2-HG Occupies the Same Space as α-KG Does in the Active Site of CeKDM7A

To gain mechanistic insights of 2-HG inhibition, we determined the structure of CeKDM7A bound with D-2-HG at 2.1 Å (Figure 2; Table S1 and Figure S2). Like other JmjC-domain-containing histone demethylase (Chen et al., 2006; Han et al., 2007), the JmjC domain of the catalytic core of CeKDM7A also forms a jelly roll motif with the Fe(II) coordinated by side chains of three highly conserved residues (D497, H495, and H567) within the JmjC domain (Figure 2A; Figure S2A). Notably, D-2-HG binds to the catalytic core in close proximity of Fe (II). We also solved the structure of CeKDM7A bound with α-KG at 2.25 Å (Figure 2B). Comparison of these two structures reveals that D-2-HG adopts a nearly identical orientation as α-KG with one notable difference: whereas the Fe (II) is coordinated by two oxygen atoms in the keto carboxyl end of α-KG, it is coordinated by one oxygen atom and a hydroxyl group in D-2-HG (Figure 2C; Figure S2B). These results provide a structural basis supporting D-2-HG as a competitor of α-KG.

Open in a separate window

Figure 2

2-HG and α-KG Bind to the Same Site in Histone Demethylases

(A) The structure of D-2-HG bound to CeKDM7A JmjC domain. D-2-HG and CeKDM7A are shown in stick and cartoon representation, respectively. Secondary structural elements of CeKDM7A are indicated. Fe (II) is colored in black, Fe (II) coordination is represented by dotted lines and water molecule is shown as orange ball.

(B) The structure of α-KG bound to CeKDM7A JmjC domain, illustrated as in (A).

(C) Superimposition of structures shown in (A) and (B). See also Figure S2.

2-HG Inhibits the Activity of Multiple Histone Demethylases In Vivo

Inhibition of histone demethylases by 2-HG in vitro and binding of 2-HG and α-KG to the same site in the catalytic center of CeKDM7A led us to determine the effect of 2-HG on genome wide histone methylation in vivo. To this end, we synthesized cell-permeable α-KG and racemic octyl-2-HG and verified their structures by NMR (Figures S3A–S3D). Addition of 10 mM octyl-2-HG to the cultured U-87MG cells resulted in a significant accumulation of intracellular 2-HG as determined by GC-MS assay (Figure S3E) and increase of dimethylation on H3K9 and H3K79 by 5- and 10-fold, respectively (Figure 3A; Figure S3F). Addition of cell-permeable octyl-α-KG reversed the increase of both H3K9 and H3K79 dimethylation, providing in vivo evidence supporting the competitive interaction between 2-HG and α-KG. We also synthesized enantiomer specific cell-permeable 2-HG and compared their inhibitory potency. Consistent with in vitro assay, treatment of U-87MG cells with either cell-permeable D- or L-2-HG increased dimethylation on both H3K9 and H3K79 with octyl-D-2-HG being less potent than octyl-L-2-HG (Figures 3B and 3C; Figures S3G and S3H).

Open in a separate window

Figure 3

Reduced Activity of IDH1 and Elevated 2-HG Increase Genome-Wide Histone Methylations and Alter Gene Expression

(A–C) Cell-permeable octyl-2-HG increases histone methylation. H3K9me2 and H3K79me2 levels of U-87MG cells treated with racemic octyl-2-HG (A), octyl-D-2-HG (B), and octyl- L-2-HG (C) were analyzed by western blotting.

(D) Histone methylation increased by IDH1^R132H overexpression can be rescued by addition of cell-permeable octyl-α-KG. Specified histone methylation levels of U-87MG cells expressing IDH1^R132H were analyzed by western blotting.

(E) Elevated H3K79 dimethylation in IDH1^R132H gliomas. IDH1 wild-type or heterozygous for R132H glioma samples were subjected to IHC analysis for H3K79me2 methylation. Scale bars represent 50 μm. Shown are representative IHC results (left) and mean values of IHC quantification (right). Error bars represent ± SD for triplicate experiments. Complete results of all 20 samples are presented in Figure S3K.

(F) Reduction of IDH1 activity activates HOXA genes. HOXA mRNA levels were analyzed by qRT-PCR in U-87MG cells after forced expression of wild-type or R132H mutant IDH1. Error bars represent ± SD for triplicate experiments. See also Figure S3.

