Combination Targeted Therapy to Disrupt Aberrant Oncogenic Signaling and Reverse Epigenetic Dysfunction in IDH2- and TET2-Mutant Acute Myeloid Leukemia

Gene discovery studies in acute myeloid leukemias (AML), lymphomas, sarcomas, brain tumors, and epithelial tumors have identified recurrent somatic mutations in genes with a role in DNA methylation (1, 2). The methylation pathway is most commonly disrupted in AML by mutations in TET2, IDH1, IDH2, DNMT3A, and WT1 (3–7). TET2 mutations abrogate 5′-methylcytosine hydroxylase activity, which is required for DNA demethylation, whereas neomorphic IDH1 and IDH2 mutations through 2-hydroxyglutarate (2HG) production competitively inhibit TET enzymatic activity (8–10). Mutations in this pathway occur in more than 30% to 35% of patients with AML, and correlative studies have shown that these disease alleles are associated with distinct transcriptional, epigenetic, and prognostic signatures (11–14). Taken together, these findings demonstrate an important role for mutations in epigenetic modifiers in AML pathogenesis and risk stratification.

Murine and human studies have shown that mutations in TET2 or in IDH1 or IDH2 are most commonly initiating events in malignant transformation that promote hematopoietic stem cell self-renewal and myeloid expansion (15–17). These data suggest that transformation requires the acquisition of additional somatic mutations, which cooperate with mutations in epigenetic regulators to promote leukemogenesis. Consistent with this hypothesis, we and others have shown that Tet2 loss and Idh1 or Idh2 mutations can cooperate with known occurring disease alleles, including Flt3^ITD, to induce AML in vivo (18–20). The resulting leukemic cells continue to bear evidence of epigenetic reprogramming, including extensive hypermethylation and transcriptional silencing of key target loci, including Gata2, which is required for normal hematopoietic differentiation. A feature of Flt3^ITD;Tet2-mutant leukemia is resistant to traditional AML chemotherapy and to FLT3 kinase inhibitors; thus, antiproliferative therapy is insufficient to effectively target leukemias marked by mutations that disrupt DNA demethylation (18, 21). As a consequence, in subsets of patients, mutations in these genes are associated with inferior survival (22–24). Toward the goal of developing epigenetic agents, we therefore sought to characterize therapies that can potentially restore the balance of DNA methylation and demethylation alone and in combination with antiproliferative therapies.

Retrospective studies in myelodysplastic syndrome (MDS) have suggested that TET2-mutant patients are more likely to respond to DNA hypomethylating agents (HMA; refs. 25, 26). However, the mechanisms by which 5-azacytidine (5-Aza) and other HMAs induce responses in MDS and AML have not been fully delineated. We therefore first sought to assess the therapeutic efficacy of 5-Aza in a transplantable model of TET2-mutant AML. We engrafted CD45.2⁺ Flt3^ITD;Tet2-mutant AML cells and CD45.1⁺ support marrow into CD45.1⁺ congenic recipients and allowed recipient mice to develop AML with engraftment of 80% to 90% CD45.2⁺ Flt3^ITD;Tet2-mutant cells and expansion of leukemic blasts in the peripheral blood, bone marrow (BM), and spleen. We then treated mice with vehicle or 5-Aza using a dose and schedule (5 mg/kg days 1–5 of 21-day cycles) similar to the clinical context. This allowed us to assess the impact of 5-Aza on Flt3^ITD;Tet2-mutant AML cells relative to wild-type cells in vivo. Four cycles of 5-Aza therapy resulted in a marked therapeutic response, with near-elimination of leukemic blasts in the BM and spleen, and restoration of megakaryocytes in the BM and germinal centers in the spleen (Fig. 1A). We observed normalization of peripheral blood cell counts and splenomegaly with 5-Aza therapy, with a significant reduction in the total white blood cell count (WBC) and spleen weight and an increase in platelet count and hematocrit (Fig. 1B–D). Morphologic examination and flow cytometric analysis of the peripheral blood confirmed that 5-Aza therapy reduced the proportion of morphologic blasts (Fig. 1C) and significantly reduced the proportion of cKit⁺Mac1⁻ immature cells in the peripheral blood (Fig. 1D and Supplementary Fig. 1A).

