The R882H substitution in the human de novo DNA methyltransferase DNMT3A disrupts allosteric regulation by the tumor supressor p53

A myriad of protein partners modulate the activity of the human DNA methyltransferase 3A (DNMT3A), whose interactions with these other proteins are frequently altered during oncogenesis. We show here that the tumor suppressor p53 decreases DNMT3A activity by forming a heterotetramer complex with DNMT3A. Mutational and modeling experiments suggested that p53 interacts with the same region in DNMT3A as does the structurally characterized DNMT3L. We observed that the p53-mediated repression of DNMT3A activity is blocked by amino acid substitutions within this interface, but surprisingly, also by a distal DNMT3A residue, R882H. DNMT3A R882H occurs frequently in various cancers, including acute myeloid leukemia, and our results suggest that the effects of R882H and other DNMT3A mutations may go beyond changes in DNMT3A methylation activity. To further understand the dynamics of how protein-protein interactions modulate DNMT3A activity, we determined that p53 has a greater affinity for DNMT3A than for DNMT3L and that p53 readily displaces DNMT3L from the DNMT3A:DNMT3L heterotetramer. Interestingly, this occurred even when the preformed DNMT3A:DNMT3L complex was actively methylating DNA. The frequently identified p53 substitutions (R248W and R273H), whereas able to regulate DNMT3A function when forming the DNMT3A:p53 heterotetramer, no longer displaced DNMT3L from the DNMT3A:DNMT3L heterotetramer. The results of our work highlight the complex interplay between DNMT3A, p53, and DNMT3L and how these interactions are further modulated by clinically derived mutations in each of the interacting partners.

Transcriptional regulation, genomic imprinting, and cellular differentiation, including 5-methylcytosine patterning on DNA (1-7), relies on diverse protein-protein and protein-nucleic acid interactions. In the crowded intracellular environment, formation of biologically significant complexes relies on the kinetic accessibility to specific protein surfaces and the thermodynamic stability of the resultant complexes (8 -11). For exam-ple, de novo DNA methylation by the DNA methyltransferase 3A (DNMT3A) involves the formation of complexes that include a wide-range of regulatory partners, such as histones, histone-modifying enzymes, transcription factors, and RNA (2,4,12,13). Such interactions are frequently altered during oncogenesis, resulting in the disruption to DNMT3A genomic localization and/or regulation of enzyme activity (14). The tumor suppressor p53 is well-known to interact with components of the epigenetic machinery, including DNMT3A; however, a functional understanding of p53-DNMT3A interactions remains largely unknown (15,16).
In addition to directly activating transcription of genes essential for cell cycle arrest and apoptosis in response to genotoxic stress, the interactions between p53 and histone modifying enzymes are a key driver of gene activation (17,18). The progressive accumulation of p53 mutations leads to the recruitment of histone modifying enzymes (19 -21). Several studies suggest a link between p53 and DNA methylation. For example, whereas DNMT1 2 (the maintenance DNA methyltransferase) represses expression of the p53 gene (22), p53 binding to DNMT1 stimulates DNMT1-mediated methylation (23). In addition, p53 represses the expression of DNMT3A and DNMT3B, whereas DNMT3A has been shown to repress the transcriptional activity of p53 (24,25). However, definitive evidence of the functional consequences of a DNMT3A:p53 complex on DNMT3A activity remains elusive. Our interest lies in exploring whether p53 binding to DNMT3A alters DNMT3A activity, and if well-studied DNMT3A and p53 mutants, such as R882H (DNMT3A) and p53 mutants R248W and R273H, alter this regulation.
Located at the dimer interface, the major DNA-binding site, the R882H substitution in DNMT3A disrupts tetramerization and processive catalysis, both of which can be restored by DNA-methyltransferase 3-like (DNMT3L) (26 -28). Additionally, R882H displays altered interactions with components of the Polycomb repressive complex 1 compared with WT DNMT3A, thereby leading to transcriptional silencing in a DNA methylation-independent manner (29). These observations suggest that compared with WT DNMT3A, the R882H substitution may lead to altered binding and/or regulation by partner proteins. Given that the p53 gene is a recurring target for mutations in a wide-range of human cancers, there is growing interest in understanding how mutations in p53 contribute to disease onset and progression (30,31). In addition to a high mutation frequency, p53 R273H and R248W form aberrant protein complexes that affect the activity of interacting partner proteins (32)(33)(34). Our goal is to understand the dynamics and functional consequences of complex assembly involving the WT catalytic domain of DNMT3A (DNMT3A WT ) and the R882H substitution (DNMT3A R882H ) under a variety of conditions. Furthermore, we seek to better understand the functional consequences of protein complexes involving two or more proteins to better understand the cellular basis of enzyme function.