R132H Mutation of IDH1 Alters Histone Methylation in Human Glioma Cells and Tumor Samples

We next ectopically expressed IDH1^R132H in U-87MG cells and determined the levels of multiple histone methylation markers. Comparing with cells expressing empty vector, the ectopic expression of wild-type increased α-KG by 20% in U-87MG cells, ectopic expression of IDH1^R132H mutant resulted in a near 60% reduction of α-KG by 60% and <20-fold increase in D-2-HG (Figure S3I). A visible increase in H3K4 monomethylation, H3K27 dimethylation, H3K4 trimethylation, H3K9 dimethylation, and H3K79 dimethylation was observed (Figure 3D; Figure S3J). Addition of cell-permeable octyl-α-KG restored histone demethylation. Together, these results indicate that in addition to CeKDM7A and KDM2A, 2-HG and mutant IDH1 inhibit wide range of histone demethylases, including those involved in the demethylation of H3K4, H3K9, H3K27, and H3K79, and both inhibitions by 2-HG and IDH1 mutant can be reversed by the addition of cell-permeable α-KG.

These results led us to determine whether IDH1 mutation could affect histone methylation in primary tumors. We analyzed H3K79 dimethylation in a panel of 20 human glioma samples, 10 containing wild-type IDH1 and 10 bearing mutated IDH1 (R132H) (see Table S2). H3K79 dimethylation levels were found to be significantly elevated in glioma samples that harbor IDH1 mutation compared to tumor samples that are similar grade but have wild-type IDH1 (Figure 3E; Figure S3K). To further substantiate this result, we determined the expression of several HOXA genes whose increased expression is associated with increased H3K79 dimethylation in MLL-rearranged mouse leukemia and human AML patients (Krivtsov et al., 2008). qRT-PCR analysis demonstrated that the expression of these HOXA genes was increased in cells with forced expression of the IDH1^R132H (Figure 3F). Collectively, these results demonstrate that either expression of mutant IDH1 or increase of 2-HG results in an inhibition of histone demethylases in vivo.

Reduction of IDH Results in an Inhibition of Histone Demethylases

Given the previous observations that mutations in IDH1 or IDH2 cause both α-KG reduction and 2-HG accumulation and the current finding that 2-HG acts as an antagonist of α-KG in vitro, we sought to determine whether reducing the activity of IDH1 and IDH2 could cause similar increase in histone methylation. To this end, we treated cells with oxalomalate, a competitive inhibitor of IDH1 and IDH2 that would decrease both cytoplasmic and mitochondrial α-KG. We found that this treatment led to a dose-dependent increase of trimethylation of H3K4, dimethylation at H3K9, H3K27, and H3K79, and a modest increase in H3K4 mono-methylation (Figure 4A; Figure S4). The differences between different histone demethylases in their responses to oxalomalate treatment probably reflect their different affinities toward α-KG. To further support the above observation, we also determined the expression of the same panel of HOXA genes (see Figure 3F) and found that expression of these HOXA genes was increased in cells treated with oxalomalate (Figure 4B) as well as in cells depleted for IDH1 by shRNA knockdown (Figure 4C). Similar conclusion was also obtained with two additional α-KG-dependent dioxygenases (see below). As both oxalomalate treatment and IDH1 knockdown reduced α-KG without 2-HG accumulation, these results indicate that inhibition of IDH1 could cause similar effect as 2-HG treatment, providing additional evidence supporting a competitive mode between α-KG and 2-HG.

Open in a separate window

Figure 4

INHIBITION of IDH1 Reduces Histone Demethylase Activities In Vivo

(A) Inhibition of IDH1 activity increases histone methylation. Specified histone methylation levels in U-87MG cells treated with increasing concentrations of oxalomalate were analyzed by western blotting.

(B and C) Reduction of IDH1 activity activates HOXA genes. HOXA mRNA levels were analyzed by qRT-PCR in U-87MG cells after knocking down IDH1 (C), and treatment of oxalomalate (B). Error bars represent ± SD for triplicate experiments. See also Figure S4.

