A Real-World Perspective on Molecular Design - Journal of Medicinal

Feb 15, 2016 - POAP: A GNU parallel based multithreaded pipeline of open babel and AutoDock suite for boosted high throughput virtual screening...
6 downloads 0 Views 8MB Size
Perspective pubs.acs.org/jmc

A Real-World Perspective on Molecular Design Miniperspective Bernd Kuhn, Wolfgang Guba, Jérôme Hert, David Banner, Caterina Bissantz, Simona Ceccarelli, Wolfgang Haap, Matthias Körner, Andreas Kuglstatter, Christian Lerner, Patrizio Mattei, Werner Neidhart, Emmanuel Pinard, Markus G. Rudolph, Tanja Schulz-Gasch, Thomas Woltering, and Martin Stahl* Roche Pharmaceutical Research and Early Development, Roche Innovation Center Basel, F. Hoffmann-La Roche Ltd., Grenzacherstrasse 124, 4070 Basel, Switzerland ABSTRACT: We present a series of small molecule drug discovery case studies where computational methods were prospectively employed to impact Roche research projects, with the aim of highlighting those methods that provide real added value. Our brief accounts encompass a broad range of methods and techniques applied to a variety of enzymes and receptors. Most of these are based on judicious application of knowledge about molecular conformations and interactions: filling of lipophilic pockets to gain affinity or selectivity, addition of polar substituents, scaffold hopping, transfer of SAR, conformation analysis, and molecular overlays. A case study of sequencedriven focused screening is presented to illustrate how appropriate preprocessing of information enables effective exploitation of prior knowledge. We conclude that qualitative statements enabling chemists to focus on promising regions of chemical space are often more impactful than quantitative prediction.



INTRODUCTION In current small molecule drug discovery, the role of molecular design appears to be firmly established. Journal articles on medicinal chemistry routinely make reference to the use of modeled binding modes, virtual screens, or predicted molecular properties. Pharma companies throughout the world have reported on their ways of deploying and using tools.1−3 Yet the real contributions of molecular design to the progress of individual projects are rarely debated and dissected in public, and convincing case studies are much scarcer than the wide distribution of molecular modeling software might suggest. By means of a series of project vignettes from more than a decade of drug discovery at Roche, we illustrate which tools and approaches have been most fruitful over the years. We demonstrate that there are common principles behind these approaches, which leads us to suggest that the overarching goals of molecular design in drug discovery could be redefined in a manner that leads to a more natural focus on impactful activities. The term “molecular design” is intimately linked to the widely accepted concept of the design cycle, which implies that drug discovery is a process of directed evolution (Figure 1). The cycle may be subdivided into the two experimental sections of synthesis and testing, and one conceptual phase. This conceptual phase begins with data analysis and ends with decisions on the next round of compounds to be synthesized. What happens between analysis and decision making is rather ill-defined. We will call this the design phase. In any actual project, the design phase is a multifaceted process, combining © 2016 American Chemical Society

Figure 1. Schematic representation of the classic design cycle with synthesis and testing as experimental components and with analysis and design as conceptual components. The arrow is broken within the design area to illustrate the discrete step of moving from the known to the unknown and virtual. This article deals with aspects of design, which can take on completely different forms depending on how new ideas are generated and filtered.

Special Issue: Computational Methods for Medicinal Chemistry Received: December 4, 2015 Published: February 15, 2016 4087

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

Table 1. Overview of the Case Studies Reported in This Article project

approach

DPP-IV FABP 4/5 Cathepsin S/L

HSL

Filling a lipophilic pocket Targeting specific residue differences in the binding site Assessment of electrostatic complementarity of protein and ligand. Visualization of halogen bonding potential Scaffold hopping Scaffold hopping Targeting binding site hotspots identified from multiple crystal structures Calculation of torsion energy potentials Hypothesis derived from homology model

GPBAR1 L-CPT1 SST5R

Hypothesis derived from homology model Bioactive shape hypothesis Sequence-based binding site similarity as a basis for focused screening (“chemogenomics”)

β-Tryptase BACE1 PDE 10A

impact

Alternative scaffold Solubility increase, alternative scaffold Affinity increase Prioritization of alternative scaffolds Novel scaffolds with potential for improved PK properties Targeted introduction of solubilizing group Prioritization of scaffolds Identification of druglike hits

reflect on how this knowledge may be utilized for design purposes. Ligands are designed such that they optimize shape complementarity in lipophilic pockets to gain affinity or selectivity. “Pocket filling” is probably the most obvious and widely applied design concept, but it is not without its challenges as the following three applications show. DPP-IV Inhibitors. Inhibitors of dipeptidyl peptidase IV (DPP-IV) have become a recommended therapy to treat patients with type 2 diabetes.8,9 An in-house HTS campaign identified the 2-aminobenzo[a]quinolizine hit 1 as a DPP-IV inhibitor with only moderate potency but interesting druglike profile (Figure 2). Initial attempts to improve the binding affinity by random chemical modifications, in times prior to public domain or in-house DPP-IV crystal structures, were not successful. About 1 year into the project, the binding mode of compound 1 in human DPP-IV was unveiled by a cocrystal structure and allowed for a more rational approach (Figure 2a). Analysis of the binding site revealed that the rather flexible nbutyl substituent points into the S1 pocket of DPP-IV, which is composed of the side chains of several aromatic and aliphatic amino acids. The high lipophilicity of the S1 pocket and the fact that proline ring systems are primarily recognized in DPP-IV substrates suggested substantial affinity gains by replacing the nbutyl substituent with a more rigid ring system. Standard medicinal chemistry replacements with a phenyl (compound 2) or pyridine (compound 4) were complemented by suggestions from virtual screening and molecular design, such as the Nsubstituted lactam 7. Surprisingly, these replacements did not significantly improve the DPP-IV binding affinity. A closer analysis of the shape complementarity of the ring analogs with the S1 pocket as well as modeled overlays with another DPP-IV inhibitor series for which the P1 substituent has been optimized,10 revealed additional space in the aromatic metaposition. Very high gains in binding affinity (40- to 70-fold) could be achieved by small methyl and fluoromethyl substitutions as illustrated by compounds 3, 5, 8, and 9. The pyridine analog 6 has significantly lower DPP-IV activity: the affinity gain from the additional methyl group is likely compensated by an unfavorable interaction of the nitrogen lone pair with the π-electron cloud of a tyrosine residue. As suggested by modeling and in analogy to substrates of DPP-IV, polar functionality is only tolerated at the ortho-position where the polar group is exposed to solvent and polar protein residues. While compounds 3, 5, 8, and 9 all have similar DPPIV affinities, 8 and 9 show the highest druglikeness due to their reduced amphiphilicity, which is caused by a more balanced

information on status and goals of the project, prior knowledge, personal experience, elements of creativity and critical filtering, and practical planning. The task of molecular design, as we understand it, is to turn this complex process into an explicit, rational and traceable one, to the extent possible. The two key criteria of utility for any molecular design approach are that they should lead to experimentally testable predictions and that whether or not these predictions turn out to be correct in the end, the experimental result adds to the understanding of the optimization space available, thus improving chances of correct prediction in an iterative manner. The primary deliverable of molecular design is an idea,4 and success is a meaningful contribution to improved compounds that interrogate a biological system. In what follows, we report 10 case studies that illustrate the contributions of molecular design in Roche drug discovery projects. An overview is provided in Table 1. Each of them begins with a brief project rationale, a problem statement, and is followed by a description of the computational approach employed and a discussion of the results obtained. All studies were conducted in a prospective manner; the outcome of the experimental results was not known at the time when computational input was provided and medicinal chemistry was planned. Most of these studies have not been published before. Thus, a set of novel compounds are shown and a series of 13 new entries to the PDB accompany this work. While reading the case studies, one should keep in mind that they are not full project accounts but focus on the role of one discipline. This means that novel compounds are occasionally introduced without a description of their genesis, in order to focus the attention on one particular phase and design problem. It also means that not all aspects of optimization receive the same amount of attention. For instance, the optimization of ADMET properties played a prominent role in all projects covered here but does not always feature prominently in the case studies.



Affinity increase while balancing polarity Affinity and selectivity increase Affinity and selectivity increase

CASE STUDIES

The origin of molecular design coincides with the advent of more routine crystal structure determination and is essentially synonymous with the exploitation of this information, one structure at a time. Today, solid empirical statements can be made about interaction and conformation preferences of proteins and small molecules by extending such analyses over thousands of small molecule and protein complex structures. We and others have compiled such knowledge,5−7 and we will not attempt a new summary here. Instead we would like to 4088

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

Figure 2. (a) Crystal structure of human DPP-IV with inhibitor 1 (PDB code 3oc0). Protein residues that form the S1 pocket and side chains that are engaged in hydrogen bonds (red dashed lines) with the amino functionality of 1 are highlighted in green. The inhibitor is shown in cyan. (b) Close-up view of the S1 pocket of the crystal structure of human DPP-IV with 9 (PDB code 3kwf). Nonbonded contacts with distances of ≤4.0 Å between the fluoromethyl group and the protein are shown with magenta lines. (c) SAR of selected compounds against DPP-IV.11,14 The values indicate enzyme inhibition IC50 values against human DPP-IV.

