Crystal structure of dihydrofolate reductase from the filarial nematode W. bancrofti in complex with NADPH and folate

Lymphatic filariasis is a debilitating illness with an estimated 50 million cases as of 2018. The majority of cases are caused by the parasitic worm W. bancrofti and additional cases by the worms B. malayi and B. timori. Dihydrofolate reductase (DHFR) is an established target in the treatment of cancer, bacterial, and protozoal infections and may be a potential target for drugs targeting parasitic worm infections, including filariasis. Recent studies have shown that known antifolate compounds, including methotrexate, inhibit the activity of W. bancrofti DHFR (WbDHFR). However, the absence of structural information for filarial DHFRs has limited the study of more in-depth structure-function relationships. We report the structure of WbDHFR complexed with NADPH and folate using X-ray diffraction data measured to 2.47 Å resolution. The structure of WbDHFR reveals the usual DHFR fold and is currently only the second nematode DHFR structure in the Protein Data Bank. The equilibrium dissociation constants for NADPH (90 ± 29 nM) and folate (23 ± 4 nM) were determined by equilibrium titrations. The interactions of known antifolates with WbDHFR were analyzed using molecular docking programs and molecular dynamics simulations. Antifolates with a hydrophobic core and extended linker formed favorable interactions with WbDHFR. These combined data should now facilitate the rational design of filarial DHFR inhibitors, which in turn can be used to determine whether DHFR is a viable drug target for filariasis and whether existing antifolates may be repurposed for its treatment.


Author summary
Lymphatic filariasis is a disease commonly characterized by urogenital swelling, or scrotal hydrocele and lymphedema, which affects nearly 50 million people worldwide as of 2018. It is caused by parasitic worms transmitted through mosquitos. Dihydrofolate reductase (DHFR) is a ubiquitous enzyme involved in folate metabolism and provides building blocks for DNA synthesis. DHFR is an established pharmacological target in the treatment of cancer and infectious diseases through the use of DHFR specific inhibitors known as antifolates. Using recombinant DHFR from the parasitic worm W. bancrofti (Wb), we a1111111111 a1111111111 a1111111111 a1111111111 a1111111111

Introduction
Lymphatic filariasis, also known as elephantiasis, is a mosquito-borne helminth infection prevalent in Sub-Saharan Africa and Southeast Asia. The disease is caused by the parasitic worms W. bancrofti (Wb), B. malayi (Bm), and B. timori [1]. Approximately 500 million people annually receive medications such as ivermectin, diethylcarbamazine, and albendazole to treat and prevent lymphatic filariasis infection [2]. Symptoms of lymphatic filariasis include swelling of the arms and legs, hydrocele, and fibrosis resulting in long-term disabilities. A significant proportion of patients carry co-infections with more than one parasite [1,[3][4][5]. Some of these coinfections include malaria and schistosomiasis, a disease caused by another parasitic worm, S. mansoni (Sm) [6,7].
Literature reports have discussed the potential of DHFR as a drug target for treating helminth infections including filariasis [17][18][19][20]. Sharma and coworkers demonstrated that three antifolate compounds reduce Bm microfilariae motility by over 99% [17][18]. Upon treatment with folate, the authors observed motility returning, suggesting that inhibition of folate metabolism was responsible for the loss of motility. Supporting this is the fact that folate is structurally similar to the substrate DHF with the exception that in folate N8 and C7 are oxidized (Fig A in S1 Text). Furthermore, recent studies have demonstrated in vitro inhibition of filarial DHFRs from Wb and Bm by antifolate compounds [21,22]. For example, methotrexate showed K I values of 0.7 ± 0.1 nM and <0.5 ± 0.3 nM against WbDHFR and BmDHFR, respectively. While methotrexate also inhibits the human DHFR (HsDHFR), and therefore is not a selective antifilarial agent, these data nevertheless suggest that antifolates may represent treatment options for filariasis [22]. However, the absence of structural information for filarial DHFRs has made the investigation of structure-function relationships difficult. Serrão and coworkers recently reported the X-ray structure of S. mansoni DHFR (SmDHFR) (PDB code: 3VCO), the only other structure currently available for a nematode DHFR [23,24].
Approximately 90% of filariasis cases are caused by Wb [25,26]. Here we report the structure of WbDHFR in complex with its cofactor NADPH and folate determined by X-ray crystallography (Fig 1). The resulting structure made it possible to dock a series of known antifolates into the active site, analyze interactions using molecular dynamics simulations, and gain insights into their interactions with WbDHFR, as a possible guide for future drug design efforts.

