Chem Biol Drug Des 2015; 85: 22–29 Special Issue Antibacterials
Computational Methods to Identify New Antibacterial Targets Martin J. McPhillie†,‡, Ricky M. Cain, Sarah Narramore, Colin W. G. Fishwick and Katie J. Simmons*,‡
the late 1930s with the production of penicillin and ending with the discovery of daptomycin in 1969, this golden age produced every class of antibiotic prescribed today (3,4).
School of Chemistry, University of Leeds, Leeds, LS2 9JT, UK *Corresponding author: Katie J. Simmons, [email protected] † Present address: Department of Molecular Biology & Biotechnology, University of Sheffield, Sheffield S10 2TN, UK ‡ These authors contributed equally to this work.
During this time, the actinomycetes provided a great source of antibacterial agents, discovered through the empirical screening of fermentation broths and extracts of microorganisms containing secondary metabolites which inhibited bacterial growth (3). Nearly all antibiotic classes used today were discovered from the actinomycetes during this golden age. Once dereplication became commonplace, emphasis shifted toward new approaches (3,5). During the 1970s, new antibiotics were discovered by synthesizing analogues of existing antibacterial drug classes, with improved wholecell activity and pharmacokinetic properties. These analogues were also able to evade some existing resistance mechanisms, for example, the introduction of the fluoroquinolone antibiotics superseded the first-in-class quinolone, nalidixic acid (5). However, as resistance mechanisms arose which compromised entire classes, such as the metallo blactamases which inactivate the b-lactam antibiotics (6), a new paradigm of drug discovery, the genomic-based approach, was introduced in the 1990s (3).
The development of resistance to all current antibiotics in the clinic means there is an urgent unmet need for novel antibacterial agents with new modes of action. One of the best ways of finding these is to identify new essential bacterial enzymes to target. The advent of a number of in silico tools has aided classical methods of discovering new antibacterial targets, and these programs are the subject of this review. Many of these tools apply a cheminformatic approach, utilizing the structural information of either ligand or protein, chemogenomic databases, and docking algorithms to identify putative antibacterial targets. Considering the wealth of potential drug targets identified from genomic research, these approaches are perfectly placed to mine this rich resource and complement drug discovery programs. Key words: cheminformatics, proteomics, therapeutic target
molecular
modeling,
Received 30 April 2014, revised 17 June 2014 and accepted for publication 18 June 2014
The ‘golden age’ of antibiotics At the turn of the 20th century, bacterial infections were among the highest causes of morbidity and mortality worldwide. Over the next 70 years, antibacterial agents would reduce the burden of infectious disease and increase life expectancy to such an extent that the US Surgeon General William H. Stewart declared in 1969: ‘. . .that we had essentially defeated infectious diseases and could close the book on them [infectious diseases]. . . ‘(1, 2)
Many of the pharmaceutical companies concurred and concentrated their efforts on other therapeutic areas, ending a period of research which would be later known as the ‘golden age’ of antibacterial drug discovery. Beginning in 22
The genomic era Applying this genomic approach to antibacterial drug discovery required sequencing the genomes of pathogens to identify conserved genes essential for their survival. The identified genes which encoded targets that generally lacked a mammalian counterpart were then used in high-throughput screens (HTS) involving chemical and combinatorial libraries to identify inhibitors of these target enzymes. Unfortunately, this target-based genomic approach has failed to deliver the number of antibacterial hit molecules that the approach promised, especially when compared with the successes seen in other therapeutic areas (7). This has translated to fewer antibacterial lead compounds and is yet to deliver a clinical candidate. Structural genomic approaches that build upon target-based genomics, such as structure-based drug discovery and in silico target identification, are hoping to revitalize the future of antibiotic discovery (8).
The ideal bacterial target A bacterial enzyme must fulfill several criteria to be classified as a suitable antibacterial drug target: (i) It is essential ª 2014 John Wiley & Sons A/S. doi: 10.1111/cbdd.12385
In silico Antibacterial Target Identification
for cell survival (i.e. enzyme inhibition results in inhibition of cell growth, death or lysis), (ii) there is no mammalian homologue with the same or similar function, (iii) it is highly conserved among bacteria species to give a useful antibacterial spectrum, and (iv) it is ‘druggable’, in that a small molecule could bind to a suitable site and exert a biological effect (3). Compounds that inhibit such an enzyme would be expected to be efficacious and selective for its target protein, as well as displaying Gram-positive and/or Gram-negative antibacterial activity. Other criteria which are non-essential but desirable are as follows: (v) a soluble and stable enzyme giving high-quality and well-diffracting crystals for X-ray crystallography and (vi) an innately low frequency of mutation so that a single point mutation does not render an inhibitor scaffold ineffective when used in monotherapy. Often choosing a suitable target is the ratelimiting step of an antibacterial drug discovery program (3). Until 2000, only ~500 drug targets had been reported, among which only 120 targets had associated drugs used in the clinic (9). The completion of the human genome and numerous pathogen genomes suggests that there are up to 40 000 genes and at least the same number of proteins, and many of which are potential targets for drug discovery (10). This is a reservoir for drug discovery and target identification, although rational approaches to mine this resource are needed (11). For example, during the genomics era, many new ‘druggable’ targets, such as FabI and FabH (7), were identified but as we have failed to convert these targets into new lead compounds, there has been a recent shift back to classical phenotypic whole-cell screens and the re-evaluation of old abandoned lead compounds. However, the whole-cell ‘black-box’ approaches exclude the use of a number of modern medicinal chemistry techniques such the rational identification of novel antibacterial targets (12). Antibacterial drug discovery programmes can either find new ways to inhibit old ‘classical’ targets or identify new ‘druggable’ targets. This review will cover recent developments in antibacterial target identification using computational methods. In particular, the review will focus on cheminformatic approaches to general target identification, including homology modeling, and pharmacophore and QSAR techniques, with applied examples from antibacterial research. In addition to primary target identification, many of these techniques are able to anticipate any offtarget effects, establishing a bioactivity spectrum of a particular molecule. Various drug target identification techniques have been developed by analyzing disease relevance, functional roles, expression profiles, and lossof-function genetics between normal and disease states (13). Most of these methods are based on the detection of sequences and functional similarity to know antibacterial targets. Parameters describing polar and apolar surface areas, surface complexity, and pocket dimensions have also been used (14). Chem Biol Drug Des 2015; 85: 22–29
Cheminformatic Approaches Toward Target Identification Many of the in silico approaches to target prediction can be classified as using either cheminformatic (structural knowledge of ligands or proteins) or bioinformatic (synergy maps/biological network models) (12) tools. A number of cheminformatic target prediction approaches exist, for example, those based on docking (15), pharmacophore (16), or ligand- and structure-based target prediction methods. In silico methods can exploit prior knowledge of ligand–target interactions, collected from the literature which is further organized in chemogenomic databases. These data are then analyzed to make knowledge-based predictions for new, untested molecules or to suggest new drug–target interactions for already marketed compounds. Ligand-based methods use the concept of shape similarity, where chemical compounds of similar shape tend to have similar biological activities (17). As such query compounds, without a known target, can be screened for shape similarity against known enzyme inhibitors present in biological databases [e.g. PubChem (18), ChEMBl (19), and KiDB (20)] to search for a putative protein targets. This method is generally fast but relies on prior knowledge of the ligands and protein structures, and is not applicable in situations where no ligand exists. Structure-based methods use the structural knowledge of proteins to predict ligand–protein interactions, where a query compound is docked against a panel of protein targets or receptors using scoring functions (termed ‘reverse’ or ‘inverse’ docking). These techniques have not progressed as rapidly as traditional molecular docking approaches and suffer from significantly increased computational time and unreliable results (21). A number of these cheminformatic tools are discussed in further detail below, in an antibacterial context. Reverse docking using potential drug target database The web-accessible potential drug target database (PDTD) is a dual-function database that associates an informatics database to a structural database of recognized and potential drug targets with known 3-D-structural information. PDTD contains 1207 entries covering 841 known and potential drug targets, with structures from the Protein Data Bank (PDB) (11). Drug targets of PDTD were categorized into types according to two conditions: therapeutic areas and biochemical criteria. The databases support an extensive searching function using PDB ID, target name and category, and related disease (Figure 1). As the PDTD is designed to search the probable binding proteins for new active compounds or existing drugs using reverse docking, it only contains proteins with known 3-D 23
McPhillie et al.
