Targeting the MTF2–MDM2 Axis Sensitizes Refractory Acute Myeloid Leukemia to Chemotherapy

Standard induction chemotherapy has been the first-line therapeutic to treat acute myeloid leukemia (AML) for about 40 years. Although standard induction chemotherapy can induce remission in most patients with AML, more than 30% of patients are unresponsive to treatment. Unfortunately, as many as 60% to 90% of patients with refractory AML will not survive their disease regardless of additional therapies (1). To move beyond generalized treatments, targeted therapeutics and biomarkers are actively being pursued to customize treatment regimens to attain complete remission and improve survival rates. Despite recent advances in genetic markers that stratify patients with AML into favorable, intermediate, and adverse risk categories, patients with refractory AML are found across all risk groups (2). This suggests that a nonmutational mechanism plays an important role in driving refractory AML. This hypothesis is further supported by insightful deep-sequencing studies which revealed that many of the common AML driver mutations are in epigenetic modifiers (3). Although these studies have relied heavily on DNA sequencing to identify key mutations, RNA expression levels and the epigenetic regulation within refractory AML cells have largely been overlooked.

Recent studies have shown that altered polycomb repressive complex 2 (PRC2)–mediated histone H3 lysine 27 trimethylation (H3K27me3) can predict disease outcome and survival in some cancers (4, 5). Yet, the mechanism by which H3K27me3-mediated epigenetic regulation affects cancer cell behavior, and thus patient survival, has not been well elucidated. In addition to the methyltransferase activity provided by either EZH1 or EZH2, SUZ12 and EED are core PRC2 proteins essential for function and complex stability. However, the PRC2 core proteins do not encode DNA binding activity, which is provided by accessory proteins such as polycomblike (PCL) family members (6). Thus, the prototypical role of PRC2 accessory proteins is to recruit the PRC2 to chromatin, and their tissue-specific expression is thought to drive lineage-specific H3K27me3 methylation. MTF2/PCL2 was recently demonstrated to recruit the PRC2 to DNA regions with a high density of unmethylated CpGs resulting in chromatin conformation–dependent binding (7). In embryonic stem cells (ESC), MTF2 functions as a prototypical PRC2 accessory protein to recruit PRC2 to the extended pluripotency network gene promoters, leading to increased H3K27me3 methylation and repression of the pluripotency network to enable ESC differentiation (6). In contrast, in hematopoietic cells, MTF2 behaves more similarly to a core PRC2 component because its loss reduces SUZ12 and EZH1/2 expression and global H3K27me3 levels (8).

Although mutational and expression studies in AML are generally performed on whole bone marrow (BM) samples, we reasoned that investigating AML cell behavior would be challenging in heterogeneous bulk BM populations containing both mature and immature cells. Therefore, to examine the role of PRC2-mediated epigenetic repression in AML, we fractionated primary AML patient-derived diagnostic samples using CD34 and CD38 markers. This strategy revealed that at diagnosis, total H3K27me3 levels in CD34⁺CD38⁻ cells are markedly reduced in refractory AML patient samples due to a loss of MTF2 expression. Using an unbiased systems approach to dissect the underlying molecular networks that drive refractory AML, we discovered that the MDM2–p53 axis is regulated by MTF2 and that inhibiting MDM2 reverses MTF2–PRC2-mediated chemoresistance associated with refractory AML.

To determine whether H3K27me3 levels vary among patients with AML, we analyzed the global H3K27me3 levels of this repressive histone mark by flow cytometry within sorted CD34⁺CD38⁻, CD34⁺CD38⁺, CD34⁺, and bulk cells from 32 diagnostic AML BM aspirates isolated from patients who underwent induction therapy (Fig. 1A; Supplementary Fig. S1A). Strikingly, we identified two patient groups based on total H3K27me3 levels: one with levels similar to matched normal immunophenotypic BM cells and the other with markedly reduced levels. Independent clinical patient follow-up performed in a blinded manner revealed that reduced H3K27me3 levels in diagnostic AML cells can predict nonresponders to induction therapy. However, the levels of H3K27me3 within the immature CD34⁺CD38⁻ population, which has been reported to enrich for leukemic stem cells (LSC; ref. 9), were the best indicator and predictor of therapeutic response (Supplementary Fig. S1B). In fact, analysis of the disease progression of this patient cohort found patients with H3K27me3-reduced CD34⁺CD38⁻ cells generally failed to achieve complete remission, otherwise referred to as nonresponders or patients with refractory AML (Fig. 1B and C; Supplementary Table S1). Furthermore, reduced H3K27me3 levels within patient CD34⁺CD38⁻ cells predicted poor patient survival based on our observation that the mean survival time of patients with normal levels of H3K27me3 (i.e., H3K27me3 basal) was three times longer than patients with reduced H3K27me3 levels (Fig. 1D; Supplementary Tables S2 and S3). To ensure the specificity of the H3K27me3 antibody (clone CST C36B11), we performed internal standard calibrated chromatin immunoprecipitation (ICeChIP; ref. 10), which demonstrated the high specificity of this antibody (Supplementary Fig. S1C).