α-KG-dependent prolyl hydroxylases and collagen prolyl-4-hydroxylases are inhibited by reduced IDH1 activity or R132H

In addition to histone demethylases, mammalian cells express a large number of dioxygenases that also utilize α-KG as a key substrate (Iyer et al., 2009; Loenarz and Schofield, 2008). To determine how broadly IDH1 mutations affect α-KG-dependent dioxygenases, we investigated the effect of reduced IDH1 activity and increased 2-HG on two additional α-KG-dependent dioxygenases; prolyl hydroxylases (PHDs) and collagen prolyl-4-hydroxylase (C-P4H).

We first determined the effect of reduced function of IDH1 on the levels of HIF-1α and endostatin. HIF-1α is a transcription factor whose function is linked to metabolism, angiogenesis and tumorigenesis, and whose protein level is downregulated under normoxic conditions by PHD-mediated hydroxylation and subsequent hydroxylation-targeted ubiquitination (Bruick and McKnight, 2001; Epstein et al., 2001; Ivan et al., 2001; Jaakkola et al., 2001). Endostatin is a 20-kDa secretory peptide implicated in inhibiting angiogenesis and tumor growth and its production is catalyzed by C-P4H (Krane, 2008; O'Reilly et al., 1997). We found that knocking down IDH1 by shRNA increased HIF-1α and decreased endostatin (Figure 5A). Likewise, inhibition of endogenous IDH1 by oxalomalate treatment similarly increased HIF-1α and decreased endostatin (Figure 5B). Conversely, ectopic expression of wild-type IDH1 decreased HIF-1α and increased endostatin (Figure 5C). In addition to PHDs, the stability and steady-state levels of HIF-1α protein are also regulated by FIH-1 (factor inhibiting HIF-1α) (Lando et al., 2002; Mahon et al., 2001), an asparaginyl hydrogenase whose activity is dependent on α-KG and could therefore also be affected by reduced function of IDH1. Unlike knocking down PHD2 that caused an evident increase of HIF-1α, knocking down FIH-1 in 293T cells did not appreciably affect the steady-state level of HIF-1α (Figure S5A). This result indicates that, at least in 293T cells, FIH-1 does not appear to play a major role in HIF-1α degradation, a notion that is consistent with a recent study in Fih-1 deletion mice showing that FIH-1 plays little or no discernible role in altering HIF stability and function (Zhang et al., 2010). We therefore conclude that the increase of HIF-1α in cells with reduced function of IDH1 is largely caused by the impairment of PHDs.

Open in a separate window

Figure 5

IDH1 Function Supports the Activity of α-KG-Dependent Dioxygenases In Vivo

(A–E) The effects of reducing or increasing IDH1 function on two α-KG-dependent dioxygenases, PHDs and C-P4H, were examined in U-87MG cells after knockdown IDH1 (A), treatment of oxalomalate, a competitive inhibitor of IDH1 (B), overexpression of wild-type IDH1 (C), and tumor-derived IDH1^R132H mutant (D and E). The protein levels of HIF-1α, endostatin, and ectopically and endogenously expressed IDH1 were determined by western blotting.

(F) Decreased endostatin in IDH1^R132H gliomas. IDH1 wild-type or heterozygous for R132H glioma samples were subjected to IHC analysis for endostatin. Shown are representative IHC results (left) and mean values of IHC quantification (right). Scale bars represent 50 μm. Error bars represent ± SD for triplicate experiments. Complete results of all 20 samples are presented in Figure S5C. See also Figure S5.

To determine how tumor-derived IDH1 mutants affect these two dioxygenases, we ectopically expressed IDH1^R132H in U-87MG cells and determined the steady-state levels of both endostatin and HIF-1α proteins. We observed a dose-dependent increase of HIF-1α and decrease of endostatin with increasing expression of IDH1^R132H (Figures 5D and 5E). Ectopic expression of IDH1^R132H did not significantly affect the levels of collagen XVIII mRNA (Figure S5B), supporting the notion that decreased endostatin protein on IDH1^R132H expression is due to inhibition of C-P4H. An evident endostatin decrease and HIF-1α increase were seen in cells in which IDH1^R132H is expressed at a level similar to that of endogenous wild-type IDH1 (lane 2, Figure 5E), excluding the possibility that observed changes in the activity of both PHDs and C-P4H dioxygenases are caused by the grossly overexpression of IDH1 mutant.