Figure 1: Overview of PDTD application (adapted from Gao et al. 11).
structures determined experimentally by the X-ray crystallographic or NMR methods. The co-ordinates of proteins were isolated from the PDB. As not all PDB structures are of equal quality, a protein structure is selected according to the following criteria when it has several redundant records in PDB: (i) select the structure without mutation and missing residues around the active site; (ii) select the structure with high resolution; and (iii) select the structure complexed with a ligand. Combined with the integrated reverse docking server TarFisDock (Target Fishing Docking) (22), PDTD has been widely used to identify binding proteins for small molecules of therapeutic relevance. The binding proteins for several molecules have been verified through bioassays and crystal structure determination for target–ligand complexes. Specifically, the PDTD was searched by TarFisDock using the antibacterial agent N-trans-caffeoyltyramine (1 Figure 2), a natural product discovered by an antibacterial screen against Helicobacter pylori. A homology search revealed that, among the fifteen protein candidates discovered by reverse docking, diaminopimelate decarboxylase and peptide deformylase (PDF) are putative binding proteins for compound 1. Enzymatic assay demonstrated compound 1 and its derivative compound 2 are the potent inhibitors against the H. pylori PDF (HpPDF) with IC50 values of 10.8 and 1.25 lM, respectively. X-ray crystal structures of HpPDF and the complexes of HpPDF with 1 and 2 were determined, indicating that these two inhibitors bind well with the HpPDF binding pocket (23). To exemplify the application
of PDTD combining with TarFisDock, this example has been uploaded to the PDTD homepage under the ‘Benchmark’ option.
INVDOCK—automated detection of protein and nucleic acid targets by inverse docking INVDOCK (24) has been developed by the Bioinformatics and Drug Design group (BIDD) at the National University of Singapore for the computer-automated identification of potential protein and nucleic acid targets of drugs, newly designed drug candidate, natural product, or other chemical compound. In addition to suggesting likely therapeutic targets of a query molecule, INVDOCK can also predict potential toxicity due to off-target effects (25). INVDOCK is designed to automatically search every entry of a protein cavity database containing 9000 protein and nucleic acid entries. All parts of the cavities are subject to docking analysis. To save CPU time, the program proceeds to the next protein cavity once a successful dock is obtained without doing exhaustive dockings within that particular cavity. Although the optimized docking pose is not sought, the prioritized search complex has been shown to find a pose close to the observed binding mode seen for the studied protein–ligand complexes. The ligands are flexibly docked by matching the ligand atoms to the center of spheres followed by a conformational optimization. Scoring of docked molecules is based on a ligand–protein interaction energy function DELPa (Figure 3).
Figure 2: Structure of Helicobacter pylori inhibitors identified using TarFisDock.
24
Chem Biol Drug Des 2015; 85: 22–29
In silico Antibacterial Target Identification
Figure 3: Overview of INVDOCK protocol.
A number of studies (15) have been carried out to validate INVDOCK. Using a group of nine known drugs, INVDOCK was able to successfully predict their known therapeutic targets, and in a separate test on a group of eight drugs, INVDOCK successfully predicted 38 known toxic sideeffects from off-target interactions, missing only five. INVDOCK has also used extensively to identify the biological target(s) of traditional Chinese medicines such as genistein (26) and natural products (27). As many of our current antibiotics are derived from natural products, INVDOCK could facilitate the rational identification of protein targets of new secondary metabolites discovered via marine sources.
UniDrug target–target identification in pathogenic bacteria The web server UniDrug-Target (24) has been specifically designed to combine bacterial biological information and computational methods to stringently identify bacterial pathogen-specific proteins as drug targets. The server compares the proteome sequences of pathogenic and non-pathogenic bacteria, and mammalian proteomes, to identify unique proteins present in pathogenic bacteria. Each potential drug target is ranked on its functional importance within the cell, for example, determining whether the protein is essential for the cell’s survival. Besides predicting pathogen-specific proteins, three new algorithms have been developed and implemented using protein sequences, domains, structures, and metabolic reactions for construction of partial metabolic networks (PMNs), determination of conservation in critical residues, and variation analysis of residues forming similar cavities in proteins sequences. Chem Biol Drug Des 2015; 85: 22–29
The UniDrug-Target server is capable of predicting drug targets for any sequenced pathogenic bacteria that have FASTA sequences and annotated information. The utility of this server was demonstrated for Mycobacterium tuberculosis (H37Rv). The UniDrug-Target server identified 265 mycobacteria pathogen-specific proteins, including 17 essential proteins which are potential drug targets. As novel pathogen-specific drug targets may have no intrinsic resistance mechanisms, future output from UniDrugTarget could be valuable in combating bacterial resistance. The server is freely available at http://117.211.115.67/ UDT/main.html. The stand-alone application (source codes) is available at http://www.bioinformatics.org/ftp/ pub/bioinfojuit/UDT.rar (Table 1).
Target identification in phenotypic bacterial growth inhibitions screens A major bottleneck in phenotypic bacterial screening is determining the mode of action of any ‘hit’ compounds, whether it is due to target-specific activity or whole-cell toxicity. In silico target prediction was used to elucidate the mode of action of a series of hit’ compounds using data mining methods (21). A total of 7812 compounds within the MDL DrugDatabaseReport (MDDR) database that were identified as ‘Antibiotic’, ‘Antibacterial’, or ‘Antifungal’ via a phenotypic screen were collated and analyzed computationally using ECFP circular fingerprints (representation of molecular structures by atom neighborhoods) and a Bayesian model (statistical model representing variables and their dependencies via a directed acyclic graph). For each compound, all targets with a Bayes score 25
McPhillie et al. Table 1: List of all applications discussed in this review detailing their source and availability Application
Source
Availability
ChEMBl Insight II
https://www.ebi.ac.uk/chembl/ http://accelrys.com/
INVDOCK
http://bidd.nus.edu.sg/group/softwares/invdock.htm
KiDB MOE
http://pdsp.med.unc.edu/kidb.php http://www.chemcomp.com/MOE-Molecular_Operating_Environment.htm
PDTD Database
http://www.dddc.ac.cn/pdtd/index.php
Phyre2
http://www.sbg.bio.ic.ac.uk/phyre2/html/page.cgi?id=index
Protein Data Bank PubChem SWISS-MODEL TarFisDock
http://www.rcsb.org/pdb/home/home.do
Freely available A number of packages available-cost varies Freely available upon registration for academic groups Freely available A number of packages available-cost varies Freely available upon registration for academic groups Freely available for academic groups only Freely available
UniDrug-Target
http://117.211.115.67/UDT/main.html (server) http://www.bioinformatics.org/ftp/pub/bioinfojuit/UDT.rar. (standalone application)
https://pubchem.ncbi.nlm.nih.gov/ http://swissmodel.expasy.org/ http://www.dddc.ac.cn/tarfisdock/
of > 30 were considered likely targets and further investigated. The study concluded that likely targets could be assigned for 6874 compounds, the most prevalent being DNA gyrase and beta-lactamases. The identification of a number of potential human targets allows the users to predict off-target effects and mitigate their effects early in the discovery process.
