Sage Journals: Discover world-class research

Abstract

For many novel epigenetics targets the chemical ligand space and structural information were limited until recently and are still largely unknown for some targets. Hit-finding campaigns are therefore dependent on large and chemically diverse libraries. In the specific case of the histone methyltransferase G9a, the authors have been able to apply an efficient process of intelligent selection of compounds for primary screening, rather than screening the full diverse deck of 900 000 compounds to identify hit compounds. A number of different virtual screening methods have been applied for the compound selection, and the results have been analyzed in the context of their individual success rates. For the primary screening of 2112 compounds, a FlashPlate assay format and full-length histone H3.1 substrate were employed. Validation of hit compounds was performed using the orthogonal fluorescence lifetime technology. Rated by purity and IC₅₀ value, 18 compounds (0.9% of compound screening deck) were finally considered validated primary G9a hits. The hit-finding approach has led to novel chemotypes being identified, which can facilitate hit-to-lead projects. This study demonstrates the power of virtual screening technologies for novel, therapeutically relevant epigenetics protein targets.

Keywords

epigenetics HMT virtual screening G9a EHMT2

Introduction

Epigenetics may be defined as the study of any potentially stable and heritable change in gene expression or cellular phenotype that occurs without changing the underlying DNA sequence.¹ Covalent modifications of DNA and histone proteins and the mechanisms by which these spatiotemporal patterns influence chromatin structure and networks of gene expression are the main focus of today’s research in epigenetics.

Epigenetic modifications are produced by the “writers,” the proteins that recognize them (the “readers”), and the “erasers,” which remove the modifications.^2,3 This description, commonly referred to as the “histone code hypothesis,” has helped to categorize the enzyme families that are involved in the chromatin regulation within cells. Among the “writers” of the histone code, the protein lysine methyltransferases (PKMTs) and protein arginine methyltransferases (PRMTs) have received the greatest attention because histone methylation is an essential function in many biological processes, including gene expression and transcriptional regulation, heterochromatin formation, and X-chromosome inactivation.⁴

Over recent years, a significantly increasing amount of data suggest that the enzymatic activities of proteins from both of these enzyme families play significant roles in cancer, inflammation, and neurodegenerative diseases, implicating vast interest for the drug discovery efforts within the pharmaceutical industry.^5-10

PKMTs and PRMTs catalyze the methylation of lysine and arginine residues. Similar to other druggable target classes like protein kinases, the PKMTs and PRMTs also share a common mechanism of catalysis, whereby the co-factor S-adenosyl-L-methionine (SAM) is the donor for the transfer of a methyl group to a nitrogen atom of either a lysine or arginine side chain.¹¹ The numbers of family members for the PKMTs and PRMTs that are encoded in the human genome have been estimated to be 52 and approximately 10 to 50, respectively.¹² For subsets of proteins with significant structural similarity, such as the SET domain histone methyltransferases (HMTs), this allows the application of target family–based approaches for the design of compound libraries, assay development, and thus targeted drug discovery.

Despite the significant interest in HMTs as potential drug targets, there are only a small number of inhibitors known and described in literature.¹² A natural product isolated from Streptomyces spp., the SAM analogue sinefungin has been one of the first nonselective HMT inhibitors identified.¹³ Virtual screening and validation of PRMT1 inhibitors have been successfully applied using a PRMT1 homology model against the arginine binding site.¹⁴ The first drug-like and selective HMT inhibitors were CARM1 inhibitors described by Allan et al.,¹⁵ but activity of these inhibitors in cell-based assays has yet to be shown.

BIX-01294, an inhibitor of the HMT G9a with an IC₅₀ = 1.6 µM, has been found to bind to the protein substrate binding site of the target, explaining the noncompetitive behavior compared to SAM.¹⁶ The compound has been identified in a high-throughput screening (HTS) campaign of 125 000 compounds and shows selectivity against other SET domain HMTs such as SETDB1 and SETD7. It has also been demonstrated that BIX-01294 reduces the H3K9me2 (histone H3 lysine 9 dimethylation) levels in mammalian cells. The quinazoline-based scaffold of BIX-01294 has been further optimized by structure–activity relationship (SAR) exploration with the support of X-ray co-crystal structures, leading to the discovery of UNC0224.¹⁷ The more potent picomolar G9a inhibitor UNC0321, recently described by Liu et al.,¹⁸ resulted from the aim to generate tool compounds for further understanding of the biological function of G9a and its role in chromatin remodeling.

G9a (EHMT2; KMT1C) represents a well-characterized example for the HMT family of lysine methyltransferases.^19,20 G9a is a SET-domain containing histone methyltransferase that specifically mono- and dimethylates histone 3 at lysine 9, H3K9,^21,22 and other protein targets,²³ including a conserved methylation motif with marked sequence similarity to H3 in G9a itself.²⁴ Methylation at H3K9 marks important events in epigenetic signaling, which lead to transcriptional silencing. For G9a, co-repressing^25-29 and co-activating³⁰ functions have been described.

A functional correlation of G9a activity and cancer progression has been postulated, and a correlation of G9a overexpression and lung cancer metastasis has been described.³¹ Knockdown of G9a in PC3 cancer cells induces reduced growth, centrosome disruption, and chromosome instability.³² Evidence for G9a-dependent epigenetic silencing of tumor suppressor genes in human breast cancer was published for MDA-MB-231 cells.³³

Virtual screening (VS)^34-39 encompasses a wide range of computational methods for selecting compounds that are predicted to be active in hit-finding campaigns. The impact of VS in drug discovery projects and its benefit compared to, for example, random HTS has been the topic of controversial discussions.^40-42 Nevertheless, there is clear evidence that VS has contributed to compounds that have reached the clinic and the market.⁴⁰ There is consensus that the success of VS is hard to predict and that it is difficult to gauge a priori which VS method is best suited for a particular project situation.³⁵ In general, it is believed that parallel use of various VS approaches promises higher success.

VS approaches can be classified into structure-based VS (SBVS) and ligand-based VS (LBVS) methods. SBVS typically uses docking algorithms to place compounds into the 3D structure of a protein binding site of interest, thereby generating 3D conformations (poses) predicting the bioactive protein-ligand complex. Subsequently, these poses are scored and subjected to postprocessing procedures, to identify the compounds that are most likely to be active (see the review by Klebe⁴³ for an in-depth overview and discussion of the state-of-the art and unresolved problems of SBVS).

The starting points for LBVS are the structures of known active compounds. LBVS includes a large portfolio of machine learning and chemoinformatics methods.⁴⁴ Despite their known limitations, fingerprint-based similarity searches are the most commonly applied LBVS methods,⁴⁵ along with 3D pharmacophore search methods.⁴⁶ Here, known active compounds are used as queries that are searched against compound databases. It has been argued that even if the target structure is available and suitable for SBVS, the application for LBVS methods can complement a SBVS campaign.³⁷ Given that, we have based our VS strategy on a selection of SBVS and LBVS approaches.

