Fragment-Based Discovery and Optimization of Enzyme Inhibitors by

Sep 20, 2017 - Structure-guided exploration of commercial chemical space using molecular docking gives access to fragment libraries that are several o...
2 downloads 0 Views 2MB Size
Article pubs.acs.org/jmc

Fragment-Based Discovery and Optimization of Enzyme Inhibitors by Docking of Commercial Chemical Space Axel Rudling,†,∥ Robert Gustafsson,†,∥ Ingrid Almlöf,‡ Evert Homan,‡ Martin Scobie,‡ Ulrika Warpman Berglund,‡ Thomas Helleday,‡ Pål Stenmark,† and Jens Carlsson*,§ †

Department of Biochemistry and Biophysics, Stockholm University, SE-106 91 Stockholm, Sweden Science for Life Laboratory, Department of Medical Biochemistry and Biophysics, Karolinska Institutet, Box 1031, SE-171 21 Solna, Sweden § Science for Life Laboratory, Department of Cell and Molecular Biology, BMC, Uppsala University, Box 596, SE-751 24 Uppsala, Sweden ‡

S Supporting Information *

ABSTRACT: Fragment-based lead discovery has emerged as a leading drug development strategy for novel therapeutic targets. Although fragment-based drug discovery benefits immensely from access to atomic-resolution information, structure-based virtual screening has rarely been used to drive fragment discovery and optimization. Here, molecular docking of 0.3 million fragments to a crystal structure of cancer target MTH1 was performed. Twenty-two predicted fragment ligands, for which analogs could be acquired commercially, were experimentally evaluated. Five fragments inhibited MTH1 with IC50 values ranging from 6 to 79 μM. Structure-based optimization guided by predicted binding modes and analogs from commercial chemical libraries yielded nanomolar inhibitors. Subsequently solved crystal structures confirmed binding modes predicted by docking for three scaffolds. Structure-guided exploration of commercial chemical space using molecular docking gives access to fragment libraries that are several orders of magnitude larger than those screened experimentally and can enable efficient optimization of hits to potent leads.



INTRODUCTION Fragment-based techniques are increasingly being integrated into the drug discovery process. Several drug candidates identified using fragment-based lead discovery (FBLD) are currently being evaluated in clinical trials, and two recently approved drugs for treatment of cancer, vemurafenib (2011) and venetoclax (2016), emerged from FBLD programs.1,2 A compilation of a large number of successful examples of FBLD from 2015 highlighted the efficiency of this technique and its widespread use in the pharmaceutical industry.2 A fragment is defined as an organic molecule with low molecular weight, corresponding to approximately half the size of a drug candidate. Conceptually, FBLD can be explained based on the nature of chemical space and the theory of molecular complexity.3−5 First, the fact that the number of possible molecules is lower for smaller compounds makes it feasible to cover a larger fraction of the chemical space spanned by fragments in a screening library than will ever be possible in libraries containing drug-like molecules.4 Second, as fragments are small and have lower molecular complexity than drug-like compounds,3,5 they are more likely to fit in a binding site and form high-quality polar interactions. Consequently, fragment screening frequently yields hit rates of ∼5% or higher,4 which is © 2017 American Chemical Society

several orders of magnitude higher than HTS. Fragments optimally complement subpockets of the target binding site but typically bind weakly due to their limited size. In the second step of FBLD, hits from fragment screening are optimized to potent leads by iterative rounds of synthesis, which is greatly benefited by structure-guided design.1,4,6 During this process, molecular weight and complexity are generally increased, and by careful monitoring of changes in ligand efficiency (LE, the free energy of binding per heavy atom7,8) for synthesized compounds, highly potent leads can be obtained as each added moiety contributes favorably to affinity. Although X-ray crystallography has become one of the cornerstones of FBLD, computational structure-based techniques such as molecular docking have rarely been used as a tools for fragment screening.4 Indeed, the use of structure-based virtual screening to predict fragment complexes has been questioned9 as it has been shown that docking scoring functions often are unable to identify the correct binding mode10 and rank fragments by affinity.11 If fragment-protein complexes could be predicted accurately, molecular docking has Received: July 17, 2017 Published: September 20, 2017 8160

DOI: 10.1021/acs.jmedchem.7b01006 J. Med. Chem. 2017, 60, 8160−8169

Journal of Medicinal Chemistry

Article

Figure 1. Binding site of MTH1 and predicted binding modes for discovered fragment ligands. (A) Crystal structure (PDB code 4N1U17) of MTH1 active site in complex with a lead-like inhibitor (carbon atoms in pale cyan) and (B−F) predicted binding modes (carbon atoms in yellow) for fragments 1, 2, 3, 4, and 5, respectively. The ligands are shown in sticks, whereas MTH1 is depicted as a ribbon with key binding site residues in sticks.

docking predictions. On the basis of the results from the prospective screen, the availability of analogs to fragments in commercial chemical space was analyzed.

the potential to search fragment databases that are orders of magnitude larger than those accessible by biophysical screening.12,13 As crystallization of weakly binding compounds in complex with the target protein can be challenging,14 the binding modes predicted by docking could also improve the efficiency of the optimization step by guiding analog design. The rapidly increasing size of commercial chemical space, which currently contains >120 million compounds,15 makes it possible for computational methods to uncover ligand chemotypes that are not present in the more size-limited empirical libraries and could also be a valuable source of analogs during optimization.16 In this work, the use of structure-based virtual screening to discover and optimize fragment ligands was applied to MTH1, an enzyme that is part of the DNA-repair machinery of the cell.17 MTH1 catalyzes hydrolysis of oxidized nucleotides (8oxo-dGTP and 2-OH-dATP) and thereby prevents incorporation of these into DNA. MTH1 is an interesting target that has been linked to cancer cell survival by several groups,17−24 while others have reported no essential role of MTH1 in cancer survival.25,26 The role of MTH1 is thus an area of intense scientific interest, and new small molecule ligands can provide useful tools to uncover its underlying biology. Due to the recent discovery of the target, few inhibitors were available at the outset of this work. Access to crystal structures of MTH1 enabled us to use an in silico fragment-based approach to discover novel inhibitors. To this end, 0.3 million commercially available fragments were screened against the active site of MTH1 using molecular docking. In order to facilitate fragment optimization, a computational approach to identify analogs from commercial chemical space was developed by integrating 2D similarity searches and scoring of binding site complementarity with docking. A set of 22 compounds from the screening library was selected for experimental evaluation, which resulted in the discovery of fragment inhibitors that were optimized based on predicted binding modes. Subsequent determination of high-resolution crystal structures for several of the discovered ligands allowed us to assess the accuracy of the



RESULTS In Silico Fragment Screening of Commercial Chemical Space for MTH1 Inhibitors. Molecular docking against MTH1 was performed to explore opportunities for computational discovery and optimization of fragment ligands. Access to analogs of screening hits has emerged as an increasingly important criterion in the design of fragment libraries and has been addressed by including fragments that can be progressed via simple chemical transformations.4 An alternative strategy explored in this work was to exploit the fact that millions of potential analogs are already available commercially. The prospective screen against MTH1 hence required virtual screening of a large fragment library and the development of a method to identify analogs in commercial chemical space. The virtual screen was carried out using DOCK3.627,28 against a crystal structure of MTH1 that had been determined in complex with an inhibitor (N4-cyclopropyl-6-(2,3dichlorophenyl)pyrimidine-2,4-diamine, TH588, Figure 1A).17 A chemical library was prepared by extracting all commercially available compounds in the ZINC fragment-like database with 0.8 were identified. In addition, an analog was only allowed to have up to six more HAs than the parent fragment. If larger substituents were permitted, changes in the predicted binding mode of the scaffold were frequently observed, which is undesirable in optimization. The number of analogs to the 5000 top-ranked fragments ranged from none to several thousand with a median of 150 per compound. In order to investigate the complementary of the analogs to the binding site, these 118 421 unique compounds were screened by docking against the MTH1 crystal structure. Analogs with better or comparable docking scores, corresponding to 0.8.