Structure-based classification of antibacterial activity It is known that antibacterial agents occupy a unique area of chemical space compared with that for other therapeutic agents (28). Quantitative structure–activity relationship (QSAR) studies have been undertaken to predict antibacterial activity within large libraries of compounds and therefore enhance in silico screening in antibacterial drug discovery programs. One study (29) established 167 2-D physicochemical and structural descriptors of 667 known drug molecules and used two statistical techniques (linear discriminant analysis and binary logistic regression) to separate them into two groups: one with antibacterial activity and one without. The authors found that six descriptors, accounting for hydrophobicity and inter- and intramolecular hydrogen bonding, were optimal in separating the two groups. The study concluded that the most significant parameter was the octanol/water partition coefficient (ClogP), suggesting that antibacterial compounds have a lower ClogP compared with other drugs. Both statistical models were highly successful at predicting antibacterial activity, therefore providing the means of screening 26
Freely available Freely available Freely available upon registration for academic groups Freely available
databases to identify antibacterial compounds on the basis easily calculated properties (29).
Support vector machine-based methods Support Vector Machines are supervised learning models with associated learning algorithms that analyze data and recognize patterns. Support Vector Machine (SVM)-based methods have been developed from amino acid sequence-derived properties of known antibacterial targets, employing a set of features of physicochemical properties directly calculated from primary protein sequence (30,31). In one example from Rashid et al., the models were trained and tested on 852 mycobacterial proteins and evaluated using a fivefold cross-validation technique. First, the support vector machine model was developed using the amino acid composition and an overall accuracy of 83% was achieved, with average accuracy of 68%. To utilize evolutionary information, a SVM model was developed using position-specific scoring matrix profiles obtained from PSI-BLAST. The overall accuracy achieved was of 87% with average accuracy of 74%. In addition, a hidden Markov model, MEME/MAST (Multiple Em for Motif Elicitation/Motif Alignment and Search Tool) and a hybrid model that combined two or more models was also developed. The authors achieved maximum overall accuracy of 87% with average accuracy of 89% using combination of PSSM based SVM model and MEME/MAST. These methods can successfully distinguish known drug targets from putative non-drug targets at an accuracy of 84% in 10-fold cross-validation test. Chem Biol Drug Des 2015; 85: 22–29
In silico Antibacterial Target Identification
Homology Modeling For cases in which the crystal structure of the particular bacterial protein is not available, construction of a homology model is often possible, provided that a crystal structure is available for a protein with substantial sequence similarity to the protein of interest. Advanced homology models use experimentally determined structures to predict the conformation of another protein that has a similar amino acid sequence. Homology modeling involves four steps: fold assignment, sequence alignment, model building, and model refinement. Several computer packages are available to perform this process automatically, for example, SWISS-MODEL (32), Phyre2 (33), and MOE (34). SWISS-MODEL and Phyre2 are freely available to academic groups. An excellent reference source for a number of different homology modeling servers can be found at: http://proteopedia.org/wiki/index.php/Homology_modeling_servers. Homology modeling is useful when a protein sequence identity is greater than ~30% is available to model with, although >50% is preferable, especially around the ligand binding site of interest. The major difficulties in producing homology models lie in the accurate prediction of flexible loops, although a large amount of research is currently being carried out to improve the accuracy of these models (35). A number of groups have generated homology models to facilitate screening of ligands for new antibacterial targets (37). The methionyl tRNA synthetase (MetRS) is strictly required for protein biosynthesis, and success was reported in developing antibacterial agents to inhibit this enzyme. A homology model of Clostridium difficile MetRS was constructed using Molecular Operating Environment (MOE) software. Aquifex aeolicus MetRS was the main template
while the query zinc binding domain was modeled using Thermus thermophilus MetRS. The model has been assessed and compared with its main template (Ramachandran, ERRAT and ProSA). The active site of the query protein has been predicted from its sequence using a detailed conservation analysis (ClustalW2). Using MOE software, suitable ligands were docked in the constructed model, including a C. difficile MetRS inhibitor REP3123 and the enzyme natural substrate, and the key active site residues and interactions were identified (Figure 4). These docking studies have validated the active site conformation in the constructed model and identified binding interactions. Dihydrofolate reductase (DHFR) has been used successfully as a drug target in the area of antibacterial, anticancer, and antimalarial therapies. It also acts as a drug target for leishmaniasis. Inhibition of DHFR leads to cell death through lack of thymine. Although the crystal structures of Leishmania major and Trypanosoma cruzi DHFRthymidylate synthase (TS) have been resolved, to date, there is no three-dimensional structural information on DHFR-TS of Leishmania donovani chagasi, which causes visceral leishmaniasis (37). Maganti et al. produced a homology model of the 3-D structure of L. donovani chagasi DHFR-TS using the homology module of Insight II (http://accelrys.com/) and investigated the structural requirements for its inhibition. Structural refinement and minimization of the generated L. donovani chagasi DHFRTS model employed the Discover 3 module of Insight II and molecular dynamic simulations. The model was further validated through use of the PROCHECK, Verify_3D, PROSA, PSQS, and ERRAT programs, which confirm that the model is reliable. Superimposition of the model structure with the templates L. major A chain, L. major B chain, and T. cruzi A chain showed root-mean-square deviations of 0.69, 0.71 and 1.11 A, respectively. Docking analysis of
Figure 4: Met-tRNA anticodon interactions with Clostridium difficile homology model. Corresponding key residues identified by analyzing homologous MetRS are highlighted in red (adapted from Al-Moubarak et al. 36).
Chem Biol Drug Des 2015; 85: 22–29
27
McPhillie et al.
the L. donovani chagasi DHFR-TS model with methotrexate enabled the author to identify specific residues in the L. donovani chagasi DHFR-TS binding pocket that play an important role in ligand or substrate binding as they have strong hydrogen bonding interactions with the ligand.
effects seen with known drugs. It is hoped in the future that further improvements in these techniques will mean they can be used to identify a new drug candidate.
References Pharmacophore Models The International Union of Pure and Applied Chemistry defines a pharmacophore to be ‘an ensemble of steric and electronic features that is necessary to ensure the optimal supramolecular interactions with a specific biological target and to trigger (or block) its biological response’. A pharmacophore model explains how structurally diverse ligands can bind to a common receptor site. Pharmacophore models can be used to identify novel ligands that will bind to the same receptor. Pharmacophore models were developed using CATALYST HypoGen and a training set of 29 diverse MetRS inhibitors to develop novel MetRS inhibitors (38). The best quantitative pharmacophore hypothesis (Hypo I) obtained a correlation coefficient of 0.975, root-mean-square deviation (RMSD) of 0.55, and cost difference (null cost-total cost) of 70.32. This Hypo I was validated by two methods, first using 104 test set molecules which resulted a correlation of 0.926 between HypoGen-estimated activities versus experimental activities and secondly by Cat-Scramble validation method. This validated pharmacophore model was further used for screening databases for discovery of new MetRS inhibitors. The new lead compounds were further analyzed for drug-like properties. A homology model of Staphylococcus aureus MetRS was built, and molecular docking studies were performed with many inhibitors using the newly built protein structure. It was found that the new leads exhibited good estimated inhibitory activity, calculated binding properties similar to experimentally proven compounds and also favorable drug-like properties.
