Medicinal Chemistry Projects Requiring Imaginative Structure-Based

Aug 16, 2016 - He worked as a Research Scientist at various companies, where he was involved in several medicinal chemistry projects. He significantly...
1 downloads 7 Views 5MB Size
Article pubs.acs.org/accounts

Medicinal Chemistry Projects Requiring Imaginative Structure-Based Drug Design Methods Nicolas Moitessier,*,† Joshua Pottel,† Eric Therrien,‡ Pablo Englebienne,§ Zhaomin Liu,† Anna Tomberg,† and Christopher R. Corbeil∥ †

Department of Chemistry, McGill University, 801 Sherbrooke Street West, Montréal, Québec, Canada H3A 0B8 Molecular Forecaster Inc., 969 Marc-Aurèle-Fortin, Laval, Québec, Canada H7L 6H9 § Royal HaskoningDHV, Laan 1914 35, 3818 EX Amersfoort, The Netherlands ∥ Human Health Therapeutics, National Research Council Canada, 6100 Royalmount Avenue, Montréal, Québec, Canada H4P 2R2 ‡

CONSPECTUS: Computational methods for docking small molecules to proteins are prominent in drug discovery. There are hundreds, if not thousands, of documented examplesand several pertinent cases within our research program. Fifteen years ago, our first docking-guided drug design project yielded nanomolar metalloproteinase inhibitors and illustrated the potential of structure-based drug design. Subsequent applications of docking programs to the design of integrin antagonists, BACE-1 inhibitors, and aminoglycosides binding to bacterial RNA demonstrated that available docking programs needed significant improvement. At that time, docking programs primarily considered flexible ligands and rigid proteins. We demonstrated that accounting for protein flexibility, employing displaceable water molecules, and using ligand-based pharmacophores improved the docking accuracy of existing methodsenabling the design of bioactive molecules. The success prompted the development of our own program, FITTED, implementing all of these aspects. The primary motivation has always been to respond to the needs of drug design studies; the majority of the concepts behind the evolution of FITTED are rooted in medicinal chemistry projects and collaborations. Several examples follow: (1) Searching for HDAC inhibitors led us to develop methods considering drug−zinc coordination and its effect on the pKa of surrounding residues. (2) Targeting covalent prolyl oligopeptidase (POP) inhibitors prompted an update to FITTED to identify reactive groups and form bonds with a given residue (e.g., a catalytic residue) when the geometry allows it. FITTEDthe first fully automated covalent docking programwas successfully applied to the discovery of four new classes of covalent POP inhibitors. As a result, efficient stereoselective syntheses of a few screening hits were prioritized rather than synthesizing large chemical libraries yielding nanomolar inhibitors. (3) In order to study the metabolism of POP inhibitors by cytochrome P450 enzymes (CYPs) for toxicology studiesthe program IMPACTS was derived from FITTED and helped us to reveal a complex metabolism with unforeseen stereocenter isomerizations. These efforts, combined with those of other docking software developers, have strengthened our understanding of the complex drug−protein binding process while providing the medicinal chemistry community with useful tools that have led to drug discoveries. In this Account, we describe our contributions over the past 15 yearswithin their historical contextto the design of drug candidates, including BACE-1 inhibitors, POP covalent inhibitors, G-quadruplex binders, and aminoglycosides binding to nucleic acids. We also remark the necessary developments of docking programs, specifically FITTED, that enabled structure-based design to flourish and yielded multiple fruitful, rational medicinal chemistry campaigns.



method to dock to f lexible proteins.2 Concurrently, Lengauer and co-workers implemented discrete water molecules for the first time.3 The perceived potential led several groups to develop docking programs. By 2008, over 60 docking programs had been reported.4 Our early medicinal chemistry projects demonstrated that different targetslike nucleic acids (e.g., bacterial RNA, targets for antibiotics) and metalloenzymes (e.g., matrix metalloproteinases (MMPs), targets for anticancer therapeutics)and various drug classeslike highly charged aminoglycosides and metalcoordinating moleculeswere treated unequally, and often

INTRODUCTION

The foundation of this Accountand our medicinal chemistry research programis the symbiotic relationship between structure-based drug design (SBDD) techniques and the drug discovery process. The benefits of SBDD on the discovery rate are well-documented, but the motivational factors driving new concepts in SBDD are often internalized or merely attributed to the competitive nature of the field.1 In contrast, the context of our research suggests that SBDD is often refined by the needs of the medicinal chemistry projects to which they contribute. In the early 1990s, docking programs were relatively simplisticused primarily to investigate the binding of small molecules to rigid proteins. In the late 1990s, to better simulate the binding process, Totrov and Abagyan reported the first © XXXX American Chemical Society

Received: April 14, 2016

A

DOI: 10.1021/acs.accounts.6b00185 Acc. Chem. Res. XXXX, XXX, XXX−XXX

Article

Accounts of Chemical Research

Figure 1. Medicinal chemistry projects and the corresponding developments of FITTED.

Figure 2. MMP inhibitors from the SBDD campaign with AutoDock.

