A model system for studying the DNMT3A hotspot mutation (DNMT3A^R882) demonstrates a causal relationship between its dominant-negative effect and leukemogenesis

Abstract

Mutation of DNA methyltransferase 3A at arginine 882 (DNMT3A^R882mut) is prevalent in hematological cancers and disorders. Recently, DNMT3A^R882mut has been shown to have hypomorphic, dominant-negative, and/or gain-of-function effects on DNA methylation under different biological contexts. However, the causal role for such a multifaceted effect of DNMT3A^R882mut in leukemogenesis remains undetermined. Here, we report TF-1 leukemia cells as a robust system useful for modeling the DNMT3A^R882mut-dependent transformation and for dissecting the cause-effect relationship between multifaceted activities of DNMT3A^R882mut and leukemic transformation. Ectopic expression of DNMT3A^R882mut and not wildtype DNMT3A promoted TF-1 cell transformation characterized by cytokine-independent growth, and induces CpG hypomethylation predominantly at enhancers. This effect was dose-dependent, acted synergistically with the isocitrate dehydrogenase 1 (IDH1) mutation, and resembled what was seen in human leukemia patients carrying DNMT3A^R882mut. The transformation- and hypomethylation-inducing capacities of DNMT3A^R882mut relied on a motif involved in heterodimerization whereas its various chromatin-binding domains were dispensable. Mutation of the heterodimerization motif that interferes with DNMT3A^R882mut binding to endogenous wildtype DNMT proteins partially reversed the CpG hypomethylation phenotype caused by DNMT3A^R882mut, thus supporting a dominant-negative mechanism in cells. In mice, bromodomain inhibition repressed gene-activation events downstream of DNMT3A^R882mut-induced CpG hypomethylation, thereby suppressing leukemogenesis mediated by DNMT3A^R882mut. Collectively, this study reports a model system useful for studying DNMT3A^R882mut, shows a requirement of the dominant-negative effect by DNMT3A^R882mut for leukemogenesis, and describes an attractive strategy for the treatment of leukemias carrying DNMT3A^R882mut.

Introduction

Aberration of the epigenomic state is commonly utilized by tumors to alter gene-expression programs and to gain growth advantage (1,2). Sequencing of primary cancer samples has identified recurrent mutations of genes involved in epigenomic regulation (2,3). In particular, somatic mutation of DNA methyltransferase 3A (DNMT3A^mut) was detected in a wide range of blood cancers including 20–30% of acute myeloid leukemia (AML) (3–7), as well as elderly individuals with clonal hematopoiesis (8–11).

DNMT3A forms a complex with accessory cofactors, serving as one of the major de novo DNA methyltransferases (12–14). DNMT3A harbors various motifs, which include a N-terminal domain (NTD) shown to interact with transcription factors (7), a Pro-Trp-Trp-Pro (PWWP) domain shown to engage methylated histone H3 lysine 36 (H3K36me) (15), an ATRX-DNMT3-DNMT3L (ADD) domain known to bind specifically to the unmodified histone H3 lysine 4 (H3K4me0) (14,16), and a C-terminal catalytic domain that methylates cytosine bases, especially those in the CpG dinucleotides (12–14). Cellular contexts such as interacting partners and chromatin states are crucial for exquisite modulation of DNMT3A’s genomic targeting and enzymatic functions. For example, DNMT3A adopts an auto-inhibitory conformation due to interaction between its ADD and methyltransferase domains, and such self-inhibition is released upon engagement of ADD to histone tails with H3K4me0 (14). The methyltransferase domain, which binds DNA using specified protein motifs (12), also contains crucial interfaces for forming DNMT dimers, tetramers and/or oligomers to regulate the methylation activities (13,14,17–22).

DNMT3A^mut is primarily heterozygotes in AMLs and shows a mutational hotspot at the Arg882 residue (DNMT3A^R882mut), which accounts for 50–60% of identified DNMT3A^mut in AMLs (2,3,7,23). Due to prevalence and clinical relevance of DNMT3A^R882mut in blood cancer and clonal hematopoiesis, considerable progress was made in understanding the mechanisms by which DNMT3A^R882mut mediates transformation. DNMT3A^R882mut is detected in hematopoietic stem/progenitor cells (HSPCs) of apparently healthy elderly individuals, supporting its role as a pre-leukemic founder mutation that provides initial selective advantage of mutant HSPC clones (8–11). We and others have shown that a cooperating genetic lesion is required for DNMT3A^R882mut or Dnmt3a loss to induce fully-blown leukemias in mice (24–28). Biochemically, partial loss-of-function, dominant-negative and gain-of-function effects have all been associated to DNMT3A^R882mut. First, DNMT3A^R882mut is a hypomorphic allele and purified DNMT3A^R882mut enzymes display reduced methyltransferase activity on CpG substrates in vitro (4,12,29,30). Particularly, the structure of the DNMT3A-DNMT3L-CpG complexes was recently solved, which revealed that the residue R882 forms interactions with both DNA substrates and a so-called ‘Target Recognition Domain’ loop, a DNMT3A motif critically involved in engaging CpG dinucleotides (12). Furthermore, the dominant-negative effect was proposed for DNMT3A^R882mut (29,31). Here, DNMT3A^R882mut associates with wildtype DNMT3A and DNMT3B, presumably interfering with the formation, stability, DNA-engaging and/or DNA-methylating activity of the whole complex. The combined hypomorphic and dominant-negative effects of DNMT3A^R882mut may explain focal CpG hypomethylation seen in leukemias harboring DNMT3A^R882mut. On the other hand, recent studies reported an altered substrate preference of DNMT3A^R882mut towards CpG sites with specific flanking sequence, which is termed as the gain-of-function effect of DNMT3A^R882mut (32). Theoretically, these above effects of DNMT3A^R882mut are not mutually exclusive, indicating that DNMT3A hotspot mutation causes redistribution of DNA methylation in the cancer genome. However, it remains elusive by what effect(s) DNMT3A^R882mut contributes to transformation.

