Gene 581 (2016) 31–42

Contents lists available at ScienceDirect

Gene

j ourna l homepage: www.e lsev ie r .com/ locate /gene

Research paper

Molecular principles behind pyrazinamide resistance due tomutations inpanD gene in Mycobacterium tuberculosis

Bharati Pandey a, Sonam Grover b, Chetna Tyagi c, Sukriti Goyal d, Salma Jamal d, Aditi Singh e,Jagdeep Kaur a, Abhinav Grover c,⁎a Department of Biotechnology, Panjab University, Chandigarh 160014, Indiab Kusuma School of Biological Sciences, Indian Institute of Technology Delhi, New Delhi 110016, Indiac School of Biotechnology, Jawaharlal Nehru University, New Delhi 110067, Indiad Department of Bioscience and Biotechnology, Banasthali University, Tonk 304022, Rajasthan, Indiae Department of Biotechnology, TERI University, Vasant Kunj, New Delhi 110070, India

Abbreviations: MD, molecular dynamics; DM, doubsquare deviation; RMSF, root mean square fluctuation.⁎ Corresponding author.

E-mail addresses: abhinavgr@gmail.com, agrover@jnu

http://dx.doi.org/10.1016/j.gene.2016.01.0240378-1119/© 2016 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 24 November 2015Received in revised form 5 January 2016Accepted 14 January 2016Available online 16 January 2016

The latest resurrection of drug resistance poses serious threat to the treatment and control of the disease. Muta-tions have been detected in panD gene in theMycobacterium tuberculosis (Mtb) strains. Mutation of histidine toarginine at residue 21 (H21R) and isoleucine to valine at residue 29 (I49V) in the non-active site of panD gene hasled to PZA resistance. This study will help in reconnoitering the mechanism of pyrazinamide (PZA) resistancecaused due to double mutation identified in the panD gene of M. tuberculosis clinical isolates. It is known thatpanD gene encodes aspartate decarboxylase essential forβ-alanine synthesis thatmakes it a potential therapeuticdrug target for tuberculosis treatment. The knowledge about the molecular mechanism conferring drug resis-tance in M. tuberculosis is scarce, which is a significant challenge in designing successful therapeutic drug. Inthis study, structural and dynamic repercussions of H21R–I49V double mutation in panD complexed with PZAhave been corroborated through docking and molecular dynamics based simulation. The double mutant (DM)shows lowdocking score and thus, lowbinding affinity for PZAas compared to the native protein. It was observedthat themutant protein exhibitsmore structuralfluctuation at the ligand binding site in comparison to the nativetype. Furthermore, the flexibility and compactness analyses indicate that the double mutation influenceinteraction of PZA with the protein. The hydrogen-bond interaction patterns further supported our results. Thecovariance and PCA analysis elucidated that the double mutation affects the collective motion of residues inphase space. The results have been presented with an explanation for the induced drug resistance conferredby the H21R–I49V double mutation in panD gene and gain valuable insight to facilitate the advent of efficienttherapeutics for combating resistance against PZA.

© 2016 Elsevier B.V. All rights reserved.

Keywords:DockingSimulationPyrazinamideRMSDRMSF

1. Introduction

Tuberculosis (TB), a deadly, airborne disease caused by the Gram-positive bacterium Mycobacterium tuberculosis (Mtb), is the secondmost deadly disease worldwide after HIV/AIDS. It also leads to highestrate of mortality in HIV positive patients being responsible for onefourth of all HIV-related deaths (Organization, 2004). Numerous globalstrategies, targets, effective diagnosis and treatment procedures forcontrolling tuberculosis have saved at least 37 million lives from 2000to 2013. But still, in 2013, 9 million people were affected from TB with1.5 million deaths recorded. Another major obstruction in effectivetreatment and eradication of TB was identified and was called as the

le mutant; RMSD, root mean

.ac.in (A. Grover).

multidrug-resistant TB (MDR-TB). An estimated 480,000 new cases ofMDR-TB that does not respond to most powerful anti-TB drugs suchas isoniazid, pyrazinamide and rifampicin, were reported (Field, 2015).

First line TB-drug for both drug susceptible and multidrug-resistanttuberculosis (MDR-TB) is Pyrazinamide (PZA) (Chiu et al., 2011). It isalso considered to be a potent companion for newer regimens that areunder development and shows a remarkable role in shortening theperiod of TB treatment. PZA is an analog of nicotinamide activated bythe action of nicotinamidase/pyrazinamidase (PZase) encoded by pncAgene and is converted to pyrazinoic acid (POA) intracellularly byhydrolyzation which inhibits the vital fatty acid synthase (Ahmad andMokaddas, 2009). The crucial role of PZA accentuates the prerequisitefor precise and prompt recognition of PZA resistance; however, withcurrent phenotypic testing it is very difficult.

L-aspartate α-decarboxylase (ADC, EC4.1.1.15), encoded by thepanD gene, is a regulatory enzyme involved in the pantothenate biosyn-thetic pathway and is an emergingpotential antibacterial and antifungal

Fig. 1. Ribbon representation showing superimposition of the native (shown in orange)and DM structure obtained from the 50 ns molecular dynamic simulations (shown inblue).

32 B. Pandey et al. / Gene 581 (2016) 31–42

target (Albert et al., 1998). It is found in plants, fungi and microorgan-isms, including Mycobacterium, but not in animals. It catalyzes theformation of β-alanine from L-aspartate in bacteria (Sharma et al.,2012). Pantothenate (vitamin B5) is synthesized from β-alanine,which is the precursor of 4′-phosphopantetheine and coenzyme A(CoA) which plays an essential role in themetabolism and biosynthesisof fatty acids, tricarboxylic acid cycle, and biosynthesis of polypeptides(Gopalan et al., 2006). Mutation in panD causes PZA resistance butwhether this interferes with functioning of pantothenate and CoAremains unclear. In vitro and clinical studies demonstrate that panD isa new target of action involved in mediating resistance to PZA (Shiet al., 2014). The presence of lipid rich cell wall in Mtbmakes it virulentand enhances antimicrobial resistance and tolerance to environmentalstress conditions (Daffé and Draper, 1997; Cox et al., 1999). The intra-cellular persistence of M. tuberculosis is dependent upon lipids andlipoglycans biosynthesis and metabolism. Thus, any alteration in thispathway poses substantial risk to the life cycle and survival of bacteria.The absence of panD gene in human and its significant role in cellularmetabolism of Mtb make it a potential drug target candidate. The ADCcoding gene is present inM. tuberculosis genome in single copy, anotherreason for it being a suitable drug target. The refined protein structureconsists of 113 amino acid residues. The remaining residues from 114to 139 are not detected in electron density map and are probablydisordered. The structure of the PanD protein from M. tuberculosis wassolved using X-ray crystallography and deposited in the PDB, at 2.99 Å

