Overexpression of MBD2 in Glioblastoma Maintains Epigenetic Silencing and Inhibits the Antiangiogenic Function of the Tumor Suppressor Gene BAI1

DNA methylation is a naturally occurring event that consists of the addition of a methyl group to the fifth carbon position of the cytosine pyrimidine ring by DNA methyltransferases. Alterations in the patterns of DNA methylation are widespread in human cancers and include genome-wide hypomethylation and the hypermethylation of CpG island–associated gene promoters, the latter of which represents one mechanism leading to the epigenetic silencing of genes in human cancers (1, 2). DNA methylation alterations have been widely reported in human glioblastoma (GBM), a highly vascularized and aggressive primary intracranial tumor (3–7). A distinct subgroup of primary GBM displays concordant hypermethylation at a large number of loci, indicating the existence of a glioma CpG island methylator phenotype (gCIMP; ref. 8). Interestingly, the subset of GBM exhibiting gCIMP is associated with isocitrate dehydrogenase (IDH) mutations, providing a link to an altered metabolic profile (9).

Methyl-CpG–binding domain (MBD) proteins interpret the DNA methylation marks and thus are critical mediators of many epigenetic processes (10–12). The MBD family comprises 5 members; MBD1–4 and MeCP2. MBD1, MBD2, and MeCP2 bind selectively to methylated CpGs and repress transcription from methylated promoters in vitro and in vivo. In contrast, MBD3 binding is not dependent on DNA methylation, and MBD4, while selective for methylated DNA, has been primarily characterized as a thymine DNA glycosylase with little role in transcriptional repression (10–12). However, the expression pattern and functional roles of MBDs in glioblastoma pathogenesis remain yet unidentified.

BAI1 is an orphan G protein–coupled receptor (GPCR)-like receptor abundantly expressed in normal brain with potent antiangiogenic and antitumorigenic properties that was initially identified in a screen for p53-regulated genes (13–17). Importantly, BAI1 and its related family members BAI2 and BAI3 were recently found to undergo somatic mutation in several cancers, including lung, breast, and ovarian cancers (18). BAI1 contains several well-defined protein modules in the N-terminus such as an integrin-binding RGD motif followed by 5 thrombospondin type 1 repeats (TSR), a hormone-binding domain, and a GPCR proteolytic cleavage site (GPS; ref. 16). The TSRs within the extracellular region of BAI1 mediate direct binding to phosphatidylserine on apoptotic cells, and BAI1 can cooperate with the engulfment and cell motility 1 (ELMO1)/dedicator of cytokinesis 1 (Dock180)/Rac to promote maximal engulfment of apoptotic cells (19). Interestingly, the ELMO1/Dock180 association is also involved in the invasive phenotype of glioma cells (20). The C-terminus is less well characterized and has a QTEV motif that mediates binding to PDZ domain–containing proteins. The N-terminal extracellular domain of BAI1 can be cleaved at the GPS site, and the resulting 120-kDa fragment, known as vasculostatin (Vstat-120), is able to inhibit angiogenesis in vitro and suppress intracranial tumor growth in vivo (14, 15). A second N-terminal cleavage site was recently identified, generating a smaller vasculostatin (Vstat-40; Cork and colleagues, manuscript submitted).

Our previous results showed that BAI1 expression was absent in most human glioma cell lines and primary glioblastoma samples examined (21), but the underlying mechanisms remain unknown. In the present study, we provide evidence that MBD2 is upregulated in glioblastomas and that it plays a central role in the epigenetic silencing of BAI1 gene expression, thereby suppressing the antiangiogenic activity of BAI1.

The primary GBM tumor samples were obtained from Emory University Hospital and were reviewed by neuropathologists (D.J. Brat and S.B. Hunter) for histologic confirmation of GBM before being included in this study. Human GBM cell lines LN71, LN229, and LN443 were originally established in our laboratory (21). Human GBM cell lines U87MG, SF188, and U251MG were obtained from the American Type Culture Collection (ATCC) and maintained as described (22). All cell lines were authenticated by the ATCC for viability, morphology, and isoenzymology. Human brain microvascular endothelial cells (HBVEC) were purchased from Cell Systems Corp. For chemical treatment, glioma cells were plated (3 × l0⁵ cells/l00-mm dish) and treated 24 hours later with 5-Aza-dC (5 μmol/L; Sigma) for 1 to 5 days.