This study aims to determine a causal relationship between the multifaceted effect of DNMT3A^R882mut and leukemic transformation. Previously, the systems reported for studying the leukemia-promoting functions of DNMT3A^R882mut used retrovirus-based or knock-in expression of DNMT3A^R882mut, often in combination with other mutations, in HSPCs (24–28), which represents a challenge to perform cause-effect relationship studies of DNMT3A^R882mut. We here report a leukemia cell line, TF-1 cells, as a straightforward, robust cell model for assaying the DNMT3A^R882mut-dependent leukemic transformation and, more importantly, for dissecting the molecular basis by which DNMT3A^R882mut mediates transformation. We show that enforced expression of DNMT3A^R882mut alone, but not DNMT3A^WT, is sufficient to induce TF-1 cell transformation characterized by cytokine-independent growth and arrested cell differentiation. In TF-1 cells, DNMT3A^R882mut induces hypomethylation of CpG sites that significantly overlap those found in primary AMLs with DNMT3A^R882mut. With this model system, we also demonstrated the dosage effect of DNMT3A^R882mut on transformation, as well as cooperativity between DNMT3A^R882mut and IDH1 mutation, the two lesions coexisting in human leukemias. Importantly, this system has allowed us to systemically analyze dependency of various functional motifs within DNMT3A^R882mut during transformation. Here we identified a heterodimerization domain (also known as DNMT3A’s heterotetramer or FF interface) (13,18–21), and not its catalytic and chromatin-binding domains, to be essential for TF-1 cell transformation. The mutation of the heterodimerization interface interfered with interaction of DNMT3A^R882mut with endogenous wildtype DNMT proteins and partially reversed CpG hypomethylation caused by DNMT3A^R882mut, supporting a dominant-negative mechanism that underlies transformation. Lastly, we found the bromodomain inhibitor suppressed the gene-activation programs associated with DNMT3A^R882mut-induced hypomethylation. Collectively, this study describes a model system for studying the DNMT3A^R882mut-related leukemogenesis, demonstrates a cause-effect relationship between the dominant-negative effect and DNMT3A^R882mut-mediated transformation, and provides an attractive therapeutic means for treatment of leukemias carrying DNMT3A^R882mut.

Materials and Methods

Cell lines

The human erythroleukemic cell line TF-1 (ATCC #CRL-2003) is cultured in the RPMI 1640 medium (Invitrogen) containing 10% of FBS and 2 ng/ml of recombinant human granulocyte-macrophage colony-stimulating factor (GM-CSF, R&D Systems) as described (12). Authentication of identities of parental and derived cell lines was ensured by the Tissue Culture Facility affiliated to UNC Lineberger Comprehensive Cancer Center with the genetic signature profiling and fingerprinting analysis. Every month, a routine examination of cell lines in culture for any possible mycoplasma contamination was performed using commercially available detection kits (Lonza). Cells with less than of 10 times of passages were used in the study.

Co-immunoprecipitation (CoIP)

CoIP was carried out as previously described (33).

DNA methylation array and data analysis

Genomic DNA was extracted using the DNeasy Blood & Tissue Kit (Qiagen), followed by bisulfite-conversion using the EZ DNA Methylation-Gold Kit (Zymo Research). DNA methylation profiling was performed by the UNC Genomics Core using the Infinium HumanMethylation450 BeadChip (lllumina) according to the manufacturer’s instructions. Methylation data were then subject to background subtraction and control normalization by executing preprocessIllumina in the R ‘minfi’ package (34). Differentially methylated CpGs were identified using dmpFinder in a categorical mode. Methylation changes were considered significant at a q-value of less than 0.05 and a beta value difference of more than 0.1. Hierarchical clustering analysis, scatter plots and density plots were generated in R using ‘pheatmap’, and ‘ggplot2’ packages as previously described (12).

Chromatin immunoprecipitation followed by deep sequencing (ChIP-Seq)

ChIP-seq was carried out using BRD4 antibody (Sigma, HPA015055) as previously described (24,35).

Animal models

All animal experiments were reviewed and approved by the Institutional Animal Care and Use Committee of UNC. Establishment and phenotypic analysis of murine leukemia models were performed as previously described (24,36,37) and in vivo dosing of I-BET151 carried out with a described procedure (35).

Data availability

The Genomics data produced by this study, including ChIP-Seq and methylome, have been deposited in Gene Expression Omnibus (GEO) under accession code GEO: GSE130094, GSE130634 and GSE71475.

The details of experimental procedures are provided in Supplementary Materials and Methods.