(bio)chemical phenomena encoded within. Developing software based on fundamental biochemical principles to solve challenges arising in experimental researchfrom which data can be used to, in turn, further improve the methodsis a cycle that is crucial to our success in medicinal chemistry. Since this Account focuses on the unmet need for certain aspects of SBDD methods in medicinal chemistry, more detailed information regarding algorithmic improvements in other software, implementations, validations, diseases, and biological targets can be found in corresponding publications.

incorrectly, by available software. Thus, we added missing features such as protein flexibility and displaceable waters to available programs for better binding simulation. Although these modifications significantly improved the docking accuracy, they also affected the performance (runtime), user-friendliness, and transferability of the enriched code. In 2007 we reported FITTED,5 a docking software integrating both protein flexibility and displaceable waters. Since then, some of the major docking programs, including Surflex,6 GOLD,7 AutoDock,8 and DOCK,9 have incorporated and improved upon these two features.10 In this Account, we wish to summarize several milestonesin the context of motivationand the current status of our research in both medicinal chemistry and docking program development (Figure 1). Ongoing medicinal chemistry projects prompted most of the improvements to the software and specifically the



MATRIX METALLOPROTEINASE INHIBITORS 1999−2001 Under Hanessian’s supervision, Moitessier and Therrien were interested in using docking programs to design MMP inhibitors B

DOI: 10.1021/acs.accounts.6b00185 Acc. Chem. Res. XXXX, XXX, XXX−XXX

Article

Accounts of Chemical Research

Figure 3. (a) Pharmacophore-oriented docking applied to integrin antagonists. (b) An example of a discovered active antagonist based on the identified pharmacophore.

as potential anticancer therapeutics.11 MMP inhibitors were designed in a traditional waywith extensive synthetic efforts at the benchbut led to no significant inhibitory potency. AutoDock3.0 and DOCK4.0 were two publicly available docking programs tested to predict the experimentally observed binding modes. When docking was used to prioritize molecules for synthesis, a number of nanomolar and subnanomolar MMP inhibitors were discovered. For example, by modification of an existing inhibitor, 1 (Figure 2), nanomolar sulfonamide derivativesexemplified by pyrrolidine-based analogue 212 and aziridine analogue 313were designed, synthesized, and biologically evaluated. AutoDock-guided studies suggested the addition of the chiral hydroxyl group in 2. The favorable hydroxyl group and the predicted favored stereochemistry were confirmed by in vitro assays. Although the synthesis of these chiral compounds was not straightforward (e.g., chiral aziridine synthesis for 3), the synthetic efforts were prioritized on the basis of AutoDock predictions. The benefits of using docking programs to guide the design of enzyme inhibitors were clearpotent inhibitors and focused (time- and cost-efficient) syntheses. Thus, we explored the use of similar programs in other medicinal chemistry projects.

docked into the apo structure, and used to orient the docking of the antagonists (Figure 3).14 Concurrently, a crystal structure of a cyclic peptide bound to this receptor was released and agreed with our proposed binding mode. A combination of this pharmacophore and focused combinatorial chemistry led to the discovery of antagonists exemplified by the xylose derivative 4 (Figure 3).14 Once more, the use of SBDD enabled us to focus on other aspects of the synthesis. In this case, to our knowledge, we reported the first combinatorial/parallel synthesis of carbohydrate-based potential drugs.



BACE-1 INHIBITORS 2002−2006: DOCKING TO FLEXIBLE PROTEINS I BACE-1 is involved in the generation of β-amyloid, which has been observed to aggregate into amyloid plaques in Alzheimer’s patients, suggesting it as a therapeutic target. With available cocrystal structures, BACE-1 was an excellent candidate for SBDD. However, the catalytic diaspartate protonation state and protein flexibility were difficult to model but critical for optimal BACE-1 inhibitor design.15,16 Using available BACE-1 crystal structures complexed with different peptides, we evaluated the accuracy of existing automated docking programs.17 Because of the flexibility of the ligands and the protein, initial docking experiments failed to provide realistic binding modes. We proposed to optimize flexible protein−ligand complexes with a genetic algorithm; conformations of both the ligand and the protein (side chains and backbone) were encoded as chromosomesproviding an accurate induced-fit docking method that was implemented within Accelrys’s Insight II. This method was successful in docking peptides cocrystallized with BACE-1 with very good accuracy, despite their large number of degrees of freedom (Figure 4). The identification of active compounds requires not only a proper binding mode but also a correct prediction of the free energy of binding, computed with a scoring function. We hypothesized that better descriptions of the contribution of water binding and of the entropic penalty upon binding would be



INTEGRIN ANTAGONISTS 2001−2003: PHARMACOPHORE-ORIENTED DOCKING In an effort to develop carbohydrate-based Arg-Gly-Asp (RGD) mimetics as integrin antagonists, a combination of computational predictions, advanced parallel and combinatorial syntheses, and cell-based assays were planned. Integrins αVβ3 and αIIbβ3 have been implicated in cancer metastasis, osteoporosis, and blood coagulation control, making them drug targets with great potential for these diseases. When a crystal structure of the unbound αVβ3 receptor was reported in the literature, AutoDock, FlexX, and DOCK were retrospectively tested to dock known antagonists, each of which featured charged functional groups (mimicking Asp and Arg residues). Unfortunately, the positively charged groups were predicted to interact with Asp150, while experiments pointed at Asp218. To correct this, a three-point pharmacophore was derived from the reported antagonists, C

DOI: 10.1021/acs.accounts.6b00185 Acc. Chem. Res. XXXX, XXX, XXX−XXX

Article

Accounts of Chemical Research

Figure 4. BACE-1/OM99-2 crystal structure (ligand, green; protein, beige; PDB code 1fkn): docked pose (purple) and 2D drawing.

necessary to discriminate between active and inactive inhibitors. A new force-field-based scoring function for BACE-1, called RankScore1 (eq 1), was developed. It included additional terms for hydrogen bonding and for the number of rotatable bondsa surrogate for entropy change upon binding:

Figure 5. Water-mediated binding of aminoglycoside to bacterial RNA.

energetically favored and disfavored waters. The accuracy of the binding mode prediction with AutoDock was significantly improved (Figure 6). This was the first step toward a novel approach for displaceable water molecules in docking programs (see below).

RankScore1 = c1EvdW + c 2Eele + c3E HB + c4Nrot + c5 (1)

Combined with our flexible protein docking protocol, RankScore1 outperformed several other methods in selecting known active compounds. The significant enrichment in active compounds demonstrated the ability of our flexible protein docking protocol to accurately identify BACE-1 inhibitors.