We determined CpG island methylation status by bisulfite sequencing and methylation-specific PCR (MS-PCR) as previously described (3, 23). Additional details, including primer sequences, are provided in Supplementary Materials and Methods.

To determine the mRNA levels of the BAI1 gene, reverse transcriptase PCR (RT-PCR) was carried out on the total RNA extracted from the cells or GBM samples as described (13). Additional details are provided in Supplementary Materials and Methods.

Western blotting was carried out as described (15). The antibodies used were mouse anti-MBD2 (Abcam; catalog no. ab45027), rabbit anti-MeCP2 (Abcam; catalog no. ab2828), goat anti-actin (Santa Cruz Biotechnology), and rabbit anti-BAI1 (21). The horseradish peroxidase–conjugated secondary antibodies and enhanced chemiluminescence were from ThermoScientific.

Immunohistochemistry (IHC) was carried out on archived formalin-fixed and paraffin-embedded human GBM resection specimens. For the tissue array study, 5 nonneoplastic brain and 54 GBM tumor specimens were sectioned and mounted on 2 slides. Sections were deparaffinized and subjected to antigen retrieval by boiling (20 minutes, 100°C) in 0.01 mol/L Tris HCl (pH 10). Slides were then incubated with a 1:200 dilution of MBD2 antibody. Immunostaining was detected with the avidin-biotin complex method, using diaminobenzidine as the chromogen (Abcam). Slides were scanned at ×40 resolution with a Nanozoomer 2.0 HT slide scanner (Hamamatsu) and staining intensity (5 fields/tumor) was quantified by the MetaMorph Premier software; MBD2 status was assessed on the basis of relative staining intensity unit [absent (0), weak (1; units, 1–75), moderate (2; units, 76–150), strong (3; units, 151–225)] and percentage of positive tumor cells [0% (0), <10% (1), 10%–50% (2), 51%–80% (3), 81%–100% (4)]. Immunoreactivity scores (IHC scores) were determined by multiplying the staining score by the percentage score to give a maximum of 12 (24).

Chromatin immunoprecipitation (ChIP) assay was carried out using a commercial kit (Cell Signaling; catalog no. 9003) with some modifications. After cross-linking, the cells were lysed and sonicated using a Misonix MX2020 sonicator (setting 15, 15 seconds for 3 times). Sonicated lysates were centrifuged at 14,000 rpm at 4°C for 15 minutes to get rid of insoluble fractions. An aliquot of the chromatin preparation was set aside and designated as input fraction. The cleared chromatin (100 μg) was immunoprecipitated with 2 μg of either anti-MBD2 or anti-MeCP2 antibody and incubated overnight at 4°C with rotation. The second day, salmon sperm DNA/Protein A/G agarose slurry was added to these samples and rocked for 4 hours at 4°C. Protein A/G immune complexes were collected and washed. Immune complexes were eluted, and DNA was recovered by DNA purification columns and analyzed by PCR. The primers used were 5′-GCT CAC TCT GAC CCT CTG CTC TTTC-3′ (forward) and 5′-AGT AGC CGA AGA ACT TTC CCT GC-3′ (reverse) for BAI1 promoter, the primers used for MGMT promoter were described previously (25). Acetyl-histone H3 (Lys9) antibody was from Cell Signaling (catalogue no. 9649S) and histone H3 (trimethyl-K9) antibody was from Abcam (catalog no. 8898).

Constructs for short hairpin RNA (shRNA) were generated with the BLOCK-iTU6 RNAi Entry Vector Kit (Invitrogen) as described (26), and primer sequences are provided in Supplementary Materials and Methods. Transient transfection of glioma cells with plasmid DNA (2 μg/60-mm dish) was carried out with Lipofectamine 2000 according to the manufacturer's instructions (Invitrogen) with minor modification (26).

This assay was conducted as previously described (14). In brief, conditioned medium (CM) from glioma cells transfected with shRNAs was collected and concentrated ×100 using an UltraCel filter (Amicon). Confluent HBVECs were incubated in 1% serum DMEM medium overnight in 12-well plates and then wounded with a 10-μL pipette tip, and detached cells were removed by PBS washes. The cells were then treated with CM collected as previously and diluted to ×10 in endothelial cell culture to induce cell migration. Initial wound width was measured; the cells were allowed to migrate for 8 hours, and wound width was measured again. The experiment was repeated independently 3 times and the significance was determined by Student t test.