The lineage⁻Sca⁺cKit⁺ (LSK) population in Flt3^ITD;Tet2-mutant AML is comprised of monomorphic CD48⁺CD150⁻ multipotent progenitors (MPP), which serve as leukemia stem cells capable of propagating the disease in serial transplantation studies (18). 5-Aza therapy tended to normalize the BM LSK compartment, with a significant reduction in CD48⁺CD150⁻ MPPs and partial restoration of the long-term hematopoietic stem cell (LT-HSC; CD48⁻CD150⁺) compartment over time (Fig. 1E). TET2 has an important role in erythroid differentiation, and Tet2 loss or mutation leads to a reduction in erythroid precursors (27, 28), with near-complete loss of megakaryocyte and erythroid progenitors (MEP) in the Flt3^ITD;Tet2-mutant model (18). 5-Aza treatment led to a reversal of anemia and thrombocytopenia and helped to restore CD71⁺Ter119⁺ erythroid precursors and MEPs (Fig. 1F and Supplementary Fig. S1B). Based on these data, we hypothesized that 5-Aza therapy might induce reversible differentiation of Flt3^ITD;Tet2-mutant AML cells, such that therapeutic efficacy is maintained only with continued treatment. Prolonged therapy (8 cycles, 24 weeks) with 5-Aza induced a durable response, with a sustained reduction in the WBC (Fig. 1G). However, following suspension of therapy, we observed recurrent AML in all treated mice, with an increase in WBC 4 weeks after the last treatment dose and expansion of blasts in the blood that could, however, again be reduced with resumption of 5-Aza therapy (Fig. 1G).

As we previously showed that the leukemic phenotype induced by Tet2 loss and Flt3^ITD is driven by aberrant hypermethylation, we next examined the impact of 5-Aza on the DNA methylation status of CD45.2⁺ Flt3^ITD;Tet2-mutant LSK cells from mice treated with vehicle or 5-Aza, using enhanced reduced representation bilsulfite sequencing (ERRBS; ref. 29). These studies showed that 5-Aza therapy reversed many of the sites with aberrant DNA hypermethylation induced by Tet2 and Flt3^ITD in leukemia derived stem-progenitor cells (Fig. 1H–I and Supplementary Fig. S1C–S1E). Taken together, these data show that 5-Aza restored a significant component of the epigenetic patterning of normal LSK cells, although we observed additional, 5-Aza–induced hypomethylation of cytosine residues beyond those affected by the Tet2 and Flt3^ITD epigenetic program (Fig. 1H and I). These data suggest that DNA hypomethylating agents promote the differentiation of Tet2-mutant AML cells while reversibly restoring normal DNA methylation patterns.

We next investigated the efficacy of targeted epigenetic therapies in IDH2-mutant AML. Preclinical studies in cell lines and patient samples have shown that small-molecule, targeted inhibition of mutant IDH1 and IDH2 can block the production of the oncometabolite 2HG and thereby induce in vitro differentiation (30, 31). Despite these important insights, the impact of small-molecule inhibition of mutant IDH enzymes in vivo, alone and in combination with other targeted AML therapies, has not been characterized in detail. We crossed a conditional mouse model that expresses the Idh2^R140Q AML disease allele from the endogenous locus (Fig. 2A and Supplementary Fig. S2A) to mice with the inducible Mx1-Cre allele and the Flt3^ITD knock-in allele. This allowed us to generate Mx1-Cre Idh2^R140QFlt3^ITD mice in which both mutant disease alleles are expressed from the endogenous loci. Consistent with previous retroviral and transgenic models (19, 20), Mx1-Cre Idh2^R140QFlt3^ITD mice developed AML. The disease was characterized by expansion of blasts in the blood, BM, and spleen, and increased levels of 2HG in the serum, leukocytosis, and splenomegaly (Fig. 2B and Supplementary Fig. S2B–S1C), similar to the phenotype of the Flt3^ITD;Tet2-mutant AML model. The leukemia was characterized by expansion of cKit⁺ cells in the peripheral blood, reduction in erythroid precursors and MEPs, and replacement of the LSK compartment by CD48⁺CD150⁻ MPPs (Fig. 2C and Supplementary Fig. S2D). Importantly, Idh2^R140QFlt3^ITD murine leukemias exhibited a similar hypermethylation phenotype to human IDH2-mutant AML, with similar alterations in locus-specific DNA methylation and gene expression (Fig. 2D and Supplementary Fig. S2E).