AMINOGLYCOSIDE ANTIBIOTICS AS BACTERIAL RNA BINDERS 2004−2006: DISPLACEABLE WATERS Another collaborative project with the Hanessian group aimed at designing aminoglycoside derivatives as potential antibiotics.18,19 Aminoglycosides bind to bacterial rRNA, consequently interfering with ribosomal protein synthesisessentially lethal to bacteria. The rise of resistant bacterial strains motivated the development of novel aminoglycosides. SBDD approaches were probed for insights. A close look at the crystal structures of aminoglycosides bound to bacterial RNA revealed that the charged ammonium groups of the antibiotics often interact with highly polarized bridging waters rather than the charged phosphate backbone (Figure 5). Computationally, when a single water molecule is involved in the binding, selecting the best pose from wet and dry docking may be sufficient. In the case of aminoglycoside−RNA complexes, however, several waters are involved. The number of combinations of waters increases exponentially, rendering the docking process highly dependent on the selection of water molecules. An alternative approachdisplacing waters during dockingwas necessary. Our approach computed the interactions between the aminoglycoside and only nonclashing crystallographic water molecules to simulate the displacement of waters clashing with aminoglycoside functional groupswithout distinction between

Figure 6. AutoDock accuracy in pose predictions with additional implementations.



FITTED 1.0 (FLEXIBILITY INDUCED THROUGH TARGETED EVOLUTIONARY DESCRIPTION) 2006−2007 At this stage, we had developed, implemented, and applied three strategies/algorithms: pharmacophore-oriented docking (integrin antagonists), docking to flexible proteins (BACE-1 inhibition), and docking in the presence of displaceable waters (aminoglycoside−RNA complexes). However, all three of these methods were implemented in various programs. We took up the challenge of creating a single integrated software and developed FITTED 1.0.5 A small set of protein−ligand complexes addressing both protein flexibility and displaceable waters was used to test the new software.5 Within this set, FITTED achieved increased accuracy over the rigid protein model. Figure 7 illustrates D

DOI: 10.1021/acs.accounts.6b00185 Acc. Chem. Res. XXXX, XXX, XXX−XXX

Article

Accounts of Chemical Research

inhibitors.21 A known motion of an α-helix of this enzyme prompted us to consider protein flexibility, a feature implemented in FITTED 1.0. To enable FITTED to perform a virtual screen (VS) on HCV polymerase, new tools were developed to reduce the size of the library and increase FITTED’s speed (Figure 8). Compounds were filtered on the basis of known reactive/toxic groups and Lipinski’s rule of five using an early version of REDUCE.22 The VS of the Maybridge Library identified 826 potential HCV binders and yielded two micromolar inhibitors.



KAINATE RECEPTOR AGONISTS 2008−2009: DOCKING TO FLEXIBLE PROTEINS II We studied the conformational changes of GluK2 kainate receptoran ionotropic glutamate receptor implicated in central nervous system disordersupon binding of full and partial agonists; the receptor can adopt an open, closed, or intermediate state. Docking of known agonists against an ensemble of crystallographic structures yielded accurate pose predictions and correctly identified the conformational state. Subsequent docking of known partial agonists showed a preference for the closed structure, when most experimentalists thought a more open structure would be preferred.23 Studies using electrophysiology confirmed their preference for the closed conformation. These investigations identified a key tyrosine in the agonist binding domain, Tyr488, that determined the open/ closed state. The information was used to design novel binders (5, Figure 9) that were predicted to variably interact with this Figure 7. Flexible docking (purple) applied to two thymidine kinase structures (crystal ligand, green; protein, beige/gray).

successful examples of features of FITTED (applied to thymidine kinase): the correct prediction of the glutamine conformation (up vs down), the occurrence of waters, and the correct ligand pose. With the development of FITTED 1.0 complete, applications followed. For example, retrospective identification of αmannosidase inhibitors was a first test of FITTED’s transferability and an opportunity to compare its accuracy to that of other docking programs.20 This study revealed that predicted zinc coordination geometries could be improved, which was eventually carried out in 2012 (see below).

Figure 9. GluK2 agonists.

tyrosine, thereby controlling the opening of the receptor.24 In vitro assays confirmed the correlation between the size of the R group and the competitive binding with glutamate and provided a better understanding of GluK2 agonism.



FITTED 2.6 2009 Following increased reports of success of docking programs from both academia and industry, we thought that it would be relevant for further improvement to capture their strengths and to illuminate sources of their failure. The impact of additional features, such as ring flexibility, were analyzed within FITTED upon re-evaluation on a larger set of complexes. This extensive



HEPATITIS C VIRUS (HCV) POLYMERASE INHIBITORS 2007−2008: VIRTUAL SCREENING In collaboration with Virochem Pharma, we proposed a structure-based identification of potential HCV polymerase

Figure 8. Filtering approach implemented in FITTED for high-throughput screening. E

DOI: 10.1021/acs.accounts.6b00185 Acc. Chem. Res. XXXX, XXX, XXX−XXX

Article

Accounts of Chemical Research study revealed that modeling of protein flexibility is critical, while, unexpectedly, displaceable waters appeared to be inconsequential.25 FITTED was comparable to the best docking programs available at this time when using rigid proteins (Figure 10). Cross-docking accuracy increased when a flexible protein was used, revealing its importancea conceptually logical conclusion.

RankScore 2(4) = c1EvdW + c 2Eele + c3E HB + c4ΔGsolvation + c5ΔGSASA + c6Nwater + c 7Erot

(2)

where EvdW, Eele, and EHB are the AMBER van der Waals, electrostatic, and hydrogen-bond protein−ligand interaction energies, respectively, ΔGsolvation is the generalized Born (GB) polar solvation free energy, ΔGSASA is the free energy related to the solvent-accessible surface area, Nwater is the number of bridging waters, Erot is a weighted score of rotatable bond types, and the cn are fitting weights. These changes to FITTED and the limited improvement demonstrated that competition is not the major driving force behind novel implementations and new discoveries. Success often plateaus in these instances, and competition unfortunately masks the true reason for failurea lack of modeling the underlying biochemistry.