We also performed immunohistochemical staining of endostatin in the same panel of 10 IDH1 wild-type and 10 IDH1 mutated gliomas (Table S2). We found that tumors with the R132H mutation expressed significantly less endostatin than those containing wild-type IDH1 (Figure 5F; Figure S5C). Together, these results demonstrate that the activities of α-KG-dependent PHDs and C-P4H are supported by the function of IDH1 and impaired by tumor-derived mutant IDH1.

2-HG Treatment Inhibits PHD and C-P4H

We next examined more directly the effect of 2-HG on these two dioxygenases in cells. Treatment of U-87MG cells with either cell-permeable racemic mixture or enantiomer-specific 2-HG increased HIF-1α and decreased endostatin (Figures 6A–6C). Similar increase of HIF-1α was also observed in 293T cells treated with octyl-D-2-HG (Figure S6A). Consistent with the inhibition of histone demethylases, octyl-D-2-HG exhibited less potent inhibition on both dioxygenases than octyl-L-2-HG and addition of cell-permeable octyl-α-KG substantially suppressed the effect of 2-HG.

Open in a separate window

Figure 6

2-HG Inhibits the Activity of α-KG-Dependent Dioxygenases In Vivo

(A–C) Cell-permeable 2-HG increases HIF-1α and decreases endostatin. U-87MG cells were treated with racemic octyl-2-HG (A), octyl-L-2-HG (B), and octyl-D-2-HG (C). The steady state levels of endostatin and HIF-1α proteins were determined by direct western blotting.

(D) DMOG treatment abolishes 2-HG effect on HIF-1α induction.

(E) PHD2 knock down abolishes 2-HG effect on HIF-1α induction.

(F) 2-HG treatment further induces HIF-1α in hypoxic U-87MG cells. See also Figure S6.

Consistent with the hypothesis that the effect of 2-HG in stabilizing HIF-1α protein in vivo is achieved by antagonizing the binding of α-KG to PHD2, treatment of cells with a cell-permeable NOG, dimethyloxalylglycine (DMOG), or knocking down PHD2 both resulted in an accumulation of HIF-1α protein and activation of several HIF-1α target genes (Figures 6D and 6E; Figures S6B–S6E), but treatment of cells with the combination of cell-permeable octyl-D-2-HG and either DMOG or siPHD2 did not cause a further increase of HIF-1α. Treatment of cells with CoCl₂, a mimetic of hypoxia and a chemical inducer of HIF-1α, accumulated HIF-1α to a level significantly higher than either DMOG or siPHD2 treatment, excluding the possibility that the lack of further HIF-1α accumulation by the octyl D-2-HG in DMOG- or siPHD2-treated cells is because that HIF-1α has already accumulated to the maximal level by DMOG or siPHD2 treatment. Further supporting the notion that 2-HG induces HIF-1α through an inhibition of PHD2 by competing off α-KG, addition of octyl-2-HG to hypoxic cells in which the activity of PHD is reduced by the decrease cosubstrate oxygen, but not impaired α-KG binding, resulted in further HIF-1α accumulation (Figure 6F).

Ectopic Expression of Tumor-Derived IDH1 and IDH2 Mutants Reduces TET-Catalyzed 5hmC Production

Very recently, a class of α-KG-dependent dioxygenase, the TET family of proteins, was discovered that catalyze the conversion of 5-methylcytosine (5mC) to 5-hydroxylmethycytosine (5hmC) (Iyer et al., 2009; Tahiliani et al., 2009). Mammalian cells express three TET genes, TET1, TET2, and TET3. Although the founding member, TET1, was initially identified as the gene fused with mixed-lineage leukemia (MLL) in chromosomal ten-eleven-translocation in rare cases of AML and suspected to be an onco-gene (Lorsbach et al., 2003; Ono et al., 2002), it was recently discovered that TET2 is mutationally inactivated in ~15% of myeloid cancers, including 22% of AML (Delhommeau et al., 2009; Langemeijer et al., 2009). Given the dependence of TET catalytic activity on α-KG, we set forth to determine whether IDH1 and IDH2 mutations and D-2-HG would affect TET activities and DNA cytosine hydroxymethylation and methylation.

