A library of 5,674 compounds from ChemDiv (diversity set) and 400 compounds from Prestwick (F2L library) were assembled for the screening, arrayed in 96-well plates at concentration of 20 mM in DMSO and stored at -20°C. When available, the hits selected for dose-response validation were repurchased from the same suppliers. The 7,8-dihydro-D-erythro-neopterin (DHNP) substrate for MtbFolB was purchased from Sigma-Aldrich. The 6-hydroxymethyl-7,8-dihydropterin hydrochloride (HP) substrate and the 7,8-dihydroxanthopterin (DHXP) reaction by-product were both purchased from Schircks Laboratories.

The coding region of MtbFolB (Rv3607c; NCBI RefSeq GeneID: 885345) was amplified by PCR using genomic DNA from Mtb H37Rv as a template, using primers 5’-TTTTCCATGGCTGACCGAATCGAA-3’ (Forward) and 5’-TTTTAAGCTTCATACCGCGCCGC-3’ (Reverse), and cloned in pET28a(+) vector using the NcoI/HindIII restriction sites (underlined in the primer sequences). The resulting plasmid pET28::folB, allowing over-expression of the native MtbFolB without tags, was transformed into competent E. coli BL21 Star (DE3) cells (Invitrogen). Transformants were selected on LB agar plates supplemented with 50 μg/mL kanamycin. Single colonies were grown overnight at 37°C in LB medium (50 mL) supplemented with 50 μg/mL kanamycin. Part of the culture (8.5 mL) was inoculated into 500 mL of LB medium supplemented with 50 μg/mL kanamycin and grown at 37°C in shaker at 180 rpm until reaching an optical density (OD_{600 nm}) of 0.4-0.6. The protein expression was induced by addition of 0.1 mM isopropyl-β-D-thiogalactoside (IPTG, Sigma-Aldrich) for 6 hours at 37°C. The cells were then harvested by centrifugation at 11,800 × g for 30 min at 4°C. Approximately 2 g of cell paste was resuspended in 10 mL of buffer A (25 mM Tris-HCl pH 8.0, 50 mM NaCl, 5% glycerol) containing a protease inhibitor cocktail tablet (Complete EDTA-free, Roche Diagnostics) and gently stirred for 30 min in the presence of 0.2 mg/mL lysozyme (Sigma Aldrich). Cells were subsequently disrupted by sonication with a 500W sonicator equipped with a 1/2” diameter probe (Qsonica), using ten pulses of 10s, with cooling intervals of 1 min, at 60% amplitude and the lysate was centrifuged at 38,900 × g for 30 min at 4°C. The supernatant was treated with 1% (w/v) streptomycin sulphate (final concentration) for 30 min to precipitate nucleic acids and ribonuclear proteins (23) under slow agitation and centrifuged at 38,900 × g for 30 min at 4°C. After centrifugation, the supernatant was dialyzed against buffer A, and loaded onto a DEAE Sepharose CL6B anionic exchange column (GE Healthcare) equilibrated with buffer A. The column was washed with 3 column volumes (CV) of buffer A and adsorbed proteins were eluted by a linear gradient (0-1M NaCl) with 14.3 CV in buffer B (25 mM Tris-HCl pH 8.0, 1M NaCl, 5% glycerol) at a flow rate of 1 mL/min. The fractions eluted from anionic exchange column were incubated with ammonium sulfate to a final concentration of 1M and loaded on a Butyl Sepharose High Performance HiLoad 16/10 (GE Healthcare) equilibrated with buffer C (25 mM Tris-HCl pH 8.0, 1M ammonium sulfate, 5% glycerol). The column was washed with 7 CV of buffer C and the adsorbed proteins were eluted by a linear gradient (1 to 0M ammonium sulfate) with 20 CV in buffer C at a flow rate of 1 mL/min. The fractions containing the enzyme, first identified by the absorbance (OD_{280 nm}) and confirmed by SDS-Page, were pooled, concentrated using an Amicon Ultra centrifugal filter with a 10 kDa cut-off (Millipore) and loaded on a HiLoad Superdex 200 26/60 size exclusion column (GE Healthcare) previously equilibrated with buffer A. Protein was isocratically eluted with 1 CV of buffer A at a flow rate of 0.3 mL/min. Fractions containing pure FolB were pooled and concentrated to 0.4-0.6 mg/mL, aliquoted and flash-frozen for conservation at -80°C.