WT (p53 WT ) and mutant p53 inhibit the DNA methylation activity of full-length and catalytic domain DNMT3A WT
Previous cell-based evidence implicates direct and indirect DNMT3A and p53 interactions (24,25). In mouse embryonic stem cells, p53 indirectly regulates DNMT3A-mediated methylation by restricting the expression of DNMT3A (24). Alternatively, direct binding of DNMT3A to p53 suppresses p53-mediated transcription of p21 in a DNA methylation-independent manner, implying that DNMT3A may allosterically regulate p53 activity (25). Based on this evidence, we sought to determine whether p53 has any effect on the DNA methylation activity of DNMT3A. Given that the DNMT3A catalytic domain and full-length enzyme have comparable kinetic parameters (k cat , K m DNA , K m AdoMet , processivity, and DNMT3L stimulation), the catalytic domain of DNMT3A has proven to be a suitable model for in vitro studies and is commonly employed (35,36). However, the N-terminal domains in full-length DNMT3A, including the ATX-DNMT3A-DNMT3L (ADD) and PWWP domains, are known to interact with numerous partner proteins that may modulate the enzymatic activities of DNMT3A (37,38). Therefore, we compared the effect of p53 WT on the methylation activity of the DNMT3A catalytic domain and fulllength enzymes by preincubating p53 with equimolar concentrations of DNMT3A for 1 h prior to initiating the reaction by the addition of poly(dI-dC). We observed comparable levels of p53 WT -mediated DNMT3A inhibition with the catalytic domain ( Fig. 1, A and B) or the full-length DNMT3A (Fig. 1, A and C). The comparable inhibition indicates that the N-terminal portion of DNMT3A is not essential for p53 WT -DNMT3A interactions and does not perturb p53 WT acting on the catalytic domain of DNMT3A. The consensus DNA binding sequence of p53 (5Ј-RRRC(A/T)(A/T)GYYY(0 -14 bases)RRRC(A/T)(A/ T)GYYY-3Ј; Aϭ adenine, Gϭ guanine, Cϭ cytosine and Tϭ thymine) differs from that of DNMT3A, which displays a strong preference toward CpG sites (39). However, it is possible that p53 WT inhibition of DNMT3A (full-length and catalytic domain) enzymatic activity is attributable to the DNA-binding ability of p53. To assess whether the inhibitory effect of p53 WT Figure 1. p53 WT -dependent inhibition of DNA methylation is specific to DNMT3A. A, fold-inhibition calculated by product formed by WT DNMT3A (full-length or catalytic domain enzymes) or M.HhaI divided by product formed by DNMT3A (full-length or catalytic domain enzymes) or M.HhaI without p53 WT from reactions in B-D. Co-incubation of DNMT3A full-length (A, blue square calculated from B) and catalytic domain (A, orange square calculated from C) enzymes with p53 WT (1:1 at 150 nM) leads to comparable levels of inhibition. Similar reactions involving the bacterial methyltransferase M.HhaI (A, green square calculated from D) with excess p53 WT (1:3) failed to inhibit DNA methylation. In all co-incubations, proteins were held at 37°C for 1 h prior to the start of the reaction by the addition of DNA (5 M bp poly(dI-dC)). Data reflect the mean Ϯ S.D. of 3 experiments; one-way analysis of variance was used to compare values of all three reactions; ***, p Ͻ 0.001; ns, p Ͼ 0.05. on DNA methylation is specific to DNMT3A, the activity of the bacterial methyltransferase M.HhaI, which recognizes 5Ј-GCGC-3Ј sites in dsDNA, was challenged with increasing concentrations of p53 WT . Although p53 WT led to roughly a 50% decrease in DNMT3A (full-length and catalytic domain) activity ( Fig. 1), DNA methylation by M.HhaI was unaltered in similar reactions involving p53 at 1:1 and 3:1 relative to M.HhaI (Fig. 1, A and D). Like DNMT3A, M.HhaI is a C-5 cytosinespecific methyltransferase that possesses a remarkably similar structure to that of the DNMT3A catalytic domain (40). Despite the shared similarities of DNMT3A and M.HhaI, here we show that p53 inhibition is specific to DNMT3A.
To further characterize the interactions between DNMT3A WT and p53, the apparent binding affinities (K D,app ) of p53 WT and p53 R248W for DNMT3A WT were determined. For comparison, we also determined the K D,app of DNMT3L for DNMT3A WT , which has a well-characterized interaction. K D,app was assessed by measuring the activity of DNMT3A WT with increasing levels of DNMT3L ( Fig. S1A) or p53 (Fig. S1B), and subsequently determining the fold-stimulation or inhibition by DNMT3L or p53 WT and p53 R248W , respectively. Although DNMT3L resulted in a K D,app of 80 Ϯ 12 nM, p53 WT displayed a nearly 5-fold stronger binding affinity for DNMT3A WT with a K D,app of 17 Ϯ 3 nM, followed by p53 R248W , which resulted in a K D,app of 41 Ϯ 6 nM (calculated from supporting Fig. 1, A and B). Thus, compared with the DNMT3A WT -DNMT3L complex, DNMT3A WT -p53 WT and -p53 R248W complexes are more energetically favorable.

