Mycobacterium tuberculosis IMPDH in Complexes with Substrates, Products and Antitubercular Compounds

Tuberculosis (TB) remains a worldwide problem and the need for new drugs is increasingly more urgent with the emergence of multidrug- and extensively-drug resistant TB. Inosine 5’-monophosphate dehydrogenase 2 (IMPDH2) from Mycobacterium tuberculosis (Mtb) is an attractive drug target. The enzyme catalyzes the conversion of inosine 5’-monophosphate into xanthosine 5’-monophosphate with the concomitant reduction of NAD+ to NADH. This reaction controls flux into the guanine nucleotide pool. We report seventeen selective IMPDH inhibitors with antitubercular activity. The crystal structures of a deletion mutant of MtbIMPDH2 in the apo form and in complex with the product XMP and substrate NAD+ are determined. We also report the structures of complexes with IMP and three structurally distinct inhibitors, including two with antitubercular activity. These structures will greatly facilitate the development of MtbIMPDH2-targeted antibiotics.


Introduction
Tuberculosis (TB) is a pandemic contagious infectious disease affecting nearly two billion people worldwide, with 5.7 million new cases reported in 2013 according to the World Health Organization [1]. The currently approved standard first-line treatment is long and requires combination therapy consisting of several antibiotics [2]. Mycobacterium tuberculosis (Mtb), the causative agent of most TB cases is a slow growing, remarkably successful pathogen capable of switching between different physiological states and adapting to disparate host environments [3,4]. The Mtb pathogen can persist in a quiescent state and survive for decades as a latent infection [5,6]. The increased prevalence of multidrug resistant (MDR) and extensively drug resistant (XDR) strains of Mtb, for which treatment options are very limited, demands the development of more effective antitubercular agents, ideally with novel mechanisms of action [6,7].
The design of antimetabolites that inhibit biosynthesis of essential metabolites within a cell is a classic approach for discovery of new antibiotics and chemotherapeutic agents. Indeed, one of the first reported TB drugs, para-aminosalicylic acid, was recently shown to disrupt folate biosynthesis [8], while the newest TB drug, bedaquiline exerts its activity through inhibition of ATP synthesis [9]. Inosine 5'-monophosphate dehydrogenase (IMPDH) lies at a key intersection of the purine biosynthesis pathway and represents an extremely attractive target since it controls flux of the guanine nucleotide pool. In many pathogens, guanine nucleotide levels are IMPDH-dependent and thus inhibition of IMPDH is a viable strategy for design of new chemotherapeutic agents [10]. The purine nucleotide biosynthetic pathway of Mtb, in common with other bacteria, contains three different enzymes from that of humans [11,12]: (1) Mtb contains both the folate dependent N 1 -(5-phospho-D-ribosyl)glycinamide (GAR) transformylase (PurN) and the formate/ATP utilizing formyl-N 1 -(5-phospho-D-ribosyl)glycinamide (FGAR) synthetase (PurT), whereas humans only contain GAR transformylase. (2) Humans use a type I 2-(formamido)-N 1 -(5-phospho-β-D-ribosyl)acetamidine (FAGM) synthetase composed only of PurL. Bacteria typically contain a type II FGAM synthetase, which is a complex of PurLQS. Orthologs for both PurL and PurQ have been identified in Mtb, as has a candidate PurS, which suggests that Mtb also utilizes the type II enzyme [11]. (3) Whereas humans convert 5-amino-1-(5-phospho-D-ribosyl)imidazole (AIR) directly to 5-amino-1-(5-phospho-D-ribosyl)imidazole-4-carboxylate (CAIR) via a class II AIR carboxylase (PurE class II), Mtb uses N 5 -carboxyaminoimidazole ribonucleotide (NCAIR) synthetase (PurK) to first convert AIR to NCAIR, then a class I AIR carboxylase converts NCAIR to CAIR. The remaining steps are common to both humans and Mtb. IMPDH catalyzes the NAD + -dependent conversion of inosine 5'-monophosphate (IMP) to xanthosine 5'-monophosphate (XMP), the first of the two step biosynthesis of guanosine 5'-monophosphate (GMP). GMP synthetase catalyzes the conversion of XMP to GMP in a reaction that also hydrolyzes glutamine to glutamate and ATP to AMP and pyrophosphate.