Results

DNMT3A^R882mut induces transformation of TF-1 cells characterized by cytokine-independent growth.

To delineate the molecular underpinnings of DNMT3A^R882mut-mediated leukemogenesis, we sought to identify a robust, straightforward cell transformation system that allows cause-effect relationship studies of DNMT3A^R882mut. We chosen to use TF-1 cells, for this leukemia line normally relies on survival-supporting cytokines to sustain cell proliferation but displays cytokine-independent growth upon acquisition of certain leukemia-related alterations such as IDH2 or TET2 mutation (38,39). To test whether DNMT3A^R882mut has transformation activities in this model, we stably expressed comparable levels of DNMT3A^WT or a prevalent form of DNMT3A^R882mut, Arg882His (DNMT3A^R882H), into TF-1 cells (Fig. 1A insert, and Supplementary Fig. 1A–B). In the presence of survival cytokines, TF-1 cells with either DNMT3A form showed the same rates of proliferation (Fig. 1A); however, under cytokine-poor conditions, only TF-1 cells with DNMT3A^R882H, not vehicle or DNMT3A^WT, demonstrated robust growth (Figure 1B). Relative to controls, TF-1 cells with DNMT3A^R882H also showed a mild but reproducible differentiation arrest in response to differentiation signals (Supplementary Fig. 1C–D). As some DNMT3A mutations found in hematological cancer are damaging ones such as frame-shifting, we asked whether TF-1 cell transformation can be recapitulated by DNMT3A loss. Indeed, knockdown of DNMT3A by independent shRNAs, and not mock, resulted in cytokine-independent proliferation (Supplementary Fig. 1E and Fig. 1C). Similar phenotypes were observed with other DNMT3A^R882mut variants (Arg882Cys and Arg882Ser; Fig. 1D). Next, we ask whether such a transformation phenotype requires continuous expression of DNMT3A^R882H. Here, a reversible expression system, which carries a pair of LoxP sites flanking the transgene (Fig. 1F, upper), was used to generate cytokine-independent TF-1 lines with DNMT3A^R882H. After cre-mediated deletion of DNMT3A^R882mut (Figure 1F, bottom; Supplementary Fig. 1F), we found that these cells no longer sustained cytokine-independent growth (Figure 1G). Collectively, we show that DNMT3A^R882mut induces cytokine-independent growth of TF-1 cells and maintenance of this phenotype relies on continuous presence of DNMT3A^R882mut.

Figure 1. R882H-mutated DNMT3A (DNMT3A^R882mut), and not the wild-type one (DNMT3A^WT), induces transformation of TF-1 cells characterized by cytokine-independent growth.

(A-B) Proliferation of TF-1 cells with stable transduction of the indicated gene in the presence (A) and absence (B) of GM-CSF. Insert in panel A, anti-Myc immunoblotting of tagged DNMT3A.

(C) Proliferation of TF-1 cells post-transduction of shRNA vector (control) or DNMT3A-targeted shRNAs under the GM-CSF-depleted culture condition.

(D) Proliferation of TF-1 cells with stable expression of vector, DNMT3A^WT or DNMT3A^R882mut (DNMT3A^R882C/S) upon GM-CSF removal. Insert, anti-Myc immunoblotting of tagged DNMT3A.

(F) A vector LEGO-iG2 (upper) that allows the cre-mediated depletion and thus reversible expression of DNMT3A^R882H as verified by immunoblotting (bottom).

(G) Proliferation of the indicated TF-1 lines upon GM-CSF removal, after depletion of DNMT3A^R882H under the cytokine-rich condition.

DNMT3A^R882mut induces a CpG hypomethylation phenotype in TF-1 cells, resembling what was seen in AML patients with DNMT3A^R882mut.

Human AMLs carrying DNMT3A^R882mut exhibit focal DNA hypomethylation (29), a phenotype also observed in murine leukemias with DNMT3A^R882mut (24). To investigate effect of DNMT3A^R882mut on DNA methylation in TF-1 cells, we performed methylome profiling with Illumina Infinium HumanMethylation450 BeadChip (450K array). Out of a total of 485,512 CpG sites, we identified 6,995 (1.44%) as differentially methylated CpG probes (DMPs) among TF-1 lines with DNMT3A^R882H relative to vector-expressing controls (Fig. 2A and Supplementary Table S1). Among the identified DMPs, almost all (6,958; 99.47%) showed hypomethylation and only 37 (0.53%) displayed hypermethylation (Fig. 2A–B). In contrast, the majority of DMPs in TF-1 cells expressing DNMT3A^WT, relative to control, showed the increased DNA methylation (Supplementary Fig. 2A). Similar patterns of focal CpG hypomethylation were seen with DNMT3A^R882C and DNMT3A^R882S (Supplementary Fig. 2B). To delineate the genomic feature of DNMT3A^R882H-associated DMPs in TF-1 cells, we related the identified DMPs to 15 chromatin modifications of K562 leukemia cells (40) and found that DNMT3A^R882H-associated DMPs with hypomethylation occurred preferentially at gene enhancers, and not promoters or heterochromatin (Fig. 2C). These results suggest enhancer as genomic regions predominantly affected by DNMT3A^R882H in TF-1 cells, which is consistent with previous findings seen in murine leukemias harboring DNMT3A^R882H or Dnmt3a knockout (24,41,42).