distribution of polar groups.11 Compound 9 was selected for clinical development (Carmegliptin). The DPP-IV example illustrates that potency gains can quickly be achieved when the binding mode is known from a crystal structure and a well-defined hydrophobic pocket is poorly occupied. Affinity gains by a single heavy atom are dramatic in this case, affirming the importance of filling small voids in buried cavities.12,13 High lipophilicity and little flexibility of the DPP-IV S1 pocket make this a compelling case for structure-based design. High affinity alone should of course not be the goal of optimization; in this case the targeted introduction of polarity was critical for a balanced PK profile and elimination of hERG inhibition. Suboptimal shape complementarity can be readily detected and visualized; thus, proper annotation and visualization of 3D complex structures can enable structure-based design with simple means. Optimizing inhibitor affinity in a more polar or more flexible binding pocket is more challenging due to the uncertainties of estimating the energetics of (de)solvation and protein induced fit and the highly directed nature of most polar interactions. Fatty Acid Binding Protein 4/5. Fatty acid binding proteins (FABPs) are cytosolic lipid binding proteins with tissue-specific distribution that are involved in uptake, metabolism, and intracellular trafficking of fatty acids.15 The isoforms FABP4 and FABP5 have been identified as potential diabetes and atherosclerosis targets based on epidemiological studies and animal knockout models.16 The desired inhibitor profile included a high degree of selectivity against FABP3. A focused screen of a set of 1200 compounds from the Roche library yielded quinoline 10 with decent activity on FABP4 (0.1 μM) but neither selectivity against FABP3 nor measurable

activity against FABP5. The goal was thus to improve FABP4 and FABP5 activities while strongly reducing FABP3 activity. A common approach to address this question is to map sequence information on the relevant targets onto the ligand binding site and to target those regions in 3D space that show amino acid differences. A crystal structure of the quinoline hit 10 in complex with FABP4 was determined early in the project, revealing an area in which a cluster of three residues differ between FABP3 and the other two isoforms (Figure 3a). This pocket appeared to be smaller for FABP3 due to the presence of three Leu side chains, which are bulkier than the set of Ile, Val, and Cys residues present in FABP4 and FABP5. Consequently, the size of the quinoline 2-substituent was increased in the series methyl < ethyl < isopropyl < piperidinyl to provoke a steric clash with FABP3. As shown in Figure 3, this strategy led to a drastic increase in the FABP3 inhibition constant from 90 nM to 10 μM, nicely correlating with the increasing size of the 2-substituent. Since the FABP4 and FABP5 binding pockets were not optimally filled with the original methyl group, the larger substituents further improved ligand affinity for these two isoforms. The effect was especially pronounced for FABP5 where the undetectable activity of the hit 10 could be converted to submicromolar inhibition with the piperidinyl moiety in 13. In the crystal structure of 13 in complex with FABP4 (Figure 3a), the quinolinepiperidine adopts a pseudoaxial conformation to fit into the selectivity pocket. Previous CSD-based conformational analyses5 revealed that this is not unusual for piperidines connected to electrondeficient aryl rings, such as pyridines or pyrimidines, suggesting little ligand strain in this case. The electron density around Cys118 had to be described with two rotamers which require 4089

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

Figure 3. (a) Binding modes of compounds 10 and 13 in FABP4 (green, PDB codes 5edb and 5edc). On the left, amino acids in the selectivity pocket that differ between FABP3, FABP4, and FABP5 are displayed. FABP3 (cyan) has Leu104, Leu115, Leu117 (PDB code 2hmb), while FABP4 (green) has Ile105, Val116, Cys118 (PDB code 5edb), and FABP5 (magenta) has Ile107, Val118, Cys120 (PDB code 1b56). Electron densities are rather weak for the methyl group in 5edb and for the piperidinyl in 5edc. Two conformations for the side chain of Cys118 and for the ligand piperidinyl were used to model the electron density of the FABP4−13 complex structure. (b) Inhibition constants (Ki) of compounds 10−13 against FABP3, FABP4, and FABP5.

already in the low nanomolar range, the design goal was to specifically improve CatL binding while maintaining good selectivity toward other cathepsins. In the CatS/L example it was helpful to analyze binding site differences at a more global level. As shown in Figure 4, cathepsin L and S show significant differences in their electrostatic potentials.20 The CatL binding site has a predominantly negative surface potential: within a radius of 7 Å around 14, six acidic Asp and Glu residues but no basic Lys or Arg residues can be found in human CatL. In contrast, the CatS binding site is more balanced. There is only one Asp residue in the human CatS site, which is compensated by one Lys residue. The related cysteine protease cathepsin K (CatK) lies between CatL and CatS in this analysis, with three Asp and no Lys/Arg residues in the binding site of the human enzyme. In an attempt to optimize electrostatic complementarity with cathepsin L, the neutral cyclopropyl linker was replaced by a positively charged azetidine unit. Due to the long-range nature of electrostatic interactions, we reasoned that introduction of a positive ligand charge even remote (>6 Å) from a negatively charged protein side chain might be beneficial for CatL binding. As shown in the pair of compounds 14, 15 in Figure 4, this chemical change indeed led to an 11-fold gain in binding affinity for CatL while the IC50 values changed only 2-fold for CatS and CatK. A subsequently determined crystal structure of 15 in CatL revealed that the azetidine ring remains solventexposed and interacts with negative protein charges only through water mediated contacts (Figure 4b). Building on previous systematic studies of halogen bonding in the S3 pocket of CatL,21 we were able to selectively optimize CatL activity further by replacing the 5-chloropyridine of 15 by a 5bromo-3-fluoropyridine unit (compound 16). In summary, almost equipotent dual CatS/L inhibitors with a good

two pseudoaxial conformations of the neighboring ligand piperidinyl group. Targeting amino acid differences in binding sites can often be achieved by focusing on direct protein−ligand contacts, for example, by designing a favorable interaction with atoms of the desired target or by introducing a steric clash with antitargets for which binding should be decreased. It is good practice to attempt to “push the boundaries” as well, i.e., to also design molecules that should clash with a protein pocket to test its flexibility. New protein conformations have been discovered in this fashion.17,18 Unfortunately, the situation is often less favorable than in the FABP3 case, where a simple steric clash with the undesired target could be introduced and was sufficient to achieve isoform specificity. More subtle conformational plasticity and indirect effects may be the dominant factors driving target selectivity, and these are virtually impossible to detect by crystallography or to predict computationally. In other cases, selectivity cannot be achieved through exploitation of amino acid differences in close proximity to the ligand. As the next example shows, it may be helpful to take into account more remote areas of the binding site and to utilize alternative ways of classifying interactions. Cathepsin S/L. Dual inhibition of the cysteine proteases cathepsin S (CatS) and L (CatL) is a potential strategy to target glomerular diseases of the kidney. Triggered by the identification of potent and selective CatS inhibitors19 from a previous project, the team embarked on a focused program to tune in additional CatL activity. Starting points were pyrrolidines of type 14 (Figure 4) in which the nitrile covalently binds to the active site cysteine, while the substituted phenylsulfone and the chloro-substituted pyridine occupy the S2 and S3 pockets, respectively. Since CatS in vitro activity was 4090

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

Figure 4. (a) Electrostatic potentials mapped onto the binding sites of human cathepsin L (left, PDB code 2yjc) and mouse cathepsin S (middle, PDB code 4bpv). Mouse crystal structures were used as surrogates for the human isoform for technical reasons. The following charged amino acids are found within 7 Å around 14: human CatL (2 Glu, 4 Asp), human CatS (1 Asp, 1 Lys), mouse CatS (1 Glu, 1 Asp, 1 Lys). Electrostatic maps were calculated with MOE using a Gaussian screened Coulomb potential.20 The color scale for the electrostatic potential ranges from −40 kcal/mol (red) to +40 kcal/mol (blue). (b) X-ray cocrystal structure of 15 with human CatL (PDB code 5f02). The newly introduced azetidine ring is highlighted in yellow. (c) IC50 values are from an enzyme inhibition assay against human cathepsin L, S, and K, as described previously.19

selectivity window toward other cathepsins could be identified through optimization of electrostatic complementarity and finetuning of halogen bonding interactions. This study exemplifies how not only ligand groups in direct contact with protein atoms but also solvent-exposed ligand groups can be optimized for binding.22,23 Engineering the electrostatic complementarity of ligands and their protein environment can be a useful strategy to gain affinity and improve selectivity against antitargets.24,25 An analysis of electrostatics should be a routine step in investigating the binding site of a new protein. Thus far, we have been concerned with optimizing interactions to gain affinity and selectivity. At the opposite end of the spectrum of structure-based design tools there are methods that aim at keeping existing interactions in place while modifying the core of the molecule. Such techniques are known as scaffold hopping methods. ReCore26,27 is one such method, building on the original concept of CAVEAT.28 ReCore suggests scaffold replacements by identifying fragments from small molecule crystal structures with matching exit vector geometries and optional pharmacophore constraints. The next two examples illustrate that scaffold hopping can be a powerful tool to access novel chemical space. β-Tryptase is a heparin-stabilized tetrameric serine protease predominantly expressed in mast cells, which plays a pivotal role in airway inflammatory and allergic responses. A reference inhibitor series from Sanofi-Aventis containing a benzylamine in the S1 pocket,29 a piperidine amide scaffold linker, and various substituted aryl rings in the S4 pocket (Figure 5) were known as potent tryptase inhibitors at the onset of our discovery program. The design goal was to identify proprietary, potent, and selective β-tryptase ligands. A crystal structure of one of the Sanofi-Aventis molecules in complex with human β-tryptase could be determined early in the project. The basic benzylamine group occupies the S1

pocket interacting with the anionic Asp207 side chain as expected for trypsin-like serine proteases, while the substituted aryl ring points into the inducible S4 pocket. The ligand segment linking these two moieties has relatively little protein contact. ReCore was therefore used to identify novel linker groups connecting the P1 and P4 aryl motifs while maintaining the acceptor functionality of the original piperidine amide (Figure 5a). The 200 top-ranking ReCore solutions were clustered and a diverse subset was visualized within the tryptase binding site. After modification of some of the proposed scaffolds for better structural fit and molecular properties, a selection of 20 frameworks was discussed with the medicinal chemists. Due to its novelty for β-tryptase and relative ease of synthetic access, the CSD scaffold BUHGEQ was selected with highest priority. The newly designed oxazolidinone molecules showed double-digit nM inhibition for human β-tryptase and a generally high selectivity toward related proteases. While the successful scaffold replacement suggested that our overlay hypothesis was correct, a cocrystal structure of an oxazolidinone analogue in β-tryptase showed a surprising flip of the central fragment (Figure 5c). The conformation of the new linker is as predicted but inverted to its mirror image. This is due to the formation of an alternative hydrogen bond: the rather weak interaction of the piperidine amide carbonyl group with the backbone nitrogen of Gly237 is replaced by a hydrogen bond to the side chain amide of Gln105 in the new series. This finding may serve as a reminder never to trust a binding mode hypothesis unless all conceivable alternatives have been explored! BACE1. We have previously described30,31 BACE1 inhibitors with a variety of cyclic amidines, which are a widely explored headgroup.32 The amidine moiety binds tightly to the catalytic aspartate residues Asp93 and Asp289 via three hydrogen bonds and offers suitable exit vectors for extending into the S2′ as well as into the S1 and S3 subsites. The S1 pocket is shallow and our 4091

