Conformational Adaptation Drives Potent, Selective and Durable Inhibition of the Human Protein Methyltransferase DOT1L


Corresponding author: Edward J. Olhava,


DOT1L is the human protein methyltransferase responsible for catalyzing the methylation of histone H3 on lysine 79 (H3K79). The ectopic activity of DOT1L, associated with the chromosomal translocation that is a universal hallmark of MLL-rearranged leukemia, is a required driver of leukemogenesis in this malignancy. Here, we present studies on the structure–activity relationship of aminonucleoside-based DOT1L inhibitors. Within this series, we find that improvements in target enzyme affinity and selectivity are driven entirely by diminution of the dissociation rate constant for the enzyme–inhibitor complex, leading to long residence times for the binary complex. The biochemical Ki and residence times measured for these inhibitors correlate well with their effects on intracellular H3K79 methylation and MLL-rearranged leukemic cell killing. Crystallographic studies reveal a conformational adaptation mechanism associated with high-affinity inhibitor binding and prolonged residence time; these studies also suggest that conformational adaptation likewise plays a critical role in natural ligand interactions with the enzyme, hence, facilitating enzyme turnover. These results provide critical insights into the role of conformational adaptation in the enzymatic mechanism of catalysis and in pharmacologic intervention for DOT1L and other members of this enzyme class.


protein lysine methyltransferase


protein methyltransferase






structure–activity relationship


surface plasmon resonance

Cellular differentiation in, for example, embryonic development is critically dependent on tightly regulated, spatial and temporal control of gene transcription. The fidelity of transcription is dependent on the gene-localized structure of chromatin (the amalgam of DNA and histone proteins that form the nucleosomal core structural units of chromosomes), which can switch between transcriptionally repressive (heterochromatic) and transcriptionally permissive (euchromatic) conformational states. The conformational transition between these chromatin states is facilitated by methylation of the DNA component of chromatin and by a collection of post-translational covalent modifications of the histone proteins (1). Among these post-translational modifications, histone methylation at specific lysine and arginine residues has proved to be particularly important in controlling gene transcription. Histone methylation is catalyzed by a class of enzymes known as the protein methyltransferases (PMTs) (2). A recent survey of the human genome suggests that there may be as many as 96 PMTs in humans, and that these enzymes divide into two structurally distinct families: the protein arginine methyltransferases (PRMTs) and the protein lysine methyltransferases (PKMTs) (3).

There are a number of human conditions in which the pluripotency that is characteristic of early embryonic development is recapitulated, such as in wound healing, hematopoiesis, immune response and the like. Likewise, in hyperproliferative diseases, such as cancers, dysregulation of gene transcription, associated with cell reprogramming, is a common hallmark of malignancy. Not surprisingly, the dysregulation of gene transcription and the attendant hyperproliferative phenotype can often be associated with genetic alterations in members of the PMT enzyme class (4). An informative example of this is provided by the hematologic malignancy known as MLL-rearranged leukemia.

MLL-rearranged leukemia is a genetically distinct form of acute leukemia that affects people at all stages of life, from infancy to late adulthood; patients with MLL-rearranged leukemia have a relatively poor prognosis and are not well treated by currently available therapies. A universal hallmark of this disease is a chromosomal translocation affecting the MLL gene on chromosome 11q23 (5). Normally, the MLL gene encodes for a PKMT that catalyzes the methylation of lysine 4 of histone H3 (H3K4) at specific gene loci (5). Gene localization is conferred by specific recognition elements within MLL, external to the catalytic domain (5). In the disease-linked translocations, the catalytic domain (SET domain) is lost and the remaining MLL protein is fused to a variety of partners, including members of the AF and ENL family of proteins such as AF4, AF9, AF10 and ENL (5,6). Thus, the MLL gene translocation is a common feature of MLL-rearranged leukemia and has been thought to play a causal role in this malignancy. The mechanistic basis for how this gene translocation might confer a pathogenic phenotype was unclear until recently. The loss of the MLL catalytic domain, resulting from the chromosomal translocation, might suggest a loss-of-function with respect to histone methylation. In fact, however, the chromosomal translocation results in a change-of-function with respect to histone methylation by the ectopic recruitment of another PKMT, DOT1L.

DOT1L catalyzes the methylation of histone H3 on lysine 79 (H3K79); H3K79 methylation is an activating mark with respect to gene transcription. DOT1L forms high-affinity complexes with the various MLL-fusion proteins via the fusion partner domains (i.e., AF4, AF9, AF10, ENL, etc.). As a consequence of the localization elements within the MLL portion of the fusion protein, the fusion protein–DOT1L complex results in ectopic gene localization of the DOT1L enzyme, where it catalyzes H3K79 methylation and consequent gene activation. Armstrong and colleagues (8) have shown that ectopic H3K79 methylation, resulting from MLL-fusion protein recruitment of DOT1L, leads to enhanced expression of a number of leukemogenic genes, including HOXA9 and MEIS1. Hence, while DOT1L is not genetically altered in the disease per se, its mislocated enzymatic activity is a direct consequence of the chromosomal translocation affecting MLL-rearranged leukemia patients; thus, DOT1L has been proposed to be a catalytic driver of leukemogenesis in this disease (8).