Next, we used RT-qPCR to test whether there is a correlation between H3K27me3 levels and PRC2 member expression. Linear regression analysis revealed a highly significant correlation between H3K27me3 levels and MTF2 mRNA expression (R² = 0.9416) within the patient cohort CD34⁺CD38⁻ cells (Fig. 1E; Supplementary Fig. S2). However, the correlation between H3K27me3 levels and other PRC2 members was significant but much weaker (Fig. 1F and G; Supplementary Fig. S2). Reduced MTF2 mRNA expression within patient CD34⁺CD38⁻ cells was also associated with poor response to standard induction chemotherapy (Fig. 1H and I) and poor survival (Supplementary Fig. S3A and S3B). Cytogenetic analysis revealed that the local patient cohort consisted of patients with AML belonging to each of the three European LeukemiaNet (ELN) risk categories (Supplementary Fig. S3C). Within the favorable cytogenetics category, MTF2 levels further stratified patients based on overall survival (Supplementary Fig. S3D); however, this was not found to be the case in the intermediate cytogenetics category (Supplementary Fig. S3E). These results were further validated using The Cancer Genome Atlas (TCGA; ref. 11) data set in which patients with AML received an induction therapy regimen similar to the treatment given to our cohort. Kaplan–Meier analysis of the TCGA data set revealed that patients with reduced levels of MTF2 within bulk BM cells had reduced overall survival (OS), whereas altered expression of other PRC2 members had no significant effect on patient outcome (Supplementary Fig. S3F–S3I; Supplementary Table S4). Furthermore, multivariate analysis using the TCGA data set confirmed that MTF2 expression is a novel risk factor that can predict OS (Supplementary Table S5).

Considering MTF2 is rarely mutated in AML, we analyzed the MTF2 promoter for evidence of hypermethylation and discovered that at least one of the two CpG islands in the MTF2 promoter was hypermethylated in MTF2-deficient AML (MD-AML) samples, whereas neither CpG island was methylated in healthy BM (H-BM) or AML samples with normal (basal) MTF2 levels (B-AML; Supplementary Fig. S4A). The close correlation between MTF2 expression and H3K27me3 levels led us to test whether downregulation of MTF2 is sufficient to reduce H3K27me3 levels. Thus, human umbilical cord blood–derived (UCB) CD34⁺CD38⁻ cells were transduced with two different MTF2 shRNA GFP-encoded lentiviruses. Flow cytometry analysis of H3K27me3 levels in GFP⁺ cells revealed markedly reduced H3K27me3 within 96 hours of MTF2 knockdown (Fig. 1J and K). Further investigation demonstrated that MTF2 knockdown results in reduced abundance of the PRC2 core proteins EZH1, EZH2, and SUZ12 (Supplementary Fig. S4B), but their mRNA levels were not affected (Supplementary Fig. S4C), suggesting that reduced MTF2 affects PRC2 levels posttranscriptionally, resulting in decreased H3K27me3. In contrast, ectopic expression of MTF2 in MTF2-deficient CD34⁺CD38⁻ AML cells reestablished global H3K27me3 levels (Supplementary Fig. S5A–S5D), further demonstrating that MTF2 expression dictates H3K27me3 levels in hematopoietic progenitors and AML.

We next used chromatin immunoprecipitation sequencing (ChIP-seq) to investigate the changes in H3K27me3 levels triggered by MTF2 deficiency. To aid in relative H3K27me3 quantification across samples, Drosophila chromatin spike-in controls were added to the samples, and H3K27me3 signal was normalized to total Histone H3 ChIP-seq signal that was performed in parallel. Although there are no differences in human total Histone H3 reads per Drosophila spike-in read between the various samples, human H3K27me3 reads per Drosophila spike-in read were reduced by 10-fold in the refractory AML patient samples compared with induction therapy–responsive MTF2 basal samples, reflecting the dramatic reduction in H3K27me3 in MTF2-deficient AML samples (Supplementary Fig. S6). A similar reduction in H3K27me3 reads per Drosophila spike-in read was observed in MTF2-knockdown CD34⁺CD38⁻ BM cells compared with the scramble shRNA control CD34⁺CD38⁻ BM cells. Principal component analysis and hierarchic clustering of H3K27me3 revealed that MTF2 knockdown CD34⁺CD38⁻ BM cells cluster close to refractory patient CD34⁺CD38⁻ BM cells, whereas scramble shRNA control CD34⁺CD38⁻ BM cells cluster with CD34⁺CD38⁻ BM cells from induction therapy–responsive patients (Fig. 1L and M). Taken together, these results demonstrate that MTF2 deficiency and reduced H3K27me3 levels in immature CD34⁺CD38⁻ cells correlate with refractory AML. Moreover, MTF2 deficiency within healthy CD34⁺CD38⁻ cells triggers an H3K27me3 landscape similar to that found within refractory CD34⁺CD38⁻ AML cells (Fig. 1L and M).

