpressing Bmi1 as compared to control (H,I). *, p < .005; **, p < .01 with Student's t test, error bars represent SD, n ≥ 3. Abbreviations: qChip, quantitative chromatin immunoprecipitation; TSS, transcription start site.

During cellular differentiation, the ratio between permissive and repressive epigenetic modifications is maintained or swiftly changed to create cell-type-specific gene expression patterns. Histone methylases and demethylases play an important role in this regulation. We reasoned that the most likely explanation for the changes in H3K27me3 levels would be either upregulation of methyltransferase EZH2 [21] or downregulation in the H3K27me3 demethylase JmjD3 [3, 22]. Expression of these genes was assessed by qRT-PCR. Decreased JmjD3 levels were observed in NS overexpressing Bmi1, while no significant changes in the expression of Ezh2 were observed (Fig. 2H, 2I). No significant changes in the expression of both JmjD3 and Ezh2 were found in NSCmon overexpressing Bmi1 as compared to controls (Fig. 2H, 2I). These data suggest that increased H3K27me3 at the Survivin promoter in NS overexpressing Bmi1 may be responsible for its transcriptional repression and they raise the possibility that this is due to downregulation of the expression of JmjD3.

Transcriptome Analysis of NS Overexpressing Bmi1

In order to gain insight into genes and pathways modulated by Bmi1 overexpression which may contribute to explain its differential impact on the cellular properties of NSC and NPC, transcriptome-wide analysis of gene expression was carried out. We isolated NSC from Nestin-Cre;STOPFloxBmi1 E16.5 neocortices and cultured them under NS-inducing conditions for two to three passages to enrich for NS-forming cells. We then analyzed their transcriptome by whole-genome Illumina platform mouse v2 and compared it to the transcriptome of NS arising from NSC isolated from nontransgenic or single transgenic littermates. Hierarchical clustering of the expression data revealed that the biological replicas were grouped according to their genotype (Fig. 3A). To define differentially expressed genes, we applied a filter consistent of statistical significance (p < .05) and a fold expression change greater than 1.5. A set of 52 genes was identified, the expression of which was deregulated in NS overexpressing Bmi1 compared to controls (supporting information Fig.S3A). The array data were validated by SYBR GREEN real-time reverse transcription PCR on independently derived NS cultures. From eight genes (Cdt, Gfap, Kiff22, Mbp, Mt3, Plk1, Sepx, and Sirt2) that were selected for RT-PCR validation on the basis of the potential interest and involvement in the observed phenotype, all were confirmed to be differentially expressed (Fig. 3B).

Figure 3.

Transcriptome analysis of neurospheres (NS) overexpressing Bmi1. Heat map representing clusters of genes differentially expressed in NestinCre;STOPFloxBmi1 NS as compared to controls, n = 2 (A). qRT-PCR analysis to validate a selection of candidate genes identified as deregulated in NestinCre;STOPFloxBmi1 NS as compared to controls in the genome-wide screening (B). qRT-PCR arrays (SA Bioscience) of genes involved in redox homeostasis and stem cell maintenance reveal significant differences between NestinCre;STOPFloxBmi1 NS and controls (C). System biology analysis—Ingenuity platform—performed on all genes differentially expressed upon Bmi1 overexpression in NSs (D). qRT-PCR analysis of gene expression in NestinCre;STOPFloxBmi1 NSCmon as compared to controls (E). *, p < .05; **, p < 0 .01; ***, p < .001 with Student's t test, Error bars represent SD, n ≥ 3, unless stated otherwise.

