D-2-Hydroxyglutarate Is Necessary and Sufficient for Isocitrate Dehydrogenase 1 Mutant–Induced MIR148A Promoter Methylation Free

Mutation in the isocitrate dehydrogenase 1 and 2 (IDH 1/2) genes defines a major subset of gliomas (1) and has been identified in a number of human cancers (2–6). The recent discovery of an association between IDH1/2 mutation and the presence of a CpG island methylation phenotype in gliomas (G-CIMP) has led to the widely accepted notion that mutant IDH induces G-CIMP, resulting in the downregulation of key tumor suppressor genes, ultimately contributing to gliomagenesis (7). This notion is supported by demonstration that stable expression of IDH1-mutant protein in normal astrocytes was sufficient to establish G-CIMP (8). Although the wild-type function of IDH1/2 is to convert isocitrate to α-ketoglutarate (α-KG; refs. 9–11), somatic mutations to IDH1/2 result in aberrant enzymatic activity, which converts α-KG into the putative oncometabolite D-2-hydroxyglutarate (D-2-HG; ref. 12). IDH1/2-mutant gain of function has been shown to result in a 10- to 100-fold increase in D-2-HG levels in mutant tumors compared with wild type (12–16) and to induce multiple pathophysiologic outcomes (17). Specifically, it has been hypothesized that generation of G-CIMP may occur through D-2-HG inhibition of the α-KG–dependent ten-eleven translocation methylcytosine dioxygenases (TET), which have a key role in the active DNA demethylation process (18, 19). Thus, mutant IDH1/2 is thought to induce methylation by diminishing active demethylation mediated by TET. However, inhibition of mutant IDH1 generation of D-2-HG is not sufficient to induce changes in the established global hypermethylation phenotype in glioma (20, 21). Overall, the role of D-2-HG in IDH-mutant–mediated hypermethylation in G-CIMP has not yet been established.

We recently established that miRNAs, specifically MIR148A, are components of G-CIMP (22). MIR148A is a member of the MIR148/152 family, which includes three members: MIR148A, MIR148B, and MIR152 (23, 24). MIR148A plays important roles in the regulation of growth, development, differentiation, and tumorigenesis (23). MIR148A downregulation appears to be related to the maintenance of multipotency in various stem cell populations (25–27), and its upregulation has been observed in early osteogenic differentiation (28), myogenic differentiation (29), and adipogenic differentiation (30). Consistent with our findings, most studies also indicate that MIR148A possesses tumor-suppressive activity in a variety of cancer cells (22, 24, 31–34). In many cases, reduced MIR148A expression in human cancers appears to result as the consequence of promoter CpG hypermethylation (22, 35–37). However, to date, there has been a paucity of experimental evidence to support the hypothesis that MIR148A transcription is driven by an independent promoter, and the sensitivity of this promoter to DNA CpG methylation has not been confirmed. Thus, this study is aimed at using MIR148A as a reporter to probe the role of D-2-HG in CpG island hypermethylation and at understanding MIR148A silencing as an important factor contributing to gliomagenesis.

The human embryonic kidney 293T cell line (293T cells) was a generous gift from Dr. Paul Mischel (University of California San Diego, La Jolla, CA) and originally obtained from the ATCC. The 293T cell line was maintained in DMEM culture medium with 10% FBS and penicillin/streptomycin at 37°C and 5% CO₂ in a tissue culture incubator. The primary human astrocytes (NHA cells) were purchased from Applied Biological Materials Inc. The culture medium PriGrow IV was purchased from Applied Biological Materials Inc. The culture and subculture of NHA cells strictly followed the manufacturer's protocol. Mycoplasma testing/cell authentication procedures were not applied.

Octyl-D-2-HG, octyl-L-2-HG (the L-enantiomer of 2-hydroxyglutarate), and octyl-α-KG were purchased from Cayman Chemical and C227 was from Xcess Biosciences, Inc. Each of these chemicals was dissolved in DMSO. 293T and NHA cells were treated with these chemicals at indicated concentrations and time courses, and control cells were treated with equal volumes of DMSO. Every 2 days (one passage), these cells were detached by trypsin-EDTA and subcultured in medium with fresh chemicals.

Previously established stable 293T cell lines were used; briefly, to achieve stable expression of IDH1^R132H (IDH1^MUT) as well as IDH1^WT, the 293T cells were infected with the following retroviral constructs, respectively: blank-pLPCX, IDH1^WT-pLPCX, and IDH1^R132H-pLPCX (22). The infected cells were selected by adding puromycin (3 μg/ml) into DMEM and single-cell clones were picked up and amplified for later experiments. For each clonal cell line, transfected gene expression was confirmed by immunostaining and Western blot analysis. These cells were cultured in normal culture medium plus puromycin, passed every 2 days upon reaching approximately 95% confluence and reseeded at approximately 40% confluence.