To test whether MTF2 deficiency confers resistance to standard induction therapy, UCB-derived Lin⁻CD34⁺ hematopoietic progenitors and AML patient–derived Lin⁻CD34⁺ leukemic cells with normal (basal) levels of MTF2 (B-AML) were transduced with MTF2 shRNA GFP-tagged lentiviruses. GFP⁺ viable cells were sorted 72 hours after transduction and then treated with the induction-therapy drugs daunorubicin or cytarabine. Forty-eight hours later, less than 6% of the scramble shRNA control UCB Lin⁻CD34⁺ hematopoietic progenitors survived, whereas more than 60% of MTF2 knockdown UCB Lin⁻CD34⁺ cells were viable (Fig. 2A and B; Supplementary Fig. S7A and S7B). Similarly, more than 25% of MTF2 knockdown B-AML cells were viable, whereas only about 1% of the scramble control B-AML cells survived (Fig. 2C and D; Supplementary Figs. S8A, S8B, and S9). Furthermore, nearly two thirds of the viable MTF2 knockdown (KD) UCB Lin⁻CD34⁺ cells and more than 50% of the MTF2 knockdown B-AML cells were PCNA⁺, demonstrating that MTF2-deficient cells continue to proliferate despite chemotherapy treatment (Fig. 2E–H; Supplementary Fig. S8C and S8D; Supplementary Fig. S10). Moreover, at least 25% of cells were BrdUrd-positive in MTF2 knockdown B-AML cells following induction drug treatment, and more than 50% of these cells maintained their cycling profile based on their DNA content, compared with scramble control cells, of which only 4% incorporated BrdUrd and 17% were still cycling (Supplementary Figs. S11 and S12). Considering both daunorubicin and cytarabine target proliferating cells by inducing DNA damage (12, 13), we examined DNA damage accumulation in scramble and MTF2 shRNA–transduced UCB Lin⁻CD34⁺ and B-AML Lin⁻CD34⁺ cells over a 48-hour period, during which the samples were treated with either induction drug. DNA damage accumulation was assessed using the alkaline comet assay, where the Olive moment (product of the tail length and the fraction of total DNA in the tail) analysis was blinded. Across all time points, the highest levels of DNA damage were found in the scramble control UCB Lin⁻CD34⁺ hematopoietic progenitors (Fig. 2I and J; Supplementary Fig. S7C and S7D) and scramble control B-AML leukemic cells (Fig. 2K and L). Conversely, rescuing MTF2 expression in CD34⁺CD38⁻ cells from refractory MD-AML cells by lentivirus-induced MTF2 expression sensitized refractory AML cells to daunorubicin and cytarabine, with viability dropping to negligible levels by 48 hours (Fig. 2M and N; Supplementary Figs. S13 and S14). Thus, MTF2 deficiency is sufficient to confer a refractory phenotype because inducing MTF2 loss in UCB Lin⁻CD34⁺ hematopoietic progenitors or chemoresponsive leukemic (B-AML) cells renders them chemoresistant and able to proliferate despite DNA damage, whereas MTF2 rescue in refractory AML cells leads to chemosensitivity.

As shown in Supplementary Fig. S4, MTF2 silencing leads to reduced PRC2 core protein abundance including the methyltransferases EZH1 and EZH2, resulting in reduced global H3K27me3. To determine whether loss of core PRC2 proteins phenocopies MTF2-mediated chemoresistance, we first knocked down the core PRC2 proteins EED or SUZ12 within UCB Lin⁻CD34⁺ cells and treated them with daunorubicin or cytarabine over 48 hours. Although less than 1% of the scramble transduced cells were viable 48 hours after treatment with daunorubicin or cytarabine, 60% or more of SUZ12 knockdown and EED knockdown Lin⁻CD34⁺ cells remained viable (Supplementary Fig. S15).

Although EZH2 is the primary PRC2 methyltransferase in adult hematopoietic cells, EZH1 compensates for EZH2 loss (14); thus, we investigated whether inhibition of the methyltransferase activity of these enzymes would confer resistance to induction-therapy drugs. Both UCB Lin⁻CD34⁺ and Lin⁻CD34⁺ leukemic cells from 2 different patients with AML with normal (basal) levels of MTF2 (B-AML) and H3K27me3 were pretreated with the dual EZH1/EZH2 inhibitor UNC1999 (15) or EZH2 inhibitor EPZ005687 (16) for 3 days followed by treatment with daunorubicin or cytarabine for 48 hours. Interestingly, whereas less than 2% of the EZH2 inhibitor–treated cells were viable 48 hours after daunorubicin or cytarabine treatment, more than 65% and 25% of UCB Lin⁻CD34⁺ and B-AML cells, respectively, treated with the dual EZH1/2 inhibitor UNC1999 were viable (Fig. 2O–R; Supplementary Fig. S16). Furthermore, more than 50% of the UNC1999-treated viable cells were PCNA⁺ (Fig. 2S–V) and showed the lowest amount of DNA damage accumulation across all time points (Fig. 2W–Z). Taken together, these results show that dual loss of EZH1 and EZH2, but not EZH2 alone, which had been recently suggested (5), is required to confer resistance to standard induction therapy within hematopoietic progenitors and AML cells.

To identify potential molecular mechanisms underlying the poor therapeutic response in MTF2-deficient cells and altered DNA damage response (DDR), we knocked down MTF2 in UCB Lin⁻CD34⁺ cells by shRNA lentiviral transduction and performed RNA sequencing (RNA-seq). Enrichment map analysis of the transcriptomic data revealed DDR, antiapoptosis, and cell-cycle pathways were affected (Fig. 3A). Further dissection of the affected DDR transcripts showed numerous genes belonging to various DNA repair pathways were upregulated within the MTF2-deficient Lin⁻CD34⁺ hematopoietic progenitors compared with the scramble control, consistent with the reduced DNA damage observed in the MTF2-deficient Lin⁻CD34⁺ hematopoietic progenitors following induction treatment (Supplementary Fig. S17A and S17B). To identify targets of the MTF2–PRC2 complex, we integrated the H3K27me3 ChIP-seq (Fig. 1L and M) with the MTF2 knockdown RNA-seq data (Fig. 3A). The resultant MTF2–PRC2 gene regulatory network (GRN) in human Lin⁻CD34⁺ hematopoietic progenitors revealed the oncogenic pathways repressed by MTF2 included the PI3K and p53 pathways that regulate cell cycle, apoptosis, DDR, and chemoresistance (Fig. 3B). Although TP53 is not a direct target of MTF2–PRC2, the gene encoding the E3-ubiquitin ligase MDM2 that targets p53 for degradation (17) is a direct target. Thus, the p53 module within the MTF2 GRN predicts that MTF2 deficiency leads to MDM2 overexpression, resulting in p53 degradation (Fig. 3C).