The recent demonstration of DOT1L enzymatic activity as a driver of MLL-rearranged leukemia (9,10) provides a compelling foundation for the development of potent, selective DOT1L inhibitors as therapeutic agents for the treatment of MLL-rearranged leukemia patients. Toward this end, we report here the design and optimization of a series of aminonucleoside inhibitors of DOT1L and their biological characteristics. The inhibitor series was initiated using mechanism-guided design principles, based on our understanding of the DOT1L enzymatic reaction mechanism, and optimized through structure-guided approaches using iterative enzyme–inhibitor complex crystal structures. Conformational adaptation of the enzyme active site attended potent inhibitor binding, and this observation was taken into account to develop a cogent structure–activity relationship (SAR) for active site-directed DOT1L inhibition. This optimization process resulted in the compound EPZ004777, which is a picomolar inhibitor of DOT1L with exquisite selectivity for its target enzyme. EPZ004777 has been shown to demonstrate potent and selective killing of MLL-rearranged leukemic cells, both in cell culture and in a highly aggressive disseminated mouse model of this disease (9). Thus, the aminonucleoside inhibitor series presented here provides a clear demonstration of the potential for potent, selective small molecule inhibitors of DOT1L to affect selective killing of MLL-rearranged leukemias.

Methods and Materials

Assays of enzymatic activity

Unless otherwise indicated, assays of DOT1L enzymatic activity were performed under balanced conditions (all substrates present at concentrations equal to their respective KM values) (11) using a radiometric assay of S-[methyl-3H] adenosyl-L-methionine transfer from S-adenosylmethionine (SAM) to chicken erythrocyte nucleosomes, as previously described (3,9). Reactions were initiated by addition of S-[methyl-3H] adenosyl-L-methionine and allowed to run at room temperature for 120 min before being quenched by the addition of 800 μm cold SAM.

Compound IC50 values were determined from assays of enzymatic activity in which compound was titrated into reaction mixtures by three-fold serial dilution from DMSO stocks. For each titration, 10 concentrations of inhibitor were used along with 100% inhibition [50 μmS-adenosylhomocysteine (SAH)] and 0% inhibition (1 μL of neat DMSO per well) controls (9). Plots of residual enzyme velocity as a function of inhibitor concentration were fit to a standard Langmuir isotherm equation (12) to derive estimates of the IC50 value of the compound. As described below, the inhibition modality of key compounds within the aminonucleoside series were tested and always found to be competitive with SAM and non-competitive with respect to nucleosome substrate. For most compounds, the Ki value was calculated from the IC50 value using the appropriate equation for competitive inhibition with respect to SAM (12).

For selected compounds, the inhibition modality with respect to the two substrates (SAM and nucleosomes) were determined by dual titration of compound and varied substrate concentration while holding the other substrate fixed at its KM value. Plots of velocity as a function of varied substrate at multiple inhibitor concentrations were globally fit to a general equation for enzyme inhibition (12) using Graphpad Prism. Selection of the modality for each data set was done by evaluating the value of α, a term related to the degree of co-operative or antico-operative interaction between substrate and inhibitor binding, as previously described (11). A value of α ≥ 10 was taken as consistent with competitive inhibition, while a value of α ≤ 0.1 was taken as consistent with uncompetitive inhibition. Values of α between 10 and 0.1 were considered to be consistent with non-competitive inhibition.

Compounds that displayed an IC50 value within 50-fold of the enzyme concentration used in initial assays ([E] = 0.25 nm) were treated as tight-binding inhibitors. In this case, the Ki value was determined by measuring the IC50 value of the compound (vide supra) at varying concentrations of enzyme from 5 to 0.25 nm. A plot of IC50 as a function of enzyme concentration was fit to a linear equation, and the y-intercept value was equivalent to Ki(1 + [S]/KM) where [S] and KM refer to SAM, the substrate with which these inhibitors compete. Knowing the values of [S] and KM used in the assay, the Ki value was then calculated from the y-intercept value (12).

Determination of ligand association and dissociation rate constants

Ligand association and dissociation rate constants were determined by surface plasmon resonance (SPR). DOT1L was stored in 20 mm Tris–HCl, 200 mm NaCl, 1 mm EDTA, 1 mm DTT, pH 7.8 and immobilized by direct amine coupling, diluting enzyme into coupling buffer containing 10 mm Hepes pH 7.4, 1 mm Tris(2-carboxyethyl)phosphine (TCEP). Immobilization run buffer contained 10 mm Hepes pH 7.4, 150 mm NaCl, 500 μm TCEP, and approximately 10 000 response units (RUs) of DOT1L was captured. A reference channel of a surface that was activated in parallel and blocked was created in a second flow cell was also created. Data was captured on either a Biacore 4000 (chip CM5; GE Healthcare Biosciences, Pittsburgh, PA, USA) or a Biorad ProteOn (chip GLM; Biorad Life Science Research, Hercules, CA, USA).

K d determinations were determined using run buffer containing 20 mm Tris pH 8.0, 10 mm NaCl, 100 mm KCl, 0.002% Tween-20, 500 μm TCEP, 2% DMSO, with the following injection parameters: 30 μL/min flow rate, with a 30 seconds association phase followed by monitoring dissociation for 30 second. Experiments were carried out at 25 °C.