An enzymatic D-2-HG assay was used to determine the intracellular D-2-HG content (38). This assay is based on the conversion of D-2-HG to α-KG in the presence of the enzyme (D)-2-hydroxyglutarate dehydrogenase (HGDH) and NAD⁺. The enzyme HGDH was a gift from Dr. von Deimling's laboratory. In brief, the 293T or NHA cell pellet was harvested in Cell Lysis Buffer (Cell Signaling Technology), and the lysate was divided into two parts, one for protein content determination by Pierce BCA Protein Assay Kit (Thermo Fisher Scientific), and another for D-2-HG assay. The lysate for D-2-HG assay was first deproteinized by adding 3 μL of Proteinase K (Qiagen) and incubating for 3 hours at 37°C. Then, 25 μL of lysate was reacted with 75 μL of assay solution at room temperature for 30 minutes. For each sample, a triplicate reaction was done. The assay buffer comprised 100 μmol/L NAD⁺ (Applichem), 0.1 μg HGDH, 5 μmol/L resazurin (Applichem), and 0.01 U/ml diaphorase (MP Biomedical) in 100 mmol/L HEPES pH 8.0. Fluorometric detection was carried out in Wallace VICTOR2 1420 MULTILABEL HTS COUNTER (PerkinElmer) with Em540nm/Ex610nm. D-2-HG content was calculated on the basis of a standard curve obtained from a range of concentrations of D-2-HG replacing the cell lysate and then expressed as D-2-HG pmol/μg protein.

Genomic DNA and total RNA (including miRNA) were isolated from 293T and NHA cells by using the AllPrep DNA/RNA/miRNA Universal Kit (Qiagen). DNA and RNAs were prepared from the same set of cells simultaneously as described by the manufacturer. Thus, the results obtained by using the DNA and RNAs are highly comparable. The concentration of Genomic DNA and total RNA were measured using NanoDrop 1000 (Thermo Fisher Scientific).

Genomic DNA (4 μL; 200 ng) was loaded onto a nitrocellulose membrane (Healthcare Life Sciences). After drying, the membrane was washed with TBS buffer with 0.05% of Triton 100 (TBPT) 3 times and then preblocked with 5% nonfat milk and 1% BSA in TBST for 1 hour at room temperature. Rabbit anti–5-hydroxymethylcytosine polyclonal antibody (ZYMO Research) was used at 1:1,000 and goat anti-rabbit IgG-horseradish peroxidase (Santa Cruz Biotechnology) was used at 1:5,000. The signal of 5-hydroxymethylcytosine (5-hmC) was generated by using the SuperSignal West Pico Chemiluminescent Substrate (Thermo Fisher Scientific) according to the manufacturer's instructions and visualized by exposure to X-ray film.

Bisulfite-converted genomic DNA was prepared by using EZ DNA Methylation-Gold Kit (ZYMO Research) and CpG island sequencing [bisulfite sequencing (Bis-Seq)] was performed by using a nested PCR protocol as described previously (22, 39). The methylation status of the MIR148A and RBP1 promoter CpG islands were assessed as described in (22) and (39), respectively.

The sequence of each sample was reviewed using Chromas Lite 2.33 (Technelysium Pty Ltd.), and CpG sites exhibiting a substantial signal for C (as compared with T) were considered methylated. For quantification of MIR148A methylation levels, the T-A cloning method was applied. Briefly, following bisulfite conversion and PCR, the amplified MIR148A CpG island products were ligated into pGEM-Teasy vector (Promega), the ligated products were identified by blue-white screening and amplified by miniprep (Invitrogen), and resulting MIR148A-pGEM-Teasy plasmids were sequenced with SP6 primer.

The level of mature MIR148A was measured as described previously (22) by using the TaqMan MicroRNA Assay (Applied Biosystems). For each sample, expression of RNU6B (as an internal loading control) was measured by using the same system. Reverse transcription of MIR148A and RNU6B was performed at the following conditions: (i) 16°C for 30 minutes; (ii) 42°C for 30 minutes; and (iii) 85°C for 5 minutes. The synthesized reverse transcription products of miRNAs were run in qPCR using the TaqMan universal PCR master mix (Applied Biosystems) and LightCycler 480 System (Roche). MIR148A expression was normalized to RNU6b serving as an internal control and presented as the relative C_t method (2^ΔΔC_t).

All experiments were repeated at least three times independently. For Bis-Seq images and 5-hmC dot blots, a typical representative figure was shown. For the quantitative data, the results were presented as mean + SEM, and differences between groups were determined using two-tailed Student t test. Differences in percent CpG island methylation were determined using χ² analysis. Statistical significance is indicated in all graphs as follows: * = P < 0.05, ** = P < 0.01, and *** = P < 0.001.