Indeed, ChIP-seq analysis demonstrated that MTF2 deficiency leads to loss of H3K27me3 at the MDM2 locus, whereas total H3 levels were unaltered (Fig. 3D), with concomitant increased MDM2 mRNA levels observed by RNA-seq (Fig. 3E). The MTF2–PRC2 GRN was validated via RT-qPCR and ChIP-qPCR (Fig. 3F and G; Supplementary Fig. S17C and S17D; Supplementary Table S6). The ChIP-qPCR validation was confirmed using both formaldehyde cross-linked ChIP (X-ChIP) and Native ChIP strategies (Supplementary Fig. S18). Functional validation using imaging flow cytometry showed that MTF2 knockdown UCB Lin⁻CD34⁺ cells had high MDM2 and low p53 levels (Fig. 3H; Supplementary Fig. S19A–S19C). Therefore, we hypothesized that MDM2 inhibitors would restore p53 levels in MTF2-deficient cells. Strikingly, p53 levels within MTF2 knockdown UCB Lin⁻CD34⁺ hematopoietic progenitors and Lin⁻CD34⁺ leukemic cells from refractory patient BM aspirates (MD-AML) were rescued by treatment with Nutlin3A or MI-773, which block MDM2–p53 interaction (refs. 18, 19; Fig. 3H; Supplementary Fig. S19). Furthermore, cell-cycle and PCNA analyses revealed that rescuing p53 levels using MDM2 inhibitors within MTF2-deficient UCB Lin⁻CD34⁺ cells decreased their proliferation rate and shifted cells into the G₀–G₁ cell-cycle stage (Supplementary Fig. S20). To test whether MTF2 regulates the MDM2–p53 axis in refractory AML patient BM cells, MTF2 expression was rescued within four MTF2-deficient Lin⁻CD34⁺ refractory patient BM samples (MD-AML1-4) by lentivirus-mediated MTF2 expression; this resulted in MDM2 repression and elevated p53 protein levels (Supplementary Fig. S21). Taken together, these results indicate that MTF2 regulates p53 levels via transcriptional repression of MDM2 and that MDM2 inhibitors rescue p53 levels in MTF2-deficient cells.

Polycomb proteins have long been known as critical negative regulators of homeotic (HOX) gene expression (20). As indicated above, reduced EZH2 expression has recently been suggested to confer chemoresistance, mediated in part by derepression of HOXA9 and HOXB7 (5). Göllner and colleagues identified a HOXA9 gene signature using Gene Set Enrichment Analysis (GSEA) in a chemoresistant AML cell line, MV4-11R, compared with the parental cell line (5). In contrast, GSEA demonstrated a negative HOXA9 enrichment score for our MTF2 knockdown UCB Lin⁻CD34⁺ cell RNA-seq data (normalized enrichment score: −1.269 and −1.138, SH3 and SH7 knockdowns, respectively), suggesting that chemoresistance in MTF2-deficient cells is independent of HOX-driven mechanisms. Considering the negative enrichment scores, we questioned whether HOXA9 and HOXB7 are derepressed in MTF2 knockdown cells; in fact, HOXA9 and HOXB7 protein levels as measured by flow cytometry were increased by 0.7-fold or less in MTF2 knockdown cells (Supplementary Fig. S22A and S22B).

To test whether MDM2 inhibition results in changes in p53 targets and HOX gene mRNA levels, we treated MTF2 knockdown UCB Lin⁻CD34⁺ cells and MD-AML Lin⁻CD34⁺ cells with MDM2 inhibitors for 24 hours and analyzed HOXA9, HOXB7, PUMA, and CDKN1A expression by RT-qPCR. Although no change in HOX gene expression was observed, a marked increase in p53 target genes PUMA and CDKN1A was detected (Supplementary Fig. S22C–S22F), consistent with the increased p53 expression upon MDM2 inhibitor treatment in MTF2-deficient cells (Fig. 3H; Supplementary Fig. S19). These data demonstrate that although MTF2 deficiency results in the loss of p53 and repression of its target genes and an increase in HOX genes, treatment with MDM2 inhibitors rescues the levels of p53 and its targets without affecting the expression of other PRC2 targets such as HOX genes.

Under normal circumstances, activation of the p53 pathway inhibits the propagation of cells that carry damaged DNA. In AML, inactivation of p53 is associated with chemoresistance, refractory disease, adverse risk group, and poor survival (11). We therefore hypothesized that rescuing p53 levels in MTF2-deficient UCB Lin⁻CD34⁺ cells with MDM2 inhibitors would sensitize them to standard induction-therapy drugs. To test this hypothesis, we transduced and sorted viable GFP⁺ MTF2 knockdown Lin⁻CD34⁺ hematopoietic cells as above and treated them with daunorubicin or cytarabine in combination with the MDM2 inhibitors Nutlin3A or MI-773 and then assayed DNA damage accumulation and cell viability. Importantly, MTF2-deficient UCB Lin⁻CD34⁺ cells exposed to the dual treatment of an MDM2 inhibitor with an induction chemotherapeutic exhibited the same amount of DNA damage accumulation as that found within scramble controls and were sensitized to induction-therapy drugs (Fig. 4A–C). Moreover, by 48 hours, refractory MD-AML cells underwent apoptosis when treated with either daunorubicin or cytarabine in combination with Nutlin3A or MI-773, comparable to induction therapy–responsive MTF2 B-AML cells (Fig. 4D and E). Interestingly, the 24-hour time point demonstrates that Nutlin3A is more efficient at sensitizing MTF2-deficient refractory AML cells to daunorubicin than MI-773, whereas the 1 μmol/L dose of MDM2 inhibitors alone had minimal effect on viability (Supplementary Fig. S23).

