Targeting Chromatin Regulators Inhibits Leukemogenic Gene Expression in NPM1 Mutant Leukemia

The clustered HOX genes are a highly conserved family of transcription factors that are expressed during early development and hematopoiesis (1–3). Specific members of the HOXA and HOXB cluster are required to maintain self-renewal properties of hematopoietic stem cells (HSC; ref. 3). Aberrant HOX gene expression is also a common feature of acute leukemias and can be activated by various oncogenes in murine leukemia models (3, 4). These genes are believed to play an important role during leukemogenesis because ectopic expression of specific members of the HOX cluster induces leukemic transformation of murine HSCs (5). In human acute myeloid leukemia (AML), aberrant HOX expression is found in 40% to 60% of the cases, whereas the specific HOX expression pattern is heterogeneous and has been used to categorize AML cases into four main groups that correlate with the presence of specific molecular markers (4). Simultaneous HOXA and HOXB cluster expression is most commonly found and frequently associated with the presence of an NPM1 mutation (4, 6, 7), one of the two most frequently mutated genes in AML (8).

Mutant NPM1 co-occurs often with other mutations in AML, most commonly in the FLT3 (60%) and/or the DNMT3A (50%) genes (9–11). Although NPM1^mut AML lacking an internal tandem duplication (ITD) in the FLT3 gene is considered a relatively favorable genotype with overall survival rates of up to 60% in younger patients, the presence of a concurrent FLT3-ITD converts this genotype into a less favorable category (9, 11, 12). The receptor tyrosine kinase FLT3 is of particular interest because (a) its mutation status is used as a prognostic indicator guiding treatment decisions in patients with AML and (b) it represents a molecular target for small-molecule inhibition. Clinical trials have shown activity of several FLT3 inhibitors against FLT3-mutated AML, but resistance occurred frequently and quickly (13). Currently, less than 50% of the (younger) patients with NPM1^mut FLT3-ITD–positive AML achieve a sustainable remission, and survival rates of patients with NPM1^mut AML >65 years of age are even more dismal (11, 14). These data highlight the need for novel therapeutic concepts, which may be less toxic and more efficacious.

NPM1 is an intracellular chaperone protein implicated in multiple cellular processes, such as proliferation and ribosome and nucleosome assembly (15, 16), but it has remained elusive how NPM1 mutations initiate and maintain AML, thus hindering the development of targeted therapeutic approaches. It is also unknown what mechanisms drive aberrant HOX gene expression in these leukemias. Our current knowledge of HOX gene regulation in leukemias has so far come from studies of MLL1-rearranged leukemias. In these leukemias, chromosomal translocations cause a fusion of the mixed-lineage leukemia gene (MLL1, also known as MLL or KMT2A) with one of more than 60 different fusion partners, resulting in the formation of an oncogenic fusion protein (17). MLL1-rearranged leukemias exhibit aberrant HOXA-cluster expression as a consequence of chromatin binding of the MLL1-fusion complex (18), which in turn requires the association with at least two other proteins, menin and LEDGF. Whereas menin facilitates LEDGF binding to the complex, the latter directly mediates chromatin binding of MLL1 (19, 20). Menin, the protein encoded by the multiple endocrine neoplasia 1 gene (MEN1), is of particular interest, as it is required by MLL1-fusion proteins for transformation and has been shown to be therapeutically targetable. Recently developed inhibitors of the menin–MLL1 interaction were demonstrated to have antileukemic activity in preclinical MLL1-rearranged leukemia models (21, 22). Although NPM1^mut AML lack MLL1-fusion proteins, studies of conditional MLL1 knockout mice have shown that wild-type MLL1 (WT-MLL1) is required for maintenance of HOX gene expression during normal hematopoiesis (23, 24). In addition, it has been shown that WT-MLL1 also requires the interaction with menin to maintain HOX expression and self-renewal properties of hematopoietic progenitors (25, 26).

The histone 3 lysine 79 (H3K79) methyltransferase DOT1L is another protein of therapeutic interest involved in HOXA cluster regulation of MLL1-rearranged leukemias (27–29). DOT1L is believed to be recruited to promoters of HOXA cluster genes via the MLL1-fusion partner that is commonly part of the DOT1L binding complex (30). Small-molecule inhibitors of DOT1L were proven to have activity against preclinical models of MLL1-rearranged leukemias (31, 32), and are currently being tested in a clinical trial (NCT01684150) with promising first results (33). DOT1L has also been linked to HOX gene control during normal hematopoiesis. Higher states of H3K79 methylation, such as dimethylation and trimethylation (me2 and me3) at the Hox locus, are associated with high expression levels of these genes in lineage⁻, Sca1⁺, c-Kit⁺ (LSK) murine hematopoietic progenitors and are converted into monomethylation (me1) as these cells mature into more committed progenitors lacking Hox expression (28). Whether Hoxb-cluster expression in LSKs is also associated with H3K79me2 and me3 is unknown to date.

We hypothesized that aberrant HOX and MEIS1 gene expression in NPM1^mut AML might be driven by similar mechanisms that control these genes in normal HSCs and that high expression of FLT3, a reported downstream target of MEIS1, is indirectly driven via elevated MEIS1 transcript levels. In the current study, we use CRISPR/Cas9 genome editing and small-molecule inhibition to identify the chromatin regulators MLL1 and DOT1L as therapeutic targets that control HOX, MEIS1, and FLT3 expression and differentiation in NPM1^mut leukemia.

