Rational design of drug-like compounds targeting Mycobacterium marinum MelF protein

The mycobacterial mel2 locus (mycobacterial enhanced infection locus, Rv1936-1941) is Mycobacterium marinum and M. tuberculosis specific, which can withstand reactive oxygen species (ROS) and reactive nitrogen species (RNS) induced stress. A library of over a million compounds was screened using in silico virtual ligand screening (VLS) to identify inhibitors against the modeled structure of MelF protein expressed by melF of mel2 locus so that M. marinum’s ability to withstand ROS/RNS stress could be reduced. The top ranked 1000 compounds were further screened to identify 178 compounds to maximize the scaffold diversity by manually evaluating the interaction of each compound with the target site. M. marinum melF was cloned, expressed and purified as maltose binding protein (MBP)-tagged recombinant protein in Escherichia coli. After establishing the flavin dependent oxidoreductase activity of MelF (~ 84 kDa), the inhibitors were screened for the inhibition of enzyme activity of whole cell lysate (WCL) and the purified MelF. Amongst these, 16 compounds could significantly inhibit the enzyme activity of purified MelF. For the six best inhibitory compounds, the minimal inhibitory concentration (MIC) was determined to be 3.4–19.4 μM and 13.5–38.8 μM for M. marinum and M. tuberculosis, respectively. Similarly, the minimal bactericidal concentration (MBC) was determined to be 6.8–38.8 μM and 27–38.8 μM against M. marinum and M. tuberculosis, respectively. One compound each in combination with isoniazid (INH) also showed synergistic inhibitory effect against M. marinum and M. tuberculosis with no cytotoxicity in HeLa cells. Interestingly, these inhibitors did not display any non-specific protein-structure destabilizing effect. Such inhibitors targeting the anti-ROS/RNS machinery may facilitate the efficient killing of replicating and nonreplicating mycobacteria inside the host cells.

Introduction Tuberculosis (TB), the leading public health problem stands along with human immunodeficiency virus (HIV) as the top most killers worldwide among the infectious diseases [1]. In 2015, 10.4 million new TB cases were estimated worldwide, of which 1.2 million people (11%) were HIV-positive [2]. TB epidemic has been exacerbated further by the emergence of multi-, extensively-and totally-drug resistant strains and there is an urgent need to devise new drugs. The most widely used strategy to identify drug targets is the whole-cell screening (WCS), which has been successfully used for the identification of new drug candidates such as diarylquinolines (TMC207) that target ATP biosynthesis, and benzathiazines (BTZ043) that target cell-wall arabinan biosynthesis [3,4]. The major disadvantage of WCS approach is the lack of precise target knowledge, which may result in the selection of unspecific toxic compounds [5]. Moreover, the compounds might be optimized for targets required only under in vitro culture conditions but are dispensable in vivo. Genome-derived, target-based approach has also been widely used to identify new anti-TB drugs [6], but it does not ensure the druggability. Virtual ligand screening (VLS) has been established as an efficient approach for drug design with a 100-1000 fold higher enrichment rate in comparison to conventional chemical screening [7,8]. Though this strategy is robust, it still requires human input in the form of identification of a druggable pocket, diversification of interacting moieties and the use of homology modeled structures [7,9].
We identified a novel M. marinum and M. tuberculosis specific mel2 locus, which can withstand enhanced ROS and RNS stress [10][11][12] and has been suggested to play a role in M. tuberculosis lipid biosynthesis [13]. Among six gene of mel2 locus identified in M. marinum and M. tuberculosis by in silico analysis; melF, melG and melH are close homologs of luxA, luxG and luxH genes involved in bioluminescence [12]. The mel2 locus confers resistance to ROS/RNS stress in laboratory media and has a similar role in activated macrophages [11,12]. The MelF protein is a putative flavin dependent oxidoreductase and the mutation in melF displays a polar effect on the downstream genes of mel2 locus [11]. The increased susceptibility of melF mutant to ROS/RNS could be partially retrieved by melF alone and completely by the entire mel2 locus [11]. The role of mel2 locus in resistance to ROS has been demonstrated for the persistence and dissemination of M. tuberculosis in C57BL/6 mice [14]. The oxidative stress has also been shown to enhance the susceptibility of mycobacterial cells to existing anti-TB drugs such as INH, clofazamine, etc. [15,16]. However, the development of new drugs is hampered by the slow growth of M. tuberculosis. Alternatively, a fast-grower, M. marinum is considered as a surrogate model to study M. tuberculosis pathogenesis [10,17] and the feasibility of using M. marinum as an anti-TB activity evaluation model has also been described [17]. Therefore, a study was designed to identify putative inhibitors by VLS analysis targeting MelF that can diminish the ability of M. marinum to withstand ROS/RNS stress and the results were validated by evaluating the flavin oxidoreductase activity of MelF protein and the growth inhibition/killing of bacteria. from www.chembridge.com. 178 drug-like compounds were purchased from ChemBridge, San Diego, USA for in vitro screening.

