Inhibition of Human Enhancer of Zeste Homolog 2 ( EZH 2 ) with tambjamine analogs

Combining computational modeling, de novo compound synthesis and in vitro and cellular assays, we have performed an inhibition study against the enhancer of zeste homolog 2 (EZH2) histone-lysine N-methyltransferase. This enzyme is an important catalytic component of the PRC2 complex whose alterations have been associated to different cancers. We introduce here several tambjamine-inspired derivatives with low micromolar in vitro activity that produce a significant decrease in histone 3 trimethylation levels in cancer cells. We demonstrate binding at the methyl transfer active Page 1 of 32 ACS Paragon Plus Environment Journal of Chemical Information and Modeling 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 site, showing, in addition, that the EZH2 isolated crystal structure is capable of being used in molecular screening studies. Altogether, this work provides a succesful molecular model that will help in the identification of new specific EZH2 inhibitors and identify a novel class of tambjamine-derived EZH2 inhibitors with promising activities for their use in cancer treatment. Page 2 of 32 ACS Paragon Plus Environment Journal of Chemical Information and Modeling 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Introduction
Enhancer of zeste homolog 2 (EZH2) is a histone-lysine N-methyltransferase enzyme, the catalytic component of the Polycomb Repressive Complex 2 (PRC2) that acts as a transcriptional repressor. It catalyzes the addition of three methyl groups, from the Sadenosyl-L-methionine (SAM) cofactor, to lysine 27 of Histone H3, one of the five main histone proteins involved in the structure of chromatin in eukaryotic cells. This modification facilitates chromatin compaction and gene silencing of tumor suppressor genes in cancer cells 1 . Since the early observation that EZH2 overexpression was associated with progression of prostate cancer 2 , several EZH2 alterations have been associated to other different cancers, including breast, colorectal and lung [3][4][5] . Moreover, EZH2 overexpression has also been associated with metastasis and poor clinical outcome 6 . Due to these evidences, we find several efforts in developing specific inhibitors for EZH2, including various preclinical studies and several human phase 1 and 2 clinical trials 3,7,8 .
Such biomedical interest has been translated in recent efforts to obtain crystal structures. Since 2013, there is public access to two human EZH2 crystal structures, protein data bank (PDB) entries 4MI0 and 4MI5. Since EZH2 without PRC2 components (SUZ12 and EED) is inactive, there is some controversy about the utility of these structures for drug design. In addition, during 2016 several structures of the human PRC2 complex were solved, including one with a pyridone inhibitor (pdb entries 5IJ7, 5IJ8) and a S-adenosyl homocysteine (SAH) cofactor (pdb entry 5HYN, clearly indicating the methyl transfer active site) 9 . Interestingly, these structures reveal an unusual binding mode. The inhibitor pocket is formed on the interface of the I-SET domain of EZH2, a stretch of 17 residues called by the authors activation loop, and the EED protein part of PRC2 complex. Comparing the EZH2/pyridone and the EZH2/SAH structures, we can observe that the only overlapping part of the inhibitor with the SAH molecule is the pyridone moiety, participating in a similar H-bond networking with Trp624.
We are involved in an anticancer drug discovery program focused at small molecules capable of facilitating the transmembrane transport of anions 10 , some of them (which we address here) inspired in the structure of marine secondary metabolites tambjamines 11 . We have synthesized and studied synthetic analogs of these natural products, which proved to possess potent anticancer activity in vitro, including activity against cancer stem cells 12 . We are also studying in depth the molecular mechanisms responsible for this activity 13 . A series of these derivatives were submitted to the Target Drug Discovery Program, within the Open innovation drug discovery panel, operated by Lilly (https://openinnovation.lilly.com/dd). Preliminary results from this program indicated important activity towards their oncology-EZH2 target. Moreover, our cellular assay showed an important reduction of histone H3 tri-methylation levels using antitrimethyl-Histone H3 (Lys27) antibody in two selected compounds. These encouraging results prompted us to carry out a study to design optimized candidates for this target using computational modeling, to synthesize these molecules and to test their in vitro efficacy.
We report here the results of this broad analysis. Our study indicates for the first time that our synthetic tambjamine molecules bind in the methyl transfer active site region of the enzyme. These results suggest that the inhibition of EZH2 may be involved in the cytotoxic properties of these molecules. Moreover, the knowledge of the binding mode, coupled to extensive state-of-the-art induced fit simulations, allowed us to design and synthesize 4 new molecules, all presenting low micromolar activity, supporting that the in silico model used is valid for the identification of novel effective EZH2 inhibitors.

