Identification of a Potential Inhibitor (MCULE-8777613195-0-12) of New Delhi Metallo-β-Lactamase-1 (NDM-1) Using In Silico and In Vitro Approaches

New Delhi metallo-β-lactamase-1 (NDM-1), expressed in different Gram-negative bacteria, is a versatile enzyme capable of hydrolyzing β-lactam rings containing antibiotics such as penicillins, cephalosporins, and even carbapenems. Multidrug resistance in bacteria mediated by NDM-1 is an emerging threat to the public health, with an enormous economic burden. There is a scarcity in the availability of specific NDM-1 inhibitors, and also a lag in the development of new inhibitors in pharmaceutical industries. In order to identify novel inhibitors of NDM-1, we screened a library of more than 20 million compounds, available at the MCULE purchasable database. Virtual screening led to the identification of six potential inhibitors, namely, MCULE-1996250788-0-2, MCULE-8777613195-0-12, MCULE-2896881895-0-14, MCULE-5843881524-0-3, MCULE-4937132985-0-1, and MCULE-7157846117-0-1. Furthermore, analyses by molecular docking and ADME properties showed that MCULE-8777613195-0-12 was the most suitable inhibitor against NDM-1. An analysis of the binding pose revealed that MCULE-8777613195-0-12 formed four hydrogen bonds with the catalytic residues of NDM-1 (His120, His122, His189, and Cys208) and interacted with other key residues. Molecular dynamics simulation and principal component analysis confirmed the stability of the NDM-1 and MCULE-8777613195-0-12 complex. The in vitro enzyme kinetics showed that the catalytic efficiency (i.e., kcat/Km) of NDM-1 on various antibiotics decreased significantly in the presence of MCULE-8777613195-0-12, due to poor catalytic proficiency (kcat) and affinity (Km). The IC50 value of MCULE-8777613195-0-12 (54.2 µM) was comparable to that of a known inhibitor, i.e., D-captopril (10.3 µM). In sum, MCULE-8777613195-0-12 may serve as a scaffold to further design/develop more potent inhibitors of NDM-1 and other β-lactamases.


Introduction
The emergence of multidrug-resistant bacterial species not only in hospitals but also in community settings poses a serious threat [1][2][3]. Such bacteria utilize various mechanisms to avoid the adverse effects of antibiotics/drugs. Some of the antibiotic resistant mechanisms include (i) the expression of β-lactamases capable of hydrolyzing β-lactam antibiotics, (ii) modification of the drug target, (iii) expression of drug efflux pumps, and (iv) alternate biosynthetic pathways. The production of β-lactamases is the most prevalent form of resistance in Gram-negative bacteria [4]. Ambler classifies β-lactamases into four classes

Validation of Docking Protocol
The legitimacy of the docking procedure was confirmed by re-docking the ligand (i.e., hydrolyzed meropenem) found in the crystal structure of NDM-1, and comparing the RMSD between the two poses (Supplementary Figure S1). The RMSD between the docked pose and the crystal structure pose was 1.4568 Å, which indicated that the redocked ligand occupied a similar binding pocket and formed native interactions with NDM-1. An RMSD value of less than 2.0 Å is considered to be acceptable for the molecular docking procedure [16]. Thus, the adopted docking protocol was accurate for the performed molecular docking between the selected ligands and NDM-1.
Furthermore, the druglikeness of the shortlisted ligands was determined by computing its Lipinskis (Pfizer, New York, NY, USA), Ghose, Veber (GSK, London, UK), Egan (Pharmacia, Uppsala, Sweden), Muegge (Bayer, Leverkusen, German) and bioavialablity scores. We found that all the shortlisted ligands followed all the above stated rules except MCULE-1996250788-0-2, in which case Lipinski's rule of five was violated ( Table 3). The bioavailability score of all the selected ligands was 0.55. In terms of medicinal chemistry, there were no PAINS alerts in all the shortlisted ligands, and no Brenk alterations, except in MCULE-2896881895-0-14 and MCULE-5843881524-0-3. The overall synthetic accessibility of all the shortlisted ligands was in the range of 3.29-4.35. Finally, in terms of leadlikeness, only MCULE-8777613195-0-12 showed zero violations, while other shortlisted ligands showed one or more violations. Thus, on the basis of the ADME properties and molecular docking analysis, the ligand MCULE-8777613195-0-12 was selected as the best molecule which may exhibit the potential to inhibit NDM-1. Therefore, MCULE-8777613195-0-12 was subjected to molecular dynamics simulation and enzyme kinetics studies. Table 3. ADMET properties of the selected molecules deduced by SwissADME.  The RMSD is measured as a deviation in the protein-ligand complex structure from its initial state [17]. Figure 3A shows the RMSD values of NDM-1 in the absence and presence of MCULE-8777613195-0-12 as a function of simulation time. The RMSD values of NDM-1 alone fluctuated significantly during the first 10 ns and then became stable for the rest of simulation. The RMSD values of NDM-1 alone, MCULE-8777613195-0-12 alone, and the MCULE-8777613195-0-12 and NDM-1 complex were fluctuating in the ranges of 0.122-0.157 nm, 0.003-009 nm, and 0.146-0.201 nm, respectively. Since the RMSD values were within the acceptable range of 0.2 nm, this suggested that the structure of the protein alone or in complex form did not deviate significantly during the MD simulation [18]. The average RMSDs of NDM-1 alone, MCULE-8777613195-0-12 alone, and the MCULE-8777613195-0-12 and NDM-1 complex were 0.148 ± 0.011 nm, 0.006 ± 0.002 nm, and 0.157 ± 0.018 nm, respectively. The formation of a stable complex between NDM-1 and MCULE-8777613195-0-12 is clearly indicated by these results.