The five most potent fragment inhibitors represented four different scaffolds based on their 2D structures. Fragments 3, 4, and 5 had unique scaffolds, whereas fragments 1 and 2 shared the same purine core. However, although fragments 1 and 2 were both based on a purine scaffold, it should be noted that the purine rings of these two compounds (Figure 1B and Figure 1C) were predicted to bind in different orientations compared to the crystal structure of MTH1 in complex with a nucleotide,34 a point that will be returned to. Fragments 3 and 4 were the most novel inhibitors, and in particular, the benzoxazole core of the latter compound was unlike both the ribonucleotide substrates and previously characterized inhibitors (Figure 1E). Finally, fragment 5 was based on a quinazoline scaffold that represented an attractive starting point for lead generation given its high LE (Figure 1F). Fragment Optimization and Crystal Structure Determination. Structure determination for the screening hits was pursued as atomic level information regarding binding modes is essential for efficient fragment optimization. However, as crystallization efforts for MTH1-fragment complexes were initially unsuccessful, the optimization step was entirely based on the predicted binding poses. Fragment 1 had an IC50 of 79 μM, which corresponded to an LE of 0.44 kcal mol−1 HA−1. Of the 56 available analogs (Table 1) in commercial chemical space, 22 passed the energy criterion and were predicted to bind in the same orientation as fragment 1. A set of seven compounds with diverse substituents in the 6position of the purine ring was finally purchased (1a−g, Table S3). It should be noted that none of these molecules were superstructures of fragment 1 but were identified as analogs based on their high 2D similarity. All the selected analogs were MTH1 inhibitors and the most potent compound (1a) had an IC50 of 170 nM, a 470-fold improvement over fragment 1 and an increased LE to 0.50 kcal mol−1 HA−1 (Figure S2). Crystallization of fragment 1 in complex MTH1 was unsuccessful, but a structure of the complex with 1a was solved at 1.85 Å resolution (Table S4 and Figure S3). There was a close agreement between the predicted binding mode for 8162

DOI: 10.1021/acs.jmedchem.7b01006 J. Med. Chem. 2017, 60, 8160−8169

Journal of Medicinal Chemistry

Article

Figure 2. Structures of optimized inhibitors with the substructure corresponding to the common atoms with the fragment colored in blue together with comparisons of predicted and experimentally determined binding modes. MTH1 is depicted as a ribbon with key binding site residues in sticks. The ligands are shown in sticks with cyan and yellow carbon atoms for the experimentally determined and predicted binding modes, respectively. (A) Compound 1a. The crystal structure of MTH1 in complex with compound 1a and the predicted binding mode of fragment 1 are shown. (B) Compound 2a. The crystal structure of MTH1 in complex with fragment 2 and the predicted binding mode of the same compound are shown. (C) Compound 4a. The crystal structure of MTH1 in complex with compound 4b and the predicted binding mode of fragment 4 are shown. (D) Compound 23, which was identified by Kettle et al.,26 is a superstructure of fragment 5. The crystal structure of compound 23 (PDB code 5ANW26) and predicted binding mode of fragment 5 are shown. ΔLE represents the difference in LE value between the optimized lead and the fragment hit.

Fragment 4 had an IC50 of 23 μM and an LE of 0.54 kcal mol−1 HA−1. As the weakly active compounds 6 and 11 had the same binding mode, the 123 commercially available analogs to these compounds were considered in fragment optimization. Of these, 62 compounds were predicted to have a conserved binding mode, but a majority had the same number of heavy atoms as fragment 4 and limited diversity. Combined with the two moderately active fragments from the primary screen, nine analogs (4, 6, 11, and 4c−h) were experimentally evaluated in the series (Table S6). The limited structure−activity relationship (SAR) suggested that the substituents in the 5- and 7position of the benzoxazole were important for activity. On the basis of the predicted binding modes, the 5-methyl extended into a small hydrophobic pocket whereas the 7-position appeared amenable to substitution with aromatic groups. As no such compounds were commercially available, two compounds with either a phenyl (4a) or furyl (4b) substituent in the 7-position were designed and custom-synthesized by a vendor. Compound 4a was the most potent inhibitor with an IC50 of 120 nM (Table S6), which was 190-fold more potent than fragment 4 and resulted in an improved LE of 0.57 kcal mol−1 HA−1. Crystallization of fragment 4 in complex with MTH1 remained unsuccessful throughout the project, but a structure for 4b was solved at 1.50 Å resolution (Table S4 and Figure S3), which confirmed the predicted binding mode and interactions with MTH1 for this inhibitor (Figure 2C, RMSD = 0.9 Å for overlapping atoms of the parent fragment and 4b). Fragment 5 was a potent inhibitor considering its small size (IC50 of 5.6 μM) and had an impressive LE of 0.61 kcal mol−1 HA−1. A total of 104 commercially available analogs were available, and 22 of these were predicted to bind in the same orientation as fragment 5. The most interesting analogs required resynthesis by the vendor, so less priority was initially given to this scaffold. During the course of our work, Kettle et al.26 disclosed a potent MTH1 inhibitor that was a superstructure of fragment 5, and from this point, optimization was not further pursued. Encouragingly, the 1.40 Å crystal structure

fragment 1 and the corresponding substructure of compound 1a. The key polar interactions of the fragment were captured, and the RMSD for the common atoms of the two compounds was 0.6 Å (Figure 2A). Fragments 2 (IC50 = 24 μM, LE = 0.54 kcal mol−1 HA−1) and 3 (IC50 = 26 μM, LE = 0.64 kcal mol−1 HA−1) were considered together in the optimization step because they had the same predicted binding modes and key interactions with MTH1 (Figure 1C and Figure 1D). These compounds had 116 commercially available analogs that passed the energy criterion, of which 35 also had a conserved binding mode. A diverse set of 19 commercially available analogs of fragment 2, which were mainly substituted in the 8-position of the purine ring, were evaluated experimentally (Table S5). Fifteen of these compounds had lower or similar activity as fragment 2. The most potent inhibitor among the four analogs with enhanced potencies had a 7-fold improved IC50 of 3.5 μM and an LE of 0.59 kcal mol−1 HA−1 (Figure S2). Subsequent to the optimization step, a structure of MTH1 in complex with fragment 2 was solved at 2.20 Å resolution (Table S4 and Figure S3). The experimentally determined binding mode for fragment 2 was rotated 90° compared to the predicted pose (RMSD = 4.1 Å), which was due to a reoriented side chain of Asn33 (Figure 2B). In addition, the hydrogen bonding network suggested altered protonation states for Asp119 and Asp120 compared to the crystal structure used in the docking screen. Redocking of fragment 2 to the determined crystal structure reproduced the experimental binding mode (RMSD = 0.3 Å), demonstrating that the discrepancy between the experimental data and the prediction was due to the protein conformational change (Figure S4). Rescreening of analogs against the new crystal structure revealed that most analogs selected based on the incorrectly predicted binding mode did not fit into the structure in the experimentally determined orientation, providing an explanation to the small improvement achieved in optimization. 8163

DOI: 10.1021/acs.jmedchem.7b01006 J. Med. Chem. 2017, 60, 8160−8169

Journal of Medicinal Chemistry

Article

Figure 3. Availability of analogs to fragments in commercial chemical space. (A) Summary of optimization against MTH1 for the five most potent hits. The bars represent the average number of analogs to the five fragments with error bars that show minimum and maximum number of analogs. The analogs were first identified using a similarity search and subsequently filtered based on docking energy and predicted binding mode. The last bar represents the average number of analogs that were experimentally evaluated. (B) Analysis of the number of available analogs to fragments in commercial chemical space. The bars represent the number of fragments that have more than a certain number of analogs based on a Tversky similarity index of >0.8 (red) or substructure search (blue).