Because MLL1 has been shown to regulate critical gene expression programs, including HOX gene expression in normal hematopoiesis and MLL1-rearranged AML, we hypothesized that it might also regulate HOX gene expression in other settings. As an initial assessment, we used an approach recently developed by Shi and colleagues that uses genome editing across exons encoding specific protein domains to determine the functional relevance of a given domain and the suitability for drug development (34). To assess potential dependencies of NPM1^mut AML on specific protein domains of MLL1 and its most similar family member MLL2 (also known as KMT2B), we designed a negative selection CRISPR/Cas9 screen interrogating multiple domains of both proteins (Supplementary Tables S1 and S2). We generated a tetracycline-inducible Cas9-expressing OCI-AML3 cell line (OCI-AML3-pCW-Cas9, Fig. 1A and B) as OCI-AML3 is the only commercially available human AML cell line harboring NPM1^mut. After transduction with 2 to 4 GFP⁺ single-guide RNAs (sgRNA) interrogating each domain, Cas9 was induced and the GFP⁺/GFP⁻ ratio assessed over a period of 15 days. Cells expressing sgRNAs targeting exon 1 that encodes the menin–LEDGF binding motif of MLL1 as defined by Huang and colleagues (35) were rapidly outcompeted by their GFP⁻ counterparts (Fig. 1C and D; fold change of GFP⁺ d0/d15 up to 100). Similar results were obtained for one of two sgRNAs targeting the first exon of RPA3, a gene required for DNA replication that is used as a positive control, whereas there was no difference in GFP expression over time in an empty-vector control guide (Fig. 1C). Having found selection against the sgRNAs targeting the menin–LEDGF binding motif, we performed SURVEYOR assays of sgRNAs targeting the menin–LEDGF binding motif and the N-HINGE-LOOP motif. The latter is located just 5′ from the menin–LEDGF locus but is considered as nonessential for the menin–MLL1 interaction and showed less negative selection in our screen. We detected indel mutations at day 3 for both sgRNAs but at day 9 only for the N-HINGE-LOOP sgRNA (Supplementary Fig. S1). These results are consistent with the selection data from our screen that demonstrate outcompeting of the cells transduced with the menin–LEDGF sgRNAs. Interestingly, sgRNAs targeting exons that encode the CXXC domain of MLL1, also known to be involved in chromatin binding and required for MLL1-fusion leukemia, were also selected against in our screen (Fig. 1C and E; refs. 36, 37). Of interest, no significant phenotype was observed for sgRNAs targeting the C-terminal SET domain of MLL1 (Fig. 1C). We validated these findings by performing a second screen using another independently engineered OCI-AML3-pCW-Cas9 clone (OCI-AML3-pCW-Cas9-C8; Fig. 1B) and obtained similar results, with sgRNAs targeting the menin–LEDGF binding motif again being the top hit (Supplementary Fig. S2A and S2B). We found no significant negative selection for any sgRNAs that interrogated the MLL2 protein domains (Fig. 1F). These data suggest that the menin–LEDGF binding motif of MLL1 but not MLL2 is required for NPM1^mut AML.

The genetic data presented above indicate that the menin–MLL1 interaction might be a therapeutic target in NPM1^mut AML. We therefore assessed whether recently developed small-molecule inhibitors of the menin–MLL1 interaction might influence proliferation and gene expression in NPM1^mut AML cells. First, we assessed the effects of MI-2-2, a small-molecule inhibitor of this protein interaction in the human NPM1^mut AML cell line OCI-AML3 (21). We observed a profound dose-dependent reduction in cell proliferation that was even more pronounced than in the MLL1-rearranged MOLM13 cells that served as a positive control (Fig. 2A). The HL60 AML cells lacking an NPM1^mut or MLL1 fusion showed only a mild cell growth–inhibitory effect at higher doses (Fig. 2A). During the work on this study, MI-503, a novel menin–MLL1 inhibitor, was described (22) that was synthesized by re-engineering the molecular scaffold of MI-2-2, resulting in enhanced drug-like properties and in vivo utility. To compare the two compounds and to define equivalent growth-inhibitory concentrations, we generated dose–response curves for both inhibitors, confirming the higher potency of MI-503 over MI-2-2 (Fig. 2A and Supplementary Fig. S3). MI-503 was used for most of the subsequent experiments.

We next assessed gene expression in the NPM1^mut AML cells and found a significant downregulation of HOXA cluster, HOXB cluster, and MEIS1 gene expression upon 4 days of treatment with MI-2-2 (Fig. 2B). Of these, MEIS1 appeared to be the most profoundly suppressed gene. This finding led us to explore a possible effect of menin–MLL1 inhibition (menin–MLL1-i) on FLT3 expression, as FLT3 is a reported downstream target of MEIS1 (38). In fact, FLT3 transcript levels were also substantially and highly significantly suppressed upon MI-2-2 treatment (Fig. 2B). To characterize the effects of pharmacologic menin–MLL1-i in more detail, we assessed apoptosis and differentiation following MI-2-2 treatment. Whereas induction of apoptosis was moderate (Supplementary Fig. S4), we observed profound differentiation following menin–MLL1-i, as reflected by an increase in CD11b expression over time as determined by flow cytometry [Fig. 2C; vehicle: 1.3%; MI-2-2: 76% (mean expression, day 7 of treatment)].