Table 1Docking results of PanD protein with PZA and POA using PatchDock and FireDock using a 4.0 Å

Protein

PatchDock and FireDock

Total ligand–receptor interactionenergy (kcal/mol)

Ligand–receptor van der Waalsenergy (kcal/mol)

A

Native-PZA –27.37 –11.61 –Native-POA –24.35 –9.99 –DM-PZA –20.75 –8.98 –DM-POA –19.38 –8.63 –

resolution (Gopalan et al., 2006). Although, the structural properties ofpanD are elaborately known, a mechanistic overview correlating themutational changes to explicit dynamic properties of the protein hasnot been discussed in detail. Strictly conserved residues involved instructure stability and activity in ADCs include Lys9, His11, Thr16,Tyr22, Gly24, Ser25, Asp29, Asn51, Arg54, Thr57, Tyr58, Gly72, Ala73,Ala74, Asp82, Ile85 and Asn111. The 3D structure is predominantly adouble-ψ β-barrel consisting of seven β-strands, two α-helices, andtwo 310 helices (Lee and Suh, 2004). Themutations that lead to reducedactivity and are associatedwith PZA resistance in pncA ofM. tuberculosishave been extensively studied (Rajendran and Sethumadhavan, 2014).Recent report demonstrated resistance to pyrazinamide being causedby single and double mutants (H21R–I49V) of panD gene (Zhanget al., 2013).

Understanding of the protein structural dynamics is important toelucidate the molecular mechanism involved in essential biologicalprocesses. Computational methods are emerging as a powerful toolto understand the underlying mechanism behind drug resistanceand target inhibition thereby expediting the process of drugdiscovery.

The aim of this study is to explicate the structural and dynamic con-sequences of the behavior of wild and mutant aspartate decarboxylaseforms and to comprehend the resistant nature of themutant protein to-wards PZA. However, there is a lack of detailed studies on this phenom-enon due to unavailability of X-ray crystal structure of the doublemutant (DM). To achieve this goal, we carried out 50 ns long simula-tions for native and DM protein structures before and after bindingwith PZA (in water solvent) to understand the dynamic behavior ofthe protein at the atomic level with time evolution.

2. Material and methods

2.1. Data set

The crystal structure of the native ADC protein (PDB code: 2C45)was retrieved from RCSB Protein Data Bank (PDB) for our study(Gopalan et al., 2006). The PDB structure contained 113 amino acidresidues along with water. In order to carry out further analysis thewater molecules were removed using Accelrys Viewerlite 5.0. Crystalstructure of themutant is not available and thuswas generated by intro-ducing a point mutation in the crystal structure through Schrödinger'sprotein preparation wizard (Sastry et al., 2013; Schrödinger, 2013).The double mutant (DM) was obtained by mutating His21 to Arg21and Ile49 to Val49 as known from Zhang et al., 2013. The 3D structureof drugs PZA and POA (CID 1046 and 1047) were downloaded fromPubChem database in .sdf format (Deursen et al., 2010). POA is formedby removal of amide group from PZA by the action of bacterial PZase.Thewild andmutant protein structureswere optimized through severalsteps of structural modifications and energy minimization. During thisprocess hydrogen atoms were added, bond lengths were improved, di-sulfide bonds were generated, capping of terminal residues were doneand selenomethionines were converted into methionines. Moleculardynamics (MD) simulationwas performed for a period of 50 ns to stabi-lize the native and DM structures to carry out further analysis.

RMSD cutoff.

Glide

CEDockingscore

Glide energy(kcal/mol)

van der Waals energy(kcal/mol)

Electrostatic energy(kcal/mol)

116.10 1578 –19.99 –13.42 –6.57129.67 1500 –20.52 –13.730 –5.2384.63 1504 –18.52 –12.78 –5.7275.32 1386 –19.24 –12.79 –4.45

Fig. 2. Illustration of H-bond interactions: A) native protein bound to PZA after docking, B) native protein boundwith PZA afterMD simulations, and C) docked doublemutant (DM) PanD-PZA complex before simulation.

33B. Pandey et al. / Gene 581 (2016) 31–42

2.2. Binding cavity prediction

Determination of the binding cavity is significantly important topredict protein function, mutation studies, and identification ofpotential drug and lead optimization. The binding pockets of the PanDvariants were compared using the Computed Atlas of Surface Topogra-phy of proteins (CASTp) server (Dundas et al., 2006). It providesdeterministic identification and computation of the solvent accessiblesurface geometry and inaccessible cavities in protein. The structuralpockets and cavities are computed in relation to area and volume bytwo approaches; solvent accessible surface model (Richards' surface)and molecular surface model (Connolly's surface). For measuringmolecular shapes, this server uses weighted Delaunay triangulationand the alpha complex framework. Calculations of the native and DMstructures were carried out with probe radius of 1.4 Å and using otherparameters with default values.

2.3. Protein–ligand docking study

Docking of all ADC variantswith PZAwas performedusing AutoDockVina 1.1.2 suite, a flexible automated docking program which employsiterated local search global optimization algorithm (Trott and Olson,2010). Implementation of new scoring function, efficient optimization,and multithreading in AutoDock Vina greatly improves the speed andaccuracy of docking. Grid maps and clusters are automaticallycomputed making it more user-friendly. The compatible pdbqt files ofproteins and ligand were prepared using AutoDock Tools 4.2.6.(Morris et al., 2009). Polar hydrogen was added and the binding free

Table 2RMSF value for the binding residues.