Several crystal structures of the catalytic domain of G9a have been solved,^18,47 offering the opportunity to apply SBVS in G9a hit-finding campaigns. Starting points for LBVS for G9a include the nonselective SAM competitive inhibitor S-adenosyl-homocysteine (SAH) and selective substrate competitive inhibitors BIX-01294 (diazepinyl-quinazolinamine derivatives), UNC0224, and UNC0321.^16-18

In this report, we describe a case study for finding new inhibitors of G9a, including assay development of two orthogonal assays, computational compound selection of a small screening deck tailored for G9a, followed by screening and follow-up characterization of the initial hits. Primary screening of the selected 2112 compounds in a radiometric FlashPlate assay and validation of hit compounds by both a G9a fluorescence lifetime (FLT) assay and the radioactive readout were employed. Hits were classified with respect to the different virtual screening methods for validating the selection approaches applied.

Materials and Methods

Biochemical Assays

FlashPlate assay

For the radiometric FlashPlate assay, full assay development was carried out, including determination of optimal enzyme and histone H3.1 concentrations, SAM K_M,app, DMSO tolerance, reaction time, and sensitivity to reference inhibitors. The G9a reaction was performed in a 50-µL reaction volume in 384-well polypropylene plates (Greiner Bio-One, Monroe, NC). The assay buffer used was 50 mM Tris-HCl (pH 8.8), 5 mM MgCl₂, 4 mM dithiothreitol (DTT), and 2% DMSO. Final assay conditions in the screen were 1.3 nM G9a (BPS Bioscience, San Diego, CA), 0.5 µM ³H-SAM (PerkinElmer, Waltham, MA), 2.5 µM unlabeled SAM (Sigma, St. Louis, MO), and 50 nM full-length histone H3.1 (BPS Bioscience) incubated for 90 min at room temperature. The reaction was stopped by an excess of unlabeled SAM and then transferred to FlashPlates (PerkinElmer). Plates were incubated for >4 h at 4° C and washed once with 0.1% Tween-20 washing buffer, and scintillation of bound product was measured in a TopCount reader (PerkinElmer). The assay was validated with two SAM-competitive reference compounds: SAH (Sigma) and sinefungin (Sigma).

Fluorescence lifetime assay

For the FLT-based screen, a G9a-specific FLEXYTE assay, developed by Almac (East Lothian, UK), was used. This approach uses a site-specifically modified peptide substrate with the long-lifetime fluorophore 9-aminoacridine.⁴⁸ The HMT reaction is coupled to a proteolytic cleavage step by Endo-LysC, where lysine methylation is masking the endoproteolytic cleavage site.⁴⁹ The assay is configured such that methylation by G9a is reported through a decrease in the fluorescence lifetime, with a change in fluorescence lifetime of ~5 ns between substrate and product. The G9a reaction was carried out in a 25-µL reaction volume in 384-well black polystyrene plates (Greiner Bio-One) for 120 min at room temperature. The assay buffer consisted of 20 mM Tris-HCl (pH 8), 25 mM NaCl, 0.025% Tween-20, and 1 mM DTT. Final assay conditions in the screen were 7.9 nM G9a (BPS Bioscience), 5 µM unlabeled SAM, and 1 µM peptide substrate (Almac). Cleavage by Endo-LysC (2 nM final concentration) was performed for 20 min. FLT measurements were performed using a NanoTaurus plate reader (Edinburgh Instruments, Livingston, UK).

Computational Compound Selection

A total of 2112 compounds, corresponding to six assay plates containing 352 cherry-picked samples each, were selected using complementary VS approaches. The entire BioFocus compound collection comprising approximately 900 000 compounds was used as a pool for the selection. Various VS methods were applied in parallel, and the results were subsequently merged. The VS campaign comprised ligand-based virtual screening methods, including pharmacophore graphs, and an integrated toolkit of chemoinformatics methods for identification of compounds that are chemically similar or related to known actives. Furthermore, SBVS, by means of docking, followed by postprocessing of the docking poses, was performed. Thresholds and parameters have been selected such that the three groups of approaches (docking, pharmacophore graphs, chemoinformatics toolkit) produced roughly the same number of prioritized compounds. Details are given in the sections below.

Ligand-based VS: pharmacophore graphs

For all ligand-based VS approaches, the 2D structures of SAH and two quinazoline derivatives (BIX-01294, UNC0224) that bind to the substrate binding pocket^16,17 were used as query structures for all searches. DiscNgine’s Chemistry Collection (version 1.0.3; DiscNgine, Romainville, France) features pharmacophore graph searching methods that resemble the concept of feature trees⁵⁰ but also incorporate newer extensions such as graph reduction methods.⁵¹ Pharmacophore graphs are abstractions of 2D molecules that represent molecules as unrooted trees. Groups of atoms are condensed into molecular features such as donors, acceptors, aromatic moieties, and so on. These features or pharmacophores form the nodes of the trees; together with the edges, these graphs can be considered generic 2D pharmacophores. Graph-theoretical methods allow researchers to rapidly compare different trees and to perform similarity searches against a vast compound database. The Pipeline-Pilot implementation of this package was employed using default settings against the three query structures. All hits with a “best score” ≥ 0.5 (DiscNgine default scoring function) were kept and subjected to postfiltering.

Postfiltering has been implemented such that compounds with high similarity to any of the query structures were prioritized but not exclusively considered. Rather than taking only a predefined number of top-ranked compounds into consideration, compounds were chosen from three similarity ranges: ]70;100], ]60;70], and ]50;60] (similarity values s are given in %; the bracket annotation ]60;70] corresponds to 60 < s ≤ 70). The ]70;100] similarity range comprised 7 compounds, all of which were accepted. From 252 compounds in the ]60;70] range and 2915 compounds in the ]50;60] similarity range, only chemically diverse subsets of 200 and 300 compounds were selected, respectively. In this way, a total of 507 compounds were selected, representing a graded similarity selection.

Ligand-based VS: integrated chemoinformatics-based toolkit

The three query structures used in the pharmacophore graph searches were also used for various similarity searches that are also often used in the context of hit expansion. All methods have been seamlessly integrated in and executed from within PipelinePilot. Selection methods included the following:

Fingerprint-based similarity searches using multiple fingerprints (MACCS, FCFP_6, FCFC_4) and Tanimoto metrics

TurboSim two-step similarity searches⁵² with various fingerprint sets as given above

Substructure searches using the Murcko scaffolds⁵³ of the query structures

Substructure searches using virtualized query structures generated from SMARTS-based ring generalization rules

Similarity searches as outlined above using virtualized query structures generated from SMIRKS-based ring assembly modification rules

Substructure and similarity searches as above using bioisosteric enumerations as query structures. Published transformation rules,⁵⁴ as implemented in PipelinePilot, were used.

The merged list of results comprised a total of 4275 compounds that were found by at least one of the search methods.

In a similar manner to the postprocessing of the pharmacophore graph results, the hits found with the chemoinformatics toolkit searches were postfiltered as follows: Four subsets were defined, and depending on the assumed relative importance, diverse subsets were selected from each of those subsets, resulting in a total of 736 compounds selected:

A total of 350 diverse compounds were selected from the set of all compounds exhibiting a Tanimoto similarity ≥0.9 (i.e., ]0.9;1.0] range) in any of the similarity searches (793 compounds).

A total of 250 compounds were selected from the 3152 compounds in the ]0.8;0.9] similarity range.