Next, we sought to validate our findings from the human OCI-AML3 cells on two independent murine conditional knock-in models of NPM1^mut leukemia. One of the leukemias is engineered to possess both an Npm1 mutation and a Flt3-ITD (Npm1^CA/+Flt3^ITD/+), whereas in the other model, secondary mutations were induced through use of a sleeping beauty transposon system (Npm1^CA/+Rosa^SB/+; refs. 39, 40). For in vitro drug testing, leukemic blasts were harvested from moribund primarily transplanted mice and cultured. As Npm1^CA/+Rosa^SB/+ cells did not grow in liquid culture medium but in methylcellulose only, we performed MI-503 treatment of these cells in colony-forming assays. Treatment for 7 and 14 days, respectively, resulted in a profound inhibitory effect on colony-forming potential in the Npm1^CA/+Rosa^SB/+ cells as well as in Mll-Af9 leukemia cells that served as a positive control, whereas Hoxa9-Meis1–transformed cells were unaffected (Fig. 2D). Similar results were obtained for Npm1^CA/+Flt3^ITD/+ cells that grow in liquid culture. These leukemias exhibited profound sensitivity to MI-503 and MI-2-2 treatment in a dose-dependent manner, which was even more pronounced than in the Mll-Af9 leukemia cells (Supplementary Fig. S3).

Next, we assessed the effects of MI-503 on Hox gene expression in both murine NPM1^mut leukemia models. Although baseline expression levels of individual Hox genes differed between the two murine leukemias, with for example lack of Hoxb2 expression in Npm1^CA/+Flt3^ITD/+ cells, we observed substantial suppression of Hoxa cluster and Hoxb cluster genes as well as Meis1. Of these, Meis1 was the most profoundly suppressed gene and again accompanied by substantial downregulation of global Flt3 transcript levels (Fig. 2E). To further explore whether the antiproliferative effects of MI-503 are causally related to Hox suppression in these leukemias, we expressed Meis1, Hoxb4, and Hoxa9-Meis1 ectopically in the Npm1^CA/+FLt3^ITD/+ murine leukemia cells. Of interest, all three scenarios rescued the Npm1^CA/+FLt3^ITD/+ cells from the menin–MLL1-i–mediated antiproliferative effect when exposed to MI-503 treatment for 11 days, as reflected by a shift of the dose–response curves toward higher concentrations and increased IC₅₀ values (Fig. 2F).

As changes in Hox gene expression were followed by myeloid differentiation in the human cells, we then determined whether MI-503 also induces differentiation in the murine leukemias. In fact, cytologic analysis of both Npm1^CA/+Rosa^SB/+ and Npm1^CA/+FLt3^ITD/+ cells revealed morphologic changes consistent with substantial myelo-monocytic differentiation (Fig. 2G and Supplementary Fig. S5).

We next sought to investigate the therapeutic effects of menin–MLL1-i on leukemia burden in vivo using a disseminated human OCI-AML3 xenotransplantation model. OCI-AML3 cells were transplanted into NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ (NSG) mice via tail-vein injection, and MI-503 treatment was initiated 7 days later. Animals were sacrificed after 7 (n = 2 per group) and 12 days (n = 3 per group) of MI-503 treatment [50 mg/kg twice daily (b.i.d.) by intraperitoneal injection (IP)]. Leukemia burden, as defined by the percentage of bone marrow cells expressing human CD45, was significantly reduced within the treated animal group compared with vehicle controls (Fig. 2H). To assess effects of MI-503 on gene expression in vivo, we analyzed sorted human CD45-positive bone marrow cells harvested from these animals for HOXB4, MEIS1, and FLT3. Twelve days of treatment resulted in dramatic suppression of these genes in the treated versus vehicle control animals (Fig. 2I).

We then explored the effects of MI-503 treatment on survival in the OCI-AML3 xenotransplantation model in a separate experiment. Treatment was initiated 7 days after transplantation at 50 mg/kg bid IP and continued for a total of 26 days excluding a 3-day break to allow recovery from local irritation at the injection site. MI-503 treatment resulted in a highly significant survival advantage compared with vehicle control animals (Fig. 2J; P = 0.0012; increase in median survival: 21%). Next, we sought to validate these findings in a murine model of NPM1^mut leukemia. Mice transplanted with secondary Npm1^CA/+Rosa^SB/+ leukemia cells were treated for 10 days with MI-503 versus vehicle control. Despite the aggressive character of this leukemia, with a disease latency of only 2 to 3 weeks, MI-503 significantly prolonged survival of leukemic animals (Supplementary Fig. S6; P = 0.0045, increase in median survival: 31%). These data indicate that therapeutic menin–MLL1-i reverses a HOX gene–dominated leukemogenic gene expression program in NPM1^mut leukemias in vitro and in vivo, thereby releasing the differentiation block in these cells and ultimately resulting in prolonged survival of mice suffering from NPM1^mut AML.

Menin associates with MLL1 at HOXA and MEIS1 promoters (19), and pharmacologic interruption of the menin–MLL1 interaction diminishes menin occupancy at the HOXA locus in MLL1-rearranged leukemias (41). We assessed the abundance of menin at HOXA, HOXB, and MEIS1 loci in the human NPM1^mut leukemia cell line before and after pharmacologic menin–MLL1-i. Chromatin immunoprecipitation (ChIP) of menin followed by qPCR demonstrated the presence of menin at expressed HOXA, HOXB, and MEIS1 gene loci, and menin was depleted from most of these loci after 5 days of MI-503 treatment (Fig. 3A). Reduction of menin enrichment was particularly pronounced at the MEIS1 locus, thereby correlating with findings from gene expression analysis (Figs. 2B and 3A).