Conclusions The rapid evolution of bacterial resistance to our current spectrum of antibacterial agents will always require a new pipeline of antibiotics. Ideally, new protein targets are also needed as the resistance mechanisms present for old targets are unlikely to be seen for these. However, identifying suitable new targets can be a very difficult task, and the use of in silico tools can greatly aid this. Ultimately, a protein’s suitability as a new antibacterial target can only be truly validated through biochemical assays and the identification of a new lead molecules with whole-cell antibacterial activity < 8 lg/mL, ideally over a broad spectrum of Gram-positive and Gram-negative bacteria. As yet this technique has not been used to successfully identify a drug which has reached the clinic, although it has been successfully used to identify proteins involved in off-target 28
1. Wenzel R.P., Edmond M.B. (2000) Managing antibiotic resistance. N Engl J Med;343:1961–1963. 2. Overbye K.M., Barrett J.F. (2005) Antibiotics: where did we go wrong. Drug Discov Today;10:45–52. 3. Silver L.L. (2011) Challenges of antibacterial discovery. Clin Microbiol Rev;24:71–109. 4. Butler M.S., Cooper M.A. (2011) Antibiotics in the clinical pipeline in 2011. J Antibiot;64:413–425. 5. Livermore D.M. (2011) Discovery research: the scientific challenge of finding new antibiotics. J Antimicrob Chemother;66:1941–1944. 6. Bush K. (2010) Alarming beta-lactamase-mediated resistance in multidrug-resistant enterobacteriaceae. Curr Opin Microbiol;13:558–564. 7. Payne D.J., Gwynn M.N., Holmes D.J., Pompliano D.L. (2007) Drugs for bad bugs: confronting the challenges of antibacterial discovery. Nat Rev Drug Discov;6: 29–40. 8. Falconer S.B., Brown E.D. (2009) New screens and targets in antibacterial drug discovery. Curr Opin Microbiol;12:497–504. 9. Hopkins A.L., Groom C.R. (2002) The druggable genome. Nat Rev Drug Discov;1:727–730. 10. Russ A.P., Lampel S. (2005) The druggable genome: an update. Drug Discov Today;10:1607–1610. 11. Gao Z., Li H., Zhang H., Liu X., Kang L., Luo X., Luo X., Zhu W., Chen K., Wang X., Jiang H. (2008) PDTD: a web-accessible protein database for drug target identification. BMC Bioinformatics;9:104. 12. Chung B.K.-S., Dick T., Lee D.-Y. (2013) In silico analyses for the discovery of tuberculosis drug targets. J Antimicrob Chemother;68:2701–2709. 13. Chan J.N.Y., Nislow C., Emili A. (2010) Recent advances and method development for drug target identification. Trends Pharmacol Sci;31:82–88. 14. Hajduk P.J., Huth J.R., Fesik S.W. (2005) Druggability indices for protein targets derived from NMR-based screening data. J Med Chem;48:2518–2525. 15. Chen Y.Z., Zhi D.G. (2001) Ligand-protein inverse docking and its potential use in the computer search of protein targets of a small molecule. Proteins; 43:217–226. 16. Schuster D. (2010) 3D pharmacophores as tools for activity profiling. Drug Discov Today Technol;7:e203– e270. 17. Bender A., Glen R.C. (2004) Molecular similarity: a key technique in molecular informatics. Org Biomol Chem;2:3204–3218. 18. Wang Y., Xiao J., Suzek T.O., Zhang J., Wang J., Zhou Z., Han L., Karapetyan K., Dracheva S., Shoemaker B.A., Bolton E., Gindulyte A., Bryant S.H. Chem Biol Drug Des 2015; 85: 22–29
In silico Antibacterial Target Identification
(2012) PubChem’s bioassay database. Nucleic Acids Res;40:D400–D412. 19. Gaulton A., Bellis L.J., Bento A.P., Chambers J., Davies M., Hersey A., Light Y., McGlinche S., Michalovich D., Al-Lazikani B., Overington J.P. (2012) ChEMBL: a large-scale bioactivity database for drug discovery. Nucleic Acids Res;40:D1100–D1107. 20. Jensen N.H., Roth B.L. (2008) Massively parallel screening of the receptorome. Comb Chem High Throughput Screen;11:420–426. 21. Koutsoukas A., Simms B., Kirchmair J., Bond P.J., Whitmore A.V., Zimmer S., Young M.P., Jenkins J.L., Glick M., Glen R.C., Bender A. (2011) From in silico target prediction to multi-target drug design: current databases, methods and applications. J Proteomics;74:2554–2574. 22. Li H., Gao Z., Kang L., Zhang H., Yang K., Yu K., Luo X., Zhu W., Chen K., Shen J., Wang X., Jiang H. (2006) TarFisDock: a web server for identifying drug targets with docking approach. Nucleic Acids Res;34: W219–W224. 23. Cai J.H., Han C., Hu T.C., Zhang J., Wu D.L., Wang F.D., Liu Y.Q., Ding J.P., Chen K.X., Yue J.M., Shen X., Jiang H.L. (2006) Peptide deformylase is a potential target for anti-Helicobacter pylori drugs: reverse docking, enzymatic assay, and X-ray crystallography validation. Protein Sci;15:2071–2081. 24. Chanumolu S.K., Rout C., Chauhan R.S. (2012) UniDrug-target: a computational tool to identify unique drug targets in pathogenic bacteria. PLoS ONE;7: e32833. 25. Ji Z.L., Wang Y., Yu L., Han L.Y., Zheng C.J., Chen Y.Z. (2006) In silico search of putative adverse drug reaction related proteins as a potential tool for facilitating drug adverse effect prediction. Toxicol Lett;164:104–112. 26. Chen X., Zhou H., Liu Y.B., Wang J.F., Li H., Ung C.Y., Han L.Y., Cao Z.W., Chen Y.Z. (2006) Database of traditional Chinese medicine and its application to studies of mechanism and to prescription validation. Br J Pharmacol;149:1092–1103. 27. Chen X., Ung C.Y., Chen Y.Z. (2003) Can an in silico drug-target search method be used to probe potential mechanisms of medicinal plant ingredients? Nat Prod Rep;20:432–444. 28. O’Shea R., Moser H.E. (2008) Physicochemical properties of antibacterial compounds: implications for drug discovery. J Med Chem;51:2871–2878.
Chem Biol Drug Des 2015; 85: 22–29
29. Cronin M.T.D., Aptula A.O., Dearden J.C., Duffy J.C., Netzeva T.I., Patel H., Rowe P.H., Schultz T.W., Wortht A.P., Voutzoulidis K., Schuurmann G. (2002) Structure-based classification of antibacterial activity. J Chem Inf Comput Sci;42:869–878. 30. Li Q., Lai L. (2007) Prediction of potential drug targets based on simple sequence properties. BMC Bioinformatics;8:353. 31. Han L.Y., Zheng C.J., Xie B., Jia J., Ma X.H., Zhu F., Lin H.H., Chen X., Chen Y.Z. (2007) Support vector machines approach for predicting druggable proteins: recent progress in its exploration and investigation of its usefulness. Drug Discov Today;12:304–313. 32. Arnold K., Bordoli L., Kopp J., Schwede T. (2006) The SWISS-MODEL workspace: a web-based environment for protein structure homology modelling. Bioinformatics;22:195–201. 33. Kelley L.A., Sternberg M.J.E. (2009) Protein structure prediction on the Web: a case study using the Phyre server. Nat Protoc;4:363–371. 34. Vilar S., Cozza G., Moro S. (2008) Medicinal chemistry and the molecular operating environment (MOE): application of QSAR and molecular docking to drug discovery. Curr Top Med Chem;8:1555–1572. 35. Kishan K.V.R. (2007) Structural biology, protein conformations and drug designing. Curr Protein Pept Sci;8:376–380. 36. Al-Moubarak E., Simons C. (2011) A homology model for Clostridium difficile methionyl tRNA synthetase: active site analysis and docking interactions. J Mol Model;17:1679–1693. 37. Maganti L., Manoharan P., Ghoshal N. (2010) Probing the structure of Leishmania donovani chagasi DHFR-TS: comparative protein modeling and proteinligand interaction studies. J Mol Model;16: 1539–1547. 38. Bharatham N., BlIaratham K., Lee K.W. (2007) Pharmacophore identification and virtual screening for methionyl-tRNA synthetase inhibitors. J Mol Graph Model;25:813–823.
Note a
http://dspace.mit.edu/bitstream/handle/1721.1/3777/MEBCS004.pdf?sequence=2.NotesNoteahttp://dspace.mit. edu/bitstream/handle/1721.1/3777/MEBCS004.pdf? sequence=2.