Computational Results
The exploration of possible binding modes between the studied tambjamine analogs and EZH2 started with the standard practice of finding possible binding sites followed by a docking procedure. From the apo EZH2 crystal (PDB entry 4MI0), we collected the first nine sites identified with SiteMap, which included the SAM catalytic binding site, and dock in each of them all the tambjamine analogs ( Figure 1). Compounds 1, 2, 3 and 4 were selected as negative controls (were found negative in the biological assays); whereas all other 7 molecules (compounds 5-11) showed experimental low μM inhibitor activities against EZH2 in the target drug discovery program performed at Lilly (Table   1). The complete results of the Glide XP docking scores for the nine discovered binding sites are shown in supporting information Table S1. All of them indicate similar (poor) scoring values between -3.0 to -6.5 kcal/mol. While absolute docking scores are often ligand dependent, low micromolar activities tend to correlate with values < -10 kcal/mol. Thus, based on the scores obtained from docking, significant inhibition activity for EZH2 could not be predicted for the studied compounds. Moreover, there seems to be no preferential binding of the ligands into the methyl transfer catalytic site.
Nevertheless, the negative controls presented overall (slightly) lower scores than the rest of molecules, introducing some differentiation between the negative controls 1-4 and the other active ligands. Since standard docking techniques do not sample the receptor conformational changes, both at the induced fit or conformational selection level (only a limited ligand flexibility is introduced), we turned into PELE for a more robust exploration of the protein-ligand energy landscape. In recent studies in mTOR 25 and BCL 26 , for example, we have shown the necessity of introducing such type of analysis in order to correlate with experimental binding affinities. The first simulation involved a non-biased exploration of the entire EZH2 surface (global exploration). In such a procedure, the ligand is placed in the bulk solvent and is allowed to explore all the space without any restriction or predefined goal. At each PELE Monte Carlo step, the system is perturbed and relaxed, introducing significant conformational changes, and the protein-ligand interaction energy is computed. The global exploration, aimed at identifying the correct binding site, was performed only for compound 8, which introduces an average molecular structure among all ligands. Figure 2A shows the nine different ligand starting positions around EZH2 (at a distances of ~15 Å from the enzyme's surface). The 2B panel of the same figure shows the results from the exploration (64 processors and 48 hours each) where we depict the protein-ligand interaction energies with respect to the (ligand) distance to an active site atom: the alpha carbon from Tyr726. By introducing conformational sampling in both the protein and the ligand, we observe now a funnel shaped energy landscape towards the methyl transfer active site; this site is now clearly preferred to the rest of the protein surface and other cavities. Following the global exploration, we performed a locally restricted simulation where, starting from the best position (lowest interaction energy) obtained in the global search, we limited the ligand to explore only the vicinity of the active site. Here, we expanded a local sampling for all the ligands shown in Figure 1, where we manually edited the changes in chemical substituents with Maestro. As an additional positive control, we also performed the local exploration for the SAH co-factor, using again the apo EZH2 crystal, and compared the best interaction energy to the recent human PRC2 complex crystal (PDB entry 5HYN), including a SAH cofactor. Figure 3 shows an excellent level of agreement between our model and the crystal, where we reproduce the overall position and main interactions. From the local sampling we extracted the best 10 structures (with lowest interaction energy) for each compound and performed additional scoring calculations with Glide.
Thus, we were able to compare the protein-ligand affinity score before and after the induced fit process. The results clearly indicate a significant improvement in the ligands scores, reaching now values in agreement to their level of activity (~ -10 kcal/mol) (Table 1). Importantly, all the negative controls still show significant lower values than the active ligands. The binding mode shows the ability of the molecule to enter in the "end" of the channel where Lys27, the histone's lysine, is methylated. We should emphasize here that this channel is collapsed in the original apo crystal, which severely restricts the model to be used in standard docking calculations, thus requiring induced fit techniques (see Table S2 for ligand RMSD along the induced fit).
Compound Initial docking PELE + docking  (IC50) distressing the cognate substrate binding. In addition, the characteristic methoxy group of these derivatives binds approximately at the methyl transfer area of EZH2, as observed when comparing our models with the recent 5HYN crystallographic structure.
As seen in Figure 4, the indole moiety participates in a hydrogen bond with Asn688, mimicking the role observed for the SAH co-factor. Moreover, recently deposited PDB entries 5T5G and 5TH7 of the SETD8 protein (with a catalytic pocket structurally similar to EZH2) showed methoxy group containing inhibitors whose positions overlap well with that of our tambjamine analogs methoxy group. All these results underscore the ability of these derivatives inhibiting the methyl transfer from SAM to Lys27.
All active compounds bear a positive charge, which seems to benefits from binding in the methyl transfer cavity, which has evolved to accommodate the positively charged histone's lysine and SAM cofactor. The increase in the binding score for the protonated form of control ligands (with respect to their deprotonated forms) seems to confirm this point. Nevertheless, the increase is moderate and we still observe a clear difference between active and inactive compounds, indicating that additional aspects, such as the imine substituent flexibility, are necessary to accommodate the molecule in the catalytic pocket. The molecular knowledge of the binding mode was further challenged by designing a second generation set of inhibitors. Based on the above local exploration, eight different snapshots were selected for molecular docking. A small library varying the imine substituent of the tambjamine analogs was created. The library contained 150 fragments selected by molecular complementarity based on the docking grid (available volume). A further selection of 19 molecules was done based on their docking scores (< 11) (see Figure S2 for details) and from this set of candidates four of them were successfully synthesized to experimentally test the validity of our calculations (compounds 12-15, Figure 5). The four molecules were tested again in the OIDD Lilly assay and all of them were found to be positive hits, displaying activities in the low micromolar range ( Table   2). The fact that all the designed molecules were found active even when significant structural changes were introduced (at the hydrogen bond acceptor/donor level), underscore the capabilities of the molecular model and overall procedure. Although the IC50 values obtained did not improve those found in the first generation of molecules, the scaffold is useful for the identification of EZH2 inhibitors and further studies with more diverse compounds should be performed. In addition, the lack of improvement might be due to the limited precision of docking scores; more elaborate free energy methods might be better suited to guide lead optimization efforts.

