Rational Design of Broad Spectrum Antibacterial Activity Based on a Clinically Relevant Enoyl-Acyl Carrier Protein (ACP) Reductase Inhibitor*

Background: The FabI inhibitor CG400549 is a promising new anti-staphylococcal drug candidate with recently validated human efficacy. Results: We revealed the molecular determinants conferring S. aureus FabI selectivity to rationally design a compound with an improved antibacterial activity spectrum. Conclusion: The 4-pyridone PT166 represents a critical step toward Gram-negative and mycobacterial coverage. Significance: We provide an approach to expand the spectrum of antimicrobial activity. Determining the molecular basis for target selectivity is of particular importance in drug discovery. The ideal antibiotic should be active against a broad spectrum of pathogenic organisms with a minimal effect on human targets. CG400549, a Staphylococcus-specific 2-pyridone compound that inhibits the enoyl-acyl carrier protein reductase (FabI), has recently been shown to possess human efficacy for the treatment of methicillin-resistant Staphylococcus aureus infections, which constitute a serious threat to human health. In this study, we solved the structures of three different FabI homologues in complex with several pyridone inhibitors, including CG400549. Based on these structures, we rationalize the 65-fold reduced affinity of CG400549 toward Escherichia coli versus S. aureus FabI and implement concepts to improve the spectrum of antibacterial activity. The identification of different conformational states along the reaction coordinate of the enzymatic hydride transfer provides an elegant visual depiction of the relationship between catalysis and inhibition, which facilitates rational inhibitor design. Ultimately, we developed the novel 4-pyridone-based FabI inhibitor PT166 that retained favorable pharmacokinetics and efficacy in a mouse model of S. aureus infection with extended activity against Gram-negative and mycobacterial organisms.

Determining the molecular basis for target selectivity is of particular importance in drug discovery. The ideal antibiotic should be active against a broad spectrum of pathogenic organisms with a minimal effect on human targets. CG400549, a Staphylococcus-specific 2-pyridone compound that inhibits the enoyl-acyl carrier protein reductase (FabI), has recently been shown to possess human efficacy for the treatment of methicillin-resistant Staphylococcus aureus infections, which constitute a serious threat to human health. In this study, we solved the structures of three different FabI homologues in complex with several pyridone inhibitors, including CG400549. Based on these structures, we rationalize the 65-fold reduced affinity of CG400549 toward Escherichia coli versus S. aureus FabI and implement concepts to improve the spectrum of antibacterial activity. The identification of different conformational states along the reaction coordinate of the enzymatic hydride transfer provides an elegant visual depiction of the relationship between catalysis and inhibition, which facilitates rational inhibitor design. Ultimately, we developed the novel 4-pyridone-based FabI inhibitor PT166 that retained favorable pharmacokinetics and efficacy in a mouse model of S. aureus infection with extended activity against Gram-negative and mycobacterial organisms.
Staphylococcus aureus can cause a variety of bacterial infections ranging from common skin infections to life-threatening pneumonia or bacteremia (1). In particular, methicillin-resistant S. aureus (MRSA) 6 poses an imminent risk to immunocompromised patients in healthcare settings all over the world. In addition, the incidence of community-acquired MRSA infections has increased among otherwise healthy individuals (1,2). The initial occurrence of S. aureus strains resistant to vancomycin, an antibiotic used to treat severe MRSA infections (3), underlines the urgent need for novel anti-staphylococcal drugs.
Isoniazid, a first-line prodrug for the treatment of tuberculosis, inhibits the type II fatty acid biosynthesis pathway of Mycobacterium tuberculosis (4). The clinical success of isoniazid validates the type II fatty acid biosynthesis pathway as an important target for the development of new antibiotics (5). Bacterial fatty acid biosynthesis differs from its mammalian counterpart and is pivotal for the production of several cellular components, such as phospholipids (6,7). In the last step of the type II fatty acid biosynthesis elongation cycle, the enoyl-acyl carrier protein (ACP) reductase (FabI) catalyzes the reduction of the trans-2-enoyl-ACP double bond ( Fig. 1) (8,9). After the discoveries that activated isoniazid and the antimicrobial agent triclosan inhibit FabI (10 -14), this enzyme was investigated as a potential broad spectrum target conserved among many pathogenic organisms (15,16). The finding that distinct bacteria such as Streptococcus pneumoniae or Vibrio cholerae utilize FabI isoenzymes, including FabK (17), FabL (18), and FabV (19) or can take up exogenous fatty acids from the host blood serum to circumvent the inhibition of FabI (20), has provided some limitations with regards to antibacterial coverage (15). Nevertheless, for several clinically relevant pathogens, such as S. aureus, Staphylococcus epidermidis, Escherichia coli, Haemophilus influenzae, Francisella tularensis, and M. tuberculosis, FabI remains a very promising drug target because these organisms require endogenous fatty acids (4,6,21,22) and contain FabI as the sole enoyl-ACP reductase (23,24). Accordingly, several drug design programs have been continued, and three S. aureus FabI (saFabI) inhibitors with different scaffolds (Fig. 1) have been advanced to clinical trials (25).
Extensive studies have been conducted on the molecular recognition of diphenyl ethers by saFabI (6,16,31). However, little is known about the binding of pyridone inhibitors such as CG400549. These compounds were intended to improve pharmacokinetic properties by replacing the metabolically labile phenol of the diphenyl ether skeleton with a stable pyridone (25). Despite the structural similarity between pyridones and diphenyl ethers, we reveal fundamental differences in the molecular recognition of these two scaffolds by FabI enzymes, which explain their respective narrow and broad spectrum activities. We report the first pyridone-bound crystal structures of saFabI, in addition to a comprehensive kinetic profile of inhibition. In particular, we solved the CG400549-bound saFabI and E. coli FabI (ecFabI) structures, which allowed us to rationalize the selectivity of this compound for the S. aureus homologue. Guided by this information, we sought to develop a compound that combined the pharmacokinetic stability of a pyridone with the broad spectrum characteristics of diphenyl ethers. The novel 4-pyridone inhibitor PT166 represents a significant step toward this goal, exhibiting extended spectrum antimicrobial activity against E. coli, F. tularensis, Proteus mirabilis, and M. tuberculosis, while retaining in vivo efficacy and favorable pharmacokinetics in a murine S. aureus thigh infection model.
Crystallization-Prior to concentrating saFabI samples from 2 to 15-19 mg/ml, the protein was incubated for 2 h at 20°C with a 12-fold molar excess of NADPH and a 20-fold molar excess of inhibitor dissolved in DMSO (CG400549 or PT173, respectively). Diffraction-quality crystals were grown in vapor diffusion experiments with a precipitant solution containing 0.1-0.2 M Li 2 SO 4 and 20 -24 w/v % PEG 3350. For CG400549, we obtained crystals of space group P2 1 2 1 2 1 with two different sets of cell parameters (the resulting structures were named  CG400549-I and CG400549-II; supplemental Table S1).
Similarly, ecFabI samples at a concentration of 13 mg/ml were incubated for 2 h at 4°C with a 10-fold molar excess of NADH and a 20-fold molar excess of CG400549 or PT166 (dissolved in DMSO), respectively. Hanging drop vapor diffusion experiments yielded diffraction-quality crystals in drops composed of 1 l of this protein/ligand mixture and 1 l of precipitant solution (0.2 M NH 4 Ac, 0.1 M CAPS, pH 10.5, and 20 w/v % PEG 8000 in the case of CG400549; 0.2 M NH 4 Ac, 0.1 M sodium citrate, pH 5.6, and 10 w/v % PEG 8000 in the case of PT166).
bpFabI samples at a concentration of 10 -30 mg/ml were incubated for 2 h at 20°C with a 10-fold molar excess of NAD ϩ and a 20-fold molar excess of PT155 (dissolved in DMSO). Sitting drop vapor diffusion experiments yielded diffraction-quality crystals in drops composed of 0.3 l of the protein/ligand mixture and 0.3 l of precipitant solution (20 w/v % PEG 3350 and 200 mM (NH 4 ) 2 HPO 4 ).