Mutational mapping suggests p53 WT interacts with the tetramer interface DNMT3A
Previous work by Wang et al. (25) suggests that DNMT3A interacts with the C-terminal tetramerization domain of p53 (amino acids 319 -393, Fig. S3). No such information is available for the region on DNMT3A that is associated. We previously used alanine scanning to identify residues on the DNMT3A tetramer interface that largely contribute to tetramer stability and alter the ability of DNMT3A to form higher order complexes (41). In a similar manner, we employed docking-based modeling of protein-protein interfaces using monomeric forms of DNMT3A (PDB code 5YX2; residues 628 -914) and p53 (PDB code 3TS8; 94 -356) to predict a surface on DNMT3A for interactions with p53. Computational models generated in ZDOCK and RosettaDock servers were used to predict the DNMT3A tetramer interface as a likely surface for DNMT3A-p53 interactions (Fig. S2) (42,43). Based on this rationale, p53 WT inhibition of the DNMT3A catalytic activity was assessed in a subset of alanine mutations on the DNMT3A tetramer interface (R729A, E733A, R736A, R771A), which are also commonly observed in acute myeloid leukemia (AML) (26) (Fig. 2A). The extent of p53 WT inhibition varied across the mutations examined in this study: the fold-inhibition of DNMT3A R771A (Fig. 2, B and C, green square) and DNMT3A E733A (Fig. 2, B and C, blue square) was greater than WT, whereas the fold-inhibition of DNMT3A R729A (Fig. 2, B and C, orange square) was less than WT and DNMT3A R736A (Fig. 2, B and C, red square) displayed no inhibition. The results obtained implicate the DNMT3A tetramer interface as a potential surface on DNMT3A for interactions with p53 and suggest that residue Arg-736 on the DNMT3A tetramer interface contributes to the necessary contacts for p53 inhibition of DNMT3A.

The regulation of DNMT3A WT by p53 WT is dominant to that of DNMT3L
Given that DNMT3A exists in several multiprotein complexes associated with transcriptional regulation, we sought to assess the extent of DNMT3A modulation in the presence of multiple regulatory partner proteins (5,44,45). Our mutational mapping and modeling results suggest that the DNMT3A tetramer interface (Fig. 2 and Fig. S2) interacts with both DNMT3L and p53. Given that the DNMT3A-DNMT3L cocrystal structure (46) presents the DNMT3A tetramer interface as an established surface for regulation of DNMT3A activity by an additional partner protein, we assessed whether DNMT3L and p53 WT regulation of DNMT3A WT is mutually exclusive using poly(dI-dC) as a substrate. After demonstrating DNMT3A WT is responsive to DNMT3L (Fig. 3A, green square) and p53 WT (Fig. 3A, blue square), we observed that p53 WT modulation of DNMT3A WT is dominant over that of DNMT3L in DNMT3A WT -DNMT3L-p53 WT co-incubations (Fig. 3A, red square). We previously showed that p53 WT inhibition of DNMT3A WT does not rely on DNA binding by p53 (Fig. 1). Therefore, the use of poly(dI-dC) as a substrate allowed us to investigate the isolated effects of p53-mediated inhibition of DNMT3A activity. Inspired by our previous studies using human promoters, we also studied the Cyclindependent Kinase Inhibitor 1A/P21 promoter, which is a common target for DNMT3A and p53 (25,102). As we observed for poly(dI-dC), p53 WT -dependent inhibition of DNMT3A WT activity is dominant over DNMT3L stimulation as DNMT3A WT -p53 WT -DNMT3L co-incubations ( Fig.  3B, yellow square) displayed comparable levels of activity as reactions consisting of DNMT3A WT -p53 WT (Fig. 3B, red square) with the Cyclin-dependent Kinase Inhibitor 1A/P21 promoter as a DNA substrate. Furthermore, equimolar concentrations of all proteins were used in DNMT3A WT -DNMT3L-p53 WT co-incubations (Fig. 3, A and B), suggesting that the dominant modulation of DNMT3A WT activity by p53 WT over DNMT3L was not due to stoichiometric differences. The stability of the DNMT3A WT :p53 WT complex is greater than the DNMT3A WT :DNMT3L complex (Fig. S1) and DNMT3L and p53 likely share a binding interface on DNMT3A (Fig. 2). To further investigate the dynamics of DNMT3A WT -DNMT3L-p53 WT interactions, we next assessed the effect of adding regulatory proteins to actively catalyzing heterotetrameric complexes, which is arguably a better representation of what occurs within the cell. We observed that addition p53 WT disrupts the DNMT3A WT :DNMT3L complex during catalysis (Fig. 3C, red square). This not only supports our observation that the DNMT3A WT :p53 WT complex is more stable, but that it can access the DNMT3A WT :DNMT3L complex and displace DNMT3L during catalysis. In contrast, adding DNMT3L to an actively methylating DNMT3A WT :p53 WT complex failed to disrupt the modulatory effect of p53 WT (Fig. 3E, red square). In fact, comparable levels of DNMT3A WT -dependent methylation activity were observed in reactions consisting of DNMT3A WT -p53 WT (Fig. 3E, blue square)

