Molecular modeling of Plasmodium falciparum peptide deformylase and structure-based pharmacophore screening for inhibitors

Abstract

Plasmodium falciparum peptide deformylase (PfPDF), a metalloenzyme which catalytically removes the N-formyl group from N-terminal methionine from polypeptide during protein maturation, is a potential drug target for antimalarial drug design but is less explored in comparison to its bacterial counterpart due to difficulties in enzyme purification, the labile nature of the metal cofactors, and the absence of crystal data of the enzyme–inhibitor complex and inhibitors. We used molecular modeling techniques to study the effect of different metal ions, Co²⁺, Zn²⁺, Ni²⁺ and Fe²⁺ towards actinonin binding and recognized PfPDF–Co²⁺–actinonin inhibitor complex as the energetically favorable structure supported by biochemical characterizations. Further, we analyzed the favorable coordination geometry of bound metal cofactor and showed that its geometry did not affect the binding mode of actinonin consistent with the crystallographic observations which suggested that FlexX docking-based virtual screening may be effectively applied to recognize PfPDF binders. Structure-based pharmacophore screening of PfPDF–Co²⁺–actinonin complex recognized five potential hits and analyzed their effectiveness in chelating metal cofactor and ability to develop similar poses in both pharmacophore fit and docking. From the geometrical properties and characteristics of antibacterial PDF inhibitors, CAP01891052 (a triazine and quinoline containing molecule) and IBS297042 (an indole and piperidine containing molecule) were prioritized as promising lead molecules to modulate PfPDF activity.

---

Introduction

The existing global resistance crisis to available antimalarials has prompted research into novel and potential drug targets and associated drug discovery to combat the prevalence of drug-resistant strains and improve treatment.¹ Classified as a severe tropical parasitic and re-emerging disease, malaria causes 1–3 million deaths (mostly children under the age of five) per annum and its severity is attributed to the deadly and drug-resistant forms of the malaria parasite, P. falciparum.² Peptide deformylase (PDF), one of the targets related to the protein maturation stage, is a metalloprotease which catalyzes the removal of the N-formyl group from N-terminal methionine of nascent polypeptide processed by protein translation machinery and this process is essential for prokaryotic cell viability.^3,4 Initially thought to be unique to bacteria, genome sequencing has unveiled the presence of PDF homologues in eukaryotes including parasites, plants and mammals.⁵ The identification of bacterial, malarial and human PDFs has facilitated a remarkable prospect for drug discovery in the antibacterial, antimalarial and anticancer areas.⁶ In P. falciparum, the process of formylation and deformylation takes place in the apicoplast^7,8 while the physiological role of the human counterpart is not yet clear but in vitro results confirmed its functionality.⁹ Almost a decade ago, human mitochondrial PDF (HsPDF) was cloned and biochemically characterized using a Co²⁺ metal cofactor to reconstitute enzymatic activity.¹⁰ Studies also advocated that PDF inhibitors may not be reaching human mitochondria and the reports of no toxicity to animal models¹¹ suggested that an efficient inhibitor design strategy can be employed to identify molecules with high specificity.

The recruitment of a metal cofactor such as Zn²⁺, Co²⁺, Ni²⁺ and Fe²⁺ in the PDF enzyme to perform deformylation can be varied among organisms.¹² Although, Fe²⁺ is known as the physiological and catalytic cofactor for most bacterial¹³ and P. falciparum PDFs and its rapid oxidation leads to PDF inactivation, thermodynamically stable PDF complexes are obtained by replacing with Ni²⁺ or Co²⁺ ions with little change in catalytic efficiency.¹⁴ This provision also helped in crystallizing numerous PDF–metal ion–inhibitor complexes and stimulated antibacterial research with implications for preclinical candidates.¹⁵ Actinonin, a hydroxamic acid pseudopeptide, was the first identified natural inhibitor of PDF isolated from a Streptomyces species¹⁶ which attained widespread attention to study the inhibitory role in different parasitic PDFs.⁶ Actinonin suffered poor bioavailability and short of in vivo efficacy in animal models but provided a metal chelator with a peptidomimetic scaffold for further exploration.⁶ Enormous efforts have been taken to synthesize inhibitors and diverse classes of compounds have been reported in the literature. A comprehensive review on PDF inhibitors with a medicinal chemistry focus was recently published by Sangshetti et al., 2015.⁶ Intriguingly, actinonin halts the growth of P. falciparum in a micromolar concentration (3 μM) and showed no effect in a rodent model or malaria under in vivo conditions.^7,17,18 In comparison to bacterial PDF enzyme whose function of formylation or deformylation in the cytosol is essential for its growth, the PfPDF present in the apicoplast is catalytically active as evident from the study of its expression in Escherichia coli and the two inhibitors, N-[(α-mercaptomethyl)caproyl]-L-lysyl-p-nitroanilide (EcPDF K_i = 19 nM) and actinonin (EcPDF K_i = 0.3 nM), which moderately inhibited PfPDF–Co²⁺ enzyme in an intraerythrocytic culture with IC₅₀ values of 57 and 2.5 μM, respectively.⁷ These experimental results provided strong evidence that PfPDF is functional and can be targeted for designing novel antimalarial drugs. However, the efforts required to synthesize and characterize PfPDF inhibitors are comparatively less than the existing antibacterial pursuit. The lower number of studies on PfPDF inhibitors can be attributed to difficulties in PfPDF enzyme purification, the exquisite sensitivity of physiological metal cofactor Fe²⁺ to molecular oxygen, and the unavailability of the PfPDF–inhibitor complex for effective structure-based design.⁶ Amina et al., 2013 have studied the binding modes of actinonin with PfPDF and employed the FlexX docking algorithm to recognize the most promising actinonin analogues (computationally developed).¹⁹ Guilloteau et al., 2002 have solved the crystal structures of PDF from E. coli, Bacillus stearothermophilus, Pseudomonas aeruginosa and Staphylococcus aureus, and compared their crystal conformations with actinonin.¹² Sangshetti et al., 2015 have suggested the urgent need for pharmacophore development studies so as to rationalize the synthesis of chemically diverse molecules.⁶ With progressive efforts focused on the development of computational methods related to pharmacophores and related antimalarial research,^20–22 we present here the molecular models of PfPDF enzyme complexed with various metal ions and studied its role towards actinonin binding. Structure-based pharmacophore screening was also carried out to recognize potential PfPDF inhibitors. To the best of our knowledge, this study is the first report on the pharmacophore development of PfPDF inhibitors.

Materials and methods

Selection of PDF structures

The structure of PDF in complex with Co²⁺ (metal cofactor) and synthesized inhibitor from Plasmodium falciparum (PfPDF; entry: 1RL4)²³ was retrieved from the RCSB Protein Data Bank (PDB).²⁴ Robien et al., 2004 synthesized (2S)-6-amino-2-[(2S)-2-[(N-hydroxyformamido)methyl]hexanamido]-N-phenyl hexanamide as an inhibitor and recorded the crystal structure (1RL4).²³ We refer to this molecule as the synthesized inhibitor throughout the study to distinguish it from rest of the molecules. The FASTA sequence of the PfPDF protein chain was used to search PDF proteins bound to actinonin and different metal cofactors from various organisms using the NCBI BLAST program.²⁵ We have chosen the blastp (protein–protein blast) algorithm to search the PDB structure database²⁴ and the resulting protein sequence alignments were scored using the BLOSUM62 matrix. From the search hits, which were initially sorted by the expect value (E-value) and sequence identity (%), a manual selection of PDF proteins was made and downloaded from the PDB.²⁴ Only a single protein chain bound with metal cofactor and actinonin in each PDF structure was retained and subsequently energy minimized (1000 iterations) using the CHARMM force field²⁶ and conjugate gradient approach via the Prepare Protein module of Accelrys Discovery Studio version 3.0 (Accelrys, San Diego, USA).²⁷

Morphing of PfPDF proteins with various metal cofactors and actinonin

We employed the protein structure morphing technique using the MatchMaker tool²⁸ of the UCSF (University of California, San Francisco) Chimera 1.10.1 program²⁹ to create PDF proteins complexed with actinonin and different metal ions. MatchMaker is a python-based structure comparison tool which relies on the alignment of protein chains to superimpose structures accordingly.²⁸ In the MatchMaker tool, the PfPDF–Co²⁺-synthesized inhibitor (1RL4) and manually selected PDF proteins were defined as the reference structure and structure(s) to be matched, respectively, in a pairwise manner. The sequence alignment was guided using the Needleman–Wunsch algorithm³⁰ and BLOSUM62 scoring matrix³¹ with gap opening and extension penalties of 12 and 1, respectively. MatchMaker imparts 30% weightage on secondary structure score and 70% weightage on residue similarity term.²⁸ This structural matching was iterated by pruning atom pairs until no pair exceeded 2.0 Å distance. In the returned superimposed PDF structures, the PfPDF protein chain from the reference and metal cofactor (Zn²⁺ – 1IX1,¹⁶ Ni²⁺ – 1LQY,³² Co²⁺ – 2OS3 (ref. 33)) along with its respective actinonin molecule from matched structures were only retained to save as PfPDF ‘morphed’ proteins. This step ensured the relative translation of metal ions and other heteroatoms (ligands) with respect to Cα atoms of the protein chain while performing 3D superimposition. The E. coli PDF (EcPDF) in complex with Ni²⁺ and actinonin molecule (1G2A)³⁴ was superimposed with PfPDF–Co²⁺–synthesized inhibitor (1RL4; selected as reference structure) to create PfPDF–Co²⁺–actinonin complex. We manually substituted Fe²⁺ ion in place of Co²⁺ in the PfPDF–Co²⁺–actinonin complex to create PfPDF–Fe²⁺–actinonin complex. Conclusively, we developed four PfPDF morph proteins viz. PfPDF–Co²⁺–actinonin (from 2OS3), PfPDF–Zn²⁺–actinonin, PfPDF–Ni²⁺–actinonin and PfPDF–Fe²⁺–actinonin complexes to analyze the effect of metal cofactor and its geometry towards actinonin binding. We used superimposed protein, PfPDF–Co²⁺–actinonin (from 1G2A) for structure-based pharmacophore modeling. The root mean square deviation (RMSD), a standard metric to judge the quality of protein structure comparisons³⁵ was calculated for Cα atoms from protein chains.

