The Folate Cycle Enzyme MTHFR Is a Critical Regulator of Cell Response to MYC-Targeting Therapies

The rewiring of cellular metabolic activities represents a major determinant of cancer progression and is considered a hallmark of cancer (1). Metabolic reprogramming supports the acquisition and maintenance of malignant properties. The integration of high-throughput omics technologies such as metabolomics and loss-of-function screening has revolutionized our understanding of how metabolic dependencies support cancer cell proliferation (2–4). Although nonproliferating cell activity depends primarily on catabolic demands, proliferating cells must balance the divergent catabolic and anabolic requirements of sustaining cellular homeostasis while duplicating mass, and thus become dependent on a plethora of metabolic pathways not typically essential for normal tissue maintenance (5). This peculiar metabolic rewiring of neoplastic cells engenders metabolic liabilities that can be exploited to design innovative therapeutic strategies, notably to increase the therapeutic index of existing anticancer therapies. For instance, it was recently observed that dietary supplementation of histidine enhances leukemic cell sensitivity to the widely used chemotherapeutic agent methotrexate, an antimetabolite that inhibits de novo nucleotide synthesis (6). This suggests that dietary supplementation can be leveraged to curtail the toxicity of anticancer therapies while maximizing on-target activity.

In addition to diet-mediated enhancement of antimetabolite-based chemotherapy, metabolic perturbation may substantially influence response to therapies targeting oncogenes involved in hijacking of neoplastic cell metabolism. In that regard, MYC represents a paradigmatic oncogene, as this transcription factor–encoding gene is deregulated in more than 50% of human cancers and reprograms many aspects of cell metabolism including glucose uptake and glycolysis, glutaminolysis, serine/glycine metabolism, and lipid biosynthesis (7). A general feature of MYC deregulation is its transcriptional regulation by superenhancer genomic regions. These clusters of enhancers are densely occupied by transcription factors and chromatin regulators, including BET bromodomain proteins and CDK7 and CDK9 kinases, and have been exploited over the past decade as essential targets owing to their MYC transcriptional regulator function. Therefore, indirect targeting of MYC transcription through inhibition of these essential regulators has shown great promise in preclinical studies, notably in acute myeloid leukemias (AML) harboring MLL gene fusions (8). Inhibitors targeting these chromatin remodelers are currently in phase I to II clinical trials in advanced solid tumors and hematologic malignancies (8–10).

The diverse metabolic alterations induced by MYC may constitute another source of unique metabolic rewiring that could be exploited to nominate new treatment strategies. These approaches may enhance cell susceptibility to its inhibition by this new class of inhibitors or enable identification of patient populations most likely to benefit from MYC-targeting therapies. Here, we present a road map for the identification of metabolic predictors of resistance to diverse chromatin-remodeling drugs in cancers. We integrate omic profiling and shRNA- and CRISPR–Cas9-based functional approaches to identify and delineate mechanisms by which metabolism perturbation affects response to MYC-targeting therapies.

BET proteins interact with acetylated histones in active regulatory domains (promoters and enhancers) and promote RNA PolII activity. Despite the general nature of this mechanism, BET inhibitors show selective effects on gene expression through suppression of MYC and MYC-related transcriptional programs (8, 11, 12). Two AML patient cohorts (TCGA-LAML, n = 198 and GSE14468, n = 526) were queried with multiple gene sets related to MYC transcriptional programs as well as vitamin, amino acid, and nucleotide metabolism using single-sample gene set enrichment analysis (ssGSEA). Correlation between gene sets was queried across all patients from each cohort to pinpoint potential cross-talk between active MYC programs and specific metabolic pathways (Fig. 1A; Supplementary Table S1). Whereas many pathways related to nucleotide (pyrimidine, purine), tetrahydrofolate salvage, lysine, cysteine/methionine, arginine/proline, alanine/aspartate/glutamate, and valine/leucine/isoleucine metabolism were strongly correlated with active MYC signatures, those associated with folate cycle and biosynthesis, histidine, glycine/serine/threonine, beta-alanine, phenylalanine/tyrosine, and tryptophan metabolism were all poorly correlated with active MYC signatures, suggesting that these pathways are not regulated by MYC and thus may be hijacked by cancer cells to undergo metabolic rewiring and escape from drug-induced MYC inhibition (Fig. 1B). To investigate this, three AML cell lines (KG1a, IMS-M2, and U937) were grown in media deprived of amino acids or vitamins belonging to the metabolic pathways poorly correlated with active MYC signatures and were then treated with the BET inhibitor OTX015 (ref. 13; Fig. 1C). Although phenylalanine and histidine starvation increased resistance to OTX015 in a subset of AML cell lines, folic acid starvation consistently enhanced resistance to OTX015 in all cell lines tested.

To investigate further the relationship between folic acid starvation and increased BET inhibitor resistance, we compiled a panel of AML cell lines harboring diverse genetic alterations. Supporting our previous observation, folic acid–starved media significantly increased the area under curve (AUC) and half-maximal inhibitory concentration (IC₅₀) of OTX015 versus standard media (Fig. 1D). The effect of folic acid removal was also evaluated in cell lines treated either with a second BET inhibitor, JQ1, or with THZ-1, a covalent CDK7 inhibitor, another key factor in transcriptional elongation at superenhancer regions and a strong inhibitor of MYC expression (ref. 14; Fig. 1E; Supplementary Fig. S1). In both conditions, folic acid withdrawal significantly dampened AML cell line sensitivity to JQ1 and THZ-1. Previous studies have focused on the role of a specific BET family member, BRD4, which is potently inhibited by JQ1 and OTX015 and serves as one of the main targets triggering their deleterious effects. Therefore, we studied selective suppression of BRD4 by shRNA in two AML cell lines. As seen with JQ1 and OTX015 treatments, there was a striking viability decrease upon BRD4 knockdown, which was significantly attenuated with folic acid starvation (Fig. 1F). To further characterize the phenotypic consequences of folic acid withdrawal on the sensitivity of AML cells to OTX015, we performed colony formation assays and established that a 125-fold decrease in folic acid concentration (from 1 mg/L to 0.008 mg/L in standard vs. conditioned media, respectively) significantly increased U937 cell colony formation in the presence of OTX015 (Fig. 1G). This was accompanied by a decrease in OTX015-induced apoptosis (Fig. 1H). Moreover, folic acid withdrawal markedly increased the IC₅₀ of OTX015 in a cohort of MLL-translocated primary patient samples with AML (Fig. 1I).