Bacterial strains and growth conditions
The departmental committee of THSTI, Faridabad has given approval to work on M. marinum strain M in biosafety-hood and M. tuberculosis H 37 Rv in biosafety level-III facility at the International Centre for Genetic Engineering and Biotechnology, New Delhi. M. marinum strain M and M. tuberculosis H 37 Rv were grown at 33˚C for~7 days and at 37˚C for~3 weeks, respectively in 7H9 Middlebrook medium supplemented with 0.5% glycerol, 10% albumin dextrose complex (ADC) and 0.05% Tween-80 (M-ADC-TW) as detailed earlier [10,14]. The mycobacterial colony-forming units (CFU) were determined on 7H10 (M-ADC) agar [10,14]. E. coli XL1 Blue and BL21 strains were grown in Luria-Bertani (LB) medium at 37˚C.

Homology modeling, validation and druggable pocket definition
Sequence of MelF (Accession Number: gi|54289551.1) was PSI-BLASTed against PDB, and 2WGK with 36% sequence identity and 98% sequence coverage was found [19]. Yet Another Scientific Artificial Reality Application (YASARA) Dynamics and Structure (YASARA Biosciences GmBH, Vienna, Austria) (v 10.12.1), software was used for homology modelling and structural refinement ( Fig 1A). The loops were built appropriately using Loop-Builder module of YASARA. The structure was refined using YASARA Dynamics and Structure in a sequential process to remove minor conformational strains by simulating the protein for 500 ps. The refined structure of MelF was validated by WHATIF server (S3 Table). The simulation was further continued using AMBER-03 force field for 8 ns. The average structure was created from snapshots from 5.2 to 8 ns time frame when the modeled structure reached a stable phase. The process of averaging the atomic coordinates also yielded the information about relatively rigid and flexible parts ( Fig 1C). The alternative structures were combined into a single representation of the ensemble to account for large scale conformational changes ( Fig 1B). The conformational changes were calculated as Root Mean Square Fluctuation (RMSF) as detailed below.
Where T = number of frames and V is coordinate set of backbone The most deviant regions are in between 150-175 residues ( Fig 1C). This region (150-175) shows high root mean square fluctuation (rmsf) thus suggesting its less ordered structural configuration. Interestingly, the structural information for residues 175-180 is missing from the crystal structure of pdb entry 5AEC (5AEC replaces PDB entry 2WGK) probably this stretch is highly flexible. None of residues from these flexible regions were part of the docking site. Similar approach was also adopted in our earlier work and by other researchers to process the PDB crystal structure of C2 domain [20,21]. The model was further validated by Whatcheck (S3 Table). The Ramachandran plot for 3D structure of MelF was also drawn (S4 Fig). The validated structure was used to identify a druggable pocket/site using Pocket-finder, Q-site finder [22] and InCaSitefinder [23]. The top-scoring pockets were scrutinized to identify a pocket with better druggability index. One such pocket on MelF was shortlisted as the best druggable pocket.