We next used this murine model of Idh2-mutant AML to test the impact of small-molecule IDH2 inhibition. IDH2 inhibition dose-dependently inhibited the ability of Idh2^R140QFlt3^ITD-mutant AML cells to serially replate in vitro at doses that correlated with reductions in cellular 2HG levels and was dependent on the presence of the Idh2 mutation (Supplementary Fig. S2F–S2H). We next engrafted CD45.2⁺ Idh2^R140QFlt3^ITD-mutant AML cells and CD45.1⁺ support marrow into CD45.1⁺ recipient mice and assessed the in vivo efficacy of AG-221, the first small-molecule inhibitor of IDH2 in AML clinical trials (32). AG-221 therapy lowered serum 2HG levels in vivo, consistent with on-target inhibition of neomorphic IDH2-mutant enzyme function (Fig. 3A). It reduced the proportion of cKit⁺ cells in the peripheral blood and increased the proportion of granulocyte macrophage progenitors (GMP), but had variable effects on blood counts and the BM (Fig. 3A and Supplementary Fig. S3A–E). Treatment induced a trend toward a decrease in CD48⁺CD150⁻ leukemic stem cells (Supplementary Fig. S3F). Consistent with the flow cytometric data, we saw evidence of maturing cells and partial restoration of splenic germinal centers with AG-221 therapy (Fig. 3B).

We next assessed the impact of small-molecule IDH2 inhibition on DNA methylation in vivo by performing ERRBS on purified CD45.2⁺ Idh2^R140QFlt3^ITD LSK cells treated with 6 weeks of AG-221 or with vehicle. AG-221 therapy induced demethylation of a subset of aberrantly hypermethylated CpGs in leukemia-derived cells (Fig. 3C and Supplementary Fig. S4A–S4C), resulting in reversal of hypermethylation toward wild-type stem cell levels (Fig 3D). AG-221 also induced additional regions of hypomethylation and even a small degree of hypermethylation, suggesting that additional epigenetic reprogramming effects may accompany the suppression of mutant IDH2 activity (Fig. 3C–D).

Both 5-Aza and AG-221 therapy significantly reduced the proportion of lineage-negative immature cells in the BM of Tet2- or Idh2-mutant AML cells, respectively, consistent with a reduction in the proportion of stem-progenitor cells (Fig. 4A). In line with the impact of these two therapies on AML differentiation, we gated on the BM leukemic cell portion (CD45.2⁺) and observed that chronic 5-Aza or AG-221 therapy normalized terminal myeloid maturation, with reversal of aberrant monocytic (Mac⁺Gr1⁻) skewing associated with TET2 mutation (28, 33) and restoration toward normal neutrophil populations (Mac⁺Gr1⁺; Fig. 4B). The data from these two therapeutic studies suggest that epigenetic therapies can restore differentiation and that therapeutic efficacy is associated with prolonged, reversible differentiation of AML cells as also observed in the in vitro setting (31).

We next performed RNA sequencing (RNA-seq) of LSK cells from Idh2^R140QFlt3^ITD mice treated with AG-221 or vehicle and from Tet2^−/−Flt3^ITD mice treated with 5-Aza or vehicle compared with control wild-type LSKs. We interrogated the effects of these two therapies on the expression of a set of genes we previously demonstrated are differentially expressed in Flt3^ITD;Tet2-mutant AML (Supplementary Table S1). AG-221 therapy in the Idh2^R140QFlt3^ITD model, and 5-Aza therapy in the Tet2^−/−;Flt3^ITD AML model significantly attenuated, but did not fully reverse this gene expression signature (Fig. 4C). Unsupervised clustering analysis using all genes with variance in expression (SD >50 percentile) showed that in both cases drug treatment shifted gene expression patterns toward normal stem-progenitor cells, with convergence of the Idh2- and Tet2-mutant transcriptional profiles as illustrated by the almost fully overlapping fields (Fig. 4C). However, the normal stem-progenitor transcriptional pattern is only partially restored by these epigenetic therapies (Fig. 4C). Consistent with these data, we found that therapy with 5-Aza or AG-221 increased the expression of a significant subset of genes that are aberrantly hypermethylated in IDH1- or IDH2-mutant human AML (Fig. 4D). Gene set enrichment analysis (GSEA) of gene expression changes revealed significant enrichment for methylation, hematopoietic stem cell, and leukemic target gene signatures with AG-221 and 5-Aza therapy, thus affecting genes important for differentiation (Supplementary Fig. S5A).