29
Computational Methods to Identify New Antibacterial Targets Martin J. McPhillie†,‡, Ricky M. Cain, Sarah Narramore, Colin W. G. Fishwick and Katie J. Simmons*,‡
the late 1930s with the production of penicillin and ending with the discovery of daptomycin in 1969, this golden age produced every class of antibiotic prescribed today (3,4).
School of Chemistry, University of Leeds, Leeds, LS2 9JT, UK *Corresponding author: Katie J. Simmons, [email protected] † Present address: Department of Molecular Biology & Biotechnology, University of Sheffield, Sheffield S10 2TN, UK ‡ These authors contributed equally to this work.
During this time, the actinomycetes provided a great source of antibacterial agents, discovered through the empirical screening of fermentation broths and extracts of microorganisms containing secondary metabolites which inhibited bacterial growth (3). Nearly all antibiotic classes used today were discovered from the actinomycetes during this golden age. Once dereplication became commonplace, emphasis shifted toward new approaches (3,5). During the 1970s, new antibiotics were discovered by synthesizing analogues of existing antibacterial drug classes, with improved wholecell activity and pharmacokinetic properties. These analogues were also able to evade some existing resistance mechanisms, for example, the introduction of the fluoroquinolone antibiotics superseded the first-in-class quinolone, nalidixic acid (5). However, as resistance mechanisms arose which compromised entire classes, such as the metallo blactamases which inactivate the b-lactam antibiotics (6), a new paradigm of drug discovery, the genomic-based approach, was introduced in the 1990s (3).
The development of resistance to all current antibiotics in the clinic means there is an urgent unmet need for novel antibacterial agents with new modes of action. One of the best ways of finding these is to identify new essential bacterial enzymes to target. The advent of a number of in silico tools has aided classical methods of discovering new antibacterial targets, and these programs are the subject of this review. Many of these tools apply a cheminformatic approach, utilizing the structural information of either ligand or protein, chemogenomic databases, and docking algorithms to identify putative antibacterial targets. Considering the wealth of potential drug targets identified from genomic research, these approaches are perfectly placed to mine this rich resource and complement drug discovery programs. Key words: cheminformatics, proteomics, therapeutic target
molecular
modeling,
Received 30 April 2014, revised 17 June 2014 and accepted for publication 18 June 2014
The ‘golden age’ of antibiotics At the turn of the 20th century, bacterial infections were among the highest causes of morbidity and mortality worldwide. Over the next 70 years, antibacterial agents would reduce the burden of infectious disease and increase life expectancy to such an extent that the US Surgeon General William H. Stewart declared in 1969: ‘. . .that we had essentially defeated infectious diseases and could close the book on them [infectious diseases]. . . ‘(1, 2)
Many of the pharmaceutical companies concurred and concentrated their efforts on other therapeutic areas, ending a period of research which would be later known as the ‘golden age’ of antibacterial drug discovery. Beginning in 22
The genomic era Applying this genomic approach to antibacterial drug discovery required sequencing the genomes of pathogens to identify conserved genes essential for their survival. The identified genes which encoded targets that generally lacked a mammalian counterpart were then used in high-throughput screens (HTS) involving chemical and combinatorial libraries to identify inhibitors of these target enzymes. Unfortunately, this target-based genomic approach has failed to deliver the number of antibacterial hit molecules that the approach promised, especially when compared with the successes seen in other therapeutic areas (7). This has translated to fewer antibacterial lead compounds and is yet to deliver a clinical candidate. Structural genomic approaches that build upon target-based genomics, such as structure-based drug discovery and in silico target identification, are hoping to revitalize the future of antibiotic discovery (8).
The ideal bacterial target A bacterial enzyme must fulfill several criteria to be classified as a suitable antibacterial drug target: (i) It is essential ª 2014 John Wiley & Sons A/S. doi: 10.1111/cbdd.12385
In silico Antibacterial Target Identification
for cell survival (i.e. enzyme inhibition results in inhibition of cell growth, death or lysis), (ii) there is no mammalian homologue with the same or similar function, (iii) it is highly conserved among bacteria species to give a useful antibacterial spectrum, and (iv) it is ‘druggable’, in that a small molecule could bind to a suitable site and exert a biological effect (3). Compounds that inhibit such an enzyme would be expected to be efficacious and selective for its target protein, as well as displaying Gram-positive and/or Gram-negative antibacterial activity. Other criteria which are non-essential but desirable are as follows: (v) a soluble and stable enzyme giving high-quality and well-diffracting crystals for X-ray crystallography and (vi) an innately low frequency of mutation so that a single point mutation does not render an inhibitor scaffold ineffective when used in monotherapy. Often choosing a suitable target is the ratelimiting step of an antibacterial drug discovery program (3). Until 2000, only ~500 drug targets had been reported, among which only 120 targets had associated drugs used in the clinic (9). The completion of the human genome and numerous pathogen genomes suggests that there are up to 40 000 genes and at least the same number of proteins, and many of which are potential targets for drug discovery (10). This is a reservoir for drug discovery and target identification, although rational approaches to mine this resource are needed (11). For example, during the genomics era, many new ‘druggable’ targets, such as FabI and FabH (7), were identified but as we have failed to convert these targets into new lead compounds, there has been a recent shift back to classical phenotypic whole-cell screens and the re-evaluation of old abandoned lead compounds. However, the whole-cell ‘black-box’ approaches exclude the use of a number of modern medicinal chemistry techniques such the rational identification of novel antibacterial targets (12). Antibacterial drug discovery programmes can either find new ways to inhibit old ‘classical’ targets or identify new ‘druggable’ targets. This review will cover recent developments in antibacterial target identification using computational methods. In particular, the review will focus on cheminformatic approaches to general target identification, including homology modeling, and pharmacophore and QSAR techniques, with applied examples from antibacterial research. In addition to primary target identification, many of these techniques are able to anticipate any offtarget effects, establishing a bioactivity spectrum of a particular molecule. Various drug target identification techniques have been developed by analyzing disease relevance, functional roles, expression profiles, and lossof-function genetics between normal and disease states (13). Most of these methods are based on the detection of sequences and functional similarity to know antibacterial targets. Parameters describing polar and apolar surface areas, surface complexity, and pocket dimensions have also been used (14). Chem Biol Drug Des 2015; 85: 22–29
Cheminformatic Approaches Toward Target Identification Many of the in silico approaches to target prediction can be classified as using either cheminformatic (structural knowledge of ligands or proteins) or bioinformatic (synergy maps/biological network models) (12) tools. A number of cheminformatic target prediction approaches exist, for example, those based on docking (15), pharmacophore (16), or ligand- and structure-based target prediction methods. In silico methods can exploit prior knowledge of ligand–target interactions, collected from the literature which is further organized in chemogenomic databases. These data are then analyzed to make knowledge-based predictions for new, untested molecules or to suggest new drug–target interactions for already marketed compounds. Ligand-based methods use the concept of shape similarity, where chemical compounds of similar shape tend to have similar biological activities (17). As such query compounds, without a known target, can be screened for shape similarity against known enzyme inhibitors present in biological databases [e.g. PubChem (18), ChEMBl (19), and KiDB (20)] to search for a putative protein targets. This method is generally fast but relies on prior knowledge of the ligands and protein structures, and is not applicable in situations where no ligand exists. Structure-based methods use the structural knowledge of proteins to predict ligand–protein interactions, where a query compound is docked against a panel of protein targets or receptors using scoring functions (termed ‘reverse’ or ‘inverse’ docking). These techniques have not progressed as rapidly as traditional molecular docking approaches and suffer from significantly increased computational time and unreliable results (21). A number of these cheminformatic tools are discussed in further detail below, in an antibacterial context. Reverse docking using potential drug target database The web-accessible potential drug target database (PDTD) is a dual-function database that associates an informatics database to a structural database of recognized and potential drug targets with known 3-D-structural information. PDTD contains 1207 entries covering 841 known and potential drug targets, with structures from the Protein Data Bank (PDB) (11). Drug targets of PDTD were categorized into types according to two conditions: therapeutic areas and biochemical criteria. The databases support an extensive searching function using PDB ID, target name and category, and related disease (Figure 1). As the PDTD is designed to search the probable binding proteins for new active compounds or existing drugs using reverse docking, it only contains proteins with known 3-D 23
McPhillie et al.