Protein Tyr22 Ser25 Val26 Thr57

Native (unbound) 0.2350 0.0898 0.0914 0.0726Native-PZA 0.2450 0.1927 0.1216 0.0883Native-POA 0.2645 0.1713 0.1008 0.0801Mutant (unbound) 0.3440 0.2424 0.1595 0.1142Mutant-PZA 0.2319 0.1480 0.1050 0.0955Mutant-POA 0.2471 0.1295 0.0777 0.0856

energies were calculated using scoring function of AutoDock Vinausing the Expression (1),

c ¼ Σib j f tit j rij� � ð1Þ

where the summation is over all the pairs of i and j atoms that canmoverelative to each other, i.e. atoms separated by 3 consecutive covalentbonds. Each atom is assigned a type ti, and a symmetric group ofinteraction functions ftitj of rij should be considered where rij is distancebetween atoms i and j. The calculated scoring function is the sum ofboth inter- and intra-molecular contribution as given by Expression(2) and will determine global minimum energy of c:

c ¼ cinter þ cintra: ð2Þ

The binding free energy is estimated from the intermolecular com-ponent of the lowest-scoring conformation as given in Expression (3):

s1 ¼ g c1−cintra1ð Þ ¼ g cinterð Þ: ð3Þ

Broyden–Fletcher–Goldfarb–Shanno (BFGS) algorithm (Peng et al.,1996) is used for the local optimization which is similar to quasi-Newton optimization approach. In this approach, proteins are treatedas rigid and ligands as flexible molecules with a number of rotatablebonds. A genetic algorithm that employs Lamarckian inheritance wasselected to find the minimum energy docked poses (Morris et al.,1998). 50 conformers were generated during docking for both nativeand DM variants. The population size was fixed to 150 and number of

Tyr58 Asn72 Asn73 Gly74 Ala75 Ala76

0.1106 0.1164 0.0947 0.1571 0.1805 0.11010.0929 0.1165 0.1609 0.1787 0.2136 0.18600.0836 0.1232 0.1719 0.2004 0.1434 0.09980.1761 0.1644 0.1633 0.2033 0.2443 0.19220.1647 0.1204 0.1081 0.1446 0.1195 0.11480.0855 0.0984 0.1508 0.2235 0.2946 0.2341

Fig. 3. Illustration of H-bond interactions: A) native protein bound to POA after docking, and B) native-POA complex after simulation.

34 B. Pandey et al. / Gene 581 (2016) 31–42

energy evaluation was set at 25,000,000, number of generations to1000, mutation rate of 0.02 and crossover rate of 0.8. Docking wascarried out with setting of grid size of 60 × 60 × 60 along the X, Y andZ-axes, with 0.375 Å spacing between the grid points. The grid wasbuilt around the protein active site which included residues Tyr22,Val26, Ser25, Thr57, Tyr58, Asn72, Asn73, Gly74, Ala75, and Ala76.Docked conformers were clustered using 2.0 Å RMS cluster tolerance.

To increase the docking accuracy, another docking was alsoperformed using geometry-based molecular docking algorithm,PatchDock (Schneidman-Duhovny et al., 2005). In order to find dockingtransformation that produce good molecular shape complementarywith high fitness, this method splits the Connolly dot surface represen-tation of protein molecules into three classes, namely, concave, convexand flat patches (Connolly, 1983b; Connolly, 1983a). Afterwards, the di-vided complementary patches between receptor and ligand arematched in order to produce candidate transformations. Evaluation ofeach candidate transformation is carried out by scoring function thatdetermines both shape complementarity and atomic desolvation ener-gy (Zhang et al., 1997). At last, a RMSD clustering was implemented toeach candidate transformation and redundant solutionswere discarded.

Fig. 4.Hydrophobic interaction pattern in A) docked native PZA complex before simulations, Bsimulations, and D) DM protein bound to PZA after MD simulations.

Highest scoring geometric fits were attained after the exhaustivetransformational search. Energyminimized protein and ligandmoleculewere formatted in PDB format and submitted with clustering rootmean-square deviations (RMSD) 4.0 Å. The PatchDock docking runwas further refined and rescored using FireDock server by allowingbackbone and side chain movement (Andrusier et al., 2007). Themolecular docking results were cross-checked with Glide module ofSchrödinger (Friesner et al., 2004).

In Glide, flexible docking program was used and extra precisiondocking between native and DM PanD variants with PZA and POA wasperformed. Total interaction energy of the protein–ligand complexeswas computed along with electrostatic energy, inter-molecular vander Waals energy and internal torsion energy. The electrostatic energydenotes the total energy of the system due to the mutual interactionof each pair of charges carried by the twomolecules. Ligand binding in-duces change in protein structure and conformational stability whichbrings variation in the thermal stability of the protein–ligand complex.The negative value of electrostatic energy suggests stable interactionbetween protein and ligand. The other non-covalent interactions suchas van der Waals interaction, H-bonding and hydrophobic interaction

) native protein bound with PZA after MD simulations, C) docked DM-PZA complex before

Fig. 5.Hydrophobic interaction pattern in A) docked native POA complex before simulations, B) native protein boundwith POA after MD simulations, C) docked DM-POA complex beforesimulations, and D) DM protein bound to POA after MD simulations.

35B. Pandey et al. / Gene 581 (2016) 31–42

play a major role in the thermodynamic stability of the ligand–proteincomplex.

2.4. Protein–ligand interaction analysis

Todeduce the interaction between the protein and bound ligand, thedocked complexes before and after simulation were used as input inprograms namely Collaborative Computational Project No. 4 (CCP46.5) graphical user interface package (http://www.ccp4.ac.uk/about.php).The interaction analysis was also performed by Ligplot+(https://www.ebi.ac.uk/thornton-srv/software/LigPlus/) and 2Ddiagrams of protein–ligand interactions were generated.