DNMT3A R882H disrupts allosteric regulation by p53
and functional DNMT3A WT :p53 WT complexes to which DNMT3L was added (Fig. 3E, red square). In sum, our results consistently suggest the regulatory effect of p53 WT on DNMT3A WT is dominant compared with DNTM3L regulation and provides insights into the dynamics of p53 WT and DNMT3L binding on DNMT3A WT .

p53 WT fails to inhibit the methylation activity of DNMT3A R882H
Identified as the most common mutation in DNMT3A in AML patients, in vitro evidence suggests the R882H substitution leads to altered binding and/or regulation by partner proteins (26). Although R882H is unable to form homotetramers, the addition of DNMT3L leads to the formation of heterotetramers and restores processive catalysis (28). In addition, immunoprecipitation experiments using HEK293T cells reveal R882H displays increased binding to components of the Polycomb repressive complex 1 compared with WT DNMT3A (29). These observations suggest that whereas R882H is located at the dimer interface ( Fig. 2A), which is distal from the tetramer interface and proximal to the DNA interface ( Fig. 2A), R882H appears to allosterically affect the ability of DNMT3A to interact with partner proteins. Although DNMT3A R882H is responsive to DNMT3L activation (Fig. 3A, green square), it is unresponsive to the modulatory effect of p53 WT in DNMT3A R882H -p53 WT (Fig. 3A, blue square) and mixed DNMT3A R882H -p53 WT -DNMT3L co-incubations ( Fig. 3A, red square). In fact, DNMT3A R882H -p53 WT (Fig. 3A, blue square) coincubations led to comparable levels of product formed as DNMT3A R882H only (Fig. 4A, f) and mixed DNMT3A R882H -p53 WT -DNMT3L co-incubations ( Fig. 3A, red square) reflected similar levels of activity as DNMT3A R882H -DNMT3L co-incubations (Fig. 3A, green square). Thus, the R882H substitution appears to disrupt the modulatory effect of p53 WT on DNA methylation. To additionally challenge this notion, we evaluated the effect of adding DNMT3L or p53 WT to actively methylating DNMT3A R882H -p53 WT or DNMT3A R882H -DNMT3L co-incubations. Consistent with the hypothesis that p53 WT fails to modulate DNMT3A R882H activity, the addition of p53 WT failed to disrupt functional DNMT3A R882H :DNMT3L heterotetramers (Fig. 3D, red square) unlike that observed in similar reactions involving DNMT3A WT (Fig. 3C, red square). However, the addition of DNMT3L failed to stimulate DNMT3A R882H in actively catalyzing DNMT3A R882H -p53 WT co-incubations (Fig. 4F, red square).

p53 R248W and p53 R273H display altered regulation of DNMT3A WT in the presence of DNMT3L
Mutated in over half of all human cancers, the majority of p53 mutations are missense mutations throughout the p53 DNA- Figure 2. DNMT3A WT tetramer interface mutants show highly variable response to p53 WT inhibition. Crystal structure of a DNMT3A WT -DNMT3L complex (adapted from PDB code 5YX2) denoting critical residues for DNMT3A WT oligomerization (A) (36,50). Although the extent of p53 WT inhibition varies across DNMT3A WT mutants harboring a single alanine substitution within the tetramer interface, the DNMT3A R736A was unresponsive to p53 WT inhibition (B and C). All reactions consisted of 150 nM DNMT3A WT and were initiated by the addition of 5 M bp poly(dI-dC). For co-incubations, DNMT3A WT and p53 WT (1:1) were preincubated for 1 h at 37°C prior to starting the reaction by the addition of substrate DNA. Fold-inhibition was calculated by product formed by DNMT3A (WT and mutants) divided by product formed by DNMT3A (WT and mutants) without p53 WT . All reactions were performed in duplicates. In B, a Student's unpaired t test was used to compare values within each set of reactions; **, p Ͻ 0.01; ns, p Ͼ 0.05. For C, a one-way analysis of variance was used to compare the values of each mutant to WT (****, p Ͻ 0.001) and across all samples (orange ****, p Ͻ 0.001). Data reflect the mean Ϯ S.D. of 3 experiments.