Figure 1: Overview of PDTD application (adapted from Gao et al. 11).
structures determined experimentally by the X-ray crystallographic or NMR methods. The co-ordinates of proteins were isolated from the PDB. As not all PDB structures are of equal quality, a protein structure is selected according to the following criteria when it has several redundant records in PDB: (i) select the structure without mutation and missing residues around the active site; (ii) select the structure with high resolution; and (iii) select the structure complexed with a ligand. Combined with the integrated reverse docking server TarFisDock (Target Fishing Docking) (22), PDTD has been widely used to identify binding proteins for small molecules of therapeutic relevance. The binding proteins for several molecules have been verified through bioassays and crystal structure determination for target–ligand complexes. Specifically, the PDTD was searched by TarFisDock using the antibacterial agent N-trans-caffeoyltyramine (1 Figure 2), a natural product discovered by an antibacterial screen against Helicobacter pylori. A homology search revealed that, among the fifteen protein candidates discovered by reverse docking, diaminopimelate decarboxylase and peptide deformylase (PDF) are putative binding proteins for compound 1. Enzymatic assay demonstrated compound 1 and its derivative compound 2 are the potent inhibitors against the H. pylori PDF (HpPDF) with IC50 values of 10.8 and 1.25 lM, respectively. X-ray crystal structures of HpPDF and the complexes of HpPDF with 1 and 2 were determined, indicating that these two inhibitors bind well with the HpPDF binding pocket (23). To exemplify the application
of PDTD combining with TarFisDock, this example has been uploaded to the PDTD homepage under the ‘Benchmark’ option.
INVDOCK—automated detection of protein and nucleic acid targets by inverse docking INVDOCK (24) has been developed by the Bioinformatics and Drug Design group (BIDD) at the National University of Singapore for the computer-automated identification of potential protein and nucleic acid targets of drugs, newly designed drug candidate, natural product, or other chemical compound. In addition to suggesting likely therapeutic targets of a query molecule, INVDOCK can also predict potential toxicity due to off-target effects (25). INVDOCK is designed to automatically search every entry of a protein cavity database containing 9000 protein and nucleic acid entries. All parts of the cavities are subject to docking analysis. To save CPU time, the program proceeds to the next protein cavity once a successful dock is obtained without doing exhaustive dockings within that particular cavity. Although the optimized docking pose is not sought, the prioritized search complex has been shown to find a pose close to the observed binding mode seen for the studied protein–ligand complexes. The ligands are flexibly docked by matching the ligand atoms to the center of spheres followed by a conformational optimization. Scoring of docked molecules is based on a ligand–protein interaction energy function DELPa (Figure 3).
Figure 2: Structure of Helicobacter pylori inhibitors identified using TarFisDock.
24
Chem Biol Drug Des 2015; 85: 22–29
In silico Antibacterial Target Identification
Figure 3: Overview of INVDOCK protocol.
A number of studies (15) have been carried out to validate INVDOCK. Using a group of nine known drugs, INVDOCK was able to successfully predict their known therapeutic targets, and in a separate test on a group of eight drugs, INVDOCK successfully predicted 38 known toxic sideeffects from off-target interactions, missing only five. INVDOCK has also used extensively to identify the biological target(s) of traditional Chinese medicines such as genistein (26) and natural products (27). As many of our current antibiotics are derived from natural products, INVDOCK could facilitate the rational identification of protein targets of new secondary metabolites discovered via marine sources.
UniDrug target–target identification in pathogenic bacteria The web server UniDrug-Target (24) has been specifically designed to combine bacterial biological information and computational methods to stringently identify bacterial pathogen-specific proteins as drug targets. The server compares the proteome sequences of pathogenic and non-pathogenic bacteria, and mammalian proteomes, to identify unique proteins present in pathogenic bacteria. Each potential drug target is ranked on its functional importance within the cell, for example, determining whether the protein is essential for the cell’s survival. Besides predicting pathogen-specific proteins, three new algorithms have been developed and implemented using protein sequences, domains, structures, and metabolic reactions for construction of partial metabolic networks (PMNs), determination of conservation in critical residues, and variation analysis of residues forming similar cavities in proteins sequences. Chem Biol Drug Des 2015; 85: 22–29
The UniDrug-Target server is capable of predicting drug targets for any sequenced pathogenic bacteria that have FASTA sequences and annotated information. The utility of this server was demonstrated for Mycobacterium tuberculosis (H37Rv). The UniDrug-Target server identified 265 mycobacteria pathogen-specific proteins, including 17 essential proteins which are potential drug targets. As novel pathogen-specific drug targets may have no intrinsic resistance mechanisms, future output from UniDrugTarget could be valuable in combating bacterial resistance. The server is freely available at http://117.211.115.67/ UDT/main.html. The stand-alone application (source codes) is available at http://www.bioinformatics.org/ftp/ pub/bioinfojuit/UDT.rar (Table 1).
Target identification in phenotypic bacterial growth inhibitions screens A major bottleneck in phenotypic bacterial screening is determining the mode of action of any ‘hit’ compounds, whether it is due to target-specific activity or whole-cell toxicity. In silico target prediction was used to elucidate the mode of action of a series of hit’ compounds using data mining methods (21). A total of 7812 compounds within the MDL DrugDatabaseReport (MDDR) database that were identified as ‘Antibiotic’, ‘Antibacterial’, or ‘Antifungal’ via a phenotypic screen were collated and analyzed computationally using ECFP circular fingerprints (representation of molecular structures by atom neighborhoods) and a Bayesian model (statistical model representing variables and their dependencies via a directed acyclic graph). For each compound, all targets with a Bayes score 25
McPhillie et al. Table 1: List of all applications discussed in this review detailing their source and availability Application
Source
Availability
ChEMBl Insight II
https://www.ebi.ac.uk/chembl/ http://accelrys.com/
INVDOCK
http://bidd.nus.edu.sg/group/softwares/invdock.htm
KiDB MOE
http://pdsp.med.unc.edu/kidb.php http://www.chemcomp.com/MOE-Molecular_Operating_Environment.htm
PDTD Database
http://www.dddc.ac.cn/pdtd/index.php
Phyre2
http://www.sbg.bio.ic.ac.uk/phyre2/html/page.cgi?id=index
Protein Data Bank PubChem SWISS-MODEL TarFisDock
http://www.rcsb.org/pdb/home/home.do
Freely available A number of packages available-cost varies Freely available upon registration for academic groups Freely available A number of packages available-cost varies Freely available upon registration for academic groups Freely available for academic groups only Freely available
UniDrug-Target
http://117.211.115.67/UDT/main.html (server) http://www.bioinformatics.org/ftp/pub/bioinfojuit/UDT.rar. (standalone application)
https://pubchem.ncbi.nlm.nih.gov/ http://swissmodel.expasy.org/ http://www.dddc.ac.cn/tarfisdock/
of > 30 were considered likely targets and further investigated. The study concluded that likely targets could be assigned for 6874 compounds, the most prevalent being DNA gyrase and beta-lactamases. The identification of a number of potential human targets allows the users to predict off-target effects and mitigate their effects early in the discovery process.