2.5. Molecular dynamics simulation

To obtain stable conformation of ligand-bound complex, all MD simu-lationswere performedwith GROMACS 5.0 software package (Berendsenet al., 1995; VanDer Spoel et al., 2005;Hess et al., 2008). GROMOS9643A1force field was used for the protein–ligand bound dynamics (vanGunsteren et al., 1996). PRODRGweb server was used to generate the to-pologies and coordinates of the ligands (SchuÈttelkopf and Van Aalten,2004). The structure of the protein–ligand complex was placed incubic water box with dimensions of 9.22614 × 9.22614 × 9.22614 usingsimple point charge water model (Wu et al., 2006). The system wasneutralized by adding sodium ions around the molecule. These ionswere incorporated by replacingwatermoleculeswith the highest electro-static potential. The whole system was energy-minimized for 50,000 cy-cles using steepest descent algorithm, terminating when maximumforce is less than 1000 kJ mol−1 nm−2. To further relax the system, theminimized system was subjected to an equilibration step by a 100 pslong ‘position restraint MD simulation’ with 1000 kJ mol−1 nm−2 con-stant force on the heavy atoms of protein and substrate under NVT condi-tions. Berendsen thermostat method was employed for temperaturecoupling (300 K, relaxation time 0.1 ps; a 100 ps pressure equilibrationMD run (NPT) with temperature set to 300 K and the pressure was kept

constant at 1 bar utilizing Parrinello–Rahman barostat (Parrinello andRahman, 1981). The linear constraint solver (LINCS) algorithm (Hesset al., 1997) was applied to fix all the bond lengths. Fast particle-meshEwald (PME) (Darden et al., 1993) electrostatics method was used tocompute long-range electrostatic interactions using a cut-off of 1.0 nm.The system was subjected to a 100 ps dynamics run with protein atomsrestrained and was followed by free production run for 50,000 ps at300 K and 1 bar pressure. Trajectories were analyzed by plotting RMSDfor each system frame using energy minimized structures as a reference.

2.6. Principal component analysis

Principal component analysis was used to obtain the overall motionof native andDMprotein residues. Calculation of eigenvector and eigen-values and their projection along the first two principal components byanalyzing MD simulation trajectories was carried out using essentialdynamics (ED) method. PCA or ED analysis (Mesentean et al., 2006)is used to reduce the dimensionality of variable describing complexdata and extract the strenuous motion in simulation. The trajectoriesfiles produced during the MD simulation were employed to describethe movement of the native and mutant PanD structures. Weused PCA to study the protein regions that are responsible for themost significant motion. The covariance matrix was estimated using asimple linear transformation in Cartesian coordinate space after theremoval of the rotational and translational movements using leastsquare fit procedure.

The elements of the positional covariancematrix were calculated bythe following Eq. (4):

Ci ¼ ðqi− qif g qj− qj

n o� �n oi; j ¼ 1;2;…3Nð Þ ð4Þ

in which qi was the Cartesian coordinate of the ith atom, and Nwas thenumber of atom in PanD protein.

Fig. 6.Graph showing the (A) RMSD, (B) Rg, (C) SASA and (D) RMSF of native unbound protein, native bound to PZA and POA, DM inunbound formandDM in complexwith PZA and POA.

36 B. Pandey et al. / Gene 581 (2016) 31–42

The symmetric matrix C was diagonalized through orthogonalcoordinate transformational matrix T, which transformed it into adiagonal matrix Λ of eigenvalues λi as given by Expression (5).

Λ ¼ TtCijT ð5Þ

where the columns were the eigenvectors describing direction of mo-tion relative to {qi}. Each eigenvector is associated with an eigenvaluerepresenting the contribution of each type of energy of each componentto the motion. The eigenvalues displayed the magnitude of the eigen-vectors along the multi-dimensional space, and the displacement of

atoms along each eigenvector represents the direction of movement ofprotein.

2.7. Image rendering and MD analysis

All protein structures and interactions were obtained using UCSFchimera v1.10.2 (Pettersen et al., 2004) and Pymol v1.7.6 (DeLano,2002). The trajectory files were analyzed by using g_rms, g_rmsf,g_sasa, g_covar and g_anaeig GROMACS services in order to analyzeRMSD, root mean square fluctuation (RMSF), solvent accessibility sur-face area (SASA) and PCA analysis. Images were also produced usingthe XMgrace program (Turner, 2005).

Fig. 7. Covariance matrix calculated form native and DM considering the backbone atom. (A) Covariance matrix for the native type and (B) covariance matrix for the DM type. Redcorresponds to a positive correlation shows the motion of the atom along the same direction and blue signifies the negative correlation indicates motion in opposite directions.

37B. Pandey et al. / Gene 581 (2016) 31–42

Fig. 8. Projection of the motion of the protein in phase space along the first two principal eigenvectors at 300 K. Native is shown in red and mutant is shown in black.

38 B. Pandey et al. / Gene 581 (2016) 31–42

39B. Pandey et al. / Gene 581 (2016) 31–42

3. Results and discussion

Microorganism becoming resistant to most of the drug due toemergence of multiple strain with various mutation and posing seriousconcern to the conventional cures and impose huge cost to individuallife and society. In order to capture the mechanism of drug resistancewe have tried to examine the structural and dynamics effect of doublemutation (H21R/I49V) in PanD. Various computational approacheshave been carried out in order to understand how the mutations inthe non-active site influence the affinity of protein–ligand complexand underlying resistance mechanism against a drug.

3.1. Generation of mutant protein

The 3D protein structure of the double mutant was not available inthe database. The DM was generated by replacing His21 by Arg21 andIle49 and Val49 in the PDB coordinate file. The native and DMwere en-ergyminimized to eliminate close contacts. To get all-atom level details,MD simulation for a period of 50 ns was instigated for the PanD nativeand DM variants. RMSD of each frame with respect to the first frameas a function of time during the 50 ns simulation were plotted. No sub-stantial deviation was noticed in the trajectories of the native and DMtill 30,000 ps. After ~35,000 ps, DM revealed striking deviation in thetrajectory till the termination of the simulation period resulting in theRMSD of ~0.29 to 0.45 nm. Stable trajectories were observed after25,000 ps in case of the native and after 35,000 ps in DM variant. Aver-age structurewas calculated from the coordinates of all the frames lyingwithin the most stable time frame. The average structure signifies theequilibrated state of the native and DM proteins. The magnitude offluctuation in the RMSD value implies that the simulation produceda stable trajectory, thus providing an appropriate groundwork forsubsequent analysis. Conformational differences at the atomicscale between the pre-MD simulated and post-simulated structureswere calculated by superimposing average molecular structures ofboth native and DM variants and yielded backbone RMSD value of0.07 Å (Fig. 1). RMSD value is less than 2.0 Å from the crystallo-graphic structure which is considered as suitable for further studies(Allen and Rizzo, 2014).