To control for bias of in vitro culture conditions on the increased BET inhibitor resistance induced by folate starvation, mice injected with MLL–AF9-driven AML cells were fed a regular or no folic acid diet before JQ1 treatment. Intracellular folic acid concentrations approximated 5–6 ng/mL/10⁶ cells irrespective of whether extracellular folic acid levels were physiologic (63 ng/mL in mouse serum) or supraphysiologic (1,000 ng/mL in standard media; Supplementary Table S2). From 40 ng/mL folic acid to complete folic acid withdrawal in the culture media, we observed a similar 2.5-fold reduction in intracellular folic acid as in leukemic blasts harvested from mice fed with a no folic acid diet. Folic acid levels in mouse serum, liver, and red blood cells were significantly decreased in animals fed for 2, 4, and 8 weeks with the folic acid–restricted diet compared with those fed a regular diet (Supplementary Fig. S2A). This was associated with increased homocysteine levels (hyperhomocysteinemia) in plasma versus those fed a regular diet, confirming that folate starvation recapitulated physiologic features reported in previous studies (refs. 15, 16; Fig. 1J). Over 8 weeks, folate withdrawal did not cause a significant drop in hematocrit, white blood cell count, or weight (Supplementary Fig. S2B). Consistent with our in vitro results, the antileukemic activity of JQ1 was markedly attenuated in animals fed the no-folate diet versus those fed a regular diet (Fig. 1K). This was associated with a significant decrease in the overall survival of folic acid–restricted mice treated with JQ1 in comparison with their regular diet–fed counterparts (Supplementary Fig. S3).

Dietary folic acid undergoes several conversion steps into intermediate metabolites before generation of 5,10-methylene-THF (5,10-CH₂ THF). 5,10-CH₂ THF is then either reduced in the folate cycle or oxidized in the formate-producing tetrahydrofolate salvage cycle, per cellular needs. Despite the fact that folic acid can be processed through these two interconnected pathways, we observed in our previous ssGSEA analyses that active MYC gene signatures were highly correlated with activation of tetrahydrofolate salvage cycle, but not folate cycle–related gene sets (Fig. 1A and B). This implies some negative regulatory relationship between MYC and the folate cycle, prompting us to investigate further the consequences of the suppression of each enzyme of the folate cycle on the sensitivity to OTX015. Five central enzymes control the multistep conversion of dietary folate into 5,10-methylene-THF (5,10-CH₂ THF) before its reduction to 5-methyl-THF (5-CH₃ THF) and recycling to THF (Fig. 2A). We infected IMS-M2 AML cells with two hairpins that markedly reduced their expression levels (Fig. 2B). Although hairpins targeting the RNAs encoding serine hydroxymethyltransferase 1 (SHMT1) and 5-methyltetrahydrofolate-homocysteine methyltransferase (MTR) did not markedly alter OTX015 response, those targeting the RNAs encoding two dihydrofolate reductase isoenzymes (DHFR and DHFR2) as well as 5,10-methylenetetrahydroflate reductase (MTHFR) which expectedly reduced the THF:folate and 5-CH₃ THF:THF ratios, respectively, substantially increased the OTX015 IC₅₀ by averages of 3- and 7-fold, respectively (Fig. 2C; Supplementary Fig. S4A and S4B). On the basis of these observations, the combination of OTX015 with the DHFR inhibitor methotretrexate was antagonistic at a wide range of concentrations across four AML cell lines: OCI-AML2, KG1a, U937, and IMS-M2 (Supplementary Fig. S5).

Given that MTHFR is the rate-limiting folate cycle enzyme and that its knockdown induced the most striking decrease in OTX015 sensitivity in IMS-M2 cells, we pursued our investigation of the effect of the two MTHFR-directed shRNAs on OTX015 response by interrogating the same AML cell lines (Fig. 2D–F) and confirmed that MTHFR suppression substantially weakened the response of all tested cell lines to increasing doses of OTX015 (Fig. 2E). The folate cycle is intimately interconnected to the methionine cycle through which it regulates methionine and homocysteine levels. Given the interdependence of both cycles, we investigated whether alteration of the methionine cycle could affect OTX015 response in AML cells as consistently as alteration of the folate cycle. We modulated the expression of S-methyl-5′-thioadenosine phosphorylase (MTAP), a critical enzyme involved in methionine salvage, in IMS-M2 and U937 AML cell lines, using two MTAP-directed shRNAs exhibiting different knockdown effectiveness on MTAP expression (Supplementary Fig. S6A). Although OTX015 resistance was enhanced by MTAP suppression in both cell lines, more sustained MTAP knockdown was required to decrease OTX015 sensitivity in U937 cells (Supplementary Fig. S6B). Given that MYC can control antiapoptotic programs, we also explored whether MTHFR suppression antagonized inhibition of the antiapoptotic protein BCL2 by venetoclax. MTHFR knockdown did not affect viability response nor cytochrome c release in two AML cell lines tested (Supplementary Fig. S7A and S7B).

Finally, we abrogated Mthfr expression in primary murine MLL–AF9-driven AML cells using two Mthfr-directed shRNAs before treatment of engrafted mice with JQ1 (Fig. 2G–I). In animals transplanted with Mthfr-depleted blasts, leukemic burden was not significantly decreased by JQ1 treatment compared with control counterparts which exhibited striking sensitivity to JQ1 (Fig. 2H). Thus, the overall survival of mice injected with Mthfr-depleted cells was significantly shorter than mice with control leukemic cells after JQ1 treatment (Fig. 2I). Similarly, Mthfr knockdown significantly reduced the antileukemic effect of JQ1 in an additional AML mouse model driven by the CBFB–MYH11 fusion oncogene (Supplementary Fig. S8). Collectively, these data nominated MTHFR as a critical folate-cycle enzyme whose impairment increases resistance to MYC-targeting therapies.