All enzymatic reactions were conducted in 384-well plates (Greiner, Cat #781900) using buffer A (25 mM Tris-HCl pH 8.0, 50 mM NaCl, 5% glycerol), in presence of 5 μM 7,8-dihydroneopterin (DHNP, Sigma-Aldrich) and 10 ng of recombinant MtbFolB (14 nM) in a total volume of 50 μL. Prior optimization experiments indicated that these concentrations of enzyme/substrate yielded reproducible end-points, while maintaining substrate saturation and minimizing the amount of enzyme needed per well. Incubations were done at room temperature (24°C) for 4h and the end-point fluorescence reading obtained with a multimode microplate reader (Envision or EnSight, Perkin Elmer) equipped with excitation 405 nm and emission 535 nm filters. Each assay plate included 4 full columns of controls: 2 columns for the positive control (1% DMSO) and 2 columns for the negative control (heat-inactivated MtbFolB), to assess the assay performance. For the negative control, an aliquot of the pure MtbFolB preparation used for the assay was taken and inactivated at 95°C for 10 min the same day.

For the screening, a fresh powder aliquot of substrate (DHNP) was added to buffer A to reach a concentration of 5 mM and solubilized by water-bath sonication for 1h in the dark. This stock was further diluted in buffer A to prepare a working solution at 12.5 μM. Stock solution of compounds at 20 mM in 100% DMSO were extemporaneously diluted at 2 mM in 100% DMSO and 0.5 µL of this intermediate dilution was added to the assay plate, which was pre-filled with 10 μL of buffer A. Using an automated dispenser, the substrate (20 μL at 12.5 μM; 5 μM final concentration) and the enzyme (10 ng in 20 μL; 14 nM final concentration) solutions were immediately added, for a total of 50 μL per well. The compounds were thus at a final concentration of 20 μM and a residual DMSO concentration of 1%. The fluorescence values obtained for each compound were normalized with that of the controls for the corresponding plates, using the average value for the positive control (1% DMSO) as reference for 0% inhibition and the average value of the negative control (heat-inactivated MtbFolB) as reference for 100% inhibition.

For determination of compound activity by dose-response, a similar procedure was followed using compound concentration starting at 100 μM and a total of 10 doses with 2-fold dilution between each dose, with 1% DMSO final concentration in each well. Each dose-response was performed as a technical duplicate and both data pooled and used together for calculations. Raw fluorescence values were plotted as a function of the compound concentration, expressed in log scale, and the data fitted against a sigmoidal dose-response model (4 parameters) using the software Prism v6 (GraphPad). The concentration required to inhibit the enzyme activity by 50% (IC₅₀) was obtained as the best fit value for the IC₅₀ parameter when the fitting converged and the 95% CI are reported. When necessary (absence of low plateau), constraints were added to the bottom value of the dose-response, constraining the value of this parameter above or equal to that of the negative control (inactivated enzyme). In such cases (no plateau), the 95% CI values were omitted (interval too large).

Enzyme inhibition studies were performed using a RF-5301 spectrofluorometer (Shimadzu), monitoring an increase in fluorescence, corresponding to the formation of HP, with excitation wavelength at 365 nm and fluorescence emission at 525 nm. The slits were 10 nm for excitation and 15 nm for emission. Reactions were conducted as previously described (20), during 6 min at 25°C using buffer A (25 mM Tris-HCl pH 8.0, 50 mM NaCl, 5% glycerol), in presence of 1.5 μM DHNP (non-saturating concentration) and 300 nM of recombinant MtbFolB in a total volume of 500 μL (quartz cuvette). The DMSO concentration was fixed at 10% and the maximal rate of enzymatic reaction (100% of MtbFolB activity) was determined in these conditions, in absence of inhibitor. As controls, buffer, substrate, inhibitor and enzyme spectra were performed under the same conditions to subtract fluorescence intensities that were not coming from the reaction product. Reaction velocities (μM.min^-1) were determined from the fluorescence kinetic measurement by fitting the data points (20) to a straight line by linear regression, using compound concentrations ranging from 50 μM to 1 mM. For each compound, velocities were then plotted as a function of concentration and the plot fitted by non-linear regression using the equation below:

in which V _i and V ₀ are, respectively, the reaction velocity in the presence and in the absence of inhibitor (I).