3.2. Binding cavity analysis

After the MD simulation, structure of the native and DM proteinswere subjected for binding pocket calculation using CASTp server. Thepredicted volumes of the binding pockets for native and DM proteinswere 230.3 Å3 and 56.51 Å3 respectively showing significant differenceswith each other. The appropriate binding pocket can accommodatea drug molecule properly and reduction in the volume of bindingcavity can hinder the ligand binding by weakening the protein–ligand interaction and thus leading to drug resistance (Lahtiet al., 2012). Significant decrease in the pocket size in the doublemutant was also confirmed using LIGSITEcsc web server (Huangand Schroeder, 2006).

3.3. Interaction analysis between the protein variants and drug molecule

The docking score value from PatchDock exemplifies geometric fit ofligand with the protein receptor. The atomic contact energy (ACE) anddocking score of simulated native and DM variants with PZA and POAwere computed and mentioned in Table 1. The obtained resultsconfirmed higher docking and ACE score for native in comparison tothe DM. Total energy of interaction between the native-PZA and POAcomplexes was ‐27.37 kcal/mol and ‐24.35 kcal/mol respectively. Thisshows good degree of complementarities between both native-PZAand POA complexes. Contrarily, the DM-PZA and POA complexes exhib-ited total energy of ‐20.75 kcal/mol and ‐19.38 kcal/mol and lesservan der Waals and electrostatic forces as compared to native ligand

complexes. A smaller binding pocket volume in the DM variant as pre-dicted by CASTp server confirms the same and leads to an obstructionin inhibitor binding. It is widely accepted that shape complementarityin the receptor–ligand complex reflects significant interaction betweenthe two and these results are in absolute concurrencewith this observa-tion (Tsai et al., 2001).

To investigate the impact of mutation on the panD gene, dockinganalysis was carried out with PZA and POA drug molecules. Thesimulated native and DM proteins were also docked using AutoDockVina and the binding energy was calculated. The binding affinity of na-tive protein with PZA was high with a binding energy of 4.8 kcal/mol.While POA showed decrease affinity towards native (binding energy:−4.3 kcal/mol), whereas its affinity towards DM decreased and similarto PZA with low binding energy of 3.9 kcal/mol, signifying weakinteraction between DM-PZA and POA. The results were further con-firmed by the values of the van der Waals energy at ‐13.42 kcal/mol,electrostatic energy at ‐6.57 kcal/mol and the glide energy (totalligand–receptor energy) at ‐19.99 kcal/mol as calculated for thenative-PZA complex using Schrödinger's Glide program (Table 1),suggesting a stable protein–ligand interaction. For the DM-PZA complex,the van der Waals energy, the electrostatic energy, and the glide energywere calculated to be ‐12.78, 5.72, and ‐18.5 kcal/mol respectively. Thevalue of vanderWaals, electrostatic and glide energy for native-POA com-plex were ‐13.29 kcal/mol, ‐7.23 kcal/mol and ‐20.52 kcal/mol respec-tively while that of DM-POA were ‐13.79 kcal/mol, ‐5.45 kcal/mol and‐19.24 kcal/mol respectively. A considerable energy difference in the na-tive and mutant-complex was observed. The results indicate that theDM-PZA and POA complexes are less stable than the native-PZA andPOA complex structures. It can be concluded that this mutation hasaltered the binding of the drug molecule in the mutant complex. Theinteraction plot generated by Ligplot shows hydrogen and hydro-phobic interactions between ligand and the protein variants. Theparameters considered for the hydrogen bond (H-bond) betweenreceptor and ligand were; distance between acceptor-donor atomsless than 3.3 Å, hydrogen acceptor atom distances not more than2.7 Å and lastly, 90° of an acceptor-donor.

The pre-simulated native-PZA and POA complexes showed the for-mation of 2.77 Å long H-bond between OG1 atom of residue Thr57and O1 atom of PZA (Fig. 2(A)) and 3.14 Å long H-bond between OG1atom of residue Thr57 and O2 atom of POA (Fig. 3(A)). Similarly, inthe pre-simulated DM-PZA complex, single H-bond was formedbetween ND2 atom of residue Asn72 and O1 atom of PZA (Fig. 2(C)).No H-bond was depicted in DM-POA complex.

Strong interactions were accounted by seven hydrophobic interac-tions involving Ser25, Tyr58, Val23, Ala74, Ala75, Gly73, and Val56were also observed in the native-PZA complex (Fig. 4(A)) and sixresidues, Val23, Ser25, Thr58, Gly73 Ala74, and Ala75 were involved inhydrophobic interactions in native-POA complex (Fig. 5(A)). However,only five residues, Val23, Gly24, Tyr22, Ser25, and Thr27 were involvedhydrophobic interaction in DM-PZA complex before simulation(Fig. 4(C)). In DM-POA complex, nine residues such as Val23, Ser25,Leu55, Val56, Thr57, Tyr58, Asn72, Gly73, and Ala75 participated inhydrophobic interaction (Fig. 5(C)).

To analyze the interaction pattern between protein and ligand, weperformed MD simulation studies and investigated the RMSD, RMSF,and radius of gyration (Rg) values to decipher the mechanism of thedrug resistance.

3.4. Investigation of flexibility and compactness in the native and DMvariants

To examine the detailed dynamics of all atoms of the native and DMcomplexes, MD simulations were carried out for a time scale of 50 ns.The simulation resulted in removal of steric clashes, optimization ofmolecular geometry and non-covalent bonding in the protein models.Backbone RMSD values of relative in comparison to the initial structure

40 B. Pandey et al. / Gene 581 (2016) 31–42

were calculated for interpreting the degree of stability and convergenceof the system. RMSD is calculated using Eq. (4):

RMSD t1; t2ð Þ ¼ 1N

Xn

i¼0ri t1ð Þ−ri t2ð Þð Þ2

� �12

ð4Þ

where, t2 is the time of the reference structure, t1 is the particulate timepoint in the simulation, ri represents atom position i at the particulartime and N is the number of atom.