The Mtb H37Rv genome contains three genes with sequence similarity to bacterial IMPDH (guaB1, guaB2, and guaB3); the functions of guaB1 and guaB3 are poorly understood and only guaB2 was shown to encode an active IMPDH [13], [14] (S1 Fig [54,55]). Consistent with this finding, the guaB2 gene is essential and cannot be rescued by the other orthologs [15]. The IMPDH activity of the guaB2 gene product, MtbIMPDH2, has been confirmed [14,16] and small molecule inhibitors have also been described [14,16,17].
The enzymatic mechanism of IMPDH has been extensively studied and consists two steps, a dehydrogenase and a hydrolase reaction ( Fig 1A) [10,18]. Upon binding of IMP and NAD + cofactor, a thioimidate enzyme-substrate adduct, E-XMP Ã , is formed via a covalent bond to the catalytic C341 (MtbIMPDH2 numbering) with concurrent production of NADH. Hydride transfer occurs to the pro-S position with the cofactor in the anti-conformation (Fig 1B). The cofactor is released and an active site mobile flap moves into the NAD + site and facilitates E-XMP Ã hydrolysis with a conserved R443 acting as a general base. Thus, the enzyme has two distinct conformations, an open form for the dehydrogenase reaction and a closed form for E-XMP Ã hydrolysis. IMPDH has multiple active site states (apoE, E•IMP, E•IMP•NAD + , E-XMP Ã •NADH adduct) that can be targeted for inhibitor design, and both the IMP/XMP and cofactor binding sites have been exploited for that purpose [10,18,19].
Human and bacterial IMPDHs differ in their affinity toward the NAD + cofactor. While the nicotinamide portion of NAD + binds similarly in human and Vibrio cholerae IMPDHs to preserve the interaction with IMP for hydride transfer, the interactions of the adenosine moiety vary dramatically. In the human type II enzyme (hIMPDH2), the adenine ring π/π stacks between a pair of aromatic residues within the same monomer as the IMP binding site (H253 and F282 in hIMPDH2; referred to as the A E -subsite; S1 Fig) and the cofactor maintains the extended conformation typically found in dehydrogenases [20]. In contrast, in V. cholera IMPDH (VcIMPDH), the adenine ring binds to the adjacent monomer and interacts with a different set of residues (referred to as the A B -subsite; S1 Fig) and the cofactor assumes a rare compact conformation [21]. It also appears that adenine binding to this site is less constrained because two different conformations are observed for this moiety in two crystal structures of VcIMPDH [21]. This provides critical information that the design of bacteria-selective inhibitors should focus on the A B -subsite. Importantly, compounds that bind with high affinity to this site should not bind to human IMPDHs.
Bacterial IMPDH-specific compounds were discovered in a high throughput screen for inhibitors of Cryptosporidum parvum IMPDH (CpIMPDH). Although C. parvum is a protozoa, its IMPDH is a bacterial-like IMPDH [22]. Multiple potent chemotypes have been developed as inhibitors of this enzyme (designated as classes A, C, D, N, P, and Q, among others) [23][24][25][26][27][28][29]. These inhibitors target the A B -subsite and thus show high selectivity for bacterial IMPDHs. Moreover, structural studies of CpIMPDH with representatives from two classes, C64 and Q21, identified an "inhibitor minimal structural motif" (IMSM) consisting of A165 and Y358' (prime denotes a residue from the adjacent monomer) that is required for compound binding [27,[30][31][32]. Bacterial IMPDHs have high sequence similarities and the IMSM motif is found in many species, including MtbIMPDH2 with A285/Y487' in the corresponding positions. However, despite this sequence conservation, the structure-activity relationships (SAR) of a given inhibitor chemotype are not preserved between different bacterial IMPDHs, thus structure-based compound optimization toward a given IMPDH cannot be performed by simple knowledge-based prediction, but requires experimental verification [21,31,32].
Numerous efforts by several groups in academia and industry directed at characterization of either apo or ligand-bound MtbIMPDH2 structures have been unsuccessful, likely due to the very low solubility of the recombinant MtbIMPDH2 protein expressed in E. coli. We have overcome this major limitation by designing a variant that lacks the non-catalytic CBS subdomain (MtbIMPDH2ΔCBS), greatly improving solubility [21] without impacting enzyme catalytic properties. Here we present the first crystal structures of MtbIMPDH2ΔCBS, including the apoenzyme, the E•XMP•NAD + complex, and complexes with IMP and three inhibitors, MAD1, P41, and Q67 ( Fig 1C). In addition, we report the SAR for MtbIMPDH2ΔCBS inhibition along with antibacterial activity of several inhibitors from structurally distinct classes.