Although we did not detect global changes of several chromatin marks upon menin–MLL1-i (Fig. 3B), we observed a locus-specific decrease of H3K4me3 at HOX and MEIS1 loci (Fig. 3C), thus demonstrating the expected changes in histone methylation due to menin–MLL1-i. Somewhat unexpectedly, we also found a significant reduction of H3K79me2 at those loci (Fig. 3D). These data are consistent with decreased binding of the menin–MLL1 complex to HOX and MEIS1 loci following pharmacologic menin–MLL1-i and also point to the H3K79 methyltransferase DOT1L as having a potential role in HOX gene regulation in NPM1^mut leukemia cells.

The data described above suggest WT-MLL1 to be important for HOX gene control in NPM1^mut AML, and also implicate DOT1L in HOX gene regulation in this disease. Recent studies of benign murine hematopoiesis demonstrated that DOT1L is another chromatin regulator critical for Hoxa-cluster regulation in early hematopoietic progenitors, whereas its role in Hoxb cluster control is not well defined (28). To explore whether DOT1L might be involved in the control of Hoxb-cluster genes, we reanalyzed gene expression data from Dot1l^fl/fl Mx1-Cre transgenic mice after polyinosinic–polycytidylic acid-induced excision compared with their normal counterparts (27). In fact, we found the anterior Hoxb cluster genes Hoxb2 and Hoxb4 to be significantly downregulated after homozygous deletion of Dot1l (Fig. 4A), whereas there was also a trend for Hoxb3 that did not reach statistical significance. In addition, there was an association between high expression levels of early Hoxb-cluster genes and higher states of H3K79 methylation (H3K79me2 and me3) in normal murine LSKs (Fig. 4B) mimicking the H3K79 profile across the expressed Hoxa cluster genes in these cells. These findings extend the concept of DOT1L being critical for Hoxa cluster regulation during normal hematopoiesis to the Hoxb cluster, which together represent also the dominant expression pattern found in NPM1^mut leukemias.

Based on these findings, we assessed the effect of EPZ4777, a small-molecule DOT1L inhibitor in NPM1^mut leukemias. DOT1L-inhibition (DOT1L-i) led to a profound reduction in cell proliferation and colony-forming potential in the human and both murine NPM1^mut leukemia models. Similar results were obtained for MLL1-fusion leukemias, whereas leukemias lacking an NPM1 mutation or MLL1 fusion such as human HL60 cells and murine Hoxa9-Meis1 in vitro–transformed cells were unaffected (Fig. 4C and D). Antiproliferative effects of DOT1L-i in the human cells were preceded by global and HOX and MEIS1 locus–specific reduction in H3K79me2 as determined by immunoblotting and ChIP-PCR, whereas global levels of other histone marks associated with transcriptional activation (H3K4me3, H3K36me3) and transcriptional repression (H3K27me2) were unchanged (Fig. 4E and F). As hypothesized, DOT1L-i resulted in significant repression of HOX genes in human and murine leukemia cells, although the pattern differed slightly among the models (Fig. 5A, left, and Fig. 5B). The murine leukemias exhibited significant suppression of Hoxa, Hoxb, and Meis1 genes (Fig. 5B), whereas the HOXB cluster was suppressed in human cells, but the HOXA cluster was only mildly affected, as assessed by RNA sequencing (Fig. 5A, left). Of note, MEIS1 was most profoundly downregulated across all NPM1^mut models. This finding was accompanied by dramatic FLT3 suppression (Fig. 5A, left, and B). To control the results from the RNA-sequencing analysis for potential bias from standard normalization in the context of global transcriptional changes as they might occur with global changes of chromatin marks, we normalized on synthetic RNA spike-in controls as proposed by Loven and colleagues and compared the results to standard normalization (42). No difference of global transcription levels was noted (data not shown), indicating that global reduction of H3K79me2 is associated not with global transcriptional downregulation in these cells but with suppression of a smaller HOX dominated gene subset. As EPZ4777 treatment resulted in profound suppression of Hox cluster genes followed by substantial growth inhibition of NPM1^mut leukemia cells, we next assessed whether retroviral overexpression of Hoxb4, Meis1, or Hoxa9-Meis1 influenced the treatment response to EPZ4777 of these leukemias. As demonstrated for menin–MLL1-i, ectopic expression of these genes abolished sensitivity of Npm1^CA/+FLt3^ITD/+ cells to DOT1L-i (Fig. 5C).

Because gene expression changes observed in response to EPZ4777 treatment were also consistent with cell differentiation (Fig. 5A, right), we next assessed the human NPM1^mut AML cells for differentiation following DOT1L-i. As expected, we observed a strong increase of CD11b expression as determined by flow cytometry and morphologic changes consistent with myeloid differentiation in the OCI-AML3 cells beginning at day 7 of treatment (Fig. 5D). Npm1^CA/+FLt3^ITD/+ and Npm1^CA/+Rosa^SB/+ cells also showed substantial monocytic cell differentiation after 14 days of pharmacologic DOT1L-i (Fig. 5E).

As HOX gene expression is associated with self-renewal properties, we next determined whether pharmacologic DOT1L-i inhibits NPM1^mut leukemia initiation in mice. Murine Npm1^CA/+Rosa^SB/+ leukemia cells were treated ex vivo with EPZ4777 or drug vehicle for 10 days, and equal numbers of viable cells were transplanted into sublethally irradiated recipient animals. A significant survival advantage was noted for the animals transplanted with treated cells compared with vehicle control in both NPM1^mut leukemia models (Fig. 5F and Supplementary Fig. S7), and animals transplanted with the EPZ4777-treated Npm1^CA/+Rosa^SB/+ leukemia cells had normal white blood cell (WBC) counts at disease onset of the vehicle control group (Fig. 5F–H). Similar results were obtained for the Npm1^CA/+FLt3^ITD/+ cells but with longer disease onset within the treated group (Supplementary Fig. S7). These data are therefore consistent with DOT1L participating in HOX gene regulation and differentiation control of NPM1^mut leukemia cells, thereby reprogramming these cells toward a state with substantially decreased leukemia-initiating and self-renewal properties.