The in vitro results led us to further test this combination therapy regimen (standard induction drugs plus MDM2 inhibition) in vivo. Mindful of the disparity between preclinical and clinical trial results (21), we designed a preclinical study that would closely mirror a clinical trial. Thus, we incorporated the following principles: AML patient-derived xenograft (PDX) mice would be treated only after the mice presented with a substantial leukemic burden, the efficacy of combination therapy would be compared against standard induction chemotherapy, and the study would be blinded. NOD/SCID IL2Rγ^null (NSG) mice were conditioned by sublethal irradiation and injected with either chemoresponsive B-AML or refractory MD-AML PDX cells. Once the transplanted mice presented with ≥20% CD45⁺CD33⁺ leukemic blasts in their peripheral blood, the mice were randomized into 4 groups, which were treated with either vehicle control, Nutlin3A, induction therapy, or combination therapy (induction therapy plus Nutlin3A). Although mice belonging to the chemoresponsive B-AML PDX cohort did not respond to vehicle control or Nutlin3A alone, the B-AML PDX mice treated with induction therapy or combination therapy survived until the experiment was terminated at 16 weeks after treatment (Supplementary Fig. S24; Supplementary Table S7). However, within the cohort of mice engrafted with refractory MD-AML PDX cells, only the mice treated with combination therapy survived until the experiment was terminated at 16 weeks after treatment (Fig. 4F; Supplementary Table S7). The weights of mice belonging to the refractory AML cohort were monitored throughout the study and reflected their survival, with weights plummeting during the 5-day treatment; only combination therapy–treated mice regained their weight and recovered (Fig. 4G). Wright–Giemsa staining of BM cells harvested at endpoint from the refractory cohort showed a stark contrast between the immature blast cells isolated from induction therapy–treated mice and the differentiated BM cells isolated from the combination therapy–treated mice (Fig. 4H). Flow cytometry analysis also confirmed a dramatic loss in the blast-containing CD45⁺CD33⁺ and CD34⁺CD38⁻ populations in the combination-therapy cohort (Supplementary Fig. S25; Supplementary Table S7).

To test whether the remaining PDX cells in the combination- treated mice within the MD-AML–refractory cohort retained residual disease, we performed secondary transplants without additional treatment. The secondary transplants showed no evidence of AML during the 16-week posttransplantation observation period, after which the experiment was terminated. Flow cytometry analysis of BM cells from secondary transplants revealed that the transplanted cells were capable of multilineage engraftment, suggesting that the combination therapy targeted bulk AML cells and LSCs while sparing some normal Lin⁻CD34⁺ stem and progenitor cells (Supplementary Fig. S26; Supplementary Table S8). Finally, flow-cytometric analysis showed the combination therapy–treated mice and their secondary transplant recipients had reduced BM cellularity, compared with DMSO, Nutlin3A alone, or induction-treated mice cohorts (Fig. 4I).

Altered epigenetic states such as H3K27me3 deficiency are observed in different types of cancers and have been correlated with poor outcome (4, 22). However, the mechanism underlying H3K27me3 deficiency and how it contributes to a poor prognosis has been unclear. Here we establish, using UCB Lin⁻CD34⁺ and patient-derived Lin⁻CD34⁺ leukemic cells, that MTF2/PRC2 epigenetically regulates the MDM2–p53 axis. Thus, MTF2 loss results in MDM2 upregulation and decreased p53-mediated apoptosis that contributes to refractory AML, which is overcome by treating the cells with induction drugs plus an MDM2 inhibitor. Furthermore, we demonstrate that MTF2 stabilizes PRC2. Therefore, loss of MTF2, EED, or SUZ12 results in chemoresistance, whereas loss of the methyltransferase EZH2 alone is insufficient to confer chemoresistance.

A recent study by Göllner and colleagues also found that low levels of H3K27me3 correlate with poor OS in AML (5); however, there are marked differences in the mechanistic explanations for this observation between the two studies. In short, Göllner and colleagues find the loss of EZH2 leads to derepression of HOXA9 and HOXB7, which then drives chemoresistance (5). Interestingly, there are many technical differences between the two studies that likely lead to the different conclusions reached. The primary difference is that although we utilized primary UCB cells and PDX-expanded patient-derived diagnostic samples without regard to specific leukemogenic driver mutations, Göllner and colleagues performed their mechanistic experiments using AML cell lines. In particular, they analyzed a drug-resistant FLT3-ITD⁺ AML cell line, MV4-11R, which demonstrated reduced EZH2 and H3K27me3 compared with the MV4-11 parental line (5). Interestingly, MV4-11R cells also encode TP53 mutations known to affect drug sensitivity. Thus, MV4-11R cells have undergone selective pressures leading to MDM2–p53 independent drug resistance, in which HOXA9 and HOXB7 play a role. Consistent with this point, we do not observe the HOXA9 gene signature in MTF2-deficient cells that was identified in MV4-11R cells. Göllner and colleagues also reported EZH2 inhibition alone drives chemoresistance (5). In contrast, we find that the refractory phenotype is only manifested when the methyltransferase activities of both EZH1 and EZH2 are inhibited. This is consistent with the fact that EZH1 can compensate for EZH2 within adult BM (14) and our finding that MTF2 is required for EZH1 and EZH2 expression in Lin⁻CD34⁺ cells. It should be noted that the inhibitor DZNep is known to have poor specificity to EZH2 (23), likely explaining Göllner and colleagues’ conclusion that EZH2 inhibition alone drives chemoresistance (5). Moreover, we demonstrate that sensitizing chemoresistant AML cells to induction drugs via MDM2 inhibitors is independent of alterations in HOXA9 and HOXB7 expression, the mechanism implicated in EZH2-mediated chemoresistance by the authors (5). Finally, while using flow-cytometric analysis of CD34⁺38⁻ cells from diagnostic patient samples, we found closer to 50% of diagnostic samples have low H3K27me3; Göllner and colleagues found more than 75% of patient samples have low H3K27me3 using IHC of bulk BM samples, suggesting the biomarker detection method is important to identify chemoresistant patients (5).