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

Figure 5. (a) Reference tryptase inhibitor and novel oxazolidinone scaffold derived from ReCore26 linker replacement. The bonds used to define the linker region are shown as bold, green sticks, and the acceptor functionality used as an additional ReCore constraint is highlighted with a red ellipse. (b) Overlay produced by ReCore in the scaffold hopping run. (c) Crystal structure overlay of the reference scaffold29 (white, PDB code 4a6l) and a newly designed oxazolidinone inhibitor (yellow, PDB code 5f03) both in human β-tryptase. Overlay of terminal groups is highlighted by red circles to emphasize similarity of overall binding modes.

Figure 6. (a) The P1 phenyl ring in 19 was replaced by a trans-cyclopropylketone in 20. The superposition of the crystal structures of 19 (PDB code 5ezz) and 21 (PDB code 5ezx) shows a perfect overlay between the P1 phenyl ring and the designed mimetic. (b) Trans-cyclopropylamide as an analogous mimetic for meta-substituted anilines: overlay of compound 21 (PDB code 5f00) and 22 (PDB code 5f01).

inhibitors contain a meta-substituted phenyl ring P1 substituent that bridges the headgroup and the P3 substituent.30 We were looking for a replacement of the P1 phenyl ring that would also improve physicochemical properties. Compound 19 in Figure 6 is highly lipophilic with a log D of 4.0, resulting in a very low kinetic solubility. ReCore suggested trans-cyclopropylketone as a polar replacement to a meta-substituted phenyl ring (CSD entry: FUQGAZ). The dihedral angles of the exit vectors only differ by 2.6°, indicating an almost perfect overlay. Incorporating this fragment into the previous compound resulted in compound 20 which had only a marginally better potency in the cell-based assay but significantly improved properties with a reduced log D and an increased solubility (Figure 6a). Aromatic amines are of special concern in medicinal chemistry because they may be associated with the formation of reactive metabolites and genotoxicity.33 In order to explore the applicability of using trans-cyclopropylketone as a replacement for meta-substituted anilines, we substituted the aniline moiety in compound 21 to produce 22. As in the

previous case, we observed a perfect match between the modeled and the experimentally determined receptor-bound conformations as well as improved solubility (Figure 6b). The β-tryptase and BACE1 examples demonstrate that scaffold replacement using CSD-derived linker motifs can identify nonobvious replacements and thus open up new chemical space for projects. Binding sites that are ideally suited for this approach have the main ligand recognition motifs at the distal ends of the molecule and few directed protein contacts in the central region. This allows for a high number of diverse design solutions and a greater tolerance for design limitations. Where this ideal situation is not given, the search for scaffold replacements essentially becomes a pharmacophore search within a very tight search space for which the number of constraints can become a limitation. Other methods of combining and connecting fragments of known inhibitors with each other may be more promising in such cases. 3D merging of fragments from different crystal structures is an elegant approach to combine interesting motifs from different molecular series. The larger and more diverse is the structural 4092

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

Figure 7. (a) Crystal structure of human PDE10A (green) with a triazolopyrimidine inhibitor (cyan, PDB code 5edh). The superimposed structures in orange are members of two other inhibitor series from the Roche PDE10 project that occupy a binding hotspot with a chlorine atom according to Scorpion network analysis39 (PDB codes 5ede and 5edg). The hotspot is shown as a cyan ellipsoid, and the detected favorable interactions are indicated with dashed lines (yellow: dispersion; pink: halogen bonding). (b) Inhibition constants for human PDE10A and ligand efficiencies (LE) for triazolopyridines 23 and 24. (c) Crystal structure of 24 in complex with human PDE10A (PDB code 5edi). The binding hotspot is highlighted by a red circle.

ellipsoid and makes the predicted interactions (Figure 7c). While in this particular case, the transfer of the Cl substituent from one series to another may appear as a straightforward exercise, the workflow described here is far more generally applicable, allowing the extraction of lessons learned from many crystal structures and the transfer of larger ligand moieties. The importance of understanding preferred, low-energy conformations in structure-based design is undisputed.40 While today’s force fields cover much of the chemical space accessed by medicinal chemistry, additional and more detailed insights into conformational preferences can be gained from statistical analysis of small molecule X-ray structures.5 However, the chemical space covered in the CSD does not necessarily overlap with that of pharmaceutically interesting compounds. One particular example is the analysis of torsion angles between heterocyclic systems, which are under-represented in the CSD. Where experimental data are too sparse for statistical analysis, quantum mechanical calculations are the method of choice. As part of the PDE10A project, we worked on a 3pyrazolylpyridazin-4-one inhibitor series (26) which was identified by an HTS. A crystal structure of 26 in complex with human PDE10A revealed important interactions of the pyridazin-4-one core with the protein, namely, a hydrogen bond of the carbonyl oxygen atom with the conserved Gln726 side chain and extensive π−π interactions of the pyridazine moiety with an aromatic clamp formed by Phe696 and Phe729. In the binding mode of 26, the two terminal phenyl groups make an intramolecular edge-to-face interaction and the pyrazole linker is twisted by ∼28° against the pyridazin-4-one plane (Figure 8). Scaffold hopping of the pyridazin-4-one core to two pyridin-2-one variants yielded the bioisosteric replacement 27, which displayed only a 3-fold lower PDE10A inhibition than 26 and the more than 300-fold less active analogue 25. Since the activity drop of 25 could not be explained from the binding pose when bound to PDE10A, we suspected ligand strain as a potential cause. Quantummechanical (QM) calculations indeed revealed a major difference in the energy profile of the torsion τ between the two heterocycles. While the energy minimum for 26 and 27 (red and blue curves) nicely coincides with the torsional angle

information, the greater are the opportunities to generate novel combinations. 3D merging of project structures was first introduced with the Breed algorithm.34 The recent design of novel DDR1/2 kinase inhibitors through combination of a fragment screening hit with known parts from another kinase inhibitor is an elegant application of this technique.35 In the example below, a halogen atom is transferred from one series to another, but the concept is equally applicable to larger ligand fragments. PDE10A. Inhibition of phosphodiesterase 10A (PDE10A) is considered an attractive target for the treatment of schizophrenia due to its high prevalence in the striatum and preclinical data demonstrating antipsychotic and procognitive effects.36 We identified triazolopyrimidines and triazolopyridines as potent and selective PDE10A inhibitors (Figure 7). Here we illustrate how binding affinity could be improved by transferring knowledge from different series. Prerequisites for the transfer of ligand motifs from one series to another are the availability of crystal structures as an overlaid set and an understanding of binding site “hotspots” from an analysis of intermolecular interactions or from mapping of activity cliffs37 onto the binding site. We predicted hotspots for the PDE10A active site based on an automated overlay procedure implemented in Proasis38 and Scorpion network analysis of favorable interactions.39 Ligand atoms with Scorpion score contributions greater than 1.5 were marked as hotspots and clustered according to their spatial distribution. The clusters were then visualized as ellipsoids whose shape indicates the distribution of the cluster members. We identified a binding hotspot for a halogen atom with favorable dispersion and halogen bonding interactions, as exemplified in two complex structures of other PDE10A inhibitor series (Figure 7a). Superposition with a PDE10A crystal structure in complex with a triazolopyrimidine suggested that this hotspot could be reached in the triazolopyridine series by substitution in the 6position. The Cl-substituted analogue of 23 displayed improved PDE10A inhibition by more than 10-fold and also had increased ligand efficiency. A crystal structure of 24 in complex with the catalytic domain of PDE10A confirmed that the additional chlorine atom does indeed occupy the cluster 4093

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

Figure 8. (a) IC50 values against human PDE10A of pyrazolyl inhibitors with pyridin-2-one, pyridazin-4-one, or pyrazin-2-one cores. (b) Crystal structure of human PDE10A with inhibitor 26 (PDB code 5i2r). Protein residues and inhibitor are shown in gray and cyan, respectively. (c) Calculated energy potential from a relaxed torsion scan around the dihedral angle τ, connecting the two central heteroarenes. Calculations were done with Gaussian 9841 using the density functional B3LYP and a cc-pVDZ basis set. The vertical dashed lines indicate the τ range (28−40°) observed in cocrystal structures of the 3-pyrazolylpyridazin-4-one core with PDE10A.