of MTH1 in complex with the inhibitor (PDB code 5ANW26) confirmed the predicted binding mode for fragment 5 (Figure 2D, RMSD = 0.9 Å for overlapping atoms of the fragment and optimized lead). A detailed analysis of the fragment optimization step for MTH1 revealed that the number of relevant analogs in commercial chemical space decreased rapidly after taking docking energy and predicted binding modes into consideration. The number of analogs to the five most potent fragment inhibitors ranged from 56 to 246 with an average of 120, which decreased to 28 molecules per fragment after the two filtering steps (Figure 3A). Interestingly, 18 of the experimentally evaluated analogs were not superstructures of the screening hits, supporting the idea of using similarity instead of strict substructure search to identify analogs. Availability of Analogs to Fragments in Commercial Chemical Space. The availability of analogs to fragments was further explored by extending the similarity calculations to every compound in the initial screening library. The protocol developed for the MTH1 screen was used to identify analogs to the 0.66 million commercially available compounds in the ZINC fragment (MW < 250 Da) database with 10−18 HAs. In analogy to the MTH1 screen, the analogs were required to have a Tversky similarity index of >0.8 to the parent fragment and a maximum of six additional HAs. For comparison, the subset of analogs that represented superstructures of each fragment was also identified. On the basis of the 2.9 million fragment-toanalog comparisons, a median of three analogs per fragment that represented superstructures was identified in commercial chemical space. If the search criterion was relaxed to also include compounds with a similar scaffold, the median number of analogs was 700, which was more than a 200-fold increase compared to the substructure search. The complete mapping of optimization routes in commercial chemical space allowed the creation of fragment sets with compounds that had more than a specific number of analogs. The number of fragments with >N analogs (N = 1, 5, 10, 50, 500, 1000, 5000, and 10 000) based on similarity and substructure searches are shown in Figure 3B. If only analogs representing superstructures were included, ∼50 000 fragments (7.6% of the fragment library) had >100 analogs and ∼10 000 of these even had >500 analogs. In

contrast, the vast majority of the 0.66 million fragments (83%) had >100 analogs based on similarity, which decreased to 0.38 (57%) and 0.27 (42%) million fragments if >500 and >1000 analogs were required, respectively.



DISCUSSION AND CONCLUSIONS The major finding of this work is the observation that structurebased virtual screening could guide both discovery and optimization of fragments to potent inhibitors of a recently discovered drug target. Molecular docking of a large fragment library to a crystal structure of MTH1 led to the discovery of 12 compounds that inhibited enzyme activity, demonstrating the potential of virtual screening to identify starting points for fragment-to-lead generation. Two compounds were optimized to nanomolar inhibitors, which supports that FBLD can be guided by predicted modes and, in some cases, be accomplished entirely from within commercial chemical space. Structure-based virtual screening has rarely been employed in fragment screening campaigns, which is surprising considering that crystal structures are available in most studies.4 One of the concerns regarding the use of fragment docking has been the accuracy of the predicted binding modes.9 In the case of MTH1, experimental coordinates of either the hit from the virtual screen or an optimized analog were obtained for four of the five most potent fragments. In three cases, the experimental structure agreed remarkably well with the predicted binding modes. In the fourth case, the ligand induced a conformational change in the binding site, illustrating the need to incorporate protein flexibility in docking algorithms, which is an area of active research.35 The distinct binding modes for the purine rings of fragments 1 and 2 compared to the drug-sized compound 8-oxo-dGMP (PDB code 3RZ034) demonstrate that chemically similar fragments can bind in completely different orientations, which further emphasizes the importance of structural information for the success of FBLD. The high hit rates and fidelity of the predicted binding modes for three disparate chemotypes provide additional support for the use of molecular docking as a tool for fragment screening.36−38 Optimization of fragments not only relies on detailed information regarding binding modes but also is critically dependent on efficient design of analogs. In the case of MTH1, 8164

DOI: 10.1021/acs.jmedchem.7b01006 J. Med. Chem. 2017, 60, 8160−8169

Journal of Medicinal Chemistry

Article

our technique for fragment optimization was remarkably successful. For two of the fragments, 200- to 500-fold improvements of IC50 were achieved after experimental evaluation of less than 10 analogs predicted by molecular docking. In both cases, the LE values were also improved, which is a rare accomplishment in fragment-to-lead optimization.6 Compound 1a, a potent inhibitor with an IC50 of 170 nM, was identified from commercial chemical space, whereas optimization of the benzoxazole scaffold represented by fragment 4 required custom synthesis of two additional compounds to reach a similar potency. Given the fact that we were unable to crystallize fragments 1, 3, and 4 in complex with MTH1, the docking predictions played a crucial role in optimization. The possibility to use modeled binding modes to guide fragment optimization will be particularly valuable for proteins that are difficult to crystallize, e.g., G protein-coupled receptors.37,39−41 The advantages of our fragment-based approach for inhibitor discovery for MTH1 are well illustrated by comparing with the results of an HTS of 29 809 compounds that was performed by Kawamura et al., which were published during the preparation of this paper.25 Twenty MTH1 ligands were discovered from the HTS, corresponding to a hit rate of 0.06%. Lead optimization was then carried out for a hit with an IC50 value of 3.5 μM, and evaluation of 131 analogs led to the discovery of a nanomolar inhibitor, which had similarities to fragment 1 and its optimized analog 1a. In our work, only 22 compounds (>1000-fold fewer than the HTS library used by Kawamura et al.) were assayed in order to discover fragment 1, which was subsequently optimized through evaluation of less than 10 analogs to a similar potency as the inhibitor identified by Kawamura et al. It should be noted that the purine analogue (N6-phenethyl-9H-purine-2,6-diamine, NPD9948) discovered by Kawamura et al. had low cellular potency in HeLa cells (EC50 = 35 μM). In line with these data, our most potent compound from the same series also failed to show appreciable activity in cells. A few caveats of the strategy employed in this work should also be noted. First of all, optimization was unsuccessful for fragments 2 and 5. A modest 7-fold improvement of activity was achieved for fragment 2 although a large number of analogs were evaluated. This was traced to the fact that the binding mode of the fragment was not correctly predicted, which led us to pursue analogs that did not fit in the active site. A second factor that, from a practical viewpoint, limited the optimization of the remarkably potent fragment 5 was that upon sourcing analogs, the requested compounds were not available or prohibitively expensive. In fact, even for the successfully optimized fragments, it was surprising to find that the number of available analogs decreased dramatically, from hundreds to only a small set of compounds, after filtering for binding mode, docking score, commercial availability, and diversity. Another limitation is that by restricting the docking screen to fragments with commercially available analogs, the diversity of the library decreases and potentially interesting chemotypes that are only represented by few compounds in the library are not considered. One solution to this problem is to generate virtual compound databases with analogs to fragments that can rapidly be synthesized, followed by docking screens to select a subset of compounds for experimental evaluation. This approach was recently applied by Rodriguez et al. and Männel et al. to identify agonists of the adenosine and dopamine receptors.42,43

In light of the results obtained for MTH1, the choice of library for virtual fragment screening and subsequent optimization can be considered to be influenced by two opposing factors. On one hand, screening of fragments regardless of their availability or access to analogs will maximize chemical space coverage. This argument would favor screening of the 166 billion fragments that have been estimated to be theoretically possible with up 17 HAs.44 The obvious drawback of this approach is that the subsequent synthesis of these compounds will be very challenging. On the other hand, one could reduce the virtual screening library to commercially available fragments and, in addition, only to those with a large number of purchasable analogs. However, wouldn’t this severely restrict the size of the screening library? Our mapping of all optimization routes in commercial chemical space suggests that this is not necessarily the case. Even if the virtual screen is restricted to the set of molecules that have at least 500 analogs, which based on the MTH1 screen will be reduced to ∼100 purchasable molecules after filtering by docking, there are still ∼380 000 unique compounds in the library. As this set is almost 2 orders of magnitude larger than the empirical screening libraries that consistently produce hits in biophysical screens, it appears possible to both access a larger part of chemical space and take advantage of commercial libraries in optimization. Virtual screening of commercial chemical space hence provides the opportunity to rapidly identify fragment ligands and optimize these to potent leads, an approach that could contribute to more efficient drug development for numerous therapeutic targets.