To the structures to be matched was applied rigid body translation T(T_x, T_y, T_z) and rotation R(R_x, R_y, R_z) to the reference structure. The former and latter proteins were considered (x₁, y₁, z₁) and (x₂, y₂, z₂) coordinates of Cα atoms, respectively. Similarly, the RMSD between identical heteroatoms of small molecules in two different conformations was computed.

Docking of actinonin and hit molecules with PfPDF proteins

The docking of actinonin and hit molecules in the PfPDF active site was executed using the FlexX docking algorithm^36,37 implemented in LeadIT, a comprehensive drug design suite (BioSolveIT, GmbH).³⁸ The PfPDF proteins were initially processed by the Receptor Preparation Wizard in which the polar hydrogens were added, crystallographic waters removed, and atom-typed. The metal ion was included as the chain receptor component to retain the metal cofactor during docking. The binding site was carved out by the definition of the actinonin bound position in the crystal structure with a radius of 10 Å. The chemical ambiguities in the selected binding site were resolved by ProToss,³⁸ an internal optimization procedure which considered residues, cofactor and the reference ligand to optimize the H bonding network of the binding site and assigned the appropriate protonation state and tautomers (rotamer flips) to residues. It was taken care that Glu199 was assigned negative charge according to structure notes.^{16,23,32–34} The important step in our study is the assignment of the appropriate metal coordination geometry to the included metal cofactor in the binding site. Unique to the FlexX and FlexXSIS (Japan) docking engines, it comprises an automated metal coordination chemistry model derived from the statistics of known metal coordination in the PDB structures.³⁸ A receptor description file (rdf) was saved.

The actinonin and hit molecules drawn in Marvin Sketch 5.10.3 (ChemAxon, LLC)³⁹ were prepared by correcting atom (including hybridization states) and bond types, adding H atoms with reasonable geometries, assigning formal charges to each atom, and finally, energy minimized using TRIPOS force field⁴⁰ with conjugate gradient approach. A library file was created in SYBYL mol2 (ref. 40) format suitable for ligand import in LeadIT.

The FlexX docking algorithm^36,37 is based on an incremental construction strategy which consists of three phases: base selection, base placement and complex construction. FlexX considers receptor (represented as interaction surface) as rigid and ligand (positioned as interaction centers) as flexible molecules. Initially, the torsional angles at acyclic single bonds are detected in the ligand to ensure ligand conformational flexibility. In the base selection phase, a base fragment having mostly three interaction centers arranged in triangular form (customizable) in the ligand is selected with emphasis on the presence of a maximum number of interaction groups and less size so as to prevent internal flexibility possibilities. In the base placement step, transformations are performed between selected triangular interaction centers from the base fragment onto a triangle of interaction points in the receptor surface to achieve ideal matches which are further filtered through predefined angular constraints. The complete-linkage hierarchical clustering method is applied to cluster initially placed base fragments in order to generate diverse base placements in the active site and a distance function (RMSD) is used to differentiate varied and refine base placements. FlexX supports only geometrically restrictive interaction types including the examination of the metal (interaction center) and its corresponding amino acid (metal acceptor) for scoring dock solutions. Once favorable placements of base fragments are sorted, the algorithm enters the third phase of complex construction wherein the remaining part of the ligand is further sub-divided at each acyclic single rotatable bond to create fragments and subsequently represented in a search tree. Hierarchically, the base fragment is considered as the parent node while the generated fragments are assigned subsequent nodes (leaves). The successive addition of fragments to the initially placed (favorably and refined) base fragment is guided using a simple greedy heuristic applied on the search tree, which attempts to identify energetically favorable k best partial placements, leading to the incremental construction of the whole ligand. After searching for new interactions, optimization of placements and clustering of the solution set, ranking of the generated dock solution is scored using a modified Böhm function⁴¹ to estimate the free energy of binding (ΔG) for protein–ligand interactions.

Here, ΔG₀ is the fixed ground term of the protein–ligand complex; the entropy caused by ligand binding (conformational entropy) is calculated by ΔG_rot and N_rot terms. The next three terms compute pairwise interactions based on the supported geometrical interaction model function f where ΔR and Δα are distance and angular conditions. Hydrophobic interactions and unfavorable steric clashes due to atom–atom contacts are estimated by the last term.^36,37

The FlexX docking parameters and chemical parameters to handle steric clashes were the defined default. A hybrid enthalpy and entropy-based docking strategy was chosen to place base fragments during simulations. The maximum number of solutions per iteration was set to 200 and the maximum number of solutions was given as 200 in order to obtain reasonable dock poses. The protein–ligand 2D interaction maps for pharmacophore fit and dock pose were generated by the Accelrys Discovery Studio (Accelrys, San Diego, USA)²⁷ and PoseView widget⁴² of LeadIT (BioSolveIT, GmbH).³⁸

The enzyme–inhibitory complexes were energy minimized using the YASARA force-field in the YASARA structure program,⁴³ to infer the reliability of the docking results.

Structure-based pharmacophore modeling of PfPDF–Co²⁺–actinonin complex

A structure-based pharmacophore model for PfPDF–Co²⁺–actinonin inhibitor complex was developed using the receptor–ligand pharmacophore generation workflow of the Accelrys Discovery Studio (Accelrys, San Diego, USA).²⁷ This modeling procedure assumes that the input receptor–ligand complex is in a low-energy conformation state suitable for eliciting activity and perceives default pharmacophore types based on the interactions made by the bound small molecule in the protein binding site. The following feature types are supported: H bond acceptor (HBA) and donor (HBD), positive (PI) and negative ionizable (NI) centers, hydrophobic (HY), ring aromatic (RA) and excluded volumes. The distance constraints (maximums: charge = 5.6 Å, H bond = 3.0 Å, hydrophobe = 5.5 Å, exclusion volume = 5.0 Å; minimum inter-feature = 2.0 Å) and both site and projection points were used in feature mapping. We requested 10 pharmacophore models and defined the maximum and minimum number of features to 6 and 4, respectively. Built on a training set of 3285 pharmacophore models, the Genetic Function Approximation (GFA) model⁴⁴ was used to compute the selectivity score with the input of the total number of features and its inter-feature distance bin values. This score represents the complexity of the generated pharmacophore hypothesis (the combination of varied feature type in a model) to demonstrate selectivity and to study the extent from randomness based on the known observations.²⁷ The resulting models were ranked by selectivity score among which the top scored model in accordance with the knowledge of experimental details of intramolecular contacts from EcPDF–Ni²⁺–actinonin inhibitor complex³⁴ was selected for further study.

Validation of structure-based pharmacophore model

The structure-based pharmacophore model was validated by a validation set comprising both known experimental inhibitors and inactive molecules. The inactive molecules can be procured by distinguishing molecules with low inhibitory activity against the target protein and random diverse molecules from the compound library or database with the fact that such molecules will be inactive apparently due to the absence of activity against a selected protein.^45,46 This selection may help to study the capability of the pharmacophore model to distinguish actives from other molecules. Due to the non-existence of experimental PfPDF inhibitors, we selected two series of hydroxamic acid derivatives tested against EcPDF viz. α-substituted hydroxamic acids and N-alkyl urea hydroxamic acids.^15,47,48 To select inactives from this experimental validation set, a cut-off of <7.8 on the logarithms of the IC₅₀ value (pIC₅₀ =−log IC₅₀) was applied. The ChEMBL database⁴⁹ was utilized to download random molecules to include as randomized set. The final validation set consisted of 33 experimental actives, 16 experimental inactives and 103 random inactives leading to 152 molecules.

We adopted the scale fit value method⁵⁰ of the hit enumeration step to recognize actives from the validation set. Due to the highly complex combination of 3D features in the selected pharmacophore model, we considered a molecule as a hit if it secures a fit value of any range and otherwise as a non-hit. This condition of active recognition was only applied to the validation set. In database screening, we espoused default fit value calculations to select top hits among hit candidates.

The fit value of a molecule is computed by the equation.

The first term is the number of pharmacophore features in the hypothesis that map the functional groups of the molecule. W is the weight associated with the feature sphere (default 1.0). The weight term included the distance between the feature centroid in the sphere and its corresponding chemical groups and tolerance is the radius of the feature sphere (default 1.6 Å).⁵⁰

Database screening of structure-based pharmacophore model

The selected pharmacophore hypothesis was used to search the Druglike Diverse database comprising 5384 molecules in the Accelrys Discovery Studio (Accelrys, San Diego, USA).²⁷ We had chosen the ‘best’ search method to select pre-generated diverse conformations using the CHARMM force field²⁶ and Poling algorithm⁵¹ (energy threshold = 20 kcal mol⁻¹). The alignment of hypothesis features with database molecules was enabled to prevent the recognition of molecules only with the mere presence of features as hits, with the minimum inter-feature distance of 0.5 Å. This screening protocol generated a list of hit molecules sorted by fit value among which the top five hits were selected for further study. The hit molecules were subsequently docked and their interactions analyzed with the PfPDF–Co²⁺ receptor as described above. Statistical analysis was performed using the SPSS v.18 (SPSS Inc.) package.⁵²

ADME properties

We assessed the ADME (Adsorption, Distribution, Metabolism and Excretion) properties of the prioritized hit molecules using PreADMET,⁵³ a web based application from Yonsei University, Republic of Korea. The PreADMET program utilizes in vitro results and its classification schemes related to Human Intestinal Absorption [HIA (%)], cell permeability from Caco-2 [human colon adenocarcinoma cells possessing multiple drug transport pathways through the intestinal epithelium (nm s⁻¹)] and MDCK cell model [Madin–Darby canine kidney cell (nm s⁻¹)] and distribution for Plasma Protein Binding [PPB (%)] and Blood–Brain Barrier penetration [BBB; C.brain/C.blood] prediction.