Seminal studies in AML defined LSCs or leukemia initiating cells (LIC) based on their ability to transplant the disease in sublethally irradiated mice. Using the limiting dilution assay, it has been shown that although LSCs/LICs are highly enriched within CD34⁺CD38⁻ cells, LSCs/LICs are also prevalent in other immunophenotypic AML subpopulations, albeit at significantly lower frequencies (24). We therefore profiled the H3K27me3 levels within the different immunophenotypic AML subpopulations from 32 diagnostic AML samples and observed that H3K27me3 levels can vary among the different BM subpopulations (Supplementary Fig. S1A and S1B). H3K27me3 levels within the CD34⁺CD38⁻ population best predicted response to induction therapy. For the first time, we demonstrate that reduced expression of the PRC2 protein MTF2 and low H3K27me3 levels within CD34⁺CD38⁻ AML cells prospectively identify refractory AML at diagnosis and provide a direct mechanism for MTF2- and PRC2-mediated chemoresistance.

AML is a clinically heterogeneous disease, with marked differences in survival following chemotherapy, based on blast cell morphology, age, and cytogenetics. Although therapeutic advances have lagged for nearly four decades, an increased number of prognostic factors have been identified that can predict patient OS, and these biomarker discoveries have driven the field's understanding of AML biology and the development of new therapeutic targets. In our study, using both our local patient cohort and the reported TCGA survival data, we demonstrate that MTF2 deficiency correlated with poor OS (Supplementary Fig. S3). Multivariate analysis of the TCGA data revealed low MTF2 expression as a prognostic marker when adjusted for effects of age, white blood cells (WBC), and cytogenetics. This is the case for TP53 wild-type patients (n = 153) and when the analysis is extended to all patients (n = 165; Supplementary Tables S5 and S9). Furthermore, MTF2 expression is capable of further stratifying patients with favorable cytogenetics (Supplementary Fig. S3). In contrast, at least in bulk BM samples, MDM2 transcript levels are not predictive of OS (Supplementary Tables S10 and S11). This may be explained by the fact that in bulk BM there is a weak inverse correlation between MDM2 and MTF2 transcript levels (Pearson r = −0.216, P = 0.005; Spearman s = −0.243, P = 0.002), suggesting a nonlinear inverse correlation. However, in CD34⁺CD38⁻ cells, we observe a stronger inverse correlation between MTF2 and MDM2 protein levels (Supplementary Figs. S19 and S21).

Mechanistically, we demonstrated that MTF2 repression is due to hypermethylation of its promoter, thus leading to derepression of MDM2, p53 degradation, and ensuing aberrant apoptosis, cell cycling, and DDR. Furthermore, rescuing MTF2 expression in primary MTF2-low LSC-enriched populations reversed chemoresistance, whereas knockdown of MTF2 in UCB Lin⁻CD34⁺ hematopoietic cells and AML Lin⁻CD34⁺ leukemic cells with normal levels of MTF2 induced resistance to induction-chemotherapy drugs. Because MTF2 plays a fundamental role in chemoresistance, MTF2 deficiency is not only correlative, but predictive of refractory AML. We also illustrate that the refractoriness observed within patients with AML with low H3K27me3 levels is driven by high MDM2 and low p53 levels.

Recent deep-sequencing studies have suggested that epigenetic alterations and changes in the epigenome, which are commonly observed across all cancers, are often associated with genetic mutations or transcriptional alterations of many genes related to cancer progression (25). Hence, targeting the epigenome is currently an area of intense research to improve upon standard-of-care chemotherapies (26, 27). Here, we demonstrate the reversibility of the AML epigenetic state by genetic manipulation of MTF2 and the use of small-molecule inhibitors against the catalytic activity of EZH1/EZH2. Furthermore, we target the dysregulated epigenetic pathway associated with refractory AML and demonstrate that MDM2 inhibitors in combination with standard induction chemotherapy provide a promising strategy to treat chemoresistant AML, a strategy recently proposed by Carvajal and colleagues (28). Moreover, in secondary transplants using the whole marrow isolated from our combination therapy–treated mice, we observed multilineage human hematopoietic cell engraftment, but no signs of disease (Supplementary Fig. S26; Supplementary Table S7). These results suggest that combination therapy preferentially targets refractory LSCs compared with healthy hematopoietic stem and progenitor cells, although additional experiments are required to directly test this hypothesis. We posit that similar strategies may be used to target the epigenetic states of other cancers.

Although TP53 dysfunction and mutations are typically associated with aggressive refractory disease (29, 30), large-scale deep-sequencing data sets show that only 2% to 8% of patients with AML have p53 mutations (11, 31). Therefore, our proposed combination therapy could effectively treat the majority of patients with refractory AML. However, our data predict that in the absence of MDM2 inhibitors, MTF2-deficient AML cells will cycle through induction therapy and accumulate additional mutations, potentially including TP53 therapy-induced mutations, which could render the cells resistant to MDM2 inhibitors (32). Because MDM2 inhibitors are now in clinical trials for a number of indications including treatment of AML, it would be prudent to screen patients with AML at diagnosis for MTF2 and H3K27me3 levels in CD34⁺CD38⁻ cells by flow cytometry to prospectively identify chemoresistant AML prior to treatment to identify candidates who will benefit the most from this targeted combination therapy.

In addition to the MDM2–p53 axis, which we found to be an underlying molecular mechanism for poor therapeutic response in MTF2-deficient cells, our MTF2 GRN identified other targets and mechanisms that may contribute to chemoresistance. Given their known roles in regulating cell-cycle progression, other potential interesting therapeutic targets include p14/ARF and Rb, as they play similar roles to that of p53. The manipulation of their expression or activity could also provide new strategies to target MTF2-deficient chemoresistant cells.