METHODS

Molecular Docking Screens. Docking was performed using DOCK3.627,28,45 and the MTH1 crystal structure with PDB code 4N1U.17 DOCK3.6 uses a flexible ligand-sampling algorithm based on pregenerated conformers of the docked compound, which are evaluated by a physics-based scoring function. The score for each conformation was calculated as the sum of the receptor−ligand electrostatic and van der Waals interaction energies, corrected for ligand desolvation.29 The three-dimensional map of the electrostatic potential in the binding site was prepared using the program Delphi,46 and unless stated otherwise, partial charges of the amino acids were based on a united atom AMBER force field.47 The side chain dipole moment of Asn33, which interacted with ligands in the available crystal structures,17,34,48,49 was increased to favor hydrogen bonding to this residue without altering its formal charge. The partial charge for the side chain oxygen was reduced by 0.4 whereas the charges were increased by 0.2 for the two polar hydrogens connected to the side chain nitrogen. This technique has been used in several past virtual screening studies to improve binding mode predictions and enhance enrichment of ligands.37,40,42,50,51The protonation states of ionizable residues Asp, Glu, Arg, and Lys in the binding site were set to their most probable states in the receptor at pH 7. Histidine protonation states were set by visual inspection on the basis of the hydrogen bonding network. The program CHEMGRID45 was used to generate a van der Waals grid based on a united atom version of the AMBER force field.47 The ligand desolvation penalty was estimated from a precalculated transfer free energy of the molecule between solvents of dielectrics 78 and 2. The desolvation energy was obtained by weighting the transfer free energy with a scaling factor that reflects the degree of burial of the docked compound in the binding site.29 Ligand sampling was based on a set of matching spheres that are placed in the binding site. A total of 45 matching spheres was generated based on the position of the cocrystallized ligand and manually adjusted to focus the sampling to the region close to residues Asn33, Asp119, and Asp120. The degree of ligand sampling was determined from bin size, bin overlap, and distance tolerance, which were set to 0.4 Å, 0.2 Å, and 8165

DOI: 10.1021/acs.jmedchem.7b01006 J. Med. Chem. 2017, 60, 8160−8169

Journal of Medicinal Chemistry

Article

5,7-Dimethyl-1,3-benzoxazol-2-amine (4) was supplied by Enamine. H NMR (400 MHz, CD3OD) δ = 6.87 (s, 1H), 6.68 (s, 1H), 2.36 (s, 1H), 2.34 (s, 1H). LC−MS (ESI+) m/z = 163 [M + H]+. 5-Methyl-7-phenylbenzo[d]oxazol-2-amine (4a) was supplied by Enamine. 1H NMR (400 MHz, DMSO-d6, ppm) δ = 7.74 (d, J = 7.3 Hz, 2H), 7.44 (t, J = 7.6 Hz, 2H), 7.33 (t, J = 7.3 Hz, 1H), 7.19 (s, 2H), 6.93 (d, J = 8.8 Hz, 2H), 2.40 (s, 3H). HPLC−MS (positive mode) m/z = 225/226 [M + H]+. 7-(Furan-2-yl)-5-methylbenzo[d]oxazol-2-amine (4b) was supplied by Enamine. 1H NMR (500 MHz, CDCl3, ppm) δ = 7.53 (s, 1H), 7.31 (s, 1H), 7.10 (s, 1H), 6.90 (d, J = 3.1 Hz, 1H), 6.55 (d, J = 1.7 Hz, 1H), 5.21 (s, 2H), 2.44 (s, 3H). HPLC−MS (positive mode) m/z = 215/216 [M + H]+. 4-Methylquinazolin-2-amine (5) was supplied by NCI. 1H NMR (400 MHz, CD3OD) δ = 8.00 (m, 1H), 7.71 (m, 1H), 7.49 (m, 1H), 7.30 (m. 1H), 2.80 (s, 3H). LC−MS (ESI+) m/z = 160 [M + H]+. The procedure described by Gad et al.17 was used for the enzymatic screening assay. The assay applied for screening purposes was based on the enzymatic hydrolysis of dGTP by purified human recombinant MTH1 to form dGMP and pyrophosphate. An excess of inorganic pyrophosphatase was added to the assay, which allows the quantification of released inorganic phosphate as a measure of product levels in a coupled enzymatic assay. Inorganic phosphate was measured using an absorbance assay based on malachite green, as previously described.57 The assay buffer in which MTH1, inorganic pyrophosphatase, and dGTP were diluted consisted of 100 mM Tris-acetate at pH 8.0, 40 mM NaCl, 10 mM Mg acetate, 0.005% Tween 20, and 1 mM DTT. The final conditions in the assay during enzymatic incubation were 0.5−2 nM recombinant human MTH1, 100 μM dGTP, and 0.2 U mL−1 inorganic pyrophosphatase. The assay volumes were 100 μL and 40 mL in the 96-well and 384-well based assays, respectively. An Echo 550 (Labcyte) was used for dispensing of compound solutions, whereas a FlexDrop (PerkinElmer) was used for dispensing of enzyme and substrate solutions. After incubation of enzymes and substrate at room temperature for 1 h, the reaction was terminated and the signal developed by the addition of the malachite green reagent using a MultiDrop (Thermo Scientific). After shaking and color development for a minimum of 15 min the plates were analyzed in a microplate reader using a filter at 630 nm (Victor 2 and 3 instruments from PerkinElmer). MTH1 Expression and Purification. MTH1 expression construct has previously been described.34 MTH1 protein for crystallographic studies was expressed in BL21 (DE3; Novagen) by induction with 0.5 mM IPTG at an OD600 of 1.0 in Terrific Broth media and further growth at 20 °C for 21 h. Bacteria were harvested by centrifugation and lysed in Lysis buffer (100 nM HEPES, pH 8.0, 500 mM NaCl, 10% (v/v) glycerol, 0.5 mM TCEP) after treatment with lysozyme, 5 mM MgSO4, DNase, and protease inhibitor cocktail (Roche) using high-pressure homogenization followed by centrifugation at 185 000 rcf for 1 h at 4 °C. His-tagged MTH1 was purified on gravity flow column (Econo-Pac chromatography column, Bio-Rad) after incubation with Ni-NTA (3.0 mL/50 mL cleared lysate) and 10 mM imidazole, washed with buffer A (20 mM HEPES buffer, pH 7.5, 500 mM NaCl, 10% glycerol, and 0.5 mM TCEP), fortified with 10 mM imidazole followed by elution of the protein with buffer A fortified with 500 mM imidazole. MTH1 containing fractions were pooled and loaded on a PD-10 desalting column (GE Healthcare) equilibrated with buffer B (20 mM HEPES, pH 7.5, 300 mM NaCl, 10% glycerol, and 0.5 mM TCEP). Eluted protein was cleaved by thrombin (GE Healthcare) overnight at 4 °C, then purified using Ni-NTA with buffer B fortified with 10 mM imidazole. Fractions containing MTH1 were pooled, loaded on a Superdex 75 16/60 column (GE Healthcare), and separated using buffer B. Fractions containing MTH1 were pooled, and purity was analyzed on SDS−PAGE. Protein was concentrated using Vivaspin 20 (Sartorious Stedim), 10 kDa MWCO at 3750 rcf at 4 °C. Concentration was determined using a calculated extinction coefficient of 28 085 M−1 cm−1. IC50 Value Determination. The described screening assay, slightly modified, was also used for IC50 determination. The compounds to be analyzed were nanodispensed from two stock solutions (10 mM and