All remaining compounds that were found independently (i.e., in consensus) by three or more of the search methods were chosen, which corresponded to 36 compounds.

From all remaining 276 compounds, a diverse subset of 100 compounds was chosen.

Structure-based VS: docking and postprocessing

The cofactor (SAM) binding site was the target for the SBVS campaign. A number of X-ray crystal structures of G9a have been solved in recent years^17,47 and deposited with the Protein Data Bank (PDB).⁵⁵ These include co-crystal structures with SAH, small-molecule inhibitors binding to the substrate binding site, and structures in complex with histone peptides. The structures with PDB codes 3nni, 3k5k, 2o8j, and 3dm1 were superimposed using molecular operating environment (MOE) and assessed for conformational conservation in the SAM binding site. It was concluded that conformational differences were minimal, and the SAM binding site was considered to be relatively rigid. Also, the crystal environment around the SAM binding site was analyzed for crystal packing effects⁵⁶ that could potentially impede the docking. It was found that there were no crystal mates within an 8-Å radius around SAH, and hence crystal packing effects were not considered. The structure was also analyzed with respect to its solvent structure. It was found that SAH binds to a few water molecules, some of which are even conserved across the superimposed G9a structures. However, all of these water molecules form contact with the SAH ligand only and not with the protein, and no water-mediated protein–ligand contacts were observed. For docking, the G9a X-ray crystal structure 3nni was taken from the PDB. The Maestro “Protein Preparation Wizard” was employed for preparing the structure for docking. All ligands and solvent molecules, particularly SAH, were removed from the structure. The receptor grid was generated using the recommended standard procedure, and default input parameters were used in all calculations. A brief validation of the docking protocol included redocking of a random 3D conformation of SAH, and the binding mode found in the X-ray crystal structure could be reproduced well.

Before docking, a prefilter step was applied to the entire BioFocus compound collection, and compounds not matching the following acceptance criteria were omitted: 150 < molecular weight < 800; AlogP < 8; LogS > −8. All compounds were prepared for docking with LigPrep, using default settings. Docking calculations were performed using Glide SP^57,58 on a 32-processor cluster running Linux.

Three output poses per input molecule were requested; however, postprocessing was performed on the 25 000 highest ranked poses, based on the GlideScore scoring function. All other poses were omitted from further processing. Two postprocessing modules were employed for triaging the 25 000 poses. The first module was implemented in PipelinePilot and aimed to capture all poses featuring the same important hydrogen bond anchors as for SAH in the adenine subpocket of the SAM binding site while exhibiting high scores for a number of scoring functions (i.e., having a high consensus score [CS]).

The PipelinePilot “Pose Analysis” component was used for annotating all hydrogen bonds between the G9a structure and the poses, and residue-based scoping was used. PilotScript was employed for flagging all poses featuring the same two essential hydrogen bonds as the G9a–SAH complex in the adenine subpocket.

For all poses, the 15% and 30% percentile consensus scores (CS30, CS15) were calculated, using the value of GlideScore associated with the poses, and the rescored values calculated with LigScore1, LigScore2, PLP1, PLP2, Jain, PMF, and PMF04 scoring functions.^59–62

A total of 114 poses with CS30 ≥5 were included in the selection, and this was referred to as the high consensus set.

A total of 271 poses featuring the two essential H-bonds and showing a less stringent CS15 ≥3 were added to the selection, and this was referred to as the subpocket anchor set.

The two merged sets compiled in this process corresponded to 255 compounds that were added to the compound selection. All poses not fulfilling these criteria but exhibiting at least a weak consensus of CS30 ≥3 were subjected to the second postprocessing module using the BioFocus Virtual Screening Toolkit. This toolkit is proprietary to BioFocus and was developed in collaboration with Chemical Computing Group. A total of 3104 poses were processed in this way. All other poses were omitted at this stage.

Using the toolkit, two terms were calculated. The first measured a snugness of fit (D50) of the ligand pose in the receptor site and the second a volume overlap (VO) between the G9a-bound SAH conformation and each pose. D50 gives an indication of the cumulative distance between ligand and receptor, and a low value implies tight binding.

High-priority and low-priority sets of poses were defined using a D50/VO scatterplot with selection criteria being validated by visual inspection of the poses. A total of 397 poses exhibiting very low D50 values and very high VO were selected. The low-priority set comprised 883 poses with relatively low D50 and relatively high VO values. From these, a chemically diverse subset was selected. The number of compounds to be selected was adjusted such that the intended number of compounds (2112) to be tested was reached. In total, 624 compounds were selected using the toolkit and added to the compound selection.

The chemical diversity of the compounds selected by VS was estimated using Murcko scaffolds and MACCS keys-based clustering. For the latter, a Tanimoto threshold of 0.75 was used. The 2112-compound pool represents a total of 1678 Murcko scaffolds and 1236 clusters. This demonstrates that the chemical diversity of the selected compounds is high. A detailed breakdown of cluster numbers in the various VS subsets is given in Table 1 .

Table 1.

Number of Compounds Selected (#Selected) Using Various VS Methods, Number of Hits after Verification (#Hits) with Associated Hit Rate (Hitrate), and Number of Finally Validated Hits (#Fhits) (i.e., compounds with fitted IC₅₀ value)

Method	#Selected	#Hits	Hitrate	#Fhits
CS/anchor postprocessing	255 (196/133)	8 (8/5)	3.1
BioFocus toolkit	616 (502/493)	20 (16/16)	3.2	3 (3/3)
BioFocus toolkit/pharmacophore graphs	1 (1/1)
Chemoinformatics TB	727 (614/144)	50 (46/26)	6.9	7 (7/6)
Chemoinformatics TB/BioFocus toolkit	7 (2/5)
Chemoinformatics TB/pharmacophore graphs	2 (2/2)
Pharmacophore graphs	504 (389/388)	17 (17/17)	3.4	8 (8/8)
Total/combined	2112 (1678/1236)	95 (85/67)	4.5	18 (18/18)

Numbers in parentheses refer to the corresponding number of Murcko scaffolds and clusters generated using MACCS keys and a Tanimoto threshold of 0.75. CS, consensus scoring; TB, tool box.

Results

G9a Assay Development

Full G9a assay development, including testing of reaction linearity, substrate titrations, SAM K_M,app determination, DMSO tolerance, and reference compound efficacy in dose–response experiments, was performed. G9a was titrated, and incorporation of labeled ³H was stopped at indicated time points by addition of an excess of unlabeled SAM (10 mM). The reactions were performed at nonlimiting concentrations of full-length histone H3.1 and SAM. For all G9a concentrations, the velocity decreased after 90 min ( Fig. 1a ). The following experiments were performed with 1.3 nM G9a and a reaction time of 90 min.

Figure 1.

(a) Reaction linearity at different G9a concentrations. Reactions were stopped by an excess of unlabeled S-adenosyl-L-methionine (SAM) at the time points shown. Mean cpm of four wells is shown. (b) Histone H3.1 titration. Mean cpm of three wells ±1 SD is shown. (c) SAM K_M,app determination.