The level of 5hmC in most cells is very low, but is substantially increased in cells transiently transfected with plasmids expressing the wild-type catalytic domain of TET (TET1-CD and TET2-CD) protein that can be readily detected by immunofluorescence using an antibody specifically recognizing 5hmC (Ito et al., 2010; Tahiliani et al., 2009) (Figure 7A). Immunofluorescence with the anti-5hmC antibody revealed that coexpression of wild-type IDH1 with TET1-CD or TET2-CD caused a significant increase of 5hmC signal (Figure 7A), suggesting that the concentration of α-KG is a rate-limiting factor of TET2 catalyzed hydroxylation of 5-methylcytosine in TET1-overexpressing cells. Notably, cotransfection of TET1-CD or TET2-CD with IDH1^R132H reduced the 5hmC signal to a barely detectable low level. Essentially the same result was also obtained for IDH2. Both TET1- and TET2-catalyzed 5mC-to-5hmC conversions were substantially increased by the coexpression with wild-type IDH2, but nearly completely inhibited by the coexpression of either IDH2^R140Q or IDH2^R172K mutants (Figures S7A and S7B). Together, these results demonstrate an inhibitory effect of mutant IDH1 and IDH2 toward the hydroxylase activity of the TET family proteins.

Open in a separate window

Figure 7

Tumor-Derived IDH1 and IDH2 Mutants Inhibit the 5hmC Production by TET1 and TET2

(A) HEK293 cells were transiently transfected with plasmids expressing indicated proteins. Thirty-six to forty hours after the transfection, cells were fixed and stained with antibodies specific to Flag to determine the expression of TET protein, to 5hmC to determine the levels of 5hmC, and to DAPI to view the cell nuclei or visualized for green fluorescence to determine the expression of IDH1 proteins. Scale bars represent 50 μm. Additional results on the inhibition of TET1 and TET2 function by IDH2 mutants are presented in Figures S7A and S7B.

(B and C) HEK293 cells were transiently transfected as described in (A). Thirty-six to forty hours after the transfection, genomic DNAs were isolated from the transfected cells, spotted on nitrocellulose membranes and immunoblotted with an antibody specific to 5hmC. Quantification of 5hmC was calculated from three independent assays. The expression of individual proteins was determined by immunoblotting as shown in the right. One representative quantification of 5hmC level determined from the assays using 50 ng genomic DNA is included and the rest of the quantifications are presented in Figures S7C–S7E. Error bars represent ± SD for triplicate experiments. See also Figure S7.

To confirm this result, we isolated genomic DNA from HEK293T cells transiently transfected with TET1 or TET2 individually or in combination with either wild-type or mutant IDH1 and IDH2, and determined 5hmC levels by dot-blot that allowed for more quantitative measurement than the immunofluorescence. These experiments demonstrate that ectopic expression of the wild-type, but not the mutant of TET1 or TET2, resulted in high levels of 5hmC in the cells comparing with cells transfected with control vector (Figures 7B and 7C). Coexpression with wild-type IDH1 or IDH2 caused a significant increase of 5hmC. For example, in the assays using 50 ng genomic DNA, TET2-catalyzed 5hmC production was increased by 149% and 166% by the coexpression of wild-type IDH1 or IDH2, respectively. In contrast, coexpression of TET2-CD with three tumor-derived mutants all caused a substantial decrease of TET2-mediated 5hmC production, resulting in a 70% reduction of 5hmC by the coexpression of IDH1^R132H, 66% reduction by both IDH2^R140Q and IDH2^R172K (Figure 7B; Figure S7C). Virtually the same result was also obtained for TET1-catalyzed 5hmC production that was increased by 222% and 203% by the coexpression of wild-type IDH1 or IDH2, respectively, but reduced by 60%, 69%, and 68% by the coexpression of IDH1^R132H, IDH2^R140Q, and IDH2^R172K, respectively (Figure 7C; Figures S7D and S7E).

Does Reduced α-KG Sensitize α-KG-Dependent Enzymes to the Inhibition by 2-HG?

