Identification of Potential Non-nucleoside MraY Inhibitors for Tuberculosis Chemotherapy Using Structure-Based Virtual Screening

Abstract

The efforts to limit the spread of the tuberculosis epidemic have been challenged by the rise of drug-resistant strains of Mycobacterium tuberculosis (Mtb), the causative agent of tuberculosis. It is critical to discover new chemical scaffolds acting on novel or unexploited targets to beat this drug-resistant pathogen. MraY (phospho-MurNAc-pentapeptide translocase or translocase I) is an in vivo validated target for antibacterials-discovery. MraY is inhibited by nucleoside-based natural products that suffer from poor in vivo efficacy. The current study is focused on discovering novel chemical entities, particularly, non-nucleoside small molecules, as MraY_Mtb inhibitors possessing antituberculosis activity. In the absence of any reported X-ray crystal structures of MraY_Mtb, we used a homology model-based virtual screening approach combined with the ligand-based e-pharmacophore screening. We screened ~12 million commercially available compounds from the ZINC15 database using GOLD software. The resulting hits were filtered using a 2-pronged screening method comprising e-pharmacophore hypotheses and docking against the MraY_Mtb homology model using Glide. Further clustering based on Glide scores and optimal binding interactions resulted in 15 in silico hits. We performed molecular dynamics (MD) simulations for the three best-ranking compounds and one other poorer-ranking compound, out of the 15 in silico hits, to analyze the interaction modes in detail. The MD simulations indicated stable interactions between the compounds and key residues in the MraY active site that are crucial for maintaining the enzymatic activity. These in silico hits could advance the antibacterial drug discovery campaign to find new MraY inhibitors for tuberculosis treatment.

Introduction

Tuberculosis (TB) is the leading cause of death from a single infectious agent, claiming 1.5 million lives worldwide in 2018.¹ The current guidelines for the treatment of drug-sensitive TB comprise a combination of four antibiotics: isoniazid (INH), rifampin (RIF), ethambutol (EMB), and pyrazinamide (PZA) over six months. Prolonged treatment regimen and astronomical costs of the chemotherapeutics cause patient non-adherence and a resultant rise in multi-drug-resistant (MDR) and extensively-drug resistant (XDR) strains of Mycobacterium tuberculosis (Mtb), the causative agent for TB. The emergence of MDR and XDR strains of Mtb is a major setback to the World Health Organization’s (WHO) goal of “The End TB Strategy” by 2030.¹ The situation calls urgently for a multi-pronged approach for new antibacterials discovery with novel mechanisms of action against new and existing targets.^2,3 Discovery of anti-TB drugs is critical to shorten the treatment regimen and reduce its cost, which then could improve patient compliance and curtail the problems of drug-resistance.

Peptidoglycan biosynthesis (PG) is an essential step in creating the robust mesh-like structure of the bacterial cell wall that is catalyzed by several enzymes that are well-precedented clinical targets for several successful antibiotics, like penicillin and vancomycin.⁴ One of the enzymes in peptidoglycan biosynthesis that is an underexploited antibiotic discovery target is phospho-N-acetyl-muramyl-pentapeptide translocase (MraY), which is a member of the polyribonucleotide nucleotidyltransferase (PNPT) superfamily of bacterial and eukaryotic integral membrane enzymes. MraY (also known as MurX for Mtb) catalyzes the second stage of peptidoglycan biosynthesis that involves the transfer of the phospho-MurNAc-pentapeptide from UDP-MurNAc-pentapeptide (UM5A) to undecaprenylphosphate (lipid carrier, C₅₅-P), giving rise to undecaprenyl-pyrophosphoryl-MurNAc-pentapeptide, known as lipid I (Figure 1).⁵ MraY is inhibited by five classes of natural nucleoside antibiotics: liposidomycins/caprazamycins, tunicamycins, muraymycins, mureidomycins, and capuramycins.^4,6,7 Studies of these natural inhibitors have aided our understanding of the catalytic mechanisms of MraY. However, none of them have been developed as clinical drug candidates, due to poor in vivo efficacy, structural complexity that demands challenging multi-step syntheses,^8,9 and target promiscuity.¹⁰ To the best of our knowledge, only a few non-nucleoside inhibitors have been reported in the literature, and those have relatively poor MraY inhibitory activity compared to the well-precedented nucleoside natural product counterparts that penetrate the cell wall very well.¹¹ The lack of structural information of MraY_Mtb in the apo state or interacting with a ligand has impeded the in vivo development of both non-nucleoside small molecules as well as nucleoside inhibitors.

Figure 1.

MraY catalyzes the formation of lipid I in PG biosynthesis. Uridine is represented by U-labeled blue pentagons and MurNAc by M-labeled brown hexagons. The P-labeled yellow circles are phosphates. The small spheres in UDP-MurNAc-pentapeptide (UM5A) are peptides. Adapted from, “Crystal Structure of MraY, an essential membrane enzyme for bacterial cell wall synthesis”, Chung BC. et al., Science 2013; 341(6149):1012–1016.

Recently, Lee and co-workers published X-crystal structures of MraY from the Gram-negative bacterium Aquifex aeolicus (MraY_AA) in the apo state¹² bound to Mg²⁺ and in complex with muraymycin D2 (MD2) and other well-known nucleoside-based inhibitors of MraY.^13,14 These findings have advanced our understanding of the structural domains involved in the binding of the MraY substrate and inhibitors. The geometry and the amino acid residues of the active site are well-conserved throughout the PNPT family (MraY homologs). These common features of the family were also observed in another recent co-crystal structure of MraY, from the Gram-positive bacterium Clostridium bolteae (MraY_CB) as a complex with tunicamycin.¹⁵ No crystal structure of MraY_Mtb (from the Mtb species) has been reported yet. Crystallization and structure determination of transmembrane proteins is challenging. However, MraY has an active site that is highly conserved across gram-positive and gram-negative bacteria and Mtb. Not surprisingly, many natural product MraY inhibitors show broad-spectrum MraY activity across different bacterial species. Still, only a few show activity against Mtb.¹³ Our interest in antitubercular agents led us to investigate a targeted approach for identifying novel narrow-spectrum TB-active small molecule inhibitors of MraY. To this end, in this work, we have chosen to use protein structure-based in silico screening, a proven, cost-effective method that has made significant strides in ligand discovery against a plethora of therapeutic targets in the last couple of decades.¹⁶ Here we report, a virtual screening approach that led to the identification of with new chemical entitities as MraY inhibitors that we think will show activity against Mtb, based on homology modeling of MraY_Mtb and molecular dynamics (MD) simulations. These studies will provide insights into using a targeted strategy to develop narrow-spectrum agents and advance the field of antibacterial discovery and development in combatting antimicrobial resistance. Divya et al. also identified a few in silico non-nucleoside hits as potential MraY inhibitors using a virtual screening approach on an Asinex database only; however, the study included only very short MD simulations.¹⁷