Quantification of the antiangiogenic responses was carried out using the directed in vivo angiogenesis assay (DIVAA) as previously described (27) with the DIVAA Inhibition Assay Kit (Trevigen; catalog no. 3450-096-IK). Collection of CM from glioma cells transfected with shRNAs was performed as in scratch-wound endothelial cell migration assay. Two microliters of ×100 concentrated CM was mixed with 18 μL Matrigel containing growth factors and filled into sterile surgical silicone tubing (angioreactors). These angioreactors were incubated at 37°C for 1 hour to allow for gel formation, before subcutaneous implantation into the dorsal flank of athymic nude mice [females, 6–8 weeks of age; National Cancer Institute (NCI), Frederick, MD]. Two weeks later, angioreactors were harvested and the Matrigel was removed and digested. Cell pellets and insoluble fractions were collected by centrifugation at 5,000× g for 2 minutes. The cell pellets were washed and incubated at 4°C overnight in 200 μL of 25 μg/mL of fluorescein isothiocyanate (FITC)-labeled Griffonia lectin (FITC-lectin), an endothelial cell–selective reagent. The relative fluorescence was measured in 96-well plates, using a Molecular Device spectrofluorometer (excitation 485 nm, emission 510 nm). The mean relative fluorescence ± SD for 8 replicate assays was determined.

Our previous studies on a limited sample set suggested loss of expression of the BAI1 tumor suppressor in human GBM specimens and cell lines (21), although no mechanism was identified. To independently confirm and investigate the extent of BAI1 loss in GBM, we analyzed large datasets from 2 brain tumor databases, namely, the NCI Repository of Molecular Brain Neoplasia Data (Rembrandt) and The Cancer Genome Atlas (TCGA). The expression of BAI1 was first determined in 28 nontumor brain tissues and 196 GBM samples (institutional diagnosis) in the Rembrandt dataset. As shown in Fig. 1A, the levels of BAI1 gene expression were significantly decreased (P < 0.01) in GBM samples compared with the nontumor tissues. In contrast, the expression of THBS1, which encodes angiogenesis inhibitor thrombospondin 1 harboring 3 TSRs, showed no change (Fig. 1A). In the TCGA dataset, the expression of BAI1 was available in a total of 424 GBM samples. Analysis of this dataset also showed a consistent and dramatic loss of BAI1 expression, with 250 samples (59%) showing more than a 2-fold decrease (Fig. 1B). In contrast, the expression of THBS1 increased in 374 samples (88%; Fig. 1B). The variation in relative THBS1 expression between the Rembrandt and TCGA databases may be due to the different probe sets and array platforms used. Taken together, these data suggest that a large fraction of primary GBMs exhibit a significant loss or downregulation of BAI1 mRNA expression.

To identify the mechanism underlying BAI1 downregulation, we first considered whether BAI1 might be located in a region of genomic loss in GBM. The BAI1 gene is located on chromosome 8q24, a region that is not reported to exhibit loss of heterozygosity in gliomas, and a fact we confirmed by analyzing the TCGA genomic dataset (results not shown). We considered next the possibility of epigenetic mechanisms of silencing of the BAI1 gene. Using the MethPrimer software, we identified a CpG island in the first exon of the BAI1 gene (Fig. 1C). Bisulfite sequencing was used to determine the methylation pattern in exon 1 in 2 nontumor brain and 6 independent GBM samples; 6 clones per sample were sequenced. Whereas the nontumor samples were mostly methylation-free, extensive methylation was detected in the GBM samples, with most CpG sites found methylated in 31 of the 36 clones sequenced (Fig. 1C). The few clones that showed minimal or no methylation could have been derived from stromal tissue within the tumor or represent tumor heterogeneity. To further confirm the bisulfite-sequencing result, the DNA methylation status of the aforementioned 6 tumors, plus an additional 6 tumors and 2 control samples from normal human brain white matter, was analyzed by MS-PCR. Eight of 12 GBM samples exhibited prominent PCR products with the methylated primer set but no products with the unmethylated primer set (Fig. 2A). In contrast, the 2 nontumoral brain samples exhibited detectable PCR bands only from the unmethylated primer set. Examination of BAI1 expression in the same tumors showed an inverse correlation between methylation and gene expression in that expression of BAI1 was only observed in those samples with some degree of unmethylated DNA, whereas the GBM samples that were completely methylated lacked BAI1 expression altogether (Fig. 2B). These data indicate that an aberrant methylation pattern in exon 1 is associated with the silencing of BAI1 expression in a subset of GBM. The glioma-associated silencing seemed to be specific for BAI1 as the expression of 2 homologs, BAI2 and BAI3, and THBS1 did not vary among human glioma cell lines (Fig. 2C). We next examined the effect of a demethylating agent, 5-Aza-dC, on BAI1 gene expression. Treatment with 5 μmol/L 5-Aza-dC for up to 5 days restored BAI1 mRNA expression in LN229 cells in a time-dependent manner (Fig. 2D). Similar reactivation was observed in 3 other BAI1-silent glioma cell lines (Fig. 2E). Because 5-Aza-dC is known to be highly effective at inducing the expression of genes inappropriately silenced by de novo methylation (28), these results suggest that DNA methylation of the BAI1 gene is likely involved in the gene silencing.