We next assessed whether these epigenetic therapies were able to specifically reduce mutant allele burden. Although both therapies were able to reduce the proportion of morphologic leukemic blasts and restore normal hematopoietic differentiation, neither therapy could reduce a major proportion of CD45.2⁺ cells (Fig. 4E). This is consistent with the persistence of the mutant AML clone (as differentiated cells) at the time of therapeutic response and a predominant in vivo differentiation effect. In addition, analysis of Tet2^−/−Flt3^ITD mice treated with 5-Aza and Idh2^R140QFlt3^ITD mice treated with AG-221 showed near-complete excision of the Tet2 or Idh2 allele without expansion of the residual nonexcised clone (Supplementary Fig. S5B–S5C). We also analyzed mice treated with 5-Aza and AG-221, by CD45.2⁺ leukemic versus CD45.1⁺ wild-type cells, and found that leukemic cells are responsible for changes in differentiation rather than simply wild-type cells expanding to restore normal hematopoiesis (Supplementary Fig. S5D). There was also no change in the expression of wild-type Flt3 with treatment that would account for the change in the leukemic phenotype (Supplementary Fig. S5E; ref. 34).

Given that neither epigenetic therapy alone was sufficient to suppress the malignant clone, we explored whether targeting the FLT3^ITD cooperating oncogene might increase therapeutic efficacy. Combination therapy with 5-Aza and the FLT3 inhibitor AC220 in Flt3^ITD;Tet2-mutant AML showed significant responses compared with vehicle or monotherapy, with reductions in spleen weight, leukocytosis, BM monocytosis, and immature or blast-like cells (Fig. 5A and Supplementary Fig. S6A–S6C). We next assessed the impact of combination therapy with AG-221 and AC220 in Idh2^R140QFlt3^ITD-mutant AML. Combined therapy with AG-221 and AC220 was more effective in reducing leukocytosis compared with AG-221 monotherapy (Fig. 5B). Importantly, combined AG-221 and AC220 therapy resulted in a further reduction in leukemic blasts in the liver, spleen, and BM compared with either therapy alone (Fig. 5C and Supplementary Fig. S7A). Moreover, combined therapy with AG-221 and AC220 improved platelet counts, increased the proportion of CD71⁺Ter119⁺ erythroid progenitors, and reduced peripheral blood immature cKit⁺ cells that neither monotherapy could completely achieve independently (Supplementary Fig. S7B–S7D). Most importantly, combined 5-Aza + AC220 or AG-221 + AC220 therapy altered hematopoietic differentiation and reduced the proportion of CD48⁺CD150⁻ MPPs compared with monotherapy, consistent with a cooperative effect of combined therapy at reducing the proportion of leukemia stem cells in vivo (Fig. 5D–E and Supplementary Fig. S7E–S7J).

We next investigated the impact of combined signaling and epigenetic therapy on DNA methylation in AML stem-progenitor cells. Combined therapy with AG-221 and AC220 resulted in more profound reversal of aberrant hypermethylation compared with AG-221 or AC220 monotherapy (Fig. 6A and Supplementary Table S2A). This includes a subset of target loci for which reversal of hypermethylation was achieved only by targeting both IDH2 and FLT3 (Fig. 6B). Similar effects were observed with 5-Aza + AC220 treatment of Tet2^−/− Flt3^ITD mice (Fig. 6A and Supplementary Fig. S8A). Combined treatment yielded proportionally greater hypomethylation and more significant reversal of aberrantly hypermethylated genes than either drug alone (Supplementary Table S2B). We previously showed that hypermethylation of the Gata2 locus is induced by concurrent mutations in Tet2 and Flt3^ITD, and we have shown that Gata2 silencing is required for AML maintenance in Flt3^ITD;Tet2-mutant AML in vivo (18). Combination therapy with AG-221 + AC220 or 5-Aza + AC220 potently reversed aberrant DNA hypermethylation at the Gata2 locus in Tet2- and Idh2-mutant AML stem-progenitor cells (Fig. 6C and Supplementary Fig. S8B), demonstrating the ability of combination therapy to restore more normal epigenetic regulation at key target loci in Tet2- or Idh2-mutant AML. Demethylation was also associated with a trend in increased Gata2 expression (Supplementary Fig. S8C–S8D). To assess the correlation of methylation with expression more globally, we annotated differentially methylated regions (DMR) to genes and correlated methylation status with differential RNA expression compared with wild-type. Genes that contain hypermethylated or hypomethylated regions are statistically different in RNA expression, and as expected hypermethylation resulted in a relative decrease in gene expression compared with wild-type (Supplementary Fig. S8E). We also ranked the genes by their log-fold change values and then checked to see if genes that contain hyper or hypo DMRs are enriched in the positive or negative leading edge of this rank list according to GSEA. We found that hypermethylation was associated with a gene set demonstrating decreased expression (Supplementary Fig. S8F).