This compound was suitable for an initial mouse study, where it showed 21% reduction of free fatty acids when dosed at 30 mg/ kg. The series of spirolactam piperidines suffered from low solubility and low metabolic stability, and the medicinal chemistry team set out to optimize these properties. To investigate whether the optimization process could be guided by a binding hypothesis, a homology model of HSL was built. HSL has little sequence homology to proteins of experimentally determined 3D structure. The three templates that were used had only 12−14% overall sequence identity and 22−24% sequence similarity to HSL. The resulting HSL model nevertheless allowed us to build an approximate binding hypothesis, since the buried active site funnel had a distinct shape and since the initial lead 30 has an equally distinct shape: simple force field calculations determined that within a polar environment, the lactam carbonyl group would have a strong preference for the pseudoequatorial position, exposing it more to the environment. Given this orientation, the sulfonamide moiety has only one preferred orientation, pointing the terminal phenyl ring into the opposite direction of the carbonyl group. Within the active site, the lactam carbonyl group fit well into the oxyanion hole, pointing the trifluoromethoxy group into a narrow lipophilic pocket. The sulfonamide substituent was consequently located at the other end of the active site tunnel in a wider lipophilic region. While this binding hypothesis was certainly only a rough sketch, it provided a rationale for the role of the lactam moiety as well as a hint that more room for variation was likely available at the sulfonamide site. However, what the team mostly needed was a position to add polarity to the lead. Sulfonamides are molecular chimeras, which are found to form hydrogen bonds as well as interact with unipolar environments within proteins.6 In this case, the homology model suggested that the sulfonamide oxygen atoms could interact with a glutamine side chain neighboring the

found in the protein-bound conformation, the preferred conformation of 25 is considerably more twisted (magenta curve, minimum at τ ≈ 90°). There is considerable ligand strain (2−3 kcal/mol) at τ ≈ 30° which would translate into a roughly 100-fold decrease in binding affinity. Subsequently synthesized compounds confirmed this observation. For example, the dihedral minimum of pyrazin-2-one containing 28 is at τ = 0°; the compound is still potent since at τ = 30° ligand strain is limited (0.3 kcal/mol). As a consequence of the good quantitative agreement between calculated ligand strains and IC50 values, we performed similar dihedral energy scans to prioritize virtual heteroaryl scaffold modifications before embarking on their synthesis (results not shown). When experimental structural information is not available, to what extent can homology models be employed as well? The following two examples illustrate how even relatively crude protein models can be used to give projects a clear direction. The key to success here is the full utilization of additional mechanistic knowledge about the target class, coupled with the understanding that the model must be interpreted at a lower level of granularity than an experimentally determined structure. Hormone sensitive lipase (HSL) plays a key role in mobilizing stored fats. Diabetic patients exhibit enhanced lipolytic activity due to up-regulation of this lipase. HSL inhibition was assumed to restore exaggerated plasma free fatty acid and triglyceride levels and thus to be of therapeutic value for the treatment of type 2 diabetes.42 At the outset of the project, all known inhibitors of HSL were suicide substrates leading to irreversible inhibition of the target. A high-throughput screen afforded sulfonylpiperidine 29, a first reversible inhibitor with weak but reproducible cellular activity (Figure 9). The potency of 29 could be significantly improved by cyclizing the amide bond to the spirolactam piperidine 30. 4094

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

Figure 9. (a) HSL HTS hit 29 led to 30 by cyclization of the amide bond. (b) Binding hypothesis of compound 30 into the binding pocket of an HSL homology model. (c) Binding hypothesis of compound 31 in the same HSL model. (d) Compound 31 was designed from 30 by replacing the sulfonylpiperidine by cyclohexanol. The binding hypothesis is further supported by the potency loss of diastereomer 32.

agonists as HTS hits, which suffered from high lipophilicity and low solubility.45 A representative compound is depicted in Figure 10 (compound 33, R = H). Since GPBAR1 binds lithocholic acid (LCA) and equally well its taurine conjugate, taurolitocholic acid (TLCA), the receptor obviously tolerates polar or even anionic side chains on naturally occurring agonists, and the question thus arose of where in the current series such polarity could be introduced. To guide medicinal chemistry efforts, a well-founded binding hypothesis was required. Due to the high degree of homology of GPBAR1 to other class A GPCRs, a homology model could be built (rhodopsin template, PDB code 1F88). However, docking an essentially featureless lipophilic compound into the transmembrane binding pocket was not a straightforward exercise. In the absence of additional information, several alternative orientations seemed likely, each suggesting a different exit vector for adding polar groups. The missing piece of information came from the fact that the team was dealing with agonists, whose molecular mechanisms of action could be compared to that of the class A GPCR rhodopsin. In rhodopsin, the β-ionone ring of retinal squeezes between transmembrane helices 5 and 6 (TM5 and TM6), and the small shift in orientation of these two helices relative to each other translates into a much larger outward rotation of TM6 on the cytosolic side of the receptor, allowing it to accommodate a G-protein. Thus, within the orthosteric binding pocket, the gap between TM5 and TM6 was assumed to be an equally important hotspot for agonism. Consequently, bile acids and TLCA were docked into the GPBAR1 model in such a way that this hotspot was occupied by a lipophilic moiety analogous to the β-ionone ring in rhodopsin. To this binding mode, representatives of the lead series could be superimposed

oxyanion hole. We hypothesized that the amide terminus of this side chain could equally well act as a hydrogen bond acceptor. We therefore designed a molecule replacing the sulfonylpiperidine with a cyclohexyl ring bearing a hydroxyl group in the axial position and a butyl chain in the equatorial position to occupy the lipophilic pocket. Indeed, compound 31 showed similar inhibitory activity as 30, with a lower number of heavy atoms and thus with significantly improved ligand efficiency. The validity of the hypothesis is demonstrated by the significant loss of activity of the corresponding enantiomer 32. The same differential activity is observed in attenuated form for the pair of compounds lacking the butyl chain (data not shown). This example illustrates how, in spite of little solid information, prior knowledge about enzymatic mechanism, about related protein structures, and about conformations can be utilized to inspire creativity and to build specific, testable hypotheses. Clearly, such hypotheses will not always be correct, but the better is the experimental design, the more may be learned about where a hypothesis needs to be adapted. In the case of HSL, the homology model was not falsified, which does not mean that it is correct but that its essential features are in accordance with experimental data. Most of all, the project team benefited from the sulfonamide replacement, which expanded chemical space in a new direction and enabled the creation of stable and potent compounds with a balanced physicochemical profile.43 G-Protein-Coupled Bile Acid Receptor 1 (GPBAR1). GPBAR1 is activated by bile acids such as litocholic acids. It plays a role in glucose homeostasis and triggers antiinflammatory activity through Kupffer cells.44 GPBAR1 is a class A GPCR and, like rhodopsin, coupled to Gα. The starting point of the GPBAR1 project was a number of potent GPBAR1 4095

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

Figure 10. (a) Homology model of GPBAR1 with TCLA (cyan) and a representative of the GPBAR1 agonist series (magenta, 33, R = H). The area where the taurine moiety binds is marked by a red circle. (b) GPBAR1 activity, lipophilicity (log D), kinetic solubility (LYSA) and microsomal clearance (CL) for several compounds of the pyridine series.

by preserving binding to the agonist hotspot. This led to a single likely binding orientation as shown in Figure 10a. The most promising exit vector for polar side chains was the para position of the phenyl ring. Figure 10b shows that compounds with various side chains were indeed potent agonists with much improved solubility and in vitro clearance values. The case of GPBAR1 illustrates again that the combination of knowledge from various sources (in this case natural ligands and mechanistic understanding of the target class) can lead to targeted hypotheses also in the absence of an experimentally determined structure of the target protein. Here, the identification of a likely exit vector effectively narrowed down the medicinal chemistry search space to particularly promising areas. With the next example we are leaving the realm of 3D target structures. In their absence, statements about bioactive conformations of course become much more vague, and so one must tread more carefully. 3D alignments of different compound classes and 3D pharmacophore generation are useful techniques, the results of which are of course strongly influenced by the quality of the input conformations. However, ligand overlays almost invariably create an erroneous

appearance of high 3D similarity. Binding modes of different structural classes are very often more different from each other than such overlays suggest. This problem has plagued the world of 3D QSAR for a long time, since overlays are trivial when structural diversity is low or they are very likely wrong. In such situations it is advisible to generate rigidified structures that allow direct testing of a bioactive conformation hypothesis. The Merck orexin project is an excellent example of such validation experiments.46 The next example shows how ligand overlays may be used as guidance without overinterpretation. Liver Carnitine Palmitoyltransferase 1. Carnitine palmitoyltransferases are mitochondrial enzymes responsible for the transfer of an acyl group of a long-chain fatty acyl-CoA to L-carnitine, facilitating transport of fatty acids across the mitochondrial membrane.47 The goal of the project was the identification of inhibitors of the liver isoform (L-CPT1) with selectivity over the muscle isoform (M-CPT1) as well as over CPT2, the enzyme catalyzing the deacetylation reaction at the inner mitochondrial membrane. In 2004, an HTS was run on LCPT1. About 5000 primary hits were characterized by eightpoint IC50 values on all three enzymes, leading to about 2200 primary hits with IC50 values below 50 μM, covering multiple 4096

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

Figure 11. (a) Manual 3D overlay of compounds representative of the major hit classes. (b) Schematic representations of two enantiomeric bioactive shape envelopes, of which the one shown to the left was predicted to be correct. (c) Compound 39 was derived from HTS hit 38. The enantiomer 39 was correctly predicted to be the active one using the shape envelope hypothesis.