Results and discussion

Structural attributes of PfPDF enzyme

We performed a structure-based pharmacophore screening of PfPDF in complex with Co²⁺ and actinonin as metal cofactor and inhibitor, respectively. The PfPDF tertiary structure exhibit a mixed α–β topology with four α-helices and eight anti-parallel β sheets and a conserved metal-binding structural motif. The metal-binding site comprises a metal ion Co²⁺ (PDB entry: 1RL4) which tetrahedrally coordinated to two histidines (His198 and His202) from the zinc hydrolase sequence (HEϕDH motif), one cysteine (Cys156) from the EGCϕS motif (ϕ represents any amino acid) and a water molecule (Fig. 1).²³ PfPDF shares 33% and 28% sequence identity with EcPDF (Fig. 2)³⁴ and Homo sapiens PDF (HsPDF), respectively. Despite highly conserved active site residues among PfPDF, EcPDF and HsPDF (multiple sequence alignment among these PDFs was shown by Sangshetti et al., 2015);⁶ significant evolutionary changes can be observed. The N-formylated group of N-terminal methionine from processed polypeptide during protein maturation acts as substrate in PDFs. In the substrate binding region of EcPDF and PfPDF, the presence of Ile195 (PfPDF) in place of Cys129 (EcPDF) decreases the volume of the cavity and the presence of Ile106 (PfPDF) in place of Ile44 (EcPDF) which are localized close to the active site region, developed a ridge on the floor of the active site and thus, participate in cavity volume reduction. Comparison of PfPDF with HsPDF indicates that the biochemical properties of amino acids present in the substrate binding pocket are very much varied (Lys104 (PfPDF) – Arg107 (HsPDF), Ile106 – Val109, Glu162 – Leu179 and Ile153 – Pro169).⁶ The knowledge of amino acids and their function related to removal of the N-formyl group at N-terminal methionine could help in designing potent inhibitors for PfPDF metalloenzyme. In addition, the actinonin (inhibitor)-bound conformation of EcPDF (PDB entry: 1G2A)³⁴ will highlight the crucial amino acids required for PfPDF inhibition.

Fig. 1 The crystal structure of PfPDF complexed with Co²⁺ (metal cofactor), active ligand (bound to active site) and inactive ligand (inter sub unit molecule existed as dimer) (PDB: 1RL4).

Fig. 2 The crystal structure of EcPDF complexed with Ni²⁺ (metal cofactor) and actinonin (inhibitor) (PDB: 1G2A).

Despite the overall structural similarities of EcPDF and PfPDF, structural studies revealed significant differences in the substrate binding regions confirmed by the absorption spectrum which showed the differences in the d–d transitions around the metal center.⁸ The changes in electrostatic and cavity shrinkage brought out by the amino acid differences between EcPDF and PfPDF enzymes noted above lead to the dramatic changes in the binding position of actinonin to PfPDF which deviated from its expected optimal binding conformation as observed in EcPDF and this can be one of the reasons for decreased affinity of the substrate or inhibitor to PfPDF.⁸ This study uncovers the mechanistic details of actinonin probable binding conformation to PfPDF enzyme and promotes PfPDF as a potential drug target to target.

Proposed catalytic mechanism of PfPDF

Based on the molecular catalytic cycle proposed in EcPDF,⁶ the catalytic mechanism carried out by PfPDF can be studied given that the residues involved in this catalysis are identical (Fig. 3). The initial state of PfPDF (apo-form) is highly stabilized by the H bonding network between two water molecules (designated as W₁ and W₂) and pocket amino acids (1). Further, the Fe²⁺ cofactor ion exhibits tetrahedral ligation to Cys156, His198, His202 and W₁. The amide from the main chain of Leu157 and the side chains of amide of Gln112 and carbonyl of Gly107 are involved in H bonding with W₂. The binding of the formyl group dissociates W₂ from PfPDF enzyme, leading to the polarization of the carbonyl oxygen from the formyl carboxylate group by establishing H bonds with Leu157 and Gln112 (2). The enzyme complex enters into the transition state by the W₁-mediated nucleophilic attack on the carbonyl oxygen of the formyl group (3). W₁ probably exists in the deprotonated state for this reaction step. The carbonyl oxygen of the formyl group is ligated to Fe²⁺, the carbonyl carbon of formyl group and the side chains of Gln112 and Leu157, and develops a tetrahedral geometry. Now, the carbonyl carbon of the formyl group moves from sp² to sp³ electronic configuration and accompanies a transition from tetrahedral to penta-coordinated Fe²⁺ center. The proton from W₁ is donated to the amide group present at the N-terminal peptide using Glu199 which acts as a general acid and subsequently assigned positive charge ensuring the nitrogen atom is vulnerable to leave the tetrahedral intermediate structure.

Fig. 3 Proposed catalytic cycle of PfPDF from the knowledge of EcPDF catalysis.

Bond cleavage events occur in the ternary enzyme–formate–peptide complex wherein the formate and the cleaved N-terminal peptide are bound to the penta-coordinated metal center and Glu199, respectively (4). Subsequently, the peptide dissociates from the enzyme complex, followed by the release of formate and intake of water molecules (W₁ and W₂), leading to the activated form of enzyme complex (5). The ability of transferring the formyl group from one formyl peptide to another peptide and its related biochemical data suggested a ‘ping-pong’⁶ or double-displacement mechanism of catalysis.

The effect of various metal ions and actinonin binding to PfPDF

The unavailability of the PfPDF structure in complex with actinonin²⁴ has drastically hampered the drug design campaign for PfPDF. The major obstacle was the difficulty in finding crystallization conditions to obtain the active PfPDF enzyme structure and the extraordinary labile nature of PfPDF towards inactivation in physiological in vitro conditions.^6,23 An initial crystallographic study on carboxy-terminal hexahistidine tagged PfPDF was prone to aggregation and numerous difficulties arise in the PfPDF purification, crystallization, diffraction and phase determination.⁶ Robien et al., 2004 have reported a crystal structure of PfPDF in complex with a synthesized inhibitor, (2S)-6-amino-2-[(2S)-2-[(N-hydroxyformamido)methyl]hexanamido]-N-phenylhexanamide. This molecule will be referred to hereafter as the ‘synthesized inhibitor’. This 2.2 Å resolution structure reported an IC₅₀ value of 130 ± 26 nM against PfPDF enzyme. Unexpectedly, the synthesized inhibitor molecule underwent proteolytic cleavage during crystal growth. The proteolyzed fragment, (2R)-2-[(N-hydroxyformamido)methyl]hexanoic acid, was bound to the active site whereas the acid hydrolyzed fragment, (2S)-6-amino-2-[(2S)-2-[(hydroxyamino)methyl]hexanamido]-N-phenylhexanamide, which existed as a dimer, was characterized as an intersubunit molecule.²³ It should be noted that the binding of Co²⁺ in the metal-binding triad center (Cys156–His198–His202) was unaltered in comparison to other metalloenzymes (Fig. 1). It is a challenging task to recognize a physiologically relevant metal ion (cofactor) of a metalloenzyme.⁵⁴ Evidence from recombinant PfPDF enzyme purification and a crystallization screen had suggested that Fe²⁺ was irreversibly oxidized to Fe³⁺ rapidly by atmospheric O₂ leading to inactive enzyme.^14,23,34 To circumvent this labile nature of PDFs, the introduction of other divalent cations such as Ni²⁺, Co²⁺ and Zn²⁺ can be used as an alternative route to obtain an enzyme with better stability and less activity.¹³ This strategy was used by Robien et al., 2004 to solve the PfPDF–Co²⁺ crystal structure.²³ Most of the bacterial PDFs utilize Fe²⁺ ion and especially, native EcPDF recruits Fe²⁺ for catalytic activity.^13,55–57 In an Fe²⁺ rich medium, EcPDF develops preferential affinity to Fe²⁺ than Zn²⁺ ion and utilizes Zn²⁺ only when Fe²⁺ is depleted from the medium. This mechanism of recruiting a metal ion depends upon the intracellular Fe²⁺ concentration.⁵⁴ The unresolved mechanistic puzzle due to the significant difference in the catalytic efficiency of EcPDF–Fe²⁺ and EcPDF–Zn²⁺ did not hamper the efforts to develop potent antibacterial PDF inhibitors. However, much less effort has been made to design PfPDF inhibitors.

To understand the role of various metal ions and their relevance to actinonin binding to PfPDF, we considered the morphing technique of protein structure comparisons.⁵⁸ The protein sequence of PfPDF in complex with synthesized inhibitor (1RL4)²³ was retrieved in the FASTA format and the PDB structure²⁴ database searched using the BLAST²⁵ interface. The sequence hits were sorted based on E-value and % identity. The hit list was further refined by emphasis on selecting the PDF enzyme with different organisms and complexed with actinonin and different metal ions. Among seven PDFs, only three structures were considered for further study (Table 1). These include Pseudomonas aeruginosa (Zn²⁺; 1IX1),¹⁶ Bacillus stearothermophilus (Ni²⁺; 1LQY)³² and Enterococcus faecalis and Streptococcus pyogenes (Co²⁺; 2OS3).³³ The EcPDF–Ni²⁺–actinonin complex (1G2A)³⁴ (Fig. 2) was not used in the morphing step to generate a new PfPDF structure due to the identical crystallized metal ion (Ni²⁺) with 1LQY.

Table 1 List of PfPDF actinonin-containing homologues obtained by BLAST similarity search

S. no.	PDB entry	Organism	Resolution (Å)	Metal ion	E-Value	Identity (%)
a Selected to create PfPDF morph proteins.
1	1IX1^a	Pseudomonas aeruginosa	1.85	Zn^2+	3 × 10^−34	38
2	1LQY^a	Bacillus stearothermophilus	1.90	Ni^2+	4 × 10^−18	33
3	2OS3^a	Enterococcus faecalis and Streptococcus pyogenes	2.26	Co^2+	6 × 10^−18	34
4	4DR9	Synechococcus elongatus	1.90	Zn^2+	1 × 10^−33	42
5	3U04	Ehrlichia chaffeensis	1.70	Zn^2+	3 × 10^−28	37
6	1SZZ	Leptospira interrogans	3.30	Zn^2+	3 × 10^−29	38
7	1Q1Y	Staphylococcus aureus	1.90	Zn^2+	1 × 10^−14	39

---

In addition, the PfPDF–Co²⁺-synthesized inhibitor (1RL4) was superimposed onto the EcPDF–Ni²⁺–actinonin complex to create the PfPDF–Co²⁺–actinonin complex (RMSD of Cα atoms = 0.879 Å). The morphing step was guided by the MatchMaker tool²⁸ of the UCSF Chimera program²⁹ in which the PfPDF–Co²⁺ complex (developed by deleting all heteroatoms present in PfPDF; 1RL4) was selected as the reference structure while three selected PDFs (1IX1, 1LQY and 2OS3) (Fig. 4) were specified as the structures to be matched. Of note, only the protein sequence-based global alignment was attempted to place the actinonin and metal ion in the PfPDF–Co²⁺ active site. The PfPDF protein chain, actinonin and various metal ions were only retained in the superimposed structures to generate PfPDF proteins with different heteroatoms viz. PfPDF–Zn²⁺–actinonin, PfPDF–Ni²⁺–actinonin and PfPDF–Co²⁺–actinonin. To study the role of Fe²⁺ binding in PfPDF, the PfPDF–Co²⁺–actinonin complex created from superimposition was manually edited to replace with the Fe²⁺ ion to generate the PfPDF–Fe²⁺–actinonin complex structure. These four PfPDF proteins (Fig. 5) were energy minimized to remove unrealistic atomic configurations. The RMSD values computed among the Cα atoms between the reference and matched structures were below 1 Å (Table 2) which signifies confidence and allows structural comparisons.