G-QUADRUPLEX BINDERS 2008−2013: DOCKING TO DNA In collaboration with the Sleiman, Autexier, and Mittermaier groups, we initiated a project aimed at developing G-quadruplex stabilizers as cancer therapeutics.30−33 Small molecules that can stabilize G-quadruplexesalternate structures to the ubiquitous DNA duplexesinhibit the telomerase enzyme active in most cancer cells and block the transcriptional activity of oncogenes. Three crystal structures of G-quadruplexes with bound ligands were available. To better address the flexibility of the Gquadruplex (nucleic acids are significantly more flexible than proteins) and the energy upon binding, we opted for a combined docking−molecular dynamics (MD) simulation strategy. FITTED was used to generate reasonable starting structures for refinement by MD simulations. Platinum(II) supramolecular squares30 and platinum(II) phenanthroimidazoles32,33 were designed (Figure 12). The docked poses revealed that square structures, such as 6, bind to the top of the G-quadruplex motif. Conversely, the nitrogen-containing side chain of 7 creates an additional interaction with the phosphate backbone, making 7 the slowest-dissociating G-quadruplex binder reported.

Figure 10. Success rates of FITTED 2.6 and others at flexible docking.

A second study was carried out to determine what role these implementations played in predicting binding affinity.26 A panel of 18 commercially available scoring functions was applied to our own well-curated, challenging data set of 209 protein−ligand complexes. As in the previous study, displaceable waters did not have a large impact and were highly system-dependent.27 On the other hand, protein flexibility had an effect (either positive or negative) on a few of the scoring functions (Figure 11). This last study led to the development of two different scoring functions implemented in FITTED:29 RankScore2 for optimizing a ligand for affinity (lead optimization) and RankScore4 for discovery of active molecules (hit identification). While the physics remained the same in the two cases, the data sets required for training were different and yielded different coefficients for the same formalism (eq 2).

Figure 11. Impact of water and protein flexibility on scoring function accuracy (Kendall tau28 is a coefficient measuring the correlation between two sets of data, here predictions and observations). F

DOI: 10.1021/acs.accounts.6b00185 Acc. Chem. Res. XXXX, XXX, XXX−XXX

Accounts of Chemical Research



Article

POP INHIBITOR TOXICOLOGY 2011−2012: METABOLISM PREDICTION Before any of our POP inhibitor lead compounds shown in Figure 14 could be moved to animal models, we evaluated their metabolic stability. Compounds 8 and 11 behaved very differently in human liver microsomes (Figure 15), suggesting that the sulfur was the site of metabolism. However, from HPLC it appeared as though there were three metabolites, pointing to aromatic oxidation of the phenyl ring. This medicinal chemistry campaign led us to consider predicting drug metabolism, specifically cytochrome P450 (CYP)-mediated oxidation. 39 We envisioned a docking component to visualize the reactive state and provide insight into the site of metabolism (SoM) of drug candidates. Our approach is unique: it models the transition state (TS) of the substrate reacting with the CYP. Furthermore, we applied a reactivity-based score derived from quantum-mechanical data, resulting in IMPACTS (Figure 16), a trivalent approach (ligand reactivity, docking, TS modeling) that demonstrated great accuracy in retrospective computations.39 We challenged our program against experts in the medicinal chemistry field. IMPACTS consistently outperformed them by 6− 7%. Although we recognize this is not a significant representation of the field, it does offer some insight into the role for such software in academia and industry. Applied to the POP inhibitor metabolism (Figure 15), IMPACTS predicted that sulfur oxidation should be the only process that occurs. A combination of experiments eventually confirmed the prediction. Furthermore, there was sufficient evidence of concomitant epimerization of the stereogenic centers, explaining the three HPLC peaks.40

Figure 12. G-quadruplex binders developed through SBDD.



PROLYL OLIGOPEPTIDASE (POP) INHIBITORS 2008−2016: COVALENT DOCKING POP is a serine protease linked to neurodegenerative disorders; its inhibition was found to restore memory and reduce αsynuclein levels in vivo. As such, it was a potential target for Alzheimer’s disease therapeutics. A large fraction of the reported POP inhibitors are covalent inhibitors that react with the catalytic serine (Ser554). SBDD would be a promising approach for new drug candidates, but automated covalent dockingidentifying favorable/unfavorable covalent bondswas impossible with existing programs. This prompted us to enable FITTED to label reactive functional groups, create the covalent bond, and compute its energy in a single run.34,35 Thus, upon docking, if the functional group (aldehyde and boronate ester in Figure 13) is geometrically accessible to the reactive residue, a covalent bond is formed; otherwise, it is assumed to be non-covalent. Other programs/workflows followed our original report, including covalentDock (AutoDock)36 and Dockovalent (DOCK).37 Using this method, we designed and synthesized POP inhibitors (Figure 14)34,35 and focused our synthetic efforts on only these chemical series rather than generating vast chemical libraries. For example, inhibitor 10 was made in six steps from commercially available chemicals.38 To our satisfaction, all three series selected through docking-guided design demonstrated nanomolar activities targeting POP with selectivity over homologous enzymes (DPP enzymes and FAP) (Figure 14).



INTEGRATING COMPUTATIONAL AND MEDICINAL CHEMISTRY 2012: THE FORECASTER PLATFORM To encourage the use of SBDD by medicinal chemists, we developed a Web-based interface, FORECASTER, that allows a user to build and customize drug design workflows (Figure 17).22 The resource precluded running multiple command-line applications to perform common drug design tasks like preparing a protein structure and setting up a ligand. As a proof of concept, FORECASTER was used by a medicinal chemist in our lab to build virtual combinatorial libraries, filter them, and extract a highly diverse library from the NCI database. These focused libraries were then docked to the estrogen receptor (ER)accurately

Figure 13. Reversible covalent inhibitors (blue) and POP residues (red). G

DOI: 10.1021/acs.accounts.6b00185 Acc. Chem. Res. XXXX, XXX, XXX−XXX

Article

Accounts of Chemical Research

Figure 14. Reversible covalent POP inhibitors discovered by automated covalent docking.