We next assessed the impact of combination therapy on the relative fitness of AML cells and wild-type hematopoietic cells in vivo. Combination therapy with AG-221 + AC220 or 5-Aza + AC220 resulted in a reduction in CD45.2⁺ leukemic cells and an increase in CD45.1⁺ normal hematopoietic cells in the BM and spleen to an extent not observed with either agent given as monotherapy (Fig. 6D and Supplementary Fig. S9A–S9C). In the case of AG-221 + AC220, there was also a small expansion of the unexcised Idh2-mutant allele clone in BM progenitor cells, further supporting loss of fitness of Flt3^ITD;Idh2^R140Q-mutant clone with combination therapy, but not with AG-221 or AC220 monotherapy (the minute fraction of unexcised Tet2 clone was too small to consistently measure; Supplementary Fig. S9D–S9E). These data demonstrate that combination epigenetic and signaling therapy can specifically target the mutant clone in vivo and abrogate the competitive advantage of AML cells and promote the expansion of normal hematopoietic cells.

The identification of neomorphic, gain-of-function mutations in IDH1, IDH2, and EZH2 has led to the first clinical trials of molecularly targeted epigenetic therapies for human cancers. As these agents enter the clinic, it is critical to better elucidate their mechanisms of action in different malignant contexts, and to use insights from preclinical and clinical trials to understand how best to use these agents. Our studies show that epigenetic therapies, specifically HMAs in TET2-mutant AML and AG-221 in IDH2-mutant AML, achieve therapeutic efficacy through induction of differentiation and reversal of aberrant methylation patterns in AML stem-progenitor cells, and that this results in progressive differentiation of mutant AML cells at the time of clinical response. Changes in methylation were associated with gene expression changes in stem-progenitor cells, although other studies also suggest the increased importance of this methylation change in programming progeny cells, setting the potential for more dramatic changes in expression in later differentiated cell types (35).

The recent report of effectiveness of the HMA decitabine in AML with its increased potency toward affecting methylation underscores the potential for combination therapies using epigenetic agents with other antileukemic therapies in AML (36, 37). The nature, kinetics, and mechanisms of therapeutic response to these agents are distinct from those observed with cytotoxic therapies and kinase inhibitors; as such it will require thoughtful use of specific criteria to measure and characterize response in the clinical context.

These data suggest the mechanism of action of these agents is to achieve therapeutic response through ongoing, chronic differentiation of mutant leukemia stem-progenitor cells. Previous studies and modeling approaches in patients with chronic myeloid leukemia treated with tyrosine kinase inhibitors (TKI) have shown that mutant stem-progenitor cells can persist for years in the setting of clinically effective targeted therapy (38), of which only a subset of patients achieve sufficient responses to maintain remissions after cessation of TKIs (39). Moreover, recent studies have shown that clonal hematopoiesis, driven by mutations in TET2, DNMT3A, IDH2, and other leukemia disease alleles, can persist for years as a preleukemic clonal state or in patients with prolonged clinical remission after cytotoxic chemotherapy (16, 17, 40–42). Detailed studies in clinical trial cohorts are needed to delineate the impact of HMAs in TET2-mutant AML and AG-221 in IDH2-mutant AML on mutational burden and clonal structure, and to determine if mutant cells can persist in a clonal, nonleukemic state in the setting of epigenetic therapies which induce differentiation of leukemia cells. Furthermore, although we observe a change in methylation with 5-Aza, its effect on RNA biology may also play a role in response (43, 44). Similarly, IDH2 inhibitors may affect other α-ketoglutarate–dependent enzymes beyond TET2 and its effects on DNA methylation that may alter treatment response (45). Although these two leukemia models with parallel and replicate experiments show similar responses, these models do not capture the full complexity of human AML, and thus not all human TET2- and IDH2-mutant leukemias may respond similarly through a differentiation effect or to combination therapy. Also, given the limitations of our model, correlations with survival were not determined, and human trials will be needed to determine whether these treatments can translate to a meaningful survival advantage in patients with de novo or relapsed, refractory AML.