Gene ontology analysis revealed that the 52 deregulated genes belong to categories such as glial cell development, myelination, ensheathment of neurons, gliogenesis, neurogenesis as well as mitotic cell cycle, selenium binding, cellular and redox homeostasis, and microtubule cytoskeleton (supporting information Fig. S3B). Deregulation of genes involved in terminal differentiation and lineage specification was consistent with delayed differentiation induced by Bmi1 overexpression in NS as reported previously [15]. Of particular interest was the potential deregulation of genes involved in redox homeostasis and stem cell maintenance. To investigate this further, qRT-PCR arrays (SA Bioscience, West Sussex, UK) were carried out on four biological replicas of transgenic NS and controls. This analysis allowed us to identify a further set of 59 genes which were differentially expressed between the two genotypes (data not shown), some of these genes are listed in supporting information Figure S3C and the most interesting ones are shown in Figure 3C. Among these, significant downregulation of Alox15, Duox1 as well as upregulation of Duox2 and Bnip3l was observed in NS overexpressing Bmi1 (Fig. 3C). System biology analysis using an Ingenuity platform was performed on all genes that were differentially expressed upon Bmi1 overexpression. Interestingly, Notch signaling, molecular mechanisms of cancer, Wnt/β-catenin, and G2/M DNA damage checkpoint regulation were listed among the most significantly deregulated canonical pathways, while G2/M DNA damage checkpoint regulation and cell death were also identified among the top tox lists (Fig. 3D).

Next we assessed whether NSCmon overexpressing Bmi1 would also show deregulation of these genes. Interestingly, genes involved in cell death, DNA damage, and oxidative stress such as, for example, Duox1 and Alox5 were downregulated or not significantly altered in these cultures while the expression of Survivin was confirmed significantly enhanced also at RNA level (Fig. 3E). These findings show that Bmi1 overexpression has a different impact on transcriptome-wide gene expression depending on the developmental stage of the cells analyzed, raising the possibility that the cellular processes regulated by these genes may also be differentially affected.

Overexpression of Bmi1 Causes Oxidative Stress-Induced DNA Damage in NS But Not in Undifferentiated NSC

We tested whether deregulated expression of genes involved in oxidative stress could be due to increased ROS production, possibly triggered by the sustained and/or ectopic proliferation observed upon Bmi1 overexpression. To this end, Reactive oxygen species (ROS) production was measured in NS 4 days after dissociation by incubating them with the fluorescent dye DCFDA followed by flow cytometry. The experiment was carried out on three independently derived E16.5 NestinCre;STOPFloxBmi1 cultures and three controls. We observed variations between independently derived cultures but a significant increase in ROS production was observed in Bmi1 overexpressing NS (1.5-fold change, related to control, Fig. 4A). To assess whether this led to increased oxidative damage to the cells contained within NS, cryosectioned NS were analyzed for the presence of 8-hydroxydeoxyguanosine (8-OH dG), a predominant form of free radical-induced oxidative lesion that is widely used as a marker of oxidative stress dependent DNA damage [23]. Quantification of the percentage of positive cells among all cells contained within a high-power field confirmed increased number of 8-OH dG + cells upon overexpression of Bmi1 (Fig. 4B, 4C, and quantification 4F). This finding was confirmed by immunolabeling with γ-H2AX, another marker of DNA damage [24], which colocalized with 8-OH dG, although 8-OH dG+ and γ-H2AX− cells also exist in these cultures (Fig. 4B inset).

Figure 4.

Oxidative stress-induced DNA damage in neurospheres (NS) but not in undifferentiated neural stem cells (NSC) upon overexpression of Bmi1. Increased ROS production in NSC/NPC contained within NS and in NSC cultured as monolayer (A). Increased number of 8-OHdG + cells in NS overexpressing Bmi1 (B, C) and colocalization of staining for 8-OHdG and γ-H2AX in a majority of cells (inset). The majority of cells showing evidence of DNA damage (γ-H2AX+) are neuronal progenitors (MAP+) in cultures overexpressing Bmi1 (D, E). Quantification of the findings is shown (F). PKH26+ NSC do not show evidence of DNA damage in NS (G, H) and the number of γ-H2AX+ cells is not increased in NSC overexpressing Bmi1 cultured as monolayer (I, J). Cells expressing Survivin do not show evidence of DNA damage in NS (K, L). Increased number of cells showing evidence of DNA damage (γ-H2AX+) is seen also in vivo in E16.5 cortices (M, N, and quantification O). Western blot analysis revealed increased pan-ubiquitylation (P) and increased ubiquitylation of H2AX (Q) but not of H2A (P) in NS overexpressing Bmi1. Increased phosphorylation of H2AX is also seen (Q). *, p < .05; **, p < .01; ***, p < 0 .001 with Student's t test. Error bars represent SD, n ≥ 3. Scale bar is 125 μm in (B, C), (G--N) and 50 μm in (D, E). Abbreviation: DAPI, 4′,6′-diamino-2-phenylindole.