Expression and purification of WbDHFR
The open reading frame of WbDHFR including an engineered N-terminal His6-tag was expressed in the E. coli LOBSTR strain [27] in terrific broth (TB) with 100 μg/mL ampicillin at 37˚C with shaking at 225 rpm. The starter culture was diluted 100-fold into 1 L of TB with 100 μg/mL ampicillin, grown to an OD 600 of 0.8-1.5 and induced with 1 mM IPTG at 37˚C. Cells were collected after 4 hours of incubation by centrifugation at 5,000 rpm for 30 minutes at 4˚C using a JA-10 rotor in an Avanti J-26S XP centrifuge. The resulting 12-16 grams of wet pellet that originated from 1 L of culture were stored at -80˚C until use.
The pellet from 1 L of culture as prepared above was resuspended in 100 mL of Mtx_Eq. buffer (8 mM Na 2 HPO 4 , 2 mM KH 2 PO 4 , 300 mM NaCl, 0.5 mM DTT, 10% glycerol, pH 7.0) and sonicated on ice for 15 minutes at 45% amplitude with 15 seconds on / 45 seconds off pulsing using the ½ inch probe on the Sonic Dismembrator FB505. The lysate was clarified by centrifugation at 17,000 rpm for 25 minutes at 4˚C in a JA-17 rotor. The lysate was loaded onto a 5 mL methotrexate-agarose resin column (Sigma Aldrich) that had been equilibrated with 10 column volumes of Mtx_Eq. buffer. The column was washed with~200 mL of Mtx_Eq. buffer and WbDHFR was eluted with~50 mL of 25 mM Tris, 300 mM NaCl, 5 mM folate, pH 8.6.
A nickel-nitriloacetic acid (Ni-NTA) column containing 10 mL of resin was equilibrated with 200 mL of 8 mM Na 2 HPO 4 , 2 mM KH 2 PO 4 , 300 mM NaCl, 10 mM imidazole at pH 7.4. The methotrexate-agarose column eluate was loaded directly onto the Ni-NTA column, and the column was washed with~150 mL wash buffer (8 mM Na 2 HPO 4 , 2 mM KH 2 PO 4 , 300 mM NaCl, 35 mM imidazole pH 7.4). The WbDHFR was eluted with 50 mL of 8 mM Na 2 HPO 4 , 2 mM KH 2 PO 4 , 300 mM NaCl, and 500 mM imidazole at pH 7.4. The Ni-NTA eluate containing WbDHFR (~20 mL) was dialyzed against 4 L of 40 mM HEPES, 100 mM NaCl, pH 7.0 for one hour and against an additional 4 L of the same buffer overnight at 4˚C. DHFR purity was assessed by SDS-PAGE (Fig B, part A in S1 Text). The concentration was determined at A 280 using the molar extinction coefficient 25,440 M -1 cm -1 [22]. The protein was concentrated to 20 mg/mL using an Amicon Ultra-15 10,000 MWCO centrifugal filter. Purification of WbDHFR yielded protein suitable for crystallization with a yield of 9-14 mg per liter of culture. The final WbDHFR preparation was filtered, aliquoted, and stored at -80˚C.

Crystallization of WbDHFR via sitting drop diffusion
WbDHFR was buffer exchanged using a 10 kDa MWCO centrifugal filter into 20 mM HEPES, 25 mM NaCl, 5 mM DTT, and 5 mM NADPH at pH 7.0 and incubated on ice for one hour. Several crystallization screens from Hampton Research and Jena Biosciences were evaluated using sitting drop vapor diffusion in MRC 96-well plates. The lower reservoir contained 50 μL of precipitant solution and drops were made with one microliter protein solution (20 mg/mL) and one microliter precipitant solution. The Jena Biosciences JBScreen Classic HTS I condition containing 100 mM tri-sodium citrate pH 5.6, 25% w/v PEG 4000, and 200 mM ammonium sulfate at 4˚C grew tiny crystals and this condition was then optimized. Increasing the pH from 5.6 to 6.6 in 0.2 pH unit increments resulted in larger, hexagonal crystals. The largest crystals grew to full size in about three weeks using a precipitant reservoir of 100 mM trisodium citrate pH 6.2, 25% w/v PEG 4000, and 200 mM ammonium sulfate. Crystals were cryo-protected in the precipitant solution containing 15% (v/v) glycerol. After transferring the crystals to the cryosolution, they were immersed in liquid nitrogen and stored in a cryo-Dewar until data collection.