Data Collection and Structure Determination-Prior to flash-freezing in liquid nitrogen, the saFabI ternary complex crystals were successively transferred into solutions composed of mother liquor supplemented with 10 and 25 v/v % ethylene glycol, respectively. Diffraction data were collected at the BESSY II MX beamline 14.1 (36) ( ϭ 0.918 Å, T ϭ 100 K) equipped with a MarMosaic 225 detector, integrated with Imosflm (CG400549-II and PT173) (37) or XDS (CG400549-I) (38), and further processed using Scala (39). The CG400549 structures were solved by molecular replacement with Phaser (40) using our previously published saFabI structure (PDB code 4ALK; lacking amino acids 196 -202) as search model (6). For PT173, the fully refined CG400549-I structure was used as a template for molecular replacement. To avoid model bias, R free flags were assigned in thin resolution shells (CG400549-I and -II) or copied from the search model (for PT173). The final structures were obtained by several alternative cycles of model building in Coot (41) and refinement in Refmac 5 (CG400549bound structures) (42) or Phenix (PT173) (43), respectively (including noncrystallographic symmetry (PT173, CG400549-II) and TLS refinement (44)). Cofactors and inhibitors could be unambiguously assigned based on the 2F o Ϫ F c and F o Ϫ F c electron density maps.
Crystals of the CG400549 and PT166 ternary ecFabI complexes were cryo-protected using the corresponding mother liquor supplemented by 25 or 30 v/v % ethylene glycol, respectively. Using a Pilatus 6M detector, diffraction data were collected at MX beamline 14.1 of the BESSY II synchrotron ( ϭ 0.918 Å, T ϭ 100 K) and at beamline 23-1 of the European Synchrotron Radiation Facility ( ϭ 1.064 Å, T ϭ 100 K), integrated using XDSAPP (45), and scaled with Scala. Initial phases were determined by molecular replacement in Phaser with our previously published ecFabI structure (PDB code 1QSG) as the search model (13). Model building in Coot and refinement using Refmac 5 (including TLS refinement) yielded the final structure. We did not model the amino acids 193 to 209/211 due to only partial and very weak electron density for the two monomers in the asymmetric unit. In addition, we note that the CG400549 electron density in subunit B was sufficient for model building but inferior compared with monomer A. Dictionaries for the cofactors and inhibitors of the S. aureus and E. coli FabI structures were computed using Grade (46,47).
bpFabI crystals were cryo-cooled in cryo-protectant containing 25 v/v % glycerol in the mother liquor. Data collection was performed at an in-house x-ray generator (MicroMax-007 HF, Rigaku) at a wavelength of 1.54 Å and recorded with an imaging plate detector (R-Axis HTC, Rigaku). Data were integrated with Imosflm and scaled in Scala. Molecular replacement was performed in Phaser using the PDB entry 3EK2 as a template. For refinement in Refmac and finally Phenix TLS parameters were created using the TLSMD server (44) and a library file supplying restraints for the cofactor and inhibitor was generated by the Prodrg server (48). The structure was refined until convergence (R/R free ϭ 14/16%) and validated using the Molprobity server (49).
To avoid model bias, omit maps were calculated prior to inclusion of cofactors and inhibitors. Data collection and refinement statistics are given in supplemental Table S1 (saFabI) and supplemental Table S2 (ecFabI and bpFabI). Distances and angles were measured for all subunits of the asymmetric unit and are given as mean values Ϯ S.D. Structural figures were prepared using PyMOL (50).
Inhibition Kinetics-Kinetics were performed on a Cary 100 spectrophotometer (Varian) at 20°C. Reaction velocities were measured by monitoring the oxidation of NAD(P)H to NAD(P) ϩ at 340 nm (⑀ ϭ 6220 M Ϫ1 cm Ϫ1 ). For saFabI, the reaction mixture was identical to that described previously for progress curve experiments (6). For ecFabI, the final reaction mixture contained ecFabI ( Table S3 or (31).
A t and A 0 are the absorbance at time t and time 0; v i and v s are the initial and steady-state velocities, and k obs is the pseudofirst order rate constant for the approach to steady state. v u is the control, uninhibited velocity, and K i app is the IC 50 [E] T and [I] T are the total enzyme and inhibitor concentrations, respectively. Thermal Shift Assay-ThermoFluor experiments were carried out in 96-well plates (Concord) using the CFX96 Real Time PCR Detection System and C1000 Thermal Cycler (Bio-Rad), as described previously (31).
Docking Studies-A computational docking and scoring procedure was used to generate putative binding modes for all pyridone inhibitors investigated in this study. The binding poses were generated with FlexX (BioSolveIT, Sankt Augustin, 2009), version 3.1.4 (53), and rescored with DrugScore X (G. Neudert and G. Klebe, University of Marburg, 2008), version 0.21, which builds on DrugScore and utilizes the DrugScore CSD potentials (54 -56). To account for the flexibility of the substrate-binding loop, we docked all inhibitors into subunits A and C of the saFabI CG400549-I structure, which represent the two experimentally observed states (for details see "Results"). The selection of the most likely binding pose and receptor was based on the DrugScore X score combined with visual inspection (supplemental Fig. S1). A comparison with the available experimental binding modes revealed root mean square deviations below 1.1 Å for the 10 best ranked binding poses, with 0.7 Å (CG400549) and 0.8 Å (PT173) for the top ranked pose, respectively (supplemental Fig. S1, A and B). These re-/cross-docking experiments confirm the validity and reliability of our computational approach.
Determination of MIC Values-MIC values were determined with the microbroth dilution assay according to the Clinical and Laboratory Standards Institute methods for antimicrobial susceptibility tests for aerobically growing bacteria (60).
Selection for Resistance-S. aureus RN4220 was grown at 37°C in Mueller-Hinton (MH) broth to late log phase (A 600 ϭ 1.2). 200 l of culture was plated on MH agar containing PT166 (2 g/ml; 5ϫ MIC). After 48 h, five resistant colonies were randomly selected, and their phenotypes were confirmed by re-growth on the same medium containing PT166. The genomic DNA was extracted and purified using the Quick g-DNA Mini Prep kit (ZYMO Research). The S. aureus fabI genes from the PT166-resistant mutants were characterized by double-stranded nucleotide sequencing of PCR products using the following primers: saFabI forward (5Ј-CTAATTAG-GCATATGTTAAATCTTGAAAACAAAACG-3Ј) and saFabI reverse (5Ј-GTAAGTGCTCGAGTTATTTAATTGCGTG-GAATCC-3Ј). Sequencing reactions were performed with the ABI PRISM BigDye Terminator Cycle Sequencing Ready Reaction kit (Applied Biosystems, Foster City, CA), and sequencing data were obtained using an Applied Biosystems 3730 DNA sequencer.
In Vivo Pharmacokinetics-Pharmacokinetic (PK) studies were conducted in female ICR mice via intraperitoneal administration of PT04 (200 mg/kg dose) or PT166 (100 mg/kg dose) in a vehicle of 40 v/v % PEG 400, 40 v/v % EtOH, 20 v/v % H 2 O. Blood samples were collected from each animal at eight time points (5,15, and 30 min and 1, 2, 4, 8, and 24 h post-injection). Three mice were sampled per time point. Plasma concentrations for each sample were measured by LC/MS/MS, and PK parameters were calculated with WinNonlin (Pharsight Corp., Mountain View, CA).
In Vivo Efficacy-Antibacterial efficacy of 5-hexyl-2-phenoxyphenol (PT04) and PT166 (Table 1) was evaluated in a neutropenic mouse thigh infection model. Six-week-old male Swiss Webster mice weighing 23-27 g were rendered neutropenic by intraperitoneal injection of cyclophosphamide 4 days (150 mg/kg) and 1 day (100 mg/kg) prior to infection. Previous studies have shown that this can produce severe neutropenia in mice for at least 5 days (61).