Additional details are provided in the Supplementary Methods and Data files.

All animal experiments were conducted with approval from the University of Ottawa Animal Care Committee, in accordance with the Canadian Council on Animal Care Standards and the Province of Ontario's Animals for Research Act. NSG mice were purchased from Jackson Labs. Mice were maintained in sterile housing conditions and given autoclaved chow and water ad libitum.

All patient samples and human UCB samples used in this study were performed after written informed consent was obtained from patients. Human UCB samples were obtained from the Canadian Blood Services (Ottawa, Canada) and the Mount Sinai Hospital Research Centre for Women's and Infants’ Health BioBank (Toronto, Canada). Informed consent and collection and research use of the human UCB samples were approved by the respective Research Ethics Boards and the Ottawa Hospital Research Ethics Board which operate according to the Declaration of Helsinki and the International Ethical Guidelines for Biomedical Research Involving Human Subjects. AML patient BM samples were collected after written informed consent was obtained from patients at The Ottawa Hospital. Procurement and use of AML BM patient samples for research was approved by The Ottawa Hospital Research Ethics Board, which operates according to the Declaration of Helsinki and the International Ethical Guidelines for Biomedical Research Involving Human Subjects.

Diagnostic BM samples from patients with AML were obtained after informed consent. Patients were treated as per the protocols at The Ottawa Hospital, Ontario, Canada. In addition to the usual supportive care, they received induction chemotherapy with idarubicin (12 mg/m² daily for 3 days) and cytarabine (200 mg/m²/day by continuous infusion for 7 days). A BM sampling was repeated upon recovery of blood counts and/or between days 25 and 40. Responsive disease was defined as <5% BM blasts. Patients who had ≥5% BM blasts were designated as failing or resistant to induction, for the purposes of this study. These patients were then treated with a salvage chemotherapy and considered for an allogeneic hematopoietic stem cell (HSC) transplant. Patients who did achieve a complete remission proceeded to receive consolidation and were considered for an allogeneic HSC transplant.

293T cells were seeded in 15-cm dishes (Corning Falcon). After 48 hours, cells were cotransfected with second-generation packaging lentiviral plasmids pMD2G, pPAX2, and either pGIPZ encoding shRNAs (Thermo Scientific) or pLenti overexpression vector (abm, Canada) in the presence of polyethylenimine (4.1 μmol/L) and NaCl (2.25 × 10⁻⁴ mol/L). Virus was produced using 40 to 80 15-cm dishes for each virus construct. Supernatant containing the virus was collected 48 and 72 hours after transfection and concentrated by ultracentrifugation and resuspended in a final volume of 300 μL of serum-free media. The viral titer for the pGIPZ and pLenti backbone viruses on each day of virus collection was ∼10⁶ and ∼10⁵ viral particles/μL, respectively.

UCB Lin⁻ or Lin⁻CD34⁺ cells or PDX-expanded primary AML BM Lin⁻CD34⁺ cells were cultured in 96-well plates (Falcon) under conditions described by Csaszar and colleagues, in Iscove's modified Dulbecco's medium (IMDM) containing BIT9500, 1% penicillin and streptomycin, stem cell factor (SCF; 100 ng/mL), thrombopoietin (50 ng/mL), FLT3 (100 ng/mL), 1% glutamax, and low-density lipoprotein (1 μg/mL); and Pabst and colleagues, in IMDM containing SCF (100 ng/mL), FLT3 (50 ng/mL), IL3 (20 ng/mL), G-CSF (20 ng/mL), 10⁻⁴ mol/L β-mercaptoethanol, and 15% BIT, respectively (33, 34). On day 1 of infection, 200,000 cells were incubated with polybrene (Sigma; 6 mg/mL) for 2 hours at 37°C, 5% CO₂, then combined with concentrated virus at a multiplicity of infection (MOI) of 100 or 30 for pGIPZ or pLenti backbone viruses, respectively. The functional MOI was previously calculated using UCB Lin⁻CD34⁺ cells. Cells were centrifuged at 400 × g for 20 minutes, then maintained at 37°C. On day 2, the infection was repeated. The functional MOI was determined based on GFP expression measured using BDFortessa or BDAstrios on both days of virus collection. Cells were grown for 3 days using a fed-batch culture system (34) and then sorted for high GFP expression and viability by negative selection of propidium iodide (BD Biosciences) using the Beckman Coulter Astrios sorter or MoFlo sorter (Beckman Coulter). The transduction of the AML cells was always performed using fresh virus. Using this protocol, we consistently transduced more than 50% of the cells.

UCB-derived Lin⁻CD34⁺ cells were transduced with either the GFP-tagged MTF2 shRNAs or a scrambled shRNA control. Transduced GFP⁺ cells were sorted 72 hours after transduction using a MoFlo sorter (Beckman Coulter). RNA was extracted from 150,000 cells per condition for RNA-seq analysis. RNA was isolated (Arcturus PicoPure Kit, LifeTech), DNase treated (Qiagen), and 2 μL of ERCC Spike-In mix (Thermo Fisher Scientific) was added to each of the GFP-tagged MTF2 shRNAs or a scrambled shRNA control. Quality of RNA was determined using a Bioanalyzer. Library preparation was performed using TruSeq Library Prep Kit (Illumina) and sequenced on a HiSeq 2000 (Illumina). Raw RNA-seq data are available in Gene Expression Omnibus (GEO; GSE98343).

