Disruptor of telomeric silencing 1-like (DOT1L) is a non-SET domain nucleosomal histone methyltransferase (HMTase) involved in the specific methylation of histone H3 at lysine 79 (H3K79) (Feng et al., 2002). The human DOT1L is a 1,537 amino acids epigenetic enzyme, comprising an HMTase catalytic N-terminal domain (a. a. 1–332, Figure 1A) and a long-disordered C-terminal segment essential for binding to nucleosomes and DNA (a. a. 333–1,537) (Min et al., 2003). Several studies have demonstrated that the DOT1L C-terminal domain also interacts with mixed lineage leukemia (MLL) fusion partners, such as ENL, AF4, AF9, and AF10 (Mueller et al., 2007; Kuntimaddi et al., 2015; Okuda et al., 2017; Song et al., 2019). Overexpression of different MLL target genes due to the recruitment of DOT1L by these fusion partners has been associated to the development of acute myeloid leukemia (AML) (Hosseini and Minucci, 2018). For this reason, DOT1L has emerged as a key drug target for the treatment of AML.

To date, several nucleoside DOT1L inhibitors have been synthetized and tested against MLL-rearranged leukemia models (Anglin and Song, 2013). These inhibitors share an adenosine moiety that facilitates their recognition by the S-adenosyl-L-methionine (SAM) cofactor pocket in the DOT1L catalytic domain. Nucleoside inhibitors inactivate DOT1L through mechanism- or non-mechanism-based inhibition processes. SAH, the demethylated product of SAM, is a potent non-mechanism-based nucleoside DOT1L inhibitor (K _i = 160–320 nM, K _D = 71 nM) (Anglin et al., 2012; Basavapathruni et al., 2012). Unfortunately, this molecule also inhibits DNA methyltransferases and other HMTases (Zhang and Zheng, 2016; Zhang et al., 2021; Talukdar et al., 2022). To increase selectivity towards DOT1L, several modifications have been made to SAH structure (Figure 1B; Supplementary Table S1) (Anglin et al., 2012; Yu et al., 2013; Spurr et al., 2016). In previous works, the amino acid fragment of SAH has been replaced by carbamate, urea, and benzimidazole groups to keep the hydrogen bond (H-bond) interaction with D161, and a bulky substituent has been added at their distal end to interact with the buried hydrophobic pocket (Yi and Ge, 2022). Moreover, the sulfide group has been replaced by a tertiary amine to mimic the methyl-sulfonium substituent of SAM and display an H-bond interaction with G163 main-chain carbonyl. From the latter, EPZ003647 (K _D = 167 nM), EPZ003696 (K _D = 1.70 nM), SYC-522/FED1 (K _D = 1.99 nM), FED2 (K _D = 26.9 nM), and EPZ-5676/pinometostat (K _i = 0.08 nM) exhibited the best inhibitory results (Anglin et al., 2012; Basavapathruni et al., 2012; Yu et al., 2012; Daigle et al., 2013). Other research groups have further replaced the adenosine moiety by a 7-deazaadenosine, leading to compounds EPZ004450 (K _i = 4.0 nM), EPZ004777 (K _D = 0.253 nM), SGC0946 (K _D = 0.056 nM), and SGC0947 (K _D = 3.35 nM) (Basavapathruni et al., 2012; Yu et al., 2012). Overall, urea/benzimidazole-containing inhibitors displayed high selectivity for DOT1L over other HMTases and potent antileukemic activity (Daigle et al., 2011; Perner et al., 2020).

In addition to their high affinity and selectivity for DOT1L, surface plasmon resonance (SPR) studies have shown that urea-containing inhibitors also exhibit long on-target residence times (τ = 1/k _off) on this epigenetic protein (Basavapathruni et al., 2012; Yu et al., 2012). Current computer-aided drug design protocols seeking novel DOT1L inhibitors have focus on the “interaction strength” (binding affinity) between the ligand and DOT1L (Raj et al., 2015; Luo et al., 2016). Nevertheless, drug-target kinetics plays a key role in the drug discovery paradigm (Tonge, 2018). Therefore, we carried out τ-Random Acceleration Molecular Dynamics (τRAMD) (Kokh et al., 2018) simulations to estimate the τ of urea-containing inhibitors and compare our predictions with reported binding kinetics experimental data. The conventional simulations performed to generate the starting replicas of the τRAMD dissociation trajectories were also analyzed to study the interaction profile of the inhibitors at the DOT1L catalytic site. This study will be useful to estimate the binding kinetics of novel DOT1L inhibitors for treating AML.