MRSA strain BAA1762 was cultured in MH broth to mid-log phase (A 600 ϭ 0.4; 3 ϫ 10 8 cfu/ml) and harvested by centrifugation. Cell pellets were resuspended in freshly sterilized heart infusion (BHI) broth to a final inoculum of 1 ϫ 10 7 cfu/ml. 50 l of this suspension (5 ϫ 10 5 cfu) was injected intramuscularly into the left thigh of each mouse, and 50 l of BHI broth was injected into the right thigh as control. The drug was administered by subcutaneous injection at 1 and 12 h post-infection (100 mg/kg dose per injection). Mice were euthanized 24 h post-infection. Each thigh muscle was subsequently collected and homogenized in saline containing 10 w/v % BHI broth. The bacterial load was determined by counting colony-forming units of serial dilutions plated on MH sheep blood agar.
For in vivo efficacy and PK studies, all animals were maintained in accordance with criteria set by the American Association for Accreditation of Laboratory Animal Care. This study was approved by the Institutional Animal Care and Use Committee (IACUC) at Stony Brook University.

CG400549 and Related 2-Pyridones Selectively Inhibit
Growth of S. aureus-The 2-pyridone CG400549 exhibits potent activity against S. aureus in contrast to many other bacteria such as E. coli, Listeria monocytogenes, Helicobacter pylori, and Pseudomonas aeruginosa (29). We confirmed this disparity by comparing growth inhibition of S. aureus and E. coli with various 2-pyridone inhibitors (Tables 1 and 2). In contrast to the Staphylococcus-specific 2-pyridones, diphenyl ethers are active against a broader spectrum of microorganisms. It is therefore important to elucidate whether the narrow spectrum behavior of 2-pyridones can be attributed to FabI-specific differences between species and why these compounds behave differently compared with the structurally similar diphenyl ethers.
In Contrast to the Diphenyl Ethers, 2-Pyridones Bind to the E-NADPH Complex and Exhibit Fast-on Kinetics-Very recently, we found that diphenyl ethers bind exclusively to the E-NADP ϩ enzyme-product complex generated via catalysis (31). Despite the obvious structural similarity between pyridone and diphenyl ether inhibitors (Fig. 1), thermal shift assays revealed preferable binding of pyridones to the E-NADPH complex ( Fig. 2A and Table 1). Thus, these inhibitors are uncompetitive with respect to NADPH but competitive with respect to the acyl substrate (Fig. 2B). Our kinetic studies confirmed this pattern of saFabI inhibition, showing enhanced inhibition with increasing concentrations of NADPH and decreasing concentrations of the substrate trans-2-octenoyl-CoA used in the assay (data not shown). Moreover, we were able to co-crystallize pyridones with saFabI using the reduced form of the cofactor. Accordingly, the saFabI target can be productively inhibited at two different stages of the catalytic cycle, as exemplified by pyridones and diphenyl ethers (Fig. 2B). This finding is of particular importance for the development of improved FabI inhibitors because most known scaffolds bind adjacent to the nicotinamide ring of the cofactor (16). The absence or presence of a buried positive charge in the E-NADPH versus E-NADP ϩ binary complex likely will have a critical influence on the true affinity of potential inhibitors (62). However, the relative potencies of inhibitors with different modes of action are highly dependent on the experimental assay condition (see below). Because NADPH binds with higher affinity to the enzyme compared with NADP ϩ , the E-NADPH binary complex exists at a higher population for most experimental assay conditions (8, 31). Thus, even though CG400549  exhibits weaker true thermodynamic affinity (K i ) than most diphenyl ethers, its apparent inhibition constant (K i app ) is much more potent in our standard enzyme assay (Table 1), requiring classical tight binding analysis ( Fig. 2C) (31).
Pyridone and diphenyl ether saFabI inhibitors also differ with respect to their apparent association and dissociation kinetics. Diphenyl ethers exhibit slow binding kinetics and bind with long residence times to their target (Fig. 2, E and F) (6,31). In contrast, progress curves of saFabI in the presence of pyridones are linear, displaying apparent rapid-onset kinetics (Fig. 2D). This is likely attributed to the higher population of E-NADPH compared with E-NADP ϩ . In fact, an estimate of the actual association rate constant yields a value in the same range as for the diphenyl ethers (31). Moreover, the rapid appearance of activity following jump dilution is highly indicative of fast off kinetics (Fig. 2F). This is consistent with the weaker true thermodynamic affinity (K i ) of pyridones compared with diphenyl ethers. In other words, differences in k off are driven by K i rather than k on .
The mode of action can have significant implications for cell growth inhibition. In open systems, substrate accumulation may eventually diminish the effect of competitive inhibitors (63). Therefore, it is important to consider the relationship between substrate concentration and K i app (Fig. 3A). In Fig. 3B, we observe a linear correlation in the double logarithmic plot of K i versus MIC for both 2-pyridones and diphenyl ethers, consistent with on-target effects. However, 100-fold higher overall ternary complex affinity (K i ϫ K d, NADP(H) ) is needed for 2-pyridones to obtain a similar cellular potency as diphenyl ethers. Although this could be due to differences in cell permeability, this observation may also arise from substrate accumulation. The MIC is a thermodynamic parameter that is essentially equivalent to a physiological apparent inhibition constant (K i app ). For competitive inhibitors, such as the pyridone compounds, the presence of high substrate concentrations may weaken the apparent affinity. Alternatively, because the rate of substrate reduction (k cat ϭ 40 min Ϫ1 ) will increase as substrate concentration increases (K S ϭ 0.75 mM), a higher proportion of E-NADP ϩ is formed, which can bind to the uncompetitive diphenyl ether compounds. At a substrate concentration range of 40 -80 times K S , the relative K i app values are very predictive of the pattern of MIC values for both 2-pyridones and diphenyl  (78). In the case of saFabI, the hydride (shown in red) is delivered by the reducing agent NADPH (6,8,9). During the second step of the reaction, the enolate intermediate is protonated, which leads to the formation of the final acyl-ACP product (the proton is depicted in blue). Examples for the most important FabI inhibitor classes are given in the boxes along with their binding mode in the saFabI active site pocket (PDB codes 4FS3 and 4ALI (6, 23); the CG400549 structure was solved during this study, PDB code 4CV1). For each of these inhibitor scaffolds, one compound is currently in clinical trials (AFN-1252, MUT056399, and CG400549) (25). One common feature of these FabI inhibitors is the formation of a hydrogen bond to Tyr-157 and the cofactor NADP(H). The oxygen atoms involved in this central interaction are colored in magenta. In addition, all depicted inhibitors directly interact with Ala-97 (the interacting inhibitor atoms are highlighted in orange).

FIGURE 2. Different mechanisms of saFabI inhibition.