Engineering IMPDH for inhibitor and structural studies
Deletion of IMPDH CBS domains facilitated crystallization of enzymes from several other organisms [21], therefore we constructed a variant of MtbIMPDH2 wherein residues E126-R252 were replaced with a GG linker. The MtbIMPDH2ΔCBS showed significantly improved solubility and crystallizability properties [21] with the steady state kinetic parameters comparable to those reported for the wild type enzyme (Table 1 [16]). Similar results were reported recently for deletion mutants of three other bacterial IMPDHs [21]. Notably, the values of K m for both IMP and NAD + for MtbIMPDH2 are significantly higher than those of the human enzymes, illustrating the functional differences between bacterial and human IMPDHs (Table 1 [33][34][35][36]).
Five P compounds and twelve Q compounds displayed significant activity against Mtb strain H37Rv in minimal BSA-free medium (MIC 20 μM, Fig 2, Tables 2 and 3). The compounds were somewhat less effective in BSA-supplemented rich media (Table 2). No active compounds were identified in the other structural series. The active compounds have significantly more polar surface area (average topological polar surface area (tPSA) 83 ± 18 Å 2 ) than the inactive compounds (average tPSA = 64 ± 4 Å 2 , p < 0.001). The active and inactive compounds have similar hydrophobicity (average cLogP = 4.1 ± 0.8 for the active versus cLogP = 4.4 ± 1.3 for the inactive). All of the active compounds were potent inhibitors of MtbIMPDH2ΔCBS, with values of K i,app ranging from 13-2000 nM ( Table 2). Notably, P67 and Q67 are the most potent inhibitors of MtbIMPDH reported to date.
Several observations suggest that antitubercular activity results from inhibition of MtbIMPDH2. First, (S)-isomers of the Q compounds inhibit bacterial IMPDHs, while the (R)isomers are inactive [27]. The racemate Q60 has approximately half the antitubercular activity as the (S)-isomer Q67, as expected if MtbIMPDH2 is the target (S7 Table). The values of MIC for ten active compounds, including all the P compounds, increased by at least a factor of 4 in the presence of guanine. This rescue provides strong evidence for the on-target activity of the compounds due to inhibition of MtbIMPDH2 in the bacteria. The values of MIC increased by lesser extents for the remaining compounds (Q9, Q22, Q27, Q42, Q46, Q49, Q59). These compounds may also engage another target. Lastly, antitubercular activity correlated roughly with the potency of MtbIMPDH2ΔCBS inhibition ( Table 2 and  All five P compounds also displayed antibacterial activity against B. anthracis (S8 Table  [ 37]). Of the eight Q compounds also tested against B. anthracis, only Q67 displayed activity against both bacteria. While P32 and Q67 have similar activity against both bacteria, P41 and P67 are 5-7-fold more effective against Mtb and P146 and P150 are 15-28-fold more effective against B. anthracis. Differences in the SARs for enzyme inhibition cannot simply account for these differences in antibacterial activity (S8 Table and S2 Fig). Therefore, differences in cellular accumulation likely determine antibacterial spectrum. Interestingly, the compounds active   Table 3. Designation "5" and "6" refers to the position in the benzoxazole ring of Q compounds.   Table 3. Structures of CpIMPDH inhibitors with antitubercular activity. against B. anthracis are significantly less hydrophobic than those active against Mtb (cLogP = 3.5 ± 0.5, p = 0.018; S8 Table [37]). The five active P compounds contain a 3-carboxamido-4-chlorophenyl ring. Remarkably, the 4-chloro substituent is also found in 21 inactive P compounds, suggesting that the 3-carboxamido-4-chlorophenyl ring is critical for antitubercular activity. The 3-piperazinylcarbonyl-4-chloro analog P94 is inactive, suggesting that the positive charge is deleterious (S5 Table). The inactive compounds include the alkene analog of P32 (P16), the ketone analog of P32 (P25) and methyloxime analog of P67 (P74). P16, P25 and P74 are also expected to be good inhibitors of MtbIMPDH2, which suggests that the oxime group confers an advantage for cellular accumulation.
All of the active Q compounds except Q22 contain the 2-(pyrid-4-yl)benzoxazole group, which largely reflects the SAR of enzyme inhibition for CpIMPDH and the consequential predominance of this group in the Q pool (70%). Nine of the active Q compounds contain a 2-chloro-substituted phenyl ring, and five of these are 2,3-dichloro substituted. Q58 and Q64 are the only two inactive compounds that contain 2,3-dichlorophenyl group. Q58 is the 2-(thiazol-2-yl)benzoxazole analog of the active compound Q36, while Q64 is the 2-(thiazol-5-yl)benzoxazole analog of the active compound Q59. These substitutions are expected to decrease the potency of enzyme inhibition by a factor of 4-20 based on CpIMPDH and BaIMPDH data (S8 Table). Therefore the 2,3-dichlorophenyl group is important for antitubercular activity in the benzoxazole scaffold.

Mechanism of MtbIMPDH2ΔCBS inhibition
Although the CpIMPDH inhibitors all bind in the cofactor site, their mechanisms of inhibition can vary depending on their relative affinities for the E•IMP and E-XMP Ã complexes. Therefore we determined the mechanism of inhibition for the representative antitubercular compounds P41 and Q67, as well as for compound MAD1, the first reported inhibitor of MtbIMPDH2 [16] (Fig 1C);. MAD1 is a third generation mycophenolic adenine nucleotide (MAD) inhibitor [16,39,40] that bears the closest resemblance to NAD + (Fig 1). All three compounds are uncompetitive inhibitors with respect to IMP as expected, implying interactions with the base of IMP are critical for binding (S9 Table). The mechanism of inhibition with respect to NAD + varies among the three compounds (S9 Table). MAD1 is an uncompetitive inhibitor, suggesting that this compound has a strong preference for the E-XMP Ã intermediate. P41 is a noncompetitive inhibitor, indicating this compound has similar affinity to both E•IMP and E-XMP Ã . Q67 is a competitive inhibitor versus NAD + , suggesting that it has a strong preference for E•IMP.