Compound filtration and VLS
We used a bimodal approach for screening drug-like compounds, which involved a consensual approach based on DOCK [24] and Vina [25] docking methods. The ChemBridge Corp. screening collection of~1 million molecules (2013-2014) was selected for in silico screening. These were filtered using FAF-Drugs2 [26] with the hard-filters. The remaining~250,000 compounds that satisfied the ADMETox filters were used for the docking studies. These sdfformatted files were converted to PDB format using Open Babel 2.2.3 [27]. Following conversion to PDB format, files containing individual compounds were created and processed using MGLTools. DOCK 3. We used DOCK 3.6 to screen non-toxic drug-like compounds against the modeled structure of MelF. Each ligand pose was scored as the sum of the receptor-ligand electrostatics (which was corrected for desolvation of ligand). In order to improve charge complementarity between the enzyme and docked molecule, the absolute magnitude of the partial atomic charges of several active site atoms were increased by 0.4 units (the net residue charges were unchanged): Tyr268OH and HOH; Thr272Oγ1 and HOγ and Thr185Oγ1 and HOγ. This method has already been used by Babaoglu et al [28]. Conformations were generated using concord. Ligand charges and initial solvation energies were calculated using AMSOL. The top 1000 molecules that were present in ranked lists of both methods were selected as consensual putative inhibitors. Among these, 178 molecules were shortlisted for experimental testing based on their docking score (only top-scoring compounds were tested), scaffold diversity and our assessment of how well they fit in the targeted site.
Cloning and expression of M. marinum MelF M. marinum strain M genomic DNA sequence (GenBank accession number NC_010612) was used to design primers 5'-ATG GAT CCA TGA TGG AGA TCG GAA TAT TCC TCA TGC-3' (forward primer) and 5'-CAA GCT TTC ACT TAT CGT CGT CAT CCT TGT AAT CGA CGA TCG CGG CTT CG-3' (reverse primer) for PCR amplification of melF. The pMAL-c5x vector that carries a MBP-tag [29] was used to clone and express M. marinum melF in E. coli. The PCR amplified melF from M. marinum genomic DNA and pMAL-c5x were subjected to restriction digestion with BamH1 and HindIII at 37˚C and ligated overnight at 16˚C with T4 DNA Ligase. The melF::pMAL-c5x plasmid DNA was transformed into E. coli XL1 Blue and selected on LB agar plates with ampicillin at 100 mg/L. The confirmed clone was transformed to E. coli BL21, inoculated in LB-glucose medium with ampicillin at 100 mg/L and incubated at 37˚C for 3-4 h, till the optical density reached~0.6 at 600 nm. The culture was induced with 1mM IPTG at 37˚C for 3 h [29]. The over-expression profile of MBP-tagged MelF protein and its solubility was confirmed by SDS-PAGE. The expression was further confirmed by Western blot analysis using rabbit anti-MBP primary (1:10000) and HRP-conjugated anti-rabbit secondary (1:10000) antibodies [29].
(CB) (20mM Tris-Cl, pH 7.4, 200mM NaCl, 1mM EDTA). After the addition of lysozyme and phenylmethylsulfonyl fluoride at a final concentration of 1mg/mL and 1mM, respectively, the bacterial cells were kept on ice for 30 min, followed by sonication and centrifugation at 10,000 rpm for 30 minutes at 4˚C. The supernatant (WCL) was collected, applied to amylose column for 3-4 h at 4˚C and washed with CB. The protein was eluted with an elution buffer (10mM maltose in CB) by collecting 15-20 fractions at 4˚C and the presence of purified MelF protein in the collected fractions was checked with 10% SDS-PAGE. The fractions containing MelF were pooled and concentrated to 1mg/mL by Amicon Ultra Centricon with a cutoff value of 30 kDa at 4˚C. The purified enzyme was run on HPLC using C-18 column (250 X 4.60 mm) (Shimadzu, Japan) with acetonitrile:water (90:10) as the mobile phase and a flow rate of 0.5 mL/min. The protein was detected at A 280 using UV-detector. To rule out non-specific mechanism of the compounds as protein aggregators, 100 μg of purified MelF protein was incubated with 30 μM of each inhibitor in 100 μL reaction mixture at 12˚C for 15 min and then loaded on native-PAGE, to see, any shift of protein band.

Flavin oxidoreductase assay
The enzyme activity was assayed by monitoring a decrease in absorption of NADH (nicotinamide adenine dinucleotide) at 340 nm (ε = 6.22 X 10 3 M -1 cm -1 ) in 50 mM Tris-Cl, pH 7.4 at 25˚C. The steady state kinetic measurements were performed in a 1-cm path length cuvette in a final volume of 1mL according to the method of Kendrew et al. [30] with slight modifications. 300μM of NADH and 80μM of FAD (flavin adenine dinucleotide) were added and the assay was initiated with the addition of 100 μg (6.09× 10 −10 mole) of enzyme present in WCL/ purified MelF. The reaction was followed at 25˚C for 5 min. The FAD was kept in the dark to prevent photo-reduction. The specific activity of the enzyme was calculated as Units=mg ¼ DA 340=min 6:22 Â mg enzyme= mL reaction mixture Where ΔA 340 represents a change in absorbance of NADH at 340 nm.
V max and K m values were determined using different substrate concentrations for 100 μg of purified enzyme by plotting the non-linear regression curve. k cat was calculated from the formula: Where k cat = Turnover number (maximum velocity per active site), E t ðmmoleÞ ¼ WtðgÞ X 10 6

Screening of inhibitors based on flavin oxidoreductase assay
The inhibitors were dissolved in DMSO to a final concentration of 100 μM in the reaction mixture. The final DMSO concentration in the reaction mixture was~0.3-0.5% and DMSO control was also run (0.5% DMSO, no other additives) to rule out non-specific inhibitory effect. A preliminary screening of 178 inhibitors was carried with the IPTG induced WCL and 20 shortlisted compounds were further tested with the purified MelF. The percent enzyme activity in presence of inhibitors was calculated as ΔA compound = change in absorbance in presence of inhibitor and ΔA control = change in absorbance of control, the enzyme activity of WCL/purified protein in absence of inhibitor. The enzyme activity of uninduced WCL was subtracted from the induced WCL to get the actual enzyme activity. To prevent auto-oxidation of NADH, all the tests were carried out in cuvettes covered with parafilm. A control was also run in which no enzyme was added to ensure that the decrease in absorbance was not due to self-oxidation of NADH.