We previously noticed that Bmi1 overexpression led to apoptosis of NPC exclusively when they committed toward a neuronal fate. We therefore assessed whether oxidative stress-induced DNA damage would specifically affect these cells. Taking advantage of the spontaneous differentiation that occurs in NS, we double labeled NS with γ-H2AX and MAP2, a marker of early neuronal differentiation. This staining revealed that indeed cells showing evidence of DNA damage were progenitor cells committed toward neuronal lineage (Fig. 4D, 4E, and quantification 4F). Accordingly, no increase in the number of 8-OH dG+ cells was seen among progenitors committed toward glial lineage such as A2B5+ cells (data not shown and quantification Fig. 4F).

To test whether undifferentiated stem cells are similarly affected by oxidative stress-induced DNA damage upon Bmi1 overexpression, we labeled NS in culture immediately after dissociation with a fluorescent dye, PKH26, which binds to the cell membrane [25]. Upon cell division, the dye is passed on to the progeny and is henceforth diluted throughout passages to eventually become undetectable after several cell cycles. This method has been shown to effectively mark slowly dividing, noncommitted stem cells in mammospheres [26] and to efficiently label NSC upon transplantation in vivo [27]. We cryosectioned NS two passages after labeling and staining against 8-OH dG revealed no evidence of oxidative stress-induced DNA damage in undifferentiated PKH26+ stem cells (Fig. 4G, 4H). Similarly, an equal number of γ-H2AX+ or 8-OH dG+ cells was detected in NSCmon in both transgenic and control cultures (Fig. 4I, 4J and data not shown). Interestingly, however, upon overexpression of Bmi1 increased ROS production was detected also in these cultures (Fig. 4A).

Survivin is an IAP which influences cell division through its action on the mitotic spindle and microtubule formation, recently reviewed in [28]. Oxidative stress and DNA damage induce expression of Survivin, which in turn plays a key role in supporting hyperproliferative conditions, for example, in preneoplastic lesions such as pterygium [29] and in cancer (reviewed in [30]). We therefore assessed how expression of Survivin and evidence of oxidative stress-induced DNA damage would relate to each other in NS. Double immunolabeling for Survivin and 8-OHdG revealed little coexpression of the two proteins in NS (Fig. 4K, 4L) with Survivin being expressed only in a small fraction of cells. On the contrary, colocalization of Survivin and 8-OHdG was readily detectable in the NSCmon (supporting information Fig. S4).

To see whether these in vitro findings reflected the in vivo situation, we analyzed whether evidence of increased DNA damage could be found in the developing neocortex of mice overexpressing Bmi1. Immunostaining for 8-OH dG on tissue sections was technically unsatisfactory, and immunostaining for γ-H2AX was performed instead. Quantification of the number of cells with positive foci showed increased number of γ-H2AX+ cells in NestinCre;STOPFloxBmi1 cortices at E16.5 (Fig. 4M, 4N, and quantification 4O).

In summary, evidence of increased ROS production and oxidative stress-induced DNA damage is found in both NSC and NPC upon overexpression of Bmi1 during embryonic development both in vitro and in vivo. However, apoptotic death was found only in NPC overexpressing Bmi1, a finding that correlated with reduced expression of Survivin.