Figure 12. (a) Analysis of the amino acid homology around the consensus drug binding site in GPCRs. Transmembrane binding pockets of biogenic amine receptors (serotonin, dopamine, histamine, opioid) are most closely related to hSST5R. The highly conserved aspartic acid (D) is marked by a red box; the tertiary amine in the piperidine ring of astemizole is believed to be in close vicinity. (b) From astemizole to compound 40, the first highly selective SSTR5 antagonist small molecule with druglike properties.

tractable hit classes. What was the role of CADD in the early phase of the project? On the basis of substructure-based clustering (MCS, MOS)48 and manually performed overlays, it could be determined that a majority of the hit classes was likely to adopt the same general shape in their bioactive conformation (Figure 11a). Since this putative bioactive shape envelope was chiral, but the hit structures were not, and in the absence of additional information, the two possible enantiomeric shape envelopes could not be distinguished (Figure 11b). A homology model of L-CPT1 allowed that distinction to be made: the compounds were too large to be fit into the acylcarnitine binding tunnel, and thus it was assumed that they would bind at the CoA binding site. This binding site fit only one of the two enantiomeric overlays. During the initial phase of hit exploration, the shape envelope shown in Figure 11b was used to prioritize the synthesis of hits and to mix and match elements of different hits with each other. The bioactive shape hypothesis was supported by the correct prediction that the Dproline derivative 39, but not its enantiomer, should be an inhibitor of L-CPT1 (Figure 11c). Once the project team had explored a broad variety of hits, medicinal chemistry quickly focused on optimization of selectivity, solubility, and stability of three promising series. From this point onward, the project was essentially a classic medicinal chemistry optimization program where predictive methods focused on molecular properties rather than binding modes and affinities.

It can be concluded that the shape analysis was a helpful exercise to discover similarities between the ligands and to initiate exercises of “mixing and matching” of ligand moieties within the project team. Generating the overlay was a stepwise, manual process that could not have been easily automated. Second, it was a good decision not to continue refining the binding model beyond the simple shape analysis. Any quantitative 3D pharmacophore hypothesis, which could have been built based on the overlay of multiple ligands, would possibly have led the team astray. Experimental validation of the L-CPT1 binding hypothesis has not been obtained to date. We were, however, able to solve crystal structures of some unselective inhibitors with CPT2, which do show a wider spread of binding modes than simple overlays would suggest.49 The absence of exploitable ligand or protein 3D information marks the transition from classic molecular design to what is often termed “cheminformatics”. Many methods in this realm have become so embedded in daily work that their hypothetical or design character is no longer realized: simple similarity algorithms are routinely used for clustering, for HTS follow-up, for the selection of focused screening sets, or for patent analysis.50,51 Cheminformatics techniques are closely linked to general information science and bioinformatics. They are most effective if implemented in such a way that they cross-link information from various different sources, provide context, and lower hurdles to data access. The following final example illustrates this way of thinking.52 4097

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

observation cannot be overrated in a field that has invested enormously in quantitative prediction methods. We believe that quantitative prediction alone is a misleading mission statement for molecular design. Computational tools, by their very nature, do of course produce numerical results, but these should never be used as such. Instead, any ranked list should be seen as raw input for further assessment within the context of the project. This principle can be applied very broadly and beyond the question of binding affinity prediction, for example, when choosing classification rather than regression models in property prediction. 2. Shaping Chemical Space. Medicinal chemistry is a discipline that creates. At any given point during a project, a team’s focus is either on expanding chemical space or on narrowing it down, for different aspects of problem solving and optimization. Broadening chemical space requires methods that create new ideas within a set of constraints. Helpful tools in such situations are query-based tools: tools for 3D scaffold hopping, pharmacophore or shape searching, or tools that recombine elements of known compounds with each other. Narrowing down chemical space can be a simple filtering process or can be based on a specific hypothesis as in the GPBAR1 case above. Within a given project context, it is important to understand whether it is required to broaden or narrow down chemical space and to choose tools and approaches accordingly (cf. legend to Figure 1). As projects progress toward candidate selection, the “amplitudes” of narrowing and broadening space typically become smaller but the concept stays the same. A rational approach in advanced lead optimization may, for example, suggest avoidance of a certain substructure or preferential use of a specific substitution pattern to avoid an off-target effect or to improve ADMET properties. 3. The Principle of Parsimony. Molecular design is a conceptual process and therefore always at risk of losing touch with reality. The scientific questions should lead to the method and not vice versa. To achieve this, it is a helpful guiding principle to keep things as simple as possible. Choosing the simplest possible explanation and the simplest possible computational protocol leads to agility and to a better focus on the key questions at hand. A case in point is the use of data derived from experiment whenever possible (e.g., the CSD for conformations) and to resort to first-principles methods only where required (e.g., torsion angles between heterocycles or tautomer preferences). 4. Annotation Is Half the Battle. The cathepsin example above shows how orthogonal perspectives are required to see design opportunities. Complex crystal structures gain value when visualized with information about favorable and unfavorable interactions, contacts formed by water molecules, flexible regions, and more. Contextual information can add value almost anywhere. A good deal of frontloading work (computational, organizational) is often required to bring data into a useful shape. Proper frontloading work can turn sophisticated queries into simple lookup processes or visualization steps. Good examples beyond binding mode visualization are the reuse of ADMET data by means of matched molecular pairs or the tailored bioinformatics approach to SST5 above. There is a significant growth potential in this area.57,58 5. Staying Close to Experiment. One way of keeping things as simple as possible is to preferentially utilize experimental data that may support a project, wherever this is meaningful. This may be done in many different ways: by

SST5R. Somatostatin (SST) is a cyclic tetradecapeptide hormone with mainly inhibitory effects on hormonal secretion such as the release of growth hormone, pancreatic insulin, glucagon, and gastrin.53 SST acts via five distinct G-proteincoupled receptors (GPCRs). When we sought selective inhibitors of the SST-receptor subtype 5 (SST5R) as potential antidiabetic agents, no small molecule antagonists were known. Since we did not want to commit substantial resources for a high-throughput screening campaign and structure-based design was out of reach, a focused screening approach seemed to be the only amenable route to quickly identify chemical entry points. A focused screen requires seed compounds for similarity searches. Rather than following the usual route to find peptidomimetics for the active tetrapeptide motif in SST,54 we chose a different approach based on a biological similarity paradigm. Class A GPCRs with known small molecule ligands were ranked with respect to the similarity of the amino acids lining the transmembrane binding pocket of the hSST5 receptor. Only the local sequence similarity around the consensus binding site was taken into account, an approach that significantly enhances resolution when comparing GPCR structures.55 The nearest neighbors of SST5R were identified to be biogenic amine receptors which typically bind a basic amine close to the conserved Asp3.32 in the transmembrane ligand pocket. We therefore tested a selection of opioid, histamine, dopamine, and serotonin receptor ligands against SSTR5 (Figure 12a). This led to a number of hits with micromolar affinity as determined by an hSST5R radioligand binding assay. From these, we selected astemizole, a second generation antihistamine, as a starting point for further optimization. The medicinal chemistry team then systematically explored the SAR of the lead compound and managed to reduce the histamine H1 affinity of astemizole by removing the benzylic substituent from the benzimidazole core.56 Potent and metabolically stable hSST5R antagonists with a high selectivity toward the other SST receptor subtypes were created in this manner (Figure 12b). In conclusion, this example shows that tailored similarity metrics in combination with systematic searching in databases of known ligands can efficiently uncover nonobvious chemical entry points. A prerequisite for such work is the appropriate data basis: annotated collections of ligands as well as properly aligned sets of transmembrane pocket sequence elements.



CONCLUSION We believe that these 10 Roche case studies, while not exhaustive, are representative of what molecular design can deliver today to small molecule drug discovery. The specific tools and methods are of broad utility for many types of projects including ones we have not covered here: fragmentbased drug design, the design of protein−protein interaction inhibitors, allosteric modulators, and more. The fundamental rules of molecular recognition stay the same, while the degree of difficulty and predictability varies. We would now like to step back to find the common themes behind the diverse tools and approaches used here. 1. Value of Qualitative Statements. Frequently, a single new idea or a pointer in a new direction is sufficient guidance for a project team. Most project impact comes from qualitative work, from sharing an insight or a hypothesis rather than a calculated number or a priority order. The importance of this 4098

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

referring to measured parameters instead of calculated ones or by utilizing existing chemical building blocks instead of designing new ones or by making full use of known ligands and SAR or related protein structures. Rational drug design has a lot to do with clever recycling. If consistently applied, these guidelines reintroduce a pragmatic focus to the current practice of molecular design. Let us look at some problematic aspects of today’s practice as well. Many computational methods introduce additional parameters and thus potential sources of error that make the predictive value harder to extract. Methods dealing with protein dynamics fall into this category. In the majority of structure-enabled projects, rough binding hypotheses or energy estimates can be established with rigid protein structures or with selective relaxation or rotation of side chains. Similar arguments hold for free energy perturbation methods. There is clear progress in this field;59−61 however, in a fast-paced project environment where qualitative answers are sufficient, the principle of parsimony would restrict their use, since there is little return for the additional complexities they introduce. Furthermore, their applicability domain is small, as significant changes in water structure or conformation are hard to account for and quantify. Molecular dynamics (MD) may be useful where an assessment of protein mobility is required and not accessible through other techniques.62,63 Great care needs to be taken not to overinterpret or misinterpret calculation results. MD trajectories cannot be validated experimentally, so extra effort is required to link such simulation results back to truly testable hypotheses, for example, in the qualitative prediction of mechanisms or protein movements that may be exploited for the design of binders. In recent years, methods attempting to simulate the water network within protein binding sites have been promoted.64−66 Here again, the direct calculation results of the methods (water positions and associated energies and in particular entropic arguments) cannot be experimentally tested, so the translation of the results in hypotheses driving drug discovery rather than providing post hoc rationalization requires extra effort. What is this extra effort? At the very least, it should be clear what the baseline is relative to which the calculations should be an improvement. In the case of water structure and dynamics, this could be classic desolvation arguments or geometric scores for crystallographic or modeled water positions.67 Without such a reference model, there is little chance to gain new insights from attempts to calculate the role of water at a higher level of theory. Every method has its applicability domain and a set of intrinsic pitfalls that must be avoided. As mentioned above, one of the most typical ones is the overly high confidence in numerical results. For example, small molecule docking to proteins can be used to prioritize compounds for experimental screening. Docking programs can assess whether a compound fits into a binding pocket in a reasonable conformation, but docking scores are notoriously uncorrelated to binding affinity data.68−71 Appropriate use of docking programs involves the use of as many additional constraints as possible and a significant investment in visual inspection and selection.72 No binding mode predicted by a docking tool should be accepted without further scrutiny. Wherever possible, pharmacophore searching or shape matching offers the testing of a more focused hypothesis.73,74