Figure 15. Metabolic stability of covalent POP inhibitors.

Figure 16. Overall approach implemented into IMPACTS for site of metabolism prediction.

in computational protocols, this interaction is modeled as a purely electrostatic or van der Waals interactionexcluding the covalent nature of metal coordinationalthough the coordination geometry is sometimes used to guide the docking process. Additionally, in close proximity to the zinc ion, a basic residue (e.g., histidine in HDACs, glutamate in MMPs) is typically present and may attract an acidic proton (Figure 18). We approximated the proton shift with two static representations and neglected the minimal energy required for the transfer: if the ligand is in proximity to the zinc ion, the proton exists on the basic residue; otherwise, the proton remains on the ligand.

identifying existing ER modulators in retrospective studies. Additionally, independent investigations of radiolabeled kinase inhibitors using a few of our programs were reported.41



HDAC INHIBITORS 2013−2015: METAL COORDINATION

In collaboration with the Gleason group, we considered histone deacetylase (HDAC) inhibitors as potential cancer therapeutics.42 HDAC classes I/II are zinc metalloenzymes; strong coordination of the zinc ion is crucial for a viable inhibitor. Often H

DOI: 10.1021/acs.accounts.6b00185 Acc. Chem. Res. XXXX, XXX, XXX−XXX

Article

Accounts of Chemical Research

Accordingly, we developed new energy functions to represent the special hydrogen bond as well as the metal coordination and implemented both in FITTED. Others followed up our efforts on docking to metalloenzymes, such as AutoDock4Zn.43 These implementations improved the overall pose prediction, the zinc binding accuracy, and the discrimination between active and inactive HDAC inhibitors. Application to the design and preparation of HDAC inhibitors followed.44



FITTED AND DRUG DISCOVERY 2013−2014: VIRTUAL SCREENING

The last wide-reaching test of the docking program was outdated, and after a variety of software modifications over 5 years, a large retrospective study was logical. Additionally, it would again provide broad-stroke insights into the strengths and weaknesses of SBDD for the next half decade. Thus, FITTED 3.6 was tested extensively on a large set of protein structures. The impacts of acknowledging protein flexibility, zinc binding, and water molecules were fully evaluated.45 The importance of protein structure was clarified from docking to rigid proteins. Taking COX-2 and FGFr1 as examples (Figure 19), the measured area under the receiver operating characteristic curve (AU-ROC) ranged from very poor to high depending on the specific protein crystal structure used. A closer look at these proteins revealed that some of the conformations hinder access to the binding site or favor binding of only a certain chemical series, leading to overall poor scores for active compounds. Including protein flexibility addressed this issue in one-third of the cases. Therefore, we suggested that when the protein flexibility is unknown and/or when a diverse set of ligands is studied, including protein flexibility should increase accuracy. However, when the ligands are similar, it is best to rely on rigid docking since these ligands are expected to bind similarly to a consistent protein conformation. As in the previous studies, using displaceable waters had little effect, reinforcing its casedependent nature and the need for individual water free energies. In some regard, the strength of the core components of SBDD have plateauedthe helpful pieces remain the uncovered intricacies of specific medicinal chemistry projects.

Figure 17. Sample workflow on the FORECASTER platform.

Figure 18. An example of the binding cycle of inhibitors to metalloenzymes.

Figure 19. Dependence of VS accuracy on water mode (left) and protein structure (right). I

DOI: 10.1021/acs.accounts.6b00185 Acc. Chem. Res. XXXX, XXX, XXX−XXX

Article

Accounts of Chemical Research

Figure 20. Roadmap for the current development of FITTED.



CONCLUSION AND PERSPECTIVE Over the years, we have reported a docking program foundation, FITTED, which was initially developed for enzyme−inhibitor binding mode prediction and subsequently expanded for application to VS, metalloenzymes, nucleic acids, covalent inhibitors, and SoM prediction. All of these implementations were driven by unmet demands for modeling specific features of the drug−macromolecule binding process. Satisfying the needs with excellent accuracy resulted in several successful drug discovery research programs, whereas a competitive focus yielded little novelty. Generally, in many laboratories the applications of docking methods have evolved from SBDD to metabolism prediction, offtarget identification, and more. In principle, any biological process encountering a binding process could be investigated using docking methods. In this context, CYP inhibition, P-gp efflux, and plasma protein binding prediction could be attainable milestones (Figure 20). Overall, our studies have shown that protein flexibility improves pose prediction and active compound discovery. Additionally, more work is needed to properly consider the binding free energy of waters. Recently, the first independent comparative study using six proteins revealed that FITTED (specifically RankScore) outperformed 15 other methods.46 This independent success story, along with well over 150 licensed academic and industrial groups, demonstrates that FITTED, along with the FORECASTER platform, is a tool medicinal chemists can use in a wide variety of drug discovery challenges with meaningful results. Furthermore, it suggests that pioneering SBDD developments are achieved primarily by addressing unmet needs in medicinal chemistry.