Fig. 4 The crystal conformation of actinonin bound to selected PDF enzymes. (A) E. faecalis and S. pyogenes – 2OS3, (B) P. aeruginosa – 1IX1 and (C) B. stearothermophilus – 1LQY.

Fig. 5 The structures of PfPDF morph proteins created from E. faecalis and S. pyogenes – 2OS3 (A), P. aeruginosa – 1IX1 (B), B. Stearothermophilus – 1LQY (C) and Fe²⁺ constructed complex (D).

Table 2 RMSD from Cα atoms of PDF protein chains computed by superimposition with PfPDF–Co²⁺-synthesized inhibitor complex (1RL4; reference structure)

S. no.	PDF protein^a	Metal ion	Organism	RMSDCα value (Å)
a Co-crystallized with actinonin inhibitor.
1	1IX1	Zn^2+	P. aeruginosa	0.756
2	1LQY	Ni^2+	B. stearothermophilus	0.876
3	20S3	Co^2+	E. faecalis and S. pyogenes	0.873

---

We had previously shown the importance of metal ion (Fe²⁺) geometry to organize the transition state required for nucleophilic attack, bond cleavage and substrate binding reaction steps of PfPDF enzyme. It will be of major interest to study the geometry of metal ions prevalent in the selected PDF structures bound with actinonin. LeadIT, a comprehensive drug design suite (BioSolveIT, GmbH),³⁸ comprises a geometry prediction tool in the Receptor Preparation Wizard (a prior step to docking) to predict the orientation of metal ions and specify the nature of restraint during FlexX^36,37 docking simulation. The entire selected PfPDF structures constituted penta-coordination in the crystallographic notes^16,32–34 which was correctly predicted by LeadIT in the PfPDF complexes of Co²⁺ from P. falciparum (1RL4) and E. faecalis and S. pyogenes (2OS3). The tetrahedral geometry was predicted in PfPDFs with Zn²⁺ (1IX1) and Ni²⁺ (1LQY) metal cofactors (Table 3). The list of metal coordinating residues and their distance computed from crystal and morphed structures are given in ESI Tables 1 and 2.†

Table 3 Known and predicted metal ion geometry of PDF enzymes

S. no.	PDB entry	Organism	Metal ion	Geometry present in crystal structure (coordination)	Geometry predicted by LeadIT package (coordination)
a This PfPDF–Fe^2+–actinonin complex was created by a superimposition technique using EcPDF–Ni^2+–actinonin (1G2A) with PfPDF–Co^2+-synthesized inhibitor complex (1RL4).
1	1RL4	P. falciparum	Co^2+	Penta	Penta
2	1IX1	P. aeruginosa	Zn^2+	Penta	Tetra
3	1LQY	B. stearothermophilus	Ni^2+	Penta	Tetra
4	2OS3	E. faecalis and S. pyogenes	Co^2+	Penta	Penta
5^a	—	—	Fe^2+	—	Penta

---

Further, we were interested in studying the effect of various metal ions with their respective geometry in influencing actinonin binding. The superimposed morph PfPDF structures revealed no significant change in the co-crystal conformations of actinonin present in the morph PfPDF proteins as shown by RMSD values (Fig. 6). Thus, these actinonin conformations indicate that metal cofactor geometry will not affect actinonin binding to PfPDF and allow the docking study to be performed regardless of the metal coordination geometry. However, we relied on the PfPDF–Co²⁺–actinonin structure obtained from the superimposition technique to mimic the active site region as closely as possible.

Fig. 6 The superimposed pose of PfPDF morph proteins.

Since enormous efforts were made to biochemically characterize EcPDF and its associated inhibitor design, structural and biochemical investigators have hypothesized about the preferential affinity of one metal ion over another and its related activity difference in EcPDFs.¹² In Borrelia burgdorferi and Lactobacillus plantarum PDFs, Zn²⁺ binds more tightly to the tetrahedral metal-binding center than Fe²⁺ and Co²⁺. It was assumed that the binding of Zn²⁺ to the tetrahedral metal center brings no substantial structural changes in the metal center upon Zn²⁺ binding than other divalent transition metals with partially filled 3d orbital configurations.⁵⁴ Some researchers have anticipated that the difference in PDF activity with Zn²⁺ and Fe²⁺ (including Co²⁺ and Ni²⁺) cofactors is its ability to develop a penta-coordination transition state.^59,60 Nguyen et al., 2007 suggested the cooperation of a metal ion and its outer shell ligands in creating the catalytic environment responsible for the PDF activity difference and attributed this to its slight variations in metal–ligand bond lengths or geometry.⁵⁴

Re-docking of actinonin to morph PfPDF structures

The LeadIT FlexX module recommends a plausibility check to re-dock the cognate ligand into the prepared protein structure and examine dock poses that should secure RMSD below 2 Å. Poses securing more than 2 Å RMSD either do not make sense or increased chances to develop incorrect poses with high likelihood.^36–38 The inspection of active sites from various selected PDFs showed that actinonin did not develop any non-covalent bond with water molecules and hence, water molecules did not play any role in actinonin binding and were removed from receptor structures to perform docking. Furthermore, the side chain of Glu199 of PfPDF was assigned negative charge in accordance with the PfPDF–Co²⁺-synthesized inhibitor complex (1RL4)²³ and other selected PDF metalloenzymes.^16,32–34 The principle bottleneck of docking scoring functions is to impose the metal center with a non-covalent parameter instead of treating its partial covalent nature. The classical scoring functions consider metals as atoms with van der Waals radius and assign point charges which would result in high electrostatic interaction and secure large desolvation penalties.⁶¹ Irwin et al., 2005 have retrospectively tested the ability of docking screens to obtain poses close to experimental conformations in five different metalloenzymes⁶¹ and showed the reproduction of close to experimental results using the DOCK 3.5.54 docking program.⁶² Few reports on the discovery of novel PDF and metalloenzyme inhibitors using conventional and FlexX docking algorithms have been published.^63,64 Interestingly, the re-docking of actinonin to EcPDF–Ni²⁺–actinonin complex (1LQY) from the Irwin et al., 2005 study generated a high-scoring docked geometry of 0.9 Å RMSD while the lowest pose with 0.7 Å was ranked fifth in the interaction energy-based rank list.⁶¹ Hence, the docking experiment of actinonin with PfPDF enzyme would procure reasonable results with the support of available literature reflecting the efficiency of the FlexX docking algorithm in metalloproteins.^61,63,64

The re-docking of actinonin among selected PfPDF enzymes (Fig. 7 and ESI Fig. 1†) showed the top scoring of the PfPDF–Co²⁺–actinonin complex with a score of −32.14 kJ mol⁻¹ whereas the Ni²⁺ complex secured a comparatively less interaction score of −19.05 kJ mol⁻¹. The PfPDF–Fe²⁺ complex possessed an interaction score of −30.95 kJ mol⁻¹ (Table 4). The interaction energy-based trend among various metal ions can be written as Co²⁺ > Fe²⁺ > Zn²⁺ > Ni²⁺. This trend was observed due to the differences in the matched functional groups among the top poses of actinonin in the respective PfPDF structures. Furthermore, this order was not similar to the trend among RMSD values which emerged from actinonin co-crystal and dock conformations (Table 4).

Fig. 7 The dock poses of actinonin in the PfPDF morph proteins.

Table 4 Re-docking of actinonin in the PfPDF–metal ion–actinonin ‘morphed’ complexes

S. no.	PfPDF morph proteins^a	Interaction score^b	Match^b	Lipo^b	Ambig^b	Clash^b	Match^c	RMSDhet value (Å)
a Redocking of actinonin molecule in the PfPDF morph proteins containing various metal ions and crystal conformations of actinonin ‘morphed’ from selected PDF structures. See Table 1.b Values are expressed in kJ mol^−1; relate the terms with FlexX ΔG equation; the rotatable term constitutes an identical score of +18.2 kJ mol^−1.c The number of matched groups in initial and dock poses.
1	Co^2+	−32.1407	−41.9359	−8.7705	−8.8883	3.8540	11	0.8584
2	Zn^2+	−27.3846	−35.8662	−10.2186	−11.0251	6.1252	10	1.5582
3	Ni^2+	−19.0493	−30.0753	−12.1221	−9.9077	9.4557	7	1.4155
4	Fe^2+	−30.9534	−40.6256	−8.8084	−8.7116	3.5921	11	1.5863

---

The energy minimized scores of the dock pose obtained from the YASARA force field⁴³ were close together in the range of −1198.19 kJ mol⁻¹ to −1237.63 kJ mol⁻¹ and this step enhanced the post-docking results and the reliability of the LeadIT docking protocol, shown in ESI Fig. 2 and Table 3.†

The PfPDF–Co²⁺ complex achieved the lowest interaction score and comparatively less RMSD of 0.858 Å. This observation was supported by the experimental results of the Co²⁺ containing PfPDF enzyme crystallized by Kumar et al., 2002 (PDB entry: 1RL4).⁸ Bracchi-Ricard and his co-workers found that the PfPDF–Fe²⁺ complex was exquisitely sensitive to molecular oxygen in a similar way to EcPDF. The replacement of Fe²⁺ by Co²⁺ metal cofactor in EcPDF led to the generation of a stable enzyme complex without compromising the catalytic efficiency of its native protein.⁷ These results inspired Kumar and his co-workers to utilize Co²⁺ as a suitable metal cofactor to crystallize PfPDF enzyme and found it to be equivalently stable.⁸ Combining the experimental outcomes related to Co²⁺ stability with catalytic activity and the predicted energetics of the PfPDF–Co²⁺–actinonin complex, it is evident that Co²⁺ is the most preferable metal cofactor next to Fe²⁺. Hence, PfPDF with Co²⁺ metal ion can be used for structure-based antimalarial drug design and we considered Co²⁺ as metal cofactor in docking and structure-based pharmacophore screenings.