Joining the loss-of-function mutations targeting fumarate hydratase (FH) and different subunits of succinate dehydrogenase (SDH) (Baysal et al., 2000; Tomlinson et al., 2002), the discovery of IDH1 and IDH2 mutations in human cancers further highlights the direct link between metabolic dysregulation and tumorigenesis. Mutations targeting FH and SDH lead to similar increase in the PHD substrate, HIF-1α (Isaacs et al., 2005; Pollard et al., 2005; Selak et al., 2005). Furthermore, succinate, the substrate of SDH that is accumulated in cells with reduced or inactivation of SDH, has been shown to directly inhibit multiple α-KG-dependent enzymes, including histone demethylases (Smith et al., 2007). Thus, a common feature of the mutations in these three metabolic enzymes is the reduced activity of α-KG-dependent dioxygenases, either indirectly by the accumulation of competitive inhibitors, namely fumarate, succinate, and 2-HG, or directly through the reduction of IDH1 and IDH2 catalytic activity in α-KG production. It will be interesting to determine whether histone and DNA methylations are similarly altered in SDH and FH mutated tumor cells.

We previously showed that mutant IDH1 retains its ability to associate with wild-type subunit and forms a catalytic inactive heterodimer, leading to the dominant inhibition of wild-type IDH1 (Zhao et al., 2009). This model is supported by the recent crystal structural analysis of wild-type:mutant IDH1 heterodimer showing that the R132H mutation hinders the conformational changes from the initial ICT-binding state to the pretransition state (Yang et al., 2010). In other words, the IDH1 enzyme activity in IDH1-mutated tumor cells is reduced to ~25% of normal cells. Production of D-2-HG by the mutant IDH1 and IDH2 homodimers raises a question of relative contribution of α-KG reduction and D-2-HG accumulation to tumorigenesis. We showed here that treatment of cells with either 2-HG enantiomer or inhibition of IDH1 and/or IDH2 activity alone in the absence of D-2-HG accumulation causes similar effects of reducing the activity of multiple α-KG-dependent dioxygenases. It is possible that IDH1 or IDH2 mutations alone do not reduce cellular level of α-KG sufficiently low to have a significant tumorigenic consequence, but nonetheless sensitize α-KG-dependent dioxygenases to the inhibitory effect by the large amounts of 2-HG accumulated in the cell. Hence, α-KG reduction and D-2-HG accumulation cooperatively contribute to tumorigenesis. The findings that a cell-permeable α-KG derivative can reverse the effects of 2-HG inhibition and that D-2-HG is a weak competitive inhibitor suggest that drugs mimicking α-KG merit exploration as a therapy for tumors that harbor an IDH1 or IDH2 mutation, either alone or in combination with inhibitors targeting mutant IDH1 and IDH2 to reduce the 2-HG production.

Crystallography Study

Purification and crystallization of His-tagged CeKDM7A were described in Supplemental Experimental Procedures. Data sets were collected on beamline BL17U at Shanghai Synchrotron Radiation Facility (SSRF). All data were processed using the program HKL2000. The crystals of CeKDM7A in complex with D-2-HG contain two molecules in one asymmetric unit. The structures of CeKDM7A with D-2-HG or α-KG were determined by molecular replacement method using CeKDM7A structure and the models were manually built with COOT. All refinements were performed using the refinement module phenix.refine of PHENIX package. The model quality was checked with the PROCHECK program, which showed good stereochemistry according to the Ramachandran plot for all structures. The structure superimposition was performed with COOT. All structure figures were generated by PyMol.

Enzymatic Assays

To assay human JHDM1A/KDM2A demethylase activity toward H3K36me2, His tagged JHDM1A was first obtained by transforming pET28a-JHDM1A into Escherichia coli BL21 and protein expression was induced by addition of 1 mM IPTG at 30° C when cell density reaches 0.5 OD₆₀₀ units. Cells were lysed by sonication and Ni-NTA agarose was used to purify His-JHDM1A fusion proteins. Histone demethylase assay was carried out by incubating 2 μg oligonucleosomes, 4 μg purified His-JHDM1A, and/or 10–50 mM L- or D-2-HG in histone demethylation buffer [50 mM HEPES (pH 8.0), 625 μM Fe(NH₄)₂(SO₄)₂, 0.1–0.5 mM α-KG, 2 mM ascorbate] at 37° C for 2–3 hr and the reactions were stopped by the addition of SDS loading buffer and subsequently analyzed by western blotting using anti-H3K36me2 antibody. To measure CeKDM7A demethylase activity toward H3K9me2 and H3K27me2, two synthetic dimethylated peptides H3K9me2 [ARTKQTARK (me2)STGGKA] and H3K27me2 [QLATKAARK (me2)SAPAS] were used as substrates. Demethylase assays were carried out in the presence of 10 μg enzyme, 1 μg peptide in 20 μl buffer 20 mM Tris-HCl (pH 7.5), 150 mM NaCl, 50 μM (NH₄)₂Fe(SO₄)₂, 100 μM α-KG, 2 mM Vc, 10 mM PMSF for 3 hr. The demethylation reaction mixture was desalted by passing through a C₁₈ ZipTip (Millipore). To examine the inhibitory effect of 2-HG, various concentrations of 2-HG were incubated with KDM7A briefly before adding other reaction mixtures. The samples were analyzed by a MALDI-TOF/TOF mass spectrometer.