and the asymmetric catalyst design platform VIRTUAL CHEMIST, and medicinal chemistry with the development of synthetic methodologies for drug synthesis and enzyme inhibitors as potential therapeutics. Joshua Pottel completed his B.Sc. in chemistry at McGill University in 2011 along with a minor in computer science. He completed a Ph.D. in chemistry at the same university under the supervision of Prof. Nicolas Moitessier and is pursuing postdoctoral studies with Prof. Brian Shoichet at UCSF. His research has generally focused on software development in chemo- and bioinformatics. He has developed the VIRTUAL CHEMIST platform and contributed to many of the tools incorporated into FORECASTER. His current research uses computational techniques to rationally predict relevant polypharmacology. Eric Therrien received his Ph.D. in chemistry from the Université de Montréal and postdoctoral training from McGill University. He worked as a Research Scientist at various companies, where he was involved in several medicinal chemistry projects. He significantly contributed to the design and synthesis of small-molecule inhibitors of diverse biological targets in the epigenetics area. He actively participated in the development, validation studies, and applications of the FITTED software and the FORECASTER platform. In 2010, he cofounded Molecular Forecaster with Nicolas Moitessier. Pablo Englebienne received his Ph.D. from McGill University in 2009 in the group of Prof. Nicolas Moitessier after earning a Licenciatura from Universidad de Buenos Aires (Argentina). During his doctoral research, he participated in the development of FORECASTER, particularly the training of RankScore2 and RankScore4 and the development of the REACT tool. Later, he performed postdoctoral research in the groups of Prof. Bert Meijer (TU Eindhoven, The Netherlands) on molecular modeling of covalent and supramolecular polymers and in the group of Prof. Thijs Vlugt (TU Delft, The Netherlands) on computational studies of diffusion of gases in liquids, liquid crystals, and porous membranes. Currently he is an expert in risk assessment of chemicals in The Netherlands.

AUTHOR INFORMATION

Corresponding Author

*Phone: 514-398-8543. Fax: 514-398-3797. E-mail: nicolas. [email protected].

Biographies

Zhaomin Liu completed his M.Sc. in chemistry at Southern Illinois University Edwardsville in 2011, supervised by Prof. Maria Kontoyianni, focusing on protein modeling and structure-based studies of G-protein coupled receptors. He is currently pursuing his Ph.D. with Prof. Nicolas Moitessier, focusing on drug design and software refinement targeting nucleic acids and on force field development, deriving molecular mechanics terms from fundamental principles.

Nicolas Moitessier received his Ph.D. from Université de Nancy I (France) and postdoctoral training with Prof. Stephen Hanessian in Montréal (Canada). He started an academic career in France before returning to Canada, where he was appointed as an Assistant Professor at McGill University in 2003 then as an Associate Professor in 2009. His research involves the development of computer-assisted chemistry predictions, with the computer-aided drug design platform FORECASTER

Anna Tomberg completed her B.Sc. in chemistry with a minor in computer science at McGill University in 2013. She continued her studies at the same university by joining the group of Prof. Nicolas Moitessier as a Ph.D. student. Her various research projects include fields such as metallo-organic catalysis, drug design and toxicity, and software development. Her current efforts focus on in silico prediction of reactive metabolites.

Notes

The authors declare the following competing financial interest(s): FITTED is distributed by Molecular Forecaster Inc., cofounded by E.T. and N.M. (FITTED is free to academic groups).

J

DOI: 10.1021/acs.accounts.6b00185 Acc. Chem. Res. XXXX, XXX, XXX−XXX

Article

Accounts of Chemical Research

(14) Moitessier, N.; Henry, C.; Chapleur, Y.; Maigret, B. Combining pharmacophore search, automated docking, and molecular dynamics simulations as a novel strategy for flexible docking. Proof of concept: Docking of arginine-glycine-aspartic acid-like compounds into the avb3 binding site. J. Med. Chem. 2004, 47, 4178−4187. (15) Polgár, T.; Keseru, G. M. Structure-based β-secretase (BACE1) inhibitors. Curr. Pharm. Des. 2014, 20, 3373−3379. (16) Cosconati, S.; Marinelli, L.; Di Leva, F. S.; La Pietra, V.; De Simone, A.; Mancini, F.; Andrisano, V.; Novellino, E.; Goodsell, D. S.; Olson, A. J. Protein flexibility in virtual screening: The BACE-1 case study. J. Chem. Inf. Model. 2012, 52, 2697−2704. (17) Moitessier, N.; Therrien, E.; Hanessian, S. A method for inducedfit docking, scoring, and ranking of flexible ligands. Application to peptidic and pseudopeptidic β-secretase (BACE 1) inhibitors. J. Med. Chem. 2006, 49, 5885−5894. (18) Hanessian, S.; Tremblay, M.; Kornienko, A.; Moitessier, N. Design, modeling and synthesis of functionalized paromamine analogs. Tetrahedron 2001, 57, 3255−3265. (19) Moitessier, N.; Westhof, E.; Hanessian, S. Docking of aminoglycosides to hydrated and flexible RNA. J. Med. Chem. 2006, 49, 1023−1033. (20) Englebienne, P.; Fiaux, H.; Kuntz, D. A.; Corbeil, C. R.; GerberLemaire, S.; Rose, D. R.; Moitessier, N. Evaluation of Docking Programs for Predicting Binding of Golgi alpha-Mannosidase II Inhibitors: A Comparison with Crystallography. Proteins: Struct., Funct., Genet. 2007, 69, 160−176. (21) Corbeil, C. R.; Englebienne, P.; Yannopoulos, C. G.; Chan, L.; Das, S. K.; Bilimoria, D.; L’Heureux, L.; Moitessier, N. Docking Ligands into Flexible and Solvated Macromolecules. 2. Development and Application of FITTED 1.5 to the Virtual Screening of Potential HCV Polymerase Inhibitors. J. Chem. Inf. Model. 2008, 48, 902−909. (22) Therrien, E.; Englebienne, P.; Arrowsmith, A. G.; MendozaSanchez, R.; Corbeil, C. R.; Weill, N.; Campagna-Slater, V.; Moitessier, N. Integrating medicinal chemistry, organic/combinatorial chemistry, and computational chemistry for the discovery of selective estrogen receptor modulatorswith FORECASTER, a novel platform for drug discovery. J. Chem. Inf. Model. 2012, 52, 210−224. (23) Fay, A. M. L.; Corbeil, C. R.; Brown, P.; Moitessier, N.; Bowie, D. Functional characterization and in silico docking of full and partial GluK2 kainate receptor agonists. Mol. Pharmacol. 2009, 75, 1096−1107. (24) Schiavini, P.; Dawe, G. B.; Bowie, D.; Moitessier, N. Discovery of novel small-molecule antagonists for GluK2. Bioorg. Med. Chem. Lett. 2015, 25, 2416−2420. (25) Corbeil, C. R.; Moitessier, N. Docking Ligands into Flexible and Solvated Macromolecules. 3. Impact of Input Ligand Conformation, Protein Flexibility, and Water Molecules on the Accuracy of Docking Programs. J. Chem. Inf. Model. 2009, 49, 997−1009. (26) Englebienne, P.; Moitessier, N. Docking ligands into flexible and solvated macromolecules. 4. Are popular scoring functions accurate for this class of proteins? J. Chem. Inf. Model. 2009, 49, 1568−1580. (27) Verdonk, M. L.; Chessari, G.; Cole, J. C.; Hartshorn, M. J.; Murray, C. W.; Nissink, J. W. M.; Taylor, R. D.; Taylor, R. Modeling water molecules in protein-ligand docking using GOLD. J. Med. Chem. 2005, 48, 6504−6515. (28) Kendall, M. A New Measure of Rank Correlation. Biometrika 1938, 30, 81−93. (29) Englebienne, P.; Moitessier, N. Docking ligands into flexible and solvated macromolecules. 5. Force-field-based prediction of binding affinities of ligands to proteins. J. Chem. Inf. Model. 2009, 49, 2564− 2571. (30) Kieltyka, R.; Englebienne, P.; Fakhoury, J.; Autexier, C.; Moitessier, N.; Sleiman, H. F. A platinum supramolecular square as an effective G-quadruplex binder and telomerase inhibitor. J. Am. Chem. Soc. 2008, 130, 10040−10041. (31) Kieltyka, R.; Fakhoury, J.; Moitessier, N.; Sleiman, H. F. Platinum phenanthroimidazole complexes as G-quadruplex DNA selective binders. Chem. - Eur. J. 2008, 14, 1145−1154. (32) Castor, K. J.; Mancini, J.; Fakhoury, J.; Weill, N.; Kieltyka, R.; Englebienne, P.; Avakyan, N.; Mittermaier, A.; Autexier, C.; Moitessier,