A, thermal shift analyses of saFabI bound to NADPH, NADP ϩ , and/or inhibitor (CG400549). The measurement variability is approximately Ϯ0.2°C. B, distinct mechanisms of saFabI inhibition. Based on our recent kinetic and structural results, we reasoned that the enzyme binds NADPH first followed by the substrate (31). In contrast to diphenyl ethers (highlighted in red), which bind to the enzyme-product complex generated via catalysis (E-NADP ϩ , depicted in blue) (31), pyridone compounds (A), PT166 (supplemental Fig. S2) and AFN-1252 (71) (all shown in red), preferentially inhibit saFabI at the enzyme-substrate complex state (E-NADPH, depicted in blue). C, representative plot of fractional velocity (v/v 0 ) as a function of inhibitor concentration for a potent pyridone (CG400549, in this example). The shape is characteristic of tight-binding inhibition. The best fit curve to the Morrison  . D, representative set of forward progress curves with pyridone-based inhibitors of saFabI. The plot depicts rapid-onset inhibition at different PT173 concentrations. E, as a reference, this plot illustrates the slow-onset inhibition of saFabI by the diphenyl ether PT04 (31). Note the clear observation of curvature that is absent in D. F, jump dilution curve for CG400549 (black) following preincubation with NADPH and saFabI. The jump dilution curve for the slow off diphenyl ether inhibitor PT52 (gray) following preincubation with NADP ϩ and saFabI is shown as reference (full recovery of activity; t R ϭ 30 min, where t R is the residence time) (31). The lack of curvature for CG400549 is consistent with rapid off kinetics. ethers (Fig. 3C). This allows us to translate K i to K i app values that are readily compared despite different modes of action. The true level of substrate accumulation in the cell is likely to be lower than predicted by this calculation. The cell contains longer chain substrates with faster turnover rates (6), which will increase the apparent affinity of diphenyl ethers. For instance, if we assume an average FabI turnover rate of 400 min Ϫ1 , the predicted substrate concentration range would be 4 -8 times K S , which is a much more reasonable estimate.
The utility of a scaffold for lead optimization is related, in part, to its intrinsic potency. We define this as the potency of a relatively unmodified scaffold, i.e. the starting point that determines how much affinity optimization is needed. For instance, PT170 and PT53 represent relatively unmodified 2-pyridone and diphenyl ether scaffolds, respectively ( Table 1). The accumulation of substrate likely weakens the intrinsic potency of the competitive 2-pyridone scaffold relative to the uncompetitive diphenyl ether scaffold by more than 20-fold (Fig. 3A). Thus, further optimization of 2-pyridones is needed to achieve potent cellular activity. As we demonstrate below, such optimization is much more readily attained in the case of saFabI compared with other FabI homologues, providing the rationale for the narrow spectrum activity of CG400549 and related 2-pyridone compounds.
Clinical Candidate CG400549 Interacts Tightly with saFabI-CG400549 binds with high affinity (K i ϭ 1.27 nM) to saFabI (Table  1). To provide insight into the underlying molecular interactions, we solved two different saFabI-NADPH-CG400549 ternary complex structures (CG400549-I and CG400549-II, respectively; unless stated otherwise, the CG400549-I structure was used for the following analyses; see also supplemental Table S1). Based on the associated 2F o Ϫ F c omit maps, we unambiguously reveal the  (Table 1) (31), the mechanism of inhibition shown in Fig. 2B, and the kinetic model described in Ref. 31. The range of substrate concentration that best correlates relative K i app to relative MIC for both classes of compounds is shaded in gray. B, double logarithmic plot depicts a strong linear correlation between K i of the overall ternary complex (K i ϫ K d, NADP(H) ) and MIC for the 2-pyridone series (F) and diphenyl ethers (E). Points corresponding to 4-pyridones (ϫ) are superimposed. Note that in this plot, the MIC for PT170 was assumed to be 550 M, which is a lower limit estimate. binding mode of CG400549 (Fig. 4A), which enables the formation of two central hydrogen bonds of the pyridone carbonyl oxygen with the Tyr-157 hydroxyl and the NADPH nicotinamide ribose 2Ј-OH at distances of 2.74 Ϯ 0.07 and 2.69 Ϯ 0.10 Å, respectively (Fig. 4B). Although reminiscent of interactions found for other inhibitor scaffolds (Fig. 1), these hydrogen bonds are 0.15 Å longer for pyridones in comparison with diphenyl ethers. This supports our hypothesis that diphenyl ethers bind to saFabI in their deprotonated form (31) leading to shorter charge-assisted hydrogen bonds. Additional long range hydrogen bonds are formed between the 3Ј-amino group of CG400549 (Fig. 1) and the main chain oxygen and nitrogen of Ala-97 at distances of 3.56 Ϯ 0.06 and 3.32 Ϯ 0.09 Å, respectively (Fig. 4B). Moreover, a water molecule is frequently bound between this B-ring NH 2 group and Ala-95 at 3.36 Ϯ 0.18 Å (Fig. 4B). Interestingly, Ala-97 is engaged in direct interactions to several potent saFabI inhibitors (Fig. 1). Triclosan and AFN-1252 form halogen and hydrogen bonds with Ala-97, respectively, and the amide group of MUT056399 was similarly suggested to interact with this residue (Fig. 1) (6, 23, 24).
The unique 5-substituent of CG400549 (atoms of the diphenyl ether and pyridone scaffolds are numbered as indicated in the two lower boxes of Fig. 1) binds to a hydrophobic pocket with the thiophene moiety trapped between the three aromatic amino acids Tyr-147, Tyr-157, and Phe-204 (Fig. 4, A and B). The side-on -stacking interaction between Tyr-157 of the catalytic triad and the thiophene ring may contribute to the high affinity of CG400549 toward saFabI (Table 1). In particular, the thiophene sulfur interacts with the edge of Tyr-157 (C ␦ ) at a distance of 3.97 Ϯ 0.11 Å (62). A similar interaction is found between this sulfur atom and Phe-204 (C ␦ ) at 4.15 Ϯ 0.16 Å (Fig.  4B). Remarkably, among 13 CG400549-resistant S. aureus strains, 10 were characterized by a single F204L mutation, whereas the residual three had no mutations in the fabi gene (30). In contrast, multiple resistance mutations are known for diphenyl ethers (8,9,26,64,65). Accordingly, the Phe-204thiophene interaction seems to be critical for the activity of CG400549.
In addition to the CG400549 structures, we solved an saFabI structure in complex with NADPH and PT173, which displays a similar affinity toward saFabI as CG400549 (K i ϭ 1.97 nM versus 1.27 nM) and contains a 5-hexyl group that mimics the natural enoyl-ACP substrate (Fig. 1) (31). Although the resolution was much lower (supplemental Table S1), clear density was observed for the cofactor and inhibitor (Fig. 4C). The binding mode of PT173 in the saFabI active site pocket is similar to CG400549, and we observed the same central hydrogen bonding network between the inhibitor, cofactor, and Tyr-157 (Fig.  4D).
Despite the relative success of CG400549, little is known about pyridone FabI inhibitors and their structure-activity relationships (SAR) (66 -68). Thus, we synthesized a series of pyridone compounds and investigated their ability to inhibit saFabI (Table 1). To rationalize the SAR results, we generated putative binding modes for all investigated inhibitors using a validated docking procedure, which could reproduce the CG400549 and PT173 binding geometries with low root mean square deviations (0.71 and 0.83 Å, respectively) (supplemental Fig. S1). Because of the lipophilic environment, bulky and hydrophobic substituents are preferred at the 5-position. Hence, PT170 is the least potent compound of our pyridone series. Replacing the 5-methyl group by a 5-hexyl group (PT172) leads to a 21-fold affinity enhancement, which is underlined by the favorable scores for the additional carbon atoms (supplemental Fig. S1). Similar to the 2Ј-chloro substituent of PT172, the 2Ј-methyl group of PT171, which is also present in CG400549, increases affinity by a factor of 2 compared with the unsubstituted analogue PT179. In comparison with the 2Ј-Cl and 2Ј-Me groups, a 2Ј-CN substituent leads to decreased potency. Interestingly, the SAR at this position is different from diphenyl ethers for which 2Ј-cyano is the best substituent (31). Pyridone 2Ј-substituents are predicted to bind in a similar orientation (supplemental Fig. S1) as observed for diphenyl ethers. The introduction of a 3Ј-amino group as present in CG400549 and PT173 further enhances the affinity of PT171 by a factor of 6, which highlights the energetically favorable character of the observed long range hydrogen bonds between those inhibitors and Ala-97 (Fig. 4B). Replacement of the PT173 5-hexyl group with the 5-substituent of CG400549 further improves the affinity of the drug candidate 1.5-fold and might be explained by the additional aromatic interactions we observed.