Overall structure
Five high resolution crystal structures were obtained for MtbIMPDH2ΔCBS (Table 4). These include the apo form, the XMP•NAD + complex, and three IMP•inhibitor complexes (E•IMP•I) with MAD1, P41, and Q67. The structure of the tertiary complex of MtbIMPDH2ΔCBS with XMP and NAD + was obtained by soaking crystals containing the MtbIMPDH2ΔCBS•IMP complex with 200 mM NAD + . Structures of protein-inhibitor complexes were obtained by co-crystallization with IMP and inhibitor. In all five structures, the first 26 or 27 N-terminal residues are disordered in every protein chain. An approximately 20-residue portion of the active site flap is also disordered, as observed for the majority of the IMPDH structures reported to date.

Structure of Apo MtbIMPDH2ΔCBS
The apo structure contains four polypeptide chains in the asymmetric unit (rmsds for Cα atoms of chain A versus other three are from 0.10 to 0.12 Å). A potassium ion is bound between subunits, interacting with six main chain carbonyls, three from the loop containing active site C341 (G336, G338, and C341) and three from the C-terminal portion of the adjacent subunit (E551', S512', and H513'). This coordination is very similar to that observed in previously reported structures [21,41,42]. A phosphate ion originating from the crystallization buffer binds in the phosphate site of IMP, anchored by hydrogen bonding interactions with the backbone nitrogen atoms of S339, G376, G397 and S398 and the side chains of S339 and Y421. This interaction network is further supplemented by water-mediated contacts. Interestingly, K454 (residue in the active site flap) occupies the position where the linker portions of P41 and Q67 bind (see below, Fig 3A). This residue must move out of the active site when an inhibitor binds. Thus the conformational dynamics of the active site flap may contribute to the differences in inhibitor affinities among IMPDHs from different sources.
MtbIMPDH2ΔCBS complex with XMP and NAD + The 1.60 Å resolution crystal structure of the product/cofactor complex contains one protein chain per asymmetric unit. Although no potassium was present in the crystallization buffer and this ion is not found in the complex, the high quality electron density maps clearly show that the product XMP is present in the active site ( Fig 3B). This indicates that the enzyme in the crystal is catalytically competent. XMP is in essentially the same orientation as observed previously in IMPDH from V. cholerae (PDB id 4X3Z) [21]. The xanthine ring contacts C341, T343, M424, G425, and E458 and also has water-mediated interactions with the main chain nitrogen atom of G336 (Fig 3B). The phosphate group of XMP interacts with S339, G376, G397, S398, and Y421, and the sugar moiety forms hydrogen bonds with D374. These residues are highly conserved in all IMPDHs, with the exception of E458, which is replaced by glutamine in eukaryotic IMPDHs. As observed for VcIMPDHΔCBS•XMP•NAD + [21], the NAD + adenosine moiety binds in the A B -subsite located at the subunit interface and interacts with residues from both monomers (Fig 3C and 3D). For comparison, binding of adenosine in the A E -subsite of the hIMPDH2 complex with NAD + and a substrate analog, CPR (hIMPDH2•CPR•NAD + ; PDB code 1NFB) is shown in Fig 3E. In MtbIMPDH2ΔCBS•XMP•NAD + the adenine ring is in the anti-orientation with respect to the sugar ring. The adenine N3A atom contacts the side chain of N289, whereas the N6A amine makes one hydrogen bond with the carbonyl group of A483' and one water-mediated contact with the side chain of E458. In addition, one side of the adenine ring participates in van der Waals interactions involving V60' and P61' and the other side π/π stacks with H286 in an edge-to-face orientation (Fig 3C). The same set of interactions is present in VcIMPDHΔCBS•XMP•NAD + , although VcIMPDH has S256 in place of N289, L26' instead of V60' and the π/π stacking with the conserved histidine residue occurs in a face-to-face orientation. The N6A amine also makes a NH 2 /π interaction with the side chain of Y487' (part of the ISMS) in MtbIMPDH2ΔCBS. This contact is not present in VcIMPDHΔCBS•XMP•NAD + because VcIMPDH has L446' in place Y487' (Fig 3D). It is also important to note the adenosine portion in MtbIMPDH2ΔCBS•XMP•NAD + is positioned deeper into the binding pocket than in VcIMPDHΔCBS•XMP•NAD + and the cofactor is in a more compact orientation (C6A-C2N distance of 9.16 Å in MtbIMPDH2ΔCBS•XMP•NAD + versus 11.41 ± 0.11 Å in VcIMPDHΔCBS•XMP•NAD + [21,43]). Interestingly, the position of the adenine group in MtbIMPDH2ΔCBS•XMP•NAD + is more similar to the VcIMPDHΔCBS structure with NAD + and a mixture of IMP and a covalent intermediate (VcIMPDHΔCBS•IMP•NAD + ; PDB id  [21]). The cofactor in VcIMPDHΔCBS•IMP•NAD + has similarly compact conformation (C6A-C2N distance of 9.42 ± 0.10 Å), with the adenine ring in an anti-conformation. Perhaps NAD + changes orientation during the catalytic cycle prior to its dissociation from the active site, as suggested for VcIMPDH [21].
The interactions of the NAD + pyrophosphate are similar to those in previously reported cofactor structures. R108 makes two direct hydrogen bonds to O3 and O2A of the phosphate moiety and the same phosphate group makes one direct hydrogen bond to the amido group of A285 (part of the IMSM) and water-mediated bonds to the main chain amido and carbonyl groups of V261. The corresponding interactions, except for R108-O3/O2A, are also found in VcIMPDHΔCBS•XMP•NAD + [21].
The nicotinamide ring of NAD + stacks against the xanthine ring of XMP and the nicotinamide ribose is anchored via hydrogen bonds between the hydroxyl groups and the conserved D283, as observed previously in other IMPDHs (Fig 3C). However, the orientation of the nicotinamide portion is opposite from that observed in other cofactor complexes. In VcIMPDHΔCBS•XMP•NAD + , the nicotinamide ring is in the anti-conformation (Fig 3D) and the carboxamide group makes hydrogen bonds with two conserved glycine residues [21,44]. This conformation is consistent with hydride transfer to the pro-S position of NAD + [45,46]. However, the nicotinamide ring flips~180°and is in a syn-conformation in MtbIMPDH2ΔCBS•XMP•NAD + (with χ N torsion angle of 5.7°; Figs 1B and 3C). The carboxamide moiety makes direct hydrogen bonds with side chains of T343, E458, and Y487' (which is a part of the IMSM) and a water-mediated interaction with the O1N atom of the phosphate group and the N7A atom of adenine (Fig 3C and 3D). The carboxamide group also makes van der Waals contacts with A285 (also a part of the IMSM). As a result of the syn-conformation, the pro-R side of the nicotinamide ring now faces IMP with its C4N carbon aligned with the C2 of the IMP hypoxanthine ring (C4N-C2 distance of 3.28 Å).
We characterized the reaction of 2 H-IMP and NAD + to determine if the stereochemistry of hydride transfer catalyzed by MtbIMPDHΔCBS was pro-R as suggested by the crystal structure. As reported previously [45,46], hIMPDH2 transferred deuterium to the pro-S position of NAD + (S4 Fig). The deuterium also transferred to the pro-S position of NAD + in the MtbIMPDH2ΔCBS catalyzed reaction (S4 Fig). These observations indicate that the orientation of NAD + observed in the XMP complex is not competent for hydride transfer and may reflect cofactor conformation stabilized by the IMSM.