Prior to activity determination, MtbFolB (300 nM) was incubated at room temperature in buffer A together with the compound (3, 9 or 20). At the corresponding time-point (0, 30 and 60 min), the substrate was added, and the enzyme velocity determined as described above (kinetic assay). Different incubation and test conditions were used while maintaining the concentration of MtbFolB fixed at 300 nM. i) Normal condition: MtbFolB was incubated with the compound at 300 μM and 2% DMSO, then the velocity measured in presence of the compound (300 μM), 1.5 μM of substrate and 2% DMSO. ii) Excess of substrate: MtbFolB was incubated with the compound at 300 μM and 2% DMSO, then the velocity measured in presence of the compound (300 μM), 3 μM (2×) of substrate and 2% DMSO. iii) Excess of inhibitor: MtbFolB was incubated with the compound at 3 mM (10×) and 15% DMSO, then the velocity measured in presence of the compound (300 μM) and 1.5 μM of substrate with DMSO at a final concentration of 2%.

All procedures for the synthesis of the hit of interests and their derivatives, along with NMR and mass spectrometry characterization data, are available in the Supplementary Methods of this article.

All docking procedures were performed using AutoDock4.2 program. To ensure that all compounds were properly docked, a 3D-grid with dimensions 50 × 50 × 50 with spacing of 0.375 Å was used to define the active site. The grid was defined in the 6-hydroxymethyl-7,8-dihydropterin (HP) binding region, which is located at the interface of two subunits, based on the octameric structure of MtbFolB obtained by crystallography (PDB: 1NBU) (24). A re-docking procedure was performed to evaluate whether the program could reproduce the ligand location found in the structure. The Lamarckian Genetic search Algorithm was used with 60 runs, and the other parameters were set to their default values, except for the number of evaluations and the number of individuals in population, which were set to 4,000,000 and 450, respectively. Docking results were prepared using PyMOL v2.5.3 (DeLano Scientific).

The activity of the compounds 3, 9, 13, 16, 20, 21, 22, and 23 was evaluated against M. tuberculosis H37Rv reference strain (ATCC 27294) using a colorimetric assay based on resazurin to determine their minimum inhibitory concentration (MIC), as previously described (25). Briefly, bacteria were seeded in 96-well plates at an OD_{600 nm} of 0.006 and incubated with the compounds at 37°C for 7 days (200 µL per well total). The resazurin solution (60 µL of 0.01%) was then added and the plates further incubated at 37°C for 2 days. A change in color from blue to pink indicated bacterial growth, and the MIC was visually defined as the lowest drug concentration that prevented the color change. Isoniazid (MIC = 2.3 µM) and rifampicin (MIC = 0.2 µM) were used as controls in this experiment. The MIC determination procedures were carried out in Difco™ Middlebrook 7H9 broth (Becton Dickinson – BD) supplemented with 10% (v/v) BBL™ Middlebrook ADC enrichment (albumin, dextrose and catalase – BD) at a final concentration of DMSO equal to 2.5%. A higher concentration of DMSO was chosen to favor compound solubilization at the highest concentrations. The compounds were evaluated using 8-point, 2-fold serial dilutions, with the maximum concentration being 200 µM for the compounds 3, 9, 16, 20, 21, 22, 23 and 20 µg/mL (equivalent to 55.4 µM) for the compound 13. Three independent experiments were conducted for each compound (each experiment included a single dose-response), and the MIC was considered as the most frequent value among the three obtained.

Based on previous work performed using FolB from S. aureus (19) and Mtb (20, 21), we established an in vitro enzymatic assay for the evaluation of compound libraries arrayed in 384-well plates. The assay is based on the fluorescence of 6-hydroxymethyl-7,8-dihydropterin (HP), the kinetic product formed by action of MtbFolB on the substrate 7,8-dihydroneopterin (DHNP). The fluorescence of the mixture containing the substrate alone was found to be stable, even after 20h incubation ( Supplementary Figure S1A ). The thermodynamic product, 7,8-dihydroxanthopterin (DHXP), has a weaker fluorescence intensity at the wavelength used ( Supplementary Figure S1B ) and is formed at a smaller rate (21), allowing for a stable end-point measurement in a time-window from 3h and up to 20h ( Supplementary Figure S1C ). The 4h time-point was selected for the screening to ensure maximum conversion of the substrate while minimizing formation of DHXP.