1.5 Å, respectively, for both matching spheres and the docked molecules. For the best scoring conformation of each docked molecule, 100 steps of rigid-body minimization were carried out. A total of 4.4 million in stock compounds from the ZINC database15 were screened and subsequently filtered based on HA count and similarity searches. The screening library was composed of 0.7 million fragment-like (2015-02-04 update of ZINC, molecular weight ≤ 250, log P ≤ 3.5, and rotatable bonds ≤ 5) and 3.7 million lead-like compounds (250 < molecular weight ≤ 350, log P ≤ 3.5, and rotatable bonds ≤ 7, 2014-06-25 update of ZINC). All compounds were prepared for docking using the ZINC database protocol.15 All illustrations of MTH1-ligand complexes were generated using PyMol.52 Molecular Similarity. RDKit (release 2014-03-01) was used for fingerprint and molecular similarity calculations.53 Two-dimensional fingerprints based on molecular descriptors and RDKits circular topology Morgan fingerprints with radius 2 and size 2048 bits were used.31,32 Compound similarities were calculated using the Tversky index, with the parameters α and β set to 1.0 and 0.0, respectively.30 The substructure searches were performed using the VF2 algorithm.54 Screening Assay. The predicted fragments and analogs were analyzed for similarities to pan assay interference compounds (PAINS). The Mobyle55 server (http://mobyle.rpbs.univ-parisdiderot.fr), which is based on the work of Baell et al.,56 was used, and none of the tested compounds contained known PAINS motifs. The tested compounds were obtained from Vitas-M, Enamine, Princeton Biomedical Research, Life Chemicals, Key Organics, Fluorochem, Specs, Maybridge, Sigma-Aldrich, and NCI (Table S1). Compound identity and purity for the five hits from the fragment screen and optimized analogs were confirmed by 1H NMR and LC− MS. 1H NMR spectra were recorded on a Bruker DRX-400 (compounds 1, 1a, 2, 2a, 3, 4, 5) and a Bruker AVANCE III 400 MHz (compounds 4a and b). Chemical shifts are expressed in parts per million (ppm) and referenced to the residual solvent peak (see below). For compounds 1, 1a, 2, 2a, 3, 4, and 5, analytical HPLC−MS was performed on an Agilent MSD mass spectrometer connected to an Agilent 1100 system: column ACE 3 C8 (50 mm × 3.0 mm); H2O (+0.1% TFA) and MeCN were used as mobile phases at a flow rate of 1 mL/min, with a gradient time of 3.0 min or column X-Terra MSC18 (50 mm × 3.0 mm); H2O (containing 10 mM NH4HCO3; pH = 10) and MeCN were used as mobile phases at a flow rate of 1 mL/min, with a gradient time of 3.0 min. For compounds 4a and 4b an Agilent 1100 series LC/MSD system with DAD/ELSD and Agilent LC/MSD VL (G1956A), SL (G1956B) mass spectrometer or Agilent 1200 series LC/MSD system with DAD/ELSD and Agilent LC/MSD SL (G6130A), SL (G6140A) mass spectrometer was used. Column Zorbax SB-C18 1.8 μm, 4.6 mm × 15 mm Rapid Resolution cartridge (PN 821975-932), acetonitrile, 0.1% formic acid, and water (0.1% formic acid) were used as mobile phases at a flow rate of 3 mL/min, with water gradient time of 0 min, 100%; 0.01 min, 100%; 1.5 min, 0%; 1.8 min 0% B; 1.81 min, 100%. The injection volume was 1 μL. All compounds were assessed to be >95% pure by HPLC−MS analysis. 8-Methyl-6-(methylsulfanyl)-7H-purin-2-amine (1) was supplied by NCI. 1H NMR (400 MHz, CD3OD) δ = 2.62 (s, 3H), 2.42 (s, 3H). LC−MS (ESI+) m/z = 196 [M + H]+. 6-[(2-Phenylethyl)sulfanyl]-7H-purin-2-amine (1a) was supplied by NCI. 1H NMR (400 MHz, DMSO-d6) δ = 7.36−7.20 (m, 5H), 3.49 (m, 2H), 2.98 (m, 2H). LC−MS (ESI+) m/z = 272 [M + H]+. 8-(Methylsulfanyl)-9H-purin-6-amine (2) was supplied by Specs. 1 H NMR (400 MHz, CD3OD) δ = 8.10 (s, 1H), 2.75 (s, 3H). LC−MS (ESI+) m/z = 182 [M + H]+. N8,N8-dimethyl-9H-purine-6,8-diamine (2a) was supplied by VitasM. 1H NMR (400 MHz, CD3OD) δ = 7.97 (s, 1H), 3.08 (s, 6H). LC− MS (ESI+) m/z = 179 [M + H]+. Imidazo[1,2-a]pyrazin-8-amine (3) was supplied by Enamine. 1H NMR (400 MHz, CD3OD) δ = 7.82 (d, J = 1.3 Hz, 1H), 7.77 (d, J = 4.7 Hz, 1H), 7.58 (d, J = 1.3 Hz, 1H), 7.22 (d, J = 4.7 Hz, 1H). LC− MS (ESI+) m/z = 135 [M + H]+.

1

8166

DOI: 10.1021/acs.jmedchem.7b01006 J. Med. Chem. 2017, 60, 8160−8169

Journal of Medicinal Chemistry



0.05 mM) in 11 concentrations, 1:3 dilution series, directly in 384-well assay plates using Echo liquid handler (Labcyte), giving a final DMSO concentration of ≤1%. A dilution series of a QC substance was included on each assay plate as well as controls lacking enzyme (negative control) or inhibitor (positive control). After preincubation with 4.8 nM MTH1, dGTP was added and the reaction mixture was incubated with shaking for 15 min at 22 °C. 10 μL/well of malachite green assay reagent was added followed by incubation with shaking for 15 min at 22 °C. The absorbance of the assay plate was read at 630 nm using a Hidex Sense plate reader. The IC50 value was determined by fitting a dose−response curve to the data points using nonlinear regression analysis and the equation Y = Ymin + (Ymax − Ymin)/(1 + 10((log IC50 − X) × Hill slope)), where Y is the read absorbance at 630 nm and X is log[compound]. For the most potent compounds the enzymatic assay reached its limit in terms of potency resolution due to tight binding conditions. Therefore, a more sensitive assay was developed with a final MTH1 concentration of 0.2 nM and a reaction time of 8 h. Crystallization and Structure Determination. 5 mM 1a, 2, or 4b and 6 mM MgCl2 were added to MTH1. Sitting drop vapor diffusion experiments at 20 °C were performed, and MTH1 (9.34 mg/ mL) was mixed with reservoir solution (30% (w/v) PEG6000, 100 mM sodium acetate, pH 3.7, and 200 mM LiSO4 for 2; 28% (w/v) PEG8000, 100 mM sodium acetate, pH 4.0, and 200 mM LiSO4 for 1a; 32% (w/v) PEG3350, 100 mM sodium acetate, pH 4.0, and 200 LiSO4 for 4b) in a 1:1 ratio (2 and 4b) or 3:5 ratio (1a). Diffraction quality crystals appeared after about 3 days, were cryoprotected by soaking them in mother liquor supplemented with 30% glycerol and 10 mM of the specified compound, and were subsequently flash-frozen in liquid nitrogen. Data collection was performed at 100 K at beamline ID23-2 at ESRF, France, for 2, at beamline I04-1 at Diamond, U.K., for 1a, and at beamline 14.1 at BESSY, Germany, for 4b. Data reduction and processing were carried out using XDS58,59 and programs from the CCP4 suite.60 The structures were solved via molecular replacement using Phaser,61 using the previously solved apo MTH1 structure as search model (PDB code 3ZR1). A few cycles of refinement in Refmac5,62,63 interspersed with manual building in Coot,64,65 were needed to complete the model. Water molecules were automatically placed in the maps, using a FO − FC Fourier difference map cutoff of 3σ, and were subsequently validated to ensure correct positioning. Crystal structures and predicted complexes were aligned using PyMOL52 with the structure used in the docking calculations as reference. The alignment was performed using the protein coordinates, and the RMSD values for the predicted fragment poses were calculated based on the corresponding atoms of the cocrystallized ligand. All structure figures were prepared using PyMOL,52 and Ramachandran statistics were generated using MolProbity.66,67 The structures have been deposited in the Protein Data Bank with accession codes 5NGR (2), 5NGS (1a), and 5NGT (4b).