In addition to the NPM1 mutation, OCI-AML3 cells harbor a DNMT3A mutation (R882C), which has been linked to aberrant HOX expression in AML in some studies (4) but was attributed to concomitant NPM1 mutations in others (43). To explore whether DNMT3A mutation status is associated with sensitivity to DOT1L-i, we included two additional cell lines with DNMT3A mutations in our study. Although the JAK2-mutated SET2 cells (which harbor the most common AML-associated DNMT3A mutation, R882H) did not show sensitivity to DOT1L-i, the OCI-AML2 cells (R635W reported in a single patient; ref. 44) were found to be tremendously sensitive (Fig. 4C). To understand these conflicting findings, we searched for other genetic factors potentially influencing sensitivity to EPZ4777 in the OCI-AML2 cells. Unexpectedly, RNA sequencing revealed the presence of an MLL–AF6 fusion transcript with breakpoints that aligned to the common breakpoint cluster region typically affected in MLL–AF6-rearranged leukemias that was confirmed by breakpoint spanning RT-PCR (Supplementary Fig. S8A). Whereas conventional cytogenetic analysis confirmed the normal 11q23 loci as reported, we detected a MLL1 split signal mapping to the derivative chromosome 1q using FISH (Supplementary Fig. S8B and S8C). These findings are consistent with a cryptic insertion of a functionally relevant MLL–AF6 fusion into the derivative chromosome 1. Also, we found the typical MLL1 fusion–related gene signature with high-level HOXA expressions that were profoundly downregulated upon DOT1L-i but lack of HOXB cluster expression (Supplementary Fig. S9). Together, these data support a previous study reporting that DOT1L controls HOXA cluster expression also in MLL1-fusion leukemia with no apparent DOT1L-recruiting activity (such as MLL–AF6; ref. 45). However, whether DNMT3A mutations are associated with sensitivity to DOT1L-i remains to be fully determined.

Our data are consistent with the menin–MLL1 interaction and DOT1L both being important regulators that control HOX, MEIS1, and FLT3 expression in NPM1^mut AML. We therefore sought to investigate the effects of combinatorial menin–MLL1-i and DOT1L-i in the three models of NPM1^mut AML. Combinatorial drug treatment in liquid culture resulted in improved growth inhibition as reflected by a shift of the dose–response curve toward lower doses compared with each of the compounds alone in the human OCI-AML3 and the murine Npm1^CA/+FLt3^ITD/+ cells (Fig. 6A and B). Of note, significant drug synergism was demonstrated for both the human and the murine NPM1^mut leukemia using the Chou–Talalay algorithm (ref. 46; Supplementary Fig. S10A and S10B; Supplementary Tables S3 and S4). Npm1^CA/+Rosa^SB/+ leukemia cells that grow only in methylcellulose also exhibited substantially superior suppression of colony-forming potential when treated with both compounds compared with single-drug treatment (Fig. 6C and D).

Gene expression analysis revealed that growth inhibition was preceded by enhanced suppression of leukemogenic gene expression after 4 days of treatment. Although suppression levels of individual HOX genes differed across the human and murine NPM1^mut leukemia models, we found that combinatorial treatment consistently suppressed MEIS1 transcript levels more than single-drug treatment (Fig. 6E and F). Although EPZ4777 does not reach full suppression of gene expression before 7 days of treatment, we noted that combination treatment resulted in an additional relative reduction of MEIS1 expression compared with single menin–MLL1-i after only 4 days (OCI-AML3, 73%, P = 0.03; Npm1^CA/+Flt3^Itd/+, 31%, P = 0.001; Npm1^CA/+Rosa^SB/+, 8%, P = 0.89). Similarly, we observed enhanced suppression of FLT3 in response to combinatorial treatment that reached significance in the OCI-AML3 and the Npm1^CA/+Flt3^Itd/+ cells (Fig. 6E and F). Consistent with these findings, OCI-AML3 cells treated with the combination regimen exhibited less RNA-Polymerase II binding to the MEIS1 transcriptional start site and gene body region than single-treated cells as assessed by ChIP-PCR (Supplementary Fig. S11), pointing to a direct inhibitory effect of MEIS1 transcription.

Next, we assessed differentiation following combinatorial menin–MLL1-i and DOT1L-i. In OCI-AML3 cells, differentiation was dramatically more pronounced with both compounds compared with single-drug treatment alone and occurred earlier. After 4 days of treatment, 77% of the cotreated OCI-AML3 cells expressed the monocytic differentiation marker CD11b compared with 54%, 15%, and 3% in the MI-2-2, EPZ4777, or vehicle control–treated cells, respectively (Fig. 6G, top plot). Flow-cytometric data were validated by cyto-morphologic analysis, which revealed more neutrophils or macrophages over blasts with 7 days of combinatorial versus single-drug treatment, particularly pronounced in the murine Npm1^CA/+Rosa^SB/+ cells (Fig. 6G bottom plot and Supplementary Fig. S12). We also detected a significant increase in apoptosis in the OCI-AML3 cells treated with both compounds compared with single treatment alone (Fig. 6H). Overall, these data suggest that combinatorial menin–MLL1-i and DOT1L-i has synergistic therapeutic effects in NPM1^mut leukemia likely conveyed by enhanced suppression of selected HOX genes, MEIS1, and FLT3 as well as induction of differentiation.