K i (inhibitory constant) determination
K i values were calculated using Dixon plots, where inverse of V max values were plotted against the inhibitor concentration using different substrate concentrations.

MIC and MBC determination
MIC of 20 inhibitors was determined for M. marinum and M. tuberculosis by a high throughput resazurin microtiter plate assay (REMA). Briefly, 100 μL of M-ADC-TW broth was dispensed/well of sterile 96-well microtiter plate (Nunc) and serial two-fold dilutions of each inhibitor were prepared directly in the plate from 1 mg/mL stock solution. To it, 100 μL of bacteria (~5 X 10 6 CFU/well) and 10 μL of resazurin solution (5 mg/mL) were added. The inhibitor control (no inhibitor) and sterile medium control were also included. The plates were incubated at 33˚C for 48 h, and 37˚C for 96 h, for M. marinum and M. tuberculosis, respectively. The concentration of inhibitors that prevented a change in color from purple to pink was taken as an indicator to determine the MIC value [32]. The in vitro growth interaction of bacteria with the inhibitors and the first line drugs such as rifampicin (RIF)/INH was determined by resazurin drugs combination microtiter assay [33]. Two-fold serial dilutions ranging from 12.5 mg/L to 0.04 mg/L for RIF/INH were combined with the same concentration of inhibitor (ranging from 12.5 mg/L to 0.04 mg/L) in wells of microtiter plates followed by the addition of bacteria and resazurin solution as described above. The fractional inhibitory concentration index (FICI) was used to evaluate the synergistic effects. The FIC index (FICI) was calculated as follows: The effects of the drug combinations were classified as synergistic (FICI 0.5), additive or no interaction (FICI > 0. , and antagonistic (FICI > 4) as described by Pagliotto et al [33].
The MBC microtiter plates were prepared similar to REMA plates, but no resazurin was added [32]. For MBC determination, the plates were incubated at 33˚C for 48 h, and 37˚C for 96 h, for M. marinum and M. tuberculosis, respectively. Then, 100 μL from each well of plates was transferred to a plate containing 100 μL/well of inhibitor-free M-ADC-TW broth. Tenfold serial dilutions from each well of MBC plate were plated on 7H10 (M-ADC) agar for CFU enumeration. The MBC was defined as the lowest concentration that resulted in killing of ! 99% of the viable bacteria in comparison with the starting inoculums. The inhibitor was considered bactericidal when the MBC/MIC ratio was 4 [32]. To determine the enhanced bactericidal effect, combinations of drugs (RIF/INH) and inhibitors were used as stated above. Briefly, two-fold serial dilutions ranging from 12.5 mg/L to 0.04 mg/L for RIF/INH were combined with the same concentration of inhibitor in wells of microtiter plates followed by the addition of bacteria as described above. The enhanced bactericidal effect was considered when the MBC/MIC ratio of combination was lesser as compared to RIF/INH alone. The MIC/MBC experiments were run in triplicate to ensure the reproducibility.

Cytotoxicity assay
The MTT (3-(4, 5-dimethyl thiazol-2yl)-2, 5-diphenyl tetrazolium bromide) assay was employed to assess the cytotoxicity of the inhibitors [34,35]. HeLa cells (Human cervical epithelial cells) were plated at a concentration of 1×10 5 /well in 100 μL of minimal essential medium (MEM) supplemented with 10% heat-inactivated fetal bovine serum (FBS) in 96-well plates. After overnight incubation at 37˚C in 5% CO 2 incubator, media was removed and the inhibitors were added in 100 μL of MEM + 10% heat-inactivated FBS at 1X and 5 X MIC concentrations. After incubation for 20 h, the cells were washed thrice with warm MEM (without FBS) and 200 μL of MTT (5mg/mL) was added. The plates were further incubated for 4 h, followed by the addition of 50 μL of DMSO (solubilizing reagent) per well, mixed thoroughly by a micropipette and left for 45 s. Presence of viable cells was visualized by the development of purple color due to formation of formazan crystals. The OD values were read at 595 nm using DMSO as a blank. A standard graph was plotted between the concentration of inhibitors and the relative cell viability.

Statistical analysis
The student's t-test was applied to find the difference in the mean enzyme activity ± SD of control and those treated with inhibitors (done in duplicate). V max and K m values were calculated along with 95% confidence interval (CI) by plotting a non-linear regression curve and the analysis was done using Stata 14.2. The P < 0.05 was considered statistically significant.