Figure 2. DNMT3A^R882mut mainly induces DNA hypo-methylation in TF-1 cells.

(A) Distribution of differentially methylated probes (DMPs), either hypo (left) or hyper-methylated (right), as defined in TF-1 stable cell lines with DNMT3A^R882H, relative to vector controls, cultured in the presence of GM-CSF.

(B) Heatmap shows relative methylation of DMPs defined in panel A among independent TF-1 lines with stable transduction of DNMT3A^R882H (RH; n=5) or empty vector (EV; n=5).

(C) Enrichment of DNMT3A^R882H-induced hypomethylated DMPs at genomic features annotated by the ENCODE K562 ChromHMM (40).

(D) Venn diagram shows significant overlapping between DMPs with the DNMT3A^R882H-induced hypo-methylation in TF-1 cells and those associated with DNMT3A^R882mut in human AMLs (29).

(E) Methylation levels indicated by beta values (y-axis) at representative DNMT3A^R882H-associated DMPs in the indicated TF-1 lines expressing EV, DNMT3A^WT or DNMT3A^R882mut (RH, RC or RS; upper panel) or among human AMLs with DNMT3A^WT or DNMT3A^mut at the R882 or non-R822 residue (bottom panel).

(F) Sanger sequencing showing CpG methylation levels at the indicated site (upper; black arrow) in parental TF-1 cells (left) or those transformed lines before (middle) and after depletion of DNMT3A^R882H (right; see also Fig. 1F–G).

To determine clinical relevance of our above finding, we compared DNMT3A^R882mut-associated DMPs found in TF-1 cells to those based on the TCGA methylome studies of AML patients (3). Here, we found a significant overlap between hypomethylated DMPs identified from TF-1 cells and human AMLs with DNMT3A^R882mut (Fig. 2D; Supplementary Table S1), as exemplified by DMPs at SH3TC2, FOXK2, and GP9 (Fig. 2E). Additionally, principal component analysis (PCA) of DNMT3A^R882H-induced hypomethylated DMPs showed that the methylation pattern of TF-1 cells with DNMT3A^R882H resembled AML patients with DNMT3A^R882mut more than those with DNMT3A^WT (Supplementary Fig. 2C). Moreover, in TF-1 lines with reversible DNMT3A^R882H expression (Figure 1F–G), we found that hypomethylation at the examined region relies on continuous presence of DNMT3A^R882H (Figure 2F; cre+ vs cre-). Together, we show that DNMT3A^R882H-expressing TF-1 cells carry methylation alterations shared by human AMLs, lending support for using TF-1 cells as a model to study DNMT3A^R882mut.

DNMT3A^R882H and IDH mutations cooperate to promote TF-1 cell transformation.

In the clinic, DNMT3A and IDH1/2 mutations frequently co-occur (3) and genetic interaction between DNMT3A^R882H and IDH2^R140Q was verified in mice (42). To query whether TF-1 cells is also suitable for studying synergy between IDH and DNMT3A mutations, we established TF-1 lines with expression of DNMT3A^R882H, IDH1^R132H, or their combination (Fig. 3A). TF-1 cells expressing either mutation showed considerable proliferation post-removal of survival cytokines; however, significantly more growth under cytokine-poor conditions was seen with cells carrying both mutations (Fig. 3B). Methylome analysis revealed that DNMT3A^R882H alone and IDH1^R132H alone predominantly induced hypo- and hypermethylation, respectively (Fig. 3C), with the commonly affected sites accounting for merely ~5–10% (Fig. 3D). This suggests that DNMT3A^R882H and IDH1^R132H target different genomic regions to induce methylation changes. Moreover, a large majority of CpGs affected by either mutation alone were no longer affected in TF-1 cells with dual mutations, while new events of hyper- and hypo-methylation at additional CpG sites were gained by these cells (Fig. 3E). This phenomenon of ‘epigenetic antagonism’ is similar to what was seen in AMLs carrying dual mutations of DNMT3A^R882H and IDH2^R140Q (42). Together, our results support TF-1 cells as a model useful for studying synergistic effect of DNMT3A and IDH1 mutations.

Figure 3. DNMT3A^R882H acts in concert with IDH1^R132H to promote transformation.

(A) Immunoblotting of Myc-tagged DNMT3A and Flag-tagged IDH1 in the indicated TF-1 stable lines.

(B) Proliferation of the indicated TF-1 lines upon GM-CSF removal.

(C) Scatter plot showing the DNMT3A^R882H-associated hypomethylated DMPs (left panel; red) and IDH1^R132H-associated hypermethylated DMPs (right; blue) found in TF-1 cells transduced with either mutation (y-axis), relative to vector controls (x-axis).

(D) Venn diagram using DNMT3A^R882H-associated hypomethylated DMPs and IDH1^R132H-associated hypermethylated DMPs found in TF-1 cells.

(E) Scatter plot showing DNA methylation in TF-1 cells expressing dual DNMT3A^R882H/IDH1^R132H mutations (y-axis), relative to vector controls (x-axis). Red and blue show the DNMT3A^R882H-associated hypomethylated DMPs and IDH1^R132H-associated hypermethylated DMPs as defined in panels C-D, respectively.