Root Mean Square Fluctuation (RMSF)
The local conformational changes in the side chain of protein are generally measured by measuring RMSF [19]. The RMSF of NDM-1 in the absence and presence of MCULE-8777613195-0-12 was measured and the result is presented in Figure 3B. At the N-and Cterminal ends, the RMSF values were higher due to the higher flexibility of terminals. Considerably higher RMSF values were shown by some amino acid residues of NDM-1, which might be due to the entry or binding of MCULE-8777613195-0-12 at the substrate binding site of NDM-1.

Radius of Gyration (Rg)
During MD simulation, the overall structure and folding state of a protein may be affected due to the binding of ligand. This can be easily measured by determining Rg as a function of simulation time [20]. Figure 4A

Root Mean Square Fluctuation (RMSF)
The local conformational changes in the side chain of protein are generally measured by measuring RMSF [19]. The RMSF of NDM-1 in the absence and presence of MCULE-8777613195-0-12 was measured and the result is presented in Figure 3B. At the N-and C-terminal ends, the RMSF values were higher due to the higher flexibility of terminals. Considerably higher RMSF values were shown by some amino acid residues of NDM-1, which might be due to the entry or binding of MCULE-8777613195-0-12 at the substrate binding site of NDM-1.

Radius of Gyration (Rg)
During MD simulation, the overall structure and folding state of a protein may be affected due to the binding of ligand. This can be easily measured by determining Rg as a function of simulation time [20]. Figure 4A   The exposure of a protein-ligand complex to its surrounding solvent can be measured by calculating the SASA. It also signifies the overall packing of a protein-ligand system and its stability during MD simulation [17]. Figure 4B depicts the behavior of the SASA during MD simulation of NDM-1 in the absence and presence of MCULE-8777613195-0-12. Some minor fluctuations in the SASA of both systems were observed; however, they remained within the acceptable limits. The SASA values of NDM-1 alone or in the presence of MCULE-8777613195-0-12 during 20-100 ns were in the range of 106-114 nm 2 and 111-120 nm 2 , respectively. The average SASA values of NDM-1 alone and the NDM-1-MCULE-8777613195-0-12 complex during 20-100 ns were 111 ± 12 nm 2 , and 117 ± 16 nm 2 , respectively. In brief, the formation of a stable NDM-1 and MCULE-8777613195-0-12 complex was suggested by the results of the SASA analysis, along with Rg.