Full-length histone H3.1 was titrated in a time course experiment ( Fig. 1b ). The G9a HMT reaction velocity dropped at all histone concentrations after 60 to 70 min. Together with the G9a titration in Fig. 1a , these data indicate a loss of enzyme activity/stability after 90 min. A full-length histone H3.1 concentration of 50 nM was used for the rest of the study because the reaction linearity was maintained over a period of 90 min. G9a reaction velocities were determined for different SAM concentrations, and velocities were plotted versus SAM concentration. Data were fitted using the Michaelis-Menten equation for determination of the SAM K_M,app value ( Fig. 1c ). A SAM K_M,app value of 1.2 µM was determined for G9a, and a SAM concentration of 3 µM was used for the final validation of the assay in dose–response experiments using product inhibition by SAH and the reference inhibitor sinefungin, which binds to the SAM binding site of G9a ( Fig. 2a ). Both compounds (structures and published IC₅₀ values are shown in Suppl. Fig. S1) were titrated starting from 300 µM maximal concentration in a 10-step 1:3 dilution series. IC₅₀ values determined for SAH and sinefungin against G9a were 4.7 µM and 500 µM, respectively. The value for SAH correlates well with published results^49,63; for sinefungin, a higher IC₅₀ value than published was observed. To ensure consistent assay sensitivity in the screening phase, an SAH sensitivity control (10 µM final concentration SAH) was included on each screening plate.

Figure 2.

(a) IC₅₀ determination of S-adenosyl-homocysteine (SAH) and sinefungin against G9a in the radiometric FlashPlate assay. (b) IC₅₀ determination of SAH and sinefungin against G9a in the G9a fluorescence lifetime (FLT) assay.

To compare the sensitivity of the radiometric FlashPlate and the G9a FLT assays, the predeveloped G9a FLEXYTE FLT assay (Almac) was validated in a dose–response experiment also using SAH and sinefungin (Fig. 2b). IC₅₀ values determined for SAH and sinefungin against G9a in the FLT assay were 3 µM and 52 µM, respectively. Assay sensitivities of both assays were comparable, with the FLT assay being slightly more sensitive for both inhibitors.

Virtual Screening Approach and Results

As described in detail in the Materials and Methods section, a total of 2112 compounds were selected by various VS methods; 872 compounds were selected using structure-based VS, 736 using the chemoinformatics toolkit, and 506 with DiscNgine’s pharmacophore trees. Chemical diversity of the subsets was monitored by counting the number of chemical clusters per group. Both Murcko scaffolds and fingerprint-based clusters (MACCS keys, T = 0.75) were used. An overview of the VS workflow is shown in Figure 3 . The overlap between the methods both in terms of individual compounds and compound classes was remarkably small (see Table 1 ). Nevertheless, all three groups of VS approaches (docking, pharmacophore trees, chemoinformatics toolkit) have identified compounds that lead to validated hits and then finally confirmed hits with associated IC₅₀ value.

Figure 3.

Virtual screening (VS) workflow and number of compounds selected by the different VS methods. Structure-based VS included prefiltering and docking with Glide in the S-adenosyl-L-methionine (SAM) binding site of G9a, followed by postprocessing of the poses. Postprocessing included consensus scoring (CS) and subpocket anchor analysis, as well as use of the BioFocus Virtual Screening Toolkit. Ligand-based VS was performed using a PipelinePilot implementation of various chemoinformatics search methods (chemoinformatics toolkit) and pharmacophore graphs. For both result sets, a graded postfiltering scheme was applied.

Radiometric G9a Screen

In total, 2112 compounds on six assay plates were screened against G9a in the radiometric FlashPlate assay. Compounds were screened in single determinations at 10 µM final concentration. Plate performance is given in Supplemental Table S1. Z′ values⁶⁴ for all plates were >0.6 with a signal-to-background (S/B) ratio >13. Mean % inhibition of 10 µM SAH controls was in the range of 40% to 60% with one outlier on plate 1. The activity distribution of control and sample wells is shown in Figure 4 . Bell-shaped distributions were obtained for all different sample categories. Distributions of full reaction controls (100%) and sample wells overlapped with the maximum of samples showing 100% Ar (= no inhibition). The maximum of the activity distribution for the IC₅₀ control samples was around 50% Ar (corresponding to 50% inhibition of G9a).

Figure 4.

Activity distributions of sample and control wells in the G9a screen. Relative activities (Ar%) are binned in 5% steps.

Hit rates in the primary screen for different cutoff values are shown in Table 2 . One hundred five compounds (5% hit rate) showed more than 50% inhibition in the primary screen, which supports the validity and success of the compound selection process. All compounds showing >30% inhibition in the primary screen were cherry picked and retested in duplicates at a 10-µM compound concentration. Mean inhibition values from the verification run were used to determine the confirmation rates, which are shown in Table 2 .

Table 2.

Hit and Confirmation Rates of G9a Screen

Phase	>30% Inhibition (Hit Rate)	>40% Inhibition (Hit Rate)	>50% Inhibition (Hit Rate)	>60% Inhibition (Hit Rate)
HTS	233 (11.0)	149 (7.1)	105 (5.0)	82 (3.9)
Confirmation	95 (41.0)	58 (39.0)	32 (30.0)	15 (18.3)

Hit and confirmation rates are given in %. HTS, high-throughput screening.

Compound Selection for Dose–Response Experiments

From the 95 confirmed hits, 32 compounds were selected for IC₅₀ determination against G9a and purity analysis by liquid chromatography/mass spectrometry (LC/MS). These 32 compounds correspond to a hit rate of 1.5% with regard to the starting screening deck of 2112 compounds. Again, a graded selection approach has been chosen (see Suppl. Fig. S2 for details). Inhibition and heavy atom count-based ligand efficiency (LEHA)⁶⁵ were used for defining the subsets.

Dose–response experiments were performed in duplicate with a maximal compound concentration of 100 µM and a 10-step 1:3 dilution series. Twenty-three compounds (1% of entire G9a screening deck) were finally validated as G9a inhibitors with IC₅₀ values between 0.6 and 55 µM. Eighteen of these compounds showed purity values >75% ( Fig. 5 ), corresponding to a hit rate of 0.9% for fully validated hits with respect to the primary screening deck. Of these compounds, 13 have been identified by the LBVS methods via query molecules that bind to the substrate binding site, suggesting that these compounds could bind at this site. Five compounds were found by docking and LBVS with SAH as a query molecule; hence, these compounds are more likely to target the SAM binding pocket.

Figure 5.

Validated G9a inhibitors with IC₅₀ values <55 µM and compound purity >75%. Based on the virtual screening (VS) selection method and the ligand-based VS (LBVS) query molecules used for each of the compounds, the putative G9a binding pocket is shown (1 = S-adenosyl-L-methionine [SAM] binding site; 2 = histone H3.1 binding site).

The compounds that were progressed to IC₅₀ experiments were tested in parallel in the orthogonal G9a FLT assay ( Fig. 6 ). The confirmation rate in the FLT assay was low, and only two compounds showed modest inhibitory activity in this assay. For compound A ( Fig. 6 , left graph), an IC₅₀ value of 1 µM was obtained in the FlashPlate assay, whereas in the FLT assay a value of 65 µM was observed. Compound B ( Fig. 6 , right graph) was confirmed in the FlashPlate assay with an IC₅₀ value of 5.5 µM, and in the FLT assay an IC₅₀ value of >90 µM was determined. Although sensitivity of the G9a FLT assay was proven to be similar to the FlashPlate assay for reference compounds SAH and sinefungin (see Fig. 2 ), differences in both assay protocols with regard to substrate and enzyme concentrations might partly explain the results. Interestingly, both compounds shown in Figure 6 were originally selected by VS as potentially targeting the SAM binding site.