Overexpression of Bmi1 Induces Enhanced Ubiquitylation of Histone H2AX in NS

Phosphorylation of H2AX is one of the initial events occurring at DNA damage loci induced by double-strand breaks (DSBs). It decorates the flanking chromatin leading to recruitment and activation of the DNA damage response (DDR) [31]. In addition to phosphorylation of H2AX, ubiquitylation of this histone subunit is an important epigenetic marker for DNA lesions in DDR [32]. Bmi1 together with Ring1B is a major histone H2A ubiquitin ligase and its role in mediating monoubiquitylation of γ-H2AX has been recently demonstrated [33, 34]. Given the increased number of γ-H2AX+ cells in NS overexpressing Bmi1, we set out to assess whether this would also be accompanied by increased ubiquitylation of histone H2 variants. Histone extraction followed by immunoblotting revealed not only increased pan-ubiquitylation (Fig. 4P) but also increased ubiquitylation on histone H2AX in NS overexpressing Bmi1 (Fig. 4Q), while a reduction in the ubiquitylation of histone H2A was noted (Fig. 4P). Probing with γH2AX confirmed increased phosphorylation of this histone subunit upon Bmi1 overexpression as observed by means of immunolabeling of NS (Fig. 4Q). No upregulation of the expression of RNF8, a RING-finger ubiquitin ligase that has also been recently reported to ubiquitylate histone H2AX, was found in these cultures (data not shown), therefore suggesting that the observed phenotype is directly linked to Bmi1 overexpression.

Bmi1 and Survivin Are Highly Expressed in Mouse and Human GBM-IC and the Epigenetic Regulation of the Survivin Promoter Is Similar to that Observed in NSC

In mouse models, neoplastic transformation of NSC and/or NPC has been shown to be a key pathogenetic event in the formation of at least a subset of high-grade glial tumors [35–37]. Moreover, cells with stem cell properties play a key role in maintaining brain tumors such as gliomas (recently reviewed in [38]) and Bmi1 overexpression in these cells is essential for tumor take after intracerebral transplantation in the mouse [13].

Advanced genomic studies have identified three main pathways in the pathogenesis of HGG RTK/PI3K pathway—where Phosphatase and tensin homolog (PTEN) LOH is the most common mutation, p53 signaling—mainly biallelic inactivation of p53, and Rb-mediated cell cycle control [39]. Indeed, high-grade glial tumors develop in a proportion of p53LoxP/LoxP;PTENLoxP/LoxP mice upon either intraventricular injection of Adeno-Cre [36] or huGFAP driven Cre expression [37]. Here, we set out to investigate whether Bmi1 is overexpressed in mouse models of HGG and whether it has a functional significance in this context. Immunostaining for Bmi1 revealed increased expression in a variable proportion of tumor cells in A-Cre;p53LoxP/LoxP;PTENLoxP/LoxP mice (Fig. 5D-5F) as compared to the weak physiological expression of Bmi1 in the surrounding brain tissue (Fig. 5A-5C). Next, we assessed Bmi1 expression in primary short-term cultures derived from two Adeno-Cre;p53LoxP/LoxP;PTENLoxP/LoxP HGG (mGBM-IC), as described [36]. These cells form NS or show multipotency when transferred to the appropriate medium and are tumorigenic when injected in the caudoputamen of syngenic mice [36]. Comparative analysis of Bmi1 expression as detected by qRT-PCR revealed a very similar expression level in these cells and in NS derived from NestinCre;STOPFloxBmi1 (Fig. 5G). Downregulation of Bmi1 expression in human GBM-IC significantly impairs proliferation in vitro, a finding that correlates with a reduced tumor take and tumor growth in vivo [13]. In agreement with these findings, reduced number of cells was counted in mGBM-IC cultures upon efficient Bmi1 knockdown (Fig. 5H, 5I).

Figure 5.

Bmi1 is overexpressed in a mouse model of HGG. Strong expression of Bmi1 is seen in a variable proportion of tumor cells in full blown HGG (A--F). Note that expression of Bmi1 is clearly higher in the neoplastic cells (D--F) than in the surrounding cerebral cortex (A--C). Scale bar is 125 μm in (A, B, D, E) and 62.5 μm in (C) and (F). qRT-PCR analysis of the expression of Bmi1 in NestinCre;STOPFloxBmi1 NSs and mGBM-IC (G). *, p < .05 with Student's t test. Error bars represent SD, n = 3. Efficient Bmi1 knockdown in mGBM-IC (H) and reduced cell counts upon knockdown in these cells (I). **, p < .01 with Student's t test. Error bars represent SD, n = 3. In mGBM-IC, the majority of Bmi1 protein is found in the chromatin fraction with a minor nuclear component while barely any protein was detected in the cytoplasm (J). qRT-PCR analysis of the expression of Survivin in mGBM-IC as compared to NestinCre;STOPFloxBmi1 NSs (K). *, p < .05; **, p < .01 with Student's t test and Bonferroni multicomparison correction. Error bars represent SD, n = 3. Abbreviations: Ctr Mon, control NSC monolayer; Ctr NS, control neurospheres; HGG, high-grade gliomas; +Ncre Mon, NestinCre;STOPFloxBmi1 NSC monolayer; +Ncre NS, NestinCre;STOPFloxBmi1 neurospheres; mGBM-IC, murine glioblastoma-initiating cells; NS, neurospheres.