Cell proliferation assay

The human leukemia cell line MV4-11 harboring the MLL-AF4 translocation was obtained from ATCC (CRL-9591). Cells were grown in Iscove’s Modified Dulbecco’s Medium with 10% fetal bovine serum (FBS). All cell culture reagents were purchased from Invitrogen Life Technologies (Carlsbad, CA, USA), and cells were maintained in a humidified incubator set to 37 °C, 5% CO2.

Cell proliferation was assessed by plating, in triplicate, exponentially growing MV4-11 cells in a 96-well plate at a density of 3 × 104 cells/well in a final volume of 150 μL. Cells were incubated in the presence of compound at increasing concentrations up to 50 μm. Viable cell numbers were determined every 3–4 days for a total of 14 days using the Guava Viacount assay (Millipore # 4000-0040). Analysis was performed on a Guava EasyCyte Plus instrument (Millipore, Billerica, MA, USA) according to the manufacturer’s protocol. On days of cell counts, growth media and compound were replaced, and cells split back to a density of 5 × 104 cells/well. Total cell number is expressed as split-adjusted viable cells per well. IC50 values were determined from concentration-dependent curves at day 14 using the Graphpad Prism software (GraphPad Software, Inc., La Jolla, CA, USA).

H3K79me2 quantitative ELISA

Exponentially growing MV4-11 cells were seeded in a 12-well plate at 2 × 105 cells/well in a final volume of 2 mL. Cells were incubated in the presence of increasing concentrations of EPZ003647, EPZ003696 or EPZ004777 up to 50 μm. Control cells were treated with 0.2% DMSO control. Cells (1–2 × 106) were harvested after 96 h of compound incubation and histones extracted, as previously described (9). An indirect enzyme-linked immunosorbent assay (ELISA) using acid extracted histones was run using matching microtiter plates (Immulon 4HBX #3855; Thermo Labsystems, Carlsbad, CA, USA). Plates were coated for total H3 and H3K79me2 detection with either 75 ng or 1500 ng/well of histones, respectively. The coating antigen was diluted in coating buffer (PBS + 0.05% BSA) for a final volume of 100 μL and allowed to incubate overnight. The plates were blocked with 300-μl blocking buffer (PBST + 2% BSA) for 2 h at RT, followed by a 2-hour incubation with 100-μl primary antibody (1:750 H3K79me2, CST 5472; or 1:5000 total H3, Abcam ab1791; Abcam, Cambridge, MA, USA) diluted in blocking buffer at RT. 100 μL of HRP tagged goat-anti-rabbit antibody (1:4000, CST 7074) in blocking buffer was added and allowed to incubate for 2 h at RT. The reaction was visualized with 100-μl 3,3′,5,5′ tetramethylbenzidine substrate (TMBS; BioFx Laboratories, Owings Mills, MD, USA) and stopped with an equivalent volume of 1 N sulfuric acid. Plates were read at 450 nm on an M5e plate spectrophotometer. Following each step, plates were washed three times with wash buffer (PBST), with an additional wash included following secondary antibody.

Quantitative real-time PCR analysis

Exponentially growing MV4-11 cells were plated in a 12-well plate at 2 × 105 cells/well in a final volume of 2 mL. Cells were incubated with increasing concentration of EPZ003647, EPZ003696 or EPZ004777 up to 50 μm for 6 days. Compound and media were refreshed on day 4, and cells split back to 5 × 105 cells/well. Cells were pelleted and processed, as previously described (9).

Structure determination of DOT1L with compound

The production of His-DOT1L-1-416 and crystallization with SAM have been reported previously (3). EPZ000004 was pre-incubated with DOT1L-1-416 in 10-fold molar excess, and then cocrystallization was set up with the hanging drop method using solution containing 100 mm sodium acetate, 1.8–2.0 m ammonium sulfate, 5 mm TCEP, pH 5.5, at 20 °C. In order to observe the protein at a physiological pH, crystals were transferred using nylon loops into a pH 7.5 Tris buffer (100 mm Tris–HCl (pH 7.5), 2.0 m ammonium sulfate, 5 mm TCEP). This transfer was done stepwise by soaking the crystals over the course of one hour in successive mixtures of acetate and Tris buffers. The crystals were equilibrated in the final Tris buffer at pH 7.5 overnight at 20 °C. The crystal were next transferred stepwise into cryoprotectant (25 mm Tris–HCl, 2.0 m ammonium sulfate, 30% glycerol, pH 7.5) and flash-frozen in liquid nitrogen. Data collection was done at beamline X12B at the National Synchrotron Light Source at Brookhaven National Labs, Upton, NY, USA.

For EPZ003696 and EPZ004777, DOT1L-bound structures were obtained by the soaking method. The DOT1L-SAM crystals were cross-linked by exposing the crystal containing hanging drop over the vapor of 1 μL of 25% glutaraldehyde, pH 3.0, for 1 h (13,14) and then soaked with mother liquor containing 1 mm compound overnight. The crystals were cryo-protected with 35% glucose in mother liquor and flash-frozen in liquid nitrogen. The diffraction data sets were collected at beamline 17U at the Shanghai Synchrotron Radiation Facility. All data were processed by HKL2000 (15). Structures were solved by molecular replacement (Phaser) (16) using the DOT1L-SAM structure (PDB code:3QOW) with the SAM molecule removed as a search model. Refinement was carried out by Refmac5 (17) and the model building was carried out by COOT (18). Detailed information of the diffraction data, refinement and structure statistics are provided in Table S1 in the Supporting Information. The 2fo-fc maps for the three ligands are shown in Figure S1A–C in the Supporting Information. The co-ordinates for the crystal structures have been deposited with PDB code of 4EK9 (DOT1L-EZP0004), 4EKG (DOT1L-EPZ003696) and 4EKI (DOT1L-EPZ004777).