Data collection and structure determination
Diffraction data for all crystals were collected at the Brookhaven National Synchrotron Light Source II (NSLS-II) on beam line AMX (17-ID-1). A total of 1,800 0.2˚images were collected spanning a phi range of 360˚. All images were indexed, integrated, and scaled using XDS [28]. Molecular replacement (MR) using the previously solved structure of mouse DHFR (PDB code 2FZJ) with high homology (~40%) was used to obtain phases. The starting mouse model was modified using Sculptor in Phenix [29] and the WbDHFR input sequence also included the 13 amino acid loop described in previous work [22]. The modified model was then used for MR using program Phaser [30]. Programs Phenix [29] and COOT [31] were used to build and refine the structure. Several iterations of model building and refinement continued until acceptable R-factors and geometric parameters were achieved. Bond lengths, angles, and clashes were assessed using MolProbity [32]. An unbiased omit 2 F o -F c map was generated using the "Iterative Omit Map Tool" in Phenix (Fig C in S1 Text) [29]. The final structure was analyzed in PyMOL and Maestro [33,34].

Molecular docking
The WbDHFR ternary structure with NADPH and folate bound served as the starting model for docking studies. The programs AutoDock VINA [35] and Glide (Schrödinger) [36] were used to dock compounds in parallel. Three-dimensional structure files (.sdf) for all antifolates were obtained from the PubChem database. AutoDockTools 1.5.6 was used to generate the appropriate input.pdbqt files for the docked ligands and WbDHFR. [37] The following center coordinates in Ångstroms were used for the docking grid in Autodock Vina [35]: x = 59.265, y = 17.612, z = -0.07. The models were analyzed in PyMOL [33]. To validate the docking methods, folate was first deleted from the ternary structure and then docked back into the active site of WbDHFR, with a resulting docking energy score of -9.1 kcal/mol (Vina) and -9.0 kcal/ mol (Glide). The predicted low docking energy scores correlate with the low K D (23 nM) that was determined for folate experimentally. Docking results also showed folate binding in a similar conformation to that observed in the experimentally determined ternary WbDHFR crystal structure, indicating that the docking approach was accurate. The antifolates methotrexate, pyrimethamine, trimethoprim, aminopterin, and trimetrexate were analyzed using molecular docking programs. The resulting docking models were used to predict contacts involved in drug-target interactions.
For comparison of docking scores obtained by another program, the same antifolates were docked into the WbDHFR crystal structure using the molecular docking program Glide [36]. The WbDHFR crystal structure was prepared using Protein Preparation in Maestro and the antifolates were prepared using Ligprep [34]. The docking grid was generated by specifying the 12 Å radius around folate in the WbDHFR active site; the docking grid coordinates in Ångstroms: X = 55.92, Y = 16.12, Z = -0.8. The antifolates were docked to the grid using Glide XP Precision Docking. Docking scores and poses for antifolates in the WbDHFR active site were examined in PyMOL [33] and Maestro [34].

Molecular dynamics simulations
The WbDHFR crystal structure, HsDHFR crystal structure (PDB code 2W3M), and the docking models for antifolates (methotrexate, aminopterin, trimetrexate, trimethoprim, pyrimethamine, and cycloguanil) were used for molecular dynamics (MD) simulations and analysis. Models were prepared by adding hydrogens, assigning charges, capping the termini, and deleting non-interacting water molecules using the Protein Preparation tool in Maestro [34]. The prepared models were energy minimized, then used to build the system for MD.
Systems for MD were prepared using System Builder in Desmond [38]. We predefined the solvent model as TIP4P. The system was neutralized by the addition of Na + and Clions, simulating concentrations of 0.15 M. The force field system was Optimized Potentials for Liquid Simulations (OPLS) and OPLS was applied to each system. Water molecules and ions around the enzyme were built into the system within the calculated orthorhombic box volume for soluble proteins. The average calculated box volume was = 240,000 Å 3 . All models were relaxed before simulation. Desmond was used to conduct MD simulations under the following conditions: canonical ensemble (NVT); Berendsen thermostat (temperature = 300˚K); and Berendsen barostat (pressure = 1.01325 bar); timescale = 10 nanoseconds (ns) [38]. We used 10 ns (10 picoseconds/interval) to allow for system equilibration confirmed by protein root mean square deviation (RMSD) plots. A total of 1000 snapshots (10 picoseconds/interval) were generated for each complex. Simulations were analyzed for equilibration and convergence by examining root mean square deviation (RMSD) versus simulation time plots. Trajectories from each simulation were analyzed using Simulation Interaction Analysis and Simulation Event Analysis programs in Desmond [38].