Article

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. ORCID

Jens Carlsson: 0000-0003-4623-2977 Author Contributions ∥

A.R. and R.G. contributed equally to this work.

Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS We thank Erik Lindahl for critical reading of the manuscript. We thank the beamline scientists at BESSY, Germany; Diamond, England; ESRF, France; Max-Lab, Sweden; and the Swiss Light Source, Switzerland, for their support in data collection. This work was supported by the Swedish e-Science Research Center (J.C.), the Science for Life Laboratory (J.C), the Knut and Alice Wallenberg Foundation (T.H. and P.S.), the Wenner-Gren Foundation, Clas Groschinskys Foundation, Åke Wibergs Foundation (P.S.), the Göran Gustafsson Foundation (T.H. and J.C.), the Swedish Children’s Cancer Foundation, the Swedish Pain Relief Foundation, the Torsten and Ragnar Söderberg Foundation (T.H.), the Swedish Cancer Society (T.H. and P.S.) and the Swedish Research Council (T.H., P.S., and J.C.). Computational resources were provided by the Swedish National Infrastructure for Computing. We thank OpenEye Scientific Software for the use of OEChem and OMEGA at no cost.



ABBREVIATIONS USED FBLD, fragment-based lead discovery; HA, heavy atom; SAR, structure−activity relationship; PAINS, pan assay interference compounds



REFERENCES

(1) Erlanson, D. A.; Fesik, S. W.; Hubbard, R. E.; Jahnke, W.; Jhoti, H. Twenty Years On: The Impact of Fragments on Drug Discovery. Nat. Rev. Drug Discovery 2016, 15, 605−619. (2) Johnson, C. N.; Erlanson, D. A.; Murray, C. W.; Rees, D. C. Fragment-to-Lead Medicinal Chemistry Publications in 2015. J. Med. Chem. 2017, 60, 89−99. (3) Hann, M. M.; Leach, A. R.; Harper, G. Molecular Complexity and Its Impact on the Probability of Finding Leads for Drug Discovery. J. Chem. Inf. Comput. Sci. 2001, 41, 856−864. (4) Keseru, G. M.; Erlanson, D. A.; Ferenczy, G. G.; Hann, M. M.; Murray, C. W.; Pickett, S. D. Design Principles for Fragment Libraries: Maximizing the Value of Learnings from Pharma Fragment-Based Drug Discovery (Fbdd) Programs for Use in Academia. J. Med. Chem. 2016, 59, 8189−8206. (5) Leach, A. R.; Hann, M. M. Molecular Complexity and FragmentBased Drug Discovery: Ten Years On. Curr. Opin. Chem. Biol. 2011, 15, 489−496. (6) Ferenczy, G. G.; Keseru, G. M. How Are Fragments Optimized? A Retrospective Analysis of 145 Fragment Optimizations. J. Med. Chem. 2013, 56, 2478−2486. (7) Hopkins, A. L.; Groom, C. R.; Alex, A. Ligand Efficiency: A Useful Metric for Lead Selection. Drug Discovery Today 2004, 9, 430− 431. (8) Kuntz, I. D.; Chen, K.; Sharp, K. A.; Kollman, P. A. The Maximal Affinity of Ligands. Proc. Natl. Acad. Sci. U. S. A. 1999, 96, 9997− 10002. (9) Scott, D. E.; Coyne, A. G.; Hudson, S. A.; Abell, C. FragmentBased Approaches in Drug Discovery and Chemical Biology. Biochemistry 2012, 51, 4990−5003.

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jmedchem.7b01006. Additional data and methods (PDF) Molecular formula strings (CSV) Atomic coordinates of predicted binding modes of fragments 1−5 (ZIP) Accession Codes

Atomic coordinates of MTH1 bound to compounds 1a, 2, and 4b (PDB codes 5NGS, 5NGR, and 5NGT) have been deposited in the Protein Data Bank. Authors will release atomic coordinates and experimental data upon article publication. 8167