Results and Discussion

Mechanism-guided design tenets

The PMTs are clearly emerging as a large target pool for human cancer indications and other diseases. Despite growing interest in these targets for drug discovery, there remains a relative paucity of pharmacologically tractable inhibitors of these enzymes (3). In particular, until recently, no inhibitors of DOT1L had been reported (9,19).

A common approach to inhibitor discovery is diversity library screening. Historically, however, the success of diversity library screening against unprecedented target classes has been quite mixed; this mixed success rate likely reflects the bias inherent to diversity library growth in that such libraries tend to be populated with pharmacophores for which success against previous target classes has been demonstrated.

Reflecting on these observations, we chose to take a multipronged approach to inhibitor discovery for DOT1L, including diversity library screening and mechanism-guided design. The latter approach proved most productive and will be the focus of the remainder of this report.

Mechanism-guided design refers to the process of designing inhibitors based on knowledge of the chemical reaction mechanism catalyzed by the target enzyme, the use of chemical intuition to infer from this the structures of key intermediate species along the reaction co-ordinate and the subsequent design of stable molecular mimics of these intermediate species (11,12,20).

In the case of DOT1L, methyl transfer from the thiomethyl moiety of SAM to the ɛ-N of the bound side chain of lysine 79 of histone H3 (H3K79) proceeds through a simple SN2 mechanism, requiring stringent alignment of the molecular orbitals of the methyl donor and acceptor atoms. The crystal structures of DOT1L bound to SAM and to the reaction product SAH illustrate a highly ordered active site with superimposable SAM and SAH configurations in the bound structures (3,21). The thiomethyl group of SAM is directed into a contiguous channel that forms the lysine-binding pocket of the enzyme; thus, facilitating facile group transfer once the ternary enzyme–SAM–histone complex has formed. Steady-state kinetic analysis suggests that DOT1L functions by a distributive mechanism, requiring dissociation of the enzyme from the histone substrate after each round of lysine 79 methylation (22). The product SAH binds to DOT1L with relatively high affinity (Ki = 320 nm), suggesting that product dissociation may be at least partially rate limiting to enzyme turnover.

The above biochemical data provides a starting point for inhibitor design. We chose to begin design based on the structure of the reaction product SAH, with the following key objectives in mind. First, we wished to reduce the polar surface area of the ligand by replacement of charged and/or polar functionalities. Second, we wished to engage recognition elements within both the SAM/SAH-binding pocket and the adjacent lysine-binding pocket to gain the affinity advantages of bisubstrate inhibitors (12). Lastly, we strove to accomplish these goals while maintaining reasonable ligand efficiency and improving the pharmacological tractability of the compound series.

Our initial approach toward the above stated objectives was to replace the highly polarizable sulfur atom, eliminate the charged groups on the amino acid portion of SAH and minimize the molecular mass contributed to by the amino acid side chain. To this end, we replaced the homocysteine moiety of SAH with a simple dimethyl amine to create a minimal pharmacophore, as represented by compound EPZ000004 (Figure 1). Despite these significant changes, compound EPZ000004 retained a reasonable degree of binding affinity for DOT1L (Ki = 38 μm); the binding free energy change (ΔΔGbinding) for this compound, relative to SAH, was only 2.9 kcal/mol (at 25 °C). Additionally, crystallographic analysis of the binary complex of DOT1L and EPZ000004 confirmed that, despite the S to N substitution, the compound binds within the active site of the enzyme in the same configuration as SAM. These data establish the utility of EPZ000004 as a founder molecule for SAR elaboration.

Figure 1.

 Chemical structures of S-adenosylhomocysteine (SAH) and selected aminonucleoside inhibitors of DOT1L.

Structure–activity relationships for SAH mimetic DOT1L inhibitors

We began modification of EPZ000004 by independent alterations to the alkyl side chains appended to the nitrogen atom of the 5′ amine. Fixing one substituent as methyl, we systematically varied the second substituent with low molecular weight substituents in hopes of more effectively engaging recognition elements near the juxtaposition of the SAM and lysine-binding pockets. To this end, a library of 25 compounds were prepared and tested for DOT1L inhibition. From this library, the compound EPZ002446 was identified, which demonstrated a DOT1L Ki of 12 μm (Figure 1).

The isopropyl group of EPZ002446 was speculated to reach into the lysine-binding pocket. Based on this assumption, we attempted to extend this functionality to further engage elements within this channel and simultaneously engage recognition elements within the amino acid-binding pocket by varying the substituent. In the course of library building for these purposes, we regularly tested key intermediate species along the synthetic route. To our surprise, the Fmoc-protected intermediate EPZ003144 displayed surprising potency for inhibition of DOT1L, with a Ki of 20 μm. The instability of this compound precluded detailed structural characterization by crystallography. This finding did, however, compel us to take the SAR in an unexpected direction, namely replacing the functionality with a short tether, linked to a large hydrophobic group. Replacing the Fmoc group with a tert-butyl phenyl urea resulted in compound EPZ003647, which showed a marked improvement in potency (Ki = 845 nm). Combining this replacement with an extension of the tether group by one methylene (from ethyl to propyl) led to even greater target engagement, as exemplified by EPZ003696, a compound with a Ki of 13 nm.