Results
Homology modeling of M. marinum MelF structure 2WGK is a Type II Baeyer-Villiger monooxygenase from Pseudomonas putida [19], which has sequence coverage of~98% and sequence identity of~36%, with M. marinum MelF protein.
Since there was a high sequence similarity between the MelF protein and 2WGK, we reasoned that a reliable homology model could be constructed. 2WGK is an FMN-dependent dimeric enzyme; therefore, M marinum MelF was modeled as a dimeric structure. Following an equilibration, the three-dimensional (3D) modeled structure of MelF protein was refined by allatom molecular dynamics simulation by optimizing its energy parameters and removing the structural strains (Fig 1B and 1C). The average MelF structure ( Fig 1A) thus derived was checked by Whatif server for its quality. The average structure was found to be in good harmony with the set parameters for a good structure with RMSD in bond angles 1.509 (S3 Table). A consensus approach was later used to find the best druggable pocket in this structure. The largest pocket providing the most number of contact atoms (630 atoms) and a site volume of 2093 Å 3 were selected for VLS study (Fig 2, S2 Fig). In addition, the sequence of M. marinum MelF was compared with M. tuberculosis H 37 Rv MelF, which showed~98% sequence identity (S3 Fig). Therefore, the proteins encoded by both the sequences are likely to be similar in structure and function independent of their origin.

Virtual screening
The ChemBridge Core library and Express-Pick Library are two mutually exclusive libraries of commercially available small drug-like compounds at the Hit2Lead.com online store of Chem-Bridge site. We selected these libraries as they provide a large set of diverse groups of compounds, which is prerequisite to carry out a large scale VLS. 1000 top ranking compounds obtained from Vina (semi-flexible VLS) and DOCK were pooled together. The docked compounds were later manually evaluated to choose molecules and shortlisted for experimental testing based on their docking score (only top-scoring compounds were tested), our assessment of how well they fit the site and their chemical diversity. Diversified set of interacting molecules were clustered and ranked based on structural, energetic and interaction mode parameters. There was a massive overall reduction in the complexity of the chemical space. Among the top six top compounds, compound # 5175552 made multiple interactions with MelF. N atom of Asp208 formed a hydrogen bond with the O2 of the compound; NE2 atom of His45 formed a hydrogen bond with N3 of the compound and the multiple hydrophobic interactions occurred between Phe7, Met9, Ala76, Ile111, Leu207 and Phe273 with the aromatic rings of compound # 5175552. Similarly, compound # 6492687 exhibited numerous identical contacts vis-à-vis compound # 5175552. Of the 14 amino acids that interacted with compound # 9125618, 10 amino acids (Phe7, Met9, His45, Glu51, Leu207, Val240, Glu242, Phe273, Ley331 and Tyr268) were common with the interacting residues of compound # 5175552. (Fig  3A, 3D and 3F). The compound # 5255825 and # 5255829 were structurally similar and this was also reflected somewhat in the binding modes (Fig 3B and 3C). The docked poses of compounds displayed partial occupancy at the same site as that of FAD/FMN (Fig 4A-4F). As per the docking poses, compound # 5255825 interacted with Phe7, Met9, Glu51, Val240, Leu207, Met278, Phe273, Val269, Tyr268, Thr272 and Leu331. About 80% contacts between compound # 5255829 and MelF residues were similar to those observed with compound # 5255825. Interactions made by compound # 9125618 and # 651374 were primarily hydrophobic in nature (Fig 3E and 3F). MelF residues interacting with compound # 6492687 and # 6513745 considerably overlapped amongst themselves as 8 out of 13 amino acids that interacted with compound # 6513745 also interacted with the compound # 6482687. Interestingly, in silico poses of the compounds exhibited a predominance of van der Waals interaction with Drug-like compounds and M. marinum MelF MelF (Fig 3). Moreover, as per the docking results, the compound # 6492687 and # 6513745 overlapped considerably with the FAD/FMN interacting site as compared to compound # 5255825 and # 5255829 (Fig 4B-4E).

Cloning, expression and purification of MelF protein
The melF clones were tested via fragment release by BamH1 and HindIII (S1  (Figs 5A and 6B). The soluble nature of the protein was established with SDS-PAGE ( Fig 5A) as most of the expressed protein was present in the WCL (soluble fraction) and~40% of the total bacterial protein was MelF. Western blot analysis revealed a clear visible band with WCL and the whole bacterial suspension ( Fig 5B) in a chemiluminescent assay thus confirming that the over-expressed protein was MBP-MelF fusion protein. Since MBP has a strong affinity for tighter binding to amylose, MelF protein was purified using MBP's affinity for maltose in a one-step purification process thus indicating that MelF was properly translated and folded in E. coli. The purified MelF ( Fig  6A) could display a sharp peak at 280 nm by HPLC. To rule out non-specific mechanism of the compounds as protein aggregators, we performed native-PAGE analysis in the presence and absence of inhibitory compounds (Fig 6B). Notably, the inhibitors did not promote protein aggregation as there was no shift in the bands of MelF in their presence.