To demonstrate the role of D-2-HG in MIR148A CpG island methylation, we treated 293T cells with membrane-permeable octyl-D-2-HG and monitored MIR148A CpG island methylation status over multiple passages of treatment. We first verified that octyl-D-2-HG can be converted to D-2-HG by endogenous cellular esterases using an enzymatic D-2-HG assay (38) that does not detect octyl-D-2-HG (Fig. 1A). When 293T cells were cultured in medium with octyl-D-2-HG (1.0 mmol/L), intracellular D-2-HG achieved a stable level approximately 20 times that of control cells (Fig. 1B), indicating expected conversion of octyl-D-2-HG to D-2-HG by endogenous cellular esterases. Although the increased D-2-HG level is lower than that generated by IDH1^MUT-293T cells (Fig. 1B), we were restricted in using higher concentrations of octyl-D-2-HG due to observed cytotoxicity of higher octyl-D-2-HG concentrations. The MIR148A CpG region did not show a detectable change in methylation status until roughly passage 30 of octyl-D-2-HG treatment, as assessed by targeted Bis-Seq. At passage 32 of the treatment, the MIR148A Bis-Seq chromatogram showed clear T/C double peaks in some CpG sites in octyl-D-2-HG–treated 293T cells (Supplementary Fig. S1), indicating partial development of CpG methylation. To quantitatively assess the degree of the methylation, we combined MIR148A Bis-Seq with T-A cloning (Fig. 1C). Untreated 293T cells were unmethylated at all 12 detected CpG sites. When treated with octyl-D-2-HG for 30 passages, some CpG sites became methylated (∼10% of total CpGs). After further octyl-D-2-HG treatment to passage 35, additional MIR148A CpG sites became methylated (above 60%). Octyl-D-2-HG treatment beyond passage 35 (up to passage 45) did not result in any further increase in MIR148A methylation levels (Supplementary Fig. S2).

To determine whether other G-CIMP CpG islands also developed methylation, we performed Bis-Seq on the RBP1 CpG island and observed methylation following 40 passages of treatment with octyl-D-2-HG (Supplementary Fig. S3). These results indicate that D-2-HG is sufficient to induce CpG island methylation of at least two G-CIMP genes.

For reference, we performed quantitative analysis by Bis-Seq/T-A cloning in previously generated 293T cells transfected with empty vector (pLPCX) or pLPCX-IDH1^R132H (IDH1^MUT) after 35 passages (22). T-A cloning revealed significantly greater CpG methylation in mutant versus vector cells, comparable with cells subjected to long-term octyl-D-2-HG treatment (Fig. 1D).

To provide further evidence supporting D-2-HG as a mediator of IDH1-mutant–induced MIR148A methylation, we investigated whether D-2-HG has the same ability to induce methylation in primary NHA cells. We confirmed that MIR148A was unmethylated in NHA cells. We found that NHA cells could not tolerate long-term treatment at concentrations exceeding 0.5 mmol/L octyl-D-2-HG. When compared with 1.0 mmol/L octyl-D-2-HG–treated 293T cells (Fig. 1B), the intracellular D-2-HG level achieved in 0.5 mmol/L octyl-D-2-HG–treated NHA cells was approximately 50% of the former, albeit still approximately 10 times that found in control NHA cells (Supplementary Fig. S4A). In contrast to untreated control NHA cells in which MIR148A CpG sites remained unmethylated following consecutive passages (Fig. 1E and Supplementary Fig. S4B, top), NHA cells treated with octyl-D-2-HG or octyl-L-2-HG showed the appearance of small C peaks in Bis-Seq chromatograms after two passages of treatment. At passage 4 of continuous octyl-D-2-HG treatment, Bis-Seq chromatograms showed obvious C peaks arising at CpG sites (Supplementary Fig. S4B, bottom). Through combined T-A cloning and Bis-Seq, we confirmed that approximately 12% of CpG sites in the evaluable region were methylated after two passages of octyl-D-2-HG treatment, which further increased to approximately 24% at passage 5 (Fig. 1E). In parallel, cells treated with octyl-L-2-HG (0.5 mmol/L) also developed partial CpG methylation (Supplementary Fig. S4C). As a control, cells treated with the same dosage of octyl-α-KG failed to develop methylation. We were unable to prolong the time of octyl-D-2-HG or -L-2-HG treatment at these concentrations, as the NHA cells could not survive beyond passages 6 to 8.

To provide evidence that TET dioxygenase activity can be inhibited by increased intracellular D-2-HG content, we performed dot blots on genomic DNA using a specific antibody against 5-hmC, the TET-mediated oxidation product of 5-methyl-cytosine (5-mC). At passage 5, 293T cells treated with octyl-D-2-HG showed a marked reduction in global 5-hmC levels (Fig. 2A), indicating inhibition of TET activity by D-2-HG; similar results were also observed in 293T cells with treatment of octyl-L-2-HG. This apparent early inhibition of TET activity as indicated by reduced 5-hmC is surprising given the delay in methylation induction until passage 30. Similarly, after two passages of octyl-D-2-HG treatment, global 5-hmC content became significantly lower than controls in NHA cells (Fig. 2B), similarly suggesting TET inhibition. Parallel treatment with octyl-L-2-HG also showed a significant decrease in 5-hmC content, whereas octyl-α-KG had no impact on 5-hmC (Fig. 2B).