In parallel studies, we sought to understand the tolerance of DOT1L inhibitors for substitutions within the nucleoside portion of the pharmacophore. In general, substitutions on the nucleoside were not well tolerated. We did, however, find that the ring nitrogen at position 7 could be substituted by carbon; the resulting deazapurine compounds demonstrated potent inhibition of DOT1L as exemplified by compound EPZ004450, an inhibitor displaying a Ki of 4 nm.

Combining the seven deazapurine moiety with the tert-butyl phenyl urea and 5′-amino isopropyl groups (vide supra) yielded compound EPZ004777. As previously described, EPZ004777 is a potent (Ki = 300 pm), selective inhibitor of DOT1L that demonstrates activity in inhibiting intracellular methylation of H3K79 and selective killing of MLL-rearranged leukemia cell lines. This compound also demonstrated a statistically significant survival advantage in an aggressive, disseminated mouse model of MLL-rearranged leukemia (9).

Conformational adaptation drives inhibitor potency and long residence time

Most enzymes rely on conformational dynamics, particularly in the vicinity of the active site, to facilitate the chemical transformations of substrate that are associated with catalysis (11,12,23). Likewise, potent inhibitor binding to enzymes often involves conformational adjustments of the inhibitor-binding pocket to achieve maximum complementarity between functionalities on the inhibitor and recognition elements within the target enzyme (24,25). Conformational adaptation appears to play a critical role in driving inhibitor potency among the aminonucleoside inhibitors of DOT1L as well. During the course of compound optimization, we have performed detailed kinetic and crystallographic studies on a subset of compounds that represent critical milestones along the path to optimal potency and selectivity.

Table 1 summarizes the kinetics of inhibitor association and dissociation with DOT1L for the enzymatic product SAH and for three key compounds (EPZ003647, EPZ003696 and EPZ004777), as measured by SPR. Several points related to the data in Table 1 are noteworthy. First, as noted above, there is a significant jump in target potency between EPZ003647 and EPZ003696 (ΔΔGbinding = 2.3–2.5 kcal/mol at 25 °C) and again between EPZ003696 and EPZ004777 (ΔΔGbinding = 2.4–2.9 kcal/mol at 25 °C). Interestingly, all of the enhancement in target potency seen across this series appears to be driven by a reduction in the dissociation rate of the inhibitors from the binary enzyme–inhibitor complex (koff), which results in a dramatic change in the residence time (τ) of the binary complex from 5 seconds to 1 h in going from EPZ003647 to EPZ004777; while the dissociation rate constants and residence times change by over three orders of magnitude among these compounds, the association rate constant (kon) is virtually invariant across the pharmacophore series and is several orders of magnitude slower than the calculated diffusion limit. More advanced inhibitors within the aminonucleoside series continue this trend, displaying residence times >24 h with no attendant change in association rate constant (data not shown; more advanced compounds will be the subject of a separate communication).

Table 1.   Thermodynamic and kinetic values for DOT1L binding by SAH and selected aminonucleoside inhibitors
Compound K i (nm) K d (nm)a k on (per m second) k off (per second)τ (s.)
  1. SAH, S-adenosylhomocysteine.

  2. a K d calculated as koff/kon.

  3. bErrors represented as standard deviation (SD).

  4. c K i determined from concentration-dependent effects of compound on enzyme activity as described in Methods and Materials; error represented as standard error of measurement (SEM).

SAH320 ± 104b71 ± 401.4 ± 0.5 (× 106)0.1 ± 0.0410 ± 1
EPZ003647845 ± 472b167 ± 1001.2 ± 0.8 (× 106)0.2 ± 0.095 ± 3
EPZ00369613 ± 7b1.7 ± 0.21.2 ± 0.1 (× 107)0.02 ± 0.00150 ± 3
EPZ0047770.30 ± 0.02c0.10 ± 0.023.0 ± 0.4 (× 106)3 ± 0.3 (× 10−4)3333 ± 300

The relatively slow association rate constants seen for SAH and the aminonucleosides also suggest a conformational gating of compound access to the SAM/SAH-binding site. The upper limit of the association rate constant for diffusion-limited small molecule binding to a protein has been estimated to be on the order of 109/m per second (23). More typically, values of the association rate constant for small molecule substrates binding to enzymes are about 107/m per second and typical values for the residence time of enzyme substrates and products are in the millisecond to second range (23). Thus, even for the natural product of the enzyme, SAH, the rates of association and dissociation are slow relative to other natural ligands of enzymes. Indeed, the crystal structures of SAM and SAH bound to DOT1L reveal an occluded active site with no clear, unobstructed pathway from bulk solvent to the active site for compound association or dissociation; thus, some conformational adaptation must attend enzyme turnover in order for substrate to access the binding pocket and for product to be released.