Figure 6.

IC₅₀ curves for two compounds that were confirmed in the primary assay and fluorescence lifetime (FLT).

The set of 18 G9a inhibitors was subjected to a fast-track version of the chemoinformatics toolkit for hit expansion mining the entire BioFocus compound inventory. Preliminary analysis of this hit expansion list identified more than 3000 compounds for additional testing. Assuming that the hit and confirmation rates would be similar to the figures presented in this article, such a comparably small effort is likely to further increase the number of active compounds against G9a. Initial SAR information and potential starting points for medicinal chemistry hit-to-lead optimization may therefore become available.

Discussion

HMT Targets in Drug Discovery

Since the discovery of the first mammalian HMT in 2000,⁶⁶ HMTs now represent an important target class for drug discovery in human disease.^6,7,67 Due to the common small-molecule co-factor SAM that is involved in the catalytic mechanism, HMTs are considered druggable by small-molecule compounds. Moreover, small molecules that bind to the substrate binding cavity have also been characterized, offering a second binding site for drug discovery programs. The huge potential for therapeutic intervention by targeting HMTs has been topic of several recent reviews,^12,68 and especially the association of certain HMTs with human cancers has received much attention.^8,69

G9a is one of the best characterized HMTs and physiological functions, cellular substrates, structural information, and involvement in diseases and development of potent small-molecule inhibitors of the enzyme have been described.^18,22,47 We used G9a as a model HMT for a knowledge-based selection from the BioFocus compound library that comprises 900 000 compounds. A computational approach using a combination of virtual screening methods was implemented to select 2112 compounds that were subsequently screened against G9a. Known structural information on the SAM binding pocket of G9a was used in parallel to similarity searching of the chemical space around known G9a inhibitors. This compound selection strategy turned out to significantly increase the number of primary, confirmed, and validated hits in the FlashPlate-based screen when compared to screening campaigns that are, for example, based on a maximal diverse compound screening deck only. The different virtual screening methods complemented each other, providing minimal overlap in the selection subsets ( Table 1 ), and these subsets were also equally represented in the hit lists. The results strengthen the rationale of the chosen strategy and demonstrate the power of combined VS approaches for new epigenetics targets.

Value of Orthogonal Assay/New Developments/Assay Technologies

Drug screening against HMTs has been facilitated by several new or adapted assay technologies. In addition to the standard radiometric assay using tritiated SAM as substrate, nonradioactive, homogeneous assay formats have recently been added to the repertoire of HMT assays. Among these, the AlphaLISA detection system from PerkinElmer and time-resolved fluorescence resonance energy transfer (TR-FRET)–based assays that were introduced by PerkinElmer and Cisbio (Bedford, MA) are dependent on methylation-specific antibodies that are available for many histone methylation sites. In addition, generic readouts of HMT activity using SAH-converting enzymes have been introduced (Transcreener EPIGEN Methyltransferase Kit from BellBrook Labs, Madison, WI). Also, label-free HMT assays have been introduced, and the RapidFire technology from Biocius (now Agilent Technologies, Santa Clara, CA) uses µSPE-MS technology for the analysis of HMT activity. A homogeneous mobility shift assay using the Caliper LabChip technology (Caliper Life Sciences, Hopkinton, MA) allowing accurate kinetic studies was recently described for the identification of small-molecule inhibitors of HMTs as well as histone demethylases (HDMs).⁴⁹ The modification of peptide substrates by G9a or the HDM LSD1 does not influence the net charge of the peptides, so activity was detected by masking/demasking of a proteolytic cleavage site that was required for a coupled Endo-LysC reaction. This novel assay technology has been a successful approach to identify and further optimize G9a inhibitors.

Although a small number of homogeneous assay technologies are currently available for HMT screening and profiling, we have decided to perform the primary screen employing the FlashPlate assay technology. The technology is the gold standard and the most sensitive method for testing the activity of HMTs. The assay interference is low, and most important, the method allows the use of complex enzyme substrates. Nucleosome extracts, full-length proteins, or peptide substrates can be used equally well. For our study, we decided to use full-length histone H3.1 protein in the primary radiometric assay and a peptide-based orthogonal secondary assay technology.

The secondary assay is based on the modulation of the fluorescence lifetime of a labeled peptide. The FLEXYTE FLT assay used in the current study is a peptide-based, homogeneous assay format that does not require antibodies. FLT is considered an attractive assay technology in comparison to other approaches because assay interferences, caused by compounds or assay reagents, are minimized.⁷⁰ Recent improvements in the field—namely, the introduction of fluorophores with appropriate long lifetimes, the development of new approaches for reporting enzyme activity through changes in fluorescence lifetime, and the availability of new FLT readers—have provided the setting for kinase, phosphatase, protease, and most recently HMT FLEXYTE FLT assays.

In this study, reference compounds that are known to bind to the SAM binding site showed IC₅₀ values comparable to the radiometric FlashPlate assay. However, the primary hit compounds showed an increased IC₅₀ in the FLT assay, showing that the sensitivity to FlashPlate hit compounds was reduced. Different molar concentrations of enzyme and different substrates used in the assays might partly explain the differences in the observed IC₅₀ values. Although the full-length histone H3.1 has been used in the radiometric assay, a histone H3.1 peptide substrate has been used in the FLT assay. To further elucidate the differences in the IC₅₀ results of the hit compounds, several studies will be performed. Alteration of the peptide substrate used in the FLT assay and competition studies with SAM as well as substrate will be employed. It can be postulated that binding of the hit compounds is dependent on the substrate, and IC₅₀ values are therefore modulated by the use of a specific substrate. Similar differences have been described by Liu et al.¹⁷ for the assumed selectivity of UNC0224. When assay conditions were adapted and identical assay conditions were used, a shift of IC₅₀ has been monitored. Our results support the practice that the development of orthogonal and complementary biochemical assays should be included from the start of a hit-finding campaign to rule out false positives and artifacts that can arise from using only one assay technology. BioFocus is currently expanding the panel of epigenetics assay formats; for example, MSA assays for HMTs and HDMs are in development.