To determine whether inhibition of endogenous D-2-HG formation by mutant IDH1 enzymatic activity can prevent IDH1-mutant–induced DNA CpG methylation of MIR148A, we employed an IDH1^R132H-selective inhibitor, C227, and treated previously established IDH1^MUT-293T cells (22) to investigate the effects of development of MIR148A CpG methylation following IDH1^MUT transfection. As we previously found (22), the MIR148A promoter region developed CpG methylation after >20 passages (Supplementary Fig. S5). At passage 5 of stable transfection, prior to the induction of MIR148A methylation, 293T cells transfected with IDH1^MUT or blank vector were treated with 2.0 μmol/L C227. This dosage effectively inhibits D-2-HG generation by IDH1^MUT-293T cells at all tested cell passages (Fig. 2C). We did not observe any detectable cytotoxicity with long-term treatment at this concentration. Importantly, in IDH1^MUT-293T cells continuously treated with C227, no obvious MIR148A methylation was detected at any time point up to passage 40 (passage 45 of transfection; Supplementary Fig. S6). After C227 treatment, we observed restored 5-hmC levels in IDH1^MUT-293T cells to the level of vector-293T cells (Fig. 2D) at passage 20 (passage 25 of transfection), suggesting restoration of TET activity. These results indicate that D-2-HG is necessary for MIR148A CpG island methylation in IDH1^MUT-293T cells. Treatment with C227 was also demonstrated to be sufficient to prevent RBP1 methylation, another established G-CIMP member (Supplementary Fig. S7).

Although our previous work had shown evidence of promoter activity within the upstream of pre-MIR148A region, the existence of an independent promoter for this had not been further validated in the experimental literature. We thus sought to experimentally confirm the location of the core promoter activity and demonstrate a direct physical interaction between the promoter and DNA-dependent RNA polymerase II (Pol II). Pol II antibody pull-down experiments using 293T cells, which highly express MIR148A, show that an endogenous DNA fragment contained within the previously identified 1.6 kb upstream fragment is associated with Pol II. The specificity of interaction between Pol II and the MIR148A promoter region was confirmed by showing that a “control” DNA fragment upstream of pre-MIR148A gene (−4757 to −4547) was not precipitated by the Pol II antibody (Fig. 3A). Our Pol II chromatin immunoprecipitation (ChIP)-PCR result is consistent with publicly available data from the UCSC Human Genome Browser (Supplementary Fig. S8), which shows a complete overlap between the CpG island and Pol II pulldown region, suggesting Pol II sensitivity to CpG methylation.

However, due to limitations of the ChIP-PCR technique in mapping precise promoter structure, we were not able to characterize the core promoter region in more detail through endogenous experiments and therefore proceeded to perform luciferase reporter assays. The MIR148A gene is localized to p15.2 of chromosome 7 (7p15.2), and is 200 kb upstream and 250 kb downstream from the nearest protein-coding genes, strongly suggesting that MIR148A is an intergenic miRNA (UCSC Human Genome Browser; Fig. 3B; Supplementary Fig. S8). Unlike intragenic miRNAs, which generally lack an independent promoter and are instead driven by the promoters of neighboring protein-encoding genes, intergenic miRNAs possess independent promoters that drive their transcription. Indeed, our initial studies indicated that the 1.6 kb upstream fragment (−1038 to −2661 bp) of the pre-MIR148A gene can drive luciferase expression, suggesting functional promoter activity (22). The 1.6-kb fragment between −1038 and −2661 bp has two potential transcriptional start sites (TSS2 and TSS3) predicted by Promoter 2.0 Prediction Server (http://www.cbs.dtu.dk/services/Promoter/), implying two potential promoters. Although another potential transcription start site, TSS1, was also predicted at −203 bp of pre-MIR148A by the program (Fig. 3B; Supplementary Fig. S8), based on the low score as well as the location within the pri-MIR148A coding region, the predicted TSS1 has a low probability to represent a TSS; thus, we did not construct a promoter reporter for the TSS1 region and focused our attention to TSS2 and TSS3.

To determine whether TSS2 or 3 were possible TSSs, we mapped the core promoter of MIR148A to a 344-bp region (−1038 to −1382 bps) upstream of pre-MIR148A gene by testing a series of deletion constructs in a luciferase reporter assay. Although deletion of the portion upstream from −1382 to −2661 bp resulted in significantly reduced promoter activity, the remaining 344-bp fragment (−1038 to −1382 bps) still possessed transcriptional activity comparable with CMV promoter activity (Fig. 3C). Moreover, once the 344-bp fragment was deleted, the remaining upstream region completely lost transcriptional activity. These results indicated that the 344 bp region around TSS2 (−1113 bp) possesses basic promoter activity. Another potential promoter may be located near the predicted region TSS3 (−1713; Fig. 3B; Supplementary Fig. S8). However, when the region immediately upstream of TSS3 was removed, luciferase expression was unaffected (Fig. 3D), indicating that the TSS3 region lacks transcriptional activity. Overall, these results strongly suggest that MIR148A transcription is driven by its own independent promoter and that significant (core) promoter activity is contained within a region −1038 to −1382 bp upstream of the pre-MIR148A gene location.

