Properly Substituted Analogues of BIX-01294 Lose Inhibition of G9a Histone Methyltransferase and Gain Selective Anti-DNA Methyltransferase 3A Activity

Chemical manipulations performed on the histone H3 lysine 9 methyltransferases (G9a/GLP) inhibitor BIX-01294 afforded novel desmethoxyquinazolines able to inhibit the DNA methyltransferase DNMT3A at low micromolar levels without any significant inhibition of DNMT1 and G9a. In KG-1 cells such compounds, when tested at sub-toxic doses, induced the luciferase re-expression in a stable construct controlled by a cytomegalovirus (CMV) promoter silenced by methylation (CMV-luc assay). Finally, in human lymphoma U-937 and RAJI cells, the N-(1-benzylpiperidin-4-yl)-2-(4-phenylpiperazin-1-yl)quinazolin-4-amine induced the highest proliferation arrest and cell death induction starting from 10 µM, in agreement with its DNMT3A inhibitory potency.


Introduction
Epigenetics studies all regulatory mechanisms involving heritable changes in gene expression without alteration of the DNA coding sequence, determining cell fate and phenotype [1][2][3][4][5]. Among the epigenetic signals, the DNA methylation occurs at the C5 position of the cytosine ring mostly in a CpG dinucleotide context, through the action of three active DNA methyltransferases (DNMTs): DNMT1, DNMT3A and DNMT3B that catalyze the transfer of a methyl group from S-adenosyl-L-methionine (AdoMet) to the C5-cytosine [6]. In addition, another related protein lacking catalytic activity, DNMT3L, physically interacts and stimulates enzymatic activities of DNMT3A or DNMT3B [7]. DNMT1 is responsible for the maintenance of methylation patterns during cell division, with a preference for hemimethylated CpG dinucleotides, and performs a crucial role in normal mammalian development, cell proliferation and survival. DNMT3A and DNMT3B are de novo methyltransferases, adding methyl groups in unmethylated genomic sequences at CpG dinucleotides, and are highly expressed in early embryonic cells, downregulated after differentiation and in adult somatic tissues, and overexpressed in cancer [1,6].
Chemical changes performed by other authors on 1 led to UNC0638 (3) [18] (Figure 1A), a potent and selective G9a/GLP inhibitor featuring the 1-iso-propyl-4-piperidinylamino moiety at C4, the cyclohexyl group at C2, the 3-(1-pyrrolidino)propyloxy chain at C7 and the methoxy group at C6 position of the quinazoline ring. Compound 3 showed nanomolar potency against G9a, excellent cell permeability and robust on-target activity in cells [18]. Interestingly, compound 3 was found to be inactive against other histone lysine methyltransferases, but displayed high micromolar activity against DNMT1 [18]. This observation suggested that the quinazoline moiety, used to develop lysine methyltransferase and demethylase inhibitors, could also be suitable to design, with appropriate substitutions, compounds able to inhibit DNMTs. In addition, we thought that such quinazolinebased compounds lacking the C6 methoxy group should not be able to bind H3K9 modifying enzymes. Thus, we prepared the series of 6,7-desmethoxy quinazolines 4-18 ( Figure 1B) by maintaining the 1-benzyl-4-piperidinylamino moiety (typical of 1 and 2) at the C4 position of the quinazoline ring and changing the substituent at C2, spanning from differently sized cyclic amines (containing or not heteroatoms and/or lipophilic groups) to openchain amines.
Here we report the evaluation of the quinazolines 4-18 against human DNMT1 (hDNMT1) and the C-terminal catalytic domain of human DNMT3A (hDNMT3A). For selected derivatives, their effect on the H3K9 methyltransferase G9a was also assessed. Some quinazolines were then tested in a cytomegalovirus-luciferase (CMV-luc) construct in KG-1 cells, repressed by methylation, in order to detect their capability to reactivate gene expression in cells [19]. Docking studies have been performed to gain insight on the binding mode of the most potent compounds on hDNMT3A.
Finally, selected derivatives were tested in human leukemia U-937 and in Burkitt's lymphoma RAJI cells, at 1-10-100 mM from 24 to 72 h, to determine their effects on proliferation and viability.