Because the impact of DNA methylation on gene silencing is often mediated through the binding of MBDs (29, 30), we then analyzed the expression of various MBD proteins in GBM by mining the expression data in the Rembrandt database. The expression of MBD1, MBD3, MBD4, and MeCP2 showed no significant difference between nontumor control and glioblastomas (Fig. 3A). In contrast, we found that MBD2 is significantly overexpressed in GBM with a more than 2-fold greater mean expression in tumors than in nontumor brain tissues. Comparison of the expression of MBD2 and MeCP2 in 424 GBM samples from the TCGA database also showed that MBD2 expression was markedly increased in a significant fraction of GBM samples (Fig 3B), whereas in a similar comparison, the mean gene expression of MeCP2 was not significantly different. To determine whether MBD2 was also overexpressed at the protein level, we applied IHC on 2 randomly selected GBM specimens and found dramatically increased MBD2 immunopositivity in tumor (GBM) versus adjacent nontumor areas (Fig 3C), consistent with the gene expression data. In contrast, MeCP2 exhibited only low to background levels of staining and most GBM tumor cells were negative for MeCP2 staining. The analysis of MBD2 protein expression in GBM was expanded in a tissue array containing 5 nontumor brain samples and 54 GBM samples (Fig 3D). Weak to moderate MBD2 expression was detected in less than 50% of the tumor cells in 6 of 54 specimens (11%); these tumors were grouped as low-expressing tumors (IHC score 0–4). The remaining 48 of 54 tumors (89%) exhibiting moderate or strong MBD2 expression in more than 50% of tumor cells were included in a high-expressing group (IHC score 6–12). Taken together, these data suggest that MBD2 is significantly overexpressed in GBM.

We next sought to determine whether there was any relationship between the levels of MBD2 and expression of BAI1 in glioma cell lines. The protein levels of MBD2 and BAI1 were determined, and our data suggested a correlation between lack of BAI1 protein expression and elevated MBD2 protein levels (Fig. 4A). We next examined the relationship between the expression of MBD2 and BAI1 in primary GBM. Among the 424 primary GBM samples for which gene expression data were available from TCGA, there was a statistically significant negative correlation between the expression of MDB2 and BAI1 (Spearman correlation coefficient of −0.095; P = 0.05; n = 424; Fig 4B). If one considers only those 373 tumors for which MBD2 was overexpressed by 1.4-fold or greater relative to normal tissues (log₂ > = 0.5), this association was even more significant (Spearman correlation coefficient of −0.14329; P = 0.0056; n = 373; Fig. 4C). Taken together, these data support a negative correlation between MBD2 and BAI1. To determine whether MBD2 played a direct role in BAI1 gene silencing, the endogenous levels of MBD2 mRNA were knocked down by transient transfection of specific shRNA expression vectors. We tested the effect of 2 MBD2-shRNAs first in LN229 cells, a glioma cell line that exhibits abundant MBD2 expression and is silent for BAI1 expression (Fig. 4A). Transfection with either shRNA exhibited a significant reduction in MBD2 protein expression (Fig. 4D) but had no effect on MeCP2 protein levels, indicating specificity of these shRNAs. Transfection of LN229 cells with the MBD2-specific shRNA expression vector led to a reactivation of BAI1 mRNA expression as compared with the nonspecific control shRNA (Fig. 4D). Similar results were observed in 2 other BAI1-silent glioma cell lines (Fig. 4E). Because MeCP2 is known to be involved in transcriptional repression of multiple genes, we also designed MeCP2-specific shRNA vectors and determined their effects on BAI1 mRNA expression. Whereas these vectors potently downregulated MeCP2 protein levels (Fig. 4F), they failed to reactivate BAI1 gene expression (Fig. 4G). Together, these results show that MBD2 contributes to BAI1 silencing.