Determination of K D of NADPH and folate
The dissociation constants (K D ) of the ligands NADPH and folate were determined by monitoring changes in tryptophan fluorescence upon increasing ligand concentrations. For dissociation constant determination, only the Ni-NTA affinity chromatography purification method (leaving out the methotrexate-agarose column step) was used to ensure WbDHFR was not bound to ligands during affinity studies. NADPH or folate was added to 400 nM WbDHFR at room temperature and fluorescence values were recorded using a Fluoromax-4 (Horiba Jobin Yvon) spectrofluorimeter (ex: 290 nm, em: 340 nm, with 5 mm and 10 mm slits, respectively).
The tested ligand concentrations ranged from 0 to 10 μM for NADPH and 0 to 4 μM for folate. The titrations were performed by adding one microliter increments of ligand to a quartz cuvette that contained 1000 μL of 400 nM enzyme in 1X MTEN (50 mM MES, 25 mM Tris, 25 mM ethanolamine, 100 mM NaCl, 2 mM DTT) buffer at pH 6.0 and allowing equilibration for 10 seconds. The data were corrected for the inner filter effect using 200 nM L-tryptophan as described previously [39]. Fluorescence intensity values were graphed against ligand concentrations using Kaleidagraph and the data were fitted to the Morrison equation [40][41][42].

Crystal optimization and structure determination
We here report the first crystal structure of DHFR from Wb, a parasitic nematode that causes lymphatic filariasis. The crystals obtained for data collection first appeared after five to eight days of incubation and continued to grow for three weeks (Fig B, part B in S1 Text). Crystals of WbDHFR grew in the presence of NADPH in the crystallization buffer. One crystal diffracted to amplitudes extending to 2.47 Å using native wavelengths (1.0 Å). The diffraction pattern confirmed that the WbDHFR crystals had a hexagonal lattice and space group P6 3 22. The data set was 99.69% complete with acceptable R-factors ( Table 1). The Matthew's coefficient predicted one molecule in the asymmetric unit (ASU) with~75% solvent content. These parameters were used to solve the phases via molecular replacement with Phaser [30]. The structure was deposited in the Protein Data Bank as 8E4F. Crystallization trials including either methotrexate or pyrimethamine did not result in crystal growth. In fact, neither co-crystallization nor soaking pre-existing crystals with antifolates provided crystals of sufficient quality for diffraction studies.

Analysis of the overall structure
The finalized crystal structure revealed a typical DHFR fold with four alpha helices and six parallel strands and one anti-parallel beta strand, similar to the human and bacterial orthologs of DHFR (Fig 2). One notable difference between the HsDHFR and WbDHFR orthologs is the absence of four amino acids near the C-terminus in WbDHFR (#160-164 in the human enzyme) [43]. The crystal structure reveals a ternary complex with both NADPH and folate present in the active site. NADPH was intentionally co-crystallized with WbDHFR while folate was used for the methotrexate-agarose column elution and remained protein-bound throughout the purification process. Omit electron density shows strong peaks at 1.0 sigma for folate in the active site (Fig C in S1 Text). The refined crystal structure of WbDHFR also shows the presence of two SO 4 2ions bound to basic residues Arg-21, Arg-117, and Arg-149. These ions likely originated from the crystallization solution. According to our model, the donor NADPH-acceptor DHF distance in WbDHFR was estimated to be 3.5 Å, (Fig D in S1 Text) similar to the hydride transfer distances in E. coli DHFR (3.6 Å) and HsDHFR (3.0 Å) [44][45].