Chemistry
Melting points were determined on a Buchi 530 melting point apparatus and are uncorrected. NMR spectra were recorded at 400 MHz (1H) and 100 MHz (13C) on a Bruker AC 400 spectrometer; chemical shifts are reported in d (ppm) units relative to the internal reference tetramethylsilane (Me 4 Si). All compounds were routinely checked by TLC and 1 H-NMR. TLC was performed on aluminium-backed silica gel plates (Merck DC, Alufolien Kieselgel 60 F 254 ) with spots visualized by UV light. All solvents were reagent grade and, when necessary, were purified and dried by standard methods. Concentration of solutions after reactions and extractions involved the use of a rotary evaporator operating at reduced pressure of , 20 Torr. Organic solutions were dried over anhydrous sodium sulphate. Elemental analysis has been used to determine purity of the described compounds, that is .95%. Analytical results are within 60.4% of the theoretical values (see Table S1 in FileS1). All chemicals were purchased from Aldrich Chimica, Milan (Italy), or from Lancaster Synthesis GmbH, Milan (Italy), and were of the highest purity.

DNMT1 Assay
His-DNMT1 (182 kDa, human) was cloned, expressed and purified as described by Lee et al. [20]. The DNMT1 assay was performed according to Gros et al. [21]. Briefly, the reaction was started by addition of 90 nM of DNMT1 on a mix containing the tested compound (up to 1% DMSO), 1 mM of a AdoMet/

DNMT3A Assay
DNMT3A enzyme inhibition was adapted from the restrictionbased fluorescence assay protocol described in Ceccaldi et al. [22]. Briefly, a 5'-labelled biotin oligonucleotide is hybridized to its complementary strand labelled with 6-carboxyfluorescein at the 39-end into a 384 well microplate (black Optiplates; Perkin Elmer) pre-coated with avidin. The duplex contains a unique CpG site overlapping with a restriction site of a methylation sensitive restriction enzyme. The C-terminal catalytic domain of human DNMT3A (residues 623-908), produced as described [7], was added in each well (200 ng/well) and mixed with the chemical compounds at desired concentrations and freshly prepared AdoMet (20 mM final concentration) to start the reaction in a total volume of 50 mL. After 1 hour incubation at 37uC each well were washed three times with PBS, 0.05% Tween-20, 500 mM NaCl and three more times with PBST. Specific fluorescent signals were detected with the methylation-sensitive restriction enzyme HpyCH4IV (NEB) as described and measured on a Perkin Elmer Envision detector. The percentage of inhibition is reported. The

CMV-luc Assay
KG-1 cell line, stably transfected with the luciferase Firefly (Luc+ from pGL3 by Promega) reporter gene under the control of the CMV promoter (from pEGFP-N1 by Clontech) partially methylated (50%), is seeded at 20,000 cell per well in 96-well plate. After 24 h of incubation in the presence of the compounds or the solvent DMSO, the induction of the promoter is measured by quantification of luciferase with the Brite-lite assay system (Perkin Elmer) according to the manufacturer protocol. The luminescence is measured on EnVision Multilabel Plate Reader (Perkin Elmer) and the data are expressed in induction factor compared to the DMSO control condition. The mean of 2-4 experiments and its standard error is reported.

G9a Assay
Human G9a (residues 786-121 0; accession II NM_006709) was expressed as N-terminal GST fusion protein in E. coli. The reaction buffer was 50 mM Tris-HCl, pH 8.5, 50 mM NaCl, 5 mM MgCl 2 , 1 mM dithiothreitol (DTT), 1 mM PMSF, and 1% DMSO. Standard substrate concentrations were 5 mM biotinylated H3 (1-21) peptide (AnaSpec) substrate and 0.1 mM AdoMet. For control compound IC 50 determinations, the test compounds were diluted in DMSO and then added to the enzyme/substrate mixture in nanoliter amounts by using an acoustic technology (Echo 500; Lab-cyte). The reaction was initiated by the addition of 3 H-AdoMet, and incubated at 30uC for 1 h. The reaction was detected by a filter-binding method. Data analysis was performed using Graph Pad Prism software for curve fits, and GraFit (Erithacus) for global fit of kinetic studies. To determine the effects of 4, 10, 13 and 14 against G9a, the quinazoline compounds were tested in a 10-dose IC 50 mode with 2-fold serial dilution starting at 400 mM. The results are summarized in Table S2 in File S1.