Using 293T cells, we investigated the role of CpG methylation in the regulation of MIR148A promoter function. We found that Pol II recruitment to the MIR148A promoter is decreased in IDH1^MUT-293T cells compared with control 293T cells (Fig. 4A). In addition, as we similarly reported earlier (22), we observed a corresponding reduction of MIR148A transcripts in IDH1^MUT-293T cells, compared with 293T cells transfected with blank vector as determined by RT-qPCR (Fig. 4B). Moreover, after 4 days of transfection of the MIR148A-pGL reporter, MIR148A promoter–driven luciferase expression is partially reduced in later passage IDH1^MUT-293T cells as compared with vector-293T cells and IDH1^WT-293T cells (Fig. 4C). To further verify the ability of MIR148a promoter methylation to silence transcription, we generated a fully methylated MIR148A-pGL reporter by in vitro treatment with the methyltransferase, SssI. When the methylated construct was transfected into 293T cells, its luciferase activity almost completely abrogated (Fig. 4D). In contrast, control constructs did not show substantial inhibition with SssI enzymatic methylation. Taken together, these results strongly suggest that MIR148A promoter activity is negatively regulated by CpG island methylation.

To confirm that octyl-D-2-HG treatment downregulates MIR148A transcription, we measured cellular MIR148A transcript levels in 293T and NHA cells by both RT-PCR and RT-qPCR. At passage 34–36, octyl-D-2-HG treatment reduced MIR148A expression compared with DMSO-treated cells (Fig. 5A). At earlier passages, octyl-D-2-HG produced no change in MIR148A expression (Supplementary Fig. S9A).

In NHA cells, at passage 4, both octyl-D-2-HG and -L-2-HG significantly inhibited MIR148A transcription (Fig. 5B). However, octyl-α-KG did not impact MIR148A expression. At passages 1 and 2, MIR148A content in NHA cells treated with octyl-D-2-HG, -L-2-HG, or -α-KG showed no change compared with untreated cells (Supplementary Fig. S9B). The occurrence of CpG methylation in the MIR148A promoter region (passage 2) immediately prior to MIR148A transcriptional inhibition (passage 4) suggests a causal role for methylation in transcriptional regulation. These results indicate that, in 293T and NHA cells, D- or L-2-HG can induce MIR148A promoter hypermethylation and reduce MIR148A expression.

We then investigated the role of long-term C227 treatment on MIR148A transcription in 293T cells. At passage 10 of C227 treatment (passage 15 of transfection), as was expected given that this is prior to the development of promoter methylation, MIR148A transcript levels were similar between IDH1^MUT- and vector-293T cells with or without application of C227 (Fig. 6A, left). However, beyond approximately passage 20 of IDH1^MUT transfection, chronic C227 application prevented MIR148A transcript reduction occurring with the establishment of MIR148A promoter methylation in IDH1^MUT-293T cells (Fig. 6A, right). At passage 40 of C227 treatment (passage 45 of transfection), parallel vector-293T cells, with or without C227 treatment, retained an unmethylated status at the MIR148A promoter (Supplementary Fig. S6) and maintained a similar level of MIR148A transcription as shown in Fig. 6A (right). At passage 35 of C227 treatment (passage 40 of IDH1^MUT transfection), IDH1^MUT-293T cells still retained an unmethylated status at the MIR148A promoter compared with the methylated status in vehicle-treated IDH1^MUT-293T cells (Supplementary Fig. S6). In these 293T cells, we assessed Pol II interaction with MIR148A promoter by ChIP-PCR. We did not observe an effect of C227 on Pol II binding with the unmethylated MIR148A promoter in vector-293T cells (Fig. 6B). Similar to passage 30 (see Fig. 4A), untreated IDH1^MUT-293T cells at passage 40 of transfection still display a decrease in Pol II binding with MIR148A promoter compared with vector-293T cells, but treatment with C227 significantly prevents the reduction of Pol II recruitment in IDH1^MUT-293T cells (Fig. 6B). These results further validate the effect of IDH1^MUT generated D-2-HG on MIR148A promoter methylation status and transcriptional regulation.

Finally, we sought to investigate whether D-2-HG–induced CpG methylation in the MIR148A promoter is reversible after withdrawal of octyl-D-2-HG treatment from 293T cells. To test the hypothesis that octyl-D-2-HG withdrawal will restore TET demethylation activity, we divided octyl-D-2-HG–treated 293T cells in which MIR148A promoter methylation had been established (passage 35) into two treatment groups: one with continued octyl-D-2-HG treatment and the other with discontinued octyl-D-2-HG treatment (de-D-2-HG). Following another 4 passages in culture, DNA and RNA were extracted from control, octyl-D-2-HG, and de-D-2-HG–treated 293T cells to analyze MIR148A DNA methylation and RNA expression. As previously shown (Fig. 1C), 293T cells with 39 passages of octyl-D-2-HG treatment demonstrated MIR148A promoter hypermethylation compared with parallel control cells, but cells in which octyl-D-2-HG treatment had been discontinued for 4 passages showed a dramatic decrease in 5-mC sites (Fig. 7A).