DOI: 10.1021/acs.jmedchem.7b01006 J. Med. Chem. 2017, 60, 8160−8169

Journal of Medicinal Chemistry

Article

(10) Verdonk, M. L.; Giangreco, I.; Hall, R. J.; Korb, O.; Mortenson, P. N.; Murray, C. W. Docking Performance of Fragments and Druglike Compounds. J. Med. Chem. 2011, 54, 5422−5431. (11) Sandor, M.; Kiss, R.; Keseru, G. M. Virtual Fragment Docking by Glide: A Validation Study on 190 Protein-Fragment Complexes. J. Chem. Inf. Model. 2010, 50, 1165−1172. (12) Barelier, S.; Eidam, O.; Fish, I.; Hollander, J.; Figaroa, F.; Nachane, R.; Irwin, J. J.; Shoichet, B. K.; Siegal, G. Increasing Chemical Space Coverage by Combining Empirical and Computational Fragment Screens. ACS Chem. Biol. 2014, 9, 1528−1535. (13) Chen, D.; Ranganathan, A.; Ijzerman, A. P.; Siegal, G.; Carlsson, J. Complementarity between in Silico and Biophysical Screening Approaches in Fragment-Based Lead Discovery against the a(2a) Adenosine Receptor. J. Chem. Inf. Model. 2013, 53, 2701−2714. (14) Hubbard, R. E.; Murray, J. B. Experiences in Fragment-Based Lead Discovery. Methods Enzymol. 2011, 493, 509−531. (15) Irwin, J. J.; Sterling, T.; Mysinger, M. M.; Bolstad, E. S.; Coleman, R. G. Zinc: A Free Tool to Discover Chemistry for Biology. J. Chem. Inf. Model. 2012, 52, 1757−1768. (16) Schulz, M. N.; Landstrom, J.; Bright, K.; Hubbard, R. E. Design of a Fragment Library That Maximally Represents Available Chemical Space. J. Comput.-Aided Mol. Des. 2011, 25, 611−620. (17) Gad, H.; Koolmeister, T.; Jemth, A. S.; Eshtad, S.; Jacques, S. A.; Strom, C. E.; Svensson, L. M.; Schultz, N.; Lundback, T.; Einarsdottir, B. O.; Saleh, A.; Gokturk, C.; Baranczewski, P.; Svensson, R.; Berntsson, R. P.; Gustafsson, R.; Stromberg, K.; Sanjiv, K.; JacquesCordonnier, M. C.; Desroses, M.; Gustavsson, A. L.; Olofsson, R.; Johansson, F.; Homan, E. J.; Loseva, O.; Brautigam, L.; Johansson, L.; Hoglund, A.; Hagenkort, A.; Pham, T.; Altun, M.; Gaugaz, F. Z.; Vikingsson, S.; Evers, B.; Henriksson, M.; Vallin, K. S.; Wallner, O. A.; Hammarstrom, L. G.; Wiita, E.; Almlof, I.; Kalderen, C.; Axelsson, H.; Djureinovic, T.; Puigvert, J. C.; Haggblad, M.; Jeppsson, F.; Martens, U.; Lundin, C.; Lundgren, B.; Granelli, I.; Jensen, A. J.; Artursson, P.; Nilsson, J. A.; Stenmark, P.; Scobie, M.; Berglund, U. W.; Helleday, T. Mth1 Inhibition Eradicates Cancer by Preventing Sanitation of the Dntp Pool. Nature 2014, 508, 215−221. (18) Warpman Berglund, U. W.; Sanjiv, K.; Gad, H.; Kalderen, C.; Koolmeister, T.; Pham, T.; Gokturk, C.; Jafari, R.; Maddalo, G.; Seashore-Ludlow, B.; Chernobrovkin, A.; Manoilov, A.; Pateras, I. S.; Rasti, A.; Jemth, A. S.; Almlof, I.; Loseva, O.; Visnes, T.; Einarsdottir, B. O.; Gaugaz, F. Z.; Saleh, A.; Platzack, B.; Wallner, O. A.; Vallin, K. S. A.; Henriksson, M.; Wakchaure, P.; Borhade, S.; Herr, P.; Kallberg, Y.; Baranczewski, P.; Homan, E. J.; Wiita, E.; Nagpal, V.; Meijer, T.; Schipper, N.; Rudd, S. G.; Brautigam, L.; Lindqvist, A.; Filppula, A.; Lee, T. C.; Artursson, P.; Nilsson, J. A.; Gorgoulis, V. G.; Lehtio, J.; Zubarev, R. A.; Scobie, M.; Helleday, T. Validation and Development of Mth1 Inhibitors for Treatment of Cancer. Ann. Oncol. 2016, 27, 2275−2283. (19) Brautigam, L.; Pudelko, L.; Jemth, A. S.; Gad, H.; Narwal, M.; Gustafsson, R.; Karsten, S.; Puigvert, J. C.; Homan, E.; Berndt, C.; Berglund, U. W.; Stenmark, P.; Helleday, T. Hypoxic Signaling and the Cellular Redox Tumor Environment Determine Sensitivity to Mth1 Inhibition. Cancer Res. 2016, 76, 2366−2375. (20) Dong, L. W.; Wang, H. G.; Niu, J. J.; Zou, M. W.; Wu, N. T.; Yu, D. B.; Wang, Y.; Zou, Z. H. Echinacoside Induces Apoptotic Cancer Cell Death by Inhibiting the Nucleotide Pool Sanitizing Enzyme Mth1. OncoTargets Ther. 2015, 8, 3649−3664. (21) Freudenthal, B. D.; Beard, W. A.; Perera, L.; Shock, D. D.; Kim, T.; Schlick, T.; Wilson, S. H. Uncovering the Polymerase-Induced Cytotoxicity of an Oxidized Nucleotide. Nature 2015, 517, 635−639. (22) Giribaldi, M. G.; Munoz, A.; Halvorsen, K.; Patel, A.; Rai, P. Mth1 Expression Is Required for Effective Transformation by Oncogenic Hras. Oncotarget 2015, 6, 11519−11529. (23) Li, L.; Song, L. J.; Liu, X. W.; Yang, X.; Li, X.; He, T.; Wang, N.; Yang, S. L. X.; Yu, C.; Yin, T.; Wen, Y. Z.; He, Z. Y.; Wei, X. W.; Su, W. J.; Wu, Q. J.; Yao, S. H.; Gong, C. Y.; Wei, Y. Q. Artificial Virus Delivers Crispr-Cas9 System for Genome Editing of Cells in Mice. ACS Nano 2017, 11, 95−111.

(24) Nakabeppu, Y. Cellular Levels of 8-Oxoguanine in Either DNA or the Nucleotide Pool Play Pivotal Roles in Carcinogenesis and Survival of Cancer Cells. Int. J. Mol. Sci. 2014, 15, 12543−12557. (25) Kawamura, T.; Kawatani, M.; Muroi, M.; Kondoh, Y.; Futamura, Y.; Aono, H.; Tanaka, M.; Honda, K.; Osada, H. Proteomic Profiling of Small-Molecule Inhibitors Reveals Dispensability of Mth1 for Cancer Cell Survival. Sci. Rep. 2016, 6, 26521. (26) Kettle, J. G.; Alwan, H.; Bista, M.; Breed, J.; Davies, N. L.; Eckersley, K.; Fillery, S.; Foote, K. M.; Goodwin, L.; Jones, D. R.; Kack, H.; Lau, A.; Nissink, J. W.; Read, J.; Scott, J. S.; Taylor, B.; Walker, G.; Wissler, L.; Wylot, M. Potent and Selective Inhibitors of Mth1 Probe Its Role in Cancer Cell Survival. J. Med. Chem. 2016, 59, 2346−2361. (27) Lorber, D. M.; Shoichet, B. K. Hierarchical Docking of Databases of Multiple Ligand Conformations. Curr. Top. Med. Chem. 2005, 5, 739−749. (28) Irwin, J. J.; Shoichet, B. K.; Mysinger, M. M.; Huang, N.; Colizzi, F.; Wassam, P.; Cao, Y. Automated Docking Screens: A Feasibility Study. J. Med. Chem. 2009, 52, 5712−5720. (29) Mysinger, M. M.; Shoichet, B. K. Rapid Context-Dependent Ligand Desolvation in Molecular Docking. J. Chem. Inf. Model. 2010, 50, 1561−1573. (30) Tversky, A. Features of Similarity. Psychol. Rev. 1977, 84, 327− 352. (31) Rogers, D.; Hahn, M. Extended-Connectivity Fingerprints. J. Chem. Inf. Model. 2010, 50, 742−754. (32) Morgan, H. L. The Generation of a Unique Machine Description for Chemical Structures-a Technique Developed at Chemical Abstracts Service. J. Chem. Doc. 1965, 5, 107−113. (33) Carlsson, J.; Coleman, R. G.; Setola, V.; Irwin, J. J.; Fan, H.; Schlessinger, A.; Sali, A.; Roth, B. L.; Shoichet, B. K. Ligand Discovery from a Dopamine D3 Receptor Homology Model and Crystal Structure. Nat. Chem. Biol. 2011, 7, 769−778. (34) Svensson, L. M.; Jemth, A. S.; Desroses, M.; Loseva, O.; Helleday, T.; Hogbom, M.; Stenmark, P. Crystal Structure of Human Mth1 and the 8-Oxo-Dgmp Product Complex. FEBS Lett. 2011, 585, 2617−2621. (35) Fischer, M.; Coleman, R. G.; Fraser, J. S.; Shoichet, B. K. Incorporation of Protein Flexibility and Conformational Energy Penalties in Docking Screens to Improve Ligand Discovery. Nat. Chem. 2014, 6, 575−583. (36) Chen, Y.; Shoichet, B. K. Molecular Docking and Ligand Specificity in Fragment-Based Inhibitor Discovery. Nat. Chem. Biol. 2009, 5, 358−364. (37) Ranganathan, A.; Heine, P.; Rudling, A.; Pluckthun, A.; Kummer, L.; Carlsson, J. Ligand Discovery for a Peptide-Binding Gpcr by Structure-Based Screening of Fragment- and Lead-Like Chemical Libraries. ACS Chem. Biol. 2017, 12, 735−745. (38) Teotico, D. G.; Babaoglu, K.; Rocklin, G. J.; Ferreira, R. S.; Giannetti, A. M.; Shoichet, B. K. Docking for Fragment Inhibitors of Ampc Beta-Lactamase. Proc. Natl. Acad. Sci. U. S. A. 2009, 106, 7455− 7460. (39) de Graaf, C.; Kooistra, A. J.; Vischer, H. F.; Katritch, V.; Kuijer, M.; Shiroishi, M.; Iwata, S.; Shimamura, T.; Stevens, R. C.; de Esch, I. J.; Leurs, R. Crystal Structure-Based Virtual Screening for FragmentLike Ligands of the Human Histamine H(1) Receptor. J. Med. Chem. 2011, 54, 8195−8206. (40) Ranganathan, A.; Stoddart, L. A.; Hill, S. J.; Carlsson, J. Fragment-Based Discovery of Subtype-Selective Adenosine Receptor Ligands from Homology Models. J. Med. Chem. 2015, 58, 9578−9590. (41) Vass, M.; Agai-Csongor, E.; Horti, F.; Keseru, G. M. Multiple Fragment Docking and Linking in Primary and Secondary Pockets of Dopamine Receptors. ACS Med. Chem. Lett. 2014, 5, 1010−1014. (42) Rodriguez, D.; Chakraborty, S.; Warnick, E.; Crane, S.; Gao, Z. G.; O’Connor, R.; Jacobson, K. A.; Carlsson, J. Structure-Based Screening of Uncharted Chemical Space for Atypical Adenosine Receptor Agonists. ACS Chem. Biol. 2016, 11, 2763−2772. (43) Mannel, B.; Jaiteh, M.; Zeifman, A.; Randakova, A.; Moller, D.; Hubner, H.; Gmeiner, P.; Carlsson, J. Structure-Guided Screening for 8168