When different ligands are compared, the resulting featurebased overlays often create an impression of high 3D similarity that is not confirmed by experiment. Binding modes often differ more than 3D similarity suggests. This should be kept in mind whenever making predictions that build on such overlays. 3D pharmacophores and shape constraints can therefore only be derived with confidence if clear statements about bioactive conformations can be made. With most QSAR approaches, there is a risk of generating self-consistent models that essentially reflect the data at hand but otherwise have no predictive power. From the medicinal chemist’s perspective, they have the additional disadvantage of being black boxes, only useful as filters, not fostering creative thinking. Best practice in molecular design is best practice in all sciences: a relentless focus on clarity, simplicity, and good experimental design. What is special about molecular design is the need to build solid hypotheses and to simultaneously foster creative thinking in medicinal chemistry. If we accept this, our focus may shift from the many semiquantitative prediction tools that we have to methods supporting this creative process.75 Further improvements in computational methods may then have less to do with science than with good software engineering and interface design. The tools are just a means to an end. Good science is what happens when they are appropriately employed.



AUTHOR INFORMATION

Corresponding Author

*Phone: +41 61 68 88421. E-mail: [email protected]. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS We thank Neil Taylor for the implementation of the Scorpion network analysis and Michael Reutlinger for the implementation of the hotspot visualization.



ABBREVIATIONS USED CPT, carnitine palmitoyltransferase; DDR, discoidin domain receptor; DPP IV, dipeptidyl peptidase IV; FABP, fatty acid binding protein; GPBAR, G-protein-coupled bile acid receptor; HSL, hormone sensitive lipase; LCA, lithocholic acid; MCS, maximum common substructure; MOS, maximum overlapping structure; SST, somatostatin; TLCA, taurolithocholic acid; TM, transmembrane



REFERENCES

(1) Stahl, M.; Guba, W.; Kansy, M. Integrating Molecular Design Resources within Modern Drug Discovery Research: The Roche Experience. Drug Discovery Today 2006, 11 (7−8), 326−333. (2) Feng, J. A.; Aliagas, I.; Bergeron, P.; Blaney, J. M.; Bradley, E. K.; Koehler, M. F. T.; Lee, M.-L.; Ortwine, D. F.; Tsui, V.; Wu, J.; Gobbi, A. An Integrated Suite of Modeling Tools That Empower Scientists in Structure- and Property-Based Drug Design. J. Comput.-Aided Mol. Des. 2015, 29 (6), 511−523. (3) Hillisch, A.; Heinrich, N.; Wild, H. Computational Chemistry in the Pharmaceutical Industry: From Childhood to Adolescence. ChemMedChem 2015, 10 (12), 1958−1962. (4) Loughney, D.; Claus, B. L.; Johnson, S. R. To Measure Is to Know: An Approach to CADD Performance Metrics. Drug Discovery Today 2011, 16 (13−14), 548−554. (5) Brameld, K. A.; Kuhn, B.; Reuter, D. C.; Stahl, M. Small Molecule Conformational Preferences Derived from Crystal Structure Data. A

4099

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

Medicinal Chemistry Focused Analysis. J. Chem. Inf. Model. 2008, 48 (1), 1−24. (6) Bissantz, C.; Kuhn, B.; Stahl, M. A Medicinal Chemist’s Guide to Molecular Interactions. J. Med. Chem. 2010, 53 (14), 5061−5084. (7) Persch, E.; Dumele, O.; Diederich, F. Molecular Recognition in Chemical and Biological Systems. Angew. Chem., Int. Ed. 2015, 54 (11), 3290−3327. (8) Ahrén, B. Emerging Dipeptidyl Peptidase-4 Inhibitors for the Treatment of Diabetes. Expert Opin. Emerging Drugs 2008, 13 (4), 593−607. (9) Gwaltney, S., II. Medicinal Chemistry Approaches to the Inhibition of Dipeptidyl Peptidase IV. Curr. Top. Med. Chem. 2008, 8 (17), 1545−1552. (10) Peters, J.-U.; Weber, S.; Kritter, S.; Weiss, P.; Wallier, A.; Boehringer, M.; Hennig, M.; Kuhn, B.; Loeffler, B.-M. Aminomethylpyrimidines as Novel DPP-IV Inhibitors: A 105-Fold Activity Increase by Optimization of Aromatic Substituents. Bioorg. Med. Chem. Lett. 2004, 14 (6), 1491−1493. (11) Mattei, P.; Boehringer, M.; Di Giorgio, P.; Fischer, H.; Hennig, M.; Huwyler, J.; Koçer, B.; Kuhn, B.; Loeffler, B. M.; MacDonald, A.; Narquizian, R.; Rauber, E.; Sebokova, E.; Sprecher, U. Discovery of Carmegliptin: A Potent and Long-Acting Dipeptidyl Peptidase IV Inhibitor for the Treatment of Type 2 Diabetes. Bioorg. Med. Chem. Lett. 2010, 20 (3), 1109−1113. (12) Stahl, M.; Böhm, H.-J. Development of Filter Functions for Protein−ligand Docking. J. Mol. Graphics Modell. 1998, 16 (3), 121− 132. (13) Banner, D. W. Principles of Enzyme-Inhibitor Design. In Methods and Principles in Medicinal Chemistry; Böhm, H.-J., Schneider, G., Eds.; Wiley-VCH Verlag GmbH & Co. KGaA: Weinheim, Germany, 2005; pp 163−185, DOI: 10.1002/3527601813.ch7. (14) Boehringer, M.; Fischer, H.; Hennig, M.; Hunziker, D.; Huwyler, J.; Kuhn, B.; Loeffler, B. M.; Luebbers, T.; Mattei, P.; Narquizian, R.; Sebokova, E.; Sprecher, U.; Wessel, H. P. Aryl- and Heteroaryl-Substituted Aminobenzo[a]quinolizines as Dipeptidyl Peptidase IV Inhibitors. Bioorg. Med. Chem. Lett. 2010, 20 (3), 1106−1108. (15) Chmurzyńska, A. The Multigene Family of Fatty Acid-Binding Proteins (FABPs): Function, Structure and Polymorphism. J. Appl. Genet. 2006, 47 (1), 39−48. (16) Furuhashi, M.; Hotamisligil, G. S. Fatty Acid-Binding Proteins: Role in Metabolic Diseases and Potential as Drug Targets. Nat. Rev. Drug Discovery 2008, 7 (6), 489−503. (17) Oefner, C.; Binggeli, A.; Breu, V.; Bur, D.; Clozel, J.-P.; D’Arcy, A.; Dorn, A.; Fischli, W.; Grüninger, F.; Güller, R.; Hirth, G.; Märki, H.; Mathews, S.; Müller, M.; Ridley, R.; Stadier, H.; Vieira, E.; Wilhelm, M.; Winkler, F.; Wostl, W. Renin Inhibition by Substituted Piperidines: A Novel Paradigm for the Inhibition of Monomeric Aspartic Proteinases? Chem. Biol. 1999, 6 (3), 127−131. (18) Carcache, D. A.; Hörtner, S. R.; Bertogg, A.; Binkert, C.; Bur, D.; Märki, H. P.; Dorn, A.; Diederich, F. De Novo Design, Synthesis, and In Vitro Evaluation of a New Class of Nonpeptidic Inhibitors of the Malarial Enzyme Plasmepsin II. ChemBioChem 2002, 3 (11), 1137−1141. (19) Hilpert, H.; Mauser, H.; Humm, R.; Anselm, L.; Kuehne, H.; Hartmann, G.; Gruener, S.; Banner, D. W.; Benz, J.; Gsell, B.; Kuglstatter, A.; Stihle, M.; Thoma, R.; Sanchez, R. A.; Iding, H.; Wirz, B.; Haap, W. Identification of Potent and Selective Cathepsin S Inhibitors Containing Different Central Cyclic Scaffolds. J. Med. Chem. 2013, 56 (23), 9789−9801. (20) Molecular Operating Environment (MOE); Chemical Computing Group Inc. (1010 Sherbooke St. West, Suite 910, Montreal, QC, H3A 2R7, Canada). (21) Hardegger, L. A.; Kuhn, B.; Spinnler, B.; Anselm, L.; Ecabert, R.; Stihle, M.; Gsell, B.; Thoma, R.; Diez, J.; Benz, J.; Plancher, J.-M.; Hartmann, G.; Banner, D. W.; Haap, W.; Diederich, F. Systematic Investigation of Halogen Bonding in Protein-Ligand Interactions. Angew. Chem., Int. Ed. 2011, 50 (1), 314−318.