Principal Component Analysis (PCA)
PCA is a widely used method to examine the global motion of target proteins in the absence and presence of their respective ligands during simulation [21]. The conformational sampling of NDM-1 alone or in the presence of MCULE-8777613195-0-12 was computed along the PC1 and PC2 projected by the Cα-atoms ( Figure 5). Each red and black dot represented a conformational state of NDM-1, while the red and black clusters indi- The exposure of a protein-ligand complex to its surrounding solvent can be measured by calculating the SASA. It also signifies the overall packing of a protein-ligand system and its stability during MD simulation [17]. Figure 4B depicts the behavior of the SASA during MD simulation of NDM-1 in the absence and presence of MCULE-8777613195-0-12. Some minor fluctuations in the SASA of both systems were observed; however, they remained within the acceptable limits. The SASA values of NDM-1 alone or in the presence of MCULE-8777613195-0-12 during 20-100 ns were in the range of 106-114 nm 2 and 111-120 nm 2 , respectively. The average SASA values of NDM-1 alone and the NDM-1-MCULE-8777613195-0-12 complex during 20-100 ns were 111 ± 12 nm 2 , and 117 ± 16 nm 2 , respectively. In brief, the formation of a stable NDM-1 and MCULE-8777613195-0-12 complex was suggested by the results of the SASA analysis, along with Rg.

Principal Component Analysis (PCA)
PCA is a widely used method to examine the global motion of target proteins in the absence and presence of their respective ligands during simulation [21]. The conformational sampling of NDM-1 alone or in the presence of MCULE-8777613195-0-12 was computed along the PC1 and PC2 projected by the Cα-atoms ( Figure 5). Each red and black dot represented a conformational state of NDM-1, while the red and black clusters indicate the presence of distinct energetically favorable conformational space. The conformational subspace occupied by NDM-1 alone spans from −15 to +15 along PC1 (30.83%), and −12 to +15 along PC2 (11.12%) ( Figure 5A). It is noticeable that the first three eigenvalues of NDM-1 alone occupied 52.1% of the conformational variances ( Figure 5B). Similarly, the conformational subspace occupied by NDM-1 in the presence of MCULE-8777613195-0-12 spans from −15 to +12 along PC1 (17.62%), and −15 to +10 along PC2 (12.14%) ( Figure 5C). The first three eigenvalues of NDM-1 in the presence of MCULE-8777613195-0-12 occupied 38.6% of the conformational variances ( Figure 5D). These results indicate that there is marginal decrease in the flexibility of NDM-1 in the presence of MCULE-8777613195-0-12, suggesting the formation of a stable protein-ligand complex.

Analysis of Enzyme Kinetics Parameters
Steady-state enzyme kinetics was performed to evaluate the potential of the identified ligand, i.e., MCULE-8777613195-0-12, to inhibit NDM-1 enzyme activity. The NDM-1 enzyme alone was found to hydrolyze different β-lactam rings containing substrates such as ampicillin, cefotaxime, imipenem, and meropenem, in addition to a chromogenic substrate nitrocefin ( Table 4). The affinity (Km; defined as the concentration of substrate at which the enzyme attains 50% of its maximum velocity), activity (kcat), and efficiency (kcat/Km) of NDM-1 against different substrates in the absence of MCULE-8777613195-0-12 were estimated to be in the ranges of 27.1-99.4 µM, 271.2-700.3 s −1 , and 3.94-10.01 µM −1 s −1 , respectively (Table 4). These results of NDM-1 kinetics in the absence of any inhibitor agreed with our previously published reports [22][23][24][25]. However, in the presence of

Analysis of Enzyme Kinetics Parameters
Steady-state enzyme kinetics was performed to evaluate the potential of the identified ligand, i.e., MCULE-8777613195-0-12, to inhibit NDM-1 enzyme activity. The NDM-1 enzyme alone was found to hydrolyze different β-lactam rings containing substrates such as ampicillin, cefotaxime, imipenem, and meropenem, in addition to a chromogenic substrate nitrocefin ( Table 4). The affinity (K m ; defined as the concentration of substrate at which the enzyme attains 50% of its maximum velocity), activity (k cat ), and efficiency (k cat /K m ) of NDM-1 against different substrates in the absence of MCULE-8777613195-0-12 were estimated to be in the ranges of 27.1-99.4 µM, 271.2-700.3 s −1 , and 3.94-10.01 µM −1 s −1 , respectively (Table 4). These results of NDM-1 kinetics in the absence of any inhibitor agreed with our previously published reports [22][23][24][25]. However, in the presence of MCULE-8777613195-0-12, the K m of NDM-1 against different substrates was increased 1.0-to 2.0-fold, the k cat values were decreased by 2.2 to 4.5 times, and the catalytic efficiency (k cat /K m ) were decreased 2.5-to 6.6-fold (Table 4). For comparison, we also determined the kinetic parameters of NDM-1 in the presence of a known inhibitor, namely, D-captopril, using nitrocefin as substrate. The K m , k cat , and k cat /K m of NDM-1 in the presence of D-captopril were estimated to be 78.6 ± 4.4 µM, 162.8 ± 16.3 s −1 , and 2.07 ± 0.15 µM −1 s −1 , respectively. It should be noted that the catalytic efficiency of NDM-1 pre-incubated with MCULE-8777613195-0-12 was comparable to that of D-captopril. These in vitro enzyme kinetics results proved that MCULE-8777613195-0-12 was a potent inhibitor of the NDM-1 enzyme. Since both K m and k cat values of NDM-1 were affected in the presence of MCULE-8777613195-0-12, a mixed kind of inhibition is anticipated.