Selectivity Panel

Selectivity and specificity of HMT inhibitors are key requirements for future therapeutic use. Nonselective inhibition of HMTs by sinefungin and product inhibition by SAH are well documented. Selectivity for G9a has been described for BIX-01294 and UNC0321.^16,17 Drug-like and highly selective HMT inhibitors were recently described for the PRMT CARM1.¹⁵ Selectivity panels reflecting the structural diversity of substrate and SAM-binding pockets and the therapeutically relevant selectivity profile are of key importance for drug discovery efforts. Compound profiling against structural similar and nonrelated members of an HMT target provides important additional information on target selectivity, and efforts are ongoing at BioFocus to set up an HMT compound profiling assay panel. Target class– specific consensus binding site models, referred to as roadmaps,^71,72 have been instrumental chemogenomics-driven approaches to address these challenges for protein kinases, G-protein-coupled receptors (GPCRs), phosphodiesterases (PDEs), and ion channels. We have recently extended this concept to HMTs, along with the development of a new computational tool referred to as EpiRoadmap. Such developments allow us to substantially improve the design of selectivity panels and support medicinal chemistry efforts in hit-to-lead programs.⁷³

Cellular Epigenetic Assays for Compound Profiling

Clearly, there is a demand for cellular readouts analyzing the methylation status of histones and target proteins of HMTs as well as transcriptional regulation of downstream targets. The cellular activity of G9a inhibitors has been shown to result in reduction of H3K9 and p53 K373 methylation. Chaetocin displayed depletion of cellular H3K9me2/3 levels but at the same time caused relatively high cell-based toxicity independent from its methyltransferase inhibitory activity. BIX-01294, as a non-SAM-competitive inhibitor of G9a with an IC₅₀ of 1.6 µM, has been shown to reduce H3K9me2 levels in mammalian cells. No cellular activity has been reported for the most potent G9a inhibitor, UNC0321, so far.

Cell-based activity of the primary screening hits within the described work has not yet been demonstrated. It is planned to evaluate the methylation events on both histone H3.1 as well as p53. The methylation of p53 and H3.1 will be monitored in parallel to overall p53 and H3.1 levels using specific antibodies. The analysis of the methylation levels of lysine 373 of p53 is of specific interest because the dimethylation of p53 K373 results in inactivation of p53, which is implicated in more than 50% of all cancers.⁷⁴

In the current case study with G9a, it has been demonstrated that the use of a combination of virtual screening methods can efficiently enrich the hit rate even in a small screening set. Each individual VS method has been proven to generate value given that all complementary methods applied have led to validated hits. The small overlap between the VS hit subsets and hit subset clusters underpins the previously mentioned notion that parallel application of various VS methods increases the likelihood of finding more potent hits. To us, this suggests that as many VS methods as possible should be combined for a VS campaign, provided that each individual method appears to be appropriate in the context of data and structures available for a particular project. For the screening of the compounds, a FlashPlate assay has been used, and a number of hit compounds have been verified in dose–response experiments. FLT has been employed for the first time as a novel orthogonal assay technology for HMTs. Although we were able to demonstrate highly similar IC₅₀ values for SAM-binding site inhibitors using both assay technologies, a larger difference in potency was measured for the remaining hit compounds, giving rise to speculation about the mechanism of G9a inhibition for this group and the binding site involved. Overall, the study demonstrates that complementary virtual screening methods can be successfully applied to epigenetics targets, and novel hit compounds can be identified from a small screening deck that is highly enriched with active compounds.

Footnotes

Acknowledgements

The authors acknowledge the valuable contributions of Kate Hilyard and Stephan Fasler (for helpful suggestions and reviewing of the manuscript); Virginie Voegtlin, Jasmin Grählert, and Céline Klein (excellent technical assistance); Vivienne Allen for support with docking calculations; Alexei Ouspenski for LC/MS; and Almac for providing the FLEXYTE peptides and support for FLT assay development.

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Supplementary material for this article is available on the Journal of Biomolecular Screening Web site at .

References

Goldberg

A. D.

Allis

C. D.

Bernstein

Epigenetics: A Landscape Takes Shape. Cell 2007, 128, 635–638.

Strahl

B. D.

Allis

C. D.

The Language of Covalent Histone Modifications. Nature 2000, 403, 41–45.

Gardner

K. E.

Allis

C. D.

Strahl

B. D.

OPERating ON Chromatin, a Colorful Language Where Context Matters. J. Mol. Biol. 2011, 409, 36–46.

Martin

Zhang

The Diverse Functions of Histone Lysine Methylation. Nat. Rev. Mol. Cell Biol. 2005, 6, 838–849.

Jones

P. A.

Baylin

S. B.

The Epigenomics of Cancer. Cell 2007, 128, 683–692.

Wilson

C. B.

Rowell

Sekimata

Epigenetic Control of T-Helper-Cell Differentiation. Nat. Rev. Immunol. 2009, 9, 91–105.

Tsankova

Renthal

Kumar

Nestler

E. J.

Epigenetic Regulation in Psychiatric Disorders. Nat. Rev. Neurosci. 2007, 8, 355–367.

Simon

J. A.

Lange

C. A.

Roles of the EZH2 Histone Methyltransferase in Cancer Epigenetics. Mutat. Res. 2008, 647, 21–29.

Jansson

Durant

S. T.

Cho

E. C.

Sheahan

Edelmann

Kessler

La Thangue

N. B.

Arginine Methylation Regulates the p53 Response. Nat. Cell Biol. 2008, 10, 1431–1439.

10.

Schneider

Bannister

A. J.

Kouzarides

Unsafe SETs: Histone Lysine Methyltransferases and Cancer. Trends Biochem. Sci. 2002, 27, 396–402.

11.

Smith

B. C.

Denu

J. M.

Chemical Mechanisms of Histone Lysine and Arginine Modifications. Biochim. Biophys. Acta 2009, 1789, 45–57.

12.

Copeland

R. A.

Solomon

M. E.

Richon

V. M.

Protein Methyltransferases as a Target Class for Drug Discovery. Nat. Rev. Drug Discov. 2009, 8, 724–732.

13.

Vedel

Lawrence

Robert-Gero

Lederer

The Antifungal Antibiotic Sinefungin as a Very Active Inhibitor of Methyltransferases and of the Transformation of Chick Embryo Fibroblasts by Rous Sarcoma Virus. Biochem. Biophys. Res. Commun. 1978, 85, 371–376.

14.

Spannhoff

Heinke

Bauer

Trojer

Metzger

Gust

Schüle

Brosch

Sippl

Jung

Target-Based Approach to Inhibitors of Histone Arginine Methyltransferases. J. Med. Chem. 2007, 50, 2319–2325.

15.

Allan

Manku

Therrien

Nguyen

Styhler

Robert

M. F.

Goulet

A. C.

Petschner

A. J.

Rahil

Robert Macleod

. N-Benzyl-1-Heteroaryl-3-(trifluoromethyl)-1H-Pyrazole-5-Carboxamides as Inhibitors of Co-activator Associated Arginine Methyltransferase 1 (CARM1). Bioorg. Med. Chem. Lett. 2009, 19, 1218–1223.

16.

Kubicek

O’Sullivan

R. J.

August

E. M.

Hickey

E. R.

Zhang

Teodoro

M. L.

Rea

Mechtler

Kowalski

J. A.

Homon

C. A.

. Reversal of H3K9me2 by a Small-Molecule Inhibitor for the G9a Histone Methyltransferase. Mol. Cell 2007, 25, 473–481.

17.

Liu

Chen

Allali-Hassani

Quinn

A. M.

Wasney

G. A.

Dong

Barsyte

Kozieradzki

Senisterra

Chau

. Discovery of a 2,4-Diamino-7-Aminoalkoxyquinazoline as a Potent and Selective Inhibitor of Histone Lysine Methyltransferase G9a. J. Med. Chem. 2009, 52, 7950–7953.

18.