Effect of DNMT3A^R882mut on CpG hypomethylation and TF-1 cell transformation is dosage-dependent.

DNMT3A^R882mut is believed to act as a founder mutation during leukemogenesis, and there is an age-dependent increase in incidence of clonal hematopoiesis among individuals carrying DNMT3A^R882mut (8,9). These studies indicate that effect of DNMT3A^R882mut might be time- and dose-dependent, with affected epigenomic alterations accumulating during the course of disease progression. To test this hypothesis, we co-expressed DNMT3A^R882mut together with GFP in an internal ribosome entry site (IRES)-based coexpression system and, based on GFP sorting, we established a set of TF-1 stable lines with increasing levels of DNMT3A^R882H–IRES-GFP expression (Supplementary Fig. 3A). By RT-PCR and immunoblotting, we confirmed serially increased DNMT3A^R882H levels in cells (Fig. 4A–B). Using 450K arrays, we found a positive correlation between DNMT3A^R882H dosage and the total number of DMPs with induced hypomethylation (Fig. 4C), as exemplified by those at SH3TC2 and GP9 (Supplemental Fig. 3B). Importantly, cultivation under cytokine-poor conditions showed the cytokine-independent outgrowth also correlated with the DNMT3A^R882H dosage (Fig. 4D). As control, we also established TF1 lines with increasing expression of DNMT3A^WT (Fig. 4A–B) and found that the DNMT3A^WT dosages had no effect on cytokine-independent growth (Fig. 4D). These results demonstrate a dose-dependent action for DNMT3A^R882mut in inducing CpG hypomethylation and TF-1 cell transformation.

Figure 4. DNMT3A^R882H promotes leukemic cell transformation and DNA hypo-methylation in a dose-dependent manner.

(A-B) RT-qPCR (A) and immunoblot (B) validation of serially increased levels of DNMT3A, either WT (blue) or R882H-mutated (red), after GFP-based sorting of TF-1 cells. Expression values shown were normalized to control. Control: parental TF-1 cells. L, M, H, VH denote low, medium, high, very high, respectively.

(C) Scatter plot showing DNA methylation levels of DMPs in TF-1 lines expressing the serially increased DNMT3A^R882H (y-axis) relative to vector controls (x-axis). DMPs were defined with a beta-value change more than 0.15 and a p-value less than 0.05. Hypo-DMPs are marked in reds, with their numbers labeled at the bottom; hyper-DMPs are marked in grey.

(D) Proliferation of TF-1 cells expressing different levels of DNMT3A upon cytokine removal.

Effect of DNMT3A^R882mut on TF-1 leukemia cell transformation is independent of its catalytic function.

Previously, it was shown that DNMT3A^R882mut is not a null mutant and shows a mildly or partially reduced methyltransferase activity as assayed in vitro (4,12,20,29,30). Moreover, recent studies reported that, relative to DNMT3A^WT, DNMT3A^R882mut exhibits the increased methylation activity in vitro towards certain CpG substrates with a specific flanking sequence, implicating a gain-of-function action (32). However, to what extent gain-of-function effect of DNMT3A^R882mut contributes to transformation remains undefined. First, we found that, in TF-1 cells with DNMT3A^R882mut, the induced hypermethylation only represents a minor event (Fig 2A). This is reminiscent of what was observed in murine DNMT3A^R882H-expressing HSPCs with a more comprehensive enhanced reduced representation bisulfite sequencing (eRRBS) platform where hypomethylation accounted for 80.8% of detected changes (24). Next, we closely examined this published eRRBS dataset and focused on sites with hypermethylation and direct binding of DNMT3A^R882H (24). Here, we did not observe significant preference towards CpG-flanking sequences (Supplementary Fig. 4A–B). Next, to testify whether the enzymatic activity retained in DNMT3A^R882mut is essential for transformation, we introduced an enzymatic-dead mutation, P709V/C710D or E756A (12,15), into DNMT3A^R882H, followed by TF-1 cell transduction (Fig. 5A). Under cytokine-poor culture conditions, neither of enzymatic-dead mutations interfered with the transforming capacity of DNMT3A^R882H (Fig. 5B). These observations support that the remaining catalytic activity harbored within DNMT3A^R882mut is not essential for transformation.

Figure 5. Effect of DNMT3A^R882H on TF-1 cell transformation is independent of its catalytic function but requires its heterodimerization interface.

(A) Immunoblot showing the level of the indicated Myc-tagged DNMT3A mutant post-transduction into TF-1 cells.

(B) Proliferation of TF-1 cells expressing the indicated DNMT3A mutant upon cytokine removal.

(C-F) Schematic representation of different truncation (C) or point mutation (E) of Myc-tagged DNMT3A^R882H, with anti-Myc immunoblots (D and F) showing protein levels of the indicated DNMT3A forms post-transduction into TF-1 cells. EV, empty vector.

(G-H) Proliferation of TF-1 cells expressing the indicated truncation (G) or point mutant form (H) of DNMT3A^R882H upon cytokine removal.

Effect of DNMT3A^R882H on TF-1 leukemia cell transformation relies on its heterodimerization interface and not chromatin-binding motifs.