Analysis of IC 50 Value
The potential of MCULE-87776613195-0-12 to inhibit NDM-1 was accessed by determining IC 50 value and comparing it with that of a known NDM-1 inhibitor, i.e., D-captropril ( Figure 6). The IC 50 values of MCULE-87776613195-0-12 and D-captopril were estimated as 54.2 ± 6.3 µM and 10.3 ± 2.7 µM, respectively. Earlier, the IC 50 values of D-captopril were reported to be 7.9-11.8 µM [25,26], which was close to the value obtained in this study. Since the IC 50 value of MCULE-87776613195-0-12 was around 5-fold higher than the known inhibitor (D-captopril), the potential of the identified drug molecule as a potent inhibitor of NDM-1 is revealed.

Discussion
Antibiotic resistance in bacteria and its global spread is an arising hazard to human health with great commercial consequences. Despite the severity of the situation, no clinical inhibitors are available against β-lactamase expressing bacteria, which is considered to be the main reason for antibiotic resistance [27,28]. Among the β-lactamases, metalloβ-lactamases such as NDM-1 are the most potent and versatile enzymes, capable of cleaving nearly all the feasible antibiotics, including carbapenems. Thus, NDM-1 is the most appropriate target in which to identify novel inhibitors in order to contain the spread of

Discussion
Antibiotic resistance in bacteria and its global spread is an arising hazard to human health with great commercial consequences. Despite the severity of the situation, no clinical inhibitors are available against β-lactamase expressing bacteria, which is considered to be the main reason for antibiotic resistance [27,28]. Among the β-lactamases, metallo-β-lactamases such as NDM-1 are the most potent and versatile enzymes, capable of cleaving nearly all the feasible antibiotics, including carbapenems. Thus, NDM-1 is the most appropriate target in which to identify novel inhibitors in order to contain the spread of antibiotic resistance in bacteria [28,29]. Earlier studies also suggest the suitability of NDM-1 as the most promising drug target to identify inhibitors such as ethylenediamine derivatives [30,31], pyridine derivatives [32], spiro-indole-thiadiazole derivatives [33], magnolol derivatives [34], pterostilbenes [35], sulfur-containing carboxylic acids [10,11], dithioazolidine derivatives [36], dipicolinic acids [37], phosphates [38], cyclic borates [7], Bi(III) compounds [39], chromones [40], sulfonamides [12], triazothioacetamides [41], and natural compounds [13]. This motivated us to screen a large database (MCULE's purchasable database, containing more than 20 million compounds) to discover novel non-β-lactam ring encompassing inhibitors against NDM-1. The existing mechanism of defense in resistant bacteria enables them to hydrolyze the β-lactam ring containing antibiotics/inhibitors. Thus, non-β-lactam-based inhibitors would be a good choice against antibiotic resistant bacteria as such, inhibitors would not be inactivated and hydrolyzed by them [1]. In this article, multi-dimensional approaches such as virtual screening, molecular docking/dynamics, principal component analysis, and in vitro enzyme kinetics were applied to identify NDM-1 inhibitors.
The X-ray crystal structure of NDM-1 revealed that NDM-1 has a αβ/βα conformation, with a deep central active site containing two Zn ions. The Zn1 is coordinated with His120, His122, and His189 in a tetrahedral geometry, while Zn2 is coordinated with Asp124, Cys208 and His250 in a trigonal pyramidal geometry [44,45]. These residues are significant in maintaining the proper orientation of di-Zn ions in the catalytic center [29]. An analysis of the docking pose of MCULE-8777613195-0-12 inside the catalytic site of NDM-1 indicates that MCULE-8777613195-0-12 interacted with NDM-1 through hydrogen bonding with key catalytic residues such as His120, His122, His189, and Cys208. In addition to the catalytic residues, some non-active residues also play significant role in maintaining the proper orientation of the substrate for feasible hydrolytic reaction. During substrate binding, Met67 moves away from the di-Zn center by~4.9 Å. This re-orientation brings Leu65 closer to the di-Zn center by~2.1 Å [29]. Trp93 along with Met67 and Leu65 facilitate the entry of substrate towards the active site. Moreover, as a result of these movements, Asn220 is pulled~1.0 Å closer to the di-Zn center, where it interacts with the carbonyl group of the substrate. Consequently, an oxy-anion hole is created at the substrate by Asn220 and Zn1, thereby facilitating hydrophilic attack by hydroxide ion, which was produced from the water molecule attached to Asp124 [25]. We also noticed that MCULE-8777613195-0-12 and NDM-1 showed a hydrophobic interaction with Trp93, and van der Waals interactions with Zn1, Zn2, Phe70, Val73, Gln123, Asp124, Lys211, Gly219, Asn220, and His250. The van der Waals interactions play significant role in determining the formation of a stable protein-ligand complex. These are distance-dependent forces; most act collectively to make an impact. The models based on the Lennard-Jones potential are useful in accurately estimating van der Waals interactions and, thus, are useful in molecular docking simulations and the virtual screening of large databases. Furthermore, the stability of the NDM-1 and MCULE-8777613195-0-12 complex was probed by MD simulation; the results (RMSD, RMSF, Rg, and SASA) suggest the formation of a stable NDM-1 and MCULE-8777613195-0-12 complex. These results were also confirmed by PCA analysis. Furthermore, the effect of MCULE-8777613195-0-12 on NDM-1 activity was evaluated on a number of substrates such as ampicillin, cefotaxime, imipenem, and meropenem, in addition to a chromogenic substrate, nitrocefin. The results confirmed that the binding of MCULE-8777613195-0-12 to NDM-1 reduced its affinity towards substrates, as well as its activity.