We first started by generating single template-based homology models of MraY_Mtb using the X-ray crystal structure of MraY_AA (Aquifex aeolicus) in complex with MD2 (PDB ID: 5CKR). The best MraY_Mtb homology model was validated by an enrichment study followed by MD simulation. Structure-based virtual screening and e-pharmacophore protocols were employed in screening ~12 million commercially available, drug-like compounds from the ZINC15 database against the best MraY_Mtb homology model. The final hits were selected based on clustering and visual inspection of ligand interactions with the MraY_Mtb model. MD simulation results revealed that the identified hits formed and maintained strong and stable interactions with the key residues of the MraY_Mtb protein that are necessary for inhibitory activity.

Results and Discussion

Homology modeling and validation

A total of 100 computational models of MraY_Mtb (UniprotKB code P9WMW7, strain ATCC 25618/ H37Rv) were generated based on sequence alignment with the 5CKR X-ray crystal structure as a template, as shown in Figure 2. A total of five models were selected based on the minimum discrete optimized protein energy (DOPE) scores. These models were refined by including all the missing hydrogens, protonating at pH 7.4, and minimizing their energies (restrained energy minimization) using the “Protein Preparation Wizard” module implemented in the Schrödinger software. The overall structural quality of these models was evaluated using Ramachandran plot analyses (Figure 3A). The Ramachandran plot showed that most of the residues of the five MraY_Mtb-homology models lie in the favored region. These homology models were further evaluated through docking of the reported well-known MraY_Mtb inhibitors, such as capuramycin, muraymycin D1 (MD1) and, caprazamycin B (CPZ-B) considering the availability of the mutagenesis data. The best homology model of MraY_Mtb in complex with MD2 was further refined by MD simulations (Supporting Information, Figures S1 and S2) using the Desmond software. The recently solved X-ray crystal structures of MraY_AA (PDB ID: 5CKR) in complex with MD2, MraY_CB (PDB ID: 5JNQ) in complex with tunicamycin, and our homology model of MraY_Mtb reveal that the geometry and the amino acid residues of the active site are well-conserved (Figure 3B).

Figure 2.

Sequence alignment of MraY_Mtb with MraY from Aquifex aeolicus, which has an available X-ray crystal structure (PDB ID: 5CKR). Identical residues are given a cyan background. The transmembrane regions (TM) are shown in ribbon representation. The TM numbering was adapted from “Crystal Structure of MraY, an essential membrane enzyme for bacterial cell wall synthesis”, Chung BC. et al., Science. 2013; 341(6149):1012–1016. The “*” sign represents the binding site residues interacting with MD2 in 5CKR.

Figure 3.

(A) Ramachandran plot analysis showing the residues in the favored regions. (B) Overlayed zoomed-in view of the active site of the MD2 ligand (carbon in cyan) in complex with the MraY_Mtb homology model (carbon in green) and MraY_CB (5JNQ, carbon in grey) and MraY_AA (5CKR, carbon in deep blue) X-ray crystal structures.

Enrichment Study for Validation of the MraY_Mtb Model

The success of docking during the virtual screening process can be evaluated by studying its capacity to enrich a small number of known active compounds that rank at the top from among a large number of decoy molecules in the database. Enrichment studies play a vital role in distinguishing between various models by testing their capacity to differentiate between actives and decoys (inactive compounds) from a mixed set of compounds. An enrichment study was conducted to ascertain the best model from a set of five homology models selected based on the DOPE score and deemed energetically favorable according to the Ramachandran plot analyses. Figure 4 shows the percentage of active ligands found, plotted as a function of the percentage of screened compounds for each of the five models. The enrichment study indicated Model 1 as the best, with an enrichment factor of 16 and high receiver operating characteristic (ROC) (0.99), in addition to high Boltzmann-enhanced discrimination of the receiver operating characteristic (BEDROCK) (0.994) scores. The ROC and BEDROCK scores vary between 0 and 1, with 1 being the ideal screen performance. The high enrichment factor (EF) of Model 1 reflects the docking calculations’ ability to find true positives from the screened database of known actives and decoys, compared to random selection.

Figure 4.

Enrichment curves of the top five MraY_Mtb homology models that were selected based on the DOPE score and Ramachandran plot.

Virtual Screening to Identify New Small-Molecule Inhibitors Against MraY_Mtb

The ZINC15 database of ~12 million commercially available compounds was screened using the GOLD suite¹⁸ by docking each molecule against the best MraY_Mtb homology model, which was obtained from the MD simulation of the MraY_Mtb–MD2 complex. The screening protocol yielded a total of 50,045 compounds, which were ranked based on their ability to dock to the receptor model using the Chemscore function. We selected the top 15,201 hits based on the Chemscore function that ranged from 29.83 to 10.0 kcal/mol. The flow diagram for the screening protocol is shown in Figure 5. In the hit selection process, we used a two-pronged strategy, standard docking and epharmacophore-based screening, to ensure the robustness of the identified hit molecules.

Figure 5.

The flow diagram for the virtual screening protocol.

For the standard precision docking (Glide SP) method, we screened the 15,201 hits against the MraY_Mtb homology model and the reported structure of MraY_AA from Aquifex aeolicus in complex with carbacaprazamycin, which will be referred to by PDB ID: 6OYH for the rest of this article. During the ongoing project, we observed that large molecular weight compounds with long flexible tail regions (e.g., caprazamycins, muraymycins, etc.) had a greater propensity to dock into the 6OYH structure with the desired pose in comparison to the other available X-crystal structures of MraY and the MraY_Mtb homology model. To address this issue and to ensure that these classes of compounds were not eliminated during the screening process, we also docked these 15,201 hits into the 6OYH structure. The MraY_Mtb homology model resulted in 43,254 docked hits, whereas 42,854 ligands were obtained from 6OYH docking. Three poses per ligand were generated during our docking protocol. The docked hits were further filtered based on the Glide cutoff score ≤−4 kcal/mol, yielding 13,704 hits for the MraY_Mtb homology model and 25,285 hits for 6OYH. The best pose per ligand was selected based on Emodel scores and the duplicates were removed. This criterion yielded 11,032 (6OYH) and 6,930 (MraY_Mtb) hits, respectively. At that stage, the two sets of hits were combined, and redundant hits were removed, resulting in a total of 11,598 hits.