Analysis of the subcellular localization of Bmi1 in these cells revealed that Bmi1 was predominantly associated with the chromatin fraction (Fig. 5J) and the results were comparable with those from the same analysis in a non-neoplastic context (Fig. 1A). Next we asked how Bmi1 overexpression would impact on Survivin expression in these cells. We found high expression levels of Survivin in a qRT-PCR assay (Fig. 5K). Moreover, qChIP experiments revealed significantly increased levels of trimethylation H3K4 in these cells while no evidence of the repressive mark H3K27me3 to the Survivin promoter was detected (Fig. 6A). High H3K27me3 and direct binding of Bmi1 to the p16 promoter region were seen in these cells (Fig. 6B), and importantly, they were both reduced and lost, respectively, upon Bmi1 knockdown (supporting information Fig. S6). qRT-PCR analysis of the expression of JmjD3 revealed significant increase of its expression in mGBM-IC upon Bmi1 knockdown (Fig. 6D), in keeping with JmjD3 being negatively regulated by Bmi1.

Figure 6.

Epigenetic modifications of Survivin in murine glioblastoma-initiating cells (mGBM-IC) overexpressing Bmi1. Schematic representation of the promoter region of survivin. PS1, PS2, and PS4 indicate location of primer pairs. (A). qChIP assay of the Survivin promoter region in mGBM-IC. The antibodies used in qChIP are indicated at the top of each panel; IgG was used as a control. Schematic representation of the p16INK4a promoter (B). Psink4a indicates the location of primer pair. qChIP assay of the p16INK4a promoter region in mGBM-IC. The antibody used in qChIP is indicated at the top of the panel; IgG was used as a control. Significant reduction of H3K27me3 in mGBM-IC upon Bmi1 knockdown as compared to si Scr (C). The binding activity of each protein is given as percentage of total input. *, p < .005; **, p < .01 with Student's t test. Error bars represent SD, n ≥ 3. Increased expression of JmjD3 upon Bmi1 knockdown in mGBM-IC (D). *, p < .005 with Student's t test. Error bars represent SD, n ≥ 3. Si Scr: control, si Bmi1: Bmi1 knockdown. Abbreviations: qChip, quantitative chromatin immunoprecipitation; TSS, transcription start site.

To assess whether these findings would mirror the situation in human HGG Bmi1 and Survivin expression was analyzed in a series of primary glioblastoma samples (Fig. 7A-7F) by immunohistochemistry. While mild nuclear staining for Bmi1 was observed throughout the tumors (Fig. 7D), stronger nuclear staining was observed in a proportion of tumor cells located around blood vessels adjacent to ischemic tumor necrosis (Fig. 7C). In the same areas, tumor cells with strong nuclear staining for Survivin were identified while the bulk of tumor cells was expressing lower levels of mainly cytoplasmic Survivin (Fig. 7E, 7F, and quantification 7G). Colocalization of both proteins in the same cells could not be confirmed with this method, as suitable conditions for double immunofluorescent labeling could not be established on paraffin-embedded archival glioblastomas. However, hGBM-IC cultures derived from the corresponding primary glioblastoma samples, and cultured for a limited number of passages were available for three of the samples analyzed. These cells formed spheres, were able to differentiate and to give rise to tumors of similar histological appearance as the primary tumor when injected into NOD-SCID mice (supporting information Fig. S7). qRT-PCR analysis revealed Bmi1 and Survivin upregulation in these cultures (Fig. 7H, 7I) and strong nuclear expression of Bmi1 and Survivin was confirmed by immunohistochemistry in hGBM-IC (Fig. 7J).

Figure 7.