The root-mean-square deviation (RMSD) and fluctuation (RMSF) analysis of the three independent 100 ns MD replicas performed for each system are reported in the Supplementary Material, as well as the statistical analysis of the τRAMD dissociation trajectories employed to compute the τ_calc (ps) (see Supplementary Table S2; Supplementary Figures S2–S5). The RMSF analysis highlighted the low mobility of binding site residues (a. a. 140–245) due to their intermolecular interaction with the bounded compounds, while the RMSD of the ligands confirmed the high stability of the DOT1L inhibitors at the active site during the simulation time. Table 1 shows the experimental [τ_exp (s)] and τRAMD calculated [τ_calc (ps)] on-target residence times for the nucleoside DOT1L inhibitors. As in other MD simulation studies to determine on-target residence times (Mollica et al., 2015; Schuetz et al., 2019), the τ values were normalized to compare the estimated residence times with the experimental data. The inhibitors-dissociation pathway and normalized log(τ) values of τ_calc and τ_exp plotted against each other are shown in Figure 2. The high coefficient of determination (R² = 0.87) demonstrated that τ_calc computed with the τRAMD method could reliably predict the experimental residence times.

In order to understand the role of the protein-ligand interaction profiles in the on-target residence times, we calculated the interaction fractions (IF) of the H-bond (HB) and hydrophobic (HI) contacts formed between DOT1L and the nucleoside inhibitors from the last 50 ns of the three independent replicas for each simulated complex. Noteworthy, no significant salt bridge or π-stacking interactions were identified during the simulations of the nine nucleoside DOT1L inhibitors. Heat maps in Figure 3 show the IF values of each intermolecular interaction formed during the analyzed simulation time. The structure depicts the DOT1L active site residues and inhibitor substituents involved in the protein-ligand interaction profile. Dark colors in the heatmaps highlight the regions with the highest frequency protein-ligand interactions, while lighter colors indicate the less frequent interaction pairs.

Contrary to SAH interaction profile, nucleoside DOT1L inhibitors did not display HB interactions with T139 and Q168. EPZ003647, SGC0947, and EPZ003696 exhibited a lower number of HBs with key residues D161 and G163 when compared to other compounds. Furthermore, these three compounds did not exhibit hydrophobic interactions with F245 due to their small amine substituent. Interestingly, these mismatches in the H-bond and hydrophobic interactions explain the low residence time predicted for EPZ004450 and a high residence time estimated for EPZ-5676.

We also computed the residence times of non-nucleoside DOT1L inhibitors CHEMBL4534250 (τ _calc = 493 ± 70 ps) and CHEMBL4590355 (τ _calc = 416 ± 47 ps) using a lower force magnitude during the τRAMD simulations. However, the predicted residence times do not correlate with the experimental information since a higher τ value was expected for CHEMBL4590355 (τ _exp > 14,400 s) than for CHEMBL4534250 (τ _exp = 2,580 s) (Chen et al., 2016).

The binding affinity of a compound towards a target is a widely used parameter employed in drug design to determine the “strength” of the target-ligand complex. Nevertheless, the time it takes for the complex to dissociate (binding kinetics) also plays a key role during the drug discovery workflow. Herein, we performed τRAMD simulations (Kokh et al., 2018) to estimate the on-target residence time (τ _calc) of nucleoside DOT1L urea-containing inhibitors. Data normalization was performed to compare the computed τRAMD residence time with previously reported experimental information. A similar data treatment has been carried out in previous studies (Mollica et al., 2015; Schuetz et al., 2019). The high correlation (R² = 0.87) demonstrated that the method reliably predicts nucleoside DOT1L inhibitors’ residence time compared to reported experimental data (τ _exp). It is worth mentioning that an increase in ionizable groups overestimates the residence time of compounds, which is why the estimated residence time (τ _calc = 1,322 ± 355 ps) and experimental data (τ _exp = 10 s) (Basavapathruni et al., 2012) of SAH were not included during the analysis.