Inhibitor binding
Each structure MtbIMPDH2ΔCBS•inhibitor complex contains one protein chain per asymmetric unit with one molecule of IMP and one molecule of inhibitor bound. The active sites are ordered and the electron densities for ligands and protein are well defined (Figs 4 and 5). The overall structures of the inhibitor complexes are very similar, with r.m.s.d. for Cα atoms of A chains ranging from 0.23 to 0.46 Å. The structures of complexes with MAD1 and Q67 contain a potassium ion, as described for the apo protein. The crystallization buffer used to obtain the IMP•P41 did not contain potassium ions ( Table 4). The protein-IMP interactions are analogous to those described above for XMP, as well as for previously determined structure of IMP complexes [21,27,29,30], and will not be discussed further.
The inhibitors have modular structures consisting of two aromatic moieties connected by different linkers (1,2,3-triazole in MAD1, urea in P41 and an amide in Q67) (Fig 1C). One aromatic portion of the inhibitor stacks against the hypoxanthine ring of IMP and interacts with residues within the nicotinamide-binding subsite (the "leftside" rings, Fig 1C). The other aromatic moiety interacts with IMSM residue Y487' in the A B -subsite (the "rightside" rings; Fig  1C) and the linker bends around IMSM residue A285. Specific interactions for each inhibitor will be described below.