Docking Studies
Prior to docking calculations, the Epik software was used to calculate the most relevant ionization and tautomeric state of compounds 4 and 14 [23]. Then the Glide program of the Schrodinger package [24] was used to dock 4 and 14 to the   DNMT3A structure (PDB 2QRV). The receptor grid generation was performed for the box with a center in the putative binding site. The size of the box was determined automatically. The extra precision mode (XP) of Glide was used for docking. The ligand scaling factor was set to 1.0. The geometry of the ligand binding site of the complex between 10 and the receptor was then optimized. The binding site was defined as 10 and all amino acid residues located within 8 Å from the ligand. All the receptor residues located within 2 Å from the binding site were used as a shell. The OPLS2005 force field was used for energy minimization. Water was used as an implicit solvent, and a maximum of 5000 iterations of the Polak-Ribier conjugate gradient minimization method was used with a convergence threshold of 0.01 kJ mol -1 Å -1 . All complex pictures were rendered employing the UCSF Chimera software [25].

Cellular Assays
U-937 and RAJI cell lines were purchased from Deutsche Sammlung für Mikroorganismen und Zellkulturen (DSZM). Cells were maintained in RPMI 1640 (Lonza) supplemented with 10% fetal calf serum (Lonza) and 1% antibiotic-antimycotic (Lonza). Cells were treated with compounds at the indicated concentrations in exponential growth phase. Proliferation and viability were assessed by trypan blue exclusion analysis at the indicated time points.