To dissect the basis underlying DNMT3A^R882mut-mediated transformation, we systematically mutated its functional motifs (Fig. 5C) that were known to mediate chromatin or protein-protein interactions (19). First, we found that, unlike deletion of the NTD or PWWP domains, deletion of ADD severely affected protein stability of DNMT3A^R882H (Fig. 5D). Thus, we also employed point mutations of the motif that did not affect stability, D333A (D329A in mouse Dnmt3a) and D529A/D531A shown to disrupt the H3K36me3- and H3K4me0-binding of PWWP and ADD, respectively (14,15) (Fig. 5E–F). Relative to DNMT3A^R882H, all derived deletion or double mutants affecting the NTD, PWWP or ADD domain showed comparable abilities to sustain cytokine-free proliferation in TF-1 cells (Fig. 5G–H).

DNMT3L is not expressed in AML (29) and DNMT3A may form high-order protein complexes with itself (17,18,20,22,29,32,43) or DNMT3B (31,44,45). DNMT3A^R882mut was proposed to act in a dominant-negative manner by ‘hijacking’ and interfering with functions of wild-type DNMT3A/B proteins via a heterodimerization interface (29,31). However, the causal role for such an effect in leukemic transformation has not been formally tested. To this end, we introduced into DNMT3A^R882H a mutation previously shown to disrupt the heterodimerization-interface-mediated interaction, i.e. F732A or Y735A (Fig. 5E and supplementary Fig. 4C) (18,20). After transduction into TF-1 cells, the DNMT3A^R882H/Y735A double mutant showed comparable protein stability (Fig. 5F) whereas DNMT3A^R882H/F732A was unstable and excluded from subsequent analyses (supplementary Fig. 4D). We found that, relative to DNMT3A^R882H, DNMT3A^R882H/Y735A displayed significantly reduced abilities to sustain cytokine-free growth of TF-1 cells (Fig. 5H, orange).

To gain mechanistic insights, we further assessed interactions of DNMT3A^R882H/F732A with endogenous DNMTs, and observed that, relative to DNMT3A^R882H, DNMT3A^R882H/Y735A is defective in forming efficient interactions with wild-type DNMT3A and DNMT3B in cells (Fig. 6A), which is in agreement with in vitro analysis of mutations affecting this heterodimerization or FF interface (18,20). Next, we examined methylation changes induced by the double mutation forms of DNMT3A^R882H. Consistent with transformation results, deletion or point mutation of PWWP and ADD did not interfere with the ability of DNMT3A^R882H to induce hypomethylation (Fig. 6B). In contrast, much fewer of CpG sites exhibited significant hypomethylation in TF-1 cells post-transduction of DNMT3A^R882H/Y735A (Fig. 6B).

Figure 6. CpG hypomethylation-inducing effect of DNMT3A^R882H requires its heterodimerization interface.

(A) CoIP with anti-Myc antibodies detects interaction of the Myc-tagged DNMT3A^R882H (middle) or DNMT3A^R882H/Y735A (right) with endogenous wildtype DNMT3B or DNMT3A proteins.

(B) Summary of the total number of hypo-methylated DMPs induced by the indicated variant forms of DNMT3A^R882H, relative to vector control (as shown in Figure 2A; with beta value decrease more than 0.1).

(C) RT-qPCR detecting expression of LMO2 in TF-1 lines with stable transduction of empty vector (EV) or the indicated DNMT3A forms at day 12 post-withdrawal of GM-CSF.

DNA methylation is implicated in regulation of numerous cellular processes including transcription, genome stability and mRNA splicing (46,47). Here, we aimed to gain a glimpse of how DNMT3A^R882H affects gene transcription. We looked closely at LMO2, a transcription factor related to leukemogenesis (48) including AML with Dnmt3a mutation (28), for LMO2 shows consistent CpG hypomethylation in TF1 cells with DNMT3A^R882mut and TCGA methylome studies of AML patients (Supplementary Fig. 5 and Supplemental Table 1). By RT-qPCR, we found expression of LMO2 activated in TF-1 cells with DNMT3A^R882H but repressed in those with DNMT3A^WT, relative to control (Fig. 6C, middle); furthermore, the heterodimerization-interface-defective mutation, Y735A, significantly interfered with the LMO2-activating effect of DNMT3A^R882H (Fig. 6C, right).

Together, we show that disrupting heterodimerization interactions severely interferes with the ability of DNMT3A^R882H to induce CpG hypomethylation, LMO2 activation, and cytokine-free growth in TF-1 cells.

Bromodomain inhibition reversed gene-activation events downstream of DNMT3A^R882H-mediated CpG hypomethylation, providing a means for treating AMLs with DNMT3A^R882H.