saFabI Conformational States Differ between Pyridone and Diphenyl Ether Ternary Complex Structures-In accordance with the distinct kinetic behavior of pyridone and diphenyl ether inhibitors, we observed considerable structural differences between the corresponding ternary complex structures. A per-residue root mean square deviation plot reveals variations in mainly three regions of the protein (Fig. 5A), the two substrate-binding loops (SBL and SBL-2; residues 194 -204 and 94 -108) and the phosphate recognition helix ␣2 (residues 40 -54), which confers the unique NADPH specificity to saFabI (6). Complete closure of the SBL has been proposed to constitute the rate-limiting step of slow binding FabI inhibition by diphenyl ethers (16). Indeed, we found more "open" SBL states in the saFabI-NADPH-CG400549 structures compared with the triclosan-bound structure (Fig. 5B). Most likely the observation of the open state can partially be attributed to steric interference between the thiophene moiety of CG400549 and the side chain of Val-201 in the SBL if it would adopt the completely "closed" state ( Fig. 5B, gray). Hence, Val-201 is shifted by 2.9 Ϯ 0.1 Å into the more open CG400549-bound structure (Fig. 5B, cyan). Interestingly, some subunits of the CG400549-I structure reveal a second, even more open SBL state; the root mean square deviation increases from 1.6 to 2.0 Å compared with the triclosan-bound reference structure, with an additional shift in helix ␣7, which we previously identified to be very FIGURE 5. Structural variations between pyridone and diphenyl ether ternary complex structures. A, structural differences between diphenyl ether and pyridone ternary complexes. Per residue root mean square deviation (RMSD) values between our triclosan-bound (PDB code 4ALI, subunit H) and CG400549-bound (CG400549-I, subunit C) structures were calculated using Theseus (79) and are plotted against the residue number. The inhibitor-bound structures of these two scaffolds differ considerably in three regions of the protein (PRH ϭ phosphate recognition helix ␣2; SBL-2 ϭ substrate-binding loop 2; SBL ϭ substrate-binding loop). B, conformational states of the SBL. The different subunits of our CG400549-I structures reveal two distinct states of the SBL and the attached helix ␣7 (shown in cyan and yellow for subunits A and C, respectively; residual parts of the protein are displayed in transparent colors for clarity). Compared with these conformations, the SBL is more closed in our triclosan-bound structure (shown in gray; PDB code 4ALI, subunit H). Detailed views of the three different conformations are displayed in the insets. Red arrows indicate the conformational changes from the closed to the open substratebinding loop states. Cyan arrows highlight backbone flips. C, conformational states of the SBL-2. Selected residues, the inhibitors, and the SBL-2 are shown for the CG400549-I (cyan; subunit A) and triclosan-bound structures (gray; PDB code 4ALI, subunit H).
flexible prior to ligand binding (Fig. 5B, yellow) (6). Both SBL conformations in the saFabI-NADPH-CG400549 complex seem to be energetically equally favorable due to the observation of subunits with both states in equilibrium. The more closed state is stabilized by a sulfate ion, which is bound to backbone amides of the SBL; this state differs from the completely closed triclosan-bound structure by a Lys-199 to Gly-200 backbone flip (Fig. 5B, right inset; K199 and G200 change by 156°and 159°, respectively), which is observed for seven of the eight CG400549-I monomers and might also be responsible for the opening movement. An additional shift of residues 202-208 with extensive variations for Gly-202 and Gly-203 leads to the most open form (Fig. 5B, left inset), which is related to the conformation observed for the AFN-1252 structure (23).
Similar to the SBL, a second loop that also contributes to the substrate binding pocket (SBL-2) was almost exclusively found to be in an open conformation in the saFabI-NADPH-CG400549 structures (Fig. 5C, cyan). In contrast, we observed alternative closed and open SBL-2 conformations for the diphenyl ether-bound structures (6). An Arg-103 to Gly-104 backbone flip (Fig. 5C, left inset; R103 and G104 change by 180°a nd 144°, respectively) clearly differentiates the two states. This flip is induced by (or induces) a 1.5 Ϯ 0.5 Å shift of Leu-102 out of the binding pocket, which in turn might be caused by (or causes) a considerable movement of the CG400549 B-ring toward this residue (Fig. 5C). The resulting differences between the pyridone and diphenyl ether binding poses (Fig. 5C) rationalize the varying SAR profiles at the 2Ј-position for both scaffolds and the success of the 3Ј-amino substituent in the case of pyridones. In particular, the 2Ј-and 3Ј-carbon atoms are relocated by 1.3 Ϯ 0.1 and 1.6 Ϯ 0.1 Å, respectively, which places the CG400549 3Ј-carbon at the position of the triclosan 4Ј-carbon. Consequently, 3Ј-pyridone and 4Ј-diphenyl ether substituents are ideally oriented to interact with the important anchor residue Ala-97. For instance, the triclosan 4Ј-Cl is halogen-bonded to the free electron pair of the Ala-97 carbonyl oxygen via the favorable linear geometry (Fig. 5C, right inset) (6,62). To enable an equally favorable angular geometry for a hydrogen bond between the CG400549 3Ј-NH 2 group and Ala-97 along one amino hydrogen atom (62), the Ala-97 carbonyl oxygen is shifted by 0.8 Ϯ 0.1 Å toward the B-ring of the inhibitor, which in turn approaches Ala-97 to reduce the interaction distance (Fig. 5C, right inset). As exemplified by these considerations, the exact knowledge of the different conformations that can be attained by a protein target is pivotal for the design of improved inhibitors.

Comparison of ecFabI and saFabI Inhibition by 2-Pyridones-
To determine whether the narrow spectrum behavior of CG400549 can be partly attributed to target-specific differences between species, we obtained and compared the structures and inhibition kinetics of CG400549 with respect to E. coli and S. aureus FabI. Indeed, CG400549 shows a 65-fold reduced affinity to ecFabI in comparison with saFabI (Tables 1 and 2), which translates into Ͼ64-fold lower antibacterial activity (Table 3). Nevertheless, the inhibitor clearly bound to the binary ecFabI-NADH complex with a similar binding geometry as observed for saFabI (Fig. 6, A and B). However, in contrast to the related saFabI ternary complex, the substrate-binding loop and the attached helix ␣7, including residues 193-214, were found to be disordered in this ecFabI structure (Fig. 6A). Based on a comparison of these structures, we propose that the saFabI residues Val-201 and Ile-207, which are located in this region of the protein, contribute to the specificity of CG400549 toward saFabI. The more bulky Ile-200 and Met-206 ecFabI residues (corresponding to saFabI positions 201 and 207, respectively) restrict the available space for the large 5-substituents as present in CG400549 (Fig. 6B). Accordingly, the elongation of the 5-substituent in 5-ethyl-2-phenoxyphenol (PT01) to 5-hexyl-2phenoxyphenol (PT04) does not enhance the affinity toward ecFabI (Table 2), whereas the affinity increases ϳ10-fold for saFabI (31). We have recently shown that these inhibitors are transition state analogues with the 5-substituent alkyl chain extending toward the fatty acyl binding channel (31). Although the k cat increases for longer enoyl substrates in the case of saFabI (31), it is similar among the different substrate chain lengths for E. coli and F. tularensis FabI, which both carry the V201I and I207M substitutions and are inhibited by PT01 and PT04 with similar potency (6,33,69,70). Thus, the enlarged binding pocket of saFabI might partially explain the specific action of the comparatively bulky CG400549 and AFN-1252 clinical candidates (Fig. 1). Interestingly, the C terminus of the ecFabI monomer located on the opposite side of the homotetrameric protein (in particular, Met-256Ј) seals this acylbinding cavity and likely restricts the side chain mobility of Met-206 (Fig. 6B). The resulting steric interference between the substrate-binding loop residues Met-206 and Ile-200 and the thiophene moiety of CG400549 presumably results in the experimentally observed enhanced mobility of this loop. In contrast, the C terminus of saFabI is shorter and lacks a residue corresponding to Met-256Ј, thus enabling Ile-207 to move away from bulky 5-substituents. Consequently, important contacts between this inhibitor and the substrate-binding loop (e.g. with Phe204; see also Fig. 4B) are more readily attained, partially explaining the selectivity of CG400549 toward the saFabI homologue.