MtbIMPDH2ΔCBS•IMP•MAD1 complex
MAD1 is a mycophenolic acid-adenosine conjugate initially designed to target human IMPDHs [16]. It was the first reported inhibitor of MtbIMPDH2, although it is a much more potent inhibitor of the human enzymes. Other mycophenolic acid-adenosine conjugates with varying linkers did not inhibit MtbIMPDH2. The mycophenolic acid-derived moiety interacts with IMP and forms hydrogen bonds with T343, G334 and G336 (Fig 4A). These residues are conserved in all IMPDHs and the corresponding interactions are also present in eukaryotic enzymes (for example, in the E-XMP Ã •mycophenolic acid complex of Chinese hamster IMPDH2 (PDB id 1JR1) [41]). MAD1 does not interact with E458, in contrast to the CpIMPDH inhibitors (see below). The absence of these contacts may account for the relatively low affinity of this compound (S9 Table). The triazole linker of MAD1 engages in hydrogen bonding interaction via the N3 atom with the main chain amide nitrogen atom of A285, while the remainder of the triazole makes van der Waals contacts with R108 and T284 (Fig 4A). The position of the adenosine portion of MAD1 in the A B -subsite is superimposable with the position of the adenosine moiety of NAD + in the XMP•NAD + complex. Consequently, interactions with H286, N289, V60', P61', A483' and Y487' are maintained (Fig 4B). Thus, MAD1 adopts an orientation analogous to the folded cofactor conformation that is preferred in bacterial IMPDHs.
The divergence of the cofactor binding sites in bacterial and eukaryotic enzymes is illustrated by comparing the structures of the MtbIMPDH2ΔCBS•IMP•MAD1 with hIMPDH2 in complex with another MAD inhibitor, C2-MAD, and ribavirin 5'-monophosphate (RVP) (PDB id 1NF7; doi:10.2210/pdb1nf7/pdb [39]). C2-MAD is a second generation mycophenolic adenine analogue that has a methylenebis(phosphonate) linker. The linker and mycophenolic acid-derived portions of the inhibitors are positioned similarly in both enzymes. However, the adenosine portion of C2-MAD rotates away from the subunit interface and binds in the A Esubsite (Fig 4C). The C2-MAD adenine ring π/π stacks between the side chains of H253 and F282 (hIMPDH2 numbering) located within a single monomer (Fig 4C). Thus, C2-MAD assumes an orientation analogous to the extended NAD + conformation preferred in eukaryotic IMPDHs (Fig 3E). Therefore, interaction of the adenine moiety with the A E -subsite may account for higher affinity of MAD inhibitors with eukaryotic IMPDHs.

MtbIMPDH2ΔCBS •IMP•P41 complex
P41 has a urea linker connecting the 3-isoprenyl-α,α-dimethylbenzyl (left side ring) and 4-chloro-3-N,N-dimethylbenzamide [26] (right side ring; Fig 1C). The hypoxanthine ring of IMP interacts with the sp 2 center and the aromatic centroid of the left side ring (Figs 1C and  5A). Both nitrogen atoms of the urea linker form hydrogen bonds with the side chain of E458 (N1/P41-OE2/E458 and N2/P41-OE2/E458 distances of 3.27 and 3.08 Å, respectively) ( Fig  5A). The interactions with E458 are observed in other inhibitor complexes and are likely to be important for inhibitor potency [21,27,29,30]. Within the A B -subsite, the right side ring interacts with the ring of Y487' in an orientation that is midway between face-to-face and edge-toface geometries. The right side ring also makes contacts with H286, V60', and P61'.
The importance of the 4-chloro substituent for antitubercular activity can be explained by the presence of a pair of orthogonal halogen/hydrogen bonds [47,48]. The Cl atom interacts   (Fig 5A). This carbonyl also makes a hydrogen bond with the side chain of S57' (O/G486'-OG/S57' distance of 2.85 Å) at an angle of 90°relative to the halogen bond. The halogen substituent also contacts the imidazole ring of H286. The structure explains the diversity of carboxamide substitutions in the P compounds with antitubercular activity. The carboxamide group of P41 does not participate in any hydrogen bonds with the protein. The two methyl groups sit in a hydrophobic pocket formed by T284, V261, A285, N289, L291 and V292. These hydrophobic interactions can explain the higher affinity of substituted amides ( Table 2, compare P34 and P67, P146 and P150).

MtbIMPDH2ΔCBS •IMP•Q67 complex
Compound Q67 contains 2,3-dichloroaniline (left side ring) and 2-(4-pyridyl)-1,3-benzoxazole moieties (right side ring) connected with an amide linker [27] (Fig 1C). The leftside ring interacts with IMP via π/π stacking and C-X/π contacts involving the 2-chloro substituent. The 3-chloro substituent contacts the carbonyl atom of M424 and interacts with G425 ( Fig 5B). These interactions explain the prevalence of the 2,3-dichloro substituents in the set of Q compounds with antitubercular activity. As in P41, the nitrogen atom of compound Q67 amide linker hydrogen bonds with one of the side chain oxygen atoms of E458 but the interaction for Q67 is stronger (2.79 Å versus 3.08 and 3.27 Å for P41).
As observed for the other inhibitors, the 4-pyridyl-1,3-benzoxazole moiety of Q67 is held in place by interactions with H286, V60', P61', and Y487' in the A B -site. In addition, Q67 makes several unique contacts: the N3 atom of the 1,3-benzoxazole group interacts with the main chain nitrogen atom of H286 and two water molecules connect the N4 atom of the 4-pyridyl substituent with the main chain nitrogen atom of R290 and the carbonyl oxygen atom of S57' (Fig 5B). Similar water-mediated interactions involving the pyridyl moiety were previously observed in the structures CpIMPDH and BaIMPDH complexes with another Q-series inhibitor, Q21 [21,27]. The additional interactions of chloro substituents, the short hydrogen bond with E458, and the network of water-mediated contacts can account for the high potency of Q67. The structure also explains why the stereochemistry of the linker is essential. The (S)methyl group faces away from E458 and interacts with L455 and M430. The methyl group in the (R) isomer is most likely oriented towards E458 and thus disrupts the crucial hydrogen bond interaction between the amide linker and E458, making this isomer inactive.