To further examine the role of MBD2 in the BAI1 gene regulation, we carried out ChIP using antibodies against MeCP2 or MBD2. ChIP assays showed that MBD2 was enriched at the BAI1 promoter in the BAI1-silent cell lines (LN229, U87MG), whereas it was not associated with the locus in BAI1-expressing cells (LN443; Fig. 5A). There was no evidence for MeCP2 association with the BAI1 promoter in any of the 3 cell lines (Fig. 5A). The inability to detect MeCP2 is not due to a lack of expression or technical aspects as these cells express high levels of MeCP2. Moreover, MeCP2 was efficiently recruited to the promoter of O⁶-methylguanine-DNA methyltransferase (MGMT), a gene to which MeCP2 has been shown to bind in a methylation-dependent manner in all 3 cell lines (Fig. 5A; refs. 25, 31). MBDs, and in particular MBD2, are components of NURD/Mi2 corepressor complexes and are thought to direct transcriptional silencing of methylated CpG islands through the recruitment of histone deacetylase (HDAC) activity (10, 11). Consistent with this, aberrant DNA methylation of CpG island promoters is associated with histone hypoacetylation and the acquisition of H3K9 methylation (1, 32). We therefore examined the status of acetylation of lysine 9 (AcH3K9) and trimethylation of lysine 9 (3MeH3K9) on histone H3 by ChIP in BAI1-expressing (SF188 and LN443) and BAI1-silent (LN229 and U87MG) cell lines. 3MeH3K9, a marker of condensed chromatin, was enriched at the BAI1 promoter in BAI1-silent cells but was absent in BAI1-expressing cells (Fig. 5B). In contrast, AcH3K9, a marker for transcriptionally active chromatin, was enriched at the BAI1 promoter in BAI1-expressing cells but was greatly reduced in BAI1-silent cells (Fig. 5B). These results show a correlation between BAI1 expression and changes in histone modification.

Next we investigated the influence of 5-Aza-dC on the association of MBD2 with the BAI1 CpG island. Treatment of the BAI1-silent cell line U251MG (Figs. 2C and 4A) with 5-Aza-dC caused a significant reduction in MBD2 occupancy at the BAI1 promoter as determined by ChIP (Fig. 5C and D). Concomitant with the loss of MBD2, 5-Aza-dC treatment also induced hyperacetylation of histone H3 in the BAI1 promoter region and a reduction in H3K9 trimethylation (Fig. 5C and D). Similar results were obtained when a shRNA against MBD2 was employed (Fig. 5C and D). Interestingly, the depletion of MBD2 did not allow for the binding of another methyl-CpG–binding protein MeCP2 (Fig. 5C). Taken together, these findings support the conclusion that MBD2 selectively binds to the BAI1 promoter and that its presence is necessary to maintain characteristics of closed chromatin and transcriptional silencing at the BAI1 locus. These data also suggest that the targeting of MBD2 may be as effective as DNA demethylating agents in restoring chromatin conformation and prompting the reactivation of BAI1 gene expression.

The aforementioned data, combined with our previous demonstration that the cleaved 120-kDa N-terminal fragment of BAI1 (vasculostatin or Vstat-120) can suppress angiogenesis in vitro and inhibit tumor growth in vivo (14, 15), suggest that reactivation of BAI1 expression by MBD2 knockdown may be a novel therapeutic approach for GBM. However, whether the levels or functionality of reactivated BAI1 are sufficient to restore the antiangiogenic activity is not known. Therefore, we first determined whether reactivation of BAI1 gene expression by MBD2 silencing could restore BAI1 protein synthesis and antiangiogenic activity as measured in an in vitro endothelial cell migration assay. BAI1-silent U251MG cells were transiently transfected with control or MBD2-specific shRNA vectors as previously, and CM was collected 3 days later. Confluent HBVECs were wounded and incubated with CM from transfected U251MG cells. Eight hours later, phase-contrast images were captured to monitor the distance traveled by HBVECs from the wound edge to the center of the wound. As a positive control, we used CM from U251MG cells stably transfected with a BAI1-expressing vector. U251MG-BAI1 cell CM dramatically inhibited wound closure as compared with untreated control (Fig. 6A; compare photographs 2 and 3), consistent with our previous results (14). Similarly, HBVECs treated with CM from MBD2-shRNA–transfected U251MG cells exhibited clearly reduced migration as compared with control shRNA-transfected cells (Fig. 6A); quantification of the migration speed showed more than 50% reduction in wound closure (Fig. 6B).