DNMT3A R882H disrupts allosteric regulation by p53
addition of DNMT3L to functional DNMT3A WT -p53 R248W (Fig. 4D, yellow square) or DNMT3A WT -p53 R273H (Fig. 4E, teal square) assembles led to an increase in DNMT3A WT activity. These findings support the notion that the regulatory effect of DNMT3L on DNMT3A WT activity is dominant over that of p53 R248W or p53 R273H . Here we show that whereas p53 WT , p53 R248W , and p53 R273H display comparable levels of DNMT3A WT inhibition, p53 R248W and p53 R273H displayed altered regulation of DNMT3A WT in the presence of DNMT3L compared with p53 WT . This presents an example in which p53 R248W and p53 R273H display altered protein-protein interactions compared with p53 WT in the context of DNA methylation and in addition to those previously reported (32)(33)(34).

p53 binds DNMT3A WT and DNMT3A R882H to form of heterotetramers
The ability of partner proteins to impact DNMT3A function depends on the formation of a complex, although this may not be sufficient. Anisotropy measurements are widely employed to assess protein-DNA, protein-protein interactions, and estimate the oligomeric state of proteins (51)(52)(53)(54)(55)(56)(57)(58). We previously used a 30-bp 5Ј 6-FAM-labeled duplex DNA (GCbox30), which contains a single recognition site for DNMT3A, to resolve the oligomeric state of DNMT3A (28,41). We relied on this approach and fluorescence anisotropy to assess the dynamics of DNMT3A WT -p53 WT interactions on DNA (GCbox30). Increasing concentrations of p53 WT or DNMT3L (Fig. 5A, purple square) to a fixed concentration of DNA-bound DNMT3A R882H (Fig. 5A, ) resulted in a corresponding increase to the initial anisotropy value, thereby suggesting the formation of higher order complexes on DNA. Under identical conditions, we observed that increasing concentrations of DNMT3L (Fig. 5A, green circle), p53 WT (Fig. 5A, red triangle), or p53 R248W (Fig. 5A, blue diamond) did not result in a detectable change to the initial anisotropy value of DNMT3A WT . DNMT3A R882H binds DNA as a homodimer and forms heterotetramers with DNMT3L on DNA (28, 36). The titration of p53 WT

DNMT3A R882H disrupts allosteric regulation by p53
to DNMT3A R882H led to similar final anisotropy values as those observed for DNMT3A WT in the absence and presence of DNMT3L, p53 WT , and p53 R248W (ϳ0.19; Fig. 5A). To confirm whether the increased anisotropy observed with DNMT3A R882H (Fig. 5A, ) is due to formation of a higher order structure (likely a tetramer), we performed gel shift assays of DNMT3A R882H with varying concentrations of p53 WT using GCbox30 as a substrate. Consistent with the results of the fluorescence anisotropy assays, an increase in the concentration of p53 WT to a fixed concentration of DNMT3A R882H (Fig. 5, B and C, lanes 3-5) led to a supershift and disappearance of the band corresponding to DNA-bound DNMT3A R882H (Fig. 5B, lane 1). In sum, our results indicate that p53 interacts with DNMT3A to form heterotetramers and that inhibition does not arise from disrupting the ability of DNMT3A to bind DNA.

Discussion
Although p53 has been extensively investigated, much less is known about whether or how this protein influences epigenetic pathways, particularly DNA methylation (17,18,30,(47)(48)(49)(50). Reports on the cross-talk between p53 and members of the DNMT family include the loss of global methylation by 5-aza-2Ј-deoxycytidine treatment induces a p53 DNA damage response pathway (59), deletion of the DNMT1 gene activates p53-mediated apoptosis (60) and p53 directly stimulates DNMT1-mediated methylation (23). In the context of the de novo DNA methyltransferases, p53 transcriptionally represses DNMT3A and DNMT3B, whereas DNMT3A inhibits p53-mediated transcription (24). Based on this evidence and the association of DNMT3A and p53 in various human cancers (26,(47)(48)(49)(50), we sought to characterize the dynamics and regulation of DNMT3A activity by p53 under various conditions. We show that p53 and the well-characterized DNMT3L bind to the same region on DNMT3A, resulting in roughly a 2-fold inhibition of DNMT3A activity. The DNMT3A-p53 interaction is modulated by well-known mutations in DNMT3A and p53. Our results provide insights into the complexity of mutation-specific variation in the regulation of protein function and elucidates a molecular basis for the distinguishing DNA methylation phenotypes associated with the R882H substitution in DNMT3A (61,62).
The interaction interface between proteins relies on welldefined single residue interactions within flat surfaces (63,64). Factors like kinetic accessibility to specific protein surfaces, thermodynamic stability of the resultant complexes, along with structural properties of protein complexes, contribute to the formation of biologically significant assembles (8 -11, 65). Complexes with partner proteins regulate DNMT3A activity (12,13,37,38,46), thereby contributing to normal and aberrant tissue-specific methylation patterns (2)(3)(4)(5). Given that DNMT3A binds the C-terminal tetramerization domain of p53 (residues 319 -393, Fig. S3) (25), we sought to identify the surface on DNMT3A that binds p53 and whether p53 directly impacts DNMT3A activity. Like DNMT3L (36), our results  1 and 5) show increasing concentrations of p53 WT (lanes 3-5) to a constant concentration of DNMT3A R882H leads to disappearance of the DNMT3A R882H band and formation of a higher order structure (see arrows, lanes [3][4][5]. In A, DNMT3A WT and DNMT3A R882H concentrations were 2.5 M. Single point anisotropy measurements were taken after increasing concentrations of DNMT3L (green circle), p53 WT (red triangle), and p53 R248W (blue diamond) were added to DNA bound (250 nM 5Ј 6-FAM-labeled GCbox30) DNMT3A WT and DNMT3A R882H and allowed to incubate at room temperature for 5 min. Measurements were taken using a fluorometer equipped with polarizing filters (excitation, 485 nm; emission, 520 nm). In B, gel shift assays were carried out as described in Holz-Schietinger et al. (28) and other than samples were run on native 4.5% polyacrylamide gels and binding reactions were performed at 37°C (lane 1). For p53 WT supershifting, varying concentrations of p53 WT were preincubated for 30 min at 37°C with DNMT3A R882H before the addition of DNA. A and C reflect the results (mean Ϯ S.D.) of 2 independent experiments.