In line with our assumption that ecFabI is a good model for all FabIs insensitive to CG400549, an alignment of the FabI sequences from clinically relevant pathogens, which can be sensitive to FabI-specific inhibitors, reveals that staphylococcal FabIs differ fundamentally from classical FabI proteins such as ecFabI (Fig. 7). Most strikingly, all nonstaphylococcal FabIs included in this comparison contain extended C termini. A comparison of the available corresponding structures clearly shows that the four amino acids following the C-terminal Lys-256 of saFabI occlude the acyl-binding cavity for classical FabIs and thus restrict the available space for bulky 5-substituents (red box in Fig. 7). In particular, the additional large hydrophobic residues at position 257 (256 for ecFabI) or 259 (Met, Leu, and Ile) will most likely interfere with bulky residues located at position 207, which thus cannot avoid the interference with large 5-substituents without the opening of the SBL (Fig. 6B). The length of the C terminus might therefore be an ideal indicator whether or not the corresponding FabI is sensitive to compounds such as CG400549 or AFN-1252. The presence of Val-201 instead of an isoleucine is an additional unique characteristic of FabIs sensitive to the more voluminous compounds and may facilitate the production of bulky branched-chain fatty acids abundant in S. aureus (Figs. 6B and 7) (6).
In addition to the tolerance for bulky 5-substituents, hydrogen bonding interactions with Ala-97 seem to be particularly favorable in the case of saFabI and are exploited by the three drug candidates, as indicated by crystallographic and computational studies (Fig. 1) (23, 24). In the case of CG400549, the more rigid Pro-96 of ecFabI (Asn-98 in the case of saFabI) slightly changes the orientation of the Ala-95 carbonyl (Ala-97 for saFabI) (Fig. 6B). This might explain the observed rotation FIGURE 6. Rationalizing the Staphylococcus-specific activity of CG400549 and designing PT166 with an extended activity spectrum. A, CG400549 (lilac) in its complex with ecFabI and NADH. The 2F o Ϫ F c omit map is depicted for CG400549 (shown as mesh at 1) and clearly reveals the presence of this molecule, although the SBL region was found to be disordered in this structure (the disordered region starts after residue 192 (colored in red) as highlighted by a red arrow; subunit A of the ecFabI-NADH-CG400549 structure is shown in yellow schematic representation). As a reference, the SBL of a superimposed ecFabI-NAD ϩtriclosan structure is shown in light green (PDB code 1QSG) (13). Together with the SBL, the loop comprising Ala-95 defines a portal toward the solvent. B, comparison between the CG400549-bound S. aureus (inhibitor in cyan, for clarity only present in the inset, amino acids in gray) and ecFabI (inhibitor in lilac, amino acids in yellow) ternary complexes. The inset contains a comparison between the CG400549 binding modes, which differ in their B-ring conformations via changes in the torsion angles between the two aromatic rings (indicated by red arrowheads). Met-256Ј of the subunit on the opposite side of the ecFabI homo-tetramer is depicted in pink. Residues Ile-200 and Met-206 of the 1QSG structure (ecFabI-NAD ϩ -triclosan) are shown as reference in green. C, putative binding mode of PT166 (yellow sticks) within the saFabI pocket (CG400549-I structure, subunit A). PT166 was docked into the saFabI binding cavity using the validated approach described under "Experimental Procedures." The radii of the blue (red) spheres indicate the values of the favorable (unfavorable) score for each individual atom, as determined with DrugScore X . Our bpFabI-NAD ϩ -PT155 structure (light green sticks) confirms this binding mode and suggests a rotation of Phe-204 to avoid the steric interference with the N-methyl group of 4-pyridone inhibitors (indicated by the red sphere). Moreover, the putative binding mode is in accordance with our PT166-bound ecFabI structure (orange sticks, subunit A). The docking results for the residual pyridone inhibitors of Table 1 are summarized in supplemental Fig. S1. D, 2F o Ϫ F c omit map (shown as mesh at 1 ) for PT166 (orange sticks) bound to ecFabI. An intersection of the ecFabI-NADH-PT166 structure (subunit A, depicted in yellow surface representation) is shown. The SBL, which usually covers the binding site (in front of the cavity), is disordered.
in the B-ring of CG400549, which seems to be required for the maintenance of the hydrogen bond to this residue. Interestingly, this interaction is solvent-exposed (Fig. 6A) and less buried in ecFabI due to the presence of Gly-97, which corresponds to Met-99 in saFabI (Fig. 6B). This observation rationalizes why the 3Ј-amino substituent leads to a decrease in affinity in the case of ecFabI (Table 2) because solvent-exposed interactions tend to be energetically less favorable (62). Thus, other classical FabI proteins harbor a more hydrophilic residue at this position compared with saFabI (Fig. 7). Furthermore, treatment of S. aureus with AFN-1252, which also binds to the saFabI-NADPH complex and forms a hydrogen bond with Ala-97 (Fig. 1), predominantly selects for an M99T mutation, and it was recently suggested that this residue confers S. aureus selectivity to AFN-1252 (23,71). Thus, the buried hydrogen bonding interaction with Ala-97 likely plays an important role in the selective affinity of CG400549 and AFN-1252 toward saFabI.

4-Pyridone PT166 Is a Potent FabI Inhibitor with Extended Spectrum in Vitro Activity and in Vivo Efficacy against S. aureus-
In contrast to saFabI, strategies to optimize the binding affinity of 2-pyridones for ecFabI and related homologues may be lim-ited due to the more constricted space in the binding crevice combined with the relatively low intrinsic potency of 2-pyridones compared with the diphenyl ether scaffold. Instead, to attain broad spectrum activity, a modified scaffold should be designed that possesses a higher intrinsic potency for all FabI homologues. One possibility is the replacement of the methylene bridge with an ether linkage, thereby changing the conformational preference prior to binding. The biologically active Ar-X-Ar conformation (Ar ϭ aromatic ring, X ϭ CH 2 or O, Figs. 4 and 6) is more readily available for the bisaryl ether system thus leading to an entropic advantage upon binding (72). This was implemented in the design of C-substituted 2-pyridone (in contrast, 2-pyridones such as CG400549 are substituted at the nitrogen atom) and 4-pyridone analogues. As expected, both prefer to bind in a ternary complex with NADPH (supplemental Fig. S2), thus preserving the 2-pyridone mode of action. Compared with the analogous N-substituted 2-pyridone PT179, the C-substituted 2-pyridone PT191 binds 10-fold less tightly to saFabI (Table 1). Although the enol form of PT191 displays a similar putative binding geometry as PT179 (supplemental Fig. S1) and more closely resembles the potent diphenyl ether PT04 (31, 66), the increased desolvation costs  (80). These organisms are known to contain FabI as the sole enoyl-ACP reductase (exception: FabV is additionally present in B. pseudomallei) (23,24,29,81). Residues that are likely responsible for the unique NADPH specificity, enhanced mobility (6), and enlarged binding pocket of saFabI are indicated by magenta, black, and red stars, respectively. The C-terminal regions occluding the acyl-binding cavity and the RKXXS motif conferring NADPH specificity are highlighted with red and magenta boxes, respectively. The blue star indicates the location of Met-99. This figure was prepared using ESPript (82).
for this more polar compound are likely responsible for the decreased affinity. In contrast, the 4-pyridones PT166 and PT159 have 3-4-fold enhanced affinities for saFabI with respect to the analogous 2-pyridones PT171 and PT420 (Table  1), consistent with a higher intrinsic potency of this scaffold. PT166 also potently inhibited ecFabI ( Table 2), suggesting that the higher intrinsic potency may indeed translate to broad spectrum activity. Thus, substituting the methylene bridge of the initially reported 4-pyridones (67) with an ether linkage (73) seems to be a successful strategy to further improve FabI pyridone inhibitors.