Survivin is highly expressed in human glioblastoma and it is coexpressed with Bmi1 in a majority of hGBM-IC. H&E staining of human glioblastoma (A, B). Strong nuclear staining for Bmi1 and Survivin is seen in cells located perivascularly in the surrounding of necrotic areas (C, E) while weaker expression of Bmi1 and mainly cytoplasmic staining for Survivin is seen in the majority of the tumor mass (D, F). Quantification is shown in (G), ***, p < .001 with Student's t test. Error bars represent SD, n ≥ 3. qRT-PCR analysis reveals correlation between high Bmi1 expression and high Survivin expression in hGBM-IC isolated from different primary tumor samples (H, I). Immunohistochemistry confirms this at the protein level (J, K, and inset). Scale bar is 62.5 μm in (A--F) and 125 μm in (J, K). Abbreviation: PV, perivascular.

In summary, our data demonstrate that increased levels of Bmi1 do not repress Survivin expression in murine or human GBM-IC and that overexpression of both proteins coexists in these cells, in a similar fashion to what observed in non-neoplastic NSC overexpressing Bmi1. These data raise the possibility that both NSC and GBM-IC are dependent on Survivin to survive the hyperproliferative state induced by Bmi1 overexpression.

DISCUSSION

It is increasingly evident that PcG genes exert a crucial role in stem cell maintenance and differentiation through chromatin modifications at the promoter region of downstream target genes. We have previously shown that conditional overexpression of Bmi1 induces increased proliferation of both NSC and NPC and that this is mediated by downregulation of both the ink4a/ARF and the p21/Foxg1 axes [15]. Here, we show that p16ink4a expression is silenced by chromatin modification induced by repressive epigenetic regulation of its promoter, a mechanism that is keeping with published literature for this locus [40]. However, we show here for the first time, a developmental stage-specific mechanism of epigenetic modifications induced by Bmi1 overexpression on the promoter of Survivin in NSC and NPC (supporting information Fig. S8).

Survivin, a member of the IAP gene protein family, plays an essential role not only as an IAP but also as a broad cytoprotector, as an effector of the cellular adaptation to stress, and as a regulator of mitosis (recently reviewed in [28]). While its role in promoting survival of cancer cells is widely recognized, its significance in a physiological context is less well-defined. However, the concept of Survivin being expressed exclusively in neoplastic cells has been increasingly challenged, indeed evidence of its expression not only in the developing embryo but also in stem cells in adult tissues has been reported (reviewed in [41]). We show here that in NS cultures, where the cellular composition is heterogeneous with a prevalence of committed and differentiated progenitor cells, the amount of H3K27me3 is increased at the promoter region of Survivin upon Bmi1 overexpression, while H3K4me3 is decreased. This series of events leads to silencing of Survivin expression and increased apoptosis within the NS. A similar increase in the levels of H3K27me3 was recently described at the promoter of E-cadherin upon Twist-mediated upregulation of Bmi1 expression, leading to silencing of the gene at the epithelial mesenchymal transition [42]. H3K27me3 is suggested to serve as a docking site to recruit PRC1 (reviewed in [43]) and in keeping with this notion we show increased direct binding of Bmi1 to the promoter region of Survivin.

Interestingly, however, we observed increased trimethylation of H3K27 at the p16ink4a promoter in NSCmon, whereas this repressive mark was not detected on the promoter region of Survivin. In addition, increased H3K4me3 was detected at the Survivin promoter. Consequently, the expression of Survivin was high in these cultures and no increase in apoptosis was noted despite the increased proliferation. Taken together these data indicate that epigenetic changes favoring chromatin condensation such as increased trimethylation of H3K27 may be a general mechanism mediating the functional outcome of elevated Bmi1 expression. However, we show here that this mechanism is cell type specific and only applies to a subset of Bmi1 target genes.

Importantly, our data also show significant reduction in the expression of JmjD3 in NS derived from NestinCre;STOPFloxBmi1 cortices while no increased expression of Ezh2 was seen, raising the possibility that JmjD3 downregulation mediates aberrant retention of H3K27me3 underlying the observed phenotype. Recent ChIP-Seq data for Bmi1 have shown that JmjD3 is a direct target of Bmi1 in HeLa cells [44], it is therefore conceivable that Bmi1 directly represses the expression of JmjD3 also in our experimental setting. In keeping with this notion, we show that knockdown of Bmi1 expression in mGBM-IC leads to increased expression of JmjD3 and loss of H3K27me3 and Bmi1 binding at the p16ink4a promoter. These data are in agreement with recent data showing activation of the ink4a locus by Ras-induced oncogenic stress via upregulation of JmjD3 [45].