Christopher R. Corbeil received his Ph.D. from McGill University under the supervision of Prof. Nicolas Moitessier. After earning his Ph.D., he joined the National Research Council Canada (NRC) as a Research Associate under the supervision of Dr. Enrico Purisima, investigating solvation and protein loop searching. Following his time at NRC, he joined Chemical Computing Group as a Research Scientist developing tools for biologics design. After 4 years, he decided to leave private industry and rejoin NRC as a Research Officer focusing on investigating tools for computer-aided biologics design.



ACKNOWLEDGMENTS We acknowledge NSERC (Discovery Grants), CIHR (Operating Grants and Drug Discovery Training Program Scholarships to J.P., Z.L., and A.T.), and FRQ-NT (scholarships to J.P.) as well as our industrial partners, ViroChem Pharma, Montréal, and AstraZeneca R&D Montréal, for financial support and fruitful discussions. We thank all of the researchers who contributed to the developments and applications of FITTED.

■ ■

DEDICATION This Account is dedicated to Prof. Steve Hanessian, a mentor and a friend. REFERENCES

(1) Special issue on 2014 Community Structure−Activity Resource (CSAR) benchmark exercises: J. Chem. Inf. Model. 2016, 56, 951−1099. (2) Totrov, M.; Abagyan, R. Flexible protein-ligand docking by global energy optimization in internal coordinates. Proteins: Struct., Funct., Genet. 1997, 29, 215−220. (3) Rarey, M.; Kramer, B.; Lengauer, T. The particle concept: Placing discrete water molecules during protein- ligand docking predictions. Proteins: Struct., Funct., Genet. 1999, 34, 17−28. (4) Moitessier, N.; Englebienne, P.; Lee, D.; Lawandi, J.; Corbeil, C. R. Towards the development of universal, fast and highly accurate docking/scoring methods: A long way to go. Br. J. Pharmacol. 2008, 153, S7−S26. (5) Corbeil, C. R.; Englebienne, P.; Moitessier, N. Docking ligands into flexible and solvated macromolecules. 1. Development and validation of FITTED 1.0. J. Chem. Inf. Model. 2007, 47, 435−449. (6) Jain, A. N. Surflex-Dock 2.1: Robust performance from ligand energetic modeling, ring flexibility, and knowledge-based search. J. Comput.-Aided Mol. Des. 2007, 21, 281−306. (7) Verdonk, M. L.; Chessari, G.; Cole, J. C.; Hartshorn, M. J.; Murray, C. W.; Nissink, J. W. M.; Taylor, R. D.; Taylor, R. Modeling water molecules in protein-ligand docking using GOLD. J. Med. Chem. 2005, 48, 6504−6515. (8) Morris, G. M.; Huey, R.; Lindstrom, W.; Sanner, M. F.; Belew, R. K.; Goodsell, D. S.; Olson, A. J. Software news and updates AutoDock4 and AutoDockTools4: Automated docking with selective receptor flexibility. J. Comput. Chem. 2009, 30, 2785−2791. (9) Dock, version 6.7; University of California: San Francisco, CA, 2015; http://dock.compbio.ucsf.edu/DOCK_6/index.htm (accessed July 25, 2016). (10) Corbeil, C. R.; Therrien, E.; Moitessier, N. Modeling reality for optimal docking of small molecules to biological targets. Curr. Comput.Aided Drug Des. 2009, 5, 241−263. (11) Hanessian, S.; Moitessier, N.; Therrien, E. A comparative docking study and the design of potentially selective MMP inhibitors. J. Comput.Aided Mol. Des. 2001, 15, 873−881. (12) Hanessian, S.; MacKay, D. B.; Moitessier, N. Design and synthesis of matrix metalloproteinase inhibitors guided by molecular modeling. Picking the S1 pocket using conformationally constrained inhibitors. J. Med. Chem. 2001, 44, 3074−3082. (13) Hanessian, S.; Moitessier, N.; Cantin, L. D. Design and synthesis of MMP inhibitors using N-arylsulfonylaziridine hydroxamic acids as constrained scaffolds. Tetrahedron 2001, 57, 6885−6900. K