Results and Discussion
For the synthesis of the quinazolines 4-18, the 2,4-dichloroquinazoline 19 [26] was treated with 4-amino-1-benzylpiperidine at room temperature providing the 4-substituted intermediate 20, which underwent C2-chloro displacement at the quinazoline ring with the proper amines at 110uC in a sealed tube to provide the desired 2,4-disubstituted quinazolines ( Figure 1C).
For compounds 4, 10, 13, and 14, the most potent against hDNMT3A, the relative EC 50 values were determined, as well as for 10 against hDNMT1 ( Figure 2). In addition, when possible their IC 50 values against G9a were calculated, to assess their DNMT3A-selectivity and to ascertain if the absence of the C6/C7 dimethoxy substitution effectively abolishes the inhibitory effect of such derivatives against H3K9 methyltransferases (10-dose IC 50 mode with 2-fold serial dilution starting at 400 mM for G9a inhibition in Table S2 in File S1). Compound 1, AdoHcy, sinefungin and chaetocin were used as reference drugs (see File S1). Compounds 1, 4, 10, 13 and 14 were also tested against G9alike protein (GLP), a related H3K9 methyltransferase, to determine the percent of inhibition at 8 mM (see File S1). As expected, among the tested compounds, only 10 was able to inhibit also DNMT1 and G9a, however it being 4-and 18-fold more selective for DNMT3A. The other compounds 4, 13 and 14 were inactive against DNMT1 (0% of inhibition at 100 mM) and showed .40-fold selectivity for DNMT3A over G9a inhibition (Figure 2). In the G9a assay, the reference compounds showed IC 50 values from 0.76 to 8.3 mM (see File S1). Against GLP, the tested compounds showed a percent of inhibition ranging from 0 to 10%, whereas the corresponding value for 1 was 91.7% (see File S1).
Compounds 4, 10, 13 and 14 were tested in KG-1 cells for their ability to reactivate gene expression, through an integrated luciferase reporter system under the control of a methylated cytomegalovirus (CMV) promoter, which inhibits its expression (CMV-luc assay) [19] ( Figure 3A). The structurally related 1 was added for comparison. The fold-induction of the luciferase signal was measured in cells incubated for 24 h in the presence of the drugs at the indicated concentrations, and normalized to the value of the non-treated cells. SGI-1027 was used as a reference drug. At 5 mM the quinazolines 10, 13 and 14 furnished 2/3-fold induction of the luciferase signal, this effect increasing at 10 mM. At 25 mM 14 showed 7.5-fold increase of the CMV-luc signal, whereas 10 and 13 were not effective because of loss of cells due to the cytotoxicity of the compounds (data not shown). Compound 4 significantly induced the luciferase signal at 10 mM (2.8-fold) and reached its highest value at 25 mM (12.5-fold). Both compounds 4 and 14 at 50 mM were ineffective to reactivate CMV-luc because of cytotoxicity (not shown). SGI-1027 displayed 17-fold of induction of the signal at 5 mM, but at 10 and 25 mM suffered from cytotoxicity giving a decrease to 10-fold and no induction, respectively ( Figure 3B). Molecular modeling studies were performed to explain the reasons behind the hDNMT3A inhibitory activity of the newly discovered analogues of 1. In particular, the most potent and structurally diverse hDNMT3A inhibitors 4 and 14 were docked into the active site of the enzyme using the Glide software which has been already successfully used to suggest a binding conformation for other DNMT3A inhibitors [29]. To this end, the 2.89 Å resolution X-ray crystal structure of the catalytic hDNMT3A-hDNMT3L tetrameric complex bound to S-adenosyl-L-homocysteine (AdoHcy) was used (PDB code 2QRV) [7].
Docking simulations achieved for 4 placed the ligand in the enzyme's binding cleft of AdoHcy with a calculated docking score of 27.232 kcal/mol (Figure 4). In this conformation, the quinazoline core scaffold is embedded in a rather lipophilic and narrow cleft made up by I639, S665, and L884 (herein referred to as L 1 , Figure 4B). Indeed, the limited room available in the aforementioned cleft would not allow hosting of the C6-and C7-methoxy groups of 1 thus explaining why the latter compound is inactive against DNMT3A.
The achieved binding pose also allowed explaining the contribution to ligand binding of the C2-and C4-substituents. In particular, the C2-diazepane moiety takes favorable hydrophobic contact with T641 and I701 residues (herein referred to as L 2 ) explaining why substitution of the C2 position with neutral azacycloalkane rings (6-10) favorably fills the L 2 cleft and still results in effective DNMT3A inhibitors. Interestingly, unlike 6-10, 4 bears a protonated nitrogen (N4 of the diazepane ring) that is able to engage with the adjacent W889 forming a cation-p interaction, which was also detected, through X-ray studies [7], for the terminal positively charged nitrogen atom of AdoHcy in the interaction with DNMT3A ( Figure 4A). In agreement with this binding orientation, the absence of hydrophobic and/or cation-p interactions in this region is detrimental for enzyme inhibition (11,12,(16)(17)(18).