To determine alterations in MIR148A transcription, we isolated mRNA in parallel from the various experimentally treated cells and performed RT-qPCR assays. We found that following octyl-D-2-HG withdrawal, MIR148A transcripts were significantly increased at passage 4 compared with the cells with continuing octyl-D-2-HG treatment (passage 39) and were not significantly different compared with control 293T cells at the same passages (Fig. 7B). These results provide further evidence for D-2-HG epigenetic regulation of MIR148A promoter activity. On the basis of our T-A cloning data, following withdrawal of octyl-D-2-HG for 4 passages, the MIR148A promoter retained a low level of apparent methylated (5-mC) sites compared with control cells, but MIR148A transcription increased to levels approaching those of control cells (Fig. 7B).

After 8 passages of octyl-D-2-HG withdrawal, the MIR148A promoter became unmethylated to a degree comparable with that of control cells at the same passage, whereas octyl-D-2-HG–treated 293T cells retained their hypermethylation status (Fig. 7C). A similar independent result was obtained via bulk sample Bis-Seq (without T-A cloning) by starting with 293T cells previously treated with octyl-D-2-HG for 40 passages (Supplementary Fig. S10). Following 8 passages of octyl-D-2-HG withdrawal, MIR148A transcripts also remained significantly increased compared with the cells with continuing octyl-D-2-HG treatment (passage 43) and were not significantly different compared with control 293T cells at the same passages (Fig. 7D). However, although the MIR148A promoter was completely unmethylated, MIR148A transcript levels did not increase markedly compared with passage 4, and in fact appear slightly reduced, although this difference was not significant. In light of the fact that both 5-mC and 5-hmC are resistant to bisulfite conversion, and therefore indistinguishable by Bis-Seq, this result suggests that the residual 5-mC signals following withdrawal of octyl-D-2-HG for 4 passages are not entirely comprised exclusively of 5-mC, but may contain 5-hmC.

To test this idea, we selected the −546 to −746 bp region of pre-MIR148A (∼250 bp downstream of TSS2) to dissect 5-hmC signals from 5-mC signals by using the EpiMark 5-hmC and 5-mC Analysis Kit (New England Biolabs). This DNA region contains two CCGG sites, which can be cut by MspI (5-mc/5-hmC nonresistant) and Hpa II (5-mc/5-hmC resistance). If the CpGs of both CCGG sites are 5-hydroxymethylated, the hydroxyls of 5-hmCs can be glucosylated to 5-gluco-hmC (5-ghmC) by the T4 Phage β-glucosyltransferase (T4-BGT), which leads to resistance to MspI digestion and therefore can be detected by PCR amplification. As expected, in control cells (passage 39 and 45), MspI or HpaII digestion completely prevented PCR amplification of the 200-bp fragment, indicating that neither 5-mC nor 5-hmC was present. In octyl-D-2-HG–treated cells at the same passage, although HpaII resistance was present, glucosylation did not prevent MSP1 digestion, indicating presence of 5mC at both CCGG sites. However, following withdrawal of octyl-D-2-HG for 4 passages, we observed the protection of glucosylation on MSP digestion, indicating that both CCGG sites contained 5-hmC signals (Fig. 7B). The 5-hmC signals disappeared following 8 passages of discontinued octyl-D-2-HG treatment (Fig. 7D). Because 5-hmC is a positive epigenetic marker for gene transcription, the temporal 5-hmC increase (albeit nonsignificant) may explain why passage 4 cells may have higher miR148a transcripts than passage 8 cells.

By inhibiting α-KG–dependent TET activity, D-2-HG has been widely hypothesized to alter the dynamic balance between DNA methylation and demethylation resulting in the DNA CpG island hypermethylation phenotype associated with IDH1/2-mutant gliomas (G-CIMP; refs. 7, 8). Turcan and colleagues reported that expression of the IDH1^R132H mutation in immortalized human astrocytes can induce development of this phenotype (8). However, to date, there has been limited direct experimental evidence to prove a causal role for D-2-HG in the establishment of G-CIMP. IDH1 mutation causes multiple metabolic abnormalities independent of D-2-HG, such as inactivation of NADPH-dependent reductive carboxylation (40), stimulation of glutamine metabolism under hypoxia (41), inhibition of ATP synthase (42), and alteration of glutathione and TCA cycle metabolite levels (43). Further adding complexity, D-2-HG does not mimic all IDH mutation effects (44), and accumulation of IDH1^MUT-generated D-2-HG is able to alter and/or dysregulate a variety of other pathways through D-2-HG–mediated inhibition of other dioxygenases, including histone lysine demethylases (45), regulators of the HIF pathway (46), and mediators of collagen protein maturation (47).