Cell assays
EZH2 is a well-established therapeutic target for several cancers, so we wanted to analyze whether our tambjamine-inspired molecules are potential chemotherapeutic compounds. Thus, in order to complement and validate the in vitro enzymatic assay, we selected two compounds and their activity in cancer cells was assessed. Lung adenocarcinoma A549 cells were treated with 2 µM of compound 5 and compound 6 for 96 h and tri-methylation levels in Lys27 of histone 3 were assessed by immunofluorescence. As observed in Figure 6, both compounds induced a very significant decrease in histone 3 tri-methylation levels in cells. After fluorescence intensity quantification and normalization, cells treated with tambjamine analogs 5 and 6 showed a relative fluorescence (in percentage) of 11.37 ± 8.5 and 12.03 ± 6.5 respectively, compared to non-treated cells. These results indicate a statistically significant effect and corroborate in cancer cells the potent EZH2 inhibition by these compounds, formerly detected in in vitro experiments ( Figure 7). Therefore, these compounds effectively inhibit the cancer therapeutic target EZH2, suggesting that they may be good candidate agents for the treatment of different cancers.

Conclusion
Molecular modeling together with in vitro and cell assays indicate the capabilities of several tambjamine-inspired derivatives to inhibit EZH2, an important catalytic component of the PRC2 complex whose alterations have been associated to different cancers. These molecules are predicted to bind in the methyl transfer active site, which has evolved to accommodate the positively charged histone's lysine and SAM cofactor.
In agreement with this, the most active compounds present a positive charge, although our simulations indicate that introducing a positive charge in the control (

Computational Methods
System preparation and docking. The crystal structure, PDBID 4MI0, was optimized with the protein preparation wizard tool from Schrodinger. 14 The protonation state of titrable residues were assigned at physiological pH with PROPKA and double-checked with the H++ server. 15 Eleven tambjamine analogs, including four negative controls (molecules 1-4), were studied in the initial docking calculations (Figure 1). Selected The local sampling focuses the sampling in one region by reducing considerably the ligand translation (up to 1.5 Å), and by keeping the ligand's perturbation direction for only one step. The backbone perturbation direction is also reduced to 1 step. The sampling was performed using 12 processors and 12 hours, producing approximately 500-1000 MC minima (notice that, due to steric clashes, the acceptance in local sampling is reduced from that of global sampling). Disregarding duplicates (RMSD < 0.2 Å), the average ligand heavy atom RMSD along the different posses obtained is significantly large, ~2.5 Å. Thus the local sampling effectively maps the different conformations adopted by the ligands in an active site.

Experimental Methods
EZH2 SPA Enzymatic Assay. Human EZH2 was co-expressed as a 5-member enzymatic complex with SUZ12, EED, AEBP2 and RbAp48 using a baculovirus/Sf9 Immunofluorescence staining. A549 cells (8x10 4 cells/mL) were cultured in a 12-well plate containing FBS-coated glass coverslips. After 24 h, they were incubated 96 h with 2 µM of two of the studied compounds (5 and 6, as shown in Figure 1). Cells were then washed twice with PBS and fixed with PBS-2% paraformaldehyde for 10 min at RT.