Previous studies have extensively characterized two common MTHFR genetic variants, C677T and A1298C, encoding MTHFR enzyme variants with reduced activity, resulting in hyperhomocysteinemia in humans. In particular, homozygous variant 677 TT and 1298 CC genotypes are present in approximately 10% of Caucasians and display only 30% and 60% of the homozygous wild-type 677 CC and 1298 AA enzyme activities, respectively (17, 18). To explore the impact of these MTHFR variants on BET inhibitor resistance, we deployed CRISPR–Cas9-based genome editing in KG1a cells to introduce these two nonsynonymous single-nucleotide polymorphisms in MTHFR, thereby generating isogenic cell lines exhibiting all variant combinations. Five KG1a clones either (i) wild-type 677 CC and 1298 AA, (ii) heterozygous 677 CT or 1298 AC, or (iii) homozygous 677 TT or 1298 CC were selected and genotyped via an allelic discrimination technique (Fig. 3A). Although clones heterozygous for any of the two variants had similar sensitivity to OTX015 as wild-type clones, 677 TT and 1298 CC homozygous KG1a clones were significantly more resistant to OTX015 than their wild-type counterparts (Fig. 3B). Furthermore, colony numbers from 677 TT and 1298 CC homozygous KG1a clones were substantially higher than those of wild-type KG1a clones upon OTX015 treatment (Fig. 3C). On the basis of these results, we divided a cohort of 16 MLL-translocated primary AML patient cells into two subgroups by MTHFR genotype: wild-type and single-heterozygous versus homozygous and compound heterozygous for any of two MTHFR variants (Supplementary Table S3). Patients displaying homozygous or compound heterozygous MTHFR genotypes for any of the two variants responded significantly less to OTX015 than those with wild-type homozygous and heterozygous MTHFR genotypes (Fig. 3D). Using the same criteria of patient subclassification based on MTHFR variant status, we observed the same significant trend of OTX015 response in a second cohort of Core Binding Factor (CBF) patients with AML (Supplementary Fig. S9A; Supplementary Table S4). Finally, we engineered a tractable MLL–AF9-driven mouse model of AML in which Mthfr genotype status was either homozygous wild-type, heterozygous, or homozygous knockout. We established that loss of a single Mthfr copy, which phenocopies in mice a partial impairment in MTHFR activity caused by nonsynonymous single-nucleotide polymorphisms on MTHFR (19), was sufficient to decrease overall leukemia burden and, most importantly, attenuate JQ1 sensitivity in MLL–AF9-driven leukemias (Fig. 3E).

Because MTHFR impairment promotes BET inhibitor resistance, we hypothesized that wild-type MTHFR overexpression may conversely enhance sensitivity of AML cells lacking proper MTHFR function due to the presence of the two genetic C677T and A1298C MTHFR variants. Exogenous DD-tagged wild-type MTHFR was overexpressed in AML cells exhibiting various 677 CC and 1298 AA genotypes (Fig. 3F). This markedly increased OTX015 sensitivity of AML cell lines displaying homozygous and compound heterozygous MTHFR genotypes for any of the two variants in comparison with cell lines exhibiting either a wild-type or heterozygous 677 CT genotype (Fig. 3G). A similar analysis performed on the isogenic MTHFR-edited KG1a clones showed that those displaying homozygous 677 TT or 1298 CC genotypes were significantly more sensitized to OTX015 upon overexpression of wild-type MTHFR than their homozygous wild-type 677 CC and 1298 AA counterparts (Fig. 3H). We then reasoned that supplementation with the end-product metabolite synthesized by MTHFR, 5-CH₃ THF, may induce a similar effect on OTX015 response as wild-type MTHFR overexpression. Despite the increased IMS-M2 and U937 cell line resistance to OTX015 in response to MTHFR depletion, we thus observed that treatment with exogenous 5-CH₃ THF restored a cytotoxic effect of OTX015 similar to that observed in shCT-infected cells (Fig. 3I). Consistent with this observation, exogenous 5-CH₃ THF substantially enhanced response to OTX015 of homozygous 677 TT or 1298 CC MTHFR KG1a clones compared with their wild-type MTHFR counterparts (Fig. 3J). In addition, exogenous 5-CH₃ THF sensitized primary patient cells exhibiting an altered MTHFR genotype to OTX015 (Fig. 3K). We finally asked whether the 5-CH₃ THF supplementation could improve sensitivity in all individuals or only in those who are folate-deficient by treating wild-type, heterozygous or homozygous Mthfr knockout MLL–AF9-driven cells with increasing concentrations of 5-CH₃ THF (Supplementary Fig. S9B). Although the sensitivity to OTX015 of wild-type Mthfr cells was not affected by a wide range of 5-CH₃ THF concentrations, a 4-fold increase in 5-CH₃ THF levels (from approximately 3 to 12 μg/mL) was required to sensitize cells whose Mthfr expression was decreased from 50% to 100%, respectively, to OTX015.

We next used metabolism profiling to characterize the effect of folate pathway alteration on global cell metabolism in U937 cells. Steady-state levels of 113 metabolites highly enriched in pathways either directly coupled to the folate cycle such as pyrimidine metabolism as well as histidine, glycine/serine/threonine, and cysteine/methionine/taurine/glutathione metabolism, or indirectly associated with folate metabolism including urea cycle–related pathways such as arginine/proline, alanine, aspartate, glutamate, and beta-alanine metabolism, were significantly changed by folate starvation (Fig. 4A; Supplementary Fig. S10A; Supplementary Table S5). Interestingly, this revealed a significant increase in S-adenosylhomocysteine (SAH), an intermediate in homocysteine synthesis from all methylation reactions involving the methionine-derived metabolite S-adenosylmethionine (SAM) as a methyl donor. shRNA-mediated MTHFR depletion induced a similar increase in SAH levels as that observed upon folate withdrawal in the AML cell lines U937 and IMS-M2 (Fig. 4B). This was associated with a reduction in SAM production and SAM:SAH ratio (Supplementary Fig. S10B and S10C). Finally, SAH supplementation increased OTX015 IC₅₀ by a magnitude similar to that measured upon folate starvation in KG1a and U937 cells (Fig. 4C). Of note, depleting methionine in media of IMS-M2 cells (by 300-fold in methionine-low “L” condition) enhanced the cytoprotective effect of SAH against OTX015 to a similar extent as that induced by folic acid starvation. In U937 and KG1a cells, this caused even greater resistance to OTX015 than folate restriction alone. Together, these data suggest that increased intracellular SAH is critical in mediating OTX015 resistance in response to impaired folate cycle activity.

SAH is a potent inhibitor of SAM-dependent methylation reactions (20). We thus hypothesized that increased SAH induced by folate cycle suppression may directly affect methylation on histone H3. Methylation profiling on H3 histone marks revealed that folic acid starvation significantly decreased dimethylation (me2) and trimethylation (me3) of H3K9 and H3K27, respectively (Fig. 4D). Furthermore, MTHFR impairment reduced H3K27me3 and H3K9me2 in human and murine AML cells (Fig. 4E–G). This decreased methylation status resulted from a significant drop in H3K27 and H3K9 methyltransferase activity upon folate starvation or MTHFR depletion (Fig. 4H).