Interactions with folate
We determined the dissociation constant (K D ) for folate to be 23 ± 4 nM, indicating tight binding affinity (Fig 3A). The low K D for folate was measured in the absence of NADPH and we expect that the K D would be even lower if NADPH was included in the folate K D determination due to the reported positive cooperative binding of folate [46]. Using tryptophan fluorescence for K D determination limits the experiment to only one ligand due to the need to correct for inner filter effects. Interactions observed for folate and residues of WbDHFR include hydrogen bonds between the carboxylic acid of Glu-32 in WbDHFR and the N3 and NA2 atoms of folate (Fig 4A and 4B; see atom labels in Fig A in S1 Text). The interaction distances between the carboxylate of Glu-32 and the N3/NA2 atoms are 2.9 Å and 2.6 Å, respectively (Fig D in S1 Text). The NH 2 of Arg-72 hydrogen bonds with the O1 and O2 atoms of folate (Fig A in S1 Text), at distances of 3.0 Å and 2.7 Å, respectively (Fig D in S1 Text). Phe-36 is involved in π-stacking and makes van der Waals contacts with the para-benzoic acid group (PABA) of folate (Fig 4A and 4B). Phe-36 likely interacts with methotrexate in a similar

Interactions with NADPH
NADPH binds in the extended conformation identified in many DHFR structures [49]. Sequence differences in WbDHFR compared with other DHFRs result in minor interaction changes between NADPH and the co-factor binding site. The K D of NADPH and WbDHFR (in the absence of folate) was determined using tryptophan fluorescence to be 90 ± 29 nM ( Fig  3B). The K D for NADPH binding to BmDHFR, a filarial nematode ortholog with 96% sequence identity to WbDHFR, was previously found to be 25 ± 24 nM [21]. These values are similar to the K D (150 ± 20 nM) reported for the E. coli ortholog [50]. In DHFR homologs from H. sapiens, E. coli, and C. albicans (PDB codes 4KD7, 2ANO, 1AOE, respectively), the hydroxyl group and backbone NH of the highly conserved threonine, which corresponds to Val-58 in WbDHFR, form hydrogen bonds with the pyrophosphate region of NADPH. In contrast, WbDHFR contains a valine residue at position 58. The backbone NH of Val-58 forms a hydrogen bond with NADPH while the valine sidechain contacts folate via van der Waals interactions (Fig E in S1 Text). In addition, HsDHFR and MmDHFR structures have a conserved arginine (Arg-77) that forms a sidechain π-cation interaction with the adenine ring. In WbDHFR, this arginine Amino acids corresponding to the Met20 loop are highlighted in red in A and in a red rectangle in B. Sequences of DHFR orthologs were obtained through UniProt and aligned with the multiple sequence alignment tool "ClustalW" within Jalview. Amino acids with similar chemical characteristics were colored using the "ClustalX" color scheme. Amino acids observed to make hydrogen bonds with either folate or NADPH are labeled "F" and "N", respectively. The first amino acid residues for Hs, Mm and Sa DHFR are denoted as #2 because these sequence entries are derived from proteins purified from the organisms rather than cloned cDNA.
https://doi.org/10.1371/journal.pntd.0011303.g002 corresponds to Lys-79. Electron density for the sidechain of Lys-79 was disordered and this side chain could not be modeled and therefore probably does not interact with NADPH.
The nicotinamide ring of NADPH (Fig 2A) is involved in π-stacking interactions with the pteridine of folate. This interaction is commonly found in ternary complexes of DHFR and highlights the cooperativity between DHF and NADPH binding sites. This stacking interaction also enhances the interactions of folate and residues Glu-32, Phe-36, and Arg-72 (Fig 4) by positioning the pteridine to maintain proper interaction distances.