Because MBD2 knockdown has the potential to reactivate the expression of other genes that are silenced in a DNA methylation–dependent manner in addition to BAI1, it is possible that the inhibitory effect on wound closure determined previously could result from the reexpression of other genes. To address this issue, we designed BAI1-shRNAs to determine whether the effect is BAI1 specific. BAI1-specific shRNAs suppressed BAI1 protein expression substantially, resulting in strongly reduced levels of Vstat-120 in the CM (Fig. 6C). We then repeated the scratch-wound assay with CM from cells transfected with both MBD2-shRNA and BAI1-shRNA. The migration of HBVECs treated with CM from U251MG cells transfected with both MBD2-shRNA and BAI1-shRNA was more than 50% faster than that of HBVECs treated with CM from cells transfected with both MBD2 and control shRNAs (Fig. 6A and B). These results show that a concomitant reduction in BAI1 expression partially abolished the antiangiogenic effect of MBD2 knockdown. We further determined the antiangiogenic activity of reactivated BAI1 in an in vivo angiogenesis assay. CMs from U251MG cells transfected with either MBD2-shRNAs alone or in combination with BAI1-shRNAs were mixed with Matrigel plus FGF-2 and VEGF (angioreactor) and s.c. implanted into athymic nude mice. Two weeks later, angioreactors were dissected and visually inspected for evidence of angiogenesis. Vascularization was readily observed, and the angiogenic response had penetrated deep into the angioreactor with CM from control shRNA transfected U251MG cells (Fig. 6D, top). In contrast, whereas an occasional angiogenic response was observed in angioreactors exposed to CM from MBD2-shRNA–transfected cells, the extent of this response was usually minimal and had only superficially penetrated the angioreactor. To directly measure the angiogenic response, we dissociated the cells in the angioreactor and used a fluorescein-labeled lectin (FITC-lectin) that specifically binds to endothelial cells to quantify the number of murine endothelial cells infiltrated into the angioreactor. A statistically significant difference in endothelial cell content was found between angioreactors exposed to CM from cells transfected with MBD2-shRNA versus MBD2-shRNA in combination with BAI1-shRNA (Fig. 6D, bottom). Taken together, these results show that restoration of BAI1 expression by MBD2 knockdown reactivates BAI1 synthesis and antiangiogenic activity.

Here we provide evidence that the BAI1 gene is epigenetically silenced in GBM and that manipulation of the silencing event can reactivate BAI1 expression and antiangiogenic tumor suppressor activity, which can be exploited for therapeutic means.

In glioma cell lines, BAI1 silencing could be reversed by treatment with the demethylating agent 5-Aza-dC. Therefore, our data support the notion that DNA methylation contributes to inactivation of the BAI1 gene. Transcriptional silencing through promoter DNA methylation has been proposed to occur through several different molecular mechanisms, such as by direct interference with transcription factor binding, by altering the structure of chromatin, and/or by recruiting MBD proteins (10, 11, 29, 30, 33). MBD proteins are critical mediators of many epigenetic processes in that they interpret the methylation marks on DNA and facilitate the establishment of a repressive chromatin environment. ChIP assays showed binding of MBD2 to the CpG island region in the BAI1 promoter specifically in cell lines where the gene was methylated and silenced, whereas there was no association of MeCP2 with the methylated BAI1 promoter. Furthermore, in BAI1-silent glioma cell lines, shRNA-directed knockdown of MBD2 resulted in the local depletion of MBD2 and restored the chromatin state to one similar to that of expressing cell lines (e.g., hyperacetylated at H3K9) and reactivated BAI1 expression. Taken together, these results support a mechanism wherein the specific binding of MBD2 to the methylated BAI1 promoter is necessary to maintain transcriptional repression of BAI1.