Discussion
The structure of MtbIMPDH2ΔCBS•XMP•NAD + confirms our recent findings that bacterial IMPDHs bind NAD + in an unusual compact conformation with the A B -subsite located at the interface between two monomers. The different locations of the A B and A E sites account for the selectivity of the CpIMPDH inhibitors for bacterial IMPDHs. All high affinity CpIMPDH inhibitors bind to bacterial enzymes using the A B site. In contrast, it appears that MAD inhibitors bind with high affinity to human IMPDH using the A E site and with low affinity to bacterial enzymes using A B site. Interestingly, the nicotinamide ring in the MtbIMPDH structure is in a syn-conformation, which is not compatible with the pro-S hydride transfer observed in the reaction. Similar syn NAD(H) conformations have been observed among other pro-S stereospecific enzymes, such as transhydrogenase [49] and UDP-galactose 4-epimerase [50]. We link dashed lines. For panels A and B 2mF o -DF c electron density map contoured at the 1 σ level for each inhibitor is shown on the right. Atoms discussed in text are labeled. (C) Overlay of three MtbIMPDH2ΔCBS inhibitor complexes. IMP is omitted for clarity. Color designations as follows: for MAD1 as in Fig 4A; for P41 and Q67 as in panels (A) and (B), respectively. doi:10.1371/journal.pone.0138976.g005 the ability to bind the syn-conformation with the presence of the IMSM that is essential for inhibitor binding [10,30,31]. While hydride transfer cannot occur in the syn-conformation, it is possible that this conformation has another regulatory role.
New antibiotics to treat tuberculosis are urgently needed. Our work identifies inhibitors of MtbIMPDH2 with encouraging antitubercular activity. We also report the first crystal structures of this promising antimicrobial drug target, including the apo form and complexes with XMP/ NAD + and three structurally distinct inhibitors. These structures will greatly facilitate the further development of MtbIMPDH2-targeted antibiotics. Our MtbIMPDH2ΔCBS inhibitor complexes provide important insights into the interactions that modulate affinity as well identify possible locations for further inhibitor optimization for both potency and cellular accumulation. Interactions with the hypoxanthine moiety of IMP, E458, and A B -subsite are especially important for high affinity and selectivity. It is important to note that while the active site flap is partially disordered in all MtbIMPDH2ΔCBS structures, different conformations of the flap are observed in the apo structure and the structures of the complexes. Thus, the flap may transiently interact with the inhibitor and these interactions may also contribute to inhibitor affinity.

Material and Methods Materials
IMP was purchased from Acros Organics. NAD + and NADH were purchased from Roche and Sigma, respectively. Tris, and common chemicals were purchased from Sigma. KCl and trichloroacetic acid were purchased from Fisher. CpIMPDH and BaIMPDH were purified as previously described [27,32]. The synthesis of the CpIMPDH inhibitors has been reported previously [23][24][25][27][28][29][30]37]. The synthesis of MAD1 is described in [16]. Crystallization reagents were purchased from Hampton Research and Microlytic.

Values of cLogP and tPSA
Values of clogP and tPSA were calculated in ChemBioDraw (Cambridgesoft Inc.). The relevant ionizations at pH 7.4 were included in the structures [37].

Cloning
The CBS domain deletion mutant (MtbIMPDH2ΔCBS) was constructed via the megaprimer cloning method [51]. Wild type MtbIMPDH2 clone in vector pMCSG7 [27] was used as a template. The E126-R252 (ΔCBS) deletion primer and the MtbIMPDH2 coding sequence forward primer were used to amplify a region of MtbIMPDH2-pMCSG7 ranging from the residue M1 to residue V261 with 5' LIC overhang, while replacing codons for residues E126-R252 with codons for GG. The resulting product was used as a megaprimer in the whole plasmid synthesis reaction, with MtbIMPDH2-pMCSG7 as template and a reverse primer encoding 3' end of MtbIMPDH2. KOD Hot Start DNA polymerase kit (EMD Millipore) was utilized in a PCR reaction. Cycling was performed at 95°for 3 min, followed by 95°for 40 sec, 53°for 40 sec, 72°f or 1.5 min for 32 cycles. The PCR product was treated with T4 polymerase (Promega), annealed into pMSCG7 vector, transformed into E. coli BL21(DE3) cells carrying the pMAGIC plasmid encoding rare E. coli tRNA (Arg (AGA/AGG)) [52] and the resulting clone sequenced. was added to initiate the reaction. After 2h incubation at room temperature, the protein was separated from small molecules by centrifugation using Amicon centrifugal filter (Millipore, 10K cutoff), and the small molecule mixtures were directly analyzed by 400MHz 1 H NMR with water suppression using presaturation pulse.