Comparison of WbDHFR and SmDHFR structures
To our knowledge, there is only one other nematode DHFR structure that is currently available in the PDB, the DHFR from Sm. In comparing WbDHFR with SmDHFR, we found that the structures are homologous (with an alignment RMSD of 1.324 Å) but differ in sequence; the amino acid sequence identity is only 32%. The alignment of the ternary WbDHFR structure with the structure of the apoenzyme SmDHFR revealed high structural similarity with the exception of the Met20 loop (Fig F in S1 Text). In SmDHFR, the Met20 loop is disordered while in WbDHFR, it exhibits a typical "closed" conformation ( Fig F in S1 Text); these loop conformations are discussed in the article by Sawaya and Kraut [51]. The difference in the Met20 loop structures likely reflects the different binding states of the two enzyme preparations used; WbDHFR is bound to folate and NADPH while the SmDHFR structure is in the unbound apo state. The relatively high RMSD between the Wb and Sm DHFRs likely results from these differences. Other DHFRs from more evolutionarily distant organisms such as mouse DHFR and HsDHFR share greater sequence identity (40-41%) with WbDHFR but lower overall structural similarity ( Table 2). In WbDHFR, Arg-72 interacts with folate's polyglutamic acid tail (Fig 4A and 4B). Similarly, SmDHFR also has an arginine (Arg-67) in this position. In WbDHFR, Glu-32 interacts with the pteridine ring of the substrate through hydrogen bonding; in SmDHFR, this position is occupied by Asp-28, which likely forms similar interactions.
Given the similarities in folate binding motifs between the SmDHFR and WbDHFR, we hypothesize that novel antifolates with improved hydrophobic-binding capability could interact with both enzymes. As described in more detail below, there are differences in amino acid side chains between the HsDHFR antifolate binding site and those of the two nematode DHFR structures described here, presenting an opportunity for the design of dual inhibitors for WbDHFR and SmDHFR that would be selective against HsDHFR.
We further explored the amino acid sequence alignments, focusing specifically on the residues within the Met20 loop, a well-studied, catalytically significant structural element in the DHFR family of enzymes (Fig 2). The Met20 loop for WbDHFR is most similar to the Met20 loop in EcDHFR (residues 10-24) [51]. A seven amino acid long motif within the Met20 loop, containing two prolines that are spaced three residues apart (PWXLPAX) is clearly present in WbDHFR (Fig 2 and Table A in S1 Text). The same motif in SmDHFR is also seven amino acids long but contains only one proline (PWKIKKD). Previous literature has suggested that longer motif length is associated with lower conformational flexibility of the Met20 loop, preventing large-scale opening and closing loop motions [23] and thus influencing the catalytic cycle [52]. Based on these observations, we predict that the Met20 loops of both SmDHFR and WbDHFR undergo conformational motions during the catalytic cycle. In contrast, HsDHFR has a longer Met20 loop with 8 residues in the motif (PWPPLRNE) and does not undergo closed to occluded transitions [23,51].
The sequence of the Met20 loop of WbDHFR compared to other DHFR orthologs (Fig 2) is largely similar with the two exceptions being Met-25 and Phe-28. Met-25, which is replaced by either leucine or isoleucine in other DHFR orthologs, makes nominal contacts with folate in both DHFR models from E. coli and W. bancrofti. Phe-28, a residue that is not conserved in other orthologs, does not make any contacts with folate but points toward the active site and could conceivably make hydrophobic contacts with a longer antifolate. Also, the Met20 loop is less flexible based on root mean square fluctuations (RMSFs) from the MD simulations for HsDHFR when compared to WbDHFR (Fig G in S1 Text). A more flexible loop, as observed for WbDHFR, may be able to sample more conformations and increase overall interactions with a longer antifolate that expands into this region.

Molecular docking of antifolate compounds
In previous studies, we tested antifolates as potential inhibitors of WbDHFR; methotrexate (K I = 0.7 ± 0.1 nM) and structurally similar aminopterin (K I = 2.1 ± 0.5 nM) were found to be the most potent inhibitors of those that were tested [22]. Other antifolates, including trimethoprim and pyrimethamine were low micromolar inhibitors. The crystal structure published here allowed us to extend these studies through molecular docking.
The docking poses of antifolates obtained from Vina and Glide were largely similar for a given antifolate (Fig H in S1 Text). The 2D ligand interaction diagrams for Glide and Vina also showed similar interactions (Figs I and J in S1 Text). VINA docking results predicted that folate analogs methotrexate and aminopterin have similar conformations and docking scores, -8.8 and -8.7 kcal/mol, respectively (Fig 5). These scores are also similar to the value predicted for folate (-9.1 kcal/mol). The Glide output energies for methotrexate (-9.1 kcal/mol) and aminopterin (-8.5 kcal/mol) also supported the VINA docking predictions (Table B in S1 Text). Methotrexate and aminopterin contain a modified pteridine ring, central hydrophobic linker, and polyglutamic acid tail, similar to the substrate DHF (Fig A in S1 Text). Here, we classified the traditional antifolates as having either a "longer" or "shorter" linker based on the number of "linker" atoms between pyrimidine/pteridine and polyglutamic acid pharmacophores. This classification includes atoms in the para-benzoic motif (8 atoms) but does not count the atoms in the pteridine ring or polyglutamic acid tail. Overall, the "shorter" linkers consist of 0-1 atoms while "longer" linkers are composed of >7 atoms. The longer linker in methotrexate and aminopterin is positioned to interact with nonpolar amino acids Phe-36, Leu-69, and Met-33 in the hydrophobic core of the active site of WbDHFR. Pyrimethamine has no such linker and trimethoprim has only a short linker consisting of one CH 2 group (Fig A in S1  Text). The docking predictions shown in Fig 5 suggested that these smaller antifolates made fewer interactions with the hydrophobic core of the enzyme active site. The docking results provide insight for WbDHFR-antifolate interactions but there are several limitations to these models. The docking scores only provide relative estimates of binding affinities and must be validated through binding assays. We observed some correlations between the docking scores and affinity measurements. For example, the experimentally determined K I values for methotrexate and pyrimethamine (0.7± 0.1 nM and 15 ± 6 μM) qualitatively correlated with the Vina (-8.8 kcal/mol and -7.9 kcal/mol) and Glide (-9.1 kcal/mol and -7.0 kcal/mol) estimated binding energies. The scoring functions for Vina and Glide are similar and rely on summing contributions of hydrophobic and hydrophilic interactions. The score differences may be due to the different force fields, OPLS and AMBER, which Glide and Vina use, respectively. While our docking comparison suggests that the OPLS force field works better for the WbDHFR structure, the docking scores only provide relative estimates of binding. The predicted absolute energies from molecular docking may not reflect experimentally determined K I values due to assumptions such as rigid-body protein receptor, no solvation, and simplified general scoring [53,54]. However, the trend in the predicted binding energies from Vina and Glide docking programs agree that antifolates with longer linkers can make more contacts with WbDHFR compared to shorter linker antifolates.