Binding Site Determination Using CASTp3.0
The 3D structure of NDM-1 bound with hydrolyzed meropenem (control ligand) was submitted to CASTp3.0 server [49] to identify the most suitable binding site present on NDM-1.

Preparation of Ligands/Protein, and Virtual Screening
The MCULE purchasable (in stock) library was used for high-throughput virtual screening (accessed on 09/03/2022). The high-throughput virtual screening was performed using "MCULE online drug discovery platform", as described previously [46]. The molecules in the library were filtered using the "Basic property filter" of the platform. Various properties such as components, inorganic atoms, rotatable bonds, chiral centers, RO5 violations, heavy atoms, N/O atom, rings, and halogen atoms were available to filter the molecules. The minimum and maximum values in the screening input parameters were defined based on the values of meropenem. A total of 20,202,562 ligands was selected for screening purposes against the NDM-1 active site. The values of sampler size and maximum number of compounds after sphere exclusion were set to 1000 and 3,000,000, respectively. The other options were set to their default values.
The 3D coordinates of the target protein (NDM-1) were downloaded from RCSB databank (PDB ID: 4EYL). The NDM-1 crystal structure with bound hydrolyzed meropenem was resolved to 1.90 Å [44]. Prior to molecular docking, protein was prepared by adding H-atoms, assigning bond orders, removing any heteroatoms including, and deleting all noncatalytic water molecules. The changes in Zn ions were maintained. A new hydrogen bond network was defined, and the energy of the system was minimized using the CHARMM36 force field. Virtual screening was performed using the AutoDock Vina-enabled MCULE screening platform. The pre-processed file of NDM-1 protein was uploaded to the MCULE screening platform. A grid box of 20.8 Å, 23.0 Å, and 24.2 Å dimensions, centered at 8.3 Å, −40.0 Å, and 6.2 Å, was used for the screening purpose. All the ligands were ranked by VINA score and the top six ligands with a VINA score of ≤ −8.0 kcal/mol were selected for further study.