Liu

Chen

Allali-Hassani

Quinn

A. M.

Wigle

T. J.

Wasney

G. A.

Dong

Senisterra

Chau

Siarheyeva

. Protein Lysine Methyltransferase G9a Inhibitors: Design, Synthesis, and Structure Activity Relationships of 2,4-Diamino-7-Aminoalkoxy-Quinazolines. J. Med. Chem. 2010, 53, 5844–5857.

19.

Shinkai

Tachibana

H3K9 Methyltransferase G9a and the Related Molecule GLP. Genes Dev. 2011, 25, 781–788.

20.

Tachibana

Sugimoto

Nozaki

Ueda

Ohta

Ohki

Fukuda

Takeda

Niida

Kato

. G9a Histone Methyltransferase Plays a Dominant Role in Euchromatic Histone H3 Lysine 9 Methylation and Is Essential for Early Embryogenesis. Genes Dev. 2002, 16, 1779–1791.

21.

Patnaik

Chin

H. G.

Esteve

P. O.

Benner

Jacobsen

S. E.

Pradhan

Substrate Specificity and Kinetic Mechanism of Mammalian G9a Histone H3 Methyltransferase. J. Biol. Chem. 2004, 279, 53248–53258.

22.

Tachibana

Sugimoto

Fukushima

Shinkai

Set Domain-Containing Protein, G9a, Is a Novel Lysine-Preferring Mammalian Histone Methyltransferase with Hyperactivity and Specific Selectivity to Lysines 9 and 27 of Histone H3. J. Biol. Chem. 2001, 276, 25309–25317.

23.

Rathert

Dhayalan

Murakami

Zhang

Tamas

Jurkowska

Komatsu

Shinkai

Cheng

Jeltsch

Protein Lysine Methyltransferase G9a Acts on Non-Histone Targets. Nat. Chem. Biol. 2008, 4, 344–346.

24.

Sampath

S. C.

Marazzi

Yap

K. L.

Sampath

S. C.

Krutchinsky

A. N.

Mecklenbrauker

Viale

Rudensky

Zhou

M. M.

Chait

B. T.

. Methylation of a Histone Mimic within the Histone Methyltransferase G9a Regulates Protein Complex Assembly. Mol. Cell 2007, 27, 596–608.

25.

Feldman

Gerson

Fang

Zhang

Shinkai

Cedar

Bergman

G9a-Mediated Irreversible Epigenetic Inactivation of Oct-3/4 during Early Embryogenesis. Nat. Cell Biol. 2006, 8, 188–194.

26.

Gyory

Fejer

Seto

Wright

K. L.

PRDI-BF1 Recruits the Histone H3 Methyltransferase G9a in Transcriptional Silencing. Nat. Immunol. 2004, 5, 299–308.

27.

Boulias

Talianidis

Functional Role of G9a-Induced Histone Methylation in Small Heterodimer Partner-Mediated Transcriptional Repression. Nucleic Acids Res. 2004, 32, 6096–6103.

28.

Fang

Miao

Xiang

Ponugoti

Treuter

Kemper

J. K.

Coordinated Recruitment of Histone Methyltransferase G9a and Other Chromatin-Modifying Enzymes in SHP-Mediated Regulation of Hepatic Bile Acid Metabolism. Mol. Cell Biol. 2007, 27, 1407–1424.

29.

Roopra

Qazi

Schoenike

Daley

T. J.

Morrison

J. F.

Localized Domains of G9a-Mediated Histone Methylation Are Required for Silencing of Neuronal Genes. Mol. Cell 2004, 14, 727–738.

30.

Lee

D. Y.

Northrop

J. P.

Kuo

M. H.

Stallcup

M. R.

Histone H3 Lysine 9 Methyltransferase G9a Is a Transcriptional Coactivator for Nuclear Receptors. J. Biol. Chem. 2006, 281, 8476–8485.

31.

Chen

M. W.

Hua

K. T.

Kao

H. J.

Chi

C. C.

Wei

L. H.

Johansson

Shiah

S. G.

Chen

P. S.

Jeng

Y. M.

Cheng

T. Y.

. H3K9 Histone Methyltransferase G9a Promotes Lung Cancer Invasion and Metastasis by Silencing the Cell Adhesion Molecule Ep-CAM. Cancer Res. 2010, 70, 7830–7840.

32.

Kondo

Shen

Ahmed

Boumber

Sekido

Haddad

B. R.

Issa

J. P.

Downregulation of Histone H3 Lysine 9 Methyltransferase G9a Induces Centrosome Disruption and Chromosome Instability in Cancer Cells. PLoS One 2008, 3, e2037.

33.

Wozniak

R. J.

Klimecki

W. T.

Lau

S. S.

Feinstein

Futscher

B. W.

5-Aza-2′-Deoxycytidine-Mediated Reductions in G9A Histone Methyltransferase and Histone H3 K9 Di-methylation Levels Are Linked to Tumor Suppressor Gene Reactivation. Oncogene 2007, 26, 77–90.

34.

McInnes

Virtual Screening Strategies in Drug Discovery. Curr. Opin. Chem. Biol. 2007, 11, 494–502.

35.

Glick

Jacoby

The Role of Computational Methods in the Identification of Bioactive Compounds. Curr. Opin. Chem. Biol. 2011, 15, 1–7.

36.

Ripphausen

Nisius

Peltason

Bajorath

Quo Vadis, Virtual Screening? A Comprehensive Survey of Prospective Applications. J. Med. Chem. 2010, 53, 8461–8467.

37.

Muegge

Oloff

Advances in Virtual Screening. Drug Discov. Today 2006, 3, 405–411.

38.

Rester

From Virtuality to Reality—Virtual Screening in Lead Discovery and Lead Optimization: A Medicinal Chemistry Perspective. Curr. Opin. Drug Discov. Dev. 2008, 11, 559–568.

39.

Langer

Hoffmann

Bryant

Lesur

Hit Finding: Towards ‘Smarter’ Approaches. Curr. Opin. Pharmacol. 2009, 9, 589–593.

40.

Clark

D. E.

What Has Virtual Screening Ever Done for Drug Discovery?

Exp. Opin. Drug Discov. 2008, 3, 841–851.

41.

Koppen

Virtual Screening: What Does It Give Us?

Curr. Opin. Drug Discov. Dev. 2009, 12, 397–407.

42.

Schneider

Virtual Screening: An Endless Staircase?

Nat. Rev. Drug Discov. 2010, 9, 273–276.

43.

Klebe

Virtual Ligand Screening: Strategies, Perspectives and Limitations. Drug Discov. Today 2006, 11, 580–594.

44.

Ripphausen

Nisius

Bajorath

State-of-the-Art in Ligand-Based Virtual Screening. Drug Discov. Today 2011, 16, 372–376.

45.

Eckert

Bajorath

Molecular Similarity Analysis in Virtual Screening: Foundations, Limitations and Novel Approaches. Drug Discov. Today 2007, 12, 225–233.

46.

Leach

A. R.

Gillet

V. J.

Lewis

R. A.

Taylor

Three-Dimensional Pharmacophore Methods in Drug Discovery. J. Med. Chem. 2010, 53, 539–558.

47.