The RMSD plot shows that a stable uniform state was attained after~25,000 ps for both native and DM variants. Difference between theRMSD values of initial and final structures was less for bound DM ascompared to the native complex thus indicating a relatively lesserrange of movement from the initial position. The RMSD plot for bothcomplexesmaintain similar pattern at several positions. The native pro-tein bound with PZA also deviates largely from its previous unboundconformation. However, the DM variant shows same pattern of fluctua-tion in the bound and unbound state after ~20,000 ps as depicted inFig. 6(A). The RMSD values of the native and DM variants were about0.35 nm and 0.3 nm respectively after 25,000 ps and remained stableafter that. A maximum deviation in RMSD was also observed up to~20,000 ps in native-POA and DM-POA and attained the highest devia-tion of ∼0.35 and 0.25 nm respectively. The RMSD structures of native-POA andDM-POA remained constant throughout the rest of time period(Fig. 6(A)).

Tomeasure the compactness of the protein, radius of gyration (Rg) iscalculatedwhich is explained as the RMS distance of the group of atomsfrom their common center of mass and was calculated as given byEq. (5):

Rg ¼ Σir2i mi

Σimiri

!ð5Þ

where, mi implies mass of atom i and ri represents position of atom iwith respect to the center of mass of the studied protein.

A steady Rg value is possible when the protein is stably folded atdifferent time points during simulation and the value will vary whenthe protein unfolds (Kumar et al., 2014). The Rg of all the frames duringthe simulation run was plotted against time and analyzed. During thefirst 5 ns, the native and DM complexes exhibited a similar pattern ofRg value. Rg value of the native and DM variants was found to be 1.48and 1.58 respectively (Fig. 6(B)). From the graph, it is revealed thatDM unbound and DM-PZA shows higher Rg value of 1.55 nm decreasesthe stability of the protein. The Rg of native-POAandDM-POAwas foundto be 1.45 and 1.4 nm respectively (Fig. 6(B)). Major fluctuations in theDM variant were observed between 20,000–25,000 ps and 35,000–4000 ps with Rg score of ~1.5 nm which can be explained by theconformational disturbance caused by the mutated and surroundingresidues. However, both, native and DM form a plateau-like curvewhen plotted against time in the end of the simulation. The Rg patternfor the native protein was found to be more stable than the DM variant.

Solvent surface accessible area (SASA) determines the proteinsurface which is in contact with the solvent molecule. A crucial role isplayed by the solvation effect in the maintenance of protein foldingand stability. SASA was calculated using the g_SASA module in theGROMACS package by rolling a spherical probe of radius 1.4 Å alongthe protein surface. Fig. 6(C) describes that the native complex has aSASA value of ~66 nm2 to ~99 nm2 in the 50,000 ps simulation periodwhereas the DM protein shows a comparatively higher SASA value.Hence, it can be inferred that the protein lost its compactness due tothe mutation and experienced more solvent contact than the nativeprotein.

3.5. Comparison of RMSF values between native and DM

Mutations are known to alter the structural flexibility of the protein.The root-mean-square-fluctuation (RMSF) plot was analyzed tounderstand the residue-wise flexibility for both native and DMcomplexes. RMSF of the protein is calculated by Expression (6):

RMSF ¼ffiffiffiffiffiffiffiffiffiffiffiffiffi3B

8π2

q �: ð6Þ

Protein regions which are highly flexible account for higher RMSFvalue whereas the low value will be described by the constrainedregions. DM protein tended to show more inconsistent fluctuationthan the native (Fig. 6(D)). The regions with residues 20–30, 45–55,60–65, 115–120 exhibited higher backbone fluctuation as comparedto the remaining regions of the trajectory. Higher fluctuation in thetrajectory indicates higher flexibility and suggested that the doublemutation affects the binding of PZA and POA and makes the backbonemore flexible to move (Table 2). Overall subsequent results havesuggested that significant conformational changes exist in the DMcomplex as compared to native type.

3.6. Interaction analysis between ligand and protein after simulation

An important role is played by the hydrogen bonds in the stability ofa protein structure. The intermolecular H-bonding pattern and otherinteractions for native and DM during the 50 ns simulation period wasanalyzed to describe the stability of the complex.

50 ns long MD simulations of native and DM variants revealsignificant outcomes of mutation on structure, interaction and dynam-ics of protein and understanding of the resistance mechanism. Afterthe simulation of the docked complex, the nature of interactionbetween the ligand and protein in both the variants changedsignificantly. In course of the complete simulation, the PZA boundwith native showed the formation of two H-bonds; one 3.004 Å longbond between N atom of Gly67 and N2 atom of PZA and another2.572 Å bond between CB atom of Asp109 and O1 atom of PZA(Fig. 2(B)). Besides H-bond interactions, PZA also showed hydrophobicinteractions with residues Ser66, Asp31, Leu32, Val106 and Val108 ofthe native protein (Fig. 4(B)).

Binding of POAwith native type was mediated through one H-bondand several hydrophobic interactions. It was observed that binding in-teraction has been reduced by single H-bond of 3.27 Å was formed byN atom of His77 and O1 atom of POA (Fig. 3(B)) and few hydrophobicinteractions (Val15, Ala18, Leu55, Ile71, Ala75, Ala74, Val79, His80,and Pro81) (Fig. 5(B)).

Furthermore, major changes were observed in the DM variantdocked with PZA and POA with loss of H-bond during simulation lead-ing to decreased stability of the complex. The number hydrophobic res-idues remained same for DM-PZA complex but the residues involved inthe interaction were found to be different than the pre-simulated DMcomplex; the residues are Val79, Ala76, Ala18, Leu78 and His77(Fig. 4(D)). Whereas in DM-POA complex the number of residues par-ticipating in hydrophobic interaction reduced to six (Met5, Leu6,Met93, Ala98, Tyr101 and Pro103) (Fig. 5(D)). The number of H-bondswas found to decrease in case of DM-PZA while it predominates innative-PZA structure making it more rigid while with POA the H-bondinteracting residue changed with greater bond length, thus weaker in-teraction. The same conclusion can be drawn from the RMSF analysisas DM shows high conformational flexibility and instability with lesserparticipation of H-bond interactions. Thus, the interaction study afterMD simulation also shows more stability of the native complex thanthe double mutant complex.