(22) Yu, T.; Tagat, J. R.; Kerekes, A. D.; Doll, R. J.; Zhang, Y.; Xiao, Y.; Esposite, S.; Belanger, D. B.; Curran, P. J.; Mandal, A. K.; Siddiqui, M. A.; Shih, N.-Y.; Basso, A. D.; Liu, M.; Gray, K.; Tevar, S.; Jones, J.; Lee, S.; Liang, L.; Ponery, S.; Smith, E. B.; Hruza, A.; Voigt, J.; Ramanathan, L.; Prosise, W.; Hu, M. Discovery of a Potent, Injectable Inhibitor of Aurora Kinases Based on the Imidazo-[1,2- a ]-Pyrazine Core. ACS Med. Chem. Lett. 2010, 1 (5), 214−218. (23) Staben, S. T.; Heffron, T. P.; Sutherlin, D. P.; Bhat, S. R.; Castanedo, G. M.; Chuckowree, I. S.; Dotson, J.; Folkes, A. J.; Friedman, L. S.; Lee, L.; Lesnick, J.; Lewis, C.; Murray, J. M.; Nonomiya, J.; Olivero, A. G.; Plise, E.; Pang, J.; Prior, W. W.; Salphati, L.; Rouge, L.; Sampath, D.; Tsui, V.; Wan, N. C.; Wang, S.; Weismann, C.; Wu, P.; Zhu, B.-Y. Structure-Based Optimization of PyrazoloPyrimidine and -Pyridine Inhibitors of PI3-Kinase. Bioorg. Med. Chem. Lett. 2010, 20 (20), 6048−6051. (24) Lee, L.-P.; Tidor, B. Optimization of Binding Electrostatics: Charge Complementarity in the Barnase-Barstar Protein Complex. Protein Sci. 2001, 10 (2), 362−377. (25) Sims, P. A.; Wong, C. F.; McCammon, J. A. Charge Optimization of the Interface between Protein Kinases and Their Ligands. J. Comput. Chem. 2004, 25 (11), 1416−1429. (26) Maass, P.; Schulz-Gasch, T.; Stahl, M.; Rarey, M. ReCore: A Fast and Versatile Method for Scaffold Hopping Based on Small Molecule Crystal Structure Conformations. J. Chem. Inf. Model. 2007, 47 (2), 390−399. (27) ReCore in LeadIT; BioSolveIT GmbH: St. Augustin, Germany, 2015. (28) Lauri, G.; Bartlett, P. A. CAVEAT: A Program to Facilitate the Design of Organic Molecules. J. Comput.-Aided Mol. Des. 1994, 8 (1), 51−66. (29) Liang, G.; Aldous, S.; Merriman, G.; Levell, J.; Pribish, J.; Cairns, J.; Chen, X.; Maignan, S.; Mathieu, M.; Tsay, J.; Sides, K.; Rebello, S.; Whitely, B.; Morize, I.; Pauls, H. W. Structure-Based Library Design and the Discovery of a Potent and Selective Mast Cell β-Tryptase Inhibitor as an Oral Therapeutic Agent. Bioorg. Med. Chem. Lett. 2012, 22 (2), 1049−1054. (30) Woltering, T. J.; Wostl, W.; Hilpert, H.; Rogers-Evans, M.; Pinard, E.; Mayweg, A.; Göbel, M.; Banner, D. W.; Benz, J.; Travagli, M.; Pollastrini, M.; Marconi, G.; Gabellieri, E.; Guba, W.; Mauser, H.; Andreini, M.; Jacobsen, H.; Power, E.; Narquizian, R. BACE1 Inhibitors: A Head Group Scan on a Series of Amides. Bioorg. Med. Chem. Lett. 2013, 23 (14), 4239−4243. (31) Hilpert, H.; Guba, W.; Woltering, T. J.; Wostl, W.; Pinard, E.; Mauser, H.; Mayweg, A. V.; Rogers-Evans, M.; Humm, R.; Krummenacher, D.; Muser, T.; Schnider, C.; Jacobsen, H.; Ozmen, L.; Bergadano, A.; Banner, D. W.; Hochstrasser, R.; Kuglstatter, A.; David-Pierson, P.; Fischer, H.; Polara, A.; Narquizian, R. β-Secretase (BACE1) Inhibitors with High in Vivo Efficacy Suitable for Clinical Evaluation in Alzheimer’s Disease. J. Med. Chem. 2013, 56 (10), 3980− 3995. (32) Oehlrich, D.; Prokopcova, H.; Gijsen, H. J. M. The Evolution of Amidine-Based Brain Penetrant BACE1 Inhibitors. Bioorg. Med. Chem. Lett. 2014, 24 (9), 2033−2045. (33) Shamovsky, I.; Ripa, L.; Börjesson, L.; Mee, C.; Nordén, B.; Hansen, P.; Hasselgren, C.; O’Donovan, M.; Sjö, P. Explanation for Main Features of Structure−Genotoxicity Relationships of Aromatic Amines by Theoretical Studies of Their Activation Pathways in CYP1A2. J. Am. Chem. Soc. 2011, 133 (40), 16168−16185. (34) Pierce, A. C.; Rao, G.; Bemis, G. W. BREED: Generating Novel Inhibitors through Hybridization of Known Ligands. Application to CDK2, P38, and HIV Protease. J. Med. Chem. 2004, 47 (11), 2768− 2775. (35) Murray, C. W.; Berdini, V.; Buck, I. M.; Carr, M. E.; Cleasby, A.; Coyle, J. E.; Curry, J. E.; Day, J. E. H.; Day, P. J.; Hearn, K.; Iqbal, A.; Lee, L. Y. W.; Martins, V.; Mortenson, P. N.; Munck, J. M.; Page, L. W.; Patel, S.; Roomans, S.; Smith, K.; Tamanini, E.; Saxty, G. Fragment-Based Discovery of Potent and Selective DDR1/2 Inhibitors. ACS Med. Chem. Lett. 2015, 6 (7), 798−803. 4100

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

(36) Schmidt, C. J.; Chapin, D. S.; Cianfrogna, J.; Corman, M. L.; Hajos, M.; Harms, J. F.; Hoffman, W. E.; Lebel, L. A.; McCarthy, S. A.; Nelson, F. R.; Proulx-LaFrance, C.; Majchrzak, M. J.; Ramirez, A. D.; Schmidt, K.; Seymour, P. A.; Siuciak, J. A.; Tingley, F. D.; Williams, R. D.; Verhoest, P. R.; Menniti, F. S. Preclinical Characterization of Selective Phosphodiesterase 10A Inhibitors: A New Therapeutic Approach to the Treatment of Schizophrenia. J. Pharmacol. Exp. Ther. 2008, 325 (2), 681−690. (37) Stumpfe, D.; Bajorath, J. Exploring Activity Cliffs in Medicinal Chemistry: Miniperspective. J. Med. Chem. 2012, 55 (7), 2932−2942. (38) Proasis, version 3; Desert Scientific Software: Sydney, Australia. (39) Kuhn, B.; Fuchs, J. E.; Reutlinger, M.; Stahl, M.; Taylor, N. R. Rationalizing Tight Ligand Binding through Cooperative Interaction Networks. J. Chem. Inf. Model. 2011, 51 (12), 3180−3198. (40) Groom, C. R.; Allen, F. H. The Cambridge Structural Database in Retrospect and Prospect. Angew. Chem., Int. Ed. 2014, 53 (3), 662− 671. (41) Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; Robb, M. A.; Cheeseman, J. R.; Zakrzewski, V. G.; Montgomery, J. A.; Stratmann, R. E.; Burant, J. C.; Dapprich, S.; Millam, J. M.; Daniels, A. D.; Kudin, K. N.; Strain, M. C.; Farkas, O.; Tomasi, J.; Barone, V.; Cossi, M.; Cammi, R.; Mennucci, B.; Pomelli, C.; Adamo, C.; Clifford, S.; Ochterski, J.; Petersson, G. A.; Ayala, P. Y.; Cui, Q.; Morokuma, K.; Malick, D. K.; Rabuck, A. D.; Raghavachari, K.; Foresman, J. B.; Cioslowski, J.; Ortiz, J. V.; Stefanov, B. B.; Liu, G.; Liashenko, A.; Piskorz, P.; Komaromi, I.; Gomperts, R.; Martin, R. L.; Fox, D. J.; Keith, T.; Al-Laham, M. A.; Peng, C. Y.; Nanayakkara, A.; Gonzalez, C.; Challacombe, M.; Gill, P. M. W.; Johnson, B. G.; Chen, W.; Wong, M. W.; Andres, J. L.; Head-Gordon, M.; Replogle, E. S.; Pople, J. A. Gaussian 98, revision A.7; Gaussian, Inc.: Pittsburgh, PA, U.S., 1998. (42) Yeaman, S. J. Hormone-Sensitive Lipase - New Roles for an Old Enzyme. Biochem. J. 2004, 379 (1), 11−22. (43) Neidhart, W.; Ackermann, J.; Amrein, K.; Conte, A.; Hunziker, D.; Kuhn, B.; Märki, H. P.; Nettekoven, M.; Ruf, A.; Schulz-Gasch, T.; Mayweg, A. V. Challenges and Rewards in Medicinal Chemistry Targeting Cardiovascular and Metabolic Diseases. Chimia 2015, 69 (7), 407−413. (44) Pols, T. W. H.; Noriega, L. G.; Nomura, M.; Auwerx, J.; Schoonjans, K. The Bile Acid Membrane Receptor TGR5 as an Emerging Target in Metabolism and Inflammation. J. Hepatol. 2011, 54 (6), 1263−1272. (45) Martin, R. E.; Bissantz, C.; Gavelle, O.; Kuratli, C.; Dehmlow, H.; Richter, H. G. F.; Obst Sander, U.; Erickson, S. D.; Kim, K.; Pietranico-Cole, S. L.; Alvarez-Sánchez, R.; Ullmer, C. 2-PhenoxyNicotinamides Are Potent Agonists at the Bile Acid Receptor GPBAR1 (TGR5). ChemMedChem 2013, 8 (4), 569−576. (46) Cox, C. D.; McGaughey, G. B.; Bogusky, M. J.; Whitman, D. B.; Ball, R. G.; Winrow, C. J.; Renger, J. J.; Coleman, P. J. Conformational Analysis of N,N-Disubstituted-1,4-Diazepane Orexin Receptor Antagonists and Implications for Receptor Binding. Bioorg. Med. Chem. Lett. 2009, 19 (11), 2997−3001. (47) McGarry, J. D.; Brown, N. F. The Mitochondrial Carnitine Palmitoyltransferase System - From Concept to Molecular Analysis. Eur. J. Biochem. 1997, 244 (1), 1−14. (48) Stahl, M.; Mauser, H. Database Clustering with a Combination of Fingerprint and Maximum Common Substructure Methods. J. Chem. Inf. Model. 2005, 45 (3), 542−548. (49) Perspicace, S.; Rufer, A. C.; Thoma, R.; Mueller, F.; Hennig, M.; Ceccarelli, S.; Schulz-Gasch, T.; Seelig, J. Isothermal Titration Calorimetry with Micelles: Thermodynamics of Inhibitor Binding to Carnitine Palmitoyltransferase 2 Membrane Protein. FEBS Open Bio 2013, 3, 204−211. (50) Hattori, K.; Wakabayashi, H.; Tamaki, K. Predicting Key Example Compounds in Competitors’ Patent Applications Using Structural Information Alone. J. Chem. Inf. Model. 2008, 48 (1), 135− 142. (51) Tyrchan, C.; Boström, J.; Giordanetto, F.; Winter, J.; Muresan, S. Exploiting Structural Information in Patent Specifications for Key Compound Prediction. J. Chem. Inf. Model. 2012, 52 (6), 1480−1489.