Ligand screening against a pharmacophore hypothesis is more computationally efficient and less time consuming than protein structure-based docking.¹⁹ It is an established tool in identifying potential hits by using knowledge about the protein target and its known active ligands. When ligands are screened in the e-pharmacophore approach, explicit matching is required for the most energetically favorable site (≥ 1.0 kcal/mol) that finds matching pharmacophores in the ligands.

We first generated the hypotheses based on three well-precedented MraY inhibitors that show activity against Mtb, capuramycin, MD1, and CPZ-B, and used the hypotheses for subsequent screening of 15,201 ligands (Figure 6). The three known MraY inhibitors were docked into the Mtb homology model and the 6OYH structure. Three poses were generated for each ligand with each protein, and their binding free energies were calculated using the Prime Molecular Mechanics/Generalized Born Surface Area (MM-GBSA) approach. The pose that had the lowest binding free energy was selected to be used to generate an e-pharmacophore hypothesis. The binding free energies for the protein-ligand complexes are provided in Table 1. A total of six hypotheses were generated manually with the PHASE module implemented in the Schrödinger software. The following hypotheses were generated: (A) MraY_Mtb–CPZ-B, (B) MraY_Mtb–capuramycin, (C) MraY_Mtb–MD1, (D) 6OYH–CPZ-B, (E) 6OYH–capuramycin, and (F) 6OYH–MD1, as shown in Figure 7.

Figure 6.

Representative structures of known MraY inhibitors.

Table 1.

Binding free-energies of the MraY_Mtb–CPZ-B, MraY_Mtb–capuramycin, MraY_Mtb–MD1, 6OYH–CPZ-B, 6OYH–capuramycin, and 6OYH–MD1 complexes.

Protein-ligand complex	ΔG Binding Free Energy MM-GBSA (kcal/mol)
MraYMtb–CPZ-B	−107.98
MraYMtb–Capuramycin	−29.23
MraYMtb–MD1	−81.37
6OYH–CPZ-B	−75.71
6OYH–Capuramycin	−64.59
6OYH–MD1	−56.57

Figure 7.

The energy-based pharmacophore features of the known MraY inhibitors CPZ-B, capuramycin, and MD1 docked into the X-ray crystal structure of 6OYH and modeled MraY_Mtb: (A) MraY_Mtb–CPZ-B, (B) MraY_Mtb–capuramycin, (C) MraY_Mtb–MD1 (D) 6OYH–CPZ-B, (E) 6OYH–capuramycin, and (F) 6OYH–MD1.

We set a criterion in which the 15,201 ligands screened were required to match a minimum of 5 out of the 9 or 10 sites that were generated in the six e-pharmacophore hypotheses with known MraY inhibitors that show TB activity. The ligands were ranked based on their fitness scores. The fitness score measures the alignment of the conformer with the established hypothesis, based on root mean square deviation (RMSD), site matching, vector alignments, and volume terms.

The top-ranked 15,201 ligands were separately aligned with the MraY_Mtb–CPZ-B, MraY_Mtb–capuramycin, and MraY_Mtb–MD1 hypotheses (Figure 7) to give 3256, 858, and 850 ligands, respectively, with a fitness score ranging from 0.4 to 1.8. A fitness cutoff score of ≥1 was applied to eliminate ligands that are less likely to fit into the receptor pocket than the known reference ligands (CPZ-B, capuramycin, and MD1). This criterion reduced the number of hits significantly to 995, 258, and 390, respectively. After combining these three sets of hits and removing the redundant hits, a total of 1350 hits were obtained from the MraY_Mtb homology model (Figure 8). We indicated earlier that large molecular weight compounds with a long tail docked well into the 6OYH structure. Keeping this in mind, we also created a hypothesis for 6OYH–CPZ-B, 6OYH–capuramycin, and 6OYH–MD1 and screened 15,201 ligands against these hypotheses, which yielded a total of 1126 hits. A comparison of the two sets of hits resulted in a total of 664 hits that were common in both the e-pharmacophores from 6OYH and MraY_Mtb. In the final step, a total of 581 common hits were identified from the two screening methods, docking (11,598 hits) and e-pharmacophore (664 hits). These 581 hits were further subjected to fingerprint-based similarity search (linear) to identify unique scaffolds and were subsequently clustered into 50 groups using Hierarchical cluster analysis implemented in Canvas.²⁰ A total of 15 representative hits were selected based on their Glide scores and protein–ligand interactions. The top three hits (Figure 9) and one poorer-ranking hit were subjected to 200 ns all-atom MD simulations to understand the putative binding mechanism and interaction profiles of the hits with the MraY_Mtb protein. In our MD simulations studies, we chose to analyze in detail the interaction modes of the best-ranked structures from the docking only for the four molecules to save computational time. We included the poorer-ranking hit (ZINC9672300) (Supporting Information, Table S1) to help understand the impact of the Glide score and the role of key protein-ligand interactions in hit selection.

Figure 8.

The flow diagram for the e-pharmacophore-based screening protocol.

Figure 9.

Chemical structures of the top three in silico hits identified as MraY_Mtb inhibitors, and one poorer-ranking compound.

Docking Analysis of In Silico Hits

The virtual screening of the compounds from the ZINC15 database using both protein structure-based and ligand-based screening (e-pharmacophore hypothesis) yielded a total of 15 potential hits (Supporting Information, Table S1), which were selected based on their ranking by Glide score and their interactions with the key residues Asp188, Asn249, and Phe256, and by visual inspection of their binding pose in the MD2 binding site of the MraY_Mtb homology model. The structure of MraY_Mtb in surface representation with inhibitor binding site hotspots (HSs) is shown in Figure 10.²¹ The four hit molecules, which also included one poorer-ranking hit (ZINC9672300) (Figure 9), that were identified based on their Glide scores and key interactions with MraY_Mtb, were subjected to MD simulations. The Glide Scores of these four compounds varied from −6.48 to −5.46 kcal/mol. The schematic representation of 2D interactions of these potential hits with MraY_Mtb and their structures in surface representation are shown in Figures 11–14.