DNMT3A R882H disrupts allosteric regulation by p53
suggest that surfaces on the catalytic domain of DNMT3A are sufficient for p53-mediated inhibition of DNMT3A (Fig. 1). Using docking-based computational models of DNMT3A (PDB 5YX2; residues 628 -914) and p53 (PDB code 3TS8; 94 -356) monomers, we identified the tetramer interface on DNMT3A as a likely surface for DNMT3A-p53 interactions (Fig. S2). We challenged this finding by determining if mutations at this interface ( Fig. 2A) (41) interfered with p53 interactions (Fig. 2, B and C). Our results suggest an overlap between DNMT3L and p53 for binding and allosteric regulation of DNMT3A activity (Tetramer interface) (Fig. 2 and Fig. S2) through the formation of heterotetramers with DNMT3A WT (Fig. 5). Our proposed complex for DNMT3A WT :p53 WT is consistent with previously resolved p53 co-crystal structures (PDB code 1KZY and 5ECG), which consist of p53 dimers bound to interacting partner proteins (66,67). The structure and functional studies on DNMT3A interactions with another protein, DNMT3L, provides a reliable "metric" to investigate a common surface on DNMT3A that facilitates allosteric regulation of enzymatic activity. We show that DNMT3A WT :p53 WT complexes (K D,app of 17 Ϯ 3 nM) are more thermodynamically stable than DNMT3A WT :DNMT3L complexes (K D,app of 80 Ϯ 12 nM). Consistent with these relative stabilities, DNMT3A WT : p53 WT complexes appear to be more kinetically stable when all three proteins are combined (DNMT3A, DNMT3L, and p53). Moreover, p53 WT can displace DNMT3L from the DNMT3A: DNMT3L heterotetrameric complex under catalytic conditions, which is arguably more relevant (Fig. 3). We propose that p53 WT replaces the outer pair of DNMT3A WT monomers (Fig.  6A, III) or DNMT3L (Fig. 6A, V) monomers to allosterically inhibit the enzymatic activity of DNMT3A WT .
Due to the energetic contributions of specific residues to protein-protein interactions (63,64,68,69), it is not surprising that mutations that alter protein complex formation have been linked to various human disorders and are over-represented among disease-causing mutations (70,71). The R882H substitution in DNMT3A is the most common recurrent mutation in AML patients (26) and DNMT3A R882H may be disruptive to protein-protein interactions (29,72,73). Although DNMT3A R882H is mildly impacted in function (28), it seems reasonable that altered interactions with partner proteins contribute to the aberrant methylation patterns observed in AML (61,62). Previous work from our lab has shown that although R882H is a functional dimer on DNA, the addition of DNMT3L restores the formation of heterotetramers, and near-normal levels of catalysis (Fig. 6A, II) (28). The comparable increase in anisotropy observed by the addition of DNMT3L or p53 to DNA-bound DNMT3A R882H (Fig. 5) supports the notion that p53 WT binds DNMT3A R882H (Fig. 5) to form heterotetramers but is unable to allosterically inhibit DNMT3A R882H activity (Figs. 3 and 6A, IV). Furthermore, p53 WT fails to displace DNMT3L monomers in DNMT3A R882H :DNMT3L heterotetramers (Fig. 3) (Fig. 6A, V). Like DNMT3A R882H , certain mutations in p53 are disruptive to protein-protein interactions and alter regulation of partner proteins relative to WT p53 (32-34, 74, 75). In addition, several observations suggest an interplay between components of the epigenetic machinery and mutations in p53 (76,77). Although located outside of the DNMT3A WT -p53 WT interface (residues 319 -393, Fig. S3) (25), we investigated how p53 R248W and p53 R273H mutations alter interactions with DNMT3A WT based on our previous findings on DNMT3A WT -p53 WT interactions and high incidence in human cancers (47)(48)(49)(50). Despite the greater stability of the DNMT3A WT :p53 R248W heterotetramers (Fig. 5) (K D,app of 41 Ϯ 6 nM) compared with the DNMT3A WT -DNMT3L complexes (K D,app of 80 Ϯ 12 nM), DNMT3L modulation of DNMT3A WT activity is dominant over that of p53 R248W and p53 R273H (Fig. 4). We propose that p53 mutations do not compromise modulation of DNMT3A WT (Fig. 6A, III). These mutations may allosterically affect the DNMT3A-interacting interface (p53 residues 319 -393, Fig. S3) (25) such that the affinity of mutant p53 in DNMT3A WT :mutant p53 complexes is compromised when presented with DNMT3L (Fig. 6A, V).
The ongoing discovery of hot spots in protein-protein interfaces, discrete regions that confer most of the binding energy, has sparked an interest in the pharmacological intervention of protein-protein interactions (78 -82). In fact, successful modulation of protein-protein interactions by small molecules has been reported in p53 (83)(84)(85)(86)(87)(88)(89). We describe how, in the context of DNMT3A interactions with partner proteins, mutations at protein-protein interfaces may lead to diverse changes in protein interactions and modulation of protein activity. In addi-  (I  and II). Similarly, p53 (blue square) interacts with DNMT3A homotetramers or homodimers (red square) to form DNMT3A (red square)-p53 (blue square) heterotetramers (III and IV). Furthermore, the addition of p53 (blue square) to DNMT3A (red square)-DNMT3L (yellow square) heterotetramers displaces monomers at the tetramer interface and leads to the formation of DNMT3A-P53 (blue square) heterotetramers (V). B, summary of the oligomeric states of DNMT3A mutants in complex with DNMT3L (28,36) and p53 as well as the effects on the catalytic function of DNMT3A. DNMT3A mutants display no change (-) or decreased (2) activity (k cat ) relative to WT. Although all the DNMT3A mutants are responsive to DNMT3L stimulation (1), DNMT3A mutants display varying p53 inhibition (2). Although DNMT3A R736A and R882H are unresponsive to p53 inhibition (-), p53 binds R882H to form heterotetramer.