The high similarity between the τ _calc values for SYC-522, EPZ004777, and SGC0946 suggests that the approach is poorly sensitive to changes made to the adenosine moiety. However, significant differences were observed in those ligands with modifications in the amino group and in the bulky substituent that occupies the hydrophobic pocket. For the first, the interaction profile analysis showed that the absence of the isopropyl substituent on the amine leads to the loss of a crucial hydrophobic interaction with F245 residue. Our results also suggest that this intermolecular interaction acts as an anchor that allows the ligand to keep H-bond interactions with D161 and G163 and increase its residence time at the catalytic site. As can be seen in the interaction fraction heatmaps (see Figure 3A), the absence of the isopropyl group also appears to substantially reduce the number of H-bonds between the protonated amine and the G163 backbone. In particular, the reduction of the aliphatic linker between the amine and the urea (EPZ003647) leads to the loss of H-bond interactions with both G163 and D161 since the substituent could not reorient and form both interactions at the same time. Although this behavior could also be attributed to a possible artifact of the molecular docking, we compared our DOT1L-EPZ003647 complex with that reported by Li and coworkers, in which the H-bond interaction with G163 was not initially displayed. Interestingly, the three independent MD simulations and τRAMD dissociation trajectories with this complex also showed a low fraction of HBs with D161 (IF = 0.33) and G163 (IF = 0.03) and a low on-target residence time of this compound (τ_calc = 140 ± 20 ps) (see Supplementary Figure S6). Therefore, this result confirms that low residence time of EPZ003647 could be due to the loss of these key H-bond interactions. On the other hand, 4-tert-butylphenyl replacement by 4-methylphenyl in FED2 resulted in the loss of interactions with L143 and Y312 residues comprising the hydrophobic pocket. As with the amine substitution, the loss of these interactions with the hydrophobic cavity could explain the low residence time for FED2.

It is well known that the equilibrium dissociation constant (K _D = k _off/k _on) and the residence time (τ = 1/k _off) are not always inversely correlated due to the variance of the association rate constant (k _on) (Copeland, 2016; Tonge, 2018). Several studies have demonstrated that compounds with similar binding affinities to a common drug target can exhibit vastly different residence times (Copeland, 2021). Nevertheless, Copeland (Copeland, 2016) found a high correlation between the K _D and k _off of the nucleoside DOT1L inhibitors reported by (Basavapathruni et al., 2012). The latter was possible because the k _on values between the different inhibitors are practically invariant. Herein, we plotted the normalized log [τ_exp (s)] and log [τ_calc (ps)] values against the normalized negative logarithm of the experimental K _D [pK _D = –log K _D (M)] (see Supplementary Figure S7). Our analysis showed that the τ_exp has a strong correlation with K _D (R² = 0.80), while a lower correlation was obtained with the τ_calc values (R² = 0.62). The decrease in the coefficient of determination between K _D and τ could be related to the association rate constant of FED2 (k _on = 3.68 × 10⁵ M⁻¹ s⁻¹), since this value is greater than those reported for the other nucleoside inhibitors (k _on = 1.20 × 10⁷ to 8.37 × 10⁶ M⁻¹ s⁻¹) (Basavapathruni et al., 2012; Yu et al., 2012). Notwithstanding the foregoing, the high correlation between these data suggests that the lack of intermolecular interactions observed for inhibitors with short τ values could also be able to explain their low binding affinities to DOT1L. Therefore, the preservation of H-bond interactions with D161 and G163, as well as the contacts with the hydrophobic pocket and F245, are key factors to design nucleoside DOT1L inhibitors with desirable binding affinities and on-target residence times.

Although EPZ004450 predicted residence time was not included during the correlation analysis due to the absence of experimental data, the interaction profile and residence time of this compound resembles those estimated for ligands that do not have an isopropylamine group (i.e., EPZ003647, SGC0947, and EPZ003696). On the other hand, the interaction and residence time (τ _calc = 3,142 ± 741 ps) predictions of EPZ-55676 are very similar to those observed with SYC-522, EPZ004777, and SGC0946, which agrees with its high residence time on DOT1L determined by SPR (τ _exp > 86,400.00 s) (Daigle et al., 2013). Therefore, it is very likely that the method can reliably estimate the residence time of this type of DOT1L inhibitors. However, this estimation might not be carried out with non-nucleoside inhibitors since the τ _calc values obtained for CHEMBL4534250 and CHEMBL4590355 do not correlate with experimental observations. In this case, both compounds exhibit a very similar interaction profile, changing only the frequency of a few hydrophobic interactions (see Supplementary Figure S8). Unlike the nucleoside inhibitors, the difference in the interaction profile does not allow us to explain the residence time estimations. Therefore, the method might not be useful for predicting the on-target residence times of non-nucleoside inhibitors. Nevertheless, more experimental determinations with non-nucleoside inhibitors will be necessary to support this conclusion.