Enzyme kinetics and screening of inhibitors
The specific activity of flavin oxidoreductase of purified MelF was somewhat higher in comparison to WCL ( Table 1). The low difference between the specific activities of enzyme from WCL and the purified form could be due to the presence of other oxidoreductases present in WCL with a similar reaction. V max and K m of the purified MelF were determined to be  Drug-like compounds and M. marinum MelF 422.36 μmole/min/mg (95% CI, 301.7-543 μmole/min/mg) and 0.155 mM (95% CI, 0.05-0.26 mM), respectively, whereas k cat value of the purified MelF was determined to be 6.94 X 10 5 min -1 (Fig 7).
While screening a total of 178 compounds with WCL, 20 were shortlisted, which displayed significant inhibition (P < 0.05) in the enzyme activity ranging from 22% to 90% (Fig 8). Such compounds were further tested to inhibit enzyme activity of the purified MelF and that was found to be greater than WCL. Out of 20 compounds, 16 were found to significantly inhibit (P < 0.05, Fig 8) the enzyme activity of purified MelF (38-98% inhibition). On the other hand, the compound # 5255827 did not inhibit (P > 0.05) the enzyme activity of purified MelF (with residual enzyme activity of~100%), though its inhibitory effect was observed in WCL. As the total protein concentrations as well as the compound concentrations in WCL and the purified MelF were equal, therefore, the effective concentration of MelF in WCL was relatively lower than the purified MelF. The compound concentration in WCL was probably adequate to inhibit the MelF activity (owing to less effective concentration of enzyme); however, this effect of compound was abolished when the enzyme concentration was increased (as expected in the purified MelF). Another possibility could also be that the compound # 5255827 might be inhibiting oxidoreductases other than MelF in WCL. Moreover, the virtual screening is not 100% accurate implying that some of the inhibitors designed might be interacting with the target(s) other than MelF.  In addition, three compounds (# 7732624, # 7362803, # 7928509) that earlier showed an enzyme inhibition with WCL could not display such inhibitory effect with the purified MelF. In fact, these compounds displayed an enhanced absorbance values at 340 nm in comparison to control. Perhaps, the increase in absorbance does not necessarily mean stimulation of the enzyme activity as these compounds might be getting altered during oxidation of NADH to NAD + thus leading to enhanced absorbance. However, in-depth investigation is required to rule out this possibility.
The intersection of Dixon's plot at all values of [S] were not on [I] axis for inhibitor # 5175552, # 5255825, # 5255829 and # 6492687 (Fig 9), which suggests competitive inhibition. On the contrary, intersection of these plots for inhibitor # 6513745 and # 9125618 was on [I] axis at [I] = -K i thus suggesting noncompetitive inhibition. Furthermore, inhibitor # 5255829 and # 5255825 displayed K i value of 9 μM and 4 μM respectively. These were followed by inhibitors # 6513745, # 9125618 and # 6492687 with K i values of 18, 24 and 25 μM respectively, whereas inhibitor # 5175552 was the least effective with K i value of 68 μM (Fig 9, S5 Fig).
The secondary structure of protein was determined using its respective CD spectra. The αhelices and β-sheets were almost constant with and without inhibitors (Fig 10, S2 Table) thus suggesting that these compounds are not denaturants. The six best compounds were further tested for MBC determination by the conventional CFU assay ( Table 2) (Table 3). It was found that three compounds (# 5175552, # 5255825 and # 6492687) showed synergistic inhibitory effect in combination with INH against M. tuberculosis with  (Tables 2 and 3) as two-fold lesser MBC/MIC ratio was observed in comparison to RIF alone. The cytotoxic effect of top six compounds was examined in HeLa cells by MTT assay (Fig  11), which revealed that three compounds (# 5175552, # 5255829 and # 6513745) were not cytotoxic at both 1X and 5X MIC concentrations, whereas # 9125618 revealed~18% cytotoxicity at 5X MIC concentration. On the other hand, two compounds (# 5255825, # 6492687) were cytotoxic at both the concentrations. It is pertinent to mention that out of three synergistic compounds identified for M. tuberculosis (Table 3), two showed cytotoxicity in HeLa cells; whereas one out of two synergistic compounds identified for M. marinum showed cytotoxicity in HeLa cells. Finally, compound # 5175552 and # 6513745, in combination with INH displayed synergistic inhibitory effect against M. tuberculosis and M. marinum, respectively with no cytotoxicity in HeLa cells. Interestingly, # 5175552 also displayed enhanced bactericidal effect (with MBC/MIC ratio of 1) in combination with RIF against M. tuberculosis. The chemical structures of six best compounds are shown in Fig 12 and their corresponding docked poses are depicted in (Fig 4A-4F). The docking scores and the basic molecular information about these inhibitors are given in S1 Table.