Figure 10.

Structure of MraY_Mtb in surface representation with inhibitor binding site HSs color-coded and labeled as follows: uridine (peach), uridine-adjacent (HS1; green), TM9b/LoopE (HS2; purple), caprolactam (HS3; pink), hydrophobic (HS4; cyan), Mg²⁺ (HS5; gold). The figure was adapted from “Chemical logic of MraY inhibition by antibacterial nucleoside natural products”, Mashalidis, EH. et al. Nat. Commun. 2019; 10(1):1–12.

Figure 11.

(A) Schematic representation of 2D interactions found in the docked pose of the MraY_Mtb–ZINC44893147 complex; (B) Structure of the docked pose of MraY_Mtb– ZINC44893147 complex in surface representation.

Figure 14.

(A) Schematic representation of 2D interactions in the docked pose of the MraY_Mtb–ZINC9672300 complex; (B) 3D structure of the docked pose of the MraY_Mtb–ZINC9672300 complex in surface representation.

It was gratifying to observe that the thienopyridine derivative (ZINC44893147) interacted strongly with the uridine binding pocket of MraY (Figures 10 and 11), which is a crucial interaction that imparts MraY activity. In the uridine binding pocket of MraY, both pyridine and thiophene moieties participated in π–π stacking interactions with the Phe256 residue that is critical for MraY activity (Figure 11). The H-bond interactions of the hydroxy group with residues Asp188 and Lys46 and the salt-bridge interaction between Asn249 and the carbonyl oxygen (C=O) imparted stability to the MraY_Mtb–ZINC44893147 complex. While the carbonyl oxygen of the acyclic amide moiety interacted with Gly258 of the HS1 pocket (Figure 10), its adjacent residue Met257 made a water-mediated interaction with carbonyl oxygen of the conjugated cyclic amide moiety.

ZINC97985272 exhibited a wide range of interactions with different residues of the protein (Figure 12A), as can be seen from the 2D-interaction diagram. A strong π–π stacking interaction with Phe256 provided stability to the complex that is imparted by the two phenyl rings of the 3-ring fused system. In addition, hydrogen-bond (H-bond) interactions with key residues like Asp185, Met257, and Gly258 of the uracil and uracil-adjacent pocket and a cation–π interaction with the Lys105 residue (HS3, Figure 10) were also present. The hydrophobic interactions between the alkane chain and the hydrophobic residues, Pro41, Val296, Val 297, Ala314, and Pro315 (HS4, Figure 10), improved its affinity towards MraY.

Figure 12.

(A) Schematic representation of 2D interactions in the docked pose of the MraY_Mtb–ZINC97985272 complex; (B) 3D structure of the docked pose of the MraY_Mtb–ZINC97985272 complex in surface representation.

The thienopyridine moiety of ZINC215762680 (Figures 10 and 13) occupied the uridine pocket of MraY via interactions with Asp188, Phe256, Asn249, and Met257. A surface representation of the MraY_Mtb–ZINC215762680 complex clearly shows the thienopyridine moiety interacting with the uridine pocket (Figures 10 and 12B). Similar to the other top compounds, the pyridine and thiophene rings were involved in a π–π stacking interaction with Phe256. The cyano group of the thienopyridine established two H-bond interactions with Met257 and Asn249, further stabilizing the complex. The carbonyl oxygen, linking the thienopyridine to the thiadiazole moiety, formed a crucial H-bond interaction with Lys105 (HS3, Figure 10).

Figure 13.

(A) Schematic representation of 2D interactions in the docked pose of the MraY_Mtb–ZINC215762680 complex; (B) 3D structure of the docked pose of the MraY_Mtb–ZINC215762680 complex in surface representation.

The 2-oxo-benzo-imidazole moiety of ZINC9672300 exhibited a favorable binding mode with MraY_Mtb, including a key π–π stacking interaction with Phe256 and H-bond interactions with Lys105, Asn182, Asp185, and Asp259 (Figure 14). The benzimidazole moiety occupied the uridine pocket known for MraY catalysis (Figure 10). The hydrophobic tail, attached with the amide, and having a thiophenol group at the terminal carbon, formed strong hydrophobic interactions with the array of hydrophobic residues Val296, Val 297, Ala314, Pro315, and Ile300 (HS4, Figure 10).

Molecular Dynamics (MD) Simulations

During docking calculations, we allowed the ligand to be flexible while the protein was kept rigid; therefore, to understand the dynamical behavior and interaction profiles of the protein-ligand complex, 200 ns MD simulations of the four in silico virtual screening hits were carried out using Desmond software.²⁶ The conformational deviations of the Cα atoms of the protein complexes with the four compounds (ZINC44893147, ZINC97985272, ZINC215762680, and ZINC9672300) can be seen in Figure 15, which shows the root mean square deviation (RMSD) vs. time. The plots show that after about 40 ns of initial fluctuation of the Cα atoms of the protein’s RMSD, the protein reached equilibrium and remained stable throughout the simulation period of 200 ns. Similarly, the RMSD plot of atom locations vs. simulation time for the heavy atoms of the ligands indicates that the ligands maintained a significantly stable state throughout the simulation after the first ~20 ns (Figure 16).

Figure 15.

RMSD plot for Cα atoms of the protein (MraY_Mtb) for the different protein–ligand systems. The protein RMSD was calculated with respect to the reference frame (the starting equilibrated structure of the protein) at time 0 ns.

Figure 16.

RMSD plot for ligand heavy-atoms for the different protein–ligand systems. The ligand RMSD was calculated with respect to the reference frame at time 0 ns.

MraY_Mtb–ZINC44893147 Complex

The dynamic fluctuations of the MraY_Mtb–ZINC44893147 bimolecular system were analyzed by RMSD and root mean square fluctuation (RMSF). The average RMSD of Cα atoms of the protein (MraY_Mtb) and ligand (ZINC44893147) were found to be 2.56 Å and 1.62 Å, respectively. This low RMSD of Cα atoms of the protein indicates that the protein was not fluctuating significantly during the simulation. Similarly, the low RMSD of the ligand suggests that the initial conformation of the ligand was maintained quite stably during the simulation. The RMSF of the protein was calculated to understand the fluctuation of amino acids interacting with ZINC44893147. The amino acids of MraY_Mtb that were involved in the interactions did not show much fluctuation in their RMSF values (Supporting Information, Figure S5), indicating the stability of the biomolecular system. The interaction histograms (Figure 17) and the protein-ligand contact contour map (Figure 18) show that Lys46, Asp102, Asn111 (H-bonding and water bridges), Asp185 (water bridges), Asp188 (H-bonding), Asn249 (H-bonding and water bridges), Phe256 (hydrophobic), and Met257 (H-bonding and water bridges) (Figure 18) were crucial amino acids that formed strong interactions with ZINC44893147. The H-bond contacts and strong π–π interaction (95% contribution) stabilized the complex. The negative average binding free-energy of ZINC44893147 (−55.57 kcal/mol), calculated by Prime MM-GBSA based on the entire trajectory of the MraY_Mtb–ZINC44893147 complex, demonstrates the strength of its binding with the MraY_Mtb protein.