These data also imply conformational adaptation of the enzyme in response to more potent (e.g., EPZ003696 and EPZ004777) inhibitor binding. Two possible conformational mechanisms are consistent with the data presented in Table 1 (24). The first is referred to as a conformational selection mechanism in which the free enzyme exists in two conformational states that are in (slow) equilibrium with one another: a state that binds inhibitor and one that does not. Binding of the inhibitor shifts the equilibrium in favor of the inhibitor-binding conformation. The second mechanism is referred to as the induced-fit mechanism. In this model, the inhibitor binds to the enzyme in a non-optimal conformational state and then induces a conformational adjustment of the enzyme to create a more complementary, tighter binding state of the binary enzyme–inhibitor complex. A third mechanism of slow association and slow dissociation results from situations in which the conformation of the enzyme does not change, but other factors limit the rate of ligand binding and dissociation (11,22). In this third situation, binding is often gated by the need for slow displacement of structured water molecules within the active site, displacement of metal-binding ligands, and similar slow processes that are distinct from conformational adjustments of the protein per se (11,22). In the present case of DOT1L interactions with high-affinity inhibitors, this third mechanistic possibility can be excluded as the crystal structure of the enzyme–inhibitor complexes displays clear evidence of protein conformational adjustments (vide infra).

As recently reviewed (24), the induced-fit mechanism is predominant among enzyme–inhibitor interactions. However, the current data do not allow us to definitively distinguish between the two conformational adaptation models described above. In either case, the low-affinity conformational state appears to be kinetically (and thermodynamically) insignificant in terms of inhibitor interactions. This is evident from the excellent agreement between the Ki values for the inhibitor series, determined from the concentration-dependent effects of compounds on enzyme activity, and the Kd values determined from the SPR-binding experiments (Table 1). The Kd values in Table 1 are calculated simply as the ratio of koff over kon, thus assuming a single-step binding and dissociation mechanism. If an intervening conformational state were kinetically significant in the inhibitor-binding pathway, one would expect to see biphasic association curves in the enzymatic assay or the SPR binding or both. The good agreement between the Ki and simply calculated Kd values here argues against a significant kinetic role for an intervening conformational state (11,12). A similar situation is encountered in other enzyme-inhibitor systems that work through an induced-fit mechanism for which the initial encounter complex is kinetically insignificant and rapidly isomerizes to the final conformational state of the tight-binding enzyme–inhibitor complex. This was observed, for example, in the case of piperidine inhibitors binding to the aspartyl protease pepsin. In this case, the binding phase for the initial encounter complex between enzyme and inhibitor could only be observed using stopped-flow measurements over the initial 5 seconds of the binding reaction (26).

Structural changes attending inhibitor binding

The inference drawn from the binding kinetics of conformational adaptation in DOT1L–inhibitor interactions is supported by crystallographic analysis of the structures of DOT1L–inhibitor binary complexes. Steady-state kinetic analysis of DOT1L inhibition by the aminonucleosides consistently demonstrated competitive inhibition with respect to SAM and non-competitive inhibition with respect to nucleosome (as exemplified for EPZ004777 in Figure 2). This kinetic characterization is consistent with the crystallographic data that demonstrates compound binding within the SAM/SAH-binding pocket for all tested compounds. As described above, the structures of the founder compound EPZ000004 and SAH are virtually superimposable (Figure 3). The more potent compounds EPZ003696 and EPZ004777 both contain an extended (i.e., propyl) tether and terminal hydrophobic groups appended to the 5′ amino nitrogen. We initially assumed that these extended structures would reach into the proximal lysine-binding channel of the enzyme, creating a bisubstrate-like mode of inhibitor interaction. Surprisingly, however, the crystal structure of DOT1L-EPZ003696 reveals an unanticipated mode of interaction with the enzyme.

Figure 2.

 Biochemical characterization of EPZ004777 inhibition of DOT1L. (A) EPZ004777 is a competitive inhibitor with respect to S-adenosylmethionine (SAM). The IC50 of EPZ004777 was determined as a function of SAM concentration relative to the Km of SAM ([SAM]/Km) and found to display a linear relationship as expected for competitive inhibition. Inset: Michaelis–Menten plot of product formation as a function of SAM concentration at various concentrations of EPZ004777. The data were fit globally to a general equation for enzyme inhibition (Copeland, 2005) and yield values of alpha of 10 ± 6 and Ki of 0.3 ± 0.03 nm. (B) EPZ004777 is a non-competitive inhibitor with respect to oligonucleosome (Nuc). The IC50 of EPZ004777 was determined as a function of Nuc concentration relative to the Km of Nuc ([Nuc]/Km) and found to be independent of [Nuc]/Km, as expected for non-competitive inhibition. Inset: Michaelis–Menten plot of product formation as a function of Nuc concentration at various concentrations of EPZ004777. The data were fit globally to a general equation for enzyme inhibition (Copeland, 2005) and yield values of alpha of 0.5 ± 0.2 and Ki of 0.3 ± 0.02 nm.

Figure 3.

 Superposition of S-adenosylmethionine (SAM) and EPZ000004 within the active site of DOT1L, demonstrating conservation of binding motif. DOT1L-SAM cocrystal structure (PDB code: 3QOW): in line presentation (carbon atoms: light blue; oxygen: red; nitrogen: dark blue). SAM: stick presentation (carbon atoms: light blue; oxygen: red; nitrogen: dark blue, sulfur: yellow). EPZ000004: stick presentation (carbon atoms: maroon; oxygen: red; nitrogen: dark blue). Hydrogen bonds between SAM and DOT1L are labeled by dashed lines.