(52) Martin, R. E.; Green, L. G.; Guba, W.; Kratochwil, N.; Christ, A. Discovery of the First Nonpeptidic, Small-Molecule, Highly Selective Somatostatin Receptor Subtype 5 Antagonists: A Chemogenomics Approach. J. Med. Chem. 2007, 50 (25), 6291−6294. (53) Reisine, T.; Bell, G. I. Molecular Biology of Somatostatin Receptors. Endocr. Rev. 1995, 16 (4), 427−442. (54) Loughlin, W. A.; Tyndall, J. D. A.; Glenn, M. P.; Fairlie, D. P. Beta-Strand Mimetics. Chem. Rev. 2004, 104 (12), 6085−6118. (55) Kratochwil, N. A.; Malherbe, P.; Lindemann, L.; Ebeling, M.; Hoener, M. C.; Mühlemann, A.; Porter, R. H. P.; Stahl, M.; Gerber, P. R. An Automated System for the Analysis of G Protein-Coupled Receptor Transmembrane Binding Pockets: Alignment, ReceptorBased Pharmacophores, and Their Application. J. Chem. Inf. Model. 2005, 45 (5), 1324−1336. (56) Martin, R. E.; Mohr, P.; Maerki, H. P.; Guba, W.; Kuratli, C.; Gavelle, O.; Binggeli, A.; Bendels, S.; Alvarez-Sánchez, R.; Alker, A.; Polonchuk, L.; Christ, A. D. Benzoxazole Piperidines as Selective and Potent Somatostatin Receptor Subtype 5 Antagonists. Bioorg. Med. Chem. Lett. 2009, 19 (21), 6106−6113. (57) Williams, A. J.; Harland, L.; Groth, P.; Pettifer, S.; Chichester, C.; Willighagen, E. L.; Evelo, C. T.; Blomberg, N.; Ecker, G.; Goble, C.; Mons, B. Open PHACTS: Semantic Interoperability for Drug Discovery. Drug Discovery Today 2012, 17 (21−22), 1188−1198. (58) Bento, A. P.; Gaulton, A.; Hersey, A.; Bellis, L. J.; Chambers, J.; Davies, M.; Kruger, F. A.; Light, Y.; Mak, L.; McGlinchey, S.; Nowotka, M.; Papadatos, G.; Santos, R.; Overington, J. P. The ChEMBL Bioactivity Database: An Update. Nucleic Acids Res. 2014, 42 (D1), D1083−D1090. (59) Mobley, D. L.; Graves, A. P.; Chodera, J. D.; McReynolds, A. C.; Shoichet, B. K.; Dill, K. A. Predicting Absolute Ligand Binding Free Energies to a Simple Model Site. J. Mol. Biol. 2007, 371 (4), 1118− 1134. (60) Chodera, J. D.; Mobley, D. L.; Shirts, M. R.; Dixon, R. W.; Branson, K.; Pande, V. S. Alchemical Free Energy Methods for Drug Discovery: Progress and Challenges. Curr. Opin. Struct. Biol. 2011, 21 (2), 150−160. (61) Wang, L.; Wu, Y.; Deng, Y.; Kim, B.; Pierce, L.; Krilov, G.; Lupyan, D.; Robinson, S.; Dahlgren, M. K.; Greenwood, J.; Romero, D. L.; Masse, C.; Knight, J. L.; Steinbrecher, T.; Beuming, T.; Damm, W.; Harder, E.; Sherman, W.; Brewer, M.; Wester, R.; Murcko, M.; Frye, L.; Farid, R.; Lin, T.; Mobley, D. L.; Jorgensen, W. L.; Berne, B. J.; Friesner, R. A.; Abel, R. Accurate and Reliable Prediction of Relative Ligand Binding Potency in Prospective Drug Discovery by Way of a Modern Free-Energy Calculation Protocol and Force Field. J. Am. Chem. Soc. 2015, 137 (7), 2695−2703. (62) Durrant, J. D.; McCammon, J. A. Molecular Dynamics Simulations and Drug Discovery. BMC Biol. 2011, 9 (1), 71. (63) Chen, Y.-C. Beware of Docking! Trends Pharmacol. Sci. 2015, 36 (2), 78−95. (64) García-Sosa, A. T.; Mancera, R. L.; Dean, P. M. WaterScore: A Novel Method for Distinguishing between Bound and Displaceable Water Molecules in the Crystal Structure of the Binding Site of Protein-Ligand Complexes. J. Mol. Model. 2003, 9 (3), 172−182. (65) Young, T.; Abel, R.; Kim, B.; Berne, B. J.; Friesner, R. A. Motifs for Molecular Recognition Exploiting Hydrophobic Enclosure in Protein−ligand Binding. Proc. Natl. Acad. Sci. U. S. A. 2007, 104 (3), 808−813. (66) Truchon, J.-F.; Pettitt, B. M.; Labute, P. A Cavity Corrected 3DRISM Functional for Accurate Solvation Free Energies. J. Chem. Theory Comput. 2014, 10 (3), 934−941. (67) Amadasi, A.; Surface, J. A.; Spyrakis, F.; Cozzini, P.; Mozzarelli, A.; Kellogg, G. E. Robust Classification of “Relevant” Water Molecules in Putative Protein Binding Sites. J. Med. Chem. 2008, 51 (4), 1063− 1067. (68) Perola, E.; Walters, W. P.; Charifson, P. S. A Detailed Comparison of Current Docking and Scoring Methods on Systems of Pharmaceutical Relevance. Proteins: Struct., Funct., Genet. 2004, 56 (2), 235−249. 4101

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102

Journal of Medicinal Chemistry

Perspective

(69) Plewczynski, D.; Łaźniewski, M.; Augustyniak, R.; Ginalski, K. Can We Trust Docking Results? Evaluation of Seven Commonly Used Programs on PDBbind Database. J. Comput. Chem. 2011, 32 (4), 742− 755. (70) Smith, R. D.; Dunbar, J. B.; Ung, P. M.-U.; Esposito, E. X.; Yang, C.-Y.; Wang, S.; Carlson, H. A. CSAR Benchmark Exercise of 2010: Combined Evaluation Across All Submitted Scoring Functions. J. Chem. Inf. Model. 2011, 51 (9), 2115−2131. (71) Smith, R. D.; Damm-Ganamet, K. L.; Dunbar, J. B.; Ahmed, A.; Chinnaswamy, K.; Delproposto, J. E.; Kubish, G. M.; Tinberg, C. E.; Khare, S. D.; Dou, J.; Doyle, L.; Stuckey, J. A.; Baker, D.; Carlson, H. A. CSAR Benchmark Exercise 2013: Evaluation of Results from a Combined Computational Protein Design, Docking, and Scoring/ Ranking Challenge. J. Chem. Inf. Model. 2015, 10.1021/acs.jcim.5b00387. (72) Schulz-Gasch, T.; Stahl, M. Scoring Functions for Protein− ligand Interactions: A Critical Perspective. Drug Discovery Today: Technol. 2004, 1 (3), 231−239. (73) Hawkins, P. C. D.; Skillman, A. G.; Nicholls, A. Comparison of Shape-Matching and Docking as Virtual Screening Tools. J. Med. Chem. 2007, 50 (1), 74−82. (74) McGaughey, G. B.; Sheridan, R. P.; Bayly, C. I.; Culberson, J. C.; Kreatsoulas, C.; Lindsley, S.; Maiorov, V.; Truchon, J.-F.; Cornell, W. D. Comparison of Topological, Shape, and Docking Methods in Virtual Screening. J. Chem. Inf. Model. 2007, 47 (4), 1504−1519. (75) Stahl, M.; Baier, S. How Many Molecules Does It Take to Tell a Story? Case Studies, Language, and an Epistemic View of Medicinal Chemistry. ChemMedChem 2015, 10 (6), 949−956.

4102

DOI: 10.1021/acs.jmedchem.5b01875 J. Med. Chem. 2016, 59, 4087−4102