Our above observations support a causal relationship between transformation and CpG hypomethylation, which is most likely due to combined hypomorphic and dominant-negative effects by DNMT3A^R882H. This further indicates that targeting events downstream of the induced CpG hypomethylation represents a therapeutic strategy. Our prior study of DNMT3A^R882H–dependent murine AML models has revealed a marked increase of histone acetylation at gene-regulatory regions carrying DNMT3A^R882H–induced CpG hypomethylation (24). Histone acetylation ‘readers’ such as BRD4 engage acetylated histones, potentiating gene activation. Using ChIP-Seq performed with DNMT3A^R882H–dependent murine AML cells, we found H3K27ac and BRD4 bound at the DNMT3A^R882H–associated target signature genes that we previously defined (24), which include leukemia-related ‘stemness’ and pro-survival oncogenes such as Mn1, Mycn and Bcl2 (Fig. 7A). Next, we queried whether BRD4 inhibition has therapeutic effect. We first examined sensitivity to I-BET151, a BRD4 inhibitor, using murine AML lines we previously established with dual DNMT3A^R882H and RAS mutations (24). A strong growth inhibition effect of I-BET151 was observed, with IC50 calculated at ~100 nM (Fig. 7B). Importantly, transcriptome analysis showed that BRD4 blockade resulted in down-regulation of DNMT3A^R882H–associated target signature genes (24) (Fig. 7C–D). As well, gene ontology (GO) analysis showed that I-BET151 caused profound changes, including up-regulation of apoptosis-related genes and down-regulation of cell cycle progression genes (Supplementary Fig. 6A). Gene set enrichment analysis (GSEA) further revealed that genes related to MYC targets, stemness, proliferation, DNA replication and cancer-associated signatures were all suppressed by I-BET151 (Fig. 7E–G; Supplementary Fig. 6B).

Figure 7. BRD4 blockade is efficient in treatment of murine AML established by combinational DNMT3A^R882H and NRAS^G12D mutations.

(A) IGV track views of BRD4 and H3K27ac ChIP-seq signals at the indicated gene in murine AML cells established by co-expressed DNMT3A^R882H and NRAS^G12D (termed as DNMT3A^R882H/NRAS^G12D AML lines) (24).

(B) Relative proliferation of the DNMT3A^R882H/NRAS^G12D AML lines post-treatment with I-BET151 relative to mock. Dashed line indicates the half maximal inhibitory concentration (IC50).

(C) GSEA showing that the DNMT3A signature genes were repressed by I-BET151. Mouse_Gene_2.0_ST arrays were performed using the DNMT3A^R882H/NRAS^G12D AML cells treated with either DMSO or 100 nM I-BET151 for 24 hours.

(D) Heatmap showing the most repressed DNMT3A signature genes by BRD4 inhibitors. Rep, replicate.

(E-G) GSEA shows downregulation of the indicated genesets related to MYC targets (E), stem cell (F) or pediatric cancer markers (G) post-treatment with I-BET151.

(H) Survival of mice transplanted with the DNMT3A^R882H/NRAS^G12D AML cells and then treated with either vehicle (n = 8) or I-BET151 (n = 6). Significance was tested by log rank test.

(I-J) White blood cell counts (I), representative images of spleens (J, top; scale bar, 1 cm) and Wright-Giemsa staining of bone marrow cells (J, bottom; scale bar, 10 μm) from mice transplanted with the DNMT3A^R882H/NRAS^G12D AML cells and treated with either vehicle or I-BET151 for 10 days.

(K-L) Kaplan-meier survival curve (K) and representative bioluminescent images (L, at day 20) of mice transplanted with DNMT3A^R882H/NRAS^G12D AML cells and treated with either vehicle, trametinib or I-BET151 alone, or their combination. Significance was tested by log rank test.

Next, we evaluated in vivo therapeutic effect of I-BET151 on murine AMLs induced by dual DNMT3A^R882H and RAS^G12D mutations. Compared to mock, I-BET151 treatment significantly prolonged survival of leukemic mice (Fig. 7H). Compared to mock, I-BET151 treatment also significantly delayed development of AML phenotypes such as splenomegaly, elevated counts of white blood cells and reduce counts of red blood cells (Fig. 7I–J, Supplementary Fig. 4C). Following I-BET151 treatment, less AML blasts were also observed in the bone marrow (Fig. 7J, bottom). Given that I-BET151 inhibits the DNMT3A^R882H-related gene-expression programs, we next ask whether inhibition of RAS signaling has additional therapeutic effect in this model. To this end, we used trametinib, an FDA-approved inhibitor of mitogen-activated protein kinase kinase (MEK) (49). Indeed, while dosing with trametinib or I-BET151 alone considerably delayed AML progression as measured by animal survival and bioluminescence imaging, their combinational treatment had more significant AML-inhibitory effect in vivo (Figure 7K–L).

Collectively, bromodomain inhibition not only represses gene-expression programs related to DNMT3A^R882H-induced hypomethylation but, importantly, suppresses development of the DNMT3A^R882H-related murine AMLs in vivo, either alone or in combination with MEK inhibitors.

Discussion

DNMT3A^mut is prevalent in patients with hematological malignancies and disorders. It is proposed that DNMT3A^mut acts as a founder mutation shaping the course of leukemia evolution and progression. The literature has documented that DNMT3A^R882mut has hypomorphic, dominant-negative and/or gain-of-function effects under different biological contexts. However, it remains undefined by what effects DNMT3A^R882mut mediates leukemogenesis. We started the investigations first by reporting TF1 cells as a robust system useful for studying DNMT3A^mut-related leukemic phenotypes, which then allowed us to next examine activities underlying DNMT3A^R882mut-induced transformation. Based on the mechanistic studies, we further show that BRD4 inhibition reverses gene-expression programs downstream of disease-relevant effect of DNMT3A^R882mut, thus suggesting a potential means of therapeutics.