To identify precisely the class of H3K27 and H3K9 methyltransferases whose decreased activity affects sensitivity to BET inhibitors, we deployed a CRISPR–Cas9 screening strategy using a dedicated library targeting 325 epigenetic regulators including DNA writers and erasers such as histone methylases/demethylases to search for depletion or enrichment of specific guides under a single JQ1 concentration (Supplementary Table S6). Single guide RNA (sgRNA) library–transduced cells were treated with JQ1 or DMSO and sampled weekly for 2 weeks of treatment to determine the composition of the sgRNA pools by deep sequencing. We identified the gene encoding a core component of the H3K27-methyltransferase PRC2 complex EED and the genes encoding two other SET-domain-containing H3K9 methyltransferases, EHMT1 and SETDB1, as top rescuer hits whose knockout rescued cells from JQ1 (Fig. 4I; Supplementary Table S7). Conversely, guides targeting three H3K9 demethylase enzyme–encoding genes, KDM1A, KDM3B, and JMJD1C, were relatively depleted in the presence of JQ1 versus control, suggesting that their loss sensitized cells to JQ1. To confirm this, two hairpins that reduced EED, EHMT1, and SETDB1 protein levels were transduced in U937 cells and markedly alleviated OTX015 sensitivity, phenocopying the cytoprotective effect of MTHFR depletion previously observed in BET inhibitor–treated cells (Fig. 4J; Supplementary Fig. S11A). Besides the critical function of EED in scaffolding the PRC2 complex, another subunit (EZH2, which can be directly targeted by small-molecule inhibitors such as EPZ-6438) triggers the catalytic methyltransferase activity of PRC2. OTX015 treatment in combination with EPZ-6438 was antagonistic at a wide range of concentrations in three AML cell lines, consistent with the identification of H3K9 and H3K27 methyltransferases, either alone or in complex, as critical mediators of response to BET inhibitors (Supplementary Fig. S11B).

Our data imply that folate cycle suppression reduces H3K27 and H3K9 methylation. This suggests that folate cycle disruption may induce transcriptomic and epigenetic rewiring of AML cells to activate a specific transcriptional program that mitigates BET inhibitor efficacy. We profiled the transcriptomes of two AML cell lines, IMS-M2 and U937, treated with OTX015 for 24 hours with or without folic acid, using RNA sequencing (RNA-seq). Consistent with the fact that folate withdrawal decreases methylation of repressive H3K27me3 and H3K9me2 histone marks, we observed a global increase in gene expression, despite OTX015 treatment, which has been shown to preferentially repress genes marked by superenhancers (412 and 706 significantly upregulated vs. 160 and 520 significantly downregulated genes in IMS-M2 and U937 cells, respectively; Supplementary Fig. S12A). Accordingly, we identified 75 and 15 genes significantly upregulated and downregulated between both folate-deprived cell lines, respectively, compared with cells in regular media with OTX015 (Fig. 5A). An open-ended enrichment analysis was conducted comparing gene expression variations (i) between DMSO and OTX015 conditions irrespective of folate status and (ii) upon folate withdrawal irrespective of DMSO or OTX015 treatment. Gene sets were considered significantly enriched based on the cutoffs P ≤ 0.05 and false discovery rate (FDR) ≤ 0.25 on the entire set of signatures from the c2, c3, and c6 (MSigDB) and ENCODE and CHEA collections (Fig. 5B). Consistent with the reported effect of BET inhibitors on MYC expression, we established that, irrespective of folate status, gene sets related to MYC transcriptional program were highly enriched (FDR < 0.25) in genes whose expression is suppressed by OTX015 in the cell lines tested (Fig. 5B, top). Conversely, many gene sets related to SPI1 transcriptional program and IFN signaling and regulatory factors (IRF), a well-reported SPI1-connected pathway (21, 22), were drastically enriched in genes whose expression is enhanced by folate withdrawal specifically with OTX015 treatment (Fig. 5B, bottom). To investigate whether OTX015-induced activation of SPI1-related transcriptional programs reflects epigenetic changes induced by folate starvation, we profiled genome-wide distribution of H3K27me3 and H3K9me2 by chromatin immunoprecipitation sequencing (ChIP-seq) of OTX015-treated IMS-M2 and U937 cells with regular or folate-restricted media. This confirmed that H3K27me3 binding signal in a 10-Kb region flanking gene transcriptional starting sites (TSS) was decreased with folic acid starvation (Supplementary Fig. S12B). A total of 3,868 and 4,214 genes were annotated within the 50-kb flanking H3K27me3 or H3K9me2-marked regions exhibiting a decreased methylation level with folate withdrawal in OTX015-treated IMS-M2 and U937 cells, respectively (Fig. 5C). By overlapping these gene sets with genes identified by RNA-seq as upregulated with folate starvation (530 and 1,140 genes in IMS-M2 and U937 cells, respectively), we narrowed down the list of genes whose upregulation was directly associated with H3K27me3 and H3K9me2 demethylation to 93 and 202 genes in IMS-M2 and U937 cells, respectively. These two gene lists were then queried by overlapping analysis with two ENCODE and CHEA datasets generated from curated target genes of transcription factors defined on the basis of published ChIP-chip, ChIP-seq, and other transcription factor–binding site profiling studies (Fig. 5D). Two gene signatures related to SPI1 target genes from ENCODE and CHEA scored among the top enriched pathways, confirming that SPI1 transcriptional program activation in response to OTX015 treatment and folate withdrawal results from H3K27 and H3K9 demethylation. Finally, using the top genes from SPI1 target–related gene sets that were upregulated upon folate withdrawal and OTX015 treatment, we designed a SPI1 consensus transcriptional target mini signature. We used qRT-PCR to assess alteration of these signature genes and confirmed induction of these SPI1 target genes in response to MTHFR depletion and OTX015 treatment (Fig. 5E). Similar induction was observed in 677 TT and 1298 AA MTHFR homozygous KG1a clones in comparison with their wild-type counterpart (Fig. 5F).

To explore functionally whether SPI1 transcriptional program activation induced by folate cycle disruption promotes resistance to BRD4 inhibition, we knocked down SPI1 in IMS-M2 and U937 cells before folate withdrawal or MTHFR suppression (Fig. 5G and H). SPI1 depletion significantly reduced OTX015 resistance in AML cells whose folate cycle activity was disturbed or MTHFR expression was suppressed (Fig. 5I and J). In addition, SPI1 depletion alleviated resistance to JQ1 of MLL–AF9 blasts injected in mice fed a folic acid–restricted diet (Supplementary Fig. S13). Similarly, Spi1 knockdown abrogated resistance to OTX015 of MLL–AF9-transformed Mthfr knockout primary murine cells both in vitro and in vivo (Fig. 5K and L). Together, these data suggest that folate cycle disruption promotes H3K27 and H3K9 demethylation, which activates a SPI1 transcriptional program capable of promoting resistance to BRD4 inhibition.