MD simulations
The MD simulations provided additional insights into the binding interactions between WbDHFR and the antifolates. System equilibration of the WbDHFR and HsDHFR simulations was confirmed through plotting RMSD over time (Figs K and L in S1 Text). Similar to the crystal structure, the MD simulations showed strong interactions between folate and Glu-32, Ile-10, and Arg-72. Additional interactions included the stacking interaction with Phe-36 and several van der Waals contacts with hydrophobic residues. The simulations provided the relative fraction or percentage of time during which the interaction was observed over 10 nanoseconds; strong interactions are those that were observed >50% of the time in the simulation.
The crystal structure, K I data, and molecular docking scores suggest that antifolates with longer linkers have better affinity for WbDHFR. When comparing total protein-ligand contacts for all MD simulations, we found that antifolates with longer linkers make more overall hydrophobic contacts in the simulation compared to antifolates with shorter linkers. Fig 6 shows a comparison of the interaction fractions for key residues for simulations conducted for WbDHFR and all antifolates. When comparing the interaction fractions, antifolates with longer linkers made more hydrophobic contacts for longer fractions or periods of time in the simulation.
MD simulations were also used to compare the WbDHFR and HsDHFR Met20 loop regions. Using average root mean square fluctuation (RMSF) values for residues within the Met20 loop, we found that WbDHFR undergoes more fluctuations compared to HsDHFR ( Fig  G in S1 Text). This finding supports the hypothesis that HsDHFR adopts a closed conformation, while WbDHFR undergoes conformational motions during the catalytic cycle.