Figure 17.

Simulation Interaction Diagram (SID) plot showing the protein–ligand interactions between the amino acid residues of the MraY_Mtb binding site and ZINC44893147.

Figure 18.

2D diagram of atomic-level interactions of ZINC44893147 with key MraY_Mtb residues during the 200 ns MD simulation.

MraY_Mtb–ZINC97985272 Complex

The average RMSD of Cα atoms of the protein (MraY_Mtb) of the MraY_Mtb–ZINC97985272 complex was found to be 2.24 Å, and Figure 15 shows that the protein reached equilibrium during the simulation. The amino acids of MraY_Mtb that interacted with ZINC97985272 were fairly stable; the RMSF value did not show many fluctuations (Supporting Information, S6). The ligand (ZINC97985272) also maintained a stable conformation throughout the simulation, as indicated in Figure 16, and it had a low average RMSD of 1.88 Å. The stabilization of the complex can be mainly attributed to the strong π–π interaction (86% contribution) between the Phe256 of MraY_Mtb and the phenyl ring of the ligand (Figure 19). The interaction histograms and protein-ligand contact contour map (Figure 20) revealed that amino acids like Asn111 (H-bonding and water bridges), Asp185 (H-bonding and water bridges), Asp188 and Asn249 (water bridges), Phe256 (hydrophobic), and Gly258 (H-bonding and water bridges) also contributed to the stability of the complex. The negative average binding free-energy of ZINC97985272 (−80.62 kcal/mol), calculated with Prime MM-GBSA for the entire trajectory of the MraY_Mtb–ZINC97985272 complex, also supports its strong binding with MraY_Mtb.

Figure 19.

SID plot showing the protein-ligand interactions between the amino acid residues of the MraY_Mtb binding site and ZINC97985272.

Figure 20.

2D diagram of atomic interactions of ZINC97985272 with key MraY_Mtb residues during the 200 ns MD simulation.

MraY_Mtb–ZINC215762680 Complex

The RMSD and RMSF were used to analyze the dynamics and structural behavior of the MraY_Mtb–ZINC215762680 complex. The Cα atoms of MraY_Mtb had an average RMSD of 2.06 Å for the simulation; the protein reached equilibration and remained equilibrated throughout the simulation. Similarly, the ligand (ZINC215762680) did not significantly change its conformation throughout the simulation, as indicated in Figure 16, and it had a low average RMSD (1.07 Å). The compound, ZINC215762680 makes crucial H-bonding and water bridges with Lys46, Ser51, Asp185, and Asp188, hydrophobic interactions with Phe256, and water bridges with Gly258 (Figure 21). These amino acid residues show only minimal fluctuation in their RMSF values (Supporting Information, S7). The MraY_Mtb–ZINC215762680 complex is mainly stabilized by the salt-bridge and strong π–π interaction (74% contribution) (Figure 22). The strong binding of the ligand to the MraY_Mtb pocket is supported by the negative average binding free-energy of ZINC215762680 (−63.41 kcal/mol), calculated with Prime MM-GBSA for the entire trajectory of the MraY_Mtb–ZINC215762680 complex.

Figure 21.

SID plot showing the protein-ligand interactions between amino acid residues of the MraY_Mtb binding site and ZINC215762680.

Figure 22.

2D diagram of atomic interactions of ZINC215762680 with key protein residues of MraY_Mtb during the 200 ns MD simulation.

MraY_Mtb–ZINC9672300 complex

Similar to what is observed with the other compounds, the Cα atoms of the protein (MraY_Mtb) did not deviate much during the simulation of the complex with ZINC9672300. A few spikes were observed in the RMSD vs. time plot for the ligand (ZINC9672300) over the course of the simulation. This can be attributed to the unusual flexibility of the benzimidazole moiety. However, the deviations were well within the acceptable range of 2–3 Å. Although most of the amino acids of MraY_Mtb that interacted with ZINC9672300 did not show much fluctuation in their RMSF values (Supporting Information, S8), a spike was noted in the region between residues 25 to 50. These residues form a part of the loop region. The interaction histograms (Figure 23) and protein-ligand contour map analysis (Figure 24) revealed that the π–π interaction only contributed 20% towards stabilizing the MraY_Mtb–ZINC9672300 complex, unlike its precursors. The amino acids that were crucial for stabilizing the interaction of MraY_Mtb with ZINC9672300 were found to be Lys46 (H-bonding and water bridges), Asp185 and Asp188 (H-bonding and water bridges), Asn249 (water bridges), Phe256 (hydrophobic), Gly258 (H-bonding and water bridges), and Val296 (hydrophobic) (Figure 24). The negative average binding free-energy (−48.61 kcal/mol) for the entire trajectory of the MraY_Mtb–ZINC9672300 complex was also significantly less negative than its counterparts.

Figure 23.

SID plot showing the protein-ligand interactions between amino acid residues of the MraY_Mtb binding site and ZINC9672300.

Figure 24.

2D diagram of atomic interactions of ZINC9672300 with key protein residues of MraY_Mtb during the 200 ns MD simulation.

The Prime MM-GBSA binding free-energies computed after the MD simulation for ZINC44893147, ZINC97985272, ZINC215762680, and ZINC9672300 revealed that ZINC97985272 exhibited the best (most negative) average binding free energy (−80.62 kcal/mol) followed by ZINC215762680 (−63.41 kcal/mol), ZINC44893147 (−55.57 kcal/mol) and ZINC9672300 (−48.61 kcal/mol).