Here we revealed folate-connected metabolic gene signatures whose activation was poorly correlated with an active MYC transcriptional program in AML. This core metabolic network is composed of the histidine, serine, threonine and glycine pathways, which serve as inputs to the folate cycle (23), the beta-alanine/histidine synthesis pathway (24), and the phenylalanine- and DHFR-dependent tetrahydrobiopterin regeneration pathway (25). We also identified the tryptophan pathway as part of this core metabolic unit, likely because this pathway produces vitamin B6, an essential cofactor for synthesis of 5-CH₃ THF from serine and THF (26).

Folate metabolism supports manifold transformations dependent on THF oxidative state, with 5,10-CH₂-THF, 5-CH₃-THF, and 10-CHO-THF each supporting distinct biosynthetic functions. 5,10-CH₂-THF is used in three ways: (i) by serine hydroxymethyltransferase (SHMT1) or thymidylate synthase (TYMS) to synthesize serine or sustain de novo thymidylate synthesis, respectively; (ii) by MTHFR for reduction to 5-CH₃-THF and entrance into the folate cycle, which stimulates the methylation cycle; or (iii) by oxidation into 10-CHO-THF to sustain the THF salvage pathway that promotes purine synthesis and maintains mitochondrial redox homeostasis (23). Although a subset of folate- and MTHFR-related gene sets were poorly correlated with MYC signatures, we demonstrate that gene sets related to the THF salvage pathway and purine synthesis were strongly associated with active MYC gene signatures. This is consistent with previous studies showing that MYC promotes expression of SHMT2, MTHFD2, and MTHFD1L mitochondrial enzymes that contribute to purine production through the THF salvage pathway (27, 28). Despite the fact that both the folate cycle and tetrahydrofolate salvage pathways are interconnected within folate metabolism, our data hereby suggest that these pathways can be uncoupled on the basis of their connectivity to MYC transcriptional programs.

Consistent with this observation, we demonstrate functionally that folate starvation promotes resistance to MYC targeting by either BRD4 or CDK7 inhibitors, or BRD4-directed shRNAs. Previous studies suggest that folate cycle, de novo thymidylate synthesis, and THF salvage pathways compete for a limiting pool of 5,10-CH₂-THF upon folate starvation. For instance, thymidylate synthesis is preserved in folate deficiency at the expense of the methylation cycle, which relies on activity of 5,10-CH₂-THF–dependent MTHFR from the folate cycle. This favorable balance toward sustained thymidylate synthesis results from the activity of a critical cytosolic 10-CHO-THF producer, MTHFD1, which translocates to the nucleus upon folate starvation to maintain de novo thymidylate biosynthesis (29, 30). Moreover, nuclear MTHFD1 also regulates transcription by direct binding to BRD4-occupied chromatin (31). These observations suggest that all folate-related metabolic pathways are not affected to the same extent by folate restriction. Although some pathways (10-CHO-THF–dependent reactions) persist without folate due to the plasticity of enzymes such as MTHFD1, the folate cycle and its associated methylation cycle are more severely impaired and thereby influence cell response to BET inhibitors.

We pinpoint MTHFR as the most critical folate-related mediator of resistance to BRD4 inhibitors and demonstrate that the presence of either of two MTHFR polymorphisms (677C>T; 1298A>C) in a homozygous state predicts sensitivity to BET inhibition. We observe that folate cycle disturbance in AML cells results in intracellular accumulation of SAH, the downstream effector of MTHFR knockdown triggering BET inhibitor resistance. Because the conversion of SAH to homocysteine is reversible but SAH-preferential, our results are consistent with the fact that hyperhomocysteinemia is accompanied by an elevation of SAH in Mthfr knockout animals (19, 32). Given that SAH is a potent inhibitor of SAM-dependent methylation, folate cycle impairment suppresses H3K27 and H3K9 methyltransferases and targets methylation across the whole genome of AML cells. We validated that a SAM-dependent H3K27-methyltransferase PRC2 complex member, EED, scored among three downstream mediators of resistance to BET inhibitors, thereby recapitulating that suppression of the PRC2 complex promotes BET inhibitor resistance in AML (33).

Decreased H3K27 and H3K9 methylation upon folate cycle alteration combined with BET inhibition activates SPI1 and IRF/IFN signaling transcriptional programs. These programs are tightly linked through physical interactions between SPI1 and IRF4 or IRF8 in hematopoietic stem and progenitor cells (34, 35). We establish that SPI1 program activation is the main downstream effector of resistance to BET inhibitors with folate or MTHFR deficiency. Indeed, inhibition of demethylase KDM1A increases resistance to BRD4 inhibition via an IRF8/SPI1-mediated epigenetic rewiring (36). This supports the idea that the folate cycle dynamically influences epigenetic AML cell resistance by favoring an adaptive transcriptional plasticity relying on widespread, SPI1-dependent transcriptional changes. The latter acts as a compensatory mechanism to sustain cell survival despite BRD4 inhibition, a canonical epigenetic regulator of enhancer regions.

Two clinical trials are currently evaluating the clinical efficacy of combining BET inhibitors and azacitidine, the standard-of-care treatment for elderly patients with AML (NCT02303782 and NCT02543879), thereby raising clinical interest in identifying relevant biomarkers of response to BET inhibitors. Although no significant prevalence of the MTHFR C677T and A1298C variants was described in AML, their frequencies were found significantly increased in patients with therapy-related AML who contracted breast cancer as primary disease or who were previously treated with cyclophosphamide-based anticancer therapies (37). This suggests that the metabolic consequences of altered MTHFR activity could trigger resistance to a wide collection of chemotherapeutic agents beyond just those known to target the folate cycle, or could even participate in therapy-induced leukemogenesis. Besides folate and MTHFR deficiencies, deficits in folate cycle enzyme cofactors like cobalamin (vitamin B12) or pyridoxal phosphate (from vitamin B6) also affect methylation and cause hyperhomocysteinemia (38, 39). Up to 38% of older adults exhibit a food-cobalamin malabsorption syndrome characterized by B12 deficiency, and the incidence of B6 deficiency in European institutionalized elderly people reaches up to 75%. Thus, our study likely underestimates the prevalence of folate cycle deficiency–induced BET inhibitor resistance among patients with AML, who are mainly elderly. Additional work is required to evaluate to what extent the effect of folate/methylation cycle disruption upon B12 and B6 deficiencies or loss-of-function mutations in enzymes controlling these metabolic pathways may, like MTHFR insufficiency, promote resistance to MYC-targeting therapies.