In order to assess the therapeutic potential of combinatorial menin–MLL1-i and DOT1L-i on primary AML patient samples, we used a human stromal cell coculture assay to maintain and treat patient samples in vitro. Of 5 de novo NPM1^mut AML patient samples (see patient details in Supplementary Table S5), 1 was not maintainable in culture so the other 4 were used for experiments. Ten days of single-drug treatment with MI-503 (2.5 μmol/L) or EPZ4777 (10 μmol/L) resulted in a profound reduction of viable cell counts compared with DMSO control in all 4 remaining primary AML samples, and we observed an enhanced antiproliferative effect for the combinatorial compared with single-drug treatment in 3 of 4 samples (Fig. 7A). Cell differentiation was also more pronounced with combination therapy (Fig. 7B and Supplementary Fig. S13). These data from primary NPM1^mut AML patient samples further support single and combinatorial menin–MLL1-i and DOT1L-i as a potentially efficacious therapeutic concept for NPM1^mut leukemia.

Our in vitro data showed a synergistic drug effect for simultaneous menin–MLL1-i and DOT1L-i in NPM1^mut leukemias. To assess the effects of combinatorial drug treatment on leukemia initiating potential, we transplanted equal numbers of viable Npm1^CA/+Rosa^SB/+ leukemia cells pretreated ex vivo with drug vehicle, EPZ4777, MI-503, or the combination into sublethally irradiated recipient mice and assessed for leukemia onset and survival. Whereas both EPZ4777 and MI-503 pretreatment led to significantly prolonged survival compared with vehicle control, we observed a significant survival advantage for the drug combination compared with either of the compounds alone (Fig. 7C). Leukemia onset was also significantly delayed as reflected by significantly lower white cell count or CD45.2 engraftment at the time of death from leukemia within the control group (Fig. 7D and Supplementary Fig. S14).

These data confirm that both menin–MLL1-i and DOT1L-i diminish leukemia initiation and indicate that simultaneous inhibition using both compounds shows enhanced effects on leukemia-initiating (stem) cells.

Despite being associated with high complete remission rates following standard chemotherapy, less than half of the patients with NPM1^mut AML achieve long-term disease-free survival due to older age at leukemia onset or concomitant adverse disease factors such as mutations in the FLT3 gene (9, 14). Research during the last decade has failed to establish mechanism-based targeted therapies specific for NPM1^mut AML, but described distinctive biological features of this disease such as a HOX dominated gene expression signature (4, 6). Efforts to therapeutically target these genes are challenging, as HOX and MEIS1 transcription factors are not amenable to direct pharmacologic inhibition. Lack of knowledge of how aberrant HOX expression is driven and maintained during NPM1^mut leukemic transformation has hindered attempts to indirectly target these genes (3).

Here, we have identified the menin–MLL1 interaction and DOT1L as therapeutic targets that control expression of HOX genes, the HOX co-factor MEIS1, and FLT3 in NPM1^mut AML. Using a CRISPR/Cas9 negative selection screen interrogating multiple MLL1 protein domains, we discovered the menin–LEDGF binding site of MLL1 as a dependency in our screen. This is the same region that is critically involved in chromatin binding of the MLL1 complex (19, 20). Also, we detected a strong phenotype for at least one of the sgRNAs targeting the CXXC domain, further pointing to chromatin binding of MLL1 as critically required for NPM1^mut leukemias (37). Furthermore, pharmacologic menin–MLL1-i diminishes menin enrichment at HOX and MEIS1 loci, resulting in dramatic suppression of HOX and MEIS1 expression. Together, these findings strongly support the concept that aberrant HOX/MEIS1 expression in NPM1^mut AML is regulated via chromatin binding of the menin–WT-MLL1 complex. Whether HOX gene activation is a direct consequence of MLL1 binding to chromatin itself or dependent on other proteins that are indirectly recruited to chromatin via MLL1 remains to be determined. It is of interest to note that none of the sgRNAs targeting the C-terminal SET domain of MLL1 containing the H3K4 methyltransferase activity exhibited any significant phenotype. These data are in line with a recent study on MLL1-fusion leukemias where selective inactivation of the SET domain did not alter transcription of HOX genes or transformation potential (47), but future studies will need to assess the importance of this domain with more detailed experiments. In addition, we found no evidence that the SET domain of MLL2, the other MLL family member interacting with menin, is involved in NPM1^mut leukemogenesis.

We have also shown that DOT1L is another chromatin modifier important for the control of HOX gene expression in NPM1^mut leukemias. We show that not only HOXA and MEIS1 but also HOXB expression is dependent on DOT1L in murine hematopoietic stem-cell progenitor cells (LSK cells) and the high expression levels that are associated with higher states of H3K79 methylation at those gene loci point to an important role for DOT1L. Although there is evidence that DOT1L is misdirected to HOXA cluster loci in MLL1-fusion leukemias via the fusion partner protein (30), the data presented here suggest that regulatory mechanisms that normally control HOX gene expression in hematopoietic cells remain critical in NPM1^mut AML. Just as we found with inhibition of the menin–MLL1 interaction, we found that pharmacologic DOT1L-i resulted in HOX downregulation, cell growth inhibition, and profound differentiation of NPM1^mut AML blasts. These data therefore extend the rationale for therapeutic DOT1L-i from MLL1-fusion leukemias to the much more prevalent NPM1^mut AML and prompt further study as to the mechanisms by which HOX genes are controlled in various genetics subtypes of AML.