Although DNA methylation has been extensively studied in GBM, the expression and regulation of MBDs have not. For the first time, we provide evidence that MBD2 is specifically overexpressed in GBM. The MBD2 gene is located on chromosome 18q21 (34); no amplification of this region has been reported in GBM, nor have genomic copy number changes or somatic mutations for MBD2 been observed in the data from the Rembrandt and TCGA glioma databases (data not shown). Therefore, the overexpression of MBD2 observed in a subset of GBM may result from a different mechanism, such as enhanced transcription. MBD2 has the greatest binding affinity for methylated DNA among MBD family proteins in vitro (34), suggesting that MBD2 may be the MBD family member with the greatest effect on gene silencing (35). Accumulating evidence shows that MBD2 is involved in the suppression of aberrantly methylated tumor suppressor genes by binding to methylated promoters (36, 37), including p14/ARF and p16/Ink4A (38, 39), 14-3-3σ (24), and GSTP1 (glutathione S-transferase p1; ref. 40). In all cases, siRNA-mediated knockdown of MBD2 resulted in the reactivation of the corresponding gene target, similar to what we observed for BAI1. Consistent with a role in tumor promotion, previous work has shown that knockout of MBD2 strongly suppresses intestinal tumorigenesis in Apc^Min mice (41). The underlying mechanism may be that deficiency of MBD2 elevates levels of the Wnt target Lect2, a Wnt pathway repressor (42).

The recent discovery of 5-hydroxymethylcytosine (5-hmC) in mammalian DNA has added a new dimension to the regulation of DNA methylation and may be particularly relevant in the pathogenesis of glioblastomas as the TET enzymes that catalyze the hydroxylation of methyl cytosine residues are among those affected by the accumulation of 2-hydroxyglutarate that accompanies IDH gene mutations (43). Although the precise function of 5-hmC in epigenetic regulation is not yet completely understood, recent work suggests that it may facilitate DNA demethylation through a base excision repair mechanism (44). Furthermore, there is emerging evidence that the binding of some MBD proteins, including MBD2b, to DNA is inhibited by 5-hmC (45, 46), and thus the conversion of 5-mC to 5-hmC may play a functional role in the dynamic regulation of gene expression. Future studies are warranted to determine the distribution of 5-hmC in the BAI1 gene regulatory regions and its role in BAI1 gene transcription.

At present, epigenetic approaches in cancer therapy have focused primarily on inhibitors of the DNA methyltransferases and histone modifiers (e.g., HDACs). Our data suggest that MBD2 may also represent a promising cancer therapeutic target. MBD2-null mice display a surprisingly weak phenotype, and global DNA methylation levels and genomic imprinting are relatively unaffected by the absence of MBD2 (47). The fact that MBD2 knockout mice are viable and resistant to tumorigenesis, coupled with the finding that downregulation of MBD2 could restore a functional BAI1 with potent antiangiogenic activity, makes MBD2 a particularly attractive target for therapeutic intervention for GBM and/or in the prevention of glioma progression. Sequence-specific antisense inhibitors of MBD2 have been shown to inhibit both anchorage-independent growth of human cancer cell lines in vitro and the growth of human tumor xenografts in vivo (48, 49). Validation of MBD2 as a viable target in GBM will require careful examination of the global impact on gene expression, as the potential therapeutic benefit will depend upon the reprogrammed transcriptome tipping the tumor toward an anti- or protumorigenic biological response. Our data show that antagonizing MBD2 elicits a global antiangiogenic response that is largely dependent upon BAI1 expression and would be expected to suppress tumor growth. In summary, our results show a functional role of MBD2 in the repression of BAI1 in GBM. This study could lead to new therapeutic prospects for the treatment of patients with brain tumors.

No potential conflicts of interest were disclosed.

D. Zhu and E.G. Van Meir conceived experiments, D. Zhu carried out all experiments, S.B. Hunter provided the GBM tissue array and histologic confirmation, and P.M. Vertino provided experimental advice. D. Zhu, P.M. Vertino, and E.G. Van Meir wrote the manuscript.

We thank all the Van Meir laboratory members for helpful comments, the Emory Biomarker CORE facility for DNA sequencing services, Dr. Daniel J. Brat for neuropathology diagnosis of GBM samples, Dr. Yuan Liu for help with statistical analysis, Dr. Y. Nakamura for the BAI1 expression vector, and Narra S. Devi and Zhaobin Zhang for technical support.

NIH R01-CA86335 (E.G. Van Meir), NIH RO1-CA077337 (P.M. Vertino), the Southeastern Brain Tumor Foundation (D. Zhu and E.G. Van Meir), and the University Research Council of Emory University (E.G. Van Meir).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.