The DOT1L-EPZ003696 structure reveals that the inhibitor reaches into a heretofore-unrecognized pocket immediately adjacent to the amino acid-binding subsite of the SAM/SAH-binding pocket of the enzyme (Figure 4). This pocket – which does not exist in the structures of the enzyme with SAM, SAH or EPZ000004 – is opened up by the tert-butyl phenyl urea functionality of EPZ003696. More specifically, the crystal structure of EPZ003696 reveals a number of novel interactions with DOT1L. For example, the charged 5′-amino group of the compound forms a hydrogen bond with the carbonyl oxygen of Gly163. The urea region occupies the binding site of the amino acid region of SAM. The terminal propyl nitrogen of the urea interacts with the side chain oxygen atom of Asp161 in the same binding mode as the carboxylate nitrogen atom in SAM. The urea carbonyl oxygen interacts with the nitrogen atom of Asn 241, similar to the interaction of one of the carboxylate oxygen atoms of SAM. In addition, the proximal nitrogen atom of the urea (attached to the aromatic ring) co-ordinates with Asp161 (Figure 4A). The steric bulk of the tert-butyl phenyl opens up the novel hydrophobic pocket by changing the side-chain conformation of Phe239, Tyr312, Met147 and Leu143, inducing a significant conformational change in the 130s’ loop to flip Thr 139 (a residue that otherwise interacts with the carboxylate terminus of SAM) away from the SAM-binding pocket (Figure 4B). The movement of Tyr312 causes a change in a loop consisting of residues 302–312 (300s’ loop). These changes result in both the 130s’ and 300s’ loops of DOT1L becoming disordered (Figure 4C).

Figure 4.

 (A). Key contacts between DOT1L protein and the urea moiety of EPZ003696 in the DOT1L-EPZ003696 cocrystal structure. S-adenosylmethionine (SAM) is superimposed on to EPZ003696. The interactions for 5′-amino and urea of EPZ003696 are labeled by dashed lines. DOT1L protein of the DOT1L-EPZ003696 cocrystal structure: in line presentation (carbon atoms: green; oxygen: red; nitrogen: dark blue). EPZ003696: stick presentation (carbon atoms: green). SAM: stick presentation (carbon atoms: light blue). (B) Opening of the hydrophobic pocket within the DOT1L-EPZ003696 cocrystal structure. Superimposition of DOT1L-SAM and DOT1L-EPZ003696 demonstrates the opening of the hydrophobic pocket by the tert-butyl phenyl group on EPZ003696. DOT1L structures are presented in Cα trace. The key residue side chains that open the pocket are displayed as in line presentation. Comparison of these residues in the DOT1L-SAM and DOT1L-EPZ003696 structures shows the conformational change and resultant opening of the hydrophobic pocket upon the binding of EPZ003696. DOT1L protein: in line presentation (carbon atoms: green for the DOT1L-EPZ003696 structure and light blue for the DOT1L-SAM structure; oxygen: red; nitrogen: dark blue). EPZ003696: stick presentation (carbon atoms: green). SAM: stick presentation (carbon atoms: light blue). (C) Loop disorder induced by binding of EPZ003696 to DOT1L. Superimposition of DOT1L-SAM and DOT1L-EPZ003696 shows the disorder in 130s’ and 300s’ loops upon the binding of EPZ003696. DOT1L protein: in line presentation (carbon atoms: green for the DOT1L-EPZ003696 structure and light blue for the DOT1L-SAM structure). EPZ003696: stick presentation (carbon atoms: green; oxygen: red; nitrogen: dark blue). SAM: stick presentation (carbon atoms: light blue; oxygen: red; nitrogen: dark blue; sulfur: yellow).

The crystal structure of the DOT1L-EPZ004777 complex reveals EPZ004777 binds to DOT1L in a similar manner as EPZ003696, with the 5′-amino isopropyl group occupying a region near that occupied by the methyl group of the thiomethyl on SAM (Figure 5).Thus, the high potency, longer residence time compounds EPZ003696 and EPZ004777 are associated with a novel conformation of the enzyme that expands the contiguous active site cavity to include a new pocket; the crystal structures of these compounds suggest that these inhibitors ‘punch through’ a protein wall to create this new pocket and to thus engage recognition elements both within the SAM/SAH-binding pocket and the newly formed hydrophobic pocket. Because this conformational change is unique to DOT1L, these compounds show excellent selectivity against other PMTs (9).

Figure 5.

 Superimposition of DOT1L-EPZ004777 and DOT1L-EPZ003696 structures. The DOT1L protein of DOT1L-EPZ004777 is in surface presentation in gray. EPZ004777 (carbon atoms: magenta) is superimposed with EZP003696 (carbon atoms: green).

In addition to the protein structural changes that attend potent inhibitor binding, it is often the case that inhibitor-induced displacement of key structural water molecules within enzyme active sites is a critical component of high-affinity ligand interactions. This may be a contributing factor in the tight-binding interactions between DOT1L and the aminonucleoside inhibitors presented here. However, the resolution of the various enzyme-inhibitor cocrystal structures is not sufficient for us to make any definitive statements with regard to the role of structural water molecule displacement in the binding of these compounds to DOT1L.