Based on our docking studies, we suggest a binding mode for the 4-pyridones similar to those observed for 2-pyridones (Fig. 6C). However, despite the enhanced affinity of PT166 with respect to PT171, the additional 4-methyl group is characterized by an unfavorable score. Thus, we hypothesized that Phe-204 will change its conformational state upon 4-pyridone binding to avoid the steric interference with the N-methyl group. Because 4-pyridones did not co-crystallize with saFabI, we initially used PT155, which carries an additional 4Ј-NH 2 group compared with PT166, and we determined a ternary complex structure with B. pseudomallei FabI (bpFabI). Indeed, this structure confirmed the predicted 4-pyridone binding mode, and Phe-203 (corresponding to the saFabI residue Phe-204) was found to adopt a different conformation presenting one of its -faces to the N-methyl group of PT155 (Fig. 6C). Finally, we were able to solve an ecFabI-NADH-PT166 structure (Fig. 6D), which further validated the proposed PT166-binding mode (Fig.  6C). Similar to the ecFabI-NADH-CG400549 structure, the substrate-binding loop was found to be disordered or in a very open conformation in these ecFabI and bpFabI structures, respectively. Interestingly, an open SBL conformation was very recently also observed in the structure of PT155 bound to InhA (74).
For the 4-pyridone analogues, in vitro MIC measurements against S. aureus RN4220 lie near the 2-pyridone linear correlation, consistent with their similar mode of action (Fig. 3B). To confirm that the main target of PT166 in vivo is indeed saFabI, we performed selection experiments, which resulted in mutations of the fabI gene in 4 of 5 sampled colonies, including the previously characterized A95V variant (Table 4) (8). Relative to CG400549, PT166 had a Ͼ50-fold lower MIC against an E. coli MG1655 strain lacking the efflux pump AcrAB, consistent with its more potent inhibition of ecFabI (Table 2). Gratifyingly, relative to CG400549, PT166 also exhibited improved activity against the Gram-negative pathogens F. tularensis and P. mirabilis (Table 3). Thus, we rationally extended the spectrum of antibacterial activity by using ecFabI as the paradigm for Gramnegative FabI homologues. Furthermore, similar optimization principles were applicable for the inhibition of mycobacterial FabI proteins. PT166 bound more potently than CG400549 to the M. tuberculosis enoyl-ACP reductase InhA with K i values of 22 and 582 nM, respectively, and translated to significantly enhanced in vitro anti-mycobacterial activity (MIC ϭ 10.5 M; Table 3).
Pyridones are a metabolically stable alternative to diphenyl ethers, which contain a hydroxyl group that can be susceptible to glucuronidation and sulfonation (75). This PK advantage may be key to the success of the clinical candidates CG400549 and AFN-1252. Importantly, PT166 maintained a superior PK profile compared with the diphenyl ether PT04 (Table 5). C max and AUC 0 -24 h (where AUC 0 -24 h is the area under the plasma concentration-time curve over 24 h) of PT166 are 9-and 5-fold higher than that of PT04, respectively, despite the fact that the dose of PT04 was double that of PT166. We also tested the efficacy of PT166 in a neutropenic mouse model of MRSA infection. As a control, no bacteria were observed in the right thigh muscle of both treated and untreated mice, confirming the lack of significant migration of bacteria. In the infected thigh, however, significant bacterial burden was observed in the different treatment groups. As expected, oxacillin, a clinical antibiotic similar to methicillin, exhibited no in vivo antibacterial efficacy (Fig. 8). However, a 100 mg/kg intramuscular dose of PT166 significantly decreased the bacterial burden in the infected thigh by 2.8 log cfu/g tissue. In comparison, the same dose of the diphenyl ether PT04 only decreased the bacterial burden by 0.9 log cfu/g tissue.

DISCUSSION
Pyridones constitute a very promising and relatively new class of FabI inhibitors (16). Compared with diphenyl ethers, CG400549 has superior pharmacokinetic properties and proven clinical efficacy against S. aureus infections. However, it also has lower activity against many other important pathogens. Understanding the molecular basis for such selectivity can guide the development of pyridone-based FabI inhibitors with broad spectrum potential.
Despite the structural similarity of the pyridone and diphenyl ether scaffolds, we observed differences with respect to their mode of inhibition, which have significant implications for the spectrum of activity. Pyridones bind predominantly to the E-NADPH complex, whereas diphenyl ethers exclusively inter-act with E-NADP ϩ generated via catalysis (Fig. 2, A and B) (31). We recently proposed that diphenyl ethers bind in a deprotonated state to their target (31). In such a scenario, the positively charged and thus electron-deficient oxidized nicotinamide ring of the cofactor forms a charge-assistedstacking interaction with the electron-rich phenolate A-ring (Fig. 1). The inability of the much less acidic pyridones to form this ionic interaction might explain their reduced affinity for E-NADP ϩ .
The diphenyl ether and pyridone-bound saFabI ternary complexes likely reflect different stages during the hydride transfer step of the enzymatic reaction (Fig. 9A). We have recently shown that diphenyl ethers are analogues of a late enolate-like transition state (31). In contrast, the pyridone structure more closely resembles the enoyl-ACP substrate (Fig. 9A) and thus FIGURE 9. Conformational states sampled along the reaction coordinates of inhibitor binding and substrate turnover. A, pyridone and diphenyl ether inhibitors resemble different species along the enzymatic reaction coordinate. In contrast to the enolate-like diphenyl ethers which are transition state analogues (31), pyridones are more substrate-like. The corresponding moieties of inhibitors and species along the reaction coordinate are highlighted in green. B, conformational states sampled by the SBL and the attached helix ␣7 of saFabI. Red arrows indicate the conformational changes we propose to occur during the enzymatic reaction (see also C). Prior to the binding of cofactor and substrate or inhibitor, the SBL is disordered, and helix ␣7 attains a very open conformation (state 1 colored in pink ϭ PDB code 4ALM, subunit B; state 2 colored in lilac ϭ PDB code 4ALM, subunit C; further details about the conformational changes upon ligand binding are provided in our previous report (6)). The more substrate-like pyridone inhibitors likely induce a conformational state between the ternary E-NADPH-S complex and the transition state of the hydride transfer (state 3 colored in yellow ϭ CG400549-I structure, subunit C; state 4 colored in cyan ϭ CG400549-I structure, subunit A; see also Fig. 5B). In contrast, the transition state analogue triclosan and several other diphenyl ethers (6,31) induce the likely fully closed state of the SBL (state 5 colored in gray ϭ PDB code 4ALI, subunit H). C, qualitative energy diagram for substrate turnover by saFabI. The numbers 1-5 indicate the conformational states (colors according to B), which are likely sampled along the reaction coordinate of the enzymatic reaction (see also B). D, approximate energy diagrams for saFabI in complex with NADPH and pyridone (black) or NADP ϩ and diphenyl ether inhibitor (red). The overall affinities of both ternary complexes are assumed to be identical. By shifting stabilization from cofactor to inhibitor, the residence time of the overall complex is increased. This rationalizes the difference in off-rate kinetics between the diphenyl ethers and pyridones. Note that, technically, k off for the pyridone complex (E-NADPH-I) is equal to k Ϫ2 because the E-NADPH complex is catalytically active. C is defined as a constant with a value greater than 1. The numbers 1-5 indicate the conformational states (colors according to B), which are likely sampled along the reaction coordinate of inhibitor binding (see also B). binds preferably in a ternary complex with the reduced cofactor, which is present prior to the hydride transfer. We envision an incremental closure of the SBL during substrate binding and hydride transfer in which the loop is fully closed at the transition state to minimize its energy (Fig. 9). In this respect, the different conformations of the SBL presented in this study define important structural snapshots along the reaction coordinate of enzyme catalysis (Fig. 5B). The observation of an opened substrate-binding site for pyridones confirms our hypothesis that these inhibitors are more substrate-like compared with the diphenyl ethers. In the light of the two alternatively ordered SBL conformations versus the disordered loop in the apoenzyme (Fig. 9B, states 1 and 2) (6, 32), it could be argued that the pyridone ternary complexes represent a state between the substrate complex and the transition state for enolate formation (Fig. 9C). The electron-donating effect of the pyridone nitrogen, which leads to a phenolate-like resonance structure, might thereby mimic the transfer of the negatively charged hydride ion (Fig. 9A).