The assay developed was then used to screen a small library of diverse molecules (total 6,074 compounds) and the fluorescence results normalized to that of the controls on a plate-by-plate basis ( Figure 2 ). Hits were selected based on a 30% inhibition threshold, yielding 19 compounds that were selected for dose-response studies ( Supplementary Figure S2 ). These molecules could be grouped into 5 independent clusters (numbered I-V), except for 4 compounds that were singletons. Hits were re-purchased or re-synthesized and tested at 10 different doses in duplicate using the same assay, with concentrations starting at 100 μM. Results are summarized in Table 1 . Based on these confirmation data, clusters I, IV and V were eliminated due to lack of activity. The activity of the singletons 2, 14 and 18 were confirmed, but we could not re-test compound 17 (not available to re-purchase). Clusters II and III were also found to be active and we decided to prioritize our efforts on the characterization of these two clusters.

The 3 active hits for this series (3, 9 and 16) had IC₅₀ values ranging from 18 to 47 μM in the dose-response assay. Interestingly, 4 other compounds from the initial screening library displayed similar structures and, although not selected from the primary screening, had percentages of inhibition close to the 30% threshold. These 4 compounds were re-purchased and tested in dose-response, revealing that all were active with IC₅₀ values within the same range as the 3 selected hits ( Table 2 ). Kinetic inhibition studies also indicated a dose-dependent inhibition of the reaction rate ( Supplementary Figure S3 ), but higher doses of compounds were required to see a reduction in velocity during the first minutes of the reaction, as shown by the higher IC₅₀ values obtained ( Table 2 ). Time-dependent inhibition studies conducted with compounds 3, 9 and 20 indicated that the inhibition of the reaction rate was stable in time for at least 1h, reaching 40 to 60% inhibition at 300 μM ( Supplementary Figure S4 ). Pre-incubating the enzyme with higher doses of compounds (up to 3 mM) did not affect the inhibition of the reaction, nor did higher substrate concentrations ( Supplementary Figure S4 ).

Because MtbFolB is able to transform the initial substrate into 3 different species (including 2 reversibly) (21), complex experiments would be needed to conclude on the inhibition mechanism. Here, we assumed that the compounds could bind within the active site (formed at the junction between 2 monomers of MtbFolB) and conducted docking studies to explore their mode of binding. The estimated free energies of the top docking poses are indicated in Table 2 . Interestingly, not all 7 derivatives adopted a similar pose and two distinct modes of binding were observed. In the first pose (compounds 9, 16, 22 and 23), the carbonyl of the pyrazol-3-one core is engaged in hydrogen-bonding with the tyrosine and lysine of the active site (Tyr54D, Lys99A; A, D refer to the MtbFolB subunit: two are necessary to form an active site), as well as π-π stacking interactions with Tyr54 ( Figure 3A ; Supplementary Figure S5 ). The N-carbonyl side-chains were found to interact either through hydrophobic contacts (9, 22) or hydrogen-bonding (16, 23) with nearby residues. In the second pose however (compounds 3, 20 and 21), only the N-carbonyl side-chains were found to interact with the Tyr54D of the active site, through π-π stacking interactions ( Figure 3B and Supplementary Figure S5 ). Although the first pose is likely to induce inhibition of the catalytic activity, the second pose appears to have less affinity for the catalytic residues, which may lead to an easy displacement of the compound by the natural substrate. The existence of this second pose, less productive, may explain the low inhibition potential of the compounds from this cluster. Indeed, no compound in this series showed inhibitory activity against M. tuberculosis (data not shown).