Our study supports that MTHFR and folate cycle deficiencies remodel the epigenetic landscape to shape transcriptional plasticity which represents a mechanism of resistance to epigenetic therapies. Their status should be carefully assessed by measurements of total plasma homocysteine or CH₃-THF levels when enrolling patients with hematologic diseases in clinical trials of BET inhibitors. Dietary supplementation with CH₃-THF may thereby represent a relatively low-risk intervention that might allow for a greater clinical benefit to curtail transcriptional adaptation of malignant cells to MYC-targeting therapies.

The U937 and KG1a cell lines were purchased from the ATCC and the OCI-AML2 cell line was purchased from DSMZ. MOLM-13 was provided by Dr. Benjamin Ebert; IMS-M2 by Pr. Hervé Dombret; and MOLM-14 cell lines by Dr. Scott Armstrong (Dana-Farber Cancer Institute, Boston, Massachusetts). Identity of all cell lines was confirmed by short tandem repeat loci profiling. All cell lines were tested negative for Mycoplasma using MycoAlert PLUS Mycoplasma Detection Kit (#LT07–705, Lonza). All cell lines, except MOLM-13 and Kasumi-3, were maintained in RPMI 1640 (Invitrogen) supplemented with 1% penicillin–streptomycin and 10% FBS (Sigma-Aldrich) at 37°C with 5% CO₂. MOLM-13 cells were maintained in RPMI 1640 supplemented with 1% penicillin–streptomycin and 20% FBS. Kasumi-3 cells were maintained in RPMI 1640 supplemented with 1% penicillin–streptomycin and 10% FBS with 20 ng per mL GM-CSF (PeproTech). The 293T cells were maintained in DMEM (Invitrogen) supplemented with 10% FBS (Sigma-Aldrich) and 100 U/mL penicillin–streptomycin (Invitrogen).

Primary patient AML blasts were collected from bone marrow aspirates or peripheral blood samples after obtaining patient written informed consent under Saint-Louis Hospital and Lille University Hospital Internal Review Board–approved protocols. These studies were conducted in accordance with recognized ethical guidelines from the Declaration of Helsinki and were approved by an institutional review board. Mononuclear cells isolated using Ficoll-Paque Plus (Amersham Biosciences) were thawed before drug treatment. These cells were maintained in RPMI 1640 medium supplemented with 20% FBS, 20 ng/mL IL3 (#200–03, PeproTech), 20 ng/mL IL6, 20 ng/mL GM-CSF (#300–03, PeproTech), 10 ng/nl G-CSF (#300–23, PeproTech), 10 ng/mL EPO (#100–64, PeproTech), 50 ng/mL TPO (#300–18, PeproTech), 100 ng/mL FLT3-Ligand (#300–19, PeproTech), and 100 ng/mL SCF (#300–07, PeproTech).

For folate starvation experiments, cell lines were washed twice in PBS and maintained for 18 hours in no-folic-acid RPMI 1640 medium (Invitrogen, #27016–021) supplemented with 10% dialyzed FBS (Sigma-Aldrich, #F0392) and 100 U/mL penicillin–streptomycin (Invitrogen) prior to treatment with drugs or infection with shRNAs. Primary patient cells were kept in the same medium as cell lines supplemented with the same concentration of cytokines as stated previously. The corresponding + folic acid RPMI 1640 medium was generated using no-folic-acid RPMI 1640 medium supplemented with 1 mg/L folic acid (Sigma-Aldrich-Aldrich, #F8758).

For functional characterization of the effect of single amino acid starvation on cell response to BET inhibitors, cell lines were washed twice in PBS and maintained for 18 hours in no folate and no amino acid RPMI 1640 medium (Genaxxon Bioscience) supplemented with all amino acids and folic acid with the exception of the amino acid to test, 10% dialyzed FBS, and 100 U/mL penicillin–streptomycin prior to treatment with BET inhibitors.

To determine relative levels of intracellular metabolites, extracts were prepared and analyzed by LC/MS-MS. Sixteen hours before extraction, 15 × 10⁶ U937 cells were plated in quadruplicate in folic acid–free RPMI 1640 supplemented with 10% dialyzed FBS and 100 U/mL penicillin/streptomycin ± 1 mg/L folic acid. Metabolites were extracted on dry ice with 4 mL of 80% methanol (−80°C), as described previously (40). Insoluble material was pelleted by centrifugation at 3,000 × g for 5 minutes, followed by two subsequent extractions of the insoluble pellet with 0.5 mL of 80% methanol, with centrifugation at 16,000 × g for 5 minutes. The 5 mL metabolite extract from the pooled supernatants was dried down under nitrogen gas using an N-EVAP (Organomation Associates, Inc).

Dried pellets were resuspended using 20 μL HPLC grade water for mass spectrometry. Ten microliters were injected and analyzed using a QTRAP 5500 triple quadrupole mass spectrometer (AB/SCIEX) coupled to a Prominence UFLC HPLC system (Shimadzu) via selected reaction monitoring (SRM) of a total of 547 endogenous water-soluble metabolites for steady-state analyses of samples. Some metabolites were targeted in both positive and negative ion mode for a total of 391 SRM transitions using pos/neg polarity switching. ESI voltage was +4,900 V in positive ion mode and −4,500 V in negative ion mode. The dwell time was 3 ms per SRM transition and the total cycle time was 1.55 seconds. Approximately 10 to 14 data points were acquired per detected metabolite. Samples were delivered to the MS via normal phase chromatography using a 4.6 mm i.d × 10 cm Amide Xbridge HILIC column (Waters Corp.) at 350 μL/minute. Gradients were run starting from 85% buffer B (HPLC grade acetonitrile) to 42% B from 0–5 minutes; 42% B to 0% B from 5–16 minutes; 0% B was held from 16–24 minutes; 0% B–85% B from 24–25 minutes; 85% B was held for 7 minutes to reequilibrate the column. Buffer A was comprised of 20 mmol/L ammonium hydroxide/20 mmol/L ammonium acetate (pH = 9.0) in 95:5 water:acetonitrile. Peak areas from the total ion current for each metabolite SRM transition were integrated using MultiQuant v2.0 software (AB/SCIEX). A Student t test was performed to assess significance of pairwise comparisons. All data are log₂-transformed and normalized against the average control condition. No or small variations in metabolite production compared with average control condition appears closest to the white color on the heat map. Decreased or increased metabolite production compared with the average control condition is represented by a gradient of color from blue to red, respectively. MetaboAnalyst software (www.metaboanalyst.ca) was used to perform pathway enrichment analysis on the set of top metabolite hits identified by steady-state profiling. No specific manual settings were needed to analyze the list of metabolites.