Structure-based classification of antibacterial activity It is known that antibacterial agents occupy a unique area of chemical space compared with that for other therapeutic agents (28). Quantitative structure–activity relationship (QSAR) studies have been undertaken to predict antibacterial activity within large libraries of compounds and therefore enhance in silico screening in antibacterial drug discovery programs. One study (29) established 167 2-D physicochemical and structural descriptors of 667 known drug molecules and used two statistical techniques (linear discriminant analysis and binary logistic regression) to separate them into two groups: one with antibacterial activity and one without. The authors found that six descriptors, accounting for hydrophobicity and inter- and intramolecular hydrogen bonding, were optimal in separating the two groups. The study concluded that the most significant parameter was the octanol/water partition coefficient (ClogP), suggesting that antibacterial compounds have a lower ClogP compared with other drugs. Both statistical models were highly successful at predicting antibacterial activity, therefore providing the means of screening 26
Freely available Freely available Freely available upon registration for academic groups Freely available
databases to identify antibacterial compounds on the basis easily calculated properties (29).
Support vector machine-based methods Support Vector Machines are supervised learning models with associated learning algorithms that analyze data and recognize patterns. Support Vector Machine (SVM)-based methods have been developed from amino acid sequence-derived properties of known antibacterial targets, employing a set of features of physicochemical properties directly calculated from primary protein sequence (30,31). In one example from Rashid et al., the models were trained and tested on 852 mycobacterial proteins and evaluated using a fivefold cross-validation technique. First, the support vector machine model was developed using the amino acid composition and an overall accuracy of 83% was achieved, with average accuracy of 68%. To utilize evolutionary information, a SVM model was developed using position-specific scoring matrix profiles obtained from PSI-BLAST. The overall accuracy achieved was of 87% with average accuracy of 74%. In addition, a hidden Markov model, MEME/MAST (Multiple Em for Motif Elicitation/Motif Alignment and Search Tool) and a hybrid model that combined two or more models was also developed. The authors achieved maximum overall accuracy of 87% with average accuracy of 89% using combination of PSSM based SVM model and MEME/MAST. These methods can successfully distinguish known drug targets from putative non-drug targets at an accuracy of 84% in 10-fold cross-validation test. Chem Biol Drug Des 2015; 85: 22–29
In silico Antibacterial Target Identification
Homology Modeling For cases in which the crystal structure of the particular bacterial protein is not available, construction of a homology model is often possible, provided that a crystal structure is available for a protein with substantial sequence similarity to the protein of interest. Advanced homology models use experimentally determined structures to predict the conformation of another protein that has a similar amino acid sequence. Homology modeling involves four steps: fold assignment, sequence alignment, model building, and model refinement. Several computer packages are available to perform this process automatically, for example, SWISS-MODEL (32), Phyre2 (33), and MOE (34). SWISS-MODEL and Phyre2 are freely available to academic groups. An excellent reference source for a number of different homology modeling servers can be found at: http://proteopedia.org/wiki/index.php/Homology_modeling_servers. Homology modeling is useful when a protein sequence identity is greater than ~30% is available to model with, although >50% is preferable, especially around the ligand binding site of interest. The major difficulties in producing homology models lie in the accurate prediction of flexible loops, although a large amount of research is currently being carried out to improve the accuracy of these models (35). A number of groups have generated homology models to facilitate screening of ligands for new antibacterial targets (37). The methionyl tRNA synthetase (MetRS) is strictly required for protein biosynthesis, and success was reported in developing antibacterial agents to inhibit this enzyme. A homology model of Clostridium difficile MetRS was constructed using Molecular Operating Environment (MOE) software. Aquifex aeolicus MetRS was the main template
while the query zinc binding domain was modeled using Thermus thermophilus MetRS. The model has been assessed and compared with its main template (Ramachandran, ERRAT and ProSA). The active site of the query protein has been predicted from its sequence using a detailed conservation analysis (ClustalW2). Using MOE software, suitable ligands were docked in the constructed model, including a C. difficile MetRS inhibitor REP3123 and the enzyme natural substrate, and the key active site residues and interactions were identified (Figure 4). These docking studies have validated the active site conformation in the constructed model and identified binding interactions. Dihydrofolate reductase (DHFR) has been used successfully as a drug target in the area of antibacterial, anticancer, and antimalarial therapies. It also acts as a drug target for leishmaniasis. Inhibition of DHFR leads to cell death through lack of thymine. Although the crystal structures of Leishmania major and Trypanosoma cruzi DHFRthymidylate synthase (TS) have been resolved, to date, there is no three-dimensional structural information on DHFR-TS of Leishmania donovani chagasi, which causes visceral leishmaniasis (37). Maganti et al. produced a homology model of the 3-D structure of L. donovani chagasi DHFR-TS using the homology module of Insight II (http://accelrys.com/) and investigated the structural requirements for its inhibition. Structural refinement and minimization of the generated L. donovani chagasi DHFRTS model employed the Discover 3 module of Insight II and molecular dynamic simulations. The model was further validated through use of the PROCHECK, Verify_3D, PROSA, PSQS, and ERRAT programs, which confirm that the model is reliable. Superimposition of the model structure with the templates L. major A chain, L. major B chain, and T. cruzi A chain showed root-mean-square deviations of 0.69, 0.71 and 1.11 A, respectively. Docking analysis of
Figure 4: Met-tRNA anticodon interactions with Clostridium difficile homology model. Corresponding key residues identified by analyzing homologous MetRS are highlighted in red (adapted from Al-Moubarak et al. 36).
Chem Biol Drug Des 2015; 85: 22–29
27
McPhillie et al.
the L. donovani chagasi DHFR-TS model with methotrexate enabled the author to identify specific residues in the L. donovani chagasi DHFR-TS binding pocket that play an important role in ligand or substrate binding as they have strong hydrogen bonding interactions with the ligand.
effects seen with known drugs. It is hoped in the future that further improvements in these techniques will mean they can be used to identify a new drug candidate.
References Pharmacophore Models The International Union of Pure and Applied Chemistry defines a pharmacophore to be ‘an ensemble of steric and electronic features that is necessary to ensure the optimal supramolecular interactions with a specific biological target and to trigger (or block) its biological response’. A pharmacophore model explains how structurally diverse ligands can bind to a common receptor site. Pharmacophore models can be used to identify novel ligands that will bind to the same receptor. Pharmacophore models were developed using CATALYST HypoGen and a training set of 29 diverse MetRS inhibitors to develop novel MetRS inhibitors (38). The best quantitative pharmacophore hypothesis (Hypo I) obtained a correlation coefficient of 0.975, root-mean-square deviation (RMSD) of 0.55, and cost difference (null cost-total cost) of 70.32. This Hypo I was validated by two methods, first using 104 test set molecules which resulted a correlation of 0.926 between HypoGen-estimated activities versus experimental activities and secondly by Cat-Scramble validation method. This validated pharmacophore model was further used for screening databases for discovery of new MetRS inhibitors. The new lead compounds were further analyzed for drug-like properties. A homology model of Staphylococcus aureus MetRS was built, and molecular docking studies were performed with many inhibitors using the newly built protein structure. It was found that the new leads exhibited good estimated inhibitory activity, calculated binding properties similar to experimentally proven compounds and also favorable drug-like properties.