The second series included 2 promising derivatives, namely 5 and 13, displaying IC₅₀ values of 2.6 and 15.4 μM, respectively. Considering the relative ease of access to these compounds and other derivatives (1 step reaction with the appropriate reagents), we decided to draw a preliminary structure activity profile of this series of compounds. In a first step, we explored the effect of varying the substituent connected to the sulfur atom of the sulfonamide (S-series, Table 3 ). Our results indicated that the activity could be retained for a variety of substituents. In particular, the pyrazole 25 and the isoxazole 27 both showed good activities (IC₅₀ values of 6.8 μM and 9.2 μM, respectively) and the longer, flexible diphenyl ether chain of 30 could also be tolerated, although the resulting inhibition was more modest (IC₅₀ value of 31.7 μM). In a second step, we explored the substituents on the amide side (N-series) and noted that variations at this position might be more constrained ( Table 4 ). While the piperidine 32 had slightly better activity compared to the reference secondary amine (IC₅₀ value of 11.3 μM against 15.4 μM), the primary amine 31 was less active (IC₅₀ value of 37.8 μM) and the imidazole 33 together with the amide 34 were both inactive. Substituting the piperidine led to either decrease (35, 36) or loss (38) of the activity. Replacing the secondary amine with an isopropyl also led to the loss of activity, as seen in compound 37. Interestingly however, the phenyl piperazine-linked bi-sulfonamide 39 maintained a good activity (IC₅₀ value of 10.6 μM), indicating that this type of bi-sulfonamide could present a viable strategy for further optimizations.

Similarly to the compounds of the pyrazol-3-one series (Cluster II), kinetic and time-dependent inhibition studies were initiated for the sulfonamide series. Despite reasonable computed logP and polar surface area (TPSA, Ų) values (< 3 and< 120, respectively; Supplementary Table S1 ), solubility issues appeared for most compounds of this family preventing studies at concentrations exceeding 20 to 40 µM, which was insufficient to complete the required experiments. We thus decided to focus instead on docking studies in order to gain more insight into the potential mode of binding of these compounds in the active site of MtbFolB. We chose to include compounds 5, 13, 27, 32 and 39 as they were among the most active derivatives identified for this series. Compounds 24 and 34, which were found to be inactive, were also included for comparison. As shown in Figure 4A , compounds 13, 32 and 39 aligned on the same pose within the active site and displayed several interactions with the tyrosine and lysine of the active site (Tyr54D, Lys99A), through π-π stacking interactions with Tyr54D and both the benzyl and thiophene rings on each side of the sulfonamide, as well as hydrogen-bonding between the Lys99A and the oxygen of the sulfonamide (distance ~3.0 Å). While the S-moiety of the sulfonamide lied within the catalytic pocket, the N-moiety rested at the entrance of the cavity, lined by alkyl residues Val55D and Ile104A ( Figure 4B ). Compound 27 had a similar orientation but displayed a different pose in the active site, so that the Lys99A was apparently not engaged in the interaction ( Supplementary Figures S6A , S7 ). Compound 5 in contrast showed an opposite orientation, with the S-moiety pointing towards the outside, similarly to the two inactive compounds 24 and 34 ( Supplementary Figure S6 ). The presence of two hydrophobic and bulky amino-acids at the entrance of the active site (Val55D and Ile104A) may explain the lower tolerance for variation on this side of the molecule, in particular the loss of activity seen for the quite rigid and polar imidazole 33 and the rigid and bulky 4-isopropylbenzyl 37. It is more complicated however to understand why the amine 31 and amide 34 are inactive solely based on these docking studies. Molecular dynamic experiments would be required to better appreciate the interactions within the active site. Other parameters might also have to be considered to explain this lack of activity, like non-specific binding elsewhere in the MtbFolB tetramer prior to the formation of the octamer, which is expected to occur upon substrate binding (24).

The main challenge with target-based screening is to identify compounds that remain active against the bacteria, especially in the case of Mtb. The results of the MIC study indicated that compound 13 showed a modest inhibitory activity against the bacillus, with a MIC of 55.4 µM. This was quite interesting as, to our knowledge, this is the first compound directed against MtbFolB that is also found to display an antimycobacterial activity. Although more experiments will be needed to more precisely evaluate the effect this compound can have on the bacteria, notably with the addition of additional clinical strains of Mtb, this finding nonetheless paves the way for further medicinal chemistry efforts to improve the antimycobacterial potency of this lead compound 13.

In summary, we have developed a high-throughput screening assay directed against MtbFolB, in an effort to identify new scaffolds that could prove effective at inhibiting this essential enzyme. We present here the results of a small 6,074-compound screening, together with the validation campaign performed on the selected hits and their newly synthesized derivatives. Through this work, we established that the assay was functional and able to select small molecules with inhibitory activities, opening prospects for larger scale screening targeting MtbFolB. We also highlighted compound 13 as a potential starting point for further studies, as it was also found to be active against Mtb in contrast to all the other hits reported in this work and elsewhere in the literature for MtbFolB inhibitors.