Three different assays were performed for TET-catalyzed 5mC-to-5hmC conversion. For in vivo assay using immunofluorescence, plasmids expressing Flag tagged TET proteins were either singularly transfected or cotransfected with plasmid expressing GFP-IDH fusion protein to HEK293T cell. Thirty-six to forty hours after transfection, cells were fixed with 4% paraformaldehyde in PBS for 15 min and then washed with cold PBS. Cells were permeabilized with 0.4% Triton X-100 in PBS for 15 min. For 5mC and 5hmC staining, DNA was denatured with 2 N HCl for 30 min. and then neutralized with 100 mM Tris-HCl (pH 8.5) for 10 min. After washing three times with PBS, samples were blocked for 1 hr with 5% goat serum, 1% BSA 0.05% Tween20 in PBS. The primary antibodies were added and incubated at 4° C overnight. After washing three times with PBS, cells were incubated with secondary antibodies and DAPI for 30 min, followed by triple wash with PBS. Images were recorded using Olympus immunofluorescence microscope DP71 and Olympus software. Antibodies to FLAG (Sigma-Aldrich, diluted at 1:1000), 5-hydroxymethylcytosine (Active Motif, Carlsbad, CA, dilution at 1:4000), 5-methylcytosine (clone 162 33 D3, Calbiochem, San Diego, CA, diluted at 1:500 1000) were purchased commercially.

For dot-blot assays, we followed the procedures described previously (Ito et al., 2010). Briefly, genomic DNA was spotted on nitrocellulose membranes. The membrane was baked at 80° C and then blocked with 5% skimmed milk in TBST for 1 hr, followed by the incubation with the anti-5hmC antibody overnight at 4° C and HRP-conjugated anti-rabbit IgG secondary antibody for 1 hr at room temperature. After washing three times with TBST, the membrane was treated with ECL and scanned by a Typhoon scanner. The quantification of dot-blot was done by Image-Quanta software (GE Healthcare).

In vitro TET-catalyzed 5mC-to-5hmC conversion was assayed as described previously (Ito et al., 2010) and described in detail in the Supplemental Experimental Procedures. Briefly, 5 μg purified proteins were incubated with 0.5 μg double-stranded oligonucleotide substrates in 50 mM HEPES (pH 8), 75 μM Fe(NH₄)₂(SO₄)₂, 2 mM ascorbate, and 0.1 mM α-KG with or without a various amount of 2-HG for 3 hr at 37° C. Oligonucleotide substrates were purified and then digested with MspI. 5′-end of the digested DNA was treated with calf alkaline phosphatase and labeled with [γ-³²P]ATP and T4 polynucleotide kinase. Labeled fragments were ethanol-precipitated and digested with 10 μg of DNase I and 10 μg Phosphodiesterase I in the presence of 15 mM MgCl₂, 2 mM CaCl₂ at 37° C. One microliter digestion product was spotted on a PEI-cellulose TLC plate (Merck) and separated in an isobutyric acid/water/ammonium hydroxide (66:20:2) running buffer.

We thank members of Fudan MCB laboratory for their valuable inputs and support throughout this study, Nara Lee for helping for assaying human histone demethylase activity, staff members of Beamline BL17U at Shanghai Synchrotron Radiation Facility (SSRF) for data collection, and Sara Jackson for the critical reading of the manuscript. This work is supported by the 985 Program from the Chinese Ministry of Education; state key development programs of basic research of China (2009CB918401, 2009CB918600), Chinese National Science Foundation grants (30971485/C0706, 30600112, 30871255, 11079016), National Basic Research Program of China (2011CB965300, 2011CB910600), Shanghai key basic research projects, China (09JC1402300), and grants from NIH (K.L.G. and Y.X.).