Conclusions The rapid evolution of bacterial resistance to our current spectrum of antibacterial agents will always require a new pipeline of antibiotics. Ideally, new protein targets are also needed as the resistance mechanisms present for old targets are unlikely to be seen for these. However, identifying suitable new targets can be a very difficult task, and the use of in silico tools can greatly aid this. Ultimately, a protein’s suitability as a new antibacterial target can only be truly validated through biochemical assays and the identification of a new lead molecules with whole-cell antibacterial activity < 8 lg/mL, ideally over a broad spectrum of Gram-positive and Gram-negative bacteria. As yet this technique has not been used to successfully identify a drug which has reached the clinic, although it has been successfully used to identify proteins involved in off-target 28
1. Wenzel R.P., Edmond M.B. (2000) Managing antibiotic resistance. N Engl J Med;343:1961–1963. 2. Overbye K.M., Barrett J.F. (2005) Antibiotics: where did we go wrong. Drug Discov Today;10:45–52. 3. Silver L.L. (2011) Challenges of antibacterial discovery. Clin Microbiol Rev;24:71–109. 4. Butler M.S., Cooper M.A. (2011) Antibiotics in the clinical pipeline in 2011. J Antibiot;64:413–425. 5. Livermore D.M. (2011) Discovery research: the scientific challenge of finding new antibiotics. J Antimicrob Chemother;66:1941–1944. 6. Bush K. (2010) Alarming beta-lactamase-mediated resistance in multidrug-resistant enterobacteriaceae. Curr Opin Microbiol;13:558–564. 7. Payne D.J., Gwynn M.N., Holmes D.J., Pompliano D.L. (2007) Drugs for bad bugs: confronting the challenges of antibacterial discovery. Nat Rev Drug Discov;6: 29–40. 8. Falconer S.B., Brown E.D. (2009) New screens and targets in antibacterial drug discovery. Curr Opin Microbiol;12:497–504. 9. Hopkins A.L., Groom C.R. (2002) The druggable genome. Nat Rev Drug Discov;1:727–730. 10. Russ A.P., Lampel S. (2005) The druggable genome: an update. Drug Discov Today;10:1607–1610. 11. Gao Z., Li H., Zhang H., Liu X., Kang L., Luo X., Luo X., Zhu W., Chen K., Wang X., Jiang H. (2008) PDTD: a web-accessible protein database for drug target identification. BMC Bioinformatics;9:104. 12. Chung B.K.-S., Dick T., Lee D.-Y. (2013) In silico analyses for the discovery of tuberculosis drug targets. J Antimicrob Chemother;68:2701–2709. 13. Chan J.N.Y., Nislow C., Emili A. (2010) Recent advances and method development for drug target identification. Trends Pharmacol Sci;31:82–88. 14. Hajduk P.J., Huth J.R., Fesik S.W. (2005) Druggability indices for protein targets derived from NMR-based screening data. J Med Chem;48:2518–2525. 15. Chen Y.Z., Zhi D.G. (2001) Ligand-protein inverse docking and its potential use in the computer search of protein targets of a small molecule. Proteins; 43:217–226. 16. Schuster D. (2010) 3D pharmacophores as tools for activity profiling. Drug Discov Today Technol;7:e203– e270. 17. Bender A., Glen R.C. (2004) Molecular similarity: a key technique in molecular informatics. Org Biomol Chem;2:3204–3218. 18. Wang Y., Xiao J., Suzek T.O., Zhang J., Wang J., Zhou Z., Han L., Karapetyan K., Dracheva S., Shoemaker B.A., Bolton E., Gindulyte A., Bryant S.H. Chem Biol Drug Des 2015; 85: 22–29
In silico Antibacterial Target Identification
(2012) PubChem’s bioassay database. Nucleic Acids Res;40:D400–D412. 19. Gaulton A., Bellis L.J., Bento A.P., Chambers J., Davies M., Hersey A., Light Y., McGlinche S., Michalovich D., Al-Lazikani B., Overington J.P. (2012) ChEMBL: a large-scale bioactivity database for drug discovery. Nucleic Acids Res;40:D1100–D1107. 20. Jensen N.H., Roth B.L. (2008) Massively parallel screening of the receptorome. Comb Chem High Throughput Screen;11:420–426. 21. Koutsoukas A., Simms B., Kirchmair J., Bond P.J., Whitmore A.V., Zimmer S., Young M.P., Jenkins J.L., Glick M., Glen R.C., Bender A. (2011) From in silico target prediction to multi-target drug design: current databases, methods and applications. J Proteomics;74:2554–2574. 22. Li H., Gao Z., Kang L., Zhang H., Yang K., Yu K., Luo X., Zhu W., Chen K., Shen J., Wang X., Jiang H. (2006) TarFisDock: a web server for identifying drug targets with docking approach. Nucleic Acids Res;34: W219–W224. 23. Cai J.H., Han C., Hu T.C., Zhang J., Wu D.L., Wang F.D., Liu Y.Q., Ding J.P., Chen K.X., Yue J.M., Shen X., Jiang H.L. (2006) Peptide deformylase is a potential target for anti-Helicobacter pylori drugs: reverse docking, enzymatic assay, and X-ray crystallography validation. Protein Sci;15:2071–2081. 24. Chanumolu S.K., Rout C., Chauhan R.S. (2012) UniDrug-target: a computational tool to identify unique drug targets in pathogenic bacteria. PLoS ONE;7: e32833. 25. Ji Z.L., Wang Y., Yu L., Han L.Y., Zheng C.J., Chen Y.Z. (2006) In silico search of putative adverse drug reaction related proteins as a potential tool for facilitating drug adverse effect prediction. Toxicol Lett;164:104–112. 26. Chen X., Zhou H., Liu Y.B., Wang J.F., Li H., Ung C.Y., Han L.Y., Cao Z.W., Chen Y.Z. (2006) Database of traditional Chinese medicine and its application to studies of mechanism and to prescription validation. Br J Pharmacol;149:1092–1103. 27. Chen X., Ung C.Y., Chen Y.Z. (2003) Can an in silico drug-target search method be used to probe potential mechanisms of medicinal plant ingredients? Nat Prod Rep;20:432–444. 28. O’Shea R., Moser H.E. (2008) Physicochemical properties of antibacterial compounds: implications for drug discovery. J Med Chem;51:2871–2878.
Chem Biol Drug Des 2015; 85: 22–29
29. Cronin M.T.D., Aptula A.O., Dearden J.C., Duffy J.C., Netzeva T.I., Patel H., Rowe P.H., Schultz T.W., Wortht A.P., Voutzoulidis K., Schuurmann G. (2002) Structure-based classification of antibacterial activity. J Chem Inf Comput Sci;42:869–878. 30. Li Q., Lai L. (2007) Prediction of potential drug targets based on simple sequence properties. BMC Bioinformatics;8:353. 31. Han L.Y., Zheng C.J., Xie B., Jia J., Ma X.H., Zhu F., Lin H.H., Chen X., Chen Y.Z. (2007) Support vector machines approach for predicting druggable proteins: recent progress in its exploration and investigation of its usefulness. Drug Discov Today;12:304–313. 32. Arnold K., Bordoli L., Kopp J., Schwede T. (2006) The SWISS-MODEL workspace: a web-based environment for protein structure homology modelling. Bioinformatics;22:195–201. 33. Kelley L.A., Sternberg M.J.E. (2009) Protein structure prediction on the Web: a case study using the Phyre server. Nat Protoc;4:363–371. 34. Vilar S., Cozza G., Moro S. (2008) Medicinal chemistry and the molecular operating environment (MOE): application of QSAR and molecular docking to drug discovery. Curr Top Med Chem;8:1555–1572. 35. Kishan K.V.R. (2007) Structural biology, protein conformations and drug designing. Curr Protein Pept Sci;8:376–380. 36. Al-Moubarak E., Simons C. (2011) A homology model for Clostridium difficile methionyl tRNA synthetase: active site analysis and docking interactions. J Mol Model;17:1679–1693. 37. Maganti L., Manoharan P., Ghoshal N. (2010) Probing the structure of Leishmania donovani chagasi DHFR-TS: comparative protein modeling and proteinligand interaction studies. J Mol Model;16: 1539–1547. 38. Bharatham N., BlIaratham K., Lee K.W. (2007) Pharmacophore identification and virtual screening for methionyl-tRNA synthetase inhibitors. J Mol Graph Model;25:813–823.
Note a
http://dspace.mit.edu/bitstream/handle/1721.1/3777/MEBCS004.pdf?sequence=2.NotesNoteahttp://dspace.mit. edu/bitstream/handle/1721.1/3777/MEBCS004.pdf? sequence=2.
29