OCI-AML2 cells were seeded into 6-well plates at a density of 3 × 10⁶ cells per well and transduced at MOI 0.2. 24 hours post-transduction, cells were replated into puromycin-containing media. A sample was collected at 48 hours post-puromycin to confirm library coverage in the transduced population. Transduced cells were expanded in puromycin for 10 days before drug, when the transduced population was split into vehicle (DMSO) and JQ1 conditions and maintained for two weeks. Deep sequencing was performed by Hudson Alpha Institute for Biotechnology on an Illumina Nextseq platform (75 bp, single-ended) to identify differences in library composition. Determinations of genetic essentiality and drug sensitization/resistance were made by evaluating differential guide compositions between the initial population and subsequent drug-treated and vehicle-treated cell populations. Briefly, the fractional representation (FR) for a guide within a sample was normalized to the sum of all guides attributed to that sample. A direct comparison between two samples entailed the quotient of the respective FRs, which we term the depletion metric (DM). The five guide-level DMs for each gene were then collapsed to gene-level scores by taking the average. Guides which totaled fewer than 200 counts for a given sample were excluded from analysis. Genetic essentiality was calculated by considering the depletion/enrichment of the vehicle-treated population over time (DMSOfinal/initial). Drug sensitization/resistance was calculated by considering the depletion/enrichment of the drug-treated population relative to the vehicle-treated population (Drug_final/DMSO_final). All depletion/enrichment effects are reported as log₂ ratios. All described manipulations were performed in R. Additional information is provided in the Supplementary Material and Methods section.

Top and bottom sgRNAs targeting the C677 and A1298 MTHFR sites (Top_C677: 5′-CACCGAAGCTGCGTGATGATGAAAT-3′ and Bottom_C677: 5′-AAACATTTCATCATCACGCAGCTTC-3′; Top_A1298: 5′-CACCGTTCAAAGACACTTTCTTCAC-3′ and Bottom_A1298: 5′-AAACGTGAAGAAAGTGTCTTTGAAC-3′) were annealed and phosphorylated as described previously (24157548), and ligated into the BbSSI-digested pSpCas9(BB)-2A-GFP (PX458) vector (Addgene, # 48138) to generate PX458_sgC677 and PX458_sgA1298 constructs. 1 × 10⁶ KG-1a cells were washed in PBS before resuspension in 100 μL Opti-MEM medium (Thermo Fisher Scientific, # 31985–047). Ten micrograms of PX458_sgC677 or PX458_sgA1298 vector and 0.3 μmol/L SSODN_C677 (5′-TGGCAGGTTACCCCAAAGGCCACCCCGAAGCAGGGAGCTTTGAGGCTGACCTGAAGCACTTGAAGGAGAAGGTGTCTGCGGGAGTCGATTTCATCATCACGCAGCTTTTCTTTGAGGCTGACACA-3′) or SSODN_A1298 (5′-TTGGGGAGCTGAAGGACTACTACCTCTTCTACCTGAAGAGCAAGTCCCCCAAGGAGGAGCTGCTGAAGATGTGGGGGGAGGAGCTGACAAGTGAAGCAAGTGTCTTTGAAGTCTTCGTTCTTTAC-3′) were added into the mix of Opti-MEM and cells to generate isogenic KG-1a clones expressing either MTHFR 677 C>T, A1298 A>C, or wild-type genetic variant, respectively.

Cells were then electroporated using a NEPA21 eletroporator (Nepagene) following a poring pulse of 150 V, length: 5 ms, interval: 50 ms, No: 2, D. rate: 10%, and polarity: +, and a transfer pulse of 20 V, length: 50 ms, interval: 50 ms, No: 5, D. rate: 40%, and polarity: ±. Electroporated cells were then incubated for 48 hours in the presence of 10 μmol/L SCR7 (Selleckchem, # S7742) prior to single-cell sorting into 96-well plates. Clones were then grown and MTHFR genetic status was screened by allelic discrimination assay.

SICER-identified significantly decreased H3K9me2 (14,651 regions in IMS-M2; 4,842 regions in U937) and H3K27me3 (4,048 regions in IMS-M2; 4,855 regions in U937) peaks in OTX-FA versus OTX+FA conditions were annotated for the 2 nearest genes using GREAT (41) within a 50kb window upstream and downstream. The union of lists of annotated genes per cell line was kept defining a list of all genes in the vicinity of lost H3K9me2 or H3K27me3 (3,961 genes for IMSM2; 4,416 genes for U937).

Significantly upregulated genes (fold change > 1.2; P < 0.05) identified by RNA-seq experiments comparing OTX-FA versus OTX+FA conditions were selected (623 genes in IMS-M2; 1,342 genes in U937) compared with the lists of genes proximal to decreased H3K9me2 and H3K27me3 regions identified previously leading to an overlap of 93 genes in IMS-M2 and 202 genes in U937. Additional information is provided in the Supplementary Material and Methods section. ChIP-seq and RNA-seq data are available at the Gene Expression Omnibus (GEO) database under accession code GSE152442.

ssGSEA was used to calculate separate enrichment scores for each pairing of a sample whose transcriptomic data was available from TCGA-LAML (42) or GSE14468 (43) and a given gene set (queried from MSigDB for MYC and KEGG-related gene signatures or manually curated from BIOCYC for other metabolic gene sets). A Pearson correlation matrix was computed between each ssGSEA score for the core MYC signature and all gene sets of interest obtained across all patients from a given cohort. Connected metabolic pathways were clustered based on the median of Pearson correlation scores obtained from each individual pathway.

The French National Committee on Animal Care reviewed and approved all mouse experiments. Sample size was influenced by historical penetrance and consistency of MLL–AF9-driven in vivo models. For the Mthfr knockout AML mouse model, BALB/cJ mice were purchased from Charles River Laboratories. Each recipient mouse was transplanted with transduced Sca1⁻/c-Kit⁺ myeloid progenitors sorted from the total bone marrow of 3 donor mice. Approximately 2 months post-transplantation, sick mice were euthanized and bone marrow harvested before flow-based sorting of MLL–AF9-positive granulo-monocytic bone marrow progenitor population (GFP⁺/Sca-1⁻/c-Kit⁺/Cd16/32⁺/Cd34⁺) and PCR-based confirmation of Mthfr knockout status. 0.2 × 10⁶ MLL–AF9-positive Mthfr^+/+, Mthfr^+/−, and Mthfr^−/− cells were then reinjected into sublethally irradiated secondary recipient mice treated daily by intraperitoneal injection with 50 mg/kg JQ1 (10% DMSO + 90% G5W).