Our data suggest that these agents have the potential for significant clinical activity without significant toxicity and that their effects can be potentiated through mechanism-based combinations with other effective anticancer therapies. We show that combined, targeted inhibition of oncogenic kinase signaling and of mutations that drive aberrant DNA methylation can increase therapeutic efficacy, and suggest the need for next-generation trials based on rational combination therapeutic approaches. Most importantly, the ability of epigenetic therapies to reversibly induce differentiation of cancer cells may serve as a platform by which these agents can increase the efficacy of different classes of anticancer agents, including small-molecule kinase inhibitors, cell surface targeting therapies, and immunotherapy.

The conditional Vav-cre⁺ Tet2^fl/flFlt3^ITD mice were previously described (18). The Idh2^R140Q mutation was targeted by a codon change from CGA to CAA in exon 4 (see Supplementary Material for details). The generated knock-in mouse was then crossed to the Mx1-cre and Flt3^ITD mice (The Jackson Laboratory). Cre expression was induced by 5 intraperitoneal injections of poly(I:C) (Invivogen) every other day at a dose of 20 μg/g starting 4 to 5 weeks after birth. All animal procedures were conducted in accordance with the Guidelines for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committees at MSKCC.

Dissected femurs and tibias were isolated. BM was flushed with a syringe into RPMI 10% FCS media or centrifuged at 8,000 rpm for 1 minute. Spleens were isolated and single-cell suspensions were made by mechanical disruption using glass slides. Cells were passed through a 70-μm strainer. Red blood cells were lysed in ammonium chloride–potassium bicarbonate lysis buffer for 10 minutes on ice. A total of 1 × 10⁶ cells were transplanted via tail-vein injection into lethally irradiated (2 × 550 Rad) C57BL/6 or CD45.1 host mice (The Jackson Laboratory).

Mice were treated daily using oral gavage with AC220 at 10 mg/kg, suspended in 5% 2-hydroxypropyl-β-cyclodextrin. 5-Aza (Sigma) was administered at 5 mg/kg dissolved in PBS using IP injection. Treatment was daily for 5 days every 21 days, for four cycles. AG-221 (provided by Agios) was administered twice daily using oral gavage at 40 mg/kg to 100 mg/kg, suspended in 0.5% methylcellulose and 0.2% Tween 80, for 4 to 6 weeks (no differences were observed between these dose levels). Pathology was obtained after fixation in 4% paraformaldehyde and slides stained with hematoxylin and eosin. Blood counts were obtained on an IDEXX ProCyte machine. Each treatment experiment (single or combination therapy) was performed with different donor disease mice (with more than 3 different donor disease mice used for each leukemic genotype) and repeated with representative experiment shown.

Cell populations were sorted using BD FACSAria and RNA isolated using TRIzol LS and RNeasy/AllPrep (Qiagen). RNA was prepared using RiboMinus from Life Technologies. The library was sequenced using Illumina HiSeq. Aligned RNA was analyzed for fold change. The raw output BAM files from the sequencer were first converted to FASTQ format using PICARD SamToFastq. Adapters were removed using cutadapt and any read shorter than 35 bp was discarded. The trimmed reads were then mapped to the Mouse genome (mm9) using TopHat (ver. 2) with a transcriptome index using the Ensembl GTF (ver 37.67) and specifying a strand-specific library. Reads that were unmapped by TopHat were then remapped using BWA bwasw (ver. 0.5.9) again to the mouse genome. The mapped reads from the two passes were merged and then quantitated with HTSeq-count with stranded and strict-intersection options.