Correlation of enzyme-inhibitor affinity with cellular activity

Inhibition of DOT1L is expected to lead to a concentration-dependent diminution of H3K79 methylation levels in treated cells. For cells bearing the chromosomal translocation of the MLL gene, the diminution of H3K79 methylation is expected to translated into reduced transcription of leukemogenic genes such as HOXA9 (27), and thus, to a concomitant inhibition of cell proliferation. We have previously demonstrated that the potent, selective DOT1L inhibitor EPZ004777, indeed, leads to concentration-dependent inhibition of intracellular H3K79 dimethylation (H3K79me2), of HOXA9 gene transcription and anti-proliferative effects selectively for MLL-rearranged leukemia cells (9). Hence, there appears to be a correlative relationship between DOT1L target engagement, reduction in intracellular formation of the enzyme product (H3K79me2) and downstream transcriptional events leading to anti-proliferative efficacy for EPZ004777. In this study, we have extended these observations to other members of the aminonucleoside series to establish more firmly the relationship between target engagement and phenotypic effects of DOT1L inhibitors.

Compounds EPZ003647, EPZ003696 and EPZ004777 provide a structurally related series of DOT1L inhibitors, spanning almost 4 orders of magnitude in target affinity, with which to test the relationship between enzyme inhibition and cellular efficacy. Various concentrations of each of these compounds were applied to MV4-11 cells bearing a chromosomal rearrangement of the MLL gene (9) and the impact on H3K79me2 level, HOXA9 message and cell proliferation were assessed at appropriate time points, taking into account the distinct kinetics of compound impact on each of these cellular parameters (9). Table 2 summarizes the results of these studies. Consistent with our mechanistic hypothesis, the data in Table 2 illustrate a clear relationship between enzyme engagement, inhibition of intracellular histone methylation, selective effects on gene transcription and antiproliferative efficacy. As was previously reported for EPZ004777, all of these DOT1L inhibitors showed selective inhibition of intracellular H3K79 methylation without effect on other histone methyl marks. Likewise, and also consistent with our working mechanistic hypothesis, the antiproliferative effects of all of these DOT1L inhibitors were specific to MLL-rearranged leukemia cell lines; other leukemia cell lines without the chromosomal rearrangement of the MLL gene were essentially unaffected (antiproliferative IC50 > 50 μm) by this series of DOT1L inhibitors. These data provide compelling evidence for a causal relationship between DOT1L inhibition and selective cell killing of MLL-rearranged leukemia cells by the aminonucleoside compound series.

Table 2.   Relationship between enzyme inhibition and cellular effects of selected aminonucleoside inhibitors of DOT1L
Compound K i (nm)Cellular IC50 μm
H3K79 methylationHOXA9 mRNAProliferation
  1. Errors represented as standard deviation (SD).

EPZ003647845 ± 47228 ± 25>50>50
EPZ00369613 ± 70.272 ± 0.0146.9 ± 0.54.9 ± 3.9
EPZ0047770.30 ± 0.020.005 ± 0.0030.83 ± 0.170.146 ± 0.072


In this paper, we have presented the design and optimization of a series of aminonucleoside inhibitors of the PKMT DOT1L. Using the crystal structures of various aminonucleoside inhibitors bound to human DOT1L, we have systematically defined the key recognition elements of ligand binding to the enzymatic active site. Conformational adaptation is a common feature of enzyme catalysis and of high-affinity ligand interactions with enzymes (24). This is clearly the case with the potent aminonucleoside inhibitors of DOT1L. We saw that a novel hydrophobic pocket, immediately adjacent to the SAM-binding site of the enzyme, is opened up to accommodate the extended aminonucleoside compounds that had been originally designed to engage the lysine-binding channel of the contiguous enzyme active site. This surprising structural adaptation results in very high-affinity binding of aminonucleosides to the enzyme and provided new directions for inhibitor optimization. The conformational adaptation mechanism of inhibitor binding demonstrated for the aminonucleosides also results in extended residence time for the optimized members of this inhibitor series. For example, EPZ004777 displays a DOT1L enzyme residence time of 60 min, as measured by SPR. A long residence time on PMT targets, such as DOT1L, may prove to be of value in demonstrating durable pharmacology in patients (24,25).

The potent and selective inhibition of DOT1L by the aminonucleoside series translates into potent and selective inhibition of intracellular H3K79 methylation and to selective cell killing for leukemic cells bearing the MLL chromosomal translocation. The quantitative correlation between target engagement, intracellular inhibition of H3K79 methylation and anti-proliferative effects is striking, and leaves little doubt that the selective phenotypic effects of these compounds are driven directly by ablation of DOT1L enzymatic activity. These data provide a solid foundation upon which further optimization of DOT1L inhibitors may be conducted. The data provide compelling proof of concept for the application of DOT1L inhibitors for selective killing of MLL-rearranged leukemias, and thus, portend the utility of DOT1L inhibitors for therapeutic intervention in this disease.


We thank Dr. Robert Gould for helpful discussions.

Conflict of Interest

All authors are employees of Epizyme Inc.