A particularly striking finding was that the profound downregulation of MEIS1 following menin–MLL1-i was accompanied by almost complete suppression of FLT3 expression. FLT3 has been previously shown to be a transcriptional target of MEIS1 (38), and thus we believe that this suppression is most likely conveyed via MEIS1 rather than a direct effect caused by MLL1 or DOT1L-i. This observation might be of particular therapeutic importance as up to 66% of NPM1^mut AMLs harbor concomitant activating FLT3 mutations associated with adverse outcome (8, 10, 11). Thus, menin–MLL1-i may be particularly attractive for patients exhibiting the unfavorable NPM1^mut FLT3-ITD genotype.

Whereas our data provide evidence that HOX genes and MEIS1 in NPM1^mut leukemias are regulated on a chromatin level via MLL1 and DOT1L, the specific role of the mutant NPM1 protein with regard to HOX and MEIS1 regulation remains elusive. However, the strong association between HOX expression in human NPM1^mut AML or genetic knock-in models of NPM1^mut leukemia in mice together with the above-presented data suggests that the mutant NPM1 protein acts upstream of menin–MLL1 and DOT1L to deregulate gene expression, at least partially analogous to the functions of MLL1-fusion proteins in MLL-rearranged leukemias. The strong oncogenic transformation potential of HOX genes in general suggests that these genes are likely required for NPM1^mut-driven leukemogenesis. Whereas formal proof of this view was lacking, our data provide strong support for this concept, as exogenous overexpression of selected HOX genes rescues antiproliferative effects of both DOT1L and menin–MLL1 inhibitors. We further show that combinatorial inhibition of menin–MLL1 and DOT1L has enhanced on-target activity as reflected by more profound HOX and MEIS1 suppression. These changes result in a synergistic antiproliferative and boosted differentiation effect, suggesting that therapeutic HOX targeting ultimately releases the differentiation block of NPM1^mut AML blasts.

The availability of menin–MLL1 and DOT1L inhibitors should enable relatively quick translation of our findings into clinical testing. However, future studies will determine whether the potent antileukemic activity can be best harnessed when introduced into current standard chemotherapy regimens, thereby overcoming possible context-specific escape mechanisms that are frequently observed in AML. A proof-of-principle study on leukemia cell lines already demonstrated drug synergism of EPZ-5676 with standard chemotherapeutic agents in vitro (48). Another interesting combination partner might be dactinomycin that directly inhibits transcriptional elongation by inhibition of RNA polymerases and was reported to have activity against NPM1^mut leukemia in an anecdotal report on a single patient (49).

In summary, our data show that NPM1^mut leukemogenesis is dependent on HOX and MEIS1 expression, which is controlled by specific chromatin-regulatory complexes. We further demonstrate that these genes are therapeutically targetable via the menin–MLL1 interaction and the H3K79 methyltransferase DOT1L. Small-molecule inhibition of these two histone modifiers releases the differentiation block of NPM1^mut leukemic blasts. Both compounds as single agents or in combination represent novel and possibly less toxic therapeutic opportunities for the relatively common NPM1^mut leukemias and represent an attractive concept particularly for patients with the prognostically adverse genotype with concomitant FLT3-ITD and the difficult-to-treat elderly population.

The AML cell lines OCI-AML3, OCI-AML2, SET2, MOLM13, and HL60, as well as 293T cells and Hs27 cells, were maintained under standard conditions. Cell line authentication testing (ATCC) was performed between March and November 2013 and verified identity and purity for all human AML cell lines used in this study. The murine leukemia models Npm1^CA/+Flt3^ITD/+ and Npm1^CA/+Rosa^SB/+ were described previously (39, 40), and cells were cultured as previously described (29). With regard to the Npm1^CA/+Rosa^SB/+ model, leukemias with two different sleeping beauty integration sites were used (A: target gene: Csf2, position: 54064482, and target gene: Mll1, position: 44649906; B: target gene: Csf2, position: 54250117; genomic positions are as per Ensemble assembly NCBIm37) for all in vitro studies. Experiments were performed using A, as these cells were found to have a slightly higher colony-forming potential in vitro. Confirmation of the results was then performed with B.

The pCW-Cas9 expression construct and the pLKO5.sgRNA.EFS.GFP vector were purchased from Addgene (#50661 and #57822). All sgRNAs in this study were designed using the Zhang laboratory's CRISPR design webpage at MIT, and sequences are listed in Supplementary Tables S1 and S2. OCI-AML3-pCW-Cas9 cells were derived by retroviral transduction of OCI-AML3 cells with pCW-Cas9, followed by puromycin selection. Cells were plated in methylcellulose to obtain single-cell–derived clones, and screened for Cas9 expression by immunoblotting using an anti-flag antibody. The MLL1 and MLL2 domain screen was performed in duplicate and repeated with two independent clones. Three days following sgRNA transduction, Cas9 was induced and GFP⁺/GFP⁻ ratio determined every 3 days for 15 days as described (34). For more details, see Supplementary Methods.

RNA purification, cDNA synthesis, qRT-PCR, immunoblotting, ChIP-PCR, flow cytometry, retroviral overexpression, and in vitro drug treatment in colony-forming or cell viability assays were performed using standard procedures. A detailed description is provided in the supplementary material. Drug synergy was assessed in viability assays comparing single versus constant ratios of combinatorial drug treatment using the Chou–Talalay method (46) and is also described in detail in the supplementary material.