DOI: 10.1021/acs.jmedchem.7b01006 J. Med. Chem. 2017, 60, 8160−8169

Journal of Medicinal Chemistry

Article

Functionally Selective D2 Dopamine Receptor Ligands from a Virtual Chemical Library. ACS Chem. Biol. 2017, DOI: 10.1021/acschembio.7b00493. (44) Ruddigkeit, L.; van Deursen, R.; Blum, L. C.; Reymond, J. L. Enumeration of 166 Billion Organic Small Molecules in the Chemical Universe Database Gdb-17. J. Chem. Inf. Model. 2012, 52, 2864−2875. (45) Meng, E. C.; Shoichet, B. K.; Kuntz, I. D. Automated Docking with Grid-Based Energy Evaluation. J. Comput. Chem. 1992, 13, 505− 524. (46) Nicholls, A.; Honig, B. A Rapid Finite Difference Algorithm, Utilizing Successive over-Relaxation to Solve the Poisson−Boltzmann Equation. J. Comput. Chem. 1991, 12, 435−445. (47) Weiner, S. J.; Kollman, P. A.; Case, D. A.; Singh, U. C.; Ghio, C.; Alagona, G.; Profeta, S.; Weiner, P. A New Force Field for Molecular Mechanical Simulation of Nucleic Acids and Proteins. J. Am. Chem. Soc. 1984, 106, 765−784. (48) Huber, K. V.; Salah, E.; Radic, B.; Gridling, M.; Elkins, J. M.; Stukalov, A.; Jemth, A. S.; Gokturk, C.; Sanjiv, K.; Stromberg, K.; Pham, T.; Berglund, U. W.; Colinge, J.; Bennett, K. L.; Loizou, J. I.; Helleday, T.; Knapp, S.; Superti-Furga, G. Stereospecific Targeting of Mth1 by (S)-Crizotinib as an Anticancer Strategy. Nature 2014, 508, 222−227. (49) Streib, M.; Kraling, K.; Richter, K.; Xie, X.; Steuber, H.; Meggers, E. An Organometallic Inhibitor for the Human Repair Enzyme 7,8-Dihydro-8-Oxoguanosine Triphosphatase. Angew. Chem., Int. Ed. 2014, 53, 305−309. (50) Huang, N.; Shoichet, B. K.; Irwin, J. J. Benchmarking Sets for Molecular Docking. J. Med. Chem. 2006, 49, 6789−6801. (51) Powers, R. A.; Morandi, F.; Shoichet, B. K. Structure-Based Discovery of a Novel, Noncovalent Inhibitor of Ampc Beta-Lactamase. Structure 2002, 10, 1013−1023. (52) DeLano, W. L. The Pymol Molecular Graphics System, version 1.4.1; Schrödinger, LLC, 2011. (53) Landrum, G. Rdkit: Open-Source Cheminformatics; GitHub and SourceForge, 2014. (54) Cordella, L. P. An Improved Algorithm for Matching Large Graphs. Proceedings of the 3rd IAPR TC-15 Workshop on Graphbased Representations in Pattern Recognition; IAPR, 2001; pp 149−159. (55) Neron, B.; Menager, H.; Maufrais, C.; Joly, N.; Maupetit, J.; Letort, S.; Carrere, S.; Tuffery, P.; Letondal, C. Mobyle: A New Full Web Bioinformatics Framework. Bioinformatics 2009, 25, 3005−3011. (56) Baell, J. B.; Holloway, G. A. New Substructure Filters for Removal of Pan Assay Interference Compounds (Pains) from Screening Libraries and for Their Exclusion in Bioassays. J. Med. Chem. 2010, 53, 2719−2740. (57) Baykov, A. A.; Evtushenko, O. A.; Avaeva, S. M. A Malachite Green Procedure for Orthophosphate Determination and Its Use in Alkaline Phosphatase-Based Enzyme Immunoassay. Anal. Biochem. 1988, 171, 266−270. (58) Kabsch, W. Integration, Scaling, Space-Group Assignment and Post-Refinement. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2010, 66, 133−144. (59) Kabsch, W. Xds. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2010, 66, 125−132. (60) Winn, M. D.; Ballard, C. C.; Cowtan, K. D.; Dodson, E. J.; Emsley, P.; Evans, P. R.; Keegan, R. M.; Krissinel, E. B.; Leslie, A. G.; McCoy, A.; McNicholas, S. J.; Murshudov, G. N.; Pannu, N. S.; Potterton, E. A.; Powell, H. R.; Read, R. J.; Vagin, A.; Wilson, K. S. Overview of the Ccp4 Suite and Current Developments. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2011, 67, 235−242. (61) McCoy, A. J.; Grosse-Kunstleve, R. W.; Adams, P. D.; Winn, M. D.; Storoni, L. C.; Read, R. J. Phaser Crystallographic Software. J. Appl. Crystallogr. 2007, 40, 658−674. (62) Murshudov, G. N.; Skubak, P.; Lebedev, A. A.; Pannu, N. S.; Steiner, R. A.; Nicholls, R. A.; Winn, M. D.; Long, F.; Vagin, A. A. Refmac5 for the Refinement of Macromolecular Crystal Structures. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2011, 67, 355−367.

(63) Murshudov, G. N.; Vagin, A. A.; Dodson, E. J. Refinement of Macromolecular Structures by the Maximum-Likelihood Method. Acta Crystallogr., Sect. D: Biol. Crystallogr. 1997, 53, 240−255. (64) Emsley, P.; Cowtan, K. Coot: Model-Building Tools for Molecular Graphics. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2004, 60, 2126−2132. (65) Emsley, P.; Lohkamp, B.; Scott, W. G.; Cowtan, K. Features and Development of Coot. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2010, 66, 486−501. (66) Chen, V. B.; Arendall, W. B., 3rd; Headd, J. J.; Keedy, D. A.; Immormino, R. M.; Kapral, G. J.; Murray, L. W.; Richardson, J. S.; Richardson, D. C. Molprobity: All-Atom Structure Validation for Macromolecular Crystallography. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2010, 66, 12−21. (67) Davis, I. W.; Leaver-Fay, A.; Chen, V. B.; Block, J. N.; Kapral, G. J.; Wang, X.; Murray, L. W.; Arendall, W. B., 3rd; Snoeyink, J.; Richardson, J. S.; Richardson, D. C. Molprobity: All-Atom Contacts and Structure Validation for Proteins and Nucleic Acids. Nucleic Acids Res. 2007, 35, W375−W383.

8169

DOI: 10.1021/acs.jmedchem.7b01006 J. Med. Chem. 2017, 60, 8160−8169