Discussion
This study has led to the identification of relatively potent small molecule inhibitors for M. marinum MelF using VLS. A model for 3D structure of MelF protein was generated and its geometry-optimized surface scrutiny by multiple methods revealed a deep pocket with very Drug-like compounds and M. marinum MelF high druggability index. Interestingly, a new and improved structure for Type II Baeyer-Villiger monooxygenase (5AEC) from P. putida has replaced the earlier 2WGK structure [19] after our VLS work was completed. However, a careful scrutiny reveals a very close similarity between the earlier 2WGK and the recently reported 5AEC structure (RMSD is~0.15). In addition, a new structure of FMN bound enzyme (4UWM) has been submitted by the same group [19]. The occupancy of the cofactor modeled in the active site of newly submitted structure (4UWM) was~0.6 thus indicating low occupancy of cofactor in the active site. This is also reflected in the poor electron density due to low affinity of FMN for Type II Baeyer-Villiger monooxygenase. A comparison of 4UWM and our docked compounds (Fig 4A-4F) also showed a partial overlap with the binding region of FMN (4UWM). In fact, the targeted druggable pocket is a juxtaposition of both cofactor (FMN) and the substrate binding sites. Notably, in silico analysis of the docked poses of our inhibitory compounds exhibited a dominance of van der Waals interaction, which has also been reported earlier [36,37] wherein the proteincompound interaction interface was deeply buried. However, an experimental validation by a crystal structure of MelF could have provided better explanation, but that is not determined till date.
VLS involves flexible-ligand docking into the pocket using the consensual approach among DOCK and Vina, which has also been successful in identifying ligands for several mycobacterial targets such as PknG (Rv1827), InhA (Rv1484), MtCM (Rv1885c), MurE (Rv2158c), DHQase (Rv2537c), L-AlaDH (Rv2780), UMPK (Rv2883c), EthR (Rv3855) [7,8,[38][39][40][41][42][43] etc. In addition, the TB structural genomics consortium (TBSGC, http://www.webtb.org) determines the 3D structures of proteins from M. tuberculosis for rational structure-based drug designing [44]. The TBSGC is also actively developing bioinformatics resources that can be used for data mining to complement the structural information. Many of M. tuberculosis protein structures determined by TBSGC are associated with the metabolic pathways and are suggested as potential attractive anti-TB drug targets, which include structural studies on urease Drug-like compounds and M. marinum MelF (Rv1848), chorismate-utilizing enzymes, arginine biosynthesis enzymes, crotonase, malate synthase (Rv1837c) and phosphoenolpyruvate carboxykinase (Rv0211), etc. [44][45][46][47]. Moreover, the atomic resolution studies of proteins from systems involving disulfide bond formation of the secreted proteins and toxin-antitoxin gene pairs can also be utilized for the structure-based drug design [44]. In this study, a series of compounds were predicted to dock into the active site and the shortlisted compounds met the Lipinski's rule of 5 for drug-likeliness [26]. Of 178 putative inhibitors shortlisted,~9% (16/178) were found to significantly inhibit the flavin oxidoreductase activity of purified MelF, whereas 1.68% (3/178) showed an MBC/MIC ratio of 2 for M. marinum and M. tuberculosis with no cytotoxic effect in HeLa cells. This further underlines the efficiency of VLS for the identification of novel protein activity modulators vis-a-vis chemical high throughput screening, whose success rate is~0.01% [20].
As observed in bioluminescent bacteria, M. marinum mel2 also facilitates the detoxification of ROS and RNS produced by the activated macrophages thus proposing that the bioluminescence systems might have been evolved from the oxidative defense mechanisms [11,12]. The in silico analysis of mel2 locus has demonstrated a remarkable similarity to bioluminescent lux genes of Vibrio harveyi and the melF of mel2 locus displays a high similarity to luxA, which has been shown to play a significant role in resistance to ROS in bioluminescent bacteria [12]. The presence of a lux-like operon, mel2, in non-bioluminescent pathogens of humans, that is, M. tuberculosis and M. marinum suggests that it could play a crucial role in mycobacterial pathogenesis [13,14], therefore, novel inhibitors against MelF flavin oxidoreductase activity were designed in the study and their bacteriostatic/bactericidal activity was examined against M. marinum and M. tuberculosis in vitro. Using M. marinum as an anti-TB activity evaluation model, Liu et al. [17] could identify an inhibitor targeting ICL (Rv0467) against both active and non-replicating M. tuberculosis. In fact, zebrafish M. marinum infection model has already been reported for investigating anti-TB agents in vivo as a high-throughput screening system [48].