The most negative average binding free-energy for ZINC97985272 had the largest contribution from the van der Waals interactions (VdW) (−78.86 kcal/mol), with other significant contributions from the Lipo term (a measure of hydrophobic interactions with water) (−21.07 kcal/mol), π–π stacking interaction (−8.04 kcal/mol), Coulombic term (Coulomb) or electrostatic interactions (−10.22 kcal/mol), and a minor contribution from the H-bonds (−2.13 kcal/mol). These contributing terms for favored interactions with MraY_Mtb and other hits are presented in Supporting Information Table S2. Overall, all the four potential hits formed stable interactions with the key residues (Asp185, Asp188, Asn249, and Phe256) in the active site of the MraY protein and showed a strong predicted affinity for MraY_Mtb.

Analysis of Calculated ADME/T Profile

The drug-like properties and toxicity profiles of the 15 in silico hits were calculated using QikProp and admetSAR, respectively, and are listed in Table 2. In the initial round of screening of the ZINC15 database of commercially available compounds, we included some ADME filtering, e.g., limiting our search to drug-like compounds with modified Lipinski rules [250 < MW < 800 Da, and LogP < 8) for our virtual screening protocol. We used a “standard” reactivity subset for our database. Most of the known naturally-occurring antibiotics and synthetic antibacterials lie in physicochemical space far outside that of small molecule therapeutics.²² The higher molecular weight and cLopP of antibiotics and antibacterials often help penetrate the bacterial cell walls. Hence, we used modified Lipinski’s rules to preclude exclusion of any potential antibacterials from our virtual screening pipeline. Most of the 15 in silico hits had similar ADME profiles to those of 95% of marketed drugs, but there were a few exceptions. The toxicity properties of the hits were predicted with good probability values using admetSAR. Hits 4 and 5 are the only compounds predicted to be toxic in the Ames test, while the remaining hits were non-Ames toxic.

Table 2.

Physicochemical properties (ADMET) of the top 15 predicted hits, calculated using QikProp and admetSAR.

Compound	QikProp Properties											admetSAR
	MWa	PSA (Å2)†	Rot. Bonds	QplogPoct (octanol/water)‡	QplogS§	QPPCaco (nm s−1)¶	QPPMDCK (nm s−1)Γ	Lipinski Rule-of-5 violations	Qplog Khsaτ	% Human Oral Absorption #	QPlogHERGΔ	Ames test	Carcinogenicity
ZINC000044893147	411.47	177.24	7	0.49	−4.24	12.22	16.19	0	−0.42	49.32	−5.109	Non-Ames toxic (0.520)	NCb (0.985)
ZINC000097985272	681.81	144.85	18	5.21	−6.33	125.60	54.44	2	0.32	69.09	−7.220	Non-Ames toxic (0.600)	NC (0.714)
ZINC000215762680	377.46	151.14	7	0.58	−4.22	5.53	9.77	0	−0.30	43.63	−5.98		Non-Ames toxic (0.570)
ZINC000100167112	552.76	116.53	20	6.63	−7.23	634.62	302.6	2	1.15	90.2	−5.77	Ames toxic (0.525)	NC (0.914)
ZINC000004945499	298.28	141.42	3	−0.19	−3.78	24.77	16.39	0	−0.45	50.81	−4.09	Ames toxic (0.520)	NC (1.000)
ZINC000002038342	260.65	138.44	2	−1.11	−1.68	32.29	23.23	0	−0.71	47.43	−3.05	Non-Ames toxic (0.810)	NC (0.620)
ZINC000017026645	467.50	178.69	6	1.79	−5.93	14.36	5.94	1	0.15	45.22	−6.09	Non-Ames toxic (0.540)	NC (0.857)
ZINC000001239561	232.23	142.54	0	−0.88	−1.40	56.13	48.28	0	−0.64	53.12	−1.85	Non-Ames toxic (0.700)	NC (0.729)
ZINC000008426376	470.67	86.09	14	6.74	−8.38	720.52	498.26	1	1.42	100.00	−6.13	Non-Ames toxic (0.600)	NC (0.929)
ZINC000059387046	307.21	150.02	2	0.00	−3.19	33.44	102.11	0	−0.59	54.25	−4.09	Non-Ames toxic (0.680)	NC (0.914)
ZINC000022204432	246.19	174.97	0	−2.96	0.47	1.51	2.49	0	−1.27	12.82	−1.43	Non-Ames toxic (0.690)	NC (0.843)
ZINC000585277591	321.37	111.97	5	0.73	−1.68	150.81	130.65	0	−0.66	70.21	−2.64	Non-Ames toxic (0.620)	NC (0.929)
ZINC000097813818	373.45	85.51	7	1.47	−1.75	298.57	148.19	0	−0.43	79.87	−5.31	Non-Ames toxic (0.550)	NC (0.914)
ZINC000013345246	337.30	135.78	5	0.97	−2.49	21.84	10.19	0	−0.69	56.59	−3.01	Non-Ames toxic (0.690)	NC (0.727)
ZINC000009672300	480.55	144.82	9	1.85	−5.32	43.94	27.11	0	−0.39	67.18	−7.75	Non-Ames toxic (0.630)	NC (0.732)

^aRecommended range, based on that of 95% of marketed drugs: Molecular weight (130–725 Da)

^†Van der Waals surface area of polar nitrogen and oxygen atoms (7–200 Ų)

^‡Predicted octanol/water partition coefficient (−2 to 6.5)

^§Predicted aqueous solubility, log S (−6.5 to 0.5)

^¶Predicted apparent Caco-2 cell permeability (<25 nm s⁻¹ poor; >500 nm s⁻¹ great)

^ΓPredicted apparent MDCK cell permeability (<25 nm s⁻¹ poor; >500 nm s⁻¹ great)

^ΔPredicted IC₅₀ value for blockage of HERG K+ channels (it is a serious concern if below −5)

^τPrediction of binding to human serum albumin (−1.5 to 1.5 acceptable)

^#Predicted qualitative human oral absorption (>80% is high; <25% is poor), NC = non-carcinogenic

Conclusions

In this work, we carried out in silico protein structure-based screening of the compounds in the ZINC15 database using a well-validated homology model of MraY_Mtb to discover small non-nucleoside MraY_Mtb inhibitors. We adopted a two-pronged strategy, standard docking and e-pharmacophore-based screening, to ensure the robustness of the final hit molecules. Our efforts resulted in the identification of 15 potential small non-nucleoside inhibitors of MraY_Mtb. These potential inhibitors are structurally distinct from the known and reported non-nucleoside inhibitors of MraY. Three top-ranking hits and one lower-ranking hit were selected for 200 ns MD simulations to understand the dynamic behavior and ligand-interactions profile of these hits complexed with MraY_Mtb.