DNMT3A R882H disrupts allosteric regulation by p53
tion, we provide examples in two important cancer-related proteins of how mutations located distally from protein-protein interfaces may affect modulation of enzymatic activity, thereby contributing to the diverse phenotypic consequences of mutations in epigenetic enzymes like DNMT3A (61,62). The findings in this study broaden our understanding of regulation of DNMT3A activity and emphasize the potential use of small molecules to target protein-protein interactions in diseases, like AML, where DNMT3A and p53 are implicated (26,90).

Computational modeling
Using a DNMT3A (PDB code 5YX2, chain A) (92) and a P53 monomer (PDB code 3TS8, chain B) (93), the protein docking server ZDOCK was initially used to predict the interface on DNMT3A involved with DNMT3A-P53 interactions (42). In addition, identical modeling was performed using a DNMT3A monomer (PDB code 5YX2,chain A) (92) and a DNMT3L monomer (PDB code 2QRV, chain B) (47) for comparison as the interface on DNMT3A for DNMT3A-DNMT3L interactions has been previously resolved (47,92). ZDOCK performs a rigidbody search of possible docking orientations between the proteins of interest (42). This docking server relies on the fast Fourier transform algorithm to perform a global docking search and explores all possible binding modalities by combining translation and rotation of the ligand. To rank each possible docking pose, ZDOCK applies a combination of shape complementarity, electrostatics, and statistical potential terms (94,95). To predict the interface on DNMT3A involved with DNMT3A-P53 interactions, known contacting residues on P53 were considered (residues 319 -393) (25). The local refinement on the RosettaDock server was then employed to perform rigidbody minimization and side chain conformation optimization (43). The local refinement function involves side chain repacking to improve rotameric side chain conformations and a Monte Carlo-based recovery of near-native protein structures (43). RosettaDock employs an energy-based scoring function that calculates the energy of interactions by amino acids (43).

Methylation assays
Assays were carried out to measure the ability of DNMT3A to incorporate tritiated methyl groups transferred from cofactor AdoMet onto DNA substrate under various experimental conditions. Reactions were carried out at 37°C in a buffer consisting of 50 mM KH 2 PO 4 /K 2 HPO 4 , 1 mM EDTA, 1 mM DTT, 0.2 mg/ml of BSA, 20 mM NaCl with saturating AdoMet (15 M) at pH 7.8. 50 M ([ 3 H]methyl-labeled:unlabeled) AdoMet stocks were made using 32 mM unlabeled AdoMet (New England Biolabs) and [ 3 H]methyl-labeled AdoMet (X Ci/mmol) supplied by PerkinElmer in 10 mM H 2 SO 4 . In all assays, 15-l aliquots were taken from a larger reaction and quenched by mixing with 0.1% SDS (1:1). Samples were spotted onto Hybond-XL membranes (GE Healthcare), washed, dried, and methylation was using a Beckman LS 6000 liquid scintillation Counter as previously established (96). Due to the large number of potential methylation sites, poly(dI-dC) is commonly used as a DNA substrate to study the enzymatic activity of DNA modifying enzymes (97)(98)(99)(100)(101). In addition, the use of poly(dI-dC) as a DNA substrate allowed us to investigate the isolated modulatory effect of p53 on DNMT3A activity. On this basis, 5 M poly(dI-dC) (Sigma-Aldrich) was used as a DNA substrate. Previous work from our lab has provided insights into the mechanism of DNMT3A activity on human promoters (102). Given that Cyclin-dependent Kinase Inhibitor 1A/P21 is a common target for DNMT3A and p53 (25, 102), we additionally employed Cyclin-dependent Kinase Inhibitor 1A/P21-pCpG L DNMT3A R882H disrupts allosteric regulation by p53 as a substrate. All radiochemical assays were performed in triplicates from a single purification and statistical analysis was performed using Prism version 7 (GraphPad).