Consensus binding pattern of actinonin bound to various metal ions of PfPDFs

The 2D PfPDF–metal ions–actinonin interaction maps (Fig. 8) were deciphered using the PoseView module⁴² of LeadIT (BioSolveIT, GmbH).³⁸ The carbonyl group situated adjacent to the hydroxamate moiety in actinonin was found to have conserved H bonding with the amide (main) chain of Leu157 in the PfPDF complexes of Co²⁺, Ni²⁺ and Fe²⁺ except for the Zn²⁺ structure. Additionally, this carbonyl group has developed an H bond with the amide side chain of Gln112 in all the PfPDF complexes. Furthermore, the carboxyl group, the other side chain group of Gln112, established a conserved H bond with hydroxyl group of the hydroxylamine moiety in the actinonin molecule. The negatively ionized residue Glu199 developed an H bond with the amide group of actinonin’s hydroxamate moiety in all the PfPDF structures except for the Ni²⁺ complex where Gly107 made the H bond instead of Glu199. The metal cofactors Co²⁺, Fe²⁺ and Zn²⁺ formed H bonds with the hydroxyl and carboxyl groups of the hydroxylamine moiety. The Ni²⁺ cofactor developed a single H bond with the carboxyl group from this moiety. Ile106 established conservativeness in H bonding by interacting with the carbonyl group present in the aliphatic octan-2-one group of actinonin. The hydrophobic isopropyl group was found to adopt different conformations owing to its flexible nature. The amide (main) chain of Gly155 in all the PfPDF complexes developed H bonds with the carbonyl group attached to the pyrrolidine ring. In the case of Co²⁺ and Zn²⁺ complexes, the same amide chain of Gly155 interacted with the hydroxyl group attached to the pyrrolidine ring using H bonds. The hydrophobic interactions can be studied from 2D interaction maps (Fig. 8) which are not discussed for the sake of brevity and the self-explanatory nature of the maps.

Fig. 8 The 2D interaction maps of dock poses obtained from re-docking to PfPDF morph proteins.

Structure-based pharmacophore model of PfPDF–Co²⁺–actinonin complex

The bioactive conformation of actinonin bound to the PfPDF–Co²⁺ complex obtained through the superimposition technique was used in structure-based pharmacophore screening to recognize compound hits having identical pharmacophore features enumerated from PfPDF–Co²⁺–actinonin (receptor–ligand) interactions (Fig. 9). The receptor–ligand pharmacophore generation protocol of the Accelrys Discovery Studio (Accelrys, San Diego, USA)²⁷ was used to develop selective pharmacophore models. We selected predefined pharmacophore feature types viz. H bond acceptor (HBA) and donor (HBD), hydrophobic (HY), positive (PI) and negative ionizable (NI) centers and ring aromatic (RA) as components of pharmacophore models. The top ten scored pharmacophore models ranked by selectivity score was requested. Ten different pharmacophore models with varied conformations of feature types were reported with an identical selectivity score of 11.978 (Table 5). Since the input actinonin molecule constituted 22 features, only 10 were selected in each pharmacophore model in accordance with the PfPDF–actinonin interactions, AAAAADDHHP (i.e. HBA = 5, HBD = 2, HY = 2, PI = 1). The selectivity score is estimated using the Genetic Function Approximation (GFA) model built from a training set of 3285 pharmacophore models.^27,44 Due to the identical selectivity scoring of pharmacophore models, we relied on the knowledge of EcPDF enzyme inhibition mechanisms obtained through crystallographic studies and inhibitor design strategy^6,12,34 from antibacterial drug design campaigns which may highlight the importance of interactions made by pocket residues and inhibitor functional groups.

Fig. 9 The PfPDF–Co²⁺–actinonin pharmacophore model (A), the pharmacophore query used in structure-based screening (B) and projected with the actinonin inhibitor (C) showing pharmacophore features and (C) with ligand actinonin.

Table 5 Summary of feature combinations of pharmacophore models derived from intramolecular contacts in PfPDF–Co²⁺–actinonin complex

S. no.	Pharmacophore model^a	Feature set
a The entire pharmacophore models constituted six features and secured identical selectivity score of 11.978.
1	Model_01	AAADHP
2	Model_02	AAADHP
3	Model_03	AADHHP
4	Model_04	AADHHP
5	Model_05	AADHHP
6	Model_06	AAADHP
7	Model_07	AAADHP
8	Model_08	AADHHP
9	Model_09	AADHHP
10	Model_10	AADDHH

---

Selection of pharmacophore feature combinations in relation to PfPDF–Co²⁺–actinonin intramolecular contacts and substrate binding pockets

To prioritize pharmacophore features embodied in the top scored models, the PfPDF–Co²⁺–actinonin complex was thoroughly studied with respect to the PfPDF substrate binding pockets and functional groups of actinonin.⁶ Structural comparison of EcPDF with PfPDF revealed three substrate binding pockets in the active site along with the adjacent metal-binding site. The substrate binding pockets are designated as S₁′, S₂′ and S₃′ and the respective functional groups in the substrate or inhibitor are referred as P₁′, P₂′ and P₃′.^6,12 Highly conserved among bacterial PDFs,¹² the S₁′ pocket is lined by hydrophobic residues and serves as binding site for methionine in which the attached formyl group will be subjected to deformylation. Good enzyme inhibition can be achieved by molecules having a corresponding P₁′ functional group (mostly a hydrophobic moiety) along with chemical groups (projected adjacently) that specifically interact with metal cofactor Fe²⁺ to ensure the penta-coordination geometry of the Fe²⁺ center.^59,65,66 The S₂′ pocket resembles an open-tunnel cavity with solvent accessibility.¹² Chemically diverse side chains connected to numerous formylated substrates can be occupied here in order to get processed by PDF proteins.⁶⁷ Similar to the S₂′ pocket, S₃′ is also exposed to solvent and is least conserved in relation to other pockets¹² which suggests that diverse functionalization at its respective P₃′ position of the inhibitor can be made.⁶⁵ It is worth mentioning that HsPDF is deficit of S₂′ and S₃′ pockets which pronounces hydrophobic depression⁹ and may not affect the PfPDF inhibitor design strategy wherein we recommend the HBA functional group connected to the hydrophobic moiety occupying the P₃′ position of the actinonin molecule to enhance selectivity of molecules against PfPDF.

Inhibitor design strategies against PDFs are largely achieved using the decomposition of functional groups in the actinonin bound (bioactive) conformation presented by crystal structures with respect to substrate binding pockets.^{6,12,32,34,59,65,66} The general characteristics of PDF inhibitors are the presence of a metal cation–chelator group which may give better binding energy, an n-butyl like group which mimics the methionine side chain of substrate (P₁′) and a side chain in concurrence to P₂′ and P₃′ sites of inhibitors. It is anticipated that P₂′ and P₃′ sites may accommodate diverse functional groups and thereby, promote additional binding energy and may confer selectivity of molecules.^6,68

The PfPDF–Co²⁺–actinonin intramolecular (protein–ligand) interactions obtained through the superimposition technique was examined with the projection of pharmacophore features suggested by the top ranked models (Fig. 10). The n-butyl chain of actinonin was projected as the hydrophobic feature (HY8) in the S₁′ pocket. The amide group from the hydroxamate moiety which developed an H bond with the negatively ionized Glu199 residue was perceived as a positive ionizable center (PI10). Of note, this feature was not explicitly mentioned in the general characteristics of PDF inhibitors but the sub-structure hydroxamate was suggested as the important metal cation-chelator group.^6,68 Due to the consideration of the hydroxamate group as a beneficial sub-structure, the hydroxyl group present in this moiety having H bonds with Gln112 and Fe²⁺ cofactor was understood and did not participate as a feature in the top 5 pharmacophore model. It was superseded by the adjacent carbonyl group which acted as HBA (HBA1) in order to coordinate an Fe²⁺ metal center in a penta-coordination fashion. The projection of HBA1 is consistent with the structures of antibiotic clinical candidates, GSK-1322322, BB-83698 and LBM-415, which possess N-hydroxy-N-methylformamide as a metal chelator group. This observation collectively indicates the requirement of a hydroxyl and carbonyl group to perform H bonding with Gln112 and ligate metal, at least for actinonin analogues. The H bond interactions made by the carbonyl group with Ile106 adjacent to the S₂′ pocket was captured as HBA (HBA3) whereas its connected amide group was selected as HBD (HBD7) which formed an H bond with the Gly155 amino acid. This connected hydroxyl–amide pair can also be noticed in the above mentioned clinical candidates. The hydroxyl group attached to the pyrrolidine ring at the P₃′ position which also developed an H bond with the Gly155 residue was projected as HBA (HBA5). It was expected that HBA5 will be placed as HBD due to the corresponding hydroxyl group. It is primarily due to the dictionary rules which catalyst supports a single feature per functional group. Finally, we selected pharmacophore model_01 which included 6 features and 17 excluded volume spheres (Fig. 11). The spatial coordinates and orientation of pharmacophore features in the selected model are given in ESI Table 4.† The space left in the active site after molecular binding is usually filled with excluded volume spheres and thereby, represents undesirable protrusions in the active site. However, the apt definition of excluded volume, applicable only to multiple structure-based pharmacophore models, is the disallowed region in the active site that was not occupied by chemical groups among active molecules and may penalize molecules that overlap with this steric region.^46,69 Studies on this spatial constraint have shown that its inclusion increases the specificity of the resulting pharmacophore model and better enrichment in virtual screenings.^70,71 This consideration is in good agreement with our objective of recognizing small molecular hits which mimic the projection of pharmacophore features derived from actinonin rather than augmenting actinonin-based similar molecules.

Fig. 10 The substrate binding pockets of PfPDF–Fe²⁺–actinonin complex and adjacent metal binding site merged with functional groups of actinonin.

Fig. 11 The inter-feature distances (Å) of perceived pharmacophore features in the model.