For gene expression analysis, biological triplicates of OCI-AML3 cells were treated for 7 days with DMSO or 10 μmol/L EPZ4777. ERCC synthetic spike-in controls were added to cell lysis buffer of each sample based on cell number and prior to RNA isolation. For fusion detection, triplicates of untreated OCI-AML2 cells were harvested for RNA purification. See Supplementary Methods for RNA purification, library construction, and sequencing analysis. Data were made publicly available at the Gene Expression Omnibus (accession code: GSE85107).

For in vivo treatment experiments, 7-to-10-week-old female NSG mice were injected via tail vein with 5 × 10⁶ OCI-AML3 cells. Animals were randomized to vehicle (25% DMSO, 25% PEG 400, and 50% normal saline) or MI-503 (50 mg/kg bid IP) starting on day 5 after transplantation. For assessment of leukemia burden, mice were sacrificed between 7 and 12 days of drug treatment and harvested bone marrow cells assessed for human CD45 expression using flow cytometry. For gene expression analysis, RNA was isolated from MACS-sorted human CD45⁺ cells. For survival analysis, treatment was initiated on day 7 for 26 days in total with a 3-day treatment break to allow partial recovery from local irritation at the injection sites. Mice were monitored daily for clinical symptoms. Moribund animals were euthanized when they displayed signs of terminal leukemic disease. All mouse experiments were approved by the Institutional Animal Care and Use Committee at Memorial Sloan Kettering Cancer Center.

1 × 10⁶Npm1 ^CA/+Rosa^SB/+ (model A) secondary leukemia cells, harvested from moribund animals, were transplanted into sublethally irradiated (200 cGy) female NSG mice via tail vein injection. Mice were randomized to vehicle or MI-503 (50 mg/kg, bid, 10 days IP), and treatment initiated 7 days after transplantation.

(a) Npm1^CA/+Flt3^ITD/+ and Npm1^CA/+Rosa^SB/+ (model A) secondary leukemia cells were treated with DMSO or EPZ4777 (10 μmol/L) for 11 days in tissue culture or (b) Npm1^CA/+Rosa^SB/+ secondary leukemia cells for 10 days with either DMSO, EPZ4777 (10 μmol/L), or 4 days of MI-503 (2.5 μmol/L) plus 6 days of DMSO or 6 days of EPZ4777 (10 μmol/L) plus 4 days of combinatorial EPZ4777 (10 μmol/L) + MI-503 (2.5 μmol/L). Pretreated viable cells (1 × 10⁶) of each group were then transplanted into sublethally irradiated female NSG (Npm1^CA/+Rosa^SB/+) or C57BL/6J (irradiated with 600cGy; Npm1^CA/+Flt3^ITD/+) recipient animals.

The coculture treatment assay of primary human AML samples was performed as previously reported (50) and is described in the supplementary material. Human AML samples were obtained under Institutional Review Board–approved protocols from patients treated at Memorial Sloan Kettering Cancer Center following written informed consent.

Statistical test computations were performed using Graph Pad Prism (v6). Statistical significance for the in vitro assays was calculated using the Student t test. Survival analysis was estimated using the Kaplan–Meier method, and P values were calculated using the log-rank test. P values were displayed as follows: *, P < 0.05; **, P < 0.005; ***, P < 0.0005. In vitro assessment of menin–MLL1-i in the NPM1^mut AML models was performed in at least two independent experiments each performed in three replicates using one of the two menin–MLL1 inhibitors (MI-2-2 or MI-503). In addition, results for most experiments were confirmed once with the other menin–MLL1 inhibitor. In vitro studies related to DOT1L-i were performed in three independent experiments each performed in three replicates.

J.E. Bradner is President at Novartis Institute of BioMedical Research. G.S. Vassiliou reports receiving a commercial research grant from Celgene and is a consultant/advisory board member for Kymab Ltd. S.A. Armstrong is a consultant/advisory board member for Epizyme, Inc., and Vitae Pharmaceuticals. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.W.M. Kühn, C.-W. Chen, S.A. Armstrong

Development of methodology: M.W.M. Kühn, Z. Feng, C.-W. Chen, A.J. Deshpande, M. Cusan, J.E. Bradner, S.A. Armstrong

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.W.M. Kühn, E. Song, A. Mupo, C. Grove, E. de Stanchina, S.A. Armstrong

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.W.M. Kühn, E. Song, A. Sinha, N. Farnoud, R. Koche, S.A. Armstrong

Writing, review, and/or revision of the manuscript: M.W.M. Kühn, N. Farnoud, J.E. Bradner, S.A. Armstrong

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Z. Feng, T. Hoshii, S.A. Armstrong

Study supervision: S.A. Armstrong

Other [contributed mouse Npm1c-driven leukemias (live cells)]: G.S. Vassiliou

We thank members of the Armstrong Laboratory for helpful comments on the article; Rowena Eng, Huiyong Zhao, and Michael Hadler for excellent technical assistance; and Olga Guryanova and Ross Levine for the SET2 and OCI-AML2 cell lines.

This study was supported by grants (to S.A. Armstrong) from the NIH (CA66996, CA140575, and CA176745) and the Leukemia and Lymphoma Society. E. de Stanchina was supported by an NIH grant (P30 CA008748). M.W.M. Kühn was supported by the German Research Foundation (DFG; KU 2688/1-1).

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.