Regarding the C4 substituent, the 1-benzyl-4-piperidinylamino group is optimally oriented to engage a long-range ionic interaction with E660 so that the terminal aromatic ring is able to form a well-oriented T-shaped charge-transfer interaction with F636 and additional hydrophobic contacts in the cleft made up by V661, C662, V683, and L726 (herein referred to as L 3 ). Also in this case, the same contacts were already detected for the adenine ring of AdoHcy in its interaction with the enzyme ( Figure 4A) [7].
Interestingly, docking results achieved for 14 (docking score 2 6.353 kcal/mol) placed the ligand in a reverted binding pose in which the C2 and C4 substituents are now lodged in the L 3 and L 2 pockets, respectively. Such a binding position places the conformationally rigid C2-phenylpiperazinyl moiety of 14 so as to recapitulate the interaction pattern established by the C4substituent of 4 and the adenine ring of AdoHcy with the enzyme (Figures 4C, 4B and 4A, respectively). Accordingly, increasing of the steric demands of the C2-substituent (15)(16)(17)(18) should disrupt the interactions with the L 3 pocket thus being detrimental for the ligand inhibitory potency. Moreover, the L 2 pocket is now favorably contacted by the C4-1-benzyl-4-piperidinylamino moiety of 14 through a parallel displaced p-p interaction with W889 aromatic ring. In this position, the above mentioned L 1 pocket does not seem to be contacted by the ligand quinazoline ring that is now forming a hydrogen-bond between its N1 atom and R887 side chain; the latter residue is also forming an additional hydrogen-bond through its backbone carbonyl oxygen with the ligand exocyclic NH in C4.
To rationalize the lack of inhibition potency of the analogues of 1 against G9a/GLP, compound 4 was docked into the X-ray structure of the GLP enzyme obtained in complex with 1 (PDB code 3FPD) [16]. Glide results indicated that 4 is able to recapitulate the experimental binding position of 1 ( Figure S1a in File S1). Interestingly, in the latter compound the C6-and C7methoxy groups are demonstrated to play a critical role in the ligand inhibition as they provide further interactions to better mimic the substrate/GLP recognition [16] (Figure S1b in File S1), thus explaining the lack of GLP inhibition by 4 in their absence.
Docking studies were also attained for compounds 4 and 14 in the X-ray structure of DNMT1 crystallized in complex with sinefungin (PDB 3SWR, Hashimoto and Cheng, unpublished data), to clarify why these two ligands are inactive against the aforementioned enzyme. Comparison of the DNMT3A and DNMT1 methyltransferase domains reveals that the room available for ligand binding is similar and, for the latter enzyme, the presence of three distinct clefts (L 1 , L 2 and L 3 ) can be also detected. Moreover, experimental data allowed to postulate that, in the case of DNMT1, the inhibitors should interact with the socalled CXXC [28] and autoinhibitory linker [29] domains (residues 646-692 and 699-733, respectively) to induce efficient inhibition. According to Glide calculations, both 4 and 14 should be able to occupy the DNMT1 methyltransferase site ( Figure S2a and S2b in File S1). Nevertheless, the residue composition of this region (i.e. negatively charged residues in the L 2 pocket) should result in low binding affinity. Also, no tight contacts are established with the CXXC and autoinhibitory domains. This should explain why, despite the overall binding site similarities between the two enzymes, 4 and 14 are potent DNMT3A inhibitors without displaying any activity against DNMT1.
Moreover, we investigated the effects of 4, 10, 13 and 14 on proliferation and viability of U-937 and RAJI lymphoma cells ( Figure 5). Globally, all compounds were cytotoxic in the low micromolar range as they decreased cell proliferation as a consequence of the loss of cell viability. IC 50 values of 4, 10, 13 and 14 on cell viability at 48 h are reported in Table 2. SGI-1027 data [28] have been added for comparison purpose. In agreement with its DNMT3A inhibitory potency, 14 showed the highest antiproliferative effect in the tested cancer cell lines, it being, when compared to SGI-1027, 2.6-fold less efficient against U-937 but 3-fold more potent against RAJI cells.

Conclusion
Starting from the quinazolines 1-3, previously reported as H3K9 methyltransferase/demethylase inhibitors, we explored the possibility to design new related derivatives 4-18 as novel DNMT inhibitors maintaining the C4-substitution typical of 1, varying the substituent at C2 position and deleting the C6/C7 dimethoxy groups, crucial for the interaction with the H3K9 modifying enzymes. When tested against hDNMT1 and the catalytic domain of hDNMT3A, the majority of 4-18 displayed high potency against DNMT3A and no inhibiting activity against DNMT1, with the only exception of 10. The most potent compounds against DNMT3A, i.e. 4, 10, 13 and 14, were up.40-times selective for DNMT3A over G9a. The same compounds, tested below their cytotoxic dose, increased the luciferase signal of an integrated reporter system in KG-1 cells silenced by a methylated cytomegalovirus (CMV) promoter (CMV-luc assay). In human lymphoma U-937 and RAJI cells, 14 displayed the highest antiproliferative and cell death inducing effects, consistently with the highest inhibitory potency showed against DNMT3A.