Transcriptome analysis of NS overexpressing Bmi1 as compared to controls revealed deregulated expression of several genes involved in DNA damage checkpoint, chromosome stability, and cell cycle progression or oxidative stress. In particular, BCL2/adenovirus E1B 19kDa interacting protein 3 (BNIP3I) is a key player in hypoxia-induced autophagy [46], Metallothionein-3 (Mt3) and selenoproteins play a role in neuronal protection and survival upon oxidative stress [47], and the Dual oxidase (DUOX) genes are responsible for the generation of H2O2 in a variety of tissues [48]. Indeed we have detected increased ROS production upon Bmi1 overexpression in both NS and NSC monolayers. However, analysis of markers expression in NS overexpressing Bmi1 showed evidence of increased oxidative DNA damage and DDR pathway activation only in progenitors committed toward a neuronal lineage but not in NSC or in progenitors committed toward a glial fate. We are currently unable to reconcile these results with published data showing increased ROS production and p53-mediated apoptosis in E18.5 Bmi1-deficient cortical neurons [49]. We do believe, however, that these results are not necessarily in contradiction with the results presented here as the increased proliferation induced by Bmi1 overexpression is most likely indirectly triggering ROS-mediated oxidative stress-induced DNA damage in our experimental setting.

Survivin plays a role in protecting dopaminergic neurons from oxidative stress [50] and, in cancer cells, activates the DNA-DSB repair by directly interacting with γ-H2AX foci [51]. It is therefore conceivable that, although the hyperproliferative condition characterizes both NS and NSC, concomitant Bmi1-mediated downregulation of Survivin upon neuronal commitment contributes to increased oxidative stress-induced DNA damage and shunts these cells into apoptotic death. These data are also in keeping with the recently defined role of Survivin in DNA-DSB repair induced by irradiation, which is mediated by direct interaction of Survivin with γ-H2AX foci [51].

Bmi1 overexpression in NSC/NPC does not elicit neoplastic transformation of these cells [15]. This does not necessarily indicate that Bmi1 does not cooperate in gliomagenesis but indicates that it does not do so in the absence of other mutations. Indeed we found enhanced Bmi1 expression in GBM-IC isolated from a murine model of HGG arising upon conditional deletion of p53 and PTEN in NSC/NPC and downregulation of Bmi1 significantly affected the proliferation of these cells. mGBM-IC revealed an open chromatin conformation at the Survivin promoter with concomitant high expression of Survivin similarly to what observed in undifferentiated NSC. Importantly, concomitant high expression of Bmi1 and Survivin was noted also in human glioblastoma samples and primary cell cultures derived thereof.

CONCLUSION

These results highlight the role of chromatin modification in modulation of Survivin expression in vitro, both in neoplastic and non-neoplastic context. It will be important to further elucidate similarities and differences in chromatin conformation and gene expression induced by overexpression of Bmi1 in normal and neoplastic brain stem cells. This will allow the design of better strategies to tackle these highly aggressive brain tumors.

Acknowledgements

We thank Denise Sheer and Axel Behrens for critically reading this article. We are grateful to the BSU staff, in particular, Anthony Price, for help in the daily care of our mouse colony and to the BI Experimental Pathology Facility for processing and cutting paraffin blocks. This work is supported by grants of the Medical Research Council U.K. (G0800020 and 85704) and Cancer Research Fund (442/1289), Barts and The London Charity to S.M. M.S.-R. is supported by the National Hospital Development Foundation, UCL ION. Grants from the NIHR Comprehensive Biomedical Research Centre (CBRC 31) and the Samantha Dickson Brain Tumor Trust (SDBTT 0805) to S.B. funded the collection of the brain tumor samples and the establishment of primary cell cultures.

DISCLOSURE OF POTENTIAL CONFLICTS OF INTEREST

The authors declare no conflict of interest.

Ancillary