MIC determinations
MIC is concentration that completely inhibits growth. Compounds MICs were determined as previously described [53]. MIC values were determined in at least triplicate according to the broth microdilution methods using compounds from DMSO stock solutions. Isoniazid was used as a positive control and DMSO was utilized as a negative control. Isolated Mtb cells (ATCC 27294) were cultured to an OD 0.2-0.3 in the required medium, then diluted to deliver approximately 1 x 10 4 bacteria per well of a 96 well clear round-bottom plate. Plates were read after 1 week with an inverted enlarging mirror plate reader and graded as either growth or no growth. GAST/Fe medium (per liter) consisted of 0.3 g of Bacto Casitone (Difco), 4.0 g of dibasic potassium phosphate, 2.0 g of citric acid, 1.0 g of L-alanine, 1.2 g of magnesium chloride hexahydrate, 0.6 g of potassium sulfate, 2.0 g of ammonium chloride, 1.80 ml of 10 sodium hydroxide, and 10.0 ml of glycerol, 0.05% Tween 80 and 0.05 g of ferric ammonium citrate adjusted to pH 6.6. 7H9/glycerol/glucose/BSA/Tween medium consisted of Middlebrook 7H9 broth base supplemented per liter with 0.2% glucose, 0.2% glycerol, 0.5% BSA fraction V, 0.08% NaCl and 0.05% Tween 80.
Details of protein expression, purification, crystallization, data collection, X-ray structure solution and refinement, steady state kinetic measurements and inhibition experiments are listed in S1 Appendix.
Supporting Information S1 Appendix. Details of protein expression, purification, crystallization, data collection, Xray structure solution and refinement, steady state kinetic measurements and inhibition experiments. MtbIMPDH3 (guaB3; gi: 444896966). Secondary structure elements derived from MtbIMPDH2ΔCBS (PDB code 4ZQR) are depicted as arrows (representing β-strands), coils (representing αand 3 10 -helices), TT (strict β-turns) and TTT (strict α-turns). The location of CBS domain is shown as a green line. The position of catalytic Cys residue is indicated by a yellow rectangle. It is important to note that MtbIMPDH3 (guaB3) does not posses the catalytic Cys residue suggesting that this protein may not be an IMPDH enzyme. (B) Sequence alignment of bacterial and human IMPDHs discussed in this study. The sequences include MtbIMPDH1 (guaB1; gi: 15608980), MtbIMPDH2 (guaB2; gi: 15610547), B. anthracis str. Ames (gi: 30253523), V. cholera O1 biovar (gi: 15640786), human type I (gi: 217035148), human type II (gi: 66933016) and C. parvum (gi: 323510309). MtbIMPDH3 was omitted. Secondary structure elements derived from BaIMPDH (PDB code 3TSB) are depicted as in panel A. The location of CBS domain is shown as a green line. The position of catalytic Cys residue is indicated by a yellow rectangle. Positions of residues involved in binding of the NAD + adenosine moiety in bacterial (A B -subsite) and eukaryotic (A E -subsite) enzymes are indicated by purple and black rectangles, respectively. In both panels identical residues are highlighted in red, and similar residues are shown as red letters. The alignment was generated using MultiAlin [54] and ESPript [55] programs. (TIFF)  Table. Structures of inactive A series amide derivatives. All values are the average of at least two determinations unless otherwise noted. a. Data from [24]. b. Data from [23]. c. Data from [37]. d. Single determination. (DOCX) S2 Table. Structures of inactive A series triazole derivatives. a. Data from [24]. b. Data from [37]. c. Single determination. (DOCX) S3 Table. SAR of enzyme inhibition for C series benzimidazole derivatives. n.d. = not determined. a. CpIMPDH data from [30]. b. Data from [25]. c. BaIMPDH data from [37]. c. Single determination. (DOCX) S4 Table. Structures of inactive D series phthalazinone derivatives. n.d. = not determined. a. Data from [28]. b. Data from [37]. c. Single determination. (DOCX) S5 Table. Structures of inactive P compounds. n.d. = not determined. a. Data from [26]. b. Data from [37]. (DOCX) S6 Table. P series: SAR of enzyme inhibition for the isopropyl and urea group. n.d. = not determined. a. Data from [26]. b. Data from [37]. (DOCX) S7 Table. Structures of inactive Q compounds. n.d. = not determined. n.d. = not determined a. Data from [27]. b. Data from [37]. (DOCX) S8 Table. CpIMPDH inhibitors with antitubercular activity: comparison of enzyme inhibition. n.d. = not determined. a. Data from [26]. b. Data from [27]. c. Data from [37]. d. Single determination. e. Two determinations. (DOCX) S9 Table. Inhibition of MtbIMPDH2ΔCBS. Ã Indicates tight binding conditions. C, competitive inhibition; UC, uncompetitive inhibition, NC, noncompetitive inhibition. (DOCX)