In this study, we used the MIR148A gene as a “reporter” to probe the role of D-2-HG in CpG island hypermethylation, and additionally, we further characterized the miRNA148a promoter region. Our results clearly indicate that: (i) octyl-D-2-HG–treated 293T and NHA cells develop MIR148A promoter CpG methylation; (ii) 5-hmC levels in cells treated with octyl-D-2-HG or -L-2-HG exhibit significant decreases, reflecting D- or L-2-HG inhibition of TET demethylation activity; (iii) 2-HG–treated cells display a reduction in MIR148A expression following development of MIR148A promoter methylation; and (iv) 2-HG withdrawal reverses methylation, increases 5-hMC, and allows transcription. These results establish that the putative oncometabolite is sufficient to induce methylation of a G-CIMP gene promoter likely via dysregulation of TET demethylating activity. That methylation could be induced in 293T cells and NHA cells is consistent with the notion that IDH1 mutation induction of CIMP is not astrocyte specific, given that IDH1-associated CIMPs are found in other malignancies such as AML (48). Upon initiation of octyl-D-2-HG treatment, NHAs, even with only 50% of D-2-HG concentration, exhibited MIR148A promoter methylation in an apparent accelerated time course compared with 293T cells (2–5 vs. 30 passages). In IDH1^MUT-293T cells, we observed a similar delay in methylation. We hypothesize these differences in sensitivity derive from inherent differences in basal methylation and demethylation activity, which could offer a potential explanation for observed variations in IDH1/2 mutation frequency among distinct tissue types.

To investigate whether D-2-HG was necessary for CpG island methylation of MIR148A in IDH1^MUT 293T cells, we used C227 to block IDH1^R132H-mutant activity. C227 is a potent and selective IDH1^R132H-mutant inhibitor first reported in the patent WO2012009678 (49). Rohle and colleagues demonstrated that AGI-5198, a compound comparable with C227, inhibits D-2-HG production and proliferation of IDH1^MUT glioma cells but not IDH1^WT glioma cells in vitro and in vivo (xenograft model; ref. 20). Our previous work has also demonstrated that stable IDH1^MUT-293T cells exhibit CpG methylation of the MIR148A promoter (22). Using this cell model, we show in this study that initiation of long-term C227 treatment at an early stage of IDH1^MUT transfection normalizes global 5-hmC content and prevented both development of MIR148A promoter methylation and inhibition of MIR148A transcription, presumably through reduced D-2-HG levels. A separate series of experiments demonstrated D-2-HG–mediated induction of promoter methylation in another gene associated with G-CIMP, RBP1, underscoring the role of D-2-HG in establishment of a global methylation phenotype. Promoter methylation of this gene was also prevented by treatment with C227. Thus, our results strongly suggest that D-2-HG is necessary for mutant IDH1–induced methylation of MIR148A, and this effect likely extends to the other G-CIMP promoters. These results corroborate previous data indicating that octyl-D-2-HG treatment leads to transcriptional repression of the miRNA 200 family in HCT116 and MCF-10A cells, although methylation was not assessed in this study (50). Despite evidence indicating inability of IDH^MUT inhibitors to reverse established methylation (20, 21), it remains to be determined whether C227 can reverse MIR148A methylation once established via inhibition of D-2-HG. Interestingly, reversal of IDH1 mutation–induced methylation by similar inhibitors was observed in an erythroleukemia model system (51).

All three members of the TET family possess a common core catalytic domain at the carboxyl terminus, comprised of a double-stranded β-helix (DSBH) domain and a cysteine-rich domain. α-KG binds with the DSBH and is necessary for TET demethylase activity (52, 53). Our 5-hmC data in response to octyl-D-2-HG exposure and withdrawal, coupled with prior evidence that that 2-HG can bind TETs competitively with a-KG and block TET activity in vivo and in vitro (19, 54), strongly indicate inhibition of TET by D-2-HG. Direct confirmation of 2-HG and TET2 physical interaction will await further studies. TETs oxidize 5-mC to 5-hmC, 5-hmC to 5-fC, and then 5-fC to 5-caC in a three-step process. Because the first step is more rapid than the latter two steps, 5-hmC signals can transiently accumulate in the TET-associated demethylation process and can function as a positive epigenetic marker of gene transcription (53). Expression levels of TET enzymes are largely decreased in various cancers, which is also consistent with a low level of 5-hmC (55). However, TET enzymes (56–58) and 5-hmC (59–61) are known to be present at high levels in neuronal cells, potentially contributing to differences in basal demethylating activity, which could be responsible for increased sensitivity in NHA cells.

Our results clearly indicate that discontinuation of octyl-D-2-HG treatment induced an increase in MIR148A transcript and, using a 5-hmC glycosylation protection assay, a transient generation of 5-hmC in at least two CCGG sites, implicating 5-hmC as a positive transcriptional regulator. However, technical limitations prevent elucidation of details on 5-hmC distribution within the MIR148A promoter region/CpG island. Despite our inability to determine 5-hmC changes throughout the CpG island, emerging evidence indicates processivity of TET enzymes (53), which suggest that the entire CpG island may undergo demethylation and thus display 5-hmC markings throughout.