DOI: 10.1021/acs.accounts.6b00185 Acc. Chem. Res. XXXX, XXX, XXX−XXX

Article

Accounts of Chemical Research N.; Sleiman, H. F. Platinum(II) phenanthroimidazoles for targeting telomeric G-quadruplexes. ChemMedChem 2012, 7, 85−94. (33) Castor, K. J.; Liu, Z.; Fakhoury, J.; Hancock, M. A.; Mittermaier, A.; Moitessier, N.; Sleiman, H. F. A platinum(ii) phenylphenanthroimidazole with an extended side-chain exhibits slow dissociation from a c-kit G-quadruplex motif. Chem. - Eur. J. 2013, 19, 17836−17845. (34) Lawandi, J.; Toumieux, S.; Seyer, V.; Campbell, P.; Thielges, S.; Juillerat-Jeanneret, L.; Moitessier, N. Constrained Peptidomimetics Reveal Detailed Geometric Requirements of Covalent Prolyl Oligopeptidase Inhibitors. J. Med. Chem. 2009, 52, 6672−6684. (35) De Cesco, S.; Deslandes, S.; Therrien, E.; Levan, D.; Cueto, M.; Schmidt, R.; Cantin, L. D.; Mittermaier, A.; Juillerat-Jeanneret, L.; Moitessier, N. Virtual screening and computational optimization for the discovery of covalent prolyl oligopeptidase inhibitors with activity in human cells. J. Med. Chem. 2012, 55, 6306−6315. (36) Ouyang, X.; Zhou, S.; Su, C. T. T.; Ge, Z.; Li, R.; Kwoh, C. K. CovalentDock: Automated covalent docking with parameterized covalent linkage energy estimation and molecular geometry constraints. J. Comput. Chem. 2013, 34, 326−336. (37) London, N.; Miller, R. M.; Krishnan, S.; Uchida, K.; Irwin, J. J.; Eidam, O.; Gibold, L.; Cimermančič, P.; Bonnet, R.; Shoichet, B. K.; Taunton, J. Covalent docking of large libraries for the discovery of chemical probes. Nat. Chem. Biol. 2014, 10, 1066−1072. (38) Mariaule, G.; De Cesco, S.; Airaghi, F.; Kurian, J.; Schiavini, P.; Rocheleau, S.; Huskić, I.; Auclair, K.; Mittermaier, A.; Moitessier, N. 3Oxo-hexahydro-1H-isoindole-4-carboxylic Acid as a Drug Chiral Bicyclic Scaffold: Structure-Based Design and Preparation of Conformationally Constrained Covalent and Noncovalent Prolyl Oligopeptidase Inhibitors. J. Med. Chem. 2016, 59, 4221−4234. (39) Campagna-Slater, V.; Pottel, J.; Therrien, E.; Cantin, L.-D.; Moitessier, N. Development of a Computational Tool to Rival Experts in the Prediction of Sites of Metabolism of Xenobiotics by P450s. J. Chem. Inf. Model. 2012, 52, 2471−2483. (40) Schiavini, P.; Pottel, J.; Moitessier, N.; Auclair, K. Metabolic Instability of Cyanothiazolidine-Based Prolyl Oligopeptidase Inhibitors: A Structural Assignment Challenge and Potential Medicinal Chemistry Implications. ChemMedChem 2015, 10, 1174−1183. (41) Bernard-Gauthier, V.; Aliaga, A.; Aliaga, A.; Boudjemeline, M.; Hopewell, R.; Kostikov, A.; Rosa-Neto, P.; Thiel, A.; Schirrmacher, R. Syntheses and evaluation of carbon-11- and fluorine-18-radiolabeled pan-tropomyosin receptor kinase (Trk) inhibitors: Exploration of the 4aza-2-oxindole scaffold as Trk PET imaging agents. ACS Chem. Neurosci. 2015, 6, 260−276. (42) Pottel, J.; Therrien, E.; Gleason, J. L.; Moitessier, N. Docking ligands into flexible and solvated macromolecules. 6. Development and application to the docking of HDACs and other zinc metalloenzymes inhibitors. J. Chem. Inf. Model. 2014, 54, 254−265. (43) Santos-Martins, D.; Forli, S.; Ramos, M. J.; Olson, A. J. AutoDock4Zn: An Improved AutoDock Force Field for Small-Molecule Docking to Zinc Metalloproteins. J. Chem. Inf. Model. 2014, 54, 2371− 2379. (44) Mendoza-Sanchez, R.; Cotnoir-White, D.; Kulpa, J.; Jutras, I.; Pottel, J.; Moitessier, N.; Mader, S.; Gleason, J. L. Design, synthesis and evaluation of antiestrogen and histone deacetylase inhibitor molecular hybrids. Bioorg. Med. Chem. 2015, 23, 7597−7606. (45) Therrien, E.; Weill, N.; Tomberg, A.; Corbeil, C. R.; Lee, D.; Moitessier, N. Docking ligands into flexible and solvated macromolecules. 7. Impact of protein flexibility and water molecules on docking-based virtual screening accuracy. J. Chem. Inf. Model. 2014, 54, 3198−3210. (46) Xu, W.; Lucke, A. J.; Fairlie, D. P. Comparing sixteen scoring functions for predicting biological activities of ligands for protein targets. J. Mol. Graphics Modell. 2015, 57, 76−88.

L

DOI: 10.1021/acs.accounts.6b00185 Acc. Chem. Res. XXXX, XXX, XXX−XXX