Validation and statistics of selected structure-based pharmacophore model

Given that actinonin constitutes more than 10 flexible groups, the requirement of 6 features in the hits and the inclusion of 17 excluded volume spheres, these geometrically stringent conditions may only recognize hits having minimum features (although we specified all 6 features) and fit over the 3D feature projections in close resemblance to the actinonin conformation. The capability to distinguish active molecules from a pool of actives and presumably inactive or random molecules determines the quality of the pharmacophore model. The concept of merging experimental active molecules with presumably inactives has been successfully employed to validate pharmacophore models in various studies and to study the specificity and sensitivity of the resulting pharmacophore models.^45,46 Due to the absence of PfPDF experimental inhibitors and the conserved active site regions of PfPDF with EcPDF, we selected experimental hydroxamic acid derivatives (actinonin analogues) having antibacterial activity as the experimental molecular set (49 molecules) to validate the presented pharmacophore model This set included two series of hydroxamic acids, α-substituted⁴⁸ and N-alkyl urea hydroxamic acids¹⁵ and comprised both active and inactive molecules for which we defined a cut-off of ≥7.8 and <7.8 pIC₅₀ (=−log IC₅₀) to distinguish as actives (33 molecules) and inactives (16 molecules). A randomly chosen 103 molecules with diverse chemical complexities were also retrieved from the ChEMBL database,⁴⁹ which acted as inactive molecules, was combined together with the experimental molecular set to create a validation set of 152 molecules (total actives = 33 and total inactives = 119 (16 experimental inactives and 103 ChEMBL data)).

Due to geometrically stringent conditions in the pharmacophore model, we slightly modified the hit enumeration step of the catalyst in which the ability of validation molecules to secure a fit value of any range was considered as hits and vice versa, instead of the usual specification of threshold in fit value to regard as hits and non-hits (for instance, compounds securing fit value above 2 are considered as hits and vice versa).^27,50,69 Similarly, a molecule was counted as a hit if one among the pre-generated conformers could be fitted on the model, according to the molecular operating environment package.^71,72 This step was necessary to perform the validation study due to the deficit in the crystallographic knowledge of PfPDF–inhibitor complexes to avoid blind analysis. The screening of pharmacophore model_01 on the validation set yielded model quality statistics^27,73 (Table 6 and ESI Table 5†). Sensitivity which measures the proportion of active molecules that were correctly recognized by the built pharmacophore model as actives, had secured only 36% due to the correct prediction of 12 active molecules out of 33 total actives (Fig. 12). This indicates that the actinonin analogues may adopt different binding conformations among which only 12 active molecules may confine binding poses close to the actinonin crystal conformation. Interestingly, the specificity metric accounts the quantity of inactives that were correctly identified by the constructed model as inactives and secured 96% due to the accurate prediction of inactives or presumably random molecules from the validation set. This observation demonstrated that the presented pharmacophore model possesses maximum likelihood of recognizing specific PfPDF molecules that share all the requested 6 pharmacophore features. An overall accuracy of 83% was achieved using this geometrical constrained complex features. A poor correlation was noted (r² = 0.029) when comparing fit value and antibacterial PDF activity among hits from the experimental validation set (Fig. 13). The poor correlation of fit value and inhibitory activities of the compound was also noted in the quantitative pharmacophore model.⁷⁴ For example Boppana et al., 2009 reported no significant correlation between the fitness score and monoamine oxidase B inhibitory activity.⁷⁴ The pharmacophore generation module computes activity of molecules based on the regression relationship between the geometric fitness and negative logarithms of the activity. The geometric fit verifies the presence and absence of features, maps and also considers a distance between the perceived feature and its feature centroid from the hypothesis to calculate fit value.⁷⁵ This may be the reason for the poor correlation of this validation set. We anticipate that parallel and rapid research on PfPDF inhibitors using medicinal chemistry and crystallographic studies will enhance the chemical diversities of molecules and generate knowledge on structure–activity relationship. The presented pharmacophore model can be enriched accordingly.

Table 6 Statistics of the selected pharmacophore model validated by validation set

Particulars	Values
a Inactives contained 16 experimental inactives and 103 ChEMBL randomized inactives.
Total number of molecules in validation set	152
Total number of actives	33
Total number of inactives^a	119
True positives	12
True negatives	114
False positives	5
False negatives	21
Sensitivity	0.363
Specificity	0.957
Accuracy	0.828

---

Fig. 12 The 2D structures of experimental validation molecules. The compound number was taken from Gao et al., 2012, followed by antibacterial pIC₅₀ and fit values.

Fig. 13 The correlation graph of fit value with antibacterial PDF activity from experimental validation set.

Database screening of selected structure-based pharmacophore model

The constructed pharmacophore model was screened in the Druglike Diverse database of the Accelrys Discovery Studio (Accelrys, San Diego, USA)²⁷ which constituted 5384 diverse molecules with pre-generated conformers. Here, we relied on fit value to consider molecules as hits. This database screening provided 33 hits which were subsequently sorted by fit value. We selected the top five hit molecules for further study (Table 7 and Fig. 14). The top hit, CD1691700 secured a fit value of 3.55. The 2D structure and the IUPAC name of the top hits are given in Table 7 and its mapped pharmacophore features are graphically shown in ESI Table 6.† These five pharmacophore hits were subsequently docked with PfPDF–Co²⁺–actinonin complex in which the actinonin binding region was specified as the docking cavity with a radius of 10 Å, followed by the enumeration of 2D PfPDF–Co²⁺–hit interaction maps.

Table 7 Small molecular hits obtained from selected structure-based pharmacophore model in the Druglike Diverse database of the Accelrys Discovery Studio

S. no.	Structure-based pharmacophore hits	2D structure	IUPAC name
1	CD1691700		Methyl-3-{11-methyl-10-oxa-3,6,8-triazatricyclo[7.3.0.0^2,6]dodeca-1(9),2,7,11-tetraene-12-amido}thiophene-2-carboxylate
2	CAP05909402		2-{[2-(2,5-Dimethoxyphenyl)ethyl]sulfamoyl}ethan-1-aminium
3	VIT1024538		N-(5-Chloro-2,4-dimethoxyphenyl)-2-(4-ethyl-6-oxo-2-phenyl-1,6-dihydropyrimidin-1-yl)acetamide
4	CAP01891052		4-Amino-6-methyl-3-{[2-oxo-2-(1,2,3,4-tetrahydroquinolin-1-yl)ethyl]sulfanyl}-4,5-dihydro-1,2,4-triazin-5-one
5	IBS297042		(1S,4S)-4-Carbamoyl-1-({[(3R)-4-chloro-2-(methoxycarbonyl)-3H-indol-3-yl]carbamoyl}methyl)piperidin-1-ium

---

Fig. 14 The pharmacophore fit of hit molecules aligned with spatial features.

Interactions of PfPDF–Co²⁺–hits based on pharmacophore fitting and docking approaches

It can be seen that molecular docking may provide flexibility in interacting with the PfPDF substrate binding region and adopt an energetically favorable conformation (Fig. 15) which was restricted by pharmacophore fitting towards alignment over the 3D projected features in the PfPDF active site (Table 8). Clearly shown by the pharmacophore fit- and docking-interaction 2D maps (Fig. 16), hits which adopt similar conformation in both methods may be considered as probable lead molecules. The ability to develop interactions with the Co²⁺ metal cofactor is the principle requisite feature^6,68 of hit molecules to prioritize them as PfPDF leads. Remarkably, the correlation between fit value and docking score (expressed in kJ mol⁻¹) among selected hits was noted as 0.61 indicating a 61% chances= for obtaining a probable hit when structure-based pharmacophore and docking methods are combined together (Fig. 17). This correlation was attributed to the force field-based optimized dock poses from its initial coordinates of LeadIT dock pose, hereby inferring the reliability of the dock and post-dock optimization techniques used in this study shown in ESI Fig. 2.†

Fig. 15 The dock view of hit molecules in the active site of PfPDF–Co²⁺ complex.

Table 8 Docking and pharmacophore fitting of the top five hit molecules

S. no.	Hit	Interaction score^a	Match^a	Lipo^a	Ambig^a	Clash^a	Rot^a	Match^b	Fit value^c
a Values are expressed in kJ mol^−1.b The number of matched groups in initial and dock poses.c Computed by pharmacophore fitting.
1	CD1691700	−24.4182	−17.3118	−9.9883	−9.3456	5.4276	1.4	10	3.551
2	CAP05909402	−17.2093	−26.6925	−6.0531	−6.0472	3.5836	12.6	12	2.191
3	VIT1024538	−13.6716	−15.9951	−9.8102	−7.6142	7.3479	7	13	1.826
4	CAP01891052	−18.7194	−21.1491	−5.6281	−5.1314	2.1892	5.6	7	1.807
5	IBS297042	−19.2692	−20.8638	−7.6546	−7.2964	4.1456	7	7	1.717

---

Fig. 16 The 2D interaction maps of PfPDF–Co²⁺–hit molecules in both pharmacophore fitting and dock pose. The heatmap of residue-wise interactions is shown at the bottom.

Fig. 17 The correlation graph of fit value and docking score among selected top hits.

The selected top hit molecules were able to contact the metal ion Co²⁺ in pharmacophore fit, which qualified them for further residue-wise interaction analysis. The entire hit molecules were able to establish H bonding with Gly155 and Leu157 amino acids in both pharmacophore fitting and dock poses.

Notably, the H bonds developed by all hits with Gly107 were abolished in the dock pose in contradiction to its presence in the pharmacophore fit. Only IBS297042 retained an H bond with Glu199 in both poses. A combination of electrostatic, hydrophobic and pi contacts were observed with the metal-binding triad (Cys156, His198 and His202). Schematic residue-wise interactions made with amino acids that participated in the PfPDF catalytic cycle and deciphered through actinonin perceived pharmacophore features are shown in Fig. 16.

The furan oxygen of CD1691700 developed a conserved H bond with Leu157 in both poses. In addition, the oxygen atom from methylacetate established a conserved H bond with Gly155. However, the tendency to lose Co²⁺ contact in dock poses raises flexibility issues with CD1691700 to promote as a likely hit. CAP05909402 containing 1,4-dimethoxybenzene has less chemical complexity relative to other hits and adopted varied conformations in both results. A 90° flip can be observed in the pose of VIT1024538 which eliminated metal contact in dock results. CAP01891052 acquired nearly similar poses with small variations in the H bonding residues. The tendency to develop penta-coordination geometry with the metal center enforces CAP01891052 as a promising hit. Captivatingly, IBS297042 adopted a similar arrangement of metal interaction observed in the hydroxamate moiety–Fe²⁺ contact pattern of PfPDF–Fe²⁺–actinonin complex. Moreover, the formation of at least 4 H bonds with key residues of PfPDF suggests IBS297042 as the most likely lead molecule of PfPDF which has similar flexible functional groups as the actinonin molecule.