For folate starvation experiments, 5-week-old male C57BL/6J mice were given regular or folate-deficient casein-reconstituted diet (U8958 Version 0262, Safe-Diets) for 4 weeks before sublethal irradiation (350 cGy) and injection with 0.2 × 10⁶ MLL–AF9-positive L-GMP cells before treatment with 35 mg/kg or 50 mg/kg JQ1 (10% DMSO + 90% G5W) for 7 days. For Mthfr knockdown experiments, MLL–AF9-positive L-GMP cells were infected with a control or two Mthfr-directed shRNAs cloned into an MSCV-miRE-SV40-eBFP vector modified from the MSCV-miRE_shBRD9_561-SV40-GFP vector (Addgene, # 75139) by substitution of the GFP cassette with an eBFP fluorescent marker. 0.2 × 10⁶ infected cells were injected into sublethally irradiated 5-week-old C57BL/6J recipient mice. Additional information is provided in the Supplementary Material and Methods section.

Statistical analysis was done using Microsoft Excel, Prism 5.03 (GraphPad), or indicated software for more dedicated analysis. Statistical significance was determined by unpaired Student t test after testing for normal distribution. For samples with significantly different variances, Welch correction was applied. Samples with nonnormal distribution (with the assumption of no Gaussian distribution of the group) were analyzed using a nonparametric Mann–Whitney test, and the level of significance (alpha) was always set at 0.05. For comparisons of three or more groups, nonparametric Kruskall–Wallis and Dunn multiple comparisons tests were used, and the level of significance was always set at 0.05.

Transcriptomic data from primary patient samples with AML are available from TCGA-LAML (GDC Data Portal, National Cancer Institute) and from the GEO database under accession number GSE14468. U937 and IMS-M2 ChIP-seq and RNA-seq data are available from the GEO database under accession number GSE152442.

J. Qi reports grants from NIH during the conduct of the study; and J. Qi is a scientific cofounder and consultant for EPIPHANES and shareholder for Zentalis. J.-B. Micol reports personal fees from Jazz Pharmaceuticals and Astellas, and grants from Novartis outside the submitted work. C. Preudhomme reports grants and personal fees from Celgene, AbbVie, Daichii Sankyo, Astellas; personal fees from Amgen and Incyte outside the submitted work. L. Benajiba reports grants from Gilead Foundation (International Hematology/Oncology Research Scholars Award) outside the submitted work. K. Stegmaier reports grants from NIH during the conduct of the study, from Novartis (grant for research unrelated to the current publication); personal fees from Rigel Pharmaceuticals (consulting fees for discussions of a targeted agent unrelated to those discussed in the publication) and Auron Therapeutics (begun consulting for Auron but have not yet received any personal fees from them) outside the submitted work. C. Lobry reports grants from Fondation Gustave Roussy and ATIP-Avenir Inserm during the conduct of the study. K.C. Wood reports grants from NIH (R01CA207083) during the conduct of the study; personal fees and other from Tavros Therapeutics (cofounder, equity holder, consultant), Element Genomics (cofounder, equity holder, consultant); and other from Celldom (cofounder and equity holder) outside the submitted work. R. Itzykson reports grants from Oncoethix, S.A. (now Merck), Association Laurette Fugain, and Gilead International Research Scholarship in Onco-haematology during the conduct of the study; in addition, R. Itzykson has a patent for EP19306350.0 issued. A. Puissant reports grants from ATIP/AVENIR French Research Program, EHA Research Grant for Nonclinical Advanced Fellow, Ligue Nationale Contre le Cancer, Mairie de Paris Emergences, ERC Starting Program (758848), St Louis Association for Leukemia Research, and INCA PLBIO (PLBIO20-246) during the conduct of the study; in addition, A. Puissant has a patent for EP19306350.0 issued. No potential conflicts of interest were disclosed by the other authors.

A. Su: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing-original draft. F. Ling: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing-original draft. C. Vaganay: Validation, investigation, methodology. G. Sodaro: Validation, investigation, visualization. C. Benaksas: Conceptualization, validation, investigation. R. Dal Bello: Conceptualization, investigation. A. Forget: Validation, investigation. B. Pardieu: Validation, investigation. K.H. Lin: Resources, formal analysis. J.C. Rutter: Conceptualization, formal analysis. C.F. Bassil: Conceptualization, formal analysis. G. Fortin: Validation, investigation. J. Pasanisi: Validation, methodology. I. Antony-Debre: Resources. G. Alexe: Resources. J.-F. Benoist: Formal analysis. A. Pruvost: Formal analysis, methodology. Y. Pikman: Resources, investigation. J. Qi: Resources. M.-H. Schlageter: Resources, supervision. J.-B. Micol: Resources, data curation. G. Roti: Resources, data curation. T. Cluzeau: Resources, data curation. H. Dombret: Resources. C. Preudhomme: Resources. N. Fenouille: Conceptualization, investigation. L. Benajiba: Conceptualization, methodology. H.M. Golan: Resources. K. Stegmaier: Conceptualization, validation. C. Lobry: Conceptualization, resources, data curation, formal analysis, supervision, validation, methodology. K.C. Wood: Conceptualization, resources, supervision, validation, investigation, writing-original draft. R. Itzykson: Conceptualization, supervision, validation, investigation. A. Puissant: Conceptualization, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing.

We are indebted to Jean-Michel Cayuela, Carole Albuquerque, Christophe Roumier, and Céline Decroocq from the Saint-Louis and Lille Tumor Banks for primary patient samples; Veronique Montcuquet, Nicolas Setterblad, Christelle Doliger, and Sophie Duchez from the Saint-Louis Research Institute Core Facility; and the technical staff from the DBA (Diagnostic Biologique Automatisé) platform of Saint-Louis Hospital. We are grateful to Dr. Lucio H. Castilla for providing us with the Cbfb-MYH11 knock-in mouse model. This work was supported by the ATIP/AVENIR French research program (to A. Puissant), the EHA research grant for Non-Clinical Advanced Fellow (to A. Puissant), the Ligue Nationale Contre le Cancer (to A. Puissant), the Mairie de Paris Emergences grants (to A. Puissant), the INCA PLBIO program (PLBIO20-246, to A. Puissant), the Fondation Gustave Roussy (to C. Lobry), and NIH 5R35 CA210030 (to K. Stegmaier). This work was also supported by grants from Oncoethix, S.A. (now Merck), Association Laurette Fugain (ALF2014-10), and Gilead International Research Scholarship in Onco-haematology (to R. Itzykson). A. Puissant is a recipient of support from the ERC Starting program (758848) and supported by the St Louis Association for Leukemia Research.

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.