Min

Lunin

V. V.

Antoshenko

Dombrovski

Zeng

Allali-Hassani

Campagna-Slater

Vedadi

Arrowsmith

C. H.

. Structural Biology of Human H3K9 Methyltransferases. PLoS One 2010, 5, e8570.

48.

Maltman

B. A.

Dunsmore

C. J.

Couturier

S. C.

Tirnaveanu

A. E.

Delbederi

McMordie

R. A.

Naredo

Ramage

Cotton

9-Aminoacridine Peptide Derivatives as Versatile Reporter Systems for Use in Fluorescence Lifetime Assays. Chem. Commun. 2010, 46, 6929–6931.

49.

Wigle

T. J.

Provencher

L. M.

Norris

J. L.

Jin

Brown

P. J.

Frye

S. V.

Janzen

W. P.

Accessing Protein Methyltransferase and Demethylase Enzymology Using Microfluidic Capillary Electrophoresis. Chem. Biol. 2010, 17, 695–704.

50.

Rarey

Dixon

J. S.

Feature Trees: A New Molecular Similarity Measure Based on Tree Matching. J. Comput. Aided Mol. Des. 1998, 12, 471–490.

51.

Birchall

Gillet

V. J.

Willett

Ducrot

Luttmann

Use of Reduced Graphs to Encode Bioisosterism for Similarity-Based Virtual Screening. J. Chem. Inf. Model 2009, 49, 1330–1346.

52.

Hert

Willett

Wilton

D. J.

Acklin

Azzaoui

Jacoby

Schuffenhauer

Enhancing the Effectiveness of Similarity-Based Virtual Screening Using Nearest-Neighbor Information. J. Med. Chem. 2005, 48, 7049–7054.

53.

Bemis

G. W.

Murcko

M. A.

The Properties of Known Drugs: 1. Molecular Frameworks. J. Med. Chem. 1996, 39, 2887–2893.

54.

Wagener

Lommerse

J. P.

The Quest for Bioisosteric Replacements. J. Chem. Inf. Model 2006, 46, 677–685.

55.

Berman

H. M.

Westbrook

Feng

Gilliland

Bhat

T. N.

Weissig

Shindyalov

I. N.

Bourne

P. E.

The Protein Data Bank. Nucleic Acids Res. 2000, 28, 235–242.

56.

Bergner

Günther

Hendlich

Klebe

Verdonk

Use of Relibase for Retrieving Complex Three-Dimensional Interaction Patterns Including Crystallographic Packing Effects. Biopolymers 2001, 61, 99–110.

57.

Friesner

R. A.

Banks

J. L.

Murphy

R. B.

Halgren

T. A.

Klicic

J. J.

Mainz

D. T.

Repasky

M. P.

Knoll

E. H.

Shaw

D. E.

Shelley

. Glide: A New Approach for Rapid, Accurate Docking and Scoring: 1. Method and Assessment of Docking Accuracy. J. Med. Chem. 2004, 47, 1739–1749.

58.

Halgren

T. A.

Murphy

R. B.

Friesner

R. A.

Beard

H. S.

Frye

L. L.

Pollard

W. T.

Banks

J. L.

Glide: A New Approach for Rapid, Accurate Docking and Scoring: 2. Enrichment Factors in Database Screening. J. Med. Chem. 2004, 47, 1750–1759.

59.

Krammer

Kirchhoff

P. D.

Jiang

Venkatachalam

C. M.

Waldman

LigScore: A Novel Scoring Function for Predicting Binding Affinities. J. Mol. Graph Model 2005, 23, 395–407.

60.

Muegge

Martin

Y. C.

A General and Fast Scoring Function for Protein-Ligand Interactions: A Simplified Potential Approach. J. Med. Chem. 1999, 42, 791–804.

61.

Muegge

PMF Scoring Revisited. J. Med. Chem. 2006, 49, 5895–5902.

62.

Jain

A. N.

Surflex: Fully Automatic Flexible Molecular Docking Using a Molecular Similarity-Based Search Engine. J. Med. Chem. 2003, 46, 499–511.

63.

Dhayalan

Dimitrova

Rathert

Jeltsch

A Continuous Protein Methyltransferase (G9a) Assay for Enzyme Activity Measurement and Inhibitor Screening. J. Biomol. Screen. 2009, 14, 1129–1133.

64.

Zhang

J. H.

Chung

T. D.

Oldenburg

K. R.

A Simple Statistical Parameter for Use in Evaluation and Validation of High Throughput Screening Assays. J. Biomol. Screen. 1999, 4, 67–73.

65.

Reynolds

C. H.

Tounge

B. A.

Bembenek

S. D.

Ligand Binding Efficiency: Trends, Physical Basis, and Implications. J. Med. Chem. 2008, 51, 2432–2438.

66.

Rea

Eisenhaber

O’Carroll

Strahl

B. D.

Sun

Z. W.

Schmid

Opravil

Mechtler

Ponting

C. P.

. Regulation of Chromatin Structure by Site-Specific Histone H3 Methyltransferases. Nature 2000, 406, 593–599.

67.

Best

J. D.

Carey

Epigenetic Therapies for Non-Oncology Indications. Drug Discov. Today 2010, 15, 1008–1014.

68.

Spannhoff

Sippl

Jung

Cancer Treatment of the Future: Inhibitors of Histone Methyltransferases. Int. J. Biochem. Cell Biol. 2009, 41, 4–11.

69.

Okada

Feng

Lin

Jiang

Coffield

V. M.

Zhang

hDOT1L Links Histone Methylation to Leukemogenesis. Cell 2005, 121, 167–178.

70.

Comley

Assay Interference a Limiting Factor in HTS?

Drug Discov. World 2003, Summer, 91–98.

71.

Harris

C. J.

Stevens

A. P.

Chemogenomics: Structuring the Drug Discovery Process to Gene Families. Drug Discov. Today 2006, 11, 880–888.

72.

Crossley

Slater

A Reductionist Approach to Chemogenomics in the Design of Drug Molecules and Focused Libraries. Chemogenomics 2006, 4, 85–108.

73.

Bergner

Sheppard

Slater

EpiRoadmap: Integration of Chemogenomics-Based Approaches for Epigenetic Targets. SBS Conference 2011 Poster M200 2011.

74.

Huang

Dorsey

Chuikov

Perez-Burgos

Zhang

Jenuwein

Reinberg

Berger

S. L.

G9a and Glp Methylate Lysine 373 in the Tumor Suppressor p53. J. Biol. Chem. 2010, 285, 9636–9641.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.18 MB

Efficient Hit-Finding Approaches for Histone Methyltransferases

Abstract

Keywords

Introduction

Materials and Methods

Biochemical Assays

FlashPlate assay

Fluorescence lifetime assay

Computational Compound Selection

Ligand-based VS: pharmacophore graphs

Ligand-based VS: integrated chemoinformatics-based toolkit

Structure-based VS: docking and postprocessing

Results

G9a Assay Development

Virtual Screening Approach and Results

Radiometric G9a Screen

Compound Selection for Dose–Response Experiments

Discussion

HMT Targets in Drug Discovery

Value of Orthogonal Assay/New Developments/Assay Technologies

Selectivity Panel

Cellular Epigenetic Assays for Compound Profiling

Footnotes

Acknowledgements

References

Supplementary Material