According to ADME prediction, the selected virtual screening hits were predicted to be well absorbed as HIA ranged between 70 and 100% (Table 9). The Caco-2 prediction revealed only CAP05909402 had probable low permeability. The MDCK values of the entire hits, suggested the low permeability in the MDCK cell permeability model. Overall, all hit molecules show moderate permeability except CAP05909402. Studies on the comparison of Caco-2 and MDCK permeability models showed that compounds possessing both high and moderate permeability predictions in these two models were assigned moderate permeable potential.⁷⁶ Hence, our prioritized hits would have moderate permeability which can be optimized by functional group modifications as suitable. The weak predictions on binding of plasma protein of hit molecules indicated its effectiveness in distribution. Furthermore, all the hit compounds were found to be CNS (Central Nervous System) inactive from the predicted BBB profile.

Table 9 ADME predictive properties of the top five selected hits

Hit	HIA^a (%)	Caco-2^b (nm s^−1)	MDCK^c (nm s^−1)	PPB^d (%)	BBB^e (C.brain/C.blood)
a Well absorbed compounds 70–100%, moderate 20–70% and weakly absorbed <20%.b Highly permeability >70 nm s^−1, middle 4–70 nm s^−1 and low permeability <4 nm s^−1.c High permeability >500 nm s^−1, middle 20–500 nm s^−1 and low permeability <25 nm s^−1.d Strongly bound chemicals >90% and weakly bound chemicals <90%.e CNS active compounds BBB >1 and inactive compounds BBB <1.
CD1691700	91.234306	22.9862	5.9182	46.729532	0.0236638
CAP05909402	87.536654	0.506337	0.590276	25.020728	0.0352205
VIT1024538	96.460491	36.4471	0.0585319	93.373546	0.338985
CAP01891052	95.045333	16.3112	4.17093	62.484171	0.15002
IBS297042	79.933868	8.76739	0.393892	13.556196	0.0479957

---

Finally, we schematically outline the possible mechanism of PfPDF inactivation (Fig. 18) based on the proposed oxygen-mediated inactivation mechanism of EcPDF by Rajagopalan and Pei, 1998.¹⁴ The inactivation mechanism is chiefly facilitated by the metal center triad. It was assumed that native PfPDF constitutes Fe²⁺ cofactor bound to water molecule (W₁ as per proposed catalytic cycle) or a hydroxide ion (1) and subsequently, Fe²⁺ may interact with molecular oxygen in the absence of substrate (2). This arrangement is also noticed in various iron metalloenzymes that reduce O₂. A ferric PfPDF intermediate will be formed by one-electron reduction on bound molecular O₂ so as to ligate superoxide ion to Fe³⁺ center (3). This superoxide ion may leave the enzyme complex to produce PfPDF–ferric deformylase complex to transform into its inactivated state (4). Alternatively, the bound superoxide ion may oxidize the adjacent Cys156 (triad) amino acid into cysteic acid (cysteine sulfonic acid; Cys-SO₃H) (5). Structure notes on Staphylococcus aureus PDF (SaPDF) showed that its metal coordinating residue Cys111 was oxidized to sulfinic acid (Cys-SO₂H).¹⁶ The exact molecular mechanism of both oxidized cysteine PDF enzyme forms are not yet clear. On the other hand, the diffused superoxide ion may disproportionate into H₂O₂ and O₂; the former species may inactivate another PfPDF enzyme via Fenton reaction and simultaneously, generate hydroxyl radical. The resulting radical may prompt cysteine oxidation of another PfPDF enzyme molecule.¹⁴ Intensive efforts are actively taken to solve this mechanistic puzzle of PDF inactivation and improvement of PDF expression protocol to isolate PDF in activated form suitable for screening small molecular and peptidomimetic inhibitors.

Fig. 18 Proposed inactivation mechanism of PfPDF using the knowledge of EcPDF oxygen mediated inactivation mechanism.

Conclusions

Structure-based pharmacophore screening of PfPDF inhibitors was carried out using the actinonin molecule and prioritized five small molecular hits. Based on the knowledge and proposal of EcPDF catalytic and inactivation mechanisms, we schematically presented the probable mechanisms of PfPDF and recognized the importance of pocket lining residues involved in the substrate binding and processing, and the amino acids which facilitated actinonin tight binding with the PfPDF active site with inhibitory potency in micromolar concentration. Further, we investigated the effect of various metal ions Co²⁺, Zn²⁺, Ni²⁺ and Fe²⁺ as PfPDF cofactors and analyzed their metal coordination geometry. We were successful in obtaining an energetically favorable pose of actinonin in the PfPDF–Co²⁺ complex compared to other selected metal ions which reinforces that the Co²⁺ will be the experimentally favored cofactor supported by native PfPDF–Co²⁺ complex to acquire catalytic efficient PfPDF for small molecular discovery even though there are biochemical reports on Fe²⁺ as physiological cofactor. We also showed that the effect of selected metal ions and its tetra- or penta-coordination geometry did not affect the reproduction of actinonin binding using FlexX docking algorithm evident from RMSD values, which suggests that docking experiments can identify probable PfPDF binders. Using the crystallographic bound (bioactive) conformation of actinonin developed through protein structural comparisons, the interaction pattern of the PfPDF–Co²⁺–actinonin complex was derived to construct a structure-based pharmacophore model. The flexible groups of actinonin, its complex six pharmacophore feature combinations and inclusion of excluded volumes offer a geometrical stringent condition for hit identification. Similar to our EcPDF structure-based modeling, we considered hydroxamic acid derivatives (actinonin analogues) having antibacterial activity to validate the structure-based pharmacophore model due to the unavailability of PfPDF inhibitors. Model statistics showed that the constructed pharmacophore model can retrieve actives with 36% confidence whereas the feature combinations may prevent false identification of molecules as actives with 96% surety, with 83% model accuracy. In concurrence with the development of tetra- or penta-coordination geometry of the Co²⁺ center, the arrangement of functional groups of the ligand to the chelate metal as well as the interaction with actinonin binding key residues and by ADME predictions, CAP01891052 and IBS297042 are concluded as the most promising PfPDF hit molecules. We believe that our presented molecular modeling strategy and applied docking and pharmacophore screens will enrich the knowledge of PfPDF inhibitors and its discovery to promote PfPDF as a most promising drug target for antimalarial research.

Conflict of interest

The authors declare that they have no conflict of interests in publishing this research article.

Acknowledgements

Anu Manhas thanks University Grants Commission (UGC) for Institutional Fellowship. Dr Sivakumar Prasanth Kumar acknowledges support from Department of Biotechnology (DBT), Govt. of India as Post-Doctoral Fellowship. Dr Prakash Chandra Jha would like to thank UGC for providing start up grants and Central University of Gujarat for providing basic computational facilities.