41B. Pandey et al. / Gene 581 (2016) 31–42

3.7. Principal component analysis

PCA analysis was used to elucidate all the strenuous movements inthe structure of native and DM PanD proteins. Magnitude of theeigenvalues represents the magnitude of dynamical fluctuations. Toget an idea of correlated residue motions in native and DM proteins,the diagonal covariance matrices of fluctuations were plotted andused to capture the degree of collinearity for each pair of atoms. Positiveregions (red) are associated with correlated residue movement,whereas negative regions (blue) indicate anti-correlated motions ofthe residues. Analysis of the covariance of backbone revealed that theDM form shows less correlation as compared to the native form(Fig. 7(A & B). The range of eigenvalues indicated that dynamicsbehavior of protein molecule is mainly defined by the first two eigen-vectors. Fig. 8 illustrates the motion projection of the native and DMproteins in phase space along the first two principal eigenvectors at300 K. It was observed that the internal protein fluctuation of the DMstructure was much higher than the native. Altogether, the study ofthe dynamical behavior provides more reliable structural informationon PanD mutation and also the insights into how protein–ligand inter-action can change due to mutations in PanD protein.

4. Conclusion

This study presented an extensive and elaborate dynamical study ofthe panD gene product and its interaction with PZA and POA, a well-known inhibitor. We carried out binding pocket analysis, solventaccessible analysis, docking analysis and 50 ns longmolecular dynamicssimulations to understand the drug resistance mechanism in themutant PanD ofM. tuberculosis. Based on docking score it was predictedthat native type showed remarkable shape complementarity andinteraction with PZA and POA while DM type of PanD exhibited inap-propriate arrangement of the binding site cavity. The time evolutionstudies of MD simulation and consequent ED analysis for the nativeand double mutant protein gives insight into the structural characteris-tics and molecular mechanism conferring drug resistance. The RMSD,RMSF, Rg, and energy value results concluded overall picture of loss ofconformational stability and altered behavior in the DM protein.Further, these long-range computational simulations indicated thatthe mutated PanD is more flexible than the native and the mutationscan alter the overall structure and dynamics of protein. This emphasizesthe significance and importance of molecular dynamics method as re-ported. The study highlights the mechanistic aspects of drug resistanceand would be useful in the discovery of novel inhibitors againsttuberculosis.

Conflicts of interest

The authors declare that they have no competing interests.

Acknowledgments

BP is thankful to the University Grants Commission (UGC) for pro-viding financial assistance in the form of Dr. D.S. Kothari Post DoctoralFellowship Scheme. Authors are also grateful to Jawaharlal NehruUniversity for providing all computational facilities. AG is grateful tothe University Grants Commission, India, for the Faculty RechargePosition.

References

Ahmad, S., Mokaddas, E., 2009. Recent advances in the diagnosis and treatment ofmultidrug-resistant tuberculosis. Respir. Med. 103, 1777–1790.

Albert, A., Dhanaraj, V., Genschel, U., Khan, G., Ramjee, M.K., Pulido, R., Sibanda, B.L., vonDelft, F., Witty, M., Blundell, T.L., 1998. Crystal structure of aspartate decarboxylaseat 2.2 Å resolution provides evidence for an ester in protein self-processing. Nat.Struct. Mol. Biol. 5, 289–293.

Allen, W.J., Rizzo, R.C., 2014. Implementation of the Hungarian algorithm to account forligand symmetry and similarity in structure-based design. J. Chem. Inf. Model. 54,518–529.

Andrusier, N., Nussinov, R., Wolfson, H.J., 2007. FireDock: fast interaction refinement inmolecular docking. Proteins: Struct., Funct., Bioinf. 69, 139–159.

Berendsen, H.J., van der Spoel, D., van Drunen, R., 1995. GROMACS: a message-passingparallel molecular dynamics implementation. Comput. Phys. Commun. 91, 43–56.

Chiu, Y.-C., Huang, S.-F., Yu, K.-W., Lee, Y.-C., Feng, J.-Y., Su, W.-J., 2011. Characteristics ofpncAmutations in multidrug-resistant tuberculosis in Taiwan. BMC Infect. dis. 11, 240.

Connolly, M.L., 1983a. Analytical molecular surface calculation. J. Appl. Crystallogr. 16,548–558.

Connolly, M.L., 1983b. Solvent-accessible surfaces of proteins and nucleic acids. Science221, 709–713.

Cox, J.S., Chen, B., McNeil, M., Jacobs, W.R., 1999. Complex lipid determines tissue-specificreplication of Mycobacterium tuberculosis in mice. Nature 402, 79–83.

Daffé, M., Draper, P., 1997. The envelope layers of mycobacteria with reference to theirpathogenicity. Adv. Microb. Physiol. 39, 131–203.

Darden, T., York, D., Pedersen, L., 1993. Particle mesh Ewald: an N·log(N) method forEwald sums in large systems. J. Chem. Phys. 98, 10089–10092.

DeLano, W.L., 2002. The PyMOL Molecular Graphics System.Deursen, R.v., Blum, L.C., Reymond, J.-L., 2010. A searchable map of PubChem. J. Chem. Inf.

Model. 50, 1924–1934.Dundas, J., Ouyang, Z., Tseng, J., Binkowski, A., Turpaz, Y., Liang, J., 2006. CASTp: computed

atlas of surface topography of proteins with structural and topographical mapping offunctionally annotated residues. Nucleic Acids Res. 34, W116–W118.

Field, S.K., 2015. Bedaquiline for the treatment of multidrug-resistant tuberculosis: greatpromise or disappointment? Ther. Adv. Chronic Dis. 2040622315582325.

Friesner, R.A., Banks, J.L., Murphy, R.B., Halgren, T.A., Klicic, J.J., Mainz, D.T., Repasky, M.P.,Knoll, E.H., Shelley, M., Perry, J.K., 2004. Glide: a new approach for rapid, accuratedocking and scoring. 1. Method and assessment of docking accuracy. J. Med. Chem.47, 1739–1749.

Gopalan, G., Chopra, S., Ranganathan, A., Swaminathan, K., 2006. Crystal structure ofuncleaved L-aspartate-α-decarboxylase from Mycobacterium tuberculosis. Proteins:Struct., Funct., Bioinf. 65, 796–802.

Hess, B., Bekker, H., Berendsen, H.J., Fraaije, J.G., 1997. LINCS: a linear constraint solver formolecular simulations. J. Comput. Chem. 18, 1463–1472.