ERRBS was performed using a protocol previously described (29). Briefly, genomic DNA was digested with MspI. DNA fragments were end-repaired, adenylated, and ligated with Illumina kits. Library fragments of 150–250 bp and 250–400 bp were isolated. Bisulfite treatment was performed using the EZ DNA Methylation Kit (Zymo Research). Libraries were amplified and sequenced on an Illumina HiSeq. Differential methylation analysis was performed on the resulting ERRBS data using methylKit (46) for finding single nucleotide differences (q-value < 0.01 and methylation percentage difference of at least 25%) and eDMR (47) for finding differentially methylated regions. Gene annotations were carried out by overlapping the CpG sites and regions with the exons, introns, UTR5, UTR3 and 5-kb upstream regions of RefSeq genes for genome assembly mm9. Synergy was determined using a generalized linear model, focusing on loci that that show differential methylation, using an adjusted P value of < 0.01 and a synergy level of at least 20% methylation (absolute methylation difference of 20% above the sum of methylation differences of both single drugs). The data for these analyses are deposited at GEO under GSE78690, GSE78691, and GSE86952.

Antibody staining and FACS analysis were performed as previously described (15). Cells were lysed in ACK lysis buffer except for CD71 Ter119 staining. Stem cell enrichment was performed using the Progenitor Cell Enrichment Kit (Stem Cell Technologies). See Supplementary Material for details.

2HG was measured as previously described (31). LC-MS/MS analysis was performed using an AB Sciex 4000 operating in negative electrospray mode. Multiple reaction monitoring data were acquired for each compound, using the following transitions: 2HG (146.9/128.8 amu), ¹³C₅-2HG (151.9/133.8 amu), and 3HMG (160.9/98.9 amu). Chromatographic separation was performed using an ion-exchange column (Fast Acid analysis, 9 μm, 7.8 × 100 mm; BioRad).

E.M. Stein is a consultant/advisory board member for Celgene and has given expert testimony for Agios. J. Travins has ownership interest (including patents) in Agios Pharmaceuticals. K. Straley has ownership interest in Agios stock. C.B. Thompson is on the board of directors of Merck and Charles River Laboratories; has ownership interest (including patents) in Merck, Agios, Charles River Laboratories, University of Michigan, University of Pennsylvania, and University of Chicago; and is a consultant/advisory board member for Agios. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.H. Shih, E.M. Stein, C. Mason, C. Thompson, A. Melnick, R.L. Levine

Development of methodology: A.H. Shih, C. Meydan, P.S. Ward, J. Travins, C. Mason, A. Melnick

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.H. Shih, K. Shank, F.E. Garrett-Bakelman, P.S. Ward, A.M. Intlekofer, A. Nazir, E.M. Stein, K. Knapp, K. Straley, A. Melnick

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.H. Shih, C. Meydan, F.E. Garrett-Bakelman, J. Glass, C. Gliser, C. Mason, K. Yen, A. Melnick, R.L. Levine

Writing, review, and/or revision of the manuscript: A.H. Shih, C. Meydan, F.E. Garrett-Bakelman, J. Glass, C. Mason, K. Yen, C.B. Thompson, A. Melnick, R.L. Levine

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Meydan, F.E. Garrett-Bakelman, A. Nazir, K. Knapp, C. Gliser, C. Mason

Study supervision: C. Mason, A. Melnick, R.L. Levine

We thank the MSK IGO for assistance with sequencing and mutation analysis, MSK Flow Cytometry core, and Caroline Sheridan of the Weill Cornell Medicine Epigenomics Core.

This work was supported by a Gabrielle's Angel Fund grant to R.L. Levine and A. Melnick, a Leukemia and Lymphoma Society (LLS) Translational Research grant to R.L. Levine, grant CA172636-01 to R.L. Levine and A. Melnick, grant CA-1U54OD020355-01 to R.L. Levine, a supplement grant to P30 CA008748 to R.L. Levine and C.B. Thompson, and by the Samuel Waxman Cancer Research Center. C.E. Mason is supported by R01HG006798 and R01NS076465; MSKCC cores are supported by P30 CA008748. A. Melnick is a Burroughs Wellcome Clinical Translational Scholar and is supported by the Sackler Center for Biomedical and Physical Sciences. A.H. Shih is supported by the Conquer Cancer Foundation, LLS, and NCI K08CA181507. F.E. Garrett-Bakelman is supported by NCI K08CA169055 and the American Society of Hematology (ASHAMFDP-20121) under the ASH-AMFDP partnership with The Robert Wood Johnson Foundation. R.L. Levine is a Scholar of the Leukemia and Lymphoma Society.

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.