Notes and references

1. C. Giglione, M. Pierre and T. Meinnel, Mol. Microbiol., 2000, 36(6), 1197–1205.
2. World Health Organization, Deaths by cause, sex and mortality stratum in WHO regions, estimates for 2001, World Health Report, Geneva, 2002.
3. C. M. Apfel, H. Locher, S. Evers, B. Takacs, C. Hubschwerlen, W. Pirson, M. G. P. Page and W. Keck, Antimicrob. Agents Chemother., 2001, 45(4), 1058–1064.
4. D. Mazel, S. Pochet and P. Marliere, EMBO J., 1994, 13(4), 914–923.
5. C. Giglione, A. Serero, M. Pierre, B. Boisson and T. Meinnel, EMBO J., 2000, 19(21), 5916–5929.
6. J. N. Sangshetti, F. A. K. Khan and D. B. Shinde, Curr. Med. Chem., 2015, 22(2), 214–236.
7. V. Bracchi-Ricard, K. T. Nguyen, Y. Zhou and D. Pei, Arch. Biochem. Biophys., 2001, 396(2), 162–170.
8. A. Kumar, K. T. Nguyen, S. Srivathsan, B. Ornstein, S. Turley, I. Hirsh, D. Pei and W. G. J. Hol, Structure, 2002, 10(3), 357–367.
9. M. D. Lee, C. Antczak, Y. Li, F. M. Sirotnak, W. G. Bornmann and D. A. Scheinberg, Biochem. Biophys. Res. Commun., 2003, 312(2), 306–315.
10. S. Escobar-Alvarez, Y. Goldgur, G. Yang, O. Ouerfelli, Y. Li and D. A. Scheinberg, J. Mol. Biol., 2009, 387(5), 1211–1228.
11. K. T. Nguyen, X. Hu, C. Colton, R. Chakrabarti, M. X. Zhu and D. Pei, Biochemistry, 2003, 42(33), 9952–9958.
12. J.-P. Guilloteau, M. Mathieu, C. Giglione, V. Blanc, A. Dupuy, M. Chevrier, P. Gil, A. Famechon, T. Meinnel and V. Mikol, J. Mol. Biol., 2002, 320(5), 951–962.
13. D. Groche, A. Becker, I. Schlichting, W. Kabsch, S. Schultz and A. F. V. Wagner, Biochem. Biophys. Res. Commun., 1998, 246(2), 342–346.
14. P. T. R. Rajagopalan and D. Pei, J. Biol. Chem., 1998, 273(35), 22305–22310.
15. C. J. Hackbarth, D. Z. Chen, J. G. Lewis, K. Clark, J. B. Mangold, J. A. Cramer, P. S. Margolis, W. Wang, J. Koehn, C. Wu, S. Lopez, G. Withers, H. Gu, E. Dunn, R. Kulathila, S. H. Pan, W. L. Porter, J. Jacobs, J. Trias, D. V. Patel, B. Weidmann, R. J. White and Z. Yuan, Antimicrob. Agents Chemother., 2002, 46(9), 2752–2764.
16. H.-J. Yoon, H. L. Kim, S.-K. Lee, H.-W. Kim, H. W. Kim, J. Y. Lee, B. Mikami and S. W. Suh, Proteins, 2004, 57(3), 639–642.
17. J. Wiesner, S. Sanderbrand, B. Altincicek, E. Beck and H. Jomaa, Trends Parasitol., 2001, 17(1), 07.
18. J. Wiesner and F. Seeber, Expert Opin. Ther. Targets, 2005, 9(1), 23–44.
19. A. Merzoug, A. Chikhi, A. Bensegueni, K. S. Hioual, H. Boucherit, E. H. Mokrani and S. Teniou, Trends Bioinf., 2013, 6(1), 17–24.
20. S. P. Kumar, P. C. Jha, Y. T. Jasrai and H. A. Pandya, J. Biomol. Struct. Dyn., 2016, 34(3), 540–559.
21. S. P. Kumar, H. A. Pandya, Y. T. Jasrai and R. M. Rawal, J. Recept. Signal Transduction, 2016, 36(2), 189–206.
22. S. P. Kumar, L. B. George, Y. T. Jasrai and H. A. Pandya, SAR QSAR Environ. Res., 2015, 26(1), 67–77.
23. M. A. Robien, K. T. Nguyen, A. Kumar, I. Hirsh, S. Turley, D. Pei and W. G. J. Hol, Protein Sci., 2004, 13(4), 1155–1163.
24. F. C. Bernstein, T. F. Koetzle and G. J. B. Williams, J. Mol. Biol., 1977, 112(3), 535–542.
25. S. F. Altschul, W. Gish, W. Miller, E. W. Myers and D. J. Lipman, J. Mol. Biol., 1990, 215(3), 403–410.
26. A. D. MacKerell Jr, D. Bashford, M. Bellott, R. L. Dunbrack Jr, J. D. Evanseck, M. J. Field, S. Fischer, J. Gao, H. Guo, S. Ha, D. Joseph-McCarthy, L. Kuchnir, K. Kuczera, F. T. K. Lau, C. Mattos, S. Michnick, T. Ngo, D. T. Nguyen, B. Prodhom, W. E. Reiher, B. Roux, M. Schlenkrich, J. C. Smith, R. Stote, J. Straub, M. Watanabe, J. Wiorkiewicz-Kuczera, D. Yin and M. Karplus, J. Phys. Chem. B, 1998, 102(18), 3586–3616.
27. Accelrys Discovery Studio Version 4.0, Accelrys, San Diego, USA, http://www.accelrys.com/products/collaborative-science/biovia-discovery-studio/.
28. E. C. Meng, E. F. Pettersen, G. S. Couch, C. C. Huang and T. E. Ferrin, BMC Bioinf., 2006, 7, 339.
29. E. F. Pettersen, T. D. Goddard, C. C. Huang, G. S. Couch, D. M. Greenblatt, E. C. Meng and T. E. Ferrin, J. Comput. Chem., 2004, 25(13), 1605–1612.
30. S. B. Needleman and C. D. Wunsch, J. Mol. Biol., 1970, 48(3), 443–453.
31. S. Henikoff and J. G. Henikoff, Proc. Natl. Acad. Sci. U. S. A., 1992, 89(22), 10915–10919.
32. A. Becker, I. Schlichting, W. Kabsch, S. Schultz and A. F. Wagner, J. Biol. Chem., 1998, 273, 11413–11416.
33. K. H. Kim, W. K. Lee, K. J. Choi and E. E. Kim, J. Korean Soc. Appl. Biol. Chem., 2014, 57(5), 565–571.
34. J. M. Clements, R. P. Beckett, A. Brown, G. Catlin, M. Lobell, S. Palan, W. Thomas, M. Whittaker, S. Wood, S. Salama, P. J. Baker, H. F. Rodgers, V. Barynin, D. W. Rice and M. G. Hunter, Antimicrob. Agents Chemother., 2001, 45(2), 563–570.
35. D. Eramian, N. Eswar, M. Y. Shen and A. Sali, Protein Sci., 2008, 17(11), 1881–1893.
36. M. Rarey, B. Kramer, T. Lengauer and G. Klebe, J. Mol. Biol., 1996, 261(3), 470–489.
37. B. Kramer, M. Rarey and T. Lengauer, Proteins Suppl., 1997, 29(S1), 221–225.
38. LeadIT 2.1.8, BioSolveIT GmbH, St. Augustin, Germany, http://www.biosolveit.de.
39. ChemAxon LLC, Marvin Sketch 5.11.4, 2012, http://www.chemaxon.com.
40. TRIPOS Associates Inc, SYBYL molecular modeling software 6.x, St. Louis, Missouri, USA, 1994.
41. H. J. Böhm, J. Comput.-Aided Mol. Des., 1994, 8(3), 243–256.
42. K. Stierand, P. C. Maab and M. Rarey, Bioinformatics, 2006, 22(14), 1710–1716.
43. E. Krieger, K. Joo, J. Lee, J. Lee, S. Raman, J. Thompson, M. Tyka, D. Baker and K. Karplus, Proteins, 2009, 77(suppl 9), 114–122.
44. D. Rogers and A. J. Hopfinger, J. Chem. Inf. Comput. Sci., 1994, 34(4), 854–866.
45. A. Garcıá-Garcıá, J. Gálvez, J. V. de Julián-Ortiz, R. Garcıá-Domenech, C. Muńoz, R. Guna and R. Borras, J. Antimicrob. Chemother., 2004, 53(1), 65–73.
46. M. Arooj, S. Sakkiah, S. Kim, V. Arulalapperumal and K. W. Lee, PLoS One, 2013, 8(4), e63030.
47. J. Gao, Y. Cheng, W. Cui, Q. Chen, F. Zhang, Y. Du and M. Ji, Med. Chem. Res., 2012, 21(8), 1597–1610.
48. R. Jain, A. Sundram, S. Lopez, G. Neckermann, C. Wu, C. Hackbarth, D. Chen, W. Wang, N. S. Ryder, B. Weidmann, D. Patel, J. Trias, R. White and Z. Yuan, Bioorg. Med. Chem. Lett., 2003, 13(23), 4223–4228.
49. ChEMBL database, European Bioinformatics Institute, http://www.ebi.ac.uk/chembl/.
50. M. O. Taha, Mixing pharmacophore modeling and classical QSAR analysis as powerful tool for lead discovery, virtual screening, in Virtual screening, ed. M. Taha, InTech, Shanghai, China, 2012.
51. A. Smellie, S. L. Teig and P. Towbin, J. Comput. Chem., 1995, 16(2), 171–187.
52. SPSS, Version 18.0, SPSS Inc, Chicago, 2007, http://www-01.ibm.com/software/analytics/spss/.
53. PreADMET program, Yonsei University, Seoul, Republic of Korea. http://preadmet.bmdrc.org.
54. K. T. Nguyen, J. C. Wu, J. A. Boylan, F. C. Gherardini and D. Pei, Arch. Biochem. Biophys., 2007, 468(2), 217–225.
55. T. Meinnel, S. Blanquet and F. Dardel, J. Mol. Biol., 1996, 262(3), 375–386.
56. P. T. R. Rajagopalan, S. Grimme and D. Pei, Biochemistry, 2000, 39(4), 779–790.
57. S. Ragusa, S. Blanquet and T. Meinnel, J. Mol. Biol., 1998, 280(3), 515–523.
58. W. Wriggers and P. Chacón, Structure, 2001, 9(9), 779–788.
59. A. Becker, I. Schlichting, W. Kabsch, D. Groche, S. Schultz and A. F. Wagner, Nat. Struct. Biol., 1998, 5(12), 1053–1058.
60. R. Jain, B. Hao, R. P. Liu and M. K. Chan, J. Am. Chem. Soc., 2005, 127(13), 4558–4559.
61. J. J. Irwin, F. M. Raushel and B. K. Shoichet, Biochemistry, 2005, 44(37), 12316–12328.
62. B. K. Shoichet, D. L. Bodian and I. D. Kuntz, J. Comput. Chem., 1992, 13(3), 380–397.
63. R. Abagyan and M. Totrov, Curr. Opin. Chem. Biol., 2001, 5(4), 375–382.
64. S. Ha, R. Andreani, A. Robbins and I. Muegge, J. Comput.-Aided Mol. Des., 2000, 14(5), 435–448.
65. K. J. Smith, C. M. Petti, K. Aubart, M. Symth, E. McManus, J. Jones, A. Fosberry, C. Lewis, M. Lonetto and S. B. Christensen, Protein Sci., 2003, 12(2), 349–360.
66. A. Kreusch, G. Spraggon, C. C. Lee, H. Klock, D. McMullan, K. Ng, T. Shin, J. Vincent, I. Warner, C. Ericson and S. A. Lesley, J. Mol. Biol., 2003, 330(2), 309–321.
67. Y. J. Hu, Y. Wei, Y. Zhou, P. T. Rajagopalan and D. Pei, Biochemistry, 1999, 38(2), 643–650.
68. Z. Yuan, J. Trias and R. J. White, Drug Discovery Today, 2001, 6(18), 954–961.
69. O. F. Guner, Pharmacophore perception, development, and use in drug design, International University Line, La Jolla, CA, 2000.
70. M. Rella, C. A. Rushworth, J. L. Guy, A. J. Turner, T. Langer and R. M. Jackson, J. Chem. Inf. Model., 2006, 46(2), 708–716.
71. D. Pandit, S. S. So and H. Sun, J. Chem. Inf. Model., 2006, 46(3), 1236–1244.
72. Molecular Operating Environment (MOE), Chemical Computing Group Inc, Montreal, Canada, http://www.chemcomp.com/software.htm.
73. D. M. W. Powers, J. Mach. Learn. Tech., 2011, 2(1), 37–63.
74. K. Boppana, P. K. Dubey, S. A. R. P. Jagarlapudia, S. Vadivelan and G. Rambabu, Eur. J. Med. Chem., 2009, 44(9), 3584–3590.
75. S. Sakkiah, C. Meganathan, Y.-S. Sohn, S. Namadevan and K. W. Lee, Int. J. Mol. Sci., 2012, 13(4), 5138–5162.
76. K. C. Sekhar, R. Syed, M. Golla, M. V. J. Kumar, N. K. Yellapu, A. R. Chippada and N. R. Chamarthi, Daru, 2014, 22(1), 76.

---

Footnote

† Electronic supplementary information (ESI) available: Geometry of metal cofactor and its residues and pharmacophore features of the model, energy minimized data of PfPDF–ligand complexes, list of validation and hits molecules, comparative view of crystal and docked conformation. See DOI: 10.1039/c6ra01071g

---