Hess, B., Kutzner, C., Van Der Spoel, D., Lindahl, E., 2008. GROMACS 4: algorithms for high-ly efficient, load-balanced, and scalable molecular simulation. J. Chem. TheoryComput. 4, 435–447.

Huang, B., Schroeder, M., 2006. LIGSITEcsc: predicting ligand binding sites using theConnolly surface and degree of conservation. BMC Struct. Biol. 6, 19.

Kumar, C.V., Swetha, R.G., Anbarasu, A., Ramaiah, S., 2014. Computational analysis revealsthe association of threonine 118methioninemutation in PMP22 resulting in CMT-1A.Adv. Bioinforma. 2014.

Lahti, J.L., Tang, G.W., Capriotti, E., Liu, T., Altman, R.B., 2012. Bioinformatics and variabilityin drug response: a protein structural perspective. J. R. Soc. Interface 9, 1409–1437.

Lee, B.I., Suh, S.W., 2004. Crystal structure of the Schiff base intermediate prior to decar-boxylation in the catalytic cycle of aspartate α-decarboxylase. J. Mol. Biol. 340, 1–7.

Mesentean, S., Fischer, S., Smith, J.C., 2006. Analyzing large-scale structural change inproteins: comparison of principal component projection and Sammon mapping.Proteins: Struct., Funct., Bioinf. 64, 210–218.

Morris, G.M., Goodsell, D.S., Halliday, R.S., Huey, R., Hart, W.E., Belew, R.K., Olson, A.J.,1998. Automated docking using a Lamarckian genetic algorithm and an empiricalbinding free energy function. J. Comput. Chem. 19, 1639–1662.

Morris, G.M., Huey, R., Lindstrom, W., Sanner, M.F., Belew, R.K., Goodsell, D.S., Olson, A.J.,2009. AutoDock4 and AutoDockTools4: automated docking with selective receptorflexibility. J. Comput. Chem. 30, 2785–2791.

Organization, W.H., 2004. Estimated Incidence, Prevalence and TB Mortality. WHO, Gene-va http://www.who.int/mediacentre/factsheets/fs104/en.

Parrinello, M., Rahman, A., 1981. Polymorphic transitions in single crystals: a newmolecular dynamics method. J. Appl. Phys. 52, 7182–7190.

Peng, C., Ayala, P.Y., Schlegel, H.B., Frisch, M.J., 1996. Using redundant internal coordinatesto optimize equilibrium geometries and transition states. J. Comput. Chem. 17, 49–56.

Pettersen, E.F., Goddard, T.D., Huang, C.C., Couch, G.S., Greenblatt, D.M., Meng, E.C., Ferrin,T.E., 2004. UCSF chimera—a visualization system for exploratory research andanalysis. J. Comput. Chem. 25, 1605–1612.

Rajendran, V., Sethumadhavan, R., 2014. Drug resistance mechanism of PncA inMycobac-terium tuberculosis. J. Biomol. Struct. Dyn. 32, 209–221.

Sastry, G.M., Adzhigirey, M., Day, T., Annabhimoju, R., Sherman, W., 2013. Protein andligand preparation: parameters, protocols, and influence on virtual screening enrich-ments. J. Comput. Aided Mol. Des. 27, 221–234.

Schneidman-Duhovny, D., Inbar, Y., Nussinov, R., Wolfson, H.J., 2005. PatchDock andSymmDock: servers for rigid and symmetric docking. Nucleic Acids Res. 33,W363–W367.

Schrödinger, L., 2013. Schrödinger Suite 2011 Protein Preparation Wizard Epik version 2.SchuÈttelkopf, A.W., Van Aalten, D.M., 2004. PRODRG: a tool for high-throughput crystallog-

raphy of protein–ligand complexes. Acta Crystallogr. D Biol. Crystallogr. 60, 1355–1363.Sharma, R., Kothapalli, R., Van Dongen, A., Swaminathan, K., 2012. Chemoinformatic iden-

tification of novel inhibitors against Mycobacterium tuberculosis L-aspartate alpha-decarboxylase. PLoS One 7.

Shi, W., Chen, J., Feng, J., Cui, P., Zhang, S., Weng, X., Zhang, W., Zhang, Y., 2014. Aspartatedecarboxylase (PanD) as a new target of pyrazinamide inMycobacterium tuberculosis.Emerg. Microbes Infect. 3, e58.

Trott, O., Olson, A.J., 2010. AutoDock Vina: improving the speed and accuracy of dockingwith a new scoring function, efficient optimization, and multithreading. J. Comput.Chem. 31, 455–461.

42 B. Pandey et al. / Gene 581 (2016) 31–42

Tsai, C.J., Norel, R., Wolfson, H.J., Maizel, J.V., Nussinov, R., 2001. Protein–Ligand Interac-tions: Energetic Contributions and Shape Complementarity. eLS.

Turner, P., 2005. XMGRACE, Version 5.1.19. Center for Coastal and Land-Margin Research.Oregon Graduate Institute of Science and Technology, Beaverton, OR.

Van Der Spoel, D., Lindahl, E., Hess, B., Groenhof, G., Mark, A.E., Berendsen, H.J., 2005.GROMACS: fast, flexible, and free. J. Comput. Chem. 26, 1701–1718.

van Gunsteren, W.F., Billeter, S., Eising, A., Hünenberger, P.H., Krüger, P., Mark, A.E., Scott,W., Tironi, I.G., 1996. Biomolecular Simulation: The {GROMOS96} Manual and UserGuide.

Wu, Y., Tepper, H.L., Voth, G.A., 2006. Flexible simple point-charge water model with im-proved liquid-state properties. J. Chem. Phys. 124, 024503.

Zhang, C., Vasmatzis, G., Cornette, J.L., DeLisi, C., 1997. Determination of atomicdesolvation energies from the structures of crystallized proteins. J. Mol. Biol. 267,707–726.

Zhang, S., Chen, J., Shi, W., Liu, W., Zhang, W., Zhang, Y., 2013. Mutations in panDencoding aspartate decarboxylase are associated with pyrazinamide resistance inMycobacterium tuberculosis. Emerg. Microbes Infect. 2, e34.