As the hydride transfer reaction proceeds, the increasing positive charge on the nicotinamide ring and the closure of the substrate-binding loop shifts the balance of ternary complex stabilization more toward the fatty acyl relative to the cofactor component, which is reflected in the relative affinities of the investigated inhibitors and cofactor forms. The transition state mimicking diphenyl ethers display affinities that are 3 orders of magnitude higher compared with analogous substrate-like 2-pyridones, whereas NADP ϩ binds ϳ1000-fold less tightly to saFabI with respect to NADPH (Fig. 9D) (31). Accordingly, the cumulative ternary complex affinities (K i ϫ K d, NADP(H) ), which also integrate cofactor affinity, of comparable E-NADP ϩ -diphenyl ether and E-NADPH-pyridone complexes are calculated to be very similar. There is only a 2-fold difference in the cumulative affinity of the E-NADP ϩ -PT04 complex compared with the analogous E-NADPH-PT179 complex. Diphenyl ethers are stabilized to a greater extent partly because of the closure of the SBL that occurs farther along the reaction coordinate. At this stage, NADP ϩ exists in the ternary complex, which also enhances the binding affinity of deprotonated diphenyl ethers. However, in comparison with E-NADPH, the steady-state concentration of E-NADP ϩ is very small due to the fast dissociation of the oxidized cofactor generated via catalysis (31). Hence, the resulting apparent inhibitor association rate (k on ⅐[E-NADP ϩ ]⅐[I]) of diphenyl ethers is slow explaining the observed slow binding phenomenon, although the actual association rate constants k on of pyridones and diphenyl ethers are very similar. Thus, in saFabI the ordering of the substrate-binding loop is likely correlated rather than causative with respect to the observation of slow-onset kinetics with diphenyl ethers. The slow-off kinetics observed with diphenyl ethers is likely the consequence of its potent thermodynamic affinity.
Among the three saFabI inhibitors investigated in clinical trials, CG400549 and AFN-1252 have been shown to be Staphylococcus-specific (23, 28 -30), whereas the diphenyl ether MUT056399 is also active against several Gram-negative pathogens (24). As depicted in Fig. 3, this is likely attributed to different modes of inhibition. Substrate accumulation weakens the intrinsic potency of competitive inhibitors in contrast to uncompetitive inhibitors. Thus, relatively unmodified diphenyl ethers are already able to potently inhibit cell growth. In contrast, further optimization of binding affinity is necessary for pyridones to achieve cellular efficacy, particularly in the presence of active efflux pumps. This argument likely extends to the naphthyridinones, such as AFN-1252, which was recently crystallized in complex with saFabI and 3Ј-NADPH (Fig. 1) (23).
Our structural data rationalize the pyridone SAR profile and clearly reveal the ability of the CG400549 5-, 2Ј-, and 3Ј-substituents to enhance its affinity toward saFabI ( Fig. 4 and Table  1). In particular, the hydrogen bond formed between the CG400549 and PT173 3Ј-amino group with Ala-97 leads to a 6-fold increase in affinity (Fig. 4B). In the case of ecFabI, this hydrogen bond is not protected from solvent exposure due to an M99G substitution (Fig. 6B), thereby reducing the affinity of such compounds and increasing the selectivity of CG400549 for saFabI ( Table 2). Because of the amino acid residues Ile-200 and Met-206 and an elongated C terminus, including Met-256Ј, ecFabI also harbors a smaller binding pocket compared with saFabI (Fig. 6B). The wider acyl cavity of saFabI facilitates the accommodation of branched and longer acyl substrates or, analogously, bulky diphenyl ether 5-substituents (6,31,33,70). Importantly, CG400549 and AFN-1252 contain large moieties at this position, which might further contribute to the Staphylococcus-specific spectrum of these two clinical trial inhibitors (Fig. 1). In fact, bulky pyridone 5-substituents interfere with the ecFabI and bpFabI SBL, although it is in a relatively closed state for saFabI with bound AFN-1252 (23) and CG400549 (Figs. 5B and 6A). This important difference might be further explained by the presence of the additional C-terminal extension in typical FabIs such as ecFabI and bpFabI, which occludes their substrate-binding sites (Fig. 7). In contrast, saFabI contains a significantly shorter C terminus compared with most other structurally characterized FabI proteins. The wider acyl cavity and thus the enhanced affinity of CG400549 can be related to the ability of saFabI to efficiently utilize bulky branched-chain fatty acyl substrates (6).
Based on the SAR profile of ecFabI, the ability to optimize binding affinity via substituents on the scaffold is very limited. A pyridone-based compound with broad spectrum activity must necessarily have higher intrinsic potency than the 2-pyridones. To achieve this, we used a 4-pyridone scaffold that retains the bridging oxygen of diphenyl ethers, thus providing an entropic advantage upon binding to FabI. Although the 4-pyridone inhibitor PT166 shares features with both the 2-pyridone and diphenyl ether scaffolds, its inhibition mechanism is the same as observed for 2-pyridones (Fig. 2B). Using the thermal shift assay, we could clearly show that this compound inhibits saFabI at the E-NADPH stage (supplemental Fig. S2), just like CG400549 (Fig. 2A). The 2-pyridone/substrate-like behavior of PT166 is further confirmed by a comparison of our ecFabI structures. In contrast to the triclosan-bound structure (13), where the SBL is in a closed state, this loop is disordered in the complexes with PT166 and CG400549 (Fig.  6). Furthermore, the structure of InhA in complex with the 4-pyridone PT155 shares a very similar open SBL conformation with the substrate-bound form of this enzyme (74,76). As expected based on these considerations, the SAR and predicted binding mode for the 4-pyridones is highly reminiscent of the 2-pyridone series, but 4-pyridones possess superior potency at the enzymatic and cellular levels (Tables 1 and 2). Importantly, the higher intrinsic potency also translates into an extended spectrum of activity for the promising lead compound PT166 (against S. aureus, E. coli, F. tularensis, P. mirabilis, and M. tuberculosis; Table 3). As with CG400549 (77), replacement of the metabolically labile hydroxyl group with a carbonyl successfully improved the pharmacokinetic profile of PT166 compared with the diphenyl ether PT04 (Table 5). Additionally, this compound primarily acts on target (Table 4), and it significantly reduced bacterial burden in a murine model of MRSA infection (Fig. 8), validating its potential as a drug lead for future optimization and development. Further in vivo studies are needed to more precisely compare the pharmacokinetics and efficacy of the diverse FabI inhibitor scaffolds.
In summary, we have elucidated the structural and mechanistic basis for selective saFabI inhibition by pyridones, including the clinical candidate CG400549. Our study has yielded an insightful glimpse into conformational states along the coordinate of the saFabI enzymatic reaction, providing yet another reminder of the elegant connection between catalysis and inhibition. Importantly, we were able to rationally design the promising lead compound PT166, which merges the pharmacokinetic advantages of a pyridone with the potential for an extended spectrum of antibacterial activity. ecFabI and InhA served as paradigms for enoyl-ACP reductase homologues in Gram-negative and mycobacterial organisms, respectively. A similar approach can be applied toward the development of much needed narrow and broad spectrum antibiotics against novel targets.