The MD results indicate that the compounds fit well into the active site and formed stable and strong interactions with the key residues Asp185, Asp188, Asn249, and Phe256 of MraY_Mtb. The strongly negative average binding free-energy for all four compounds received contributions from the van der Waals interactions, the Lipo term, H-bonds, π–π stacking interaction, and Coulombic interactions. In the future, we plan to test the selected 15 predicted hit compounds to determine their MraY_Mtb inhibitory activities, and if they are found to be significant, the compounds will also be tested for their ability to inhibit Mtb growth and survival.

Materials and Methods

Template Selection and Sequence Alignment

We used an experimental X-ray crystal structure of MraY from Aquifex aeolicus (PDB ID: 5CKR, MraY_AA, available in the RCSB Protein Data Bank, which is 32% identical and 50% similar to MraY_Mtb, and has >91% sequence coverage) as a template, to build the homology models of MraY_Mtb (UniprotKB code P9WMW7). The ClustalW²³ algorithm was used to align sequences of target/templates that were manually edited to avoid sequence gaps in the conserved and transmembrane (TM) regions (Figure 2).

Homology modeling and validation

Computational models of MraY_Mtb (UniprotKB code P9WMW7, strain ATCC 25618/ H37Rv) were constructed utilizing a homology modeling approach with the Modeller software.²⁴ MD2, the co-crystallized ligand of 5CKR, along with its three conserved water molecules were retained during our model generation. Based on the DOPE score and Ramachandran plot, we selected the five best MraY_Mtb homology models, which showed most of the residues in the favored region (Figure 3). The best homology models were further validated through docking of known available MraY_Mtb inhibitors.

Enrichment study

A database of MraY inhibitors was built from 15 previously papers^25–39 that reported an IC₅₀ value for MraY inhibitory activity. The compounds having an IC₅₀ ≤1 μM were classified as actives. The cutoff criterion yielded 155 active compounds. The second dataset of 1922 drug-like compounds with an average molecular weight of 400 Daltons was downloaded from the Schrödinger website. These compounds, which we assumed to be inactive, were used to form the decoy set. A total of 2049 compounds was used for the enrichment study. Docking was carried out using the Standard Precision (SP) docking method in Glide⁴⁰ software with flexible ligand sampling. The OPLS3e (optimized potential for liquid simulations 3) force field was used in our calculations.

Ligand preparation

A thorough literature search was conducted to determine MraY inhibitors that have good activity against Mtb. We narrowed the list down to three ligands, MD1, CPZ-B, and capuramycin. These compounds have good anti-TB activity (against the H₃₇R_V strain) (1.56 μg/mL, 3.13 μg/mL, and 6.25 μg/mL, respectively).⁴¹ The ligands were prepared in LigPrep using the OPLS3e (optimized potential for liquid simulations 3) force field.⁴² Ionization states were generated using Epik⁴³ at physical pH (7.4).

Generation of docked poses

The grid for the 6OYH structure was generated using the OPLS3e force field using default parameters. The three prepared ligands, MD1, CPZ-B, and capuramycin, were docked into the grid using default parameters. Although MD1 and CPZ-B docked well into 6OYH with docking scores of −9.88 kcal/mol and −9.49 kcal/mol, respectively, the docking pose of capuramycin was not similar to that of the experimental X-ray crystal structure of MraY from Aquifex aeolicus in complex with capuramycin (PDB ID: 6OYZ). To resolve this issue, we further carried out induced-fit⁴⁴ docking of capuramycin with 6OYH using carbacaprazamycin as the reference ligand. One of the best poses was further refined using the Prime refinement⁴⁵ module (which implements minimization Monte Carlo). The refined docked pose of capuramycin overlaid well with the experimental 6OYH–co-crystallized ligand. These three complexes were used for the generation of e-pharmacophore hypotheses.

Our efforts to dock the three ligands to the MraY_Mtb homology model using SP docking and induced-fit docking protocol proved challenging. Capuramycin could be docked to the receptor using a rigid SP docking protocol. However, for the flexible CPZ-B ligand, we had to devise a different strategy. The carbacaprazamycin ligand was extracted from the experimental X-ray crystal structure of 6OYH and was merged with the aligned MraY_Mtb homology model. The resulting complex was further refined using minimization Monte Carlo to obtain an energy-minimized complex. Subsequently, an induced-fit docking of CPZ-B was carried out with the homology model using the MraY_Mtb–carbacaprazamycin complex as a reference. Gratifyingly, the strategy was fruitful to generate the desired pose of MraY_Mtb–CPZ-B.

Despite our best efforts, MD1 could not be docked into the MraY_Mtb homology model; hence we adopted a different strategy. Since MD1 is an analog of MD2, which is methylated at its C-2” hydroxy of the amino ribosyl moiety (Figure 6), we decided to modify the 2D structure of the MD2 ligand of the MraY_Mtb homology model to form MD1, using the 3D-build suite of Schrödinger. The modified MraY_Mtb–MD1 complex was further refined using the Prime refinement module (minimization Monte Carlo). The complex was then directly used for the e-pharmacophore hypothesis generation.

Development of e-pharmacophore models

Pharmacophore sites were manually generated with PHASE⁴⁶ using the docked poses. The default set of six chemical features, hydrogen bond acceptor (A), hydrogen bond donor (D), hydrophobic (H), negative ionizable (N), positive ionizable (P), and aromatic ring (R), were used to derive the important pharmacophore features. Hydrogen bond acceptor sites were represented as vectors along the hydrogen bond axis as per the hybridization of the acceptor atom. Hydrogen bond donors were represented as projected points, located at the corresponding hydrogen bond acceptor positions in the binding site. Projected points allowed the possibility for structurally dissimilar active compounds to form hydrogen bonds to the same location, regardless of their point of origin and directionality.

Protein-structure and e-pharmacophore based virtual screening

We screened in silico the commercially available, drug-like compounds in the ZINC15 database with modified Lipinski rules [250 < MW <800 Da, and LogP < 8) for their potential binding with the MraY_Mtb receptor model using the GOLD suite,¹⁸ considering its Chemscore function. We applied the early termination option during VS if the RMSD difference for the first three poses of each ligand was < 1.5 Å.