P53 assays
To test modulation of DNMT3A (full-length, catalytic domain, and mutant enzymes) or M.HhaI methylation activity by P53 WT , proteins were preincubated in reaction buffer with AdoMet for 1 h at 37°C before initiating the reaction by the addition of substrate DNA. Reactions were then run for 1 h, stopped as stated above, and methylation was counted. All proteins (DNMT3A, M.HhaI, or P53 WT ) were at a ratio of 1:1 (150 nM) in the 1-h preincubation. Fold-inhibition was calculated by product formed by DNMT3A variants (or M.HhaI) divided by product formed by DNMT3A variants (or M.HhaI) without P53 WT . Statistical analysis was performed using GraphPad software (version 6.0).

DNMT3L and P53 assays
To evaluate regulation of DNMT3A activity by DNMT3L or P53 under a binary (protein pairs) approach, proteins (1:1:1 at 150 nM) were preincubated in reaction buffer with AdoMet for 1 h at 37°C before starting the reaction by the addition of substrate DNA. To assess modulation of DNMT3A activity under a co-complex (groups of proteins) approach, DNMT3A was preincubated with DNMT3L (or P53) (1:1 at 150 nM) for 1 h at 37°C prior to initiating the reaction by the addition of substrate DNA and enzyme were allowed to carry out catalysis for 30 min before the addition of P53 (or DNMT3L) (150 nM). Reactions were then monitored for an additional hour, stopped, and methylation was counted as stated above.

Apparent binding affinities (K D,app )
Modulation of DNMT3A activity by varying DNMT3L, P53 WT , or P53 R248W concentrations from 10 to 300 nM with 10 nM DNMT3A was tested to determine apparent affinities (K D,app ). DNMT3A WT was preincubated with varying concentrations of DNMT3L, P53 WT , or P53 R248W in reaction buffer with AdoMet for 1 h at 37°C prior initiating the reaction by the addition of substrate DNA. Following the addition of DNA, reactions were run for 1 h at 37°C, stopped, and methylation was counted as stated above. Fold-stimulation was calculated by product formed by DNMT3A WT with DNMT3L divided by product formed by DNMT3A WT without DNMT3L. Fold-inhibition by P53 WT or P53 R248W was calculated as stated above. The data were fit to a one site-specific binding equation using GraphPad software (version 6.0).

Fluorescence anisotropy
Fluorescence anisotropy measurements were performed using a Horiba Fluoromax fluorescence spectrophotometer equipped with excitation and emission polarizers (excitation, 485 nm; emission, 520 nm). Fluorescence anisotropy measurements of DNA-bound DNMT3A WT or DNMT3A R882H (both at 2.5 M) were taken following the titration of DNMT3L (DNMT3A WT and DNMT3A R882H reactions), P53 WT (DNMT3A WT and DNMT3A R882H reactions), or P53 R248W (DNMT3A WT reactions). Anisotropy values were obtained following a 5-min preincubation at room temperature. The sub-strate DNA (Gcbox30) consisted of a fluorescein (6-FAM) label on the 5Ј end of the top strand of the duplex (5Ј/6-FAM/TGG-ATATCTAGGGGCGCTATGATATCT-3Ј) and was supplied by Integrated DNA Technologies; the recognition site for DNMT3A is underlined.

Electrophoretic mobility shift assay (EMSA)
Experiments were carried out as described in Holz-Schietinger et al. (28). In brief, DNMT3A R882H (150 nM) was incubated at 37°C for 15 min with 200 nM duplex 5Ј 6-FAM GCbox30 in reaction buffer with 50 M Sinefungin (Sigma-Aldrich) and 10% glycerol. For P53 WT super shifting, varying concentrations of P53 WT were preincubated with DNMT3A R882H under identical conditions for 30 min at 37°C before the addition of GCbox30. Samples were run on a native 4.5% (75:1) polyacrylamide gel in 0.25ϫ Tris boric acid/EDTA, pH 7.8, at 250 V for 50 min. Gels were visualized for fluorescein using a Typhoon scanner and data were analyzed using ImageJ.