For ChIP-seq, PDX-expanded CD34⁺CD38⁻ sorted cells (no lineage depletion was performed) were cross-linked with 1% formaldehyde for 10 minutes at room temperature. Samples were sheared using a Covaris sonicator (cycles/burst = 200; time: 480 minutes; temperature: 6°C to 8°C; water level: 12; peak power: 175) until DNA reached a final size of 100 to 300 bp. To ensure protein stability, the percentage of SDS in the lysis buffer was reduced to 0.1%. Upon sonication, one volume of dilution buffer (1% Triton X-100, 1.2 mmol/L EDTA, and 16.7 mmol/L Tris–HCL pH8.1) was added to neutralize the SDS. 750 ng of Drosophila chromatin spike-in (Active Motif) was added to each sonicated sample. Four micrograms of anti-H3K27me3 antibody (Cell Signaling Technology, c36B11) or H3 (Abcam, ab1791) was bound to preblocked protein A magnetic beads (Millipore) in combination with 2 μg of spike-in antibody (Active Motif) for 12 hours. The beads were then combined with sonicated sample containing Drosophila chromatin spike-in and incubated overnight. After incubation, beads were collected and washed with low-salt buffer (0.1% SDS, 1% Triton X-100, 2 mmol/L EDTA, 20 mmol/L Tris–HCL ph8.1, 150 nmol/L NaCL) and TE buffer (1 mol/L Tris–HCL ph8.0, 0.5 mol/L EDTA, 10% SDS and ddH₂O). The DNA–antibody complexes were eluted at 65°C while shaking at 1,250 g. Cross-links were reversed overnight at 65°C. Samples were treated with Proteinase K (Fisher Scientific) and RNase A (Fisher Scientific), and DNA was purified using phenol-chloroform. The purified DNA for each sample was resuspended in 15 μL. All ChIP-seq experiments were cell number normalized, and 150,000 cells per biological sample were used for each H3 and H3K27me3 ChIP experiment. Raw ChIP-seq data are available in GEO (GSE98378). Sequencing data can be found in GEO (GSE98380).

Briefly, female NSG mice (>7 week old) were sublethally irradiated with 300 Rads (Gammacell 3000) and transplanted via tail vein with 1 million bulk AML patient BM cells, which had been previously expanded and harvested from NSG mice. Upon the presence of >20% CD45⁺CD33⁺ cells in the peripheral blood (∼50–57 days after transplantation), mice were randomized into 4 treatment groups [n = 6 PDX samples (4 from refractory AML and 2 from basal AML); n = 2 mice per group; n = 8 mice total per group] and treated with (i) DMSO, (ii) Nutlin3A, (iii) induction therapy, or (iv) combination therapy. The person who administered the treatment intravenously via tail vein was blinded to the contents of the syringes. Mice belonging to the induction-therapy cohort were treated based on the previously published 5 + 3 treatment regimen (35) with the following modifications: (i) daunorubicin was used, (ii) drugs were administered intravenously via the tail vein and delivered in 2 × 200 μL doses 30 minutes apart, (iii) mice were fed a water-rich diet consisting of ice cubes, liquid gelatin, and autoclaved chow soaked in sterile water, and (iv) bedding was changed daily to minimize exposure to excreted drugs.

All data were expressed as mean ± SEM or SD. Data were analyzed using Prism 6.0 (GraphPad Software). Two-way ANOVA measured statistical significance between the conditions. Kaplan–Meier curves were calculated using GraphPad Prism 6.0, and significance was determined by log-rank test (Mantel–Cox). Pearson correlation was performed using SPSS v20.0. Multivariate Cox proportional hazards models were used to test the association of MDM2 and MTF2 with OS in parallel models. Age, risk cytogenetics, and WBC were included as identified important covariates in these models. The interaction between time and each of the covariates was included in the model one at a time, to assess if the effect of the given covariate is constant over time. There was no evidence for violation of proportional hazard assumption (P > 0.1). Adjusted hazard ratios with respective 95% confidence intervals were reported as results. SAS 9.3 was used to run the multivariable models. A P value of <0.05 was used as a cutoff to indicate statistical significance.

A.J. Ruthenburg has ownership interest (including stock, patents, etc.) in Epicypher Inc. and is a consultant/advisory board member for the same. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H.B. Maganti, H. Jrade, M. Sabloff, C.Y. Ito, W.L. Stanford

Development of methodology: H.B. Maganti, H. Jrade, W.L. Stanford

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H.B. Maganti, H. Jrade, C. Cafariello, J.L. Manias Rothberg, J. Yockell-Lelièvre, H.L. Battaion, S.T. Khan, Y. Li, A.T. Grzybowski, A.J. Ruthenburg, M. Sabloff

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H.B. Maganti, H. Jrade, C.J. Porter, J. Yockell-Lelièvre, A.T. Grzybowski, E. Sabri, F.J. Dilworth, T.J. Perkins, M. Sabloff, C.Y. Ito, W.L. Stanford

Writing, review, and/or revision of the manuscript: H.B. Maganti, H. Jrade, C. Cafariello, J. Yockell-Lelièvre, H.L. Battaion, A.T. Grzybowski, T.J. Perkins, M. Sabloff, C.Y. Ito, W.L. Stanford

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H.B. Maganti, H. Jrade, M. Sabloff

Study supervision: C.Y. Ito, W.L. Stanford

Other (conducted experiments and analyzed data): J.P. Howard

We thank the OHRI Stem Core Facility for their assistance with FACS and sequencing and the University of Ottawa/Ottawa Heart Institute Animal Care Facility for their invaluable assistance with the mouse experiments. This work was supported by operating grants from the Canadian Cancer Society Research Institute (2011-700822; 704783), Canadian Institutes of Health Research (MOP 142474), and the Cancer Research Society (CRS18330) to W.L. Stanford and C.Y. Ito, and the National Institutes of Health (R01-GM115945) to A.J. Ruthenburg. W.L. Stanford was supported by a Tier 1 Canada Research Chair in Integrative Stem Cell Biology.