Molecular Docking and Validation of Protocol
The molecular docking of the top-ranked ligands against NDM-1 was again performed using AutoDock4.2, as suggested earlier [50,51]. Briefly, the molecular docking was performed inside a grid box with dimensions of 20.8 Å, 23.0 Å, and 24.2 Å, centered at 8.3 Å, −40.0 Å, and 6.2 Å with a spacing between grid points of 0.375 Å. Lamarck Genetic Algorithm (LGA) was used for the global search while the Solis-Wets method was used for the local search of the binding site of ligands inside NDM-1's active site. A total of 2.5 × 10 6 energy calculations were performed for each docking run, and a total of 50 docking runs were computed. The values of population size, translational step, torsion steps, and quaternions were set to 150, 0.2 Å, 5, and 5, respectively. The molecular interaction between the ligand and protein was identified using Discovery Studio Visualizer 4.1. The dissociation constant (K d ) for the NDM-1 and ligand interaction was calculated from the docking energy (∆G) using the following relation, and as defined earlier [52,53]: where R and T are the universal gas constant and temperature, respectively.
The validation of the molecular docking protocol was performed by re-docking the ligand (meropenem), which was bound to the NDM-1 active site in its crystal structure. For this purpose, the bound ligand was first extracted from the crystal structure and then docked again to the active site of NDM-1 using the same set of docking parameters. Finally, the root mean square deviation (RMSD) between the docked and crystal structure poses was computed by super-imposing the two structures.

Determination of ADME Properties
The pharmaco-kinetic, i.e., adsorption, distribution, metabolism, and excretion (ADME), properties of the top six ligands were determined using SwissADME [54]. In addition, the druglikeness of the selected ligands was determined by performing several tests such as those of Lipinski, Ghose, Veber, Egan, PAINS, and Muegge. The top-rated ligand was identified which had favorable ADME properties and passed all the aforementioned tests for druglikeness. Lipinski's rule of five states that a druglike molecule should have a molecular mass of less than 500 Da; it should have no more than 5 hydrogen bond donors and 10 hydrogen bond acceptors; and its lipophilicity (i.e., MlogP) should be less than 5. Similarly, the Ghose filter suggests a druglike molecule should have a molecular weight in the range of 160-480 Da, its lipiphilicity (WlogP) should be in the range of −0.4 to 5.6, with a molar refractivity ≤130, and the number of atoms should be in the range of 20-70 [42]. Likewise, Veber's filter indicates the number of rotatable bonds should be ≤10 and the total polar surface area should be ≤140 [43]. Further, Egan's filter shows that lipophilicity (WlogP) should be ≤5.88 and the total polar surface area should be ≤131. Furthermore, Muegge's filter suggests that the molecular weight of a druglike molecule should be within the range of 200-600 Da, its lipophilicity (XlogP3) should be in the range of −2 to 5, its total polar surface area should be ≤150, the number of rings should be ≤7, the number of carbon should be >4, the number of heteroatoms should be >1, the number of rotatable bonds should be ≤15, the number of hydrogen bond acceptors should be ≤10, and the number of hydrogen bonds should be ≤5.

Molecular Dynamics (MD) Simulation
Molecular dynamics (MD) simulation was performed to evaluate the stability of the NDM-ligand complex using Gromacs 2020.4 installed on a workstation powered by Intel Xenon E3-1245 with eight cores, a 3.50 GHz processor, 32 GB RAM, and an NVIDIA Quadro P5000 GPU card [55,56]. The pdb2gmx command of GROMACS was used to generate protein topology using a CHARMM36-all atom forcefield, and TIP3P water molecules, while CHARMM General Force Field (CGenFF) was used to generate the ligand's topology. A dodecahedron box was used to perform the MD simulation after placing the NDM-1 and ligand complex at the center, at least 1.0 nm away from the boundaries of the box. A total of 17,014 TIP3P water molecules were used to solvate the simulation box, and 5 Na+ ions were added to neutralize the system. Furthermore, NaCl (150 mM) was added to imitate the physiological conditions. A maximum 50,000 steps was used to minimize the energy of the system by the steepest descent minimization method. Furthermore, isothermal-isochroic (NVT) and isothermal-isobaric (NPT) ensembles were accomplished at a temperature of 300 K and a pressure of 1.0 bar, which were maintained throughout using a Berendsen thermostat and Parrinello-Rahman barostat. Finally, a production run of 100 ns was performed on the equilibrated system, with a time-step of 2 fs that was fixed using a leap-frog integrator. The algorithm for NVT, NPT, and the production runs was constrained using LINCS. Finally, the MD simulation results were investigated for root mean square deviation (RMSD), root mean square fluctuation (RMSF), radius of gyration (Rg), and solvent accessible surface area (SASA) [20,55]. All the experiments were performed independently in triplicates and the results are reported as mean ± standard errors.