Similar to mycobacterial MelF flavin oxidoreductase, M. tuberculosis dezaflavin-dependent nitroreducase (Ddn, Rv3547) and its two homologues Rv1261c and Rv1558 (which encode for an F 420 H 2 -dependent quinone reductase function, thus leading to the production of dihydroquinones) has been demonstrated to provide protection to mycobacterial cells from the oxidative stress and bactericidal agents [49]. Notably, the M. tuberculosis F 420 -deficient Ddn mutants defective in the formation of dezaflavin were found to be hypersensitive to INH, clofazimine and moxifloxacin thus proposing that the inhibitors that can inhibit Ddn or dezaflavin biosynthesis could synergize with the existing anti-TB drugs, as observed in this study also. These findings indicate the utility of exploiting MelF as a target for designing the inhibitors that might enhance the susceptibility of mycobacterial cells to oxidative stress thus leading to bacterial killing. We also speculate that the partial inhibition of MelF protein by inhibitors could possibly lead to accumulation of toxic intermediates thus resulting into bacterial killing.
Interestingly, Dwyer et al. [50] hypothesized that all bactericidal antibiotics kill bacteria by generating ROS/RNS, which has been demonstrated for the drugs such as delamanid, PA-824 and clofazimine for their activity against M. tuberculosis [15,16]. Perhaps, this explains the rational for the evolution of enzymes such as MelF flavin oxidoreductase and Ddn in imparting protection to mycobacterial cells against ROS/RNS. Further supporting the role of ROS in M. tuberculosis killing, Grant et al. [51] documented that even a small change in dissolved oxygen concentration (20%) could affect killing of mycobacterial persisters, a subpopulation of bacteria that is phenotypically resistant to killing by most of the antibiotics but still sensitive to high quantities of hydroxyl radicals. Notably, one inhibitor each for M. marinum and M. tuberculosis displayed synergistic inhibition in combination with INH, and two inhibitors displayed an enhanced bactericidal effect for M. tuberculosis in combination with RIF, with no cytotoxic effect in HeLa cells. The possible mechanism for such effects could be the triggering of ROS/RNS (produced by INH/RIF), which becomes more acute with the simultaneous blocking of MelF activity (with anti-ROS/RNS machinery) by the inhibitors. These findings suggest a need to further explore new drug-like molecules with the synergistic inhibitory effect and that may improve the efficacy of existing anti-TB drugs.
In addition to MelF flavin oxidoreductase and Ddn, other metabolic enzymes of M. tuberculosis have also been suggested as putative drug targets, which include lipoamide dehydrogenase (Lpd) and dihydrolipoamide acyltrasferase (DlaT), the components of pyruvate dehydrogenase [52] and M. tuberculosis peroxynitrite redutase/peroxidase function to resist RNS generated by the host cells [53]. By targeting the Lpd and DlaT, Bryk et al. [54,55] identified inhibitors that were selectively active against nonreplicating M. tuberculosis. Similarly, inhibitors of the respiratory chain and ATP synthesis have also been reported [56], which are probably the best-validated M. tuberculosis drug targets due to clinical success of the ATP synthase inhibitor TMC207 (bedaquiline) and the first-line TB drug pyrazinamide, which disrupts the proton motive force [57].
We anticipate that the inhibitors inhibiting the flavin oxidoreductase activity of M. marinum MelF would also inhibit M. tuberculosis MelF activity. Several researchers have targeted similar enzyme systems of E. coli and M. tuberculosis. For example, drugs such as sulfonylureas, imidazolinones and benzoyl esters, the traditional inhibitors of E. coli acetohydroxyacid synthases (AHAS) also inhibit M. tuberculosis AHAS [58]. Furthermore, the use of M. tuberculosis 3-isopropylmalate dehydrogenase (IPMDH) has been exploited as a drug target [59], which has homologues in other bacterial species such as E. coli and Thermus thermophilus and the inhibitory effect of IPMDH inhibitors was found to be independent of the origin of IPMDH. Similarly, the inhibitors targeting M. marinum MelF in the study also inhibited M. tuberculosis in vitro independent of the origin of MelF.
In conclusion, we cloned, expressed and purified M. marinum MelF protein. This is the first report that establishes the flavin dependent oxidoreductase activity of M. marinum MelF. The potential inhibitors targeting MelF were identified through in silico analysis, which inhibited the flavin oxidoreductase activity of MelF and showed bacteriostatic/bactericidal activity against M. marinum and M. tuberculosis in vitro. Interestingly, two inhibitors also displayed synergistic inhibitory effect in combination with INH against M. marinum/M. tuberculosis with no cytotoxic effect in HeLa cells. Further work is in progress to examine the effect of these inhibitors in human and mouse activated macrophages infected with M. marinum and M. tuberculosis.