Next, a two-pronged Glide docking approach was used in the selection procedure to identify more robust hits. In the first approach, the 15,201 hits from GOLD docking were further screened against the MraY_Mtb homology model. Since the experimental X-ray crystal structure of MraY_AA from Aquifex aeolicus in complex with carbacaprazamycin (PDB ID: 6OYH) could dock large molecular weight compounds with flexible tail regions with the desired pose, unlike other MraY crystal structures, we also screened the 15,201 hits against 6OYH. The compounds were prepared using the ligand preparation module, and their energies were minimized using the OPLS3e force field. Ionization states were generated using Epik at physiological pH (7.4). Receptor grids were created for the 6OYH and MraY_Mtb homology model using the receptor grid generation tool in the Glide application of the Schrödinger suite. The 15,201 prepared ligands were then docked on the 6OYH structure and MraY_Mtb homology model using the SP scoring function (Glide SP), recording the three top-scoring poses for each ligand. The van der Waals radius-scaling factor and partial charge cutoff were kept to the default settings. No additional constraints were used for docking.

In the e-pharmacophore approach, six hypotheses were generated based on three well-precedented MraY inhibitors, CPZ-B, capuramycin, and MD1, with the PHASE module implemented in the Schrödinger software. The top-ranked 15,201 ligands obtained from GOLD docking were separately aligned with the generated hypotheses to find matching pharmacophores in the ligands. The screened ligands were required to match a minimum of 5 out of the 9 or 10 sites that were generated in the six e-pharmacophore hypotheses. The ligands were then ranked based on their fitness score that measures the alignment of the conformer with the established hypothesis. All ligands that had a fitness score < 1 were removed because they were less likely to fit into the receptor pocket than the known reference ligands (CPZ-B, capuramycin, and MD1).

Binding energy calculations

The Prime Molecular Mechanics/Generalized Born Surface Area (MM-GBSA) approach was used to calculate ligand-binding energies for the MraY_Mtb–CPZ-B, MraY_Mtb–capuramycin, MraY_Mtb–MD1, 6OYH–CPZ-B, 6OYH–capuramycin, and 6OYH–MD1 complexes using a VSGB2.0 solvent model with the OPLS3e force field.

Clustering

Overall, 581 ligands were found in common from the two screening methods, Glide docking and e-pharmacophore hypothesis screening. To get access to unique scaffolds, we generated a linear 2D fingerprint for each of the 581 hits using the Canvas module of Schrödinger and performed similarity clustering using the Hierarchical method.

Molecular dynamics simulation

The MD simulations were carried out using the Desmond module of Schrödinger suite 2019.²⁶ The MraY_Mtb in complex with one of the final hits was embedded in a POPC (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine) bilayer and solvated with an 11 Å TIP3P⁴⁷ water buffer using the OPLS_2005 (optimized potentials for liquid simulations) force field implemented in Desmond, Schrödinger. The system was neutralized by adding chloride ions as needed and 0.15 M NaCl was added to the system. The system was equilibrated using the previously published protocol.^48,49 In brief, the system was simulated for 1 ns using Brownian dynamics in the NVT ensemble at 10 K with the restraint of 50 kcal/mol on solute heavy atoms. Secondly, a 300 ps simulation was run in the NVT ensemble using the Berendsen thermostat (10 K) while retaining the restraint on solute heavy atoms. Thirdly, a 300 ps simulation was run in the NPT ensemble using the Berendsen thermostat (10 K) and barostat (1 atm) while restraints were retained. Over the next 300 ps, the system was gradually heated to 300 K. A final 5 ns simulation was performed in which all restraints were removed. The NPγT ensemble with a temperature of 300 K and a pressure of 1 bar was applied in all the simulations. The simulation length of the production run was 200 ns. The OPLS_2005 force field parameters were used in all simulations. The long-range electrostatic interactions were calculated using the particle mesh Ewald method. The cutoff radius for Coulomb interactions was 9.0 Å. The Langevin coupling schemes were used for the pressure and temperature controls used for the 200 ns production run. Nonbonded forces were calculated using the RESPA integrator and the trajectories were saved at 13.3 ps intervals for analysis. The dynamical behavior and interactions between the ligand and protein were analyzed using the Simulation Interaction Diagram tool implemented in the Desmond MD package. The stability of the MD simulations was monitored by looking at the RMSD of the ligand and protein atom positions in time and by RMSF.

In silico ADME/T profile

The predicted absorption, distribution, metabolism, and excretion (ADME) properties of the potential MraY_Mtb inhibitors were calculated using the QikProp module (QikProp, version 12.3.013, Schrödinger, LLC, New York, NY, 2020) implemented in the Schrödinger software. The important toxicity parameters were calculated using admetSAR version 2.0.⁵⁰ The QikProp module predicts pharmaceutically relevant properties of the given molecules, and also provides the recommended values for the properties based on the analysis of 95% of marketed drugs used in their training set. These recommended values are used for the comparison. The potential hits were preprocessed using LigPrep. The QikProp calculation was run in normal mode. In addition, the toxicity profile was calculated using an online server of admetSAR software, with SMILES notation as input.

Abbreviations

AA

Aquifex aeolicus

BEDROC

Boltzmann-enhanced discrimination of the receiver operating characteristic

C₅₅-P

Undecaprenyl phosphate

CB

Clostridium bolteae

CPZ-B

Caprazamycin B

DOPE

Discrete optimized protein energy

EF

Enrichment factor

EMB

Ethambutol

INH

Isoniazid

MD

Molecular dynamics

MD1

Muraymycin D1

MD2

Muraymycin D2

MDR

Multi-drug-resistant

MraY

Phospho-MurNAc-pentapeptide translocase or translocase I

Mtb

Mycobacterium tuberculosis

MurNAc

N-Acetylmuramic acid

NIAID

National Institute of Allergy and Infectious Diseases

NIH

National Institutes of Health

OPLS

Optimized potentials for liquid simulations

PG

Peptidoglycan

PNPT

Polyribonucleotide nucleotidyltransferase

POPC

1-Palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine

PZA

Pyrazinamide

RIF

Rifampin

RMSD

Root mean square deviation

RMSF

Root mean square fluctuation

ROC

Receiver operating characteristic

SID

Simulation interaction diagram

SP

Standard precision

TB

Tuberculosis

UDP

Uridine diphosphate

UM5A

UDP-MurNAc-pentapeptide

WHO

World Health Organization

XDR

Extensively-drug resistant

XSEDE

Extreme Science and Engineering Discovery Environment