Principal Component Analysis (PCA)
Principal component analysis (PCA) or essential dynamics (ED) is a widely used method to compute the conformational flexibility of protein in the presence of ligands by measuring their collective motions. In this study, the PCA of NDM-1 in the presence of the top-rated ligand was performed using Bio3D, as reported previously [57,58]. In PCA, the translational and rotational motions of the protein were first removed. Then, the atomic coordinates' positional covariance matrix and its eigenvectors were computed by superimposing the coordinates of the protein onto a reference structure. Later, a diagonal matrix of eigenvalues was generated by diagonalizing the calculated symmetric matrix by an orthogonal coordinate transformation matrix. In this matrix, each eigenvector represents an eigenvalue associated with the total mean-square fluctuation of the system, along the corresponding eigenvector. The covariance matric (C) is calculated using the following relation. 2, 3, . . . ., 3N) where N, x i/j , and <x i/j > represent the number of Cα-atoms, the Cartesian coordinate of the ith/jth Cα-atom, and time average of all the conformations, respectively.

Determination of Enzyme Kinetics Parameters
The purified NDM-1 was obtained from GenScript (New Jersey, NJ, USA) and its zinc content was determined using PAR (4-(2-Pyridylazo) resorcinol) assay, as described elsewhere [23]. The concentration of NDM-1 was determined spectrophotometrically using a molar extinction coefficient of 27,800 M −1 cm −1 . The stock solution of inhibitor was prepared in DMSO and then diluted in the assay buffer (final DMSO concentration was less than 0.5%). Furthermore, the steady-state enzyme kinetics was performed to evaluate the hydrolytic activity of NDM-1 in the presence of the top-rated ligand, as reported previously [25]. The change in the molar extinction coefficients of the antibiotics/substrates upon hydrolysis were ∆ε 486 = +15,000 M −1 cm −1 for nitrocefin, ∆ε 235 = −900 M −1 cm −1 for ampicillin, ∆ε 264 = −7,250 M −1 cm −1 for cefotaxime, ∆ε 295 = −10,500 M −1 cm −1 for imipenem, and ∆ε 297 = −10,940 M −1 cm −1 for meropenem. The enzymatic reaction was performed in 50 mM HEPES buffer (pH 7.0) containing NaCl (0.25 M), and ZnCl 2 (0.1 mM) at 30 • C. BSA (with no hydrolytic activity of its own) was added at a. concentration of 20 µg/ml to the reaction buffer to prevent the denaturation of NDM-1. Different concentrations of antibiotics/substrates were incubated with varying concentrations (0.1 to 2 nM) of NDM-1. The initial velocities were calculated from the observed change in absorbance upon antibiotic/substrate hydrolysis and the kinetic parameters (K m and k cat ) were calculated using the Michaelis-Menten equations: where v, V max , [S], and [E] are initial velocity, maximum velocity, substrate concentration, and enzyme concentration, respectively. Three independent experiments were performed and the results are reported as mean ± standard error.

Determination of IC 50
The IC 50 value of a ligand/drug is defined as the concentration at which the activity of an enzyme is reduced by 50%. In this study, the IC 50 values of the top-rated ligand and D-captopril (as control) were determined by observing the hydrolysis of 100 µM nitrocefin at 486 nm, as published earlier [26]. Briefly, the NDM-1 enzyme (0.5 nM) was incubated for 5 min at 30 • C with varying concentrations (0.001 to 1000 µM) of the ligand and D-captopril. The change in absorbance due to the hydrolysis of nitrocefin was converted into enzyme activity and a plot of activity versus log (ligand) concentration was plotted to compute the IC 50 values, using the following relation in a Sigma plot: Y = (max − min) 1 + 10 ((logIC 50 −X)×Hill slope) where X represents inhibitor concentration, Y represents %inhibition data, IC 50 is the concentration of substrate at which the activity is reduced by 50%, and Hill slope is the slope factor of the plot.
The results are reported as mean ± standard error of three independent experiments.