Compared to the previously described in vivo and primary cell systems for studying DNMT3A^R882mut (24–28), TF-1 cells is straightforward and suitable for cause-effect relationship studies to delineate the role for multifaceted effects of DNMT3A^R882mut during leukemic transformation. Via a systematic interrogation of various motifs in DNMT3A^R882mut with this system, we show that the transformation-promoting function of DNMT3A^R882mut is independent of its residual catalytic activity (as assayed by double mutants carrying a enzyme-dead mutation), suggesting that CpG methylation gains at certain specific sequence contexts, as proposed in the gain-of-function model (32), is unlikely to contribute to leukemogenesis. In contrast, mutating a heterodimerization interface not only interfered with physical associations of DNMT3A^R882mut to endogenous wildtype DNMT3A/B proteins but also largely abolished the ability of DNMT3A^R882mut in inducing CpG hypomethylation and TF-1 cell transformation. These findings support a causal relationship between the dominant-negative effect of DNMT3A^R882mut and transformation. Furthermore, purified DNMT3A^R882mut enzymes were known to be hypomorphic (4,12,20,29,30); recent structural studies further showed that R882 forms interactions with both DNA substrates and a so-called ‘Target Recognition Domain’ loop, which indicated its role in substrate binding and/or enzymatic functions (12). Our results thus support that combined hypomorphic and dominant-negative effects of DNMT3A^R882mut lead to CpG methylation decrease/loss, which is most likely to contribute to leukemogenesis. In agreement, CpG hypomethylation and not hypermethylation was detected as a major event in TF-1 cells and murine AML models carrying DNMT3A^R882mut (12,24). Despite advances, exact molecular mechanism by DNMT3A^R882mut induces CpG hypomethylation in cells remains to be fully understood. As well, since blood cells do not express DNMT3L (29), investigation of leukemia-related DNMT3A^mut requires using physiologically relevant contexts such as homo-tetramers or oligomers of DNMT3A (18,21) or complexes of DNMT3A with DNMT3B (31,44,45) or other putative partners.

Additionally, TF-1 cells act as a system useful for modeling various other biological effects of DNMT3A^mut. First, CpG hypomethylations induced by DNMT3A^R882mut in TF-1 cells resemble those seen in leukemia patients with DNMT3A^R882mut, suggesting a clinical relevance of findings from this model. Furthermore, there is a dose-dependent effect by DNMT3A^R882mut on CpG hypomethylation and transformation (cytokine-free proliferation) in TF-1 cells, which provides an explanation for age-related increase in incidence of clonal hematopoiesis seen among elderly individuals. We observed synergy between DNMT3A^R882mut and IDH1 mutation in TF-1 cells, as well as a phenomenon of ‘epigenetic antagonism’ that resembles what was seen in human AML cells carrying dual DNMT3A^R882H and IDH2 mutations (42). Future investigation, however, is warranted to delineate interplays between DNMT3A^R882mut and coexisting IDH mutations for inducing epigenomic alterations in both DNA methylation (including 5mC and 5hmC) and histone modifications. Lastly, missense mutations at non-R882 residues account for ~40–50% of DNMT3A^mut in blood cancer and disorders. Considering many mutant forms for such a non-R882 category of DNMT3A^mut, TF-1 cells shall provide a simple yet robust system for studying their biological effects as shown recently with several DNMT3A^mut known to affect DNMT3A’s DNA-binding (12). Together, TF-1 cells represent a straightforward, robust and disease-relevant model system for investigating numerous unsolved issues relating DNMT3A^mut.

This study not only demonstrated a requirement for the dominant-negative effect of DNMT3A^R882mut (most likely to act in concert with a hypomorphic nature of DNMT3A^R882mut) during leukemic transformation but also explored potential therapeutics. For the latter, our mechanistic studies suggest that intervention shall be developed for targeting events associated with DNMT3A^R882mut-induced CpG hypomethylation. Here, we first used a transgene-reversible system to show that DNMT3A^R882mut-induced TF-1 cell transformation relies on its continuous expression; as well, we find CpG hypomethylation correlated to transcriptional activation of LMO2, a leukemia-related oncogene, thereby providing a glimpse of how induced CpG hypomethylation influences gene expression in TF-1 cells. It is consistent to previous reports showing that sites with DNMT3A^R882mut-induced CpG hypomethylation are enriched in cis-regulatory elements such as enhancers and display increased histone acetylation, which then recruits ‘reader’ protein complexes to potentiate gene activation (24,50). Besides DOT1L complexes (24,50), we here show that BRD4, another histone acetylation ‘reader’, also binds to DNMT3A^R882H-associated gene targets (notably Mn1, Mycn and Bcl2) in murine AMLs and that pharmacological inhibition of BRD4 suppressed activation of DNMT3A^R882H-related gene signatures. BRD4 inhibitors efficiently suppressed in vitro and in vivo growth of murine AMLs carrying dual DNMT3A^R882mut and RAS mutations. Note that DNA methylation potentially regulates numerous biological processes including gene transcription, genome stability, chromatin architecture and looping, mRNA splicing, and silencing of repetitive DNA elements (46,47). Determination of putative effects of DNMT3A^R882H-related CpG hypomethylation on these crucial processes warrants future investigation.

Statement of Significance.

Findings highlight a model system to study the functional impact of a hotspot mutation DNMT3A R882 in leukemia