Insights for rational design of WbDHFR inhibitors
Experimental data have demonstrated that traditional antifolate substrate analogs can inhibit WbDHFR. Subsequent molecular docking predictions using the WbDHFR X-ray structure . In cases where the interaction fraction is >1.0, the residues (or atoms) may be involved in more than one interaction. Common interactions are color-coded including hydrogen bonds (green), hydrophobic contacts (purple), ionic bonds (pink), and water bridges (blue) (D). Comparison of interaction fractions shows that the long linker antifolates make more hydrophobic contacts.
https://doi.org/10.1371/journal.pntd.0011303.g006 PLOS NEGLECTED TROPICAL DISEASES corroborated this finding and revealed potential strategies for drug design. As described below, there are amino acid side chain differences between the folate binding sites of HsDHFR and WbDHFR (Fig 7), suggesting opportunities for developing WbDHFR selective inhibitors. There are several differences between the folate binding sites of WbDHFR and HsDHFR (PDB 2W3M). Docking the antifolates to HsDHFR and WbDHFR with Glide and Vina shows that methotrexate, a HsDHFR-tailored inhibitor, is predicted to bind with high affinity to both enzymes (Table C in S1 Text). Glide predicts that trimetrexate has a higher affinity for WbDHFR than HsDHFR. Two hydrophilic residues that form side chain hydrogen bonds to folate in the human structure (Gln-35 and Asn-64) have hydrophobic counterparts in the WbDHFR structure (Ala-37 and Phe-66) (Fig 7). One similarity between the WbDHFR and SmDHFR is that they share active site hydrophobic residues unique to these nematode orthologs such as Met-33 and Phe-66 in WbDHFR and the corresponding Met-29 and Phe-61 in SmDHFR. Therefore, small molecules with hydrophobic linker motifs may serve as dual inhibitors for both SmDHFR and WbDHFR enzymes while lacking affinity for the human enzyme.
An antifolate where the carbonyl O atom (Fig A in S1 Text) has been replaced with a nonpolar functional group or compounds with longer and more hydrophobic polyglutamic acid tails could be more selective for WbDHFR over HsDHFR. Based on the docking predictions shown in Fig 5 an effective strategy to develop antifolates targeting WbDHFR could exploit the side chains of residues Glu-32, Arg-72, Tyr-120, and Ser-61 and the backbone carbonyls of Ile-10 and Ile-114.
The active site of WbDHFR contains hydrophobic regions consisting of residues Phe-66, Ala-37, and Val-58 (Fig 7). These residues may also be used to anchor aliphatic substitutions in novel antifolates, since they are not present in the human ortholog, in which hydrophilic side chains occupy these positions. The predictions shown here and in previous studies have noted the importance of linker length and hydrophobic moieties connecting the diaminopyrimidine pharmacophore in the design of antifolates [44,55,56]. In further support of this importance, our docking models for both trimetrexate and trimethoprim (both containing a trimethoxyphenyl functional group) predict a more negative docking score for trimetrexate (-8.6 kcal/mol) than for the shorter trimethoprim (-7.6 kcal/mol). This may again be due to the linker length of trimetrexate and its subsequent ability to extend further, thus taking advantage of additional hydrophobic residues such as Leu-69 and Phe-66 (Fig 5). We determined the IC 50 value of trimetrexate experimentally to be 0.49 ± 0.16 μM (Fig M in S1 Text), which lies between the previously determined IC 50 values for trimethoprim and methotrexate (83 ± 25 μM and 0.018 ± 0.003 μM, respectively). The docking results and the biochemical data suggest that interactions with the more hydrophobic residues in the active site of WbDHFR are important for efficient inhibition.
We now have the crystal structure of WbDHFR with folate and NADPH bound in the active site. Our studies were limited to mostly computational methods when exploring the linker length of effective antifolate inhibitors for WbDHFR. Therefore, future experiments may involve solving the crystal structure of WbDHFR bound to antifolates with different linker lengths. In summary, inspection of the WbDHFR crystal structure, docking predictions, and results from MD simulations suggest that novel antifolates designed to contain a central hydrophobic linker directed to interact with the now well-established WbDHFR target regions could help make antifolate therapy a reality in the treatment of filarial nematode infections. Omit electron density for active site residues of WbDHFR, folate, and NADPH. Electron density is contoured to 1.0 sigma. Fig D. Distances in Å between atoms of folate (yellow) and residues of WbDHFR (cyan) that form hydrogen bonds are shown. The two amino acids responsible for forming hydrogen bonds with folate, Glu-32 and Arg-72, are shown as sticks. The predicted distance of the hydride transfer from NADPH to folate is also shown. The distances were measured in PyMOL. Fig E. Alignment of ternary structures of HsDHFR and WbDHFR demonstrates the similarity of ligand conformations for both folate and NADPH. The ternary structure for HsDHFR (PDB: 2W3M, green) was obtained from the PDB and was aligned with WbDHFR in PyMOL. The conformation of folate for both WbDHFR and HsDHFR is very similar; however, the adenine moiety in NADPH is differently positioned in the two structures. Fig F. Alignment of SmDHFR apoenzyme (PDB: 3VCO, magenta) and the WbDHFR ternary structure (PDB: 8E4F, cyan) done in PyMOL. Folate and NADPH from WbDHFR are shown in yellow. The Met20 of the ternary WbDHFR structure is in the closed conformation while the Met20 loop of SmDHFR is in the disordered conformation. Table A: Met20 loop, B-factor, and steady state parameter data for four DHFR homologs. Fig G. Comparison of root mean square fluctuation (RMSF) values for residues within the Met20 loop region  in Angstroms: WbDHFR (blue) and HsDHFR (red). WbDHFR shows more fluctuation in the Met20 loop region whereas HsDHFR shows less fluctuation. Fig H. Alignment of docking poses of antifolates obtained from Glide and Vina. The docking poses of the antifolates from Glide (green) and Vina (violet) with the receptor WbDHFR (cyan). The NADPH cofactor is seen as yellow. Table B: Summary of interactions observed in molecular docking models obtained from both Autodock and Glide. Gubler.