In previous work, we established the existence of a CpG island in close proximity with the MIR148A coding region as a component of the glioma-associated methylation phenotype, G-CIMP. These data provided primary evidence that MIR148A possesses tumor-suppressive activity and is epigenetically silenced in IDH1/2-mutant gliomas (22). Analysis of the pre-MIR148A 5′ upstream region using a promoter prediction program (Promoter 2.0 Prediction Server, http://www.cbs.dtu.dk/services/Promoter/) identified three potential TSS at −203 bp (TSS1), −1113 bp (TSS2), and −1713 bp (TSS3) respectively (Fig. 3B; Supplementary Fig. S8). Data provided by Li and colleagues (37) indicate that there is methylation-sensitive transcriptional regulatory activity in the region located at −155 to +218. On the basis of its location downstream of TSS1, their data do not indicate involvement of TSS1 in transcriptional regulation of miRNA148a. Therefore, we did not attempt to identify promoter activity near the predicted TSS1 because its borderline score and location with the pri-miRNA transcript coding sequence strongly argues against TSS1 being the true start site. Instead, we focused our attention to TSS2 and TSS3.

Previously, we cloned a 1623-bp DNA fragment upstream of the MIR148A coding region, which contains TSS2 and TSS3 and has been verified to possess strong promoter activity (22). Thus, in the current study, we used ChIP-PCR assay to establish the presence of physical interaction between Pol II and a putative MIR148A promoter region in 293T cells (Figs. 3 and 4), providing greater evidence that MIR148A transcription is driven by an independent promoter. RNA polymerase II (Pol II) catalyzes DNA transcription to form the precursors of a majority of miRNA transcripts as well as mRNAs (62). These Pol II ChIP-PCR experiments demonstrate that the endogenous promoter region of MIR148A, in 293T cells, is physically occupied by Pol II (Fig. 3A), that occupancy is sensitive to CpG methylation (Fig. 4A), and that decreasing occupancy was followed by a decrease in endogenous MIR148A expression (Fig. 4B). This assay provides the most direct evidence to mark an active promoter site by probing the interaction between the DNA promoter and DNA-dependent RNA transcriptase (63). Moreover, using a serial deletion assay of the promoter-luciferase reporter expressed in 293T cells, a 344-bp fragment around TSS2 was mapped as a core region of MIR148A promoter activity, while the DNA fragment around TSS3 did not have detectable transcriptional activity (Fig. 3C and D). Our identified active region for the MIR148A promoter overlaps with a CpG island identified upstream of MIR148A by the UCSC Human Genome Browser (see Fig. 3B; Supplementary Fig. S8). Previously, we established a 293T cell model with stable IDH1^MUT expression in which we observed MIR148A CpG island methylation compared with blank vector transfected 293T cells (22). Using this model, we observed that CpG island methylation abolished Pol II-promoter binding in the 293T-expressing IDH1^MUT protein (Fig. 4A), suggesting that MIR148A CpG island methylation impairs Pol II–mediated MIR148A transcription. Moreover, when a premethylated MIR148A promoter-reporter created in vitro using CpG methyltransferase SssI was transfected into 293T cells, the promoter activity was reduced by >80% compared with the mock reporter, indicating a direct causal consequence of MIR148A promoter methylation on its transcriptional activity. Taken together, these results strongly support the conclusion that MIR148A expression is driven by an independent, methylation-sensitive promoter.

In conclusion, the current study demonstrates that D-2-HG, the IDH^MUT-generated oncometabolite, is necessary and sufficient to induce methylation of a G-CIMP gene promoter: This study provides insight into the mechanisms by which mutant IDH dysregulates global methylation patterns. These findings support the widely held notion by which the IDH1/2 mutation induces oncogenesis via epigenetic silencing of important tumor-suppressive genes such as MIR148A.

T.F. Cloughesy is the chief medical officer at Agile Research Foundation and is a consultant/advisory board member for Abbvie, BMS, Boston BioMedical, Celgene, Cortice, GW Pharma, Lilly, Merck, Novocure, Tocagen, VBI Vaccines, and Wellcome Trust. A. Lai reports receiving a commercial research grant from Genentech/Roche, has received speakers bureau honoraria from Abbvie, and is a consultant/advisory board member for Merck. No potential conflicts of interest were disclosed by the other authors.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

Conception and design: T. Li, A. Lai

Development of methodology: T. Li, H.I. Kornblum, P.L. Nghiemphu, A. Lai

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Li, A. Lai

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Li, C.D. Cox, B.H. Ozer, T.J. Lai, A. Lai

Writing, review, and/or revision of the manuscript: T. Li, C.D. Cox, B.H. Ozer, N.T. Nguyen, T.J. Lai, H.I. Kornblum, L.M. Liau, P.L. Nghiemphu, T.F. Cloughesy, A. Lai

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Li, C.D. Cox, N.T. Nguyen, H.N. Nguyen, S. Li, F. Liu, T.F. Cloughesy, A. Lai

Study supervision: T. Li, A. Lai

This study is supported by NIH (R01CA179071 to A. Lai), UCLA SPORE in Brain Cancer